ARTICLES Neural Correlates of the Dual-Pathway Model for ADHD in Adolescents ent psycho nattention.Compared with healthy contol subjects.neve and Methods:Inalongitudinalcommunity-basedcc ndeper The paselin was as. symptom domain. AmJP3chGy2020177844-854dat101176/aoplaio.2020.19020183 Attention deficit hyperactivity disorder (ADHD)is immediate 50%66%of cases persisting into adulthood (2.3).This dis- with ADHD in neuroimaging studies (12),one hypothesis order has been characterized by its significant heterogeneity stinct d ns (There into the fronto-dor striatal circut identification of the neural abnormalities underlying these for motivational dysfunction(5).Previous be heterogeneous impairments may improve both the diagnostic havioral studies have reported that children with ADHDhave accuracy and motivational deficits(13,14).both of whic ependen has leading to the symptoms of ADHD(5-7).including cogitive motivational deficits are independent from(119)orasso dysfunctions,such as deficits in working memory (8),at- ciated witheachother(20,21).Recent studiesseem tosuggest a ru n(stop-signal rea ample Am J Psychiatry 177:9.September 2020
Neural Correlates of the Dual-Pathway Model for ADHD in Adolescents Chun Shen, M.Sc., Qiang Luo, Ph.D., Tianye Jia, Ph.D., Qi Zhao, M.Sc., Sylvane Desrivières, Ph.D., Erin Burke Quinlan, Ph.D., Tobias Banaschewski, M.D., Ph.D., Sabina Millenet, D.Psych., Arun L.W. Bokde, Ph.D., Christian Büchel, M.D., Herta Flor, Ph.D., Vincent Frouin, Ph.D., Hugh Garavan, Ph.D., Penny Gowland, Ph.D., Andreas Heinz, M.D., Ph.D., Bernd Ittermann, Ph.D., Jean-Luc Martinot, M.D., Ph.D., Eric Artiges, M.D., Ph.D., Marie-Laure Paillère-Martinot, M.D., Ph.D., Frauke Nees, Ph.D., Dimitri Papadopoulos Orfanos, Ph.D., Tomás Paus, M.D., Ph.D., Luise Poustka, M.D., Juliane H. Fröhner, M.Sc., Michael N. Smolka, M.D., Henrik Walter, M.D., Ph.D., Robert Whelan, Ph.D., Fei Li, M.D., Ph.D., Jianfeng Feng, Ph.D., Gunter Schumann, M.D., Barbara J. Sahakian, Ph.D., D.Sc., for the IMAGEN consortium Objective: The dual-pathway model has been proposed to explain the heterogeneity in symptoms of attention deficit hyperactivity disorder (ADHD) by two independent psychological pathways based on distinct brain circuits. The authors sought to test whether the hypothesized cognitive and motivational pathways have separable neural correlates. Methods: In a longitudinal community-based cohort of 1,963 adolescents, the neuroanatomical correlates of ADHD were identified by a voxel-wise association analysis and then validated using an independent clinical sample (99 nevermedicated patients with ADHD, 56 medicated patients with ADHD, and 267 healthy control subjects). The cognitive and motivational pathways were assessed by neuropsychological tests of working memory, intrasubject variability, stop-signal reaction time, and delay discounting. The associations were tested between the identified neuroanatomical correlates and both ADHD symptoms 2 years later and the polygenic risk score for ADHD. Results: Gray matter volumes of both a prefrontal cluster and a posterior occipital cluster were negatively associated with inattention. Compared with healthy control subjects, nevermedicated patients, but not medicated patients, had significantly lower gray matter volumes in these two clusters. Working memory and intrasubject variability were associated with the posterior occipital cluster, and delay discounting was independently associated with both clusters. The baseline gray matter volume of the posterior occipital cluster predicted the inattention symptoms in a 2-year follow-up and was associated with the genetic risk for ADHD. Conclusions: The dual-pathway model has both shared and separable neuroanatomical correlates, and the shared correlate in the occipital cortex has the potential to serve as an imaging trait marker of ADHD, especially the inattention symptom domain. Am J Psychiatry 2020; 177:844–854; doi: 10.1176/appi.ajp.2020.19020183 Attention deficit hyperactivity disorder (ADHD) is one of the most common neurodevelopmental disorders, affecting 5.9%–7.1% children and adolescents worldwide (1), with 50%–66% of cases persisting into adulthood (2, 3). This disorder has been characterized by its significant heterogeneity, as patients receiving the diagnosis often present neuropsychological impairments in