ADHD Medications and Preadolescent Brain Structure: Patterns of Cortical Attenuation from the ABCD Study

Attention-deficit/hyperactivity disorder (ADHD) is the most common neurodevelopmental disorder in the U.S., and the stimulant and nonstimulant medications used to treat ADHD are among the most widely prescribed treatments in youth. Stimulants—including amphetamine-based (AMP) and methylphenidate-based (MPH) medications—act primarily on dopaminergic and noradrenergic systems, while nonstimulants (NS) more selectively target noradrenergic pathways. Although pharmacotherapy is the most clinically effective treatment, its neurostructural effects remain poorly understood. Leveraging the Adolescent Brain Cognitive Development Study (ABCD Study®), we used a machine learning approach to identify neuroanatomical targets of medications, followed by linear mixed-effects modeling to estimate the effects of ADHD status and medication class (AMP, MPH, NS) on cortical thickness, surface area, and cortical and subcortical volumes. ADHD was not associated with statistically significant differences; however, a consistent pattern emerged in which AMP and MPH effects attenuated ADHD effects, suggesting that stimulant medications may attenuate ADHD-related cortical patterns. NS medications showed a similar, albeit weaker, effect pattern. Notably, AMP and/or MPH use was associated with significant effects in the right entorhinal cortex and the right banks of the superior temporal sulcus, potentially reflecting overcompensatory effects, as well as in the left posterior cingulate, possibly indicating de novo medication-related differences.

ADHD, amphetamine, methylphenidate, nonstimulants, MRI. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 3

Attention-deficit hyperactivity disorder (ADHD) is the most prevalent neurodevelopmental disorder in the United States, with estimates of the biological prevalence ranging from 2 to 8%. 1–3 Untreated, ADHD can substantially interfere with normal cognitive and functional pediatric development. 4 Among all treatment methods, pharmacologic therapy is the first-line treatment approach and has the most substantial evidence for improving symptoms of inattention, impulsivity, and hyperactivity, while reducing negative impacts on academic, social, and occupational functioning. 5,6 Accordingly, approximately 6% of U.S. children ages 3 to 17 years are treated with one or more medications for ADHD, making these medications some of the most common neuropsychiatric medications prescribed during childhood. 7,8 ADHD medications have been hypothesized to “normalize” differences in brain structure and function. Broadly, medications include stimulant (e.g., amphetamine-based [AMP] and methylphenidate-based [MPH]) medications as the preferred first-line agents and nonstimulant [NS] medications (e.g., norepinephrine reuptake inhibitors and alpha-2-adrenergic agonists) as second-line agents if patients fail to respond to stimulants, experience intolerable side effects, or prefer nonstimulant options. Despite stimulants being the first-line treatment, 40% of children with ADHD display a preferential response to either AMP or MPH, while 30% of children fail to respond. 9–11 The reasons behind variability in ADHD medication response remain unclear. 5,11 Understanding how these medications affect brain phenotypes may help advance precision biomarkers for predicting treatment response. 12 Stimulant medications exert their effects primarily by increasing synaptic levels of dopamine (DA) and norepinephrine (NE) through the inhibition of transporters (e.g., DAT and NET). However, AMP and MPH differ in their additional mechanisms of action: AMP also inhibits monoamine breakdown pathways (e.g., VMAT2 and MAO), while MPH displays weak agonism of serotonergic 5-HT1A and alpha-2-adrenergic receptors. 13 Nonstimulant medications, including alpha-2-adrenergic agonists (e.g., clonidine, guanfacine) and norepinephrine reuptake inhibitors (e.g., atomoxetine, viloxazine), primarily modulate NE and lack the direct DA activity of stimulants. Despite the widespread use of ADHD medications in the U.S. and worldwide, research on their effects on MRI brain phenotypes has been constrained by small, homogeneous samples that lack the statistical power to capture nuanced effects on brain structure and sample diversity to produce generalizable findings. Relatively few studies have examined structural MRI (sMRI) outcomes related to ADHD medication use, in contrast to the larger body of functional MRI literature. 14 Among sMRI studies, stimulant-associated attenuation has been observed in frontal, cingulate, and parietal-occipital cortical regions, 15,16 in basal ganglia and cerebellar subregions, 17–19 and in the corpus callosum and global white matter. 20,21 However, generalizability across these studies remains poor, with inconsistent and null findings. 14 While meta-analyses have linked stimulant treatment in pediatric ADHD to attenuation in right hemisphere volumes of basal ganglia subregions (e.g., globus pallidus, putamen, lentiform), 22,23 they have failed to identify replicable brain differences attributable to ADHD pathology itself, likely due to inter-study variability. 24 However, the prefrontal cortex and basal ganglia are among the most frequently reported grey matter regions exhibiting ADHD-related differences in the literature. 25,26 These findings align well with the roles of DA and NE in mesocortical and noradrenergic pathways that support prefrontal function and executive control, as well as mesolimbic pathways involved in emotion and reward processing. 27 These dopaminergic pathways modulate the affective and cognitive loops of the cortico-striato-thalamo-cortical (CSTC) circuit, where dysregulation is hypothesized to contribute to the pathophysiology of ADHD and morphology in related brain regions. 28,29 Given differences in neurotransmitter pathways and regional brain variation in transporter abundance, secondary receptor interactions, and catabolic enzymes, it is plausible that stimulants and nonstimulants may achieve (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et a similar improvements in ADHD symptoms via distinct neurostructural changes ( Figure 1 ). 13,30,31 However studies have directly compared the neurostructural effects of different ADHD medication classe s (i.e., AM P MPH vs. NS), and few have examined the brain effects of any single class using sufficiently powered sam pl Thus, our objective was first to use an explainable artificial intelligence (xAI) framework for a data- dri approach to identifying potentially medication-sensitive brain regio ns, and then to characterize the d ist neurostructural effects of AMP, MPH, and NS in relation to ADHD status in these regions. We leveraged a la and demographically diverse sample of preadolescents from the NIH- funded Adolescent Brain and Co gni Development Study (ABCD Study® ). We hypothesized that stimulants (AMP and MPH) would be asso cia with attenuation of ADHD- related morphological differences, particularly in the ventromedial prefrontal cor cingulate cortex, and basal ganglia. In cont rast, we expected nonstimulant effects to be more limited, ref lec primarily noradrenergic pathways and possibly affecting parietal regions. Figure 1. Schematic of neurotransmission systems relevant in ADHD pharmacology. A) Dopaminergic and noradren pathways in the brain. Figure adapted from Faraone et. al, 2015. 30 Note: Panel A of this figure is not displayed in the bioR Preprint due to copyright laws. B) Neurotransmitter transporters and receptors relevant to ADHD medications ex varying concentrations throughout the cortex. Figure adapted from Hansen et al., 2022. 31 Abbreviations: DA = dopam (dopaminergic); NE = norepinephrine (noradrenergic); DAT= dopamine tra nsporter; NET = norepinephrine transporte HT1A = serotonin receptor subtype 1A; AMP = amphetamine; MPH = methylphenidate; NS = nonstimulant.

