Growing out of attention deficit hyperactivity disorder: insights from the ‘remitted’ brain.

Abstract

We consider developmental and cognitive models to explain why some children ‘grow out’ of attention deficit hyperactivity disorder (ADHD) by adulthood. The first model views remission as a convergence towards more typical brain function and structure. In support, some studies find that adult remitters are indistinguishable from those who were never affected in the neural substrates of ‘top-down’ mechanisms of cognitive control, some ‘bottom-up’ processes of vigilance/response preparation, prefrontal cortical morphology and intrinsic functional connectivity. A second model postulates that remission is driven by the recruitment of new brain systems that compensate for ADHD symptoms. It draws support from demonstrations of atypical, but possibly beneficial, patterns of connectivity within the cognitive control network in adult remitters. The final model holds that some childhood ADHD anomalies show lifelong persistence, regardless of adult outcome, supported by shared reports of anomalies in remitters and persisters in posterior cerebral and striato-thalamic regions. The models are compatible: different processes driving remission might occur in different brain regions. These models provide a framework for future studies which might inform novel treatments to ‘accelerate’ remission.

Introduction.

There has been a recent surge of interest in the mechanisms that may explain why some children, but not others, outgrow attention deficit hyperactivity disorder (ADHD) as they move into adulthood. This new focus complements an earlier concentration on adults with ADHD symptoms that have persisted from childhood. Studying adults whose childhood ADHD has remitted is important for several reasons. First, it is reasonable to surmise that understanding the mechanisms of remission occurring over a period of years could inform treatment strategies to promote remission over a shorter time frame. Such novel treatment approaches are greatly needed. While, psychostimulant medication is the mainstay of treatment and has proven efficacy in the short- to medium-term, its long-term impact is less clear (Craig et al., 2015; Fredriksen et al., 2013). Additionally, meta-analyses suggest that behavioral interventions have minimal efficacy for core symptoms of ADHD when probably blinded ratings are considered although there are beneficial effects on other outcomes, including parenting and childhood conduct problems (Daley et al., 2014; Sonuga-Barke et al., 2013). Secondly, an understanding of remission might provide biomarkers to help predict outcome and thus identify children likely to remit or persist. Such biomarkers are needed as we cannot accurately predict an individual child’s likely outcome using routinely available clinical, sociodemographic and neuropsychological data. While baseline symptom severity and multiple comorbidities are associated with adult outcomes, these factors, when combined, account for between 15–20% of the variance in outcome (Biederman et al., 2011; Cheung et al., 2015; Taylor et al., 1996). There are mixed findings on whether or not socioeconomic status, general intelligence, family environment and cognitive skills are associated with adult outcome (Biederman et al., 2009; Biederman et al., 2011; Brocki et al., 2007; Cheung et al., 2015; Hart et al., 1995; Langley et al., 2010; Molina et al., 2009; Weiss and Hechtman, 1993). We thus turn to objective measures of the brain and ask if these can help predict clinical outcome. Lastly, insights into remission from ADHD could inform our understanding of recovery from other neurodevelopmental disorders, such as Tourette’s, oppositional defiant disorder and perhaps even autistic spectrum disorders.

In this review, we first consider definitions of remission and summarize recent estimates of its prevalence. We then outline developmental and cognitive models of remission of ADHD and review the literature using these concepts. Due to the limited data, quantitative meta-analyses are not possible. Notably, nine of the 17 imaging studies included were published in the past three years highlighting the burgeoning interest in this topic. Our focus is on understanding remission of childhood ADHD; the extensive literature on adult ADHD persisting from childhood is reviewed elsewhere (Alderson et al., 2013; Bonvicini et al., 2016; Davidson, 2008; Franke et al., 2012) and herein discussed only when relevant. We also do not include the debate about the possibility of adult onset of ADHD which has been discussed extensively elsewhere (Agnew-Blais et al., 2016; Caye et al., 2017; Caye et al., 2016; Faraone and Biederman, 2016; Sibley et al., 2017).

Section 1: Remission: definitions and prevalence.

Most studies in this review define remission as no longer meeting DSM diagnostic criteria for ADHD. This approach reflects clinical practice which often relies on categorical diagnoses to guide treatment but has several drawbacks. Firstly, this definition means that an adult could have up to four symptoms of ADHD causing substantial impairment and still be considered to be ‘remitted’. A recent change in the DSM means this individual would now be categorized as having ADHD in partial remission. While a welcome addition, this category is too recent to have been incorporated into the studies in this review.