distinct domains (4). Therefore, identification of the neural abnormalities underlying these heterogeneous impairments may improve both the diagnostic accuracy and the treatment efficiency of this disorder. To account for such heterogeneity, a dual-pathway model has suggested two separable pathophysiological pathways leading to the symptoms of ADHD (5–7), including cognitive dysfunctions, such as deficits in working memory (8), attention regulation (intrasubject variability) (9), and response inhibition (stop-signal reaction time) (10), and motivational dysfunction, such as preferring small immediate rewards over larger delayed rewards (delay discounting) (11). As frontostriatal dysfunction has frequently been associated with ADHD in neuroimaging studies (12), one hypothesis has been proposed that these two pathways can be dissociated into the fronto-dorsal striatal circuit, responsible for cognitive dysfunction, and the fronto-ventral striatal circuit, responsible for motivational dysfunction (5). Previous behavioral studies have reported that children with ADHD have cognitive and motivational deficits (13, 14), both of which independently contribute to ADHD symptoms (15–17). However, it is still under debate whether the cognitive and motivational deficits are independent from (18, 19) or associated with each other (20, 21). Recent studies seem to suggest a functional overlap between these two pathways; for example, working memory training could also improve delay 844 ajp.psychiatryonline.org Am J Psychiatry 177:9, September 2020 ARTICLES
SHEN ET AL discounting (22.23).In neuroi ralidated a t tool for mental health problems in children and adolescents (30)and has been demonstrated discussed (24):for example.a 2016 meta-analysis (25)re in IMAGEN to be a promising assessment for ADHD 山四少 symptoms(31-34).The hyperactivity-inattention subscale com for ve items covering three key symptom Our main goal in the pr sent study then wastotest whether (Cronbach's alpha=0.75)is at an acceptable level (35) these two pathways are linked to ADHD symptoms by shared The ADHD total score was the total score of all five items; tomical co elate Given tha score was calculated using twe nd activity-i other three items restless.”fidgetv.”and "reflective" As used in nationwide epidemiological studies (36),a hree nd class ication was established fo the sDQu medication were 、This is the cu >6,109% first study to our knowledge to assess the independent as parent-report SDQ because it is more reliable than the sociations of the identihed neuroanatomical correlates with hild self-report version,and the parent-report SDQ also gnitive d working ass op-sgin nd 5 fo D)(30.36 mal s s at hoth a 14 and findings to be an intermediate phenotype of ADHD(27).we 6 were classified into the persistent adHd oup and those with normal scores at both ages were classified into elates ted to explair the typically developed control group were Delay discounting.the monetary choice ou aire (37 an efficient and reliable measurement of delay discounting METHODS in ains 27c e items pittin Participant “g3 logiudinalatudy2ofaoescemtbrhndeaopmtape sed cohor e levels of large).Higher k coefficients in a hyperbolic discounting ewhere (28).W con sent wa reward an high participants (9524%of them male)who had completed eat age our analyses. magingdata were available were included in the analysis (Ta able 1) Working memory.Spatial working memory,as as by th Clinical cohort.AdHd-200is a multice er clinical study (29 approved by the local research ethics review boards at each task to measure participants'ability to preserve spatial in- center.A total of 233 patients nith ADHD and 267 typically formation(40)is widely used in studies of ADHD in children ope cts41[53% (he number of available were included in the analysis (see 1and Table S1in theonlinesupplement).Ofthe ADHDpatients, Intrasubject variability and stop-signal reaction time.Intra- 129 had the combined subtyp subject variability and eight the hyperacti e subtype;56 v k42 tiona missing for78 patients A full-scale IO was an in stimated hy the standard deviation of reaction time in clusion criterion (see Table S2 in the online supplement). successful go trials.Stop-signal reaction time was estimated naire (SDO ytgoresponse time.Partieipants nal reasth .September 2020