Participants Cross-sectional data were collected as a part of the ongoing ABCD Study and included in the ABCD 5 .1 d release (doi: 10.15154/z563-zd24 ). The ABCD Study enrolled more than 11,800 children 9 and 10 years of (mean age = 9.49; 48% female) between 2016 and 2018 for a 10-year longitudinal study. 32 Participants w recruited at 21 study sites across the United States, aiming to represent nationwide sociodemographic dive rsi Experimental and consent procedures are approved and overseen by the institutional revie w board (IR B) human research protection programs at the University of California, San Diego. Each study site received local approval. Participants provided written assent, and their legal guardian provided written consent to parti cip See Garavan et al. (2018) 33 and Volkow et al. (2018) 34 for additional details. Criteria included English profi cie absence of severe sensory, neurological, medical, or intellectual limitations, and completion of an MRI scan. 35 t al. 4 er, n o P vs. ples. 14 driven istinct a large nitive ciated cortex, lecting energic ioRxiv exhibit amine rter; 5 - .1 data of age w ere rsity. 33 B) and al IRB cipate. ciency, (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 5 The present study analyzes a subset of participant ABCD baseline enrollment data at ages 9 and 10 years, including structural magnetic resonance imaging (sMRI), ADHD diagnostic criteria, and medication usage. For this study, we excluded participants based on sMRI quality control procedures, 36 incidental radiological findings, 37 and other neurologic or psychiatric medications. Participants with missing covariate data were included in the machine learning (ML) stage of our analysis, which did not incorporate confounders; however, they were excluded from the LME modeling stage (N=5). Additionally, to address within-family correlation, we semi-randomly selected one sibling per family for the LME modeling stage, with the caveat that we preferentially retained the sibling with ADHD. Details of the present study's exclusion criteria are in Supplemental Figure 1. Sample characteristics for our study are described in Table 1. Table 1. Study Sample Characteristics Analysis 1: ML-Sample (N=1310) Analysis 2: LME-Sample (N=8873) ADHD Classification ADHD+RXa 753 (57.5%) 699 (7.9%) ADHD-NoRXb 557 (42.5%) 543 (6.1%) Control 0 (0.0%) 7630 (86.0%) ADHD Medication Use AMP 275 (21.0%) 257 (2.9%) MPH 360 (27.5%) 333 (3.8%) NS 161 (12.3%) 151 (1.7%) Age (Months) Mean (SD) 119 (7.43) 119 (7.41) Median [Min, Max] 119 [107, 132] 119 [107, 133] Sex at Birth Male 916 (69.9%) 4635 (52.2%) Female 394 (30.1%) 4237 (47.8%) Race/Ethnicity Non-Hispanic white < 680c 4497 (50.7%) Non-Hispanic Black 227 (17.3%) 1320 (14.9%) Hispanic/Latinx 244 (18.6%) 1923 (21.7%) (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 6 Non-Hispanic Asian < 10c 204 (2.3%) Other 159 (12.1%) 928 (10.5%) Household Income <$50K USD 420 (32.1%) 2453 (27.6%) $50K to $100K USD 336 (25.6%) 2287 (25.8%) ≥$100K USD 447 (34.1%) 3363 (37.9%) Don't know/Refuse to answer 107 (8.2%) 769 (8.7%) Parent Education < HS Diploma 56 (4.3%) 460 (5.2%) HS Diploma/GED 135 (10.3%) 880 (9.9%) Some College 410 (31.3%) 2299 (25.9%) Bachelor’s degree 331 (25.3%) 2189 (24.7%) Post Graduate Degree 378 (28.9%) 3034 (34.2%) Missing/Refuse to Answer 0 (0.0%) 10 (0.1%) MRI Manufacturer GE Medical Systems 303 (23.1%) 2223 (25.1%) Philips Medical Systems 144 (11.0%) 1103 (12.4%) Siemens 863 (65.9%) 5546 (62.5%) a ADHD+RX = medicated ADHD group (i.e., participants using at least one ADHD medication) b ADHD-noRX = unmedicated ADHD group (i.e., participants who met ADHD diagnostic criteria and did not report ADHD medication use) c Small cell sizes (< 10) were masked in accordance with ABCD data-use guidelines. Secondary cell suppression is implemented to ensure category frequencies cannot be recalculated by subtraction. Percentages for masked or suppressed cells are not shown. Abbreviations: ML = machine learning; LME = linear mixed-effects modeling; AMP = amphetamine; MPH = methylphenidate; NS = nonstimulant; HS = high school ADHD and Medication Classification Caregivers completed the computerized Kiddie Schedule for Affective Disorders and Schizophrenia for DSM-5 (KSADS-5) to identify participants meeting ADHD diagnostic criteria, and the Medication History Questionnaire ( M H Q ) , u s e d t o d e t e r m i n e A M P , M P H , a n d / o r N S m e d i c a t i o n u s e . 38 All participants taking antidepressant, antipsychotic, anxiolytic, antiepileptic, or other neurologic/psychiatric medications were excluded to reduce potential bias introduced by other neuroactive drugs or conditions. All participants currently using at least one (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 7 AD HD m e d i c a t i on w e r e c l a s s i fi e d a s t h e m e d i c a t e d AD HD g r o u p (AD HD + R X ) . T hi s a p p r oa c h w a s n e c e s s a r y because KSADS does not account for whether reported symptoms reflect behavior while on or off medication, which could otherwise lead to a substantial under-classification of individuals with ADHD well-controlled by medication. From the ADHD+RX group, we identified and removed two subjects using an NS medication for Tic Disorder, rather than ADHD. Participants who met ADHD diagnostic criteria and did not report ADHD medication use were labeled as the unmedicated ADHD group (ADHD-noRX). The remaining participants were labeled the non-ADHD control (Control) group. Neuroimaging Data Scanning protocol details have been reported by Casey et al. (2018). 39 A harmonized MRI protocol was used across sites, utilizing 3T scanners manufactured by Siemens, Philips, or GE. Motion compliance training and real- time, prospective motion correction were used to minimize motion distortion. T1-weighted (T1w) images consisting of 176 slices with 1 mm 3 isotropic resolution were acquired using a magnetization-prepared rapid acquisition gradient echo (MPRAGE) sequence. 