It has also been argued that studies based on symptom dimensions may prove more revealing that those using diagnostic categories alone. In line with this concept, epidemiological evidence suggests that the symptoms of ADHD can be considered dimensionally, lying at the extreme end of a continuous distribution of symptoms (Lubke et al., 2009; Polderman et al., 2007). Arguably such symptoms dimensions might prove easier to tie to underlying. objective cognitive constructs that are also continuously distributed (Insel et al., 2010). Assessing age-related change in symptoms of inattention and impulsivity also provides a more precise measure of clinical course than is obtained through considering changes in diagnostic status alone. We thus emphasize results pertaining to symptom dimensions when available.

What is the prevalence of remission in ADHD? A meta-analysis of data to 2005 found an age dependent decline of ADHD, with around 85% attaining remission of the syndrome by adulthood (Faraone et al., 2006). However nearly half of these ‘syndromic’ remitters still had symptoms that caused impairment. More recent studies return similar, if slightly lower rates of remission, finding that the syndrome remits by early adulthood in around 55–70% of cases rising to ~ 80% in later adulthood –Table 1 and Figure 1 (Biederman et al., 2010; Halperin et al., 2008; Klein et al., 2012; Molina et al., 2009; Shaw et al., 2013). Of these, 25–48% show impairing symptoms not numerous enough to meet current diagnostic criteria.

Table 1.

Recent prospective studies of childhood ADHD. Rates of remission (syndromic), persistence of the full syndrome (diagnosed using DSM-IVR throughout) are given. Rates are also given for ‘residual’ or partially remitted ADHD, with the definition used in each study.

Reference	Numbe r	Age: Mean, SD (years)		Remissio n of syndrome (full and partial remission )	ADHD persistence
		Baselin e	Follow -up		Partial remission	Full syndrome
Klein et al 2013	135 (all male)	8 (1)	41 (3)	68%	Subthreshol d (3, 4 or 5 symptoms) 48% (N=58 of 121)	Total: 22% (N=30) Combined: 7% (9) Inattentive: 7% (N=10) Hyperactive/Impulsive : 8% (N=11)
Shaw et al 2013	92	10 (3)	24 (4)	60%	Subthreshol d (3, 4 or 5 symptoms in either domain) 36% (N=33)	Total: 40% (N=37) Combined: 13% (N=12) Inattentive: 23 % (N=21) Hyperactive/Impulsive : 4% (N=4)
Molina et al 2009	436	9 (1)	14 (1) 6 year F/U 17 (1.2) 8 year F/U	57% 70%	Not reported Not reported	Total: 43% Total: 30% (N=132) Combined: 8.8 % (N=38) Inattentive: 18.9% (N=83) Hyperactive/Impulsive : 2.5% (N=11)
Halperin et al 2008	98	9 (1)	18.3 (2)	55%	Subthreshol d (4 or 5 symptoms in either domain) 26% (N=25)	Total persistent- 45% (N=44) Combined: 14% (N=14) Inattentive:21% (N=21) Hyperactive/Impulsive : 9% (N=9)
Biederma n et al 2010	110 (14 and all males)-	11 (1)	22 (4)	65%	Symptom persistence (3, 4 or 5 symptoms in either domain)- 22% (N=24) Functional impaired (but few symptoms)- 15% (N=16) Remitted on Medication – 6%(N=7) Full remission – 22% (N=24)	Total: 35% (N=39)
Francx et al 2015	129	12 (2)	17 (3)	15%	Symptom persistence (3, 4 or 5 symptoms in either domain)- 9% (N=11) Full remission, 6 % (N=8)	Total: 85% (N=110) Combined: 36% (N=47) Inattentive: 40% (N=52) Hyperactive/Impulsive : 9% (N=11)

Figure 1:

Rates of persistence of the full syndrome of ADHD (DSM-IVR definition throughout) and impairing symptoms (different thresholds used). The average age at final assessment is indicated on the x axis. Studies (1) Klein et al 2013; (2) Shaw et al 2013; (3) Halperin et al 2008; (4) Biederman et al 2010; (5) Molina et al 2009 ([a]at age 14, [b] at age 17); (6) Francx et al 2015. Not included is a study by Bode et al 2015, which showed 100% syndrome remission during adolescence.