discounting (22, 23). In neuroimaging studies, a large-scale brain system beyond the frontostriatal model has also been discussed (24); for example, a 2016 meta-analysis (25) reported structural abnormalities in ADHD patients in both the right basal ganglia/insula and prefrontal cortex as well as in the left occipital lobe. Ourmaingoalin the present study, then,was to testwhether these two pathways are linked to ADHD symptoms by shared and/or separable neuroanatomical correlates. Given that ADHD has been considered an extreme of a quantitative trait (26),wefirstanalyzedalarge-scale population-based sample to identify its neuroanatomical correlates, and then validated the findings using an independent clinical sample. With both medicated and never-medicated patients with ADHD in this clinical sample, we were also able to assess the effects of medication on these neuroanatomical correlates. This is the first study, to our knowledge, to assess the independent associations of the identified neuroanatomical correlates with both cognitive deficits (i.e., working memory, intrasubject variability, stop-signal reaction time) and motivational deficits (i.e., delay discounting). To demonstrate the potential of our findings to be an intermediate phenotype of ADHD (27), we further tested whether the identified neuroanatomical correlates contributed to explaining ADHD symptoms 2 years later and whether these correlates were associated with genetic risks for the disorder. METHODS Participants Population-based cohort. IMAGEN is a community-based longitudinal study of adolescent brain development. Details on the recruitment procedure have been published elsewhere (28).Written informed consent was obtained from all participants and their legal guardians. A total of 1,963 participants (952 [49%] of them male) who had completed psychometric assessments and for whom baseline (i.e., at age 14) quality-controlled neuroimaging data were available were included in the analysis (Table 1). Clinical cohort.ADHD-200 is a multicenter clinical study (29) approved by the local research ethics review boards at each center. A total of 233 patients with ADHD and 267 typically developed control subjects (141 [53%] of themmale;mean age, 11.98 years [SD=3.04]) for whom quality-controlled MRI data were available were included in the analysis (see eMethods 1 andTable S1in the online supplement). Of theADHD patients, 129 had the combined subtype, 96 the inattentive subtype, and eight the hyperactive/impulsive subtype; 56 were medicated, 99 were never medicated, and medication information was missing for 78 patients. A full-scale IQ score .80 was an inclusion criterion (see Table S2 in the online supplement). Measurements ADHD. The Strengths and Difficulties Questionnaire (SDQ), administered at both baseline and follow-up in IMAGEN, is a validated assessment tool for mental health problems in children and adolescents (30) and has been demonstrated in IMAGEN to be a promising assessment for ADHD symptoms (31–34). The hyperactivity-inattention subscale is composed of five items covering three key symptom domains for ADHD; the subscale’s internal consistency (Cronbach’s alpha=0.75) is at an acceptable level (35). The ADHD total score was the total score of all five items; the inattention score was calculated using two items (“poor concentration” and “good attention”), and the hyperactivity-impulsivity score was estimated using the other three items (“restless,” “fidgety,” and “reflective”). As used in nationwide epidemiological studies (36), a three-band classification was established for the SDQ using a cut-off score of 6 (normal: scores ,6, 80%; borderline: score of 6, 10%; abnormal: scores .6, 10%). We used the parent-report SDQ because it is more reliable than the child self-report version, and the parent-report SDQ also has a stronger association with clinical assessments (reported odds ratios of 32.3 and 5 for ADHD) (30, 36). Participants who had abnormal scores at both ages 14 and 16 were classified into the persistent ADHD group, and those with normal scores at both ages were classified into the typically developed control group. Delay discounting. The Monetary Choice Questionnaire (37), an efficient and reliable measurement of delay discounting that has been validated in adolescents (38), was administered at baseline. It contains 27 dichotomous-choice