39 The ABCD Data Analysis, Informatics, and Resource Center inspected individual T1w images for poor image quality (imgincl_t1w_include) and incidence of abnormal clinical findings (mrif_score). The cortical surface of participants' T1w images was reconstructed using FreeSurfer v7.1.1 and registered to the Desikan-Killaney Atlas to quantify cortical thickness, surface area, and volume across 68 cortical brain regions of interest (ROIs) and to the Aseg Atlas to quantify volumes in 14 subcortical ROIs. 36 Demographic Data and Covariates The age, sex, and household size of the ABCD cohort closely match the distribution of 9- and 10-year-olds in the U.S. Census Bureau’s American Community Survey. The racial breakdown is similar, except for the underrepresentation of Asian, American Indian/Alaska Native, and Native Hawaiian/Pacific Islander children in the ABCD. 40 Covariates for our study included these demographic and socioeconomic variables: child’s age (in months), child’s sex at birth (male or female), race/ethnicity (Non-hispanic white, Non-hispanic Black, Hispanic/Latinx, Asian, or Other), average household income ( •$100K USD, $50K and <$100K USD, <$50K USD, or Don’t Know/Refuse to Answer), and highest household education ( Post-Graduate Degree, Bachelor Degree, Some College, High School Diploma/GED, or Less than High School Diploma ). The Other category of the combined race/ethnicity variable in ABCD includes caregiver-identified participants as American Indian/Native American, Alaska Native, Native Hawaiian, Guamanian, Samoan, other Pacific Islander, or other race. Analyses Each analysis followed a two-step approach: (1) a region-of-interest (ROI) selection phase using bootstrap- enhanced elastic net (ENet) regression to identify regions potentially sensitive to medication effects (i.e., those that differ between medicated and unmedicated ADHD groups), and (2) linear mixed-effects (LME) regression modeling applied to the selected ROIs to estimate the effects of ADHD status and the specific medication classes (i . e . , A M P , M P H, a nd NS ) . T hi s w or k fl o w i s d e p i c t e d i n Figure 2 and was applied three times to measures of cortical thickness, surface area, and volume. All analyses were conducted in R Version 4.4.1 (R Core Team, 2024) using the glmnet version 4.1-8 package 41 for ENet models and the lme4 version 1.1-35.5 package 42 for LME models, with model diagnostics assessed using the performance version 0.12.2 package. 43 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 8 Figure 2. Workflow of our two-step analysis pipeline. A) First, a Machine Learning phase is used to identify brain regions that are potentially sensitive to ADHD medications. B) Second, LME models are used to measure the effects of ADHD and medication classes on identified regions. Abbreviations: ADHD+RX = medicated ADHD group ; ADHD-noRX = unmedicated ADHD group; LME = mixed effects modeling; AMP = amphetamine; MPH = methylphenidate; NS = nonstimulant. Analysis 1: Brain Feature Selection through Machine Learning To identify potential neuroanatomic targets of ADHD medications, we applied bootstrap-enhanced elastic net (ENet) regression modeling to three sMRI metrics across 68 cortical ROIs and 14 subcortical ROI volumes to predict ADHD medication status (ADHD+RX vs. ADHD-noRX) as a logistic target. Each ML step was run separately for each sMRI metric (e.g., cortical thickness, surface area, volume). ENet regression was chosen because combining L 1 and L2 regularization can accommodate the high degree of autocorrelation characteristic of MRI data. That is, it strikes a balance between sparsifying model coefficients (i.e., performing feature selection for dimensionality reduction) and retaining coefficients for correlated features (e.g., similar brain regions). This method, proposed by Bunea et al. (2011), 44 involves repeatedly bootstrapping ENet regression and has been shown to improve MRI feature selection compared to traditional ENet regression. Analyses were conducted in R (v4.4.1) using the glmnet (v4.1-8) package, which standardizes features. Specifically, we performed 1,000 bootstraps of cross-validated ENet regression, ensuring each iteration preserved a stratified ratio of ADHD+RX to ADHD-noRX subjects in both the 80% training and 20% testing sets. The L 1:L2 ratio, or , was set at 0.5 to ensure models balanced sparsity and shrinkage, as noted above. Within each iteration,  was fine-tuned via 10-fold cross-validation across the 80% training set, selecting the regularization parameter that yielded the lowest cross-validated mean squared error. Non-zero coefficients from each model iteration indicated whether an ROI was selected. To account for model quality, we weighted each selected ROI by its model’s performance on the held-out 20% testing set, using the area under the curve (AUC Test). These weighted selection scores were then averaged across bootstrap iterations to generate a feature stability weight for each ROI. Building on the approach of Bunea et al. (2011) 44 and Abram et al. (2016) 45 in retaining ROIs selected in 50% of (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 9 bootstrap iterations, we set our feature stability threshold at 0.25—equivalent to selection in 50% of iterations weighted by a model performing at chance level (AUCTest = 0.50). Analysis 2: Linear mixed-effects (LME) modelling Selected ROIs were included in the regression modeling stage to assess the effects of our primary dependent variables: (1) ADHD status, as well as medication class effects of (2) AMP, (3) MPH, and (4) NS on the brain in these ROIs (i.e., independent variables). ADHD status and each medication class were included as binary variables since some subjects display polypharmacy (i.e., use of more than one ADHD medication class). Each regression model included fixed effects of demographic covariates and MRI scanner manufacturer, with a random effect of MRI scanner serial number. Due to the small categorical counts, we combined the Asian and Other race/ethnicity categories into a single categorical variable (Asian/Other) for regression analysis. Inclusion of MRI scanner manufacturer (Siemens, Philips, or GE) accounts for differences in scanner hardware and software in analyses. Instead of an ABCD site effect, we included a random effect that accounted for the 29 unique MRI scanners using their serial numbers, which has been demonstrated to adequately model the significant scanner- related effects observed in the ABCD Study. 46 See Supplemental Methods for model details. Additionally, we included total intracranial volume as a confounder in analyses of cortical and subcortical volumes. After removing participants with missing covariate data, we randomly selected one child per family. Importantly, there was no evidence of problematic collinearity for LME models, with Variance Inflation Factor (VIF) values of 2.01 for ADHD, 1.34 for AMP, 1.50 for MPH, and 1.17 for NS in our data. False discovery rate (FDR) correction was performed separately for each predictor (e.g., ADHD, AMP, MPH, NS) across each set of models (i.e., cortical thickness, surface area, volume) using the Benjamini-Hochberg method. 47 Effect sizes are reported as standardized beta coefficients, allowing for direct comparison across brain features. Given the limitations of relying solely on statistical significance in large population-based studies, 48,49 we first describe patterns in the beta coefficients—comparing the direction and average magnitude of effects for ADHD status with those of AMP, MPH, and NS—before reporting statistical significance.