Section 2: Developmental-cognitive models of remission.

Remission has both developmental and cognitive dimensions. To understand remission, we thus need to identify both the cognitive mechanisms that underpin remission and the nature of the neurodevelopmental change these mechanisms undergo. We first consider models of neurodevelopmental change as these highlight the importance of studying the ‘remitted brain’. Three neurodevelopmental models have been used implicitly or explicitly to explain remission of ADHD. The first model, which we call ‘convergence/normalization’, posits that remission is due to convergence towards more typical brain function and structure (Halperin and Schulz, 2006; Shaw et al., 2013). Thus, remission occurs due to the rectification of early anomalies in brain structure and function whereas persistence is linked with persisting neural anomalies. A second model postulates that remission is driven by the recruitment of new brain systems that help the individual overcome the core symptoms of ADHD (Francx et al., 2015a).. We call this the ‘compensation/neural reorganization’ model. A final model, here called the ‘fixed anomaly’ model, argues that the neural anomalies of ADHD leave an indelible mark on the brain which persists regardless of the clinical course of ADHD (Proal et al., 2011). These models are not incompatible. For example, one group has argued that while ADHD is caused by subcortical anomalies which persist throughout the lifespan, its variable clinical course is determined by plasticity of the cerebral cortex (Halperin and Schulz, 2006).

All three models predict atypical neural features in adults with persistent ADHD. Critically, however, the models make different predictions about the brain in remitted ADHD- see Table 2. In ‘convergence/normalization’ models, the remitted brain will resemble the never affected brain. In models positing compensation/neural reorganization, the remitted brain will differ from both the never affected and the persistent ADHD brain, but in different ways. In this model, the differences between the remitted and never affected brain are beneficial; differences between the persistent and never affected brain are not. Finally, in the fixed anomalies model, the persistent and remitted brain will resemble each other, and both will differ from the never affected brain. To test these models, it is thus critical to include individuals whose childhood ADHD has at least partly remitted and not just adults with ADHD, all of whom have the persistent disorder.

Table 2:

Developmental models of remission in ADHD and their different predictions about the remitted adult brain.

Model	Persistent brain	Remitted adult brain
Compensation/ neural reorganization	Atypical	Remitted ≠ persistent ≠ never affected brain
Normalization/convergence	Atypical	Remitted = never affected brain
Fixed anomalies ‘scar’	Atypical	Remitted = persistent ≠ never affected brain

The ideal design for defining the neurodevelopmental models of remission would be to prospectively acquire clinical assessments and neuroimaging in tandem from childhood into adulthood. This is rarely achieved as most methods for defining brain function and connectivity have only been in widespread use for the past decade and have changed rapidly. More frequently, a mixed design has been used in which the participant is diagnosed as a child and then followed into adulthood at which stage the first brain imaging is acquired. The fully prospective and mixed designs both avoid the retrospective determination of the age of onset of symptoms which can be unreliable (Mannuzza et al., 2002). The fully prospective design also has the additional advantage of allowing us to conclude that any neural changes reflect, if not drive clinical course, as both imaging and clinical status are acquired in tandem. When imaging data are acquired in adulthood only, interpretation is more complicated. For example, consider the case when the remitted group is found in adulthood to resemble the never affected group. This could represent a convergence towards more typical dimensions accompanying remission. However, it could also arise if the remitted group had more typical neural features in childhood that have been carried forward into adulthood. Further, atypical brain features in adults with persistent ADHD are usually taken to reflect anomalies persisting from childhood. However, without childhood imaging, one cannot rule out the possibility that these anomalies reflect emergent neural changes driven by persistent symptoms. These limitations notwithstanding, important initial insights into remission can still be gained through cross-sectional studies of adults.