items pitting a smaller immediate reward against a larger delayed reward for three levels of reward magnitude (small, medium, and large). Higher k coefficients in a hyperbolic discounting equation for each rewardlevel represent greater preference for small immediate rewards and higher impulsivity (see eMethods 2 in the online supplement). The geometric mean was calculated and logarithmically transformed to use in our analyses. Working memory. Spatial working memory, as assessed by the Cambridge Neuropsychological Testing Automated Battery (39), was measured at baseline. This self-ordered searching task to measure participants’ ability to preserve spatial information (40) is widely used in studies of ADHD in children and adolescents (41). The number of errors was used as an index of working memory. Intrasubject variability and stop-signal reaction time. Intrasubject variability and stop-signal reaction time were obtained by behavioral data for the stop-signal functional MRI (fMRI) task (42) (N=1,846). Intrasubject variability was estimated by the standard deviation of reaction time in successful go trials. Stop-signal reaction time was estimated by subtracting the mean stop-signal latency from the mean correct go response time. Participants who had less than 50% correct hits and who had negative stop-signal reaction time were excluded. Am J Psychiatry 177:9, September 2020 ajp.psychiatryonline.org 845 SHEN ET AL
NEURAL CORRELATES OF THE DUAL-PATHWAY MODEL FOR ADHD IN ADOLESCENTS TABLE1.Characteristics of the study population in the IMAGEN cohort FWE d p th Characteristic or Measure Baseline (N=1.963) 48.5 728 480% Neuropsychological assc analysis.Sep- Mean ere con SD Mean SD Age (years 14.43 0.40 16.47 0.57 (working memory.intrasubject variability stop-signal reaction time,and delay dis- y-mpycre counting)and both ADHDsymptoms and gray Inattention score 192 lling tes ndedn N N terval was given by 5,000 bootstraps.Next,we ADHD included other variables as covar aesfor the aon ana ne varia a sig 941 fter controlling for other variables this as Abnorma 8.5 4.1 sociation is not independent ofother variables Mean SD out is contributed by some common facto between cognitive and motivational Stop-signal reaction time 186.43 61.90 Prospective association analysis.We extended GEN par ider significant associations between the baseline Structural MRI features and ADHD svmptoms 2 years later.In these re- acqui ition and and corresponding baselin Tweighted maiio-preparedaqu ymptoms, sand gray matt was collected using 3-T scanners and preprocessed using the performance (i.e,a significant AR2with p<0.05). s perto etween the pe Genetic Data Genotyping was carried out from blood drawn from total intracranial volume.and site.Sigificance of the results ntormation was col was given by 10,000 random permutations (reported as Beadchin Human Ge p-perm)and wasv edby the comparisons be enwe totaling 506.932 single-nucleotide the R packanp ize with polymorphisms available for establishing the polygenic risk ersistent ADHD goun)(46) score (PRS)for ADHD (see eMethods 4 in the online supplement) lates 3519 Statistical analysis the di (4Z Voxel-wise brain-wide association analysis.A whole-brain summary statistics were downloaded from the Psychiatric Genomics Consortium(http://www.med.unc.edu/pgc/results ay matt The primary analyses are ba on the IMAGEN.Age ore a total intr nial volun (48) and site were considered as covariates.IQ is not recom info(4).Associations of PRS with the neurop -hological mended as a variable to be controlled in cognitive studieso variables were tested by partial correlation analyses while ing for age te, ations with natter v ers were asses 846 aip.psychiatryonline org Am J Psychiatry 177-9.September 2020