Sample Characteristics Consistent with prior epidemiological findings in ADHD, 7 our sample included a higher prevalence of male than female youth in both the medicated (Male: 73.1%; Female: 26.9%) and unmedicated ADHD groups (Male: 66.9%; Female: 32.9%), along with fewer participants identifying as Asian. Additionally, in line with prior findings, 7 females and Hispanic/Latinx individuals within the ADHD groups were less likely to receive ADHD medication treatment. See Supplemental Table 1. Analysis 1: Brain Feature Selection through Machine Learning Cortical and Subcortical Brain Regions Relevant to ADHD Medication Status Under our feature stability weight threshold of 0.25, our ML approach identified nine regions for cortical thickness, two regions for cortical surface area, and twelve regions for cortical and subcortical volumes relevant for differentiating medicated and unmedicated ADHD participants ( Figure 3) . Regions exhibiting relevant cortical thickness differences between ADHD+RX and ADHD-noRX individuals included the right caudal anterior cingulate, right entorhinal, right inferior temporal, left isthmus cingulate, left middle temporal, left precuneus, left postcentral, left supramarginal, and right temporal pole. The average AUC Test across bootstrap iterations was 0.51 for cortical thickness. For surface area, relevant regions included the right banks of the (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al superior temporal sulcus and the left posterior cingulate; the average AUCTest was 0.54. For cortical and subcor volume, relevant regions included the right banks of the superior temporal sulcus, right cuneus, left i nfe parietal, right inferior temporal, right lingual, left pars triangularis, left posterior cingulate, left superior p ari right temporal pole, right accumbens area, left amygdala, and right putamen; the average AUC Test was 0.51. average AUCs for our three bootstrap-enhanced ENet models were all near chance, underscoring the limitat of standard ENet models in reliably identifying features. This highlights the importance of boo tst enhancement, which improves feature importance metrics (e.g., our feature stability weight) by mea su feature selection frequency across resampled datasets. Post hoc, we discovered that approximately 100 bootstr would have been sufficient for our data, as the feature stability weight leveled off around this p (Supplemental Figures 2-4). Figure 3. Brain Features Relevant to ADHD Medication Usage. Feature Stability Weights of brain regions predicting ADHD medication status (i.e., ADHD+RX vs. ADHD–NoRX) using bootstrap-like ENet regression. A feature stability weight of 0.25 was used as our threshold; a higher feature stability weight indicates a more reliable selection across bootstraps and better model performance. Abbreviations: ADHD+RX = medicated ADHD group; ADHD-noRX = unmedicated ADHD group. al. 10 ortical nferior arietal, 1. The tations tstrap- suring tstraps p oint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 11 Analysis 2: Linear mixed-effects (LME) modelling Under our data-driven ML approach, brain regions identified as potentially medication-sensitive were then used in linear mixed-effects (LME) modeling to measure the effects of (1) ADHD Status, (2) AMP use, (3) MPH use, and (4) NS use. To explore broad, non-inferential patterns of ADHD and medication effects on brain structure, we describe the direction and magnitude of standardized beta coefficients across regions before considering statistical significance. Patterns in ADHD and Medications Effects on Brain Structure All standardized beta coefficients for the primary predictors—regardless of statistical significance—are mapped across brain regions for cortical thickness, surface area, and volume in Figure 4, and for subcortical volume in Supplemental Figure 5 to illustrate the broader patterns of associations between ADHD and medication effects with brain structure outcomes. These effects are also plotted with 95% confidence intervals in Figure 5. For cortical thickness, individuals with ADHD showed smaller cortical thickness compared to non-ADHD controls in the right caudal anterior cingulate, right inferior temporal, left isthmus cingulate, left middle temporal, left postcentral, and right temporal pole ( Figure 4A and Figure 5A ). In contrast, individuals with ADHD showed greater cortical thickness compared to controls in the right entorhinal cortex, left precuneus, and left supramarginal gyrus. AMP use was related to attenuation of ADHD cortical thickness differences (i.e., opposite directionality of ADHD status effect; or in other words, improvement compared to ADHD phenotype) in 8 of the 9 ROIs, with the exception of the left precuneus. Similarly, MPH use was related to attenuation in all 9 of the ROIs. NS use was also related to attenuation in 8 of the 9 ROIs, with the exception of the right temporal pole (Figure 4A and Figure 5A ). Across the 9 cortical thickness ROIs, the average absolute values of the standardized betas were 0.013 ± 0.007 for ADHD, 0.018 ± 0.010 for AMP, 0.016 ± 0.010 for MPH, and 0.0095 for NS. For surface area, individuals with ADHD had a larger surface area of the right banks of the superior temporal sulcus and left posterior cingulate ( Figure 4B and Figure 5B ). AMP, MPH, and NS use attenuated these differences in both ROIs. For surface area ROIs, the average absolute values of the standardized beta coefficients were 0.010 ± 0.012 for ADHD, 0.046 ± 0.008 for AMP, 0.035 ± 0.005 for MPH, and 0.010 ± 0.007 for NS. For cortical and subcortical volumes, individuals with ADHD had smaller volumes in the right cuneus, right inferior temporal, right lingual, left pars triangularis, left superior parietal, right temporal pole, left amygdala, and right putamen (Figure 4C, Figure 5C-D, Supplemental Figure 5). In addition, those with ADHD had larger volumes as compared to controls in the right banks of the superior temporal sulcus, left inferior parietal, left posterior cingulate, and right accumbens area. AMP use a ttenuated ADHD effects in 8 of 12 ROIs, MPH use in 9 of 12 ROIs, and NS use in 9 of 12 ROIs. The primary exception for medication effects was the volume of the left pars triangularis, which was not attenuated by any of the three medication classes. Medications showed mixed associations with volumes in the left superior parietal cortex, right accumbens area, left amygdala, and right putamen, with at least one medication in the attenuating direction and at least one in the ADHD direction. Not considering the three subcortical regions, the medication attenuation rate for the cortical volumes was like that of cortical thickness and surface area. Across these cortical and subcortical volume ROIs, the average absolute standardized beta coefficients were 0.013 ± 0.008 for ADHD, 0.016 ± 0.011 for AMP, 0.017 ± 0.007 for MPH, and 0.008 ± 0.006 for NS. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al Figure 4. Effects of ADHD and medications mapped onto the Desikan-Killiany Atlas. Standardized b eta coefficient ADHD status, amphetamine (AMP), methylphenidate (MPH), and nonstimulant (NS) use from LME regression models ac cortical brain features selected by ML, including a) cortical thickness, b) surface area, and c) cortical volume. Red in dic positive beta coefficients (i.e., higher values in the ADHD or medication- exposed groups), and blue indicates nega coefficients. Cortical regions not selected during the ML stage (and therefore not included in LME modeling) are colore grey. Abbreviations: AMP = amphetamine; MPH = methylphenidate; NS = nonstimulant. Significant Stimulant Effects in Cortical Regions After false discovery rate (FDR) correction, significant effects were observed only for AMP and MPH us cortical regions (denoted with a star in Figure 5). Both stimulants were associated with reduced surface are the right banks of the superior temporal sulcus and the left posterior cingulate ( Figure 5B). Additionally, A use was linked to increased cortical thick ness in the right entorhinal cortex and decreased volume in t he posterior cingulate following FDR correction ( Figure 5A and 5C). No effects of ADHD status or NS use re ac significance after FDR correction. Additionally, no significant effects on sub cortical volumes were observed a FDR correction. al. 12 nts for across dicates egative ored in use in rea in , AMP he le ft ached d after (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al Figure 5. Forest plots of the effects of ADHD and medications. Standardized beta coefficient with 95% confidence inter plotted for ADHD status, amphetamine (AMP), methylphenidate (MPH), and nonstimulant (NS) use from LME regres models for ML-selected cortical a) thickness, b) surface area, c) volume, and d) subcortical volume brain features. E ffects reached statistical significance after FDR correction are denoted with a star. Abbreviations: AMP = amphetamine; MP methylphenidate; NS = nonstimulant.

This study is among the first to investigate brain structure differences associated with ADHD medication use large, demographically diverse sample. By integrating machine learning to identify medication- sensitive reg with linear mixed-effects regression modeling, we demonstrate opposing effects of ADHD status and medica use in parietal, occipital, and temporal regions. Our analytic approach is further strengthened by a theore tic grounded framework that allows us to disentangle the distinct effects of different ADHD medication cla w h i l e a c c o u n t i n g f o r t h e c o m m o n p o l y p h a r m a c y ( e . g . , A M P + N S o r M P H + N S ) o b s e r v e d i n r e a l -w populations. Medications Appear to Attenuate ADHD Brain Differences We highlight the utility of our conceptual approach, which utilizes machine learning (ML) to identify po ten medication-sensitive regions in youth with ADHD, within an explainable artificial intelligence (xAI) fram ew We then further investigated how these subjects differ from those without ADHD (controls) and the eff ect specific medication classes on morphology. The resulting attenuation patterns are strengthened by the fa ct al. 13 tervals ression cts that PH = se in a egions ication tically classes world tential ework. ects of ct that (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 14 the ML phase was conducted without data from controls, and the selected brain features were identified solely for their ability to distinguish medicated from unmedicated individuals with ADHD. Notably, the morphology of many medication-sensitive brain regions appeared to be normalized (i.e., more like that of controls) by medication use, highlighting the utility of our combined data- and theory-driven approach. The cortical regions identified as likely medication targets were primarily located within the parietal, occipital, and temporal lobes, with a relative paucity of involvement in the frontal lobe, despite long-standing hypotheses about the role of prefrontal regions in ADHD pathology. 25,26 One possibility is that prefrontal regions continue to exhibit ADHD- related structural differences but are not the primary targets of pharmacological intervention, that these differences in prefrontal regions emerge later during adolescence, 50 or that their extended developmental trajectory into early adulthood 51 may render them less susceptible to medication effects during the period of preadolescence studied here. In our study of preadolescents, we found no statistically significant effects attributable to ADHD across the cortical and subcortical features examined when accounting for medication use, consistent with prior meta- analytic findings. 24 Rather than relying solely on significance, we also examined the magnitude and direction of brain morphology associated with ADHD. In doing so, we observed both larger and smaller cortical brain region f e a t u r e s r e l a t i v e t o c o n t r o l s , h i g h l i g h t i n g a l a c k o f g l o b a l e f f e c t s o f A D H D a t t h i s a g e . F u r t h e r m o r e , w h e n focusing on the patterns regarding the direction of effects, our findings align with prior research suggesting that ADHD medications may help “normalize” cortical structure. 14 Moreover, AMP and MPH exhibited parallel patterns across cortical regions, with the direction of the effects aligning with potential attenuation of ADHD- related structural differences when compared to cont rols. For instance, although participants with ADHD showed reduced cortical thickness in several regions of the temporal lobe and the cingulate cortex when compared to control youth, stimulant usage was associated with relative increases in cortical thickness in these regions. NSs showed a similar pattern of attenuation, but with smaller beta coefficients compared to stimulants, suggesting weaker effects. However, this attenuation pattern for ADHD medications was not observed in subcortical regions, where medication effects appeared more variable. Prior ABCD Study research by Wu et al. (2024) 52 identified differences in right insula thickness and left nucleus accumbens volume between stimulant-treated individuals with low ADHD symptoms (N = 273) and unmedicated individuals with ADHD. However, these regions were not identified as potential primary medication targets in our study, potentially due to differences in analytic approach or the broader scope of our investigation—such as disaggregating stimulant classes, including a larger number of stimulant users, and examining both stimulant and nonstimulant medication effects. Additionally, in a small ABCD Study subsample, Kaminski et al. (2024) 53 reported that stimulant use was associated with altered functional connectivity between striatal regions and several large-scale brain networks, including the cingulo-opercular, salience, fronto-parietal, cingulo-parietal, somatomotor, dorsal attention, ventral attention, and visual networks. While our ML approach identified two potential striatal targets of ADHD