Cognition forms the second dimension of remission. Three classes of cognitive mechanisms for remission have been proposed. The first holds that remission is due to the late maturation of the neural substrate that supports ‘top-down’ processes, particularly cognitive control (Halperin and Schulz, 2006). The second holds that changes in ‘bottom-up’, more automatic processes, such as maintaining vigilance and response preparation, drive remission (Cheung et al., 2016; Michelini et al., 2016). Closely aligned with the concept of ‘bottom-up’ drivers of remission, are models that view remission as tied to changes in the processing of rewards and temporal information (Mackie et al., 2007; Wetterling et al., 2015). The third model posits that neural processes mediating remission will center on the so-called default mode network (DMN) (Castellanos and Aoki, 2016; Konrad and Eickhoff, 2010). The DMN is prominent during task free periods that evoke introspective processing. It is argued that intrusions of the DMN into task-oriented processing produce the characteristic cognitive deficits in ADHD, including deficient sustained attention and distractibility (Fassbender et al., 2009; Rosenberg et al., 2015; Wen et al., 2012). While this model of ADHD has mostly been applied to the childhood disorder, we ask if the DMN and its interactions with ‘task-positive’ networks might also relate to ADHD adult outcome.

Many individuals with ADHD are treated with psychostimulant medication, fewer with non-stimulant medications, such as atomoxetine and some with behavioral interventions such as parent training. Psychostimulant medication has been most extensively studied and has clear efficacy (Chan et al., 2016; Storebø et al., 2015). Thus, we focus on psychostimulant medication, asking how it might relate to the different models of remission. Firstly, psychostimulant medication could be viewed as leading to shifts towards more typical brain structure and function, for example through its effects on dopaminergic tone (Schrantee et al., 2016). The treatment might also promote shifts towards more typical neural features by allowing the child to participate more fully in academics and by promoting beneficial interactions with peers and family (Kortekaas-Rijlaarsdam et al., 2018). In support of this concept, most, but not all studies, suggest that children with ADHD on psychostimulant medication have more typical anatomic dimensions than their unmedicated peers with ADHD (Friedman and Rapoport, 2015; Spencer et al., 2013). Psychostimulants are also associated with more typical brain function, at least in the acute phase when psychostimulant medication is active (Rubia et al., 2014). Nearly all of these studies are observational and thus casual links between psychostimulants and alterations in brain structure/function cannot be inferred, as third factors could be at play, such as symptom severity. Nonetheless, the hypothesis emerges that psychostimulant medication could be a driver of a convergence towards more typical brain structure and function. There is much less evidence that psychostimulant medication leads to neural reorganization, insofar as the medication mostly shifts brain function towards typical levels, rather than recruiting entirely novel brain regions to compensate for the neural dysfunctions that underlie core symptoms.

Psychostimulant medication is also important to consider methodologically, given that adults with remitted or persistent ADHD are likely to differ in their medication exposure. Most neuroanatomic studies of remission have ensured the main results hold when psychostimulant medication history is entered as a covariate or when analyses are repeated after excluding those taking regular medication. For functional imaging studies, psychostimulant medication is usually temporarily stopped (at least the day before the scan) to attenuate its acute effects. To consider chronic effects of psychostimulant medication, some studies have also repeated analyses after excluding those on medication.

Two additional factors are important to consider; imaging quality and comorbid disorders. Neuroimaging is prone to artifacts resulting from head motion and it is extremely challenging to integrate data acquired on different scanners across different sites at different times. Quality control of imaging data is a rapidly evolving field, with no consensus approach. For neuroanatomic data, it is currently recommended that both the ‘raw’ image input and the final segmentations are visually expected by at least one, and ideally several trained and reliable raters. Recently, automated tools have been introduced which promise a more standardized approach (Esteban et al., 2017; White et al., 2018). For functional imaging studies, finer control of artefacts likely related to head motion is feasible, which allows the exclusion of data that show excessive head motion or other artifacts. In fMRI head motion can also be measured in real time and then used to exclude data or be used as a covariate in analyses. Head motion is not the only challenge. As mentioned above, the integration of neuroimaging data acquired across time on different scanners using sequences that emphasize different tissue contrast properties is also difficult. This challenge is not insurmountable but much remains to be done (Bocchetta et al., 2015; Nir et al., 2013).

ADHD is often accompanied by other neuropsychiatric disorders. In neuroimaging studies, adults with psychotic disorders or substance dependence are mostly excluded, as these disorders have well-documented effects on brain structure and function. However, adults with a history of childhood ADHD also report high rates of mood and anxiety disorders, particularly when ADHD has persisted (Kessler et al., 2006; Uchida et al., 2015). While all studies have tried to consider this factor, usually by repeating analyses after excluding those with these disorder, none have been powered to consider systematically these important comorbidities.