Structural MRI The MRI acquisition protocols and quality controls in IMAGEN have been describedin detail (28). A high-resolution T1-weighted magnetization-prepared gradient echo sequence was collected using 3-T scanners and preprocessed using the VBM8 toolbox, as reported previously (43) (see eMethods 3 in the online supplement). Genetic Data Genotyping was carried out from blood drawn from IMAGEN participants (28). Genotype information was collected at 582,982 markers using the Illumina Human Genotyping BeadChip. After quality control, 1,790 cases were included in our sample, totaling 506,932 single-nucleotide polymorphisms available for establishing the polygenic risk score (PRS) for ADHD (see eMethods 4 in the online supplement). Statistical Analysis Voxel-wise brain-wide association analysis. A whole-brain analysis was conducted at the voxel level using the general linear model in SPM12 to identify clusters with gray matter volume associated with the ADHD total score at baseline in IMAGEN. Age, sex, handedness, total intracranial volume, and site were considered as covariates. IQ is not recommended as a variable to be controlled in cognitive studies of neurodevelopmental disorders, since it is often affected by the disorder (44). An uncorrected p threshold of 0.001 at voxel level, with a cluster-level family-wise error (FWE) corrected p threshold of 0.05, was applied to identify significant clusters (45). Neuropsychological association analysis. Separate partial correlation analyses were conducted between neuropsychological measures (working memory, intrasubject variability, stop-signal reaction time, and delay discounting) and both ADHD symptoms and gray matter volumes of the significant clusters, controlling for age, sex, handedness, total intracranial volume, and site. Confidence interval was given by 5,000 bootstraps. Next, we included other variables as covariates for the association analysis of one variable. If a significant association becomes insignificant after controlling for other variables, this association is not independent of other variables but is contributed by some common factor shared between cognitive and motivational deficits. Prospective association analysis. We extended our analysis to ADHD symptoms at age 16 in the IMAGEN participants. Hierarchical multiple regression was applied to identify significant associations between the baseline features and ADHD symptoms 2 years later. In these regression models with covariates and corresponding baseline symptoms, the behavioral variables and gray matter volumes of the significant clusters were entered one by one. A variable was retained in the model ifit significantly elevated the model performance (i.e., a significant DR2 with p,0.05). Analysis of covariance was performed between the persistent ADHD group and the typically developed control group in IMAGEN while controlling for sex, handedness, total intracranial volume, and site. Significance of the results was given by 10,000 random permutations (reported as p-perm) and was validated by the comparisons between wellmatched samples (healthy control subjects were selected by the R package MatchIt to match the sample size with the persistent ADHD group) (46). Polygenic analysis. The latest genome-wide association meta-analysis of 20,183 patients with ADHD and 35,191 control subjects was used as the discovery data set (47); the summary statistics were downloaded from the Psychiatric Genomics Consortium (http://www.med.unc.edu/pgc/resultsand-downloads). The primary analyses are based on the threshold of p,0.50, since it maximally captures phenotypic variance (48), using PRS software (PRSice; http://prsice. info/) (49). Associations of PRS with the neuropsychological variables were tested by partial correlation analyses while controlling for age, sex, and site, and its associations with gray matter volumes of the significant clusters were assessed by TABLE 1. Characteristics of the study population in the IMAGEN cohorta Characteristic or Measure Baseline (N=1,963) 2-Year Follow-Up (N=1,518) N %N% Male 952 48.5% 728 48.0% Mean SD Mean SD Age (years) 14.43 0.40 16.47 0.57 Hyperactivity-inattention subscale on parent SDQ Total score 2.97 2.29 2.39 2.05 Hyperactivity-impulsivity score 0.70 1.05 0.47 0.87 Inattention score 2.27 1.65 1.92 1.57 N %N % ADHD categories by hyperactivity-inattention total scoreb Normal 1,690 86.1 1,394 91.8 Borderline 107 5.5 64 4.1 Abnormal 166 8.5 62 4.1 Mean SD Delay discounting –1.98 0.61 Working memory 19.45 14.00 Intrasubject variabilityc 119.38 30.96 Stop-signal reaction timec 186.43 61.90 a ADHD=attention deficit hyperactivity disorder; SDQ=Strengths and Difficulties Questionnaire. b Scores were categorized as follows: normal: score ,6; borderline: score of 6; abnormal: score .6. c N=1,846. 846 ajp.psychiatryonline.org Am J Psychiatry 177:9, September 2020 NEURAL CORRELATES OF THE DUAL-PATHWAY MODEL FOR ADHD IN ADOLESCENTS