medications (the right putamen and right accumbens area), we found no significant effects of ADHD or medication use in these regions after FDR correction. Moreover, we observed no consistent patterns of ADHD attenuation between medication classes in subcortical regions. Despite the basal ganglia’s frequent implication in ADHD pathogenesis, 25,26 our findings suggest that these medications may not alter striatal region structure during preadole scence, or that such effects may not lead to such morphological differences as assessed here. Stimulants May Overcompensate Across medication-sensitive brain regions, the average absolute effects of AMP and MPH on cortical thickness, surface area, cortical volume, and subcortical volume were larger than those associated with ADHD. Interestingly, several cortical regions—such as the right entorhinal cortex, the right banks of the superior (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 15 temporal sulcus, and the left posterior cingulate—showed statistically significant effects of AMP and/or MPH. In the entorhinal cortex and banks of the superior temporal sulcus, these effects were in the direction of attenuation, suggesting a potential normalization of ADHD-related structural differences. However, in these regions, the magnitude of stimulant-related effects exceeded those associated with ADHD diagnosis, raising the possibility of neurobiological overcompensation. NS medications exhibited similar directional effects to stimulants but with smaller magnitudes, consistent with their distinct pharmacological profile—namely, their primary action on noradrenergic pathways and downstream engagement of dopaminergic pathways. 54 This also aligns with clinical evidence indicating that NSs generally yield more modest improvements in ADHD symptoms. 5 Bernanke et al. (2022), 55 using data from the same wave of the ABCD Study, identified reduced posterior cingulate surface area as the largest ADHD-related structural difference, but they did not account for ADHD medication use. In contrast, our study found significant reductions in left posterior cingulate surface area associated with both amphetamine (AMP) and methylphenidate (MPH) use, while the effect of ADHD in this region was near zero. This suggests that stimulants may relate to de novo alterations in the posterior cingulate, raising the possibility that previously reported differences may reflect medication effects rather than intrinsic ADHD-related pathology. Additionally, we observed that AMP use was associated with reduced posterior cingulate volume, implying that volumetric normalization may be primarily driven by surface area reduction rather than cortical thinning. Given that total surface area peaks during preadolescence, this morphometric feature may represent a particularly sensitive target for pharmacological effects during this developmental period. 51 AMP exhibited more statistically significant effects, including unique associations with entorhinal cortical thickness and posterior cingulate volume, but not significant with MPH. This may reflect pharmacodynamic differences, as some literature suggests AMP leads to greater cytosolic dopamine accumulation than MPH. 56 Because the spatial distribution of AMP-only effects (e.g., entorhinal cortex thickness, posterior cingulate volume) aligns with regions of higher dopamine transporter (DAT) density, these observed stimulant effects may be driven by dopaminergic mechanisms. 31 In contrast, stimulant effects observed in the superior temporal sulcus may be driven more by noradrenergic mechanisms, given its higher relative expression of norepinephrine transporter (NET). 31 Altogether, these findings emphasize the need to account for medications as key confounding variables in future ADHD neuroimaging research. Our flexible LME modeling approach enabled us to isolate the unique effects of ADHD, AMP, MPH, and NS medications while handling polypharmacy. Unlike conventional approaches (e.g., grouping participants into mutually exclusive medication categories, aggregating stimulants, or excluding individuals with ADHD polypharmacy), our framework accommodates overlapping medication use. This approach reduces collinearity, preserves sample size, and more accurately reflects real-world treatment patterns, thereby enhancing interpretability and ecological validity. However, in theory, the observed medication-related effects on brain structure could also reflect unmeasured confounders related to medication treatment. For example, these findings could stem from who initiates treatment (e.g., individuals with brain features linked to greater severity or specific symptom profiles) or who maintains treatment (e.g., those with features associated with stimulant responsivity). Future longitudinal research is essential to determine whether these brain structure differences precede or follow the initiation of stimulant use. Strengths and Limitations Although it lacks the rigor of a randomized controlled trial, the large and demographically diverse ABCD Study cohort offers an unparalleled opportunity to investigate pharmacologic effects under naturalistic conditions, including imperfect medication adherence. Additionally, our xAI framework used a data-driven feature selection approach to identify potential neuroanatomical targets of medications. This focus of our xAI framework may (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 16 limit the identification of regions relevant to ADHD that are relatively independent of medication effects. Finally, to our knowledge, this is the first structural MRI study to model major ADHD medication classes concurrently by disaggregating stimulant classes and including nonstimulant medication effects. This has enabled us to identify both common and distinct neuroanatomical effects of these medications, representing a critical advancement in precision psychiatric medicine. A key limitation of our analysis is its focus on a single developmental time point (ages 9–10 years) within a longitudinal study. ADHD has been hypothesized to reflect brain developmental delays, with phenotypic differences becoming more pronounced over the course of adolescence. Future longitudinal work is needed to assess how ADHD- and medication-related neurobiological differences evolve or resolve across later developmental stages. While our classification approach captures all individuals prescribed ADHD medication, it remains possible that some cases reflect overdiagnosis within the community. The ABCD Study does not survey medication history prior to the study period; therefore, the lifetime exposure to ADHD medications is unknown. We did not examine the potential moderating effects of medication dosage or disease severity, which are priorities for future research. Finally, due to the relatively small NS subsample size, we analyzed NS effects as a single categorical class, rather than disaggregating alpha-agonists and norepinephrine reuptake inhibitors, which are pharmacologically distinct.