Section 3: Literature review

We now review the literature, organizing studies around the developmental and cognitive models of remission. Table 3 gives a summary of the 17 studies included in this review. The search strategy to identify the relevant literature using PubMed was:- (ADHD OR attention deficit hyperactivity disorder) AND (outcome OR persist* OR remission OR remitted OR follow-up OR prospective OR history) AND (neuroimaging OR DTI OR *MRI* OR neurophys* OR cerebellum OR cortex). Further constraints were language of publication (English), year (2000 onwards). This returned identified 710 articles; all abstracts were examined (by PS) retaining only those that included neuroimaging on adolescent or adult subjects who had remitted from childhood ADHD.

Table 3:

Summary of neuroimaging studies included. The final three columns indicate the degree to which each model of ADHD outcome is supported by the study. Some of the major strengths and limitations are noted in the final columns.

Referenc e	Participants (N)	Age (year) M(SD)		Outcomes	Models					Comments	Association with hyperactivity- impulsivity and inattention
	Persistent ADHD /Remitted /Never Affected	Entr y	Follo w-up		Converge/ ‘Normalize’	Compensat e		Fixed trait
Task based imaging
Schulz et al 2017	11/16/28	7–11	24 (2)	fMRI: stimulus and response conflict task	+++ Fronto- parietal activation during high cognitive demand		-		-	Strengths: Tested core ADHD cognitive deficit Limitations: Modest N of remitters	Not reported
	Persistent ADHD /Remitted /Never Affected	Entr y	Follo w-up		Converge/ ‘Normalize’		Compensat e		Fixed trait
Szekely et al 2017	fMRI:24/40/84 MEG-25/26/46	9 (1)	24(4)	MEG and f RI: response inhibition	+++ Prefrontal inhibitory processing		-		+++ Striatal activity	Strengths: Multimodal Large sample size Limitations: Adult endpoint imaging only	Hyperactivity- impulsivity (not inattention) associated with atypical inferior frontal activity in fMRI and MEG
Schneider et al 2010	11/8/17	N/A	32(10 )	fMRI: NoGo in continuous performance task	++ More typical activation in fronto- parietal regions in remitters		-		-	Strengths: medication naive Limitations: First clinical assessment in adulthood Small N of remitters	Inattention, hyperactivity, impulsivity scores all correlated with decreased fronto- striato-parietal hypoactivation, and with increased insula activity
	Persistent ADHD /Remitted /Never Affected	Entr y	Follo w-up		Converge/ ‘Normalize’		Compensat e		Fixed trait
Clerkin et al 2013	16/19/32	9(1)	24(2)	fMRI: cued response preparation (‘bottom-up’)	-		+ Thalamo- cortical connectivit y during response preparation		++ Thalamic activation	Strengths: Direct test of an influential model of ADHD remission Limitations: ROI (thalamus) connectivity analyses only	Not reported
Michelini et al 2016	87/23/169	13	19 (3)	EEG: Eriksen Flanker task	+++ Error related negativity (bottom- up)		-		+ Conflict monitoring - N2 signal (top- down)	Strengths: First use of electrophysiology in adult remitters. Limitations: Modest N of remitters	Not reported
	Persistent ADHD /Remitted /Never Affected	Entr y	Follo w-up		Converge/ ‘Normalize’		Compensat e		Fixed trait
Cheung et al 2016	87/23/169	13	19 (3)	EEG: cued flanker continuous performance task	+++ Preparation -vigilance: contingent negative variation (bottom- up)				+ Executive, ‘top-down’ processes (NoGo P3 amplitude)	As above	Not reported
Wetterlin g et al 2015	17/15/22	Not give n	21(4)	fMRI: reward, punishment processing	+++ Fronto- striatal connectivit y		++ Pre-frontal processing of punishmen t		-	Strengths: Only examination of reward processing in remitters Limitations: Modest N of remitters	Not reported