SHEN ET AL TABLE2.Associations of neuropsychological variables with ADHD symptoms and gray matter volumes of the signficant clusters ADHD Symptoms Gray Matter Volume Total Score Inattention Prefrontal Cluster Posterior occipital Cluster Measure 95%C 95%C1 95%C1 95%C 95%C1 Norking memory019*015.0240.09*0.05.0140.21*0.16025-0.04 0.08.0.01 -0.08*-0.12.-0.03 01001 n0y0.16*0.12.0.210.070.03.0120.180.13.0.22-0.04-0.09.0.005-0.07… -0.11-0.02 010*+0.05.0.150.05 oretedfo 0.001,0.10011*0.06.015-0.05-0.09.-0.002-0.05-0.09.-3.3e-4 0120.08,0170.080.03.013012*0.07,0.16-0.04-0.09,0.005-0.05 -010.-0.01 Gray mat ite and total ic p<0.05."p<0.01.p<0.00 additionally controlling for handedness and totalintracranial correlations were not confounded by one another(Table 2) volume. There was no significant correlation between ADHI and discounting rate was positively associated with working ADHD-200 clinical sample.Using a mask of the significant memory errors(r=0.13,df1952,p<0.00L,95%C1-0.08,0.17) ci) variability (r=0.09.df=1835, simificant clusters by comparing patients with control Neuroanatomical Correlates of Inattention in subjects;2)which ADHD patient subtype had the lowest gray a Population-Based Cohort control subjects:and3)whether medication had any remedial =3:3.357 voxels:peak t=-4.29.df-1950.cluster-level effect on the reduced gray matter volumes of the significant WE<0.001)and the posterior occipital cortex (x=-15, ers by group comparisons of ver-medicated patients =91.5,z=112 voxels;peakt 32.df=1 and site. and anterior insula,and the posterior occipital cluster was mainly in the left cuneus and extended to the left calcarine RESULTS tex (Figure 1).Thes not confounded by a的baine Ca地oiag became insignificant after controlling for the inattention memory,intrasubject variability,and delay discounting score but remained significant after controlling for the were positively associated with ADHD symptoms,and the hyperactivity-impulsivity score (prefrontal cluster:r=-0.08, .September 2020
additionally controlling for handedness and total intracranial volume. Validation. We applied the same preprocessing pipeline of structural neuroimaging data as that used in IMAGEN to the ADHD-200 clinical sample. Using a mask of the significant clusters identified in IMAGEN, the gray matter volume of each cluster was extracted for analyses. We tested 1) whether patients with ADHD had lower gray matter volumes of the significant clusters by comparing patients with control subjects; 2) which ADHD patient subtype had the lowest gray matter volumes of the significant clusters by comparing between two ADHD subtypes (hyperactive/impulsive subtype was excluded because of a small sample size of eight) and control subjects; and 3) whethermedication had any remedial effect on the reduced gray matter volumes of the significant clusters by group comparisons of never-medicated patients, medicated patients, and control subjects. All analyses were controlled for age, sex, handedness, total intracranial volume, and site. RESULTS Descriptive Statistics In the IMAGEN cohort at baseline (Table 1), working memory, intrasubject variability, and delay discounting were positively associated with ADHD symptoms, and the correlations were not confounded by one another (Table 2). There was no significant correlation between ADHD symptoms and stop-signal reaction time (p.0.05), and therefore it was not included in further analyses. Delay discounting rate was positively associated with working memory errors (r=0.13, df=1952, p,0.001, 95% CI=0.08, 0.17) and increased intrasubject variability (r=0.09, df=1835, p,0.001, 95% CI=0.04, 0.13). Neuroanatomical Correlates of Inattention in a Population-Based Cohort In IMAGEN at baseline, we found that higher ADHD total score was associated with lower gray matter volumes of two brain clusters in both the prefrontal cortex (x=219.5, y=49.5, z=3; 3,357 voxels; peak t=24.29, df=1950, cluster-level pFWE,0.001) and the posterior occipital cortex (x=21.5, y=91.5, z=15; 1,295 voxels; peak t=24.32, df=1950, cluster-level pFWE=0.025). The prefrontal cluster covered the left ventromedial prefrontal cortex, dorsal anterior cingulate cortex, and anterior insula, and the posterior occipital cluster was mainly in the left cuneus and extended to the left calcarine cortex (Figure 1). These associations were