O u r s t u d y e m p l o y e d a n i n n o v a t i v e a p p r o a c h t o m a p p i n g the effects of ADHD medications on preadolescent cortical and subcortical brain development, identifying novel medication-related associations. We contribute several key findings to the field: (1) AMP and MPH appear to broadly attenuate ADHD-related effects in the cortex, and (2) NS exhibit similar but weaker effects to those of stimulants. Additional longitudinal research is needed to investigate whether stimulants may be associated with overcompensatory effects in specific regions of the cortex, particularly in the posterior cingulate, banks of the superior temporal sulcus, and entorhinal cortex, which may be due to their roles in dopaminergic and noradrenergic pathways. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 17 Funding The research described in this article was supported by the National Institutes of Health [NIDA U01DA041048]. Acknowledgments A special thanks to the participants and families of the ABCD Study. We would like to acknowledge the ABCD Consortium staff for their efforts in collecting data. We would also like to acknowledge Alethea de Jesus and Yara Akiel, who contributed to the data cleaning and formatting of the ABCD tabulated data prior to data analysis. Competing Interests The authors declare no competing interests. Author Contributions Conceptualization: LNO, KLB, SLK, BSP, MMH. Data Curation: LNO. Formal Analysis: LNO. Funding Acquisition: MMH. Investigation: LNO. Methodology: LNO, KLB. Supervision: MMH. Visualization: LNO. Writing – Original Draft Preparation: LNO, MMH. Writing – Review & Editing: LNO, KLB, SLK, BSP, MMH. Data Availability Statement Data used in the preparation of this article were obtained from the Adolescent Brain Cognitive Development (ABCD) Study (https://abcdstudy.org), held in the NIMH Data Archive (NDA). This is a multisite, longitudinal study designed to recruit more than 10,000 children aged 9–10 and follow them over 10 years into early adulthood. The ABCD Study is supported by the National Institutes of Health Grants [U01DA041022, U01DA041028, U01DA041048, U01DA041089, U01DA041106, U01DA041117, U01DA041120, U01DA041134, U01DA041148, U01DA041156, U01DA041174, U24DA041123, U24DA041147]. A complete list of supporters is available at https://abcdstudy.org/nih-collaborators. A list of participating sites and a comprehensive list of study investigators can be found at https://abcdstudy.org/principal-investigators.html. ABCD consortium investigators designed and implemented the study and/or provided data, but did not necessarily participate in the analysis or writing of this report. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH or ABCD consortium investigators. The ABCD data repository grows and changes over time. Qualified researchers can request access to ABCD shared data from the ABCD Data Access Committee (DAC). (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 18

1. Erskine, H. E. et al. Research Review: Epidemiological modelling of attention-deficit/hyperactivity disorder and conduct disorder for the Global Burden of Disease Study 2010. Journal of Child Psychology and Psychiatry 54, 1263–1274 (2013). 2. Polanczyk, G. V., Salum, G. A., Sugaya, L. S., Caye, A. & Rohde, L. A. Annual research review: A meta- analysis of the worldwide prevalence of mental disorders in children and adolescents. J Child Psychol Psychiatry 56, 345–365 (2015). 3. Thomas, R., Sanders, S., Doust, J., Beller, E. & Glasziou, P. Prevalence of Attention-Deficit/Hyperactivity Disorder: A Systematic Review and Meta-analysis. Pediatrics 135, e994–e1001 (2015). 4. Nigg, J. T., Sibley, M. H., Thapar, A. & Karalunas, S. L. Development of ADHD: Etiology, Heterogeneity, and Early Life Course. Annual Review of Developmental Psychology 2, 559–583 (2020). 5. Peterson, B. S. et al. Treatments for ADHD in Children and Adolescents: A Systematic Review. Pediatrics e2024065787 (2024) doi:10.1542/peds.2024-065787. 6. Boland, H. et al. A literature review and meta-analysis on the effects of ADHD medications on functional outcomes. Journal of Psychiatric Research 123, 21–30 (2020). 7. Danielson, M. L. et al. ADHD Prevalence Among U.S. Children and Adolescents in 2022: Diagnosis, Severity, Co-Occurring Disorders, and Treatment. J Clin Child Adolesc Psychol 53, 343–360 (2024). 8. Bitsko, R. H. Mental Health Surveillance Among Children — United States, 2013–2019. MMWR Suppl 71, (2022). 9. Santosh, P. J. & Taylor, E. Stimulant drugs. Eur Child Adolesc Psychiatry 9 Suppl 1, I27-43 (2000). 10. Hodgkins, P., Shaw, M., Coghill, D. & Hechtman, L. Amfetamine and methylphenidate medications for attention-deficit/hyperactivity disorder: complementary treatment options. Eur Child Adolesc Psychiatry 21, 477–492 (2012). (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 19 11. Arnold, L. E. Methyiphenidate vs. amphetamine: Comparative review. J Atten Disord 3, 200–211 (2000). 12. Buitelaar, J. et al. Toward Precision Medicine in ADHD. Front Behav Neurosci 16, 900981 (2022). 13. Faraone, S. V. The pharmacology of amphetamine and methylphenidate: Relevance to the neurobiology of attention-deficit/hyperactivity disorder and other psychiatric comorbidities. Neurosci Biobehav Rev 87, 255– 270 (2018). 14. Spencer, T. J. et al. Effect of Psychostimulants on Brain Structure and Function in ADHD: A Qualitative Literature Review of MRI-Based Neuroimaging Studies. J Clin Psychiatry 74, 902–917 (2013). 15. Shaw, P. et al. Psychostimulant treatment and the developing cortex in attention deficit hyperactivity disorder. Am J Psychiatry 166, 58–63 (2009). 16. Pliszka, S. R., Lancaster, J., Liotti, M. & Semrud-Clikeman, M. Volumetric MRI differences in treatment-naïve vs chronically treated children with ADHD. Neurology 67, 1023–1027 (2006). 17. Bledsoe, J., Semrud-Clikeman, M. & Pliszka, S. R. An MRI Study of the Cerebellar Vermis in Chronically- Treated and Treatment-Naïve Children with ADHD-Combined Type. Biol Psychiatry 65, 620–624 (2009). 18. Sobel, L. J. et al. Basal ganglia surface morphology and the effects of stimulant medications in youth with attention deficit hyperactivity disorder. Am J Psychiatry 167, 977–986 (2010). 19. Ivanov, I., Murrough, J. W., Bansal, R., Hao, X. & Peterson, B. S. Cerebellar Morphology and the Effects of Stimulant Medications in Youths with Attention Deficit-Hyperactivity Disorder. Neuropsychopharmacol 39, 718–726 (2014). 