	Persistent ADHD /Remitted /Never Affected	Entr y	Follo w-up		Converge/ ‘Normalize’		Compensat e		Fixed trait
Resting state
Mattfield et al 2014	13/22/17	10 (3)	28 (5)	fMRI : default mode and its interconnectio ns	++ DMN internal connectivit y	-		++ DMN connection to other regions		Strengths: First rsFMRI study of adult outcome Clinical FU from childhood into adulthood Limitations: Seed based analyses only	Not reported
	Persistent ADHD /Remitted /Never Affected	Entr y	Follo w-up		Converge/ ‘Normalize’	Compensat e		Fixed trait
Francx et al 2015	74 (persistent symptoms)/55 (improving symptoms; 19 reached remission)/10 0	12	18 (3)	fMRI: cognitive control network only	-	+++ Higher connectivit y between cognitive and other brain networks in remitters vs controls		-		Strengths: Large overall sample size FU over adolescence Limitations: Examined only cognitive network Few syndromic remitters	Hyperactivity- impulsivity (not inattention) associated with higher connectivity between cognitive control and other networks
Sudre et al 2017	fMRI-41/fMRI- 35/fMRI-71	9 (1)	24(4)	MEG and fMRI: default mode and its interconnectio ns	+++ DMN internal and external connectivit y	-		-		Strengths: Multimodal Large sample size Limitations: Adult endpoint only	Anomalies associated with inattentive symptoms
	MEG-32/MEG- 35/MEG-58
White matter microstructure
	Persistent ADHD /Remitted /Never Affected	Entr y	Follo w-up		Converge/ ‘Normalize’	Compensat e		Fixed trait
Cortese et al 2013	15/25/47	8 (2)	41 (3)	DTI - analyzed at voxel level	+	-		++		Strengths: Clinical FU from childhood into middle age with low attrition rate Limitations: Males only Low N of directions in DTI	Not reported
Bode et al 2015	0/30/30	15 (1)	23(1)	DTI - analyzed at voxel level	?	?		?		Strengths Based on population birth cohort Medication naive Limitations: No persisters in study for contrast 1.5T scanner	Not reported
	Persistent ADHD /Remitted /Never Affected	Entr y	Follo w-up		Converge/ ‘Normalize’	Compensat e		Fixed trait
Francx et al 2015	59 (persistent symptoms)/42 (improving symptoms)/40	12	18 (3)	DTI - analyzed at voxel level	+++	-		-		Strengths: Large overall sample size Symptom level analyses Limitations: Low N of directions Few syndromic remitters	Persistent hyperactive- impulsive (but not inattentive) symptoms correlated with white matter microstructural anomalies
Shaw et al 2015	32/43/74	10(3 )	23(4)	DTI - analyzed at tract level	+++	-		-		Strengths: Clinical FU from childhood to adulthood Large N of DTI directions Limitations: Tract based analyses only	Persistent inattention correlated with anomalies
Neuroanatomic
	Persistent ADHD /Remitted /Never Affected	Entr y	Follo w-up		Converge/ ‘Normalize’	Compensat e		Fixed trait
Shaw et al 2013	37/55/184	10 (3)	24(3)	Cortical thickness	+++ Cortical attention, cognitive control regions converged to typical dimensions in remitters	-		-		Strengths: Combined prospective clinical FU and imaging from childhood into adulthood Limitations: 1.5T scanner	Persistent inattention correlated with degree of cortical convergence
Mackie et al 2007	16/16/32	10 (3)	14(4)	Cerebellar subregional volumes	++ Trajectory of hemisphere s	-		++ Trajectory of vermis		As above	Combined presentation showed most atypical trajectories
	Persistent ADHD /Remitted /Never Affected	Entr y	Follo w-up		Converge/ ‘Normalize’	Compensat e		Fixed trait
Proal et al 2013	17/26/57	8 (2)	41 (3)	Cortical thickness/ VBM	++ Volumes of thalamus/ cerebellum	-		++ Thickness of posterior cortex		Strengths: Clinical FU from childhood into middle age Limitations: Males only Two scanners used	Not reported