not confounded by either site (see Figure S1 in the online supplement) or IQ (see eResults 2 in the online supplement). These associations became insignificant after controlling for the inattention score but remained significant after controlling for the hyperactivity-impulsivity score (prefrontal cluster: r=20.08, TABLE 2. Associations of neuropsychological variables with ADHD symptoms and gray matter volumes of the significant clustersa ADHD Symptoms Gray Matter Volume Total Score HyperactivityImpulsivity Inattention Prefrontal Cluster Posterior occipital Cluster Measure r 95% CI r 95% CI r 95% CI r 95% CI r 95% CI Working memoryb 0.19*** 0.15, 0.24 0.09*** 0.05, 0.14 0.21*** 0.16, 0.25 20.04 20.08, 0.01 20.08*** 20.12, 20.03 Delay discountingb 0.13*** 0.08, 0.17 0.06** 0.02, 0.11 0.14*** 0.09.0.18 20.07** 20.11, 20.02 20.06* 20.10, 20.01 Intrasubject variabilityc 0.14*** 0.10, 0.19 0.09*** 0.04, 0.14 0.14*** 0.10, 0.19 20.05* 20.10, 20.01 20.06** 20.11, 20.01 Working memory corrected for delay discounting and intrasubject variability 0.16*** 0.12, 0.21 0.07** 0.03, 0.12 0.18*** 0.13, 0.22 20.04 20.09, 0.005 20.07** 20.11, 20.02 Delay discounting corrected for working memory and intrasubject variability 0.10*** 0.05, 0.15 0.05* 0.001, 0.10 0.11*** 0.06, 0.15 20.05* 20.09, 20.002 20.05* 20.09, 23.3e–4 Intrasubject variability corrected for working memory and delay discounting 0.12*** 0.08, 0.17 0.08** 0.03, 0.13 0.12*** 0.07, 0.16 20.04 20.09, 0.005 20.05* 20.10, 20.01 a ADHD symptoms were adjusted for age, sex, and site. Gray matter volume was adjusted for age, sex, handedness, site, and total intracranial volume. Confidence intervals were estimated by bootstrap 5,000 times. b N=1,963. c N=1,846. *p,0.05. **p,0.01. ***p,0.001. Am J Psychiatry 177:9, September 2020 ajp.psychiatryonline.org 847 SHEN ET AL
NEURAL CORRELATES OF THE DUAL-PATHWAY MODEL FOR ADHD IN ADOLESCENTS FIGURE1 Significa ain clusters asso ciated with ADHD total score in a population-based cohort had reduced ra matter vol umes of both the prefrontal as compared with the [SD=103]comparedwith6.23 mL[SD119:F=6.37,df ter (2.06 mL SD=0.351 com. paredwith2.19mL[SD=0.28]; ng the onata9e14Nl26i000 nificant results with even of005 larger effect sizes were found ster 9.5 95.2 29. 1950 4.32 0.001 asing matchcd-group 0251 No clust supplement). f=1949,p=0.002,95%CI=-0.03,-0.1 ate Phenotypes With mical Correlates of Inattention Selectively In the IMAGEN sample,we found that higher PRS for ADHD was associated with higher ADHD total score at baseline Variability,or Delay Discounting In IMAGEN at baseline,we found that a larger number of working memory errors was CL-02007.95%C002L0109 d0.6 matter volume of the posterior occipital cluster only (r=-0.07,df=1831,p=0.005)(Table 2).Similar to working .06,df=1777,p=0.009,95%C1=-0106,-0.015) memory,increased intrasubject variability was associate lower gray matte Validation using an adHd clinical cohort In the ADHD-200 sample,we confirmed that patients had delay discounting (r=0.05.df=1831.p=0.027(Table 2). ower gray matter volumes in both the prefrontal (3.86 mL SD-167 compared wi 40m king memory ith 1.28 ml SD0.291r928.dfL.49Lp-0.002:,0.019 (Figure2B. ter:r=-0.05,df=1831,p=0.049)(Table2). ese volumetric reductions were nonsignificant in patients ective Associations With Inattention 2 Ye sLate onding ADHD sympto clister:F=1292 dfl 354 00020035 age 14,working memory and delay discountingat age 14were xccipital cluster:F=729,df=1 354.p=0.007:p0.20) selectively associated with inattention (t=2.35, d1505 (Figure 2C-D). and hyperactivity-impulsivity(24. y-204d404.AR2-0002 p=0.042)and gray matter volume of the posterior oc cipital cluster (t=-3.55,df=1404,AR2=0.005,P<0.001) olumes of both clusters,the medicated patients(N56)had associate with inattention 2 years late lumes,and the never-medicated gray matter volumes 848 aip psychiatryonline.oro AmJPsychiatry 177:9.September 2020