20. Schnoebelen, S., Semrud-Clikeman, M. & Pliszka, S. R. Corpus callosum anatomy in chronically treated and stimulant naïve ADHD. J Atten Disord 14, 256–266 (2010). 21. Castellanos, F. X. et al. Developmental trajectories of brain volume abnormalities in children and adolescents with attention-deficit/hyperactivity disorder. JAMA 288, 1740–1748 (2002). (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 20 22. Nakao, T., Radua, J., Rubia, K. & Mataix-Cols, D. Gray Matter Volume Abnormalities in ADHD: Voxel-Based Meta-Analysis Exploring the Effects of Age and Stimulant Medication. AJP 168, 1154–1163 (2011). 23. Frodl, T. & Skokauskas, N. Meta-analysis of struct ural MRI studies in children and adults with attention deficit hyperactivity disorder indicates treatment effects. Acta Psychiatrica Scandinavica 125, 114–126 (2012). 24. Samea, F. et al. Brain alterations in children/adolescents with ADHD revisited: A neuroimaging meta- analysis of 96 structural and functional studies. Neurosci Biobehav Rev 100, 1–8 (2019). 25. Albajara Sáenz, A., Villemonteix, T. & Massat, I. Structural and functional neuroimaging in attention- deficit/hyperactivity disorder. Developmental Medicine & Child Neurology 61, 399–405 (2019). 26. Firouzabadi, F. D. et al. Neuroimaging in Attention-Deficit/Hyperactivity Disorder: Recent Advances. American Journal of Roentgenology 218, 321–332 (2022). 27. da Silva, B. S. et al. An overview on neurobiology and therapeutics of attention-deficit/hyperactivity disorder. Discov Ment Health 3, 2 (2023). 28. Jiang, X., Liu, L., Ji, H. & Zhu, Y. Association of Affected Neurocircuitry With Deficit of Response Inhibition and Delayed Gratification in Attention Deficit Hyperactivity Disorder: A Narrative Review. Front. Hum. Neurosci. 12, (2018). 29. Posner, J., Park, C. & Wang, Z. Connecting the Dots : A Review of Resting Connectivity MRI Studies in Attention-Deficit/Hyperactivity Disorder. Neuropsychol Rev 24, 3–15 (2014). 30. Faraone, S. V. et al. Attention-deficit/hyperactivity disorder. Nat Rev Dis Primers 1, 15020 (2015). 31. Hansen, J. Y. et al. Mapping neurotransmitter systems to the structural and functional organization of the human neocortex. Nat Neurosci 25, 1569–1581 (2022). 32. Jernigan, T. L. & Brown, S. A. Introduction. Developmental Cognitive Neuroscience 32, 1–3 (2018). 33. Garavan, H. et al. Recruiting the ABCD sample: Design considerations and procedures. Dev Cogn Neurosci 32, 16–22 (2018). (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 21 34. Volkow, N. D. et al. The conception of the ABCD study: From substance use to a broad NIH collaboration. Dev Cogn Neurosci 32, 4–7 (2018). 35. Palmer, C. E. et al. Microstructural development from 9 to 14 years: Evidence from the ABCD Study. Dev Cogn Neurosci 53, 101044 (2022). 36. Hagler, D. J. et al. Image processing and analysis methods for the Adolescent Brain Cognitive Development Study. Neuroimage 202, 116091 (2019). 37. Li, Y. et al. Rates of Incidental Findings in Brain Magnetic Resonance Imaging in Children. JAMA Neurol 78, 578 (2021). 38. Barch, D. M. et al. Demographic, physical and mental health assessments in the adolescent brain and cognitive development study: Rationale and description. Developmental Cognitive Neuroscience 32, 55–66 (2018). 39. Casey, B. J. et al. The Adolescent Brain Cognitive Development (ABCD) study: Imaging acquisition across 21 sites. Dev Cogn Neurosci 32, 43–54 (2018). 40. Heeringa, S. G. & Berglund, P. A. A Guide for Population-based Analysis of the Adolescent Brain Cognitive Development (ABCD) Study Baseline Data. Preprint at https://doi.org/10.1101/2020.02.10.942011 (2020). 41. Friedman, J. et al. glmnet: Lasso and Elastic-Net Regularized Generalized Linear Models. 4.1-8 https://doi.org/10.32614/CRAN.package.glmnet (2008). 42. Bates, D., Mächler, M., Bolker, B. & Walker, S. Fitting Linear Mixed-Effects Models using lme4. Preprint at https://doi.org/10.48550/arXiv.1406.5823 (2014). 43. Lüdecke, D., Ben-Shachar, M., Patil, I., Waggoner, P. & Makowski, D. performance: An R Package for Assessment, Comparison and Testing of Statistical Models. JOSS 6, 3139 (2021). 44. Bunea, F. et al. Penalized least squares regression methods and applications to neuroimaging. NeuroImage 55, 1519–1527 (2011). (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 22 45. Abram, S. V. et al. Bootstrap Enhanced Penalized Regression for Variable Selection with Neuroimaging Data. Front. Neurosci. 10, (2016). 46. Dudley, J. A. et al. ABCD_Harmonizer: An Open-source Tool for Mapping and Controlling for Scanner Induced Variance in the Adolescent Brain Cognitive Development Study. Neuroinformatics 21, 323–337 (2023). 47. Benjamini, Y. & Hochberg, Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society: Series B (Methodological) 57, 289–300 (1995). 48. Aarts, S., Van Den Akker, M. & Winkens, B. The importance of effect sizes. European Journal of General Practice 20, 61–64 (2014). 49. Carey, E. G., Ridler, I., Ford, T. J. & Stringaris, A. Editorial Perspective: When is a ‘small effect’ actually large and impactful? Journal of Child Psychology and Psychiatry 64, 1643–1647 (2023). 50. Halperin, J. M. & Schulz, K. P. Revisiting the role of the prefrontal cortex in the pathophysiology of attention- deficit/hyperactivity disorder. Psychological Bulletin 132, 560–581 (2006). 51. Bethlehem, R. a. I. et al. Brain charts for the human lifespan. Nature 604, 525–533 (2022). 52. Wu, F. et al. Stimulant medications in children with ADHD normalize the structure of brain regions associated with attention and reward. Neuropsychopharmacol. (2024) doi:10.1038/s41386-024-01831-4. 53. Kaminski, A., Xie, H., Hawkins, B. & Vaidya, C. J. Change in striatal functional connectivity networks across 2 years due to stimulant exposure in childhood ADHD: results from the ABCD sample. Transl Psychiatry 14, 1–11 (2024). 54. Wilens, T. E. Mechanism of Action of Agents Used in Attention-Deficit/Hyperactivity Disorder. (2006). 55. Bernanke, J. et al. Structural brain measures among children with and without ADHD in the Adolescent Brain and Cognitive Development Study cohort: a cross-sectional US population-based study. The Lancet Psychiatry 9, 222–231 (2022). (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint Overholtzer et al. 23 56. Schiffer, W. k. et al. Therapeutic doses of amphetamine or methylphenidate differentially increase synaptic and extracellular dopamine. Synapse 59, 243–251 (2006). (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 14, 2025. ; https://doi.org/10.1101/2025.11.13.687018doi: bioRxiv preprint