df=1949, p,0.001, 95% CI=20.03, 20.13; posterior occipital cluster: r=20.07, df=1949, p=0.002, 95% CI=20.03, 20.11). Neuroanatomical Correlates of Inattention Selectively Associated With Working Memory, Intrasubject Variability, or Delay Discounting In IMAGEN at baseline, we found that a larger number of working memory errors was associated with lower gray matter volume of the posterior occipital cluster even after controlling for intrasubject variability and delay discounting (r=20.07, df=1831, p=0.005) (Table 2). Similar to working memory, increased intrasubject variability was associated with lower gray matter volume of the posterior occipital cluster even after controlling for working memory and delay discounting (r=20.05, df=1831, p=0.027) (Table 2). Greater delay discounting rate was associated with lower graymatter volumes of both clusters even after controlling for working memory and intrasubject variability (prefrontal cluster: r=20.05, df=1831, p=0.042; posterior occipital cluster: r=20.05, df=1831, p=0.049) (Table 2). Prospective Associations With Inattention 2 Years Later After controlling for the corresponding ADHD symptom at age 14, working memory and delay discounting at age 14 were selectively associated with inattention (t=2.35, df=1505, p=0.019) and hyperactivity-impulsivity (t=2.24, df=1505, p=0.025) at age 16. In the multivariate regression model, we found that bothworkingmemory (t=2.04, df=1404,DR2 =0.002, p=0.042) and gray matter volume of the posterior occipital cluster (t=23.55, df=1404, DR2 =0.005, p,0.001) at age 14 were associated with inattention 2 years later (Table 3). Adolescents with persistent ADHD symptoms (N=29) had reduced gray matter volumes of both the prefrontal cluster as compared with the typically developed control subjects (N=1,278; 5.63 mL [SD=1.03] compared with 6.23 mL [SD=1.19]; F=6.37, df=1, 1295, p-perm=0.012; partial eta-squared [h2 p]=0.005) and the posterior occipital cluster (2.06 mL SD=0.35] compared with 2.19mL [SD=0.28]; F=5.12,df=1,1295,p-perm=0.022; h2 p=0.004; see Figure S2 in the online supplement). Significant results with even larger effect sizes were found using matched-group comparisons (29 compared with 58; see eResults 3 in the online supplement). Associations of Neuropsychological and Neuroanatomical Intermediate Phenotypes With Polygenic Risk for ADHD In the IMAGEN sample, we found that higher PRS for ADHD was associated with higher ADHD total score at baseline (r=0.14, df=1779, p,0.001, 95% CI=0.097, 0.188), more working memory errors ( r=0.07, df=1779, p=0.002, 95% CI=0.026, 0.121), greater delay discounting rate (r=0.06, df=1779, p=0.007, 95% CI=0.021, 0.109), and lower gray matter volume of the posterior occipital cluster only (r=20.06, df=1777, p=0.009, 95% CI=20.106, 20.015). Validation Using an ADHD Clinical Cohort In the ADHD-200 sample, we confirmed that patients had lower gray matter volumes in both the prefrontal (3.86 mL [SD=1.67] compared with 4.40 mL [SD=1.56]; F=12.18, df=1, 491, p,0.001; h2 p=0.024) (Figure 2A) and the posterior occipital clusters (1.21 mL [SD=0.30] compared with 1.28 mL [SD=0.29]; F=9.28, df=1, 491, p=0.002;h2 p=0.019) (Figure 2B). These volumetric reductions were nonsignificant in patients with the combined subtype (N=129) and significant only in patients with the inattentive subtype (N=96; prefrontal cluster: F=12.92, df=1, 354, p,0.001; h2 p=0.035; posterior occipital cluster: F=7.29, df=1, 354, p=0.007; h2 p=0.20) (Figure 2C–D). Medication Effects In the ADHD-200 sample, we found that the typically developed control subjects (N=267) had the highest gray matter volumes of both clusters, the medicated patients (N=56) had intermediate gray matter volumes, and the never-medicated patients (N=99) had the lowest gray matter volumes. FIGURE 1. Significant brain clusters associated with ADHD total score in a population-based cohorta a The results were given by a voxel-wise whole brain analysis using the IMAGEN cohort at age 14 (N=1,963). Age, sex, handedness, total intracranial volume, and site were used as covariates. An uncorrected p threshold of 0.001 at voxel level, with a cluster-level family-wise error (FWE) corrected p threshold of 0.05, was applied to identify significant clusters. Two clusters were found to be negatively associated with the ADHD total score: the prefrontal cluster (x=219.5, y=49.5, z=3; 3,357 voxels; peak t=24.29, df=1950, cluster-level pFWE,0.001) and the posterior occipital cortex (x=21.5, y=91.5, z=15; 1,295 voxels; peak t=24.32, df=1950, cluster-level pFWE=0.025). No clusters were found to be positively associated with the ADHD total score. 848 ajp.psychiatryonline.org Am J Psychiatry 177:9, September 2020 NEURAL CORRELATES OF THE DUAL-PATHWAY MODEL FOR ADHD IN ADOLESCENTS