a. Remission as the maturation of the neural substrates of ‘top-down’ cognition.

Findings are mixed on whether the course of ADHD symptoms can be explained by change in ‘top-down’ cognitive processes, as measured by behavior. A review of the literature until 2013, and some later reports concluded that behavioral measures of ‘top-down’ processes in childhood, including as response inhibition, interference control, working memory and planning, did not predict later remission (Coghill et al., 2014; McAuley et al., 2014; van Lieshout et al., 2013). There are, however, discrepant findings: three recent studies found that working memory and lower response time variability significantly and substantially predicted symptom improvement from childhood into late adolescence (Karalunas et al., 2017; Sjöwall et al., 2017; van Lieshout et al., 2017). Here, however, we focus on studies that parsed the neural substrate of such ‘top-down’ processes in remitters.

Most studies have used functional magnetic resonance imaging (fMRI), which examines the change in the blood oxygen level-dependent signals across the brain when the subject is either performing a probe of executive function or is simply at rest. Analyses focus either on activity within specific regions or consider functional connectivity across the brain, as defined through the temporally correlated level of blood oxygen level-dependent signals. Other studies have used magnetoencephalography (MEG) which noninvasively detects the magnetic fields evoked by synchronized current flow in neuronal populations. This defines the neuronal oscillations that synchronize brain activity across the brain, supporting key cognitive functions in ADHD (Gu et al., 2015; Ward, 2003). Finally, electroencephalography (EEG) has also been applied to the understanding of remission. This technique non-invasively measures the voltage fluctuations that result from neuronal activity. Each imaging modality has its strengths: for example, the exquisite temporal resolution of MEG and EEG, on the order of milliseconds, complements the excellent spatial resolution of fMRI.

The first study we consider supported the concept that remission stems from the maturation of the neural substrate of ‘top-down’ cognitive processes (Schulz et al., 2017). This fMRI study drew contrasts between 11 adults in remission against 16 adults with persistent ADHD and 29 never affected controls at a mean age of 24 years. The key finding was that under conditions of high cognitive demand (with combined stimulus and response conflicts) patterns of brain activity in the remitted group did not differ from the never affected controls. By contrast, the persistent ADHD group showed hypo-activation compared to controls within inferior prefrontal-parietal lobular areas that support cognitive control. This study of a demanding cognitive process provided a more definitive examination of cortical processes in remission than earlier, smaller studies from this group (Schulz et al.,2005) and argues for ‘normalization’. Similar findings emerge from a fMRI study of 19 adults with a history of childhood ADHD, of whom eight were in partial remission (Schneider et al., 2010). During response inhibition, the degree of hypo-activation of a fronto-parieto-striatal network was associated with the severity of adult symptoms of inattention and impulsivity; fewer symptoms were associated with more typical patterns of activation.

A recent multimodal imaging study also found that the brain activity underpinning ‘top-down’ processes did not differ between adults with remitted ADHD and those who were never affected (Szekely et al., 2017) The study probed response inhibition using the stop signal task, in which the subject responds to a stimulus unless a warning signal appears. Around 180 participants, followed clinically from childhood to adulthood, completed this task in fMRI, MEG or both at an average of 24 years. The study first examined activity in regions supporting response inhibition- the inferior frontal gyri- and found that activity in these regions reflected adult outcome. The persistent group differed significantly from both the remitters and never affected group in contrasts of brain activation during failed versus successful inhibition. However, the remitted and never affected groups did not differ. Furthermore, no differences in prefrontal activity emerged when data were analyzed on the basis of childhood history, combining the persistent and remitted groups in a contrast against never affected controls. For the striatum, the inverse pattern held. Here, there were no differences between groups defined on the basis of adult outcome. Instead, differences emerged when data were analyzed on the basis of childhood history. Thus, those who had ADHD as children showed reduced right caudate activity when successfully inhibiting and increased activity in the right caudate when failing to inhibit. Thus, at a striatal but not cortical level, anomalies reflected a childhood history of ADHD rather than current adult status. In whole brain analyses unconstrained by a priori regions of interest, cerebellar activity during successful inhibition reflected adult outcome, an unpredicted but interesting finding we discuss later.

The exquisite temporal resolution of MEG was then used to further parse cortical inhibitory processes. These analyses were conducted at a whole brain level using appropriate adjustments for multiple testing. An association was found between adult symptoms of hyperactivity-impulsivity and neural responses during successful inhibition in the right inferior prefrontal cortex only. Those with persistent symptoms showed less theta band activity in this region than the remitters. This group difference was confined to a time window of 300 to 350 milliseconds after the stop signal, which covers the time period during which inhibition usually occurs. In short, using MEG it was shown that during an inhibitory time window, individuals with persistent ADHD had decreased theta activity compared to remitters within the right inferior frontal cortex. This study controlled for scanner artifacts, removing data that showed traces of motion, ensured that the outcome groups did not differ in the amount of residual motion, and also considered residual motion in analyses. The findings also held after controlling for comorbidities and psychostimulant treatment.

In summary, three studies of response inhibition find that adult remission is tied to patterns of brain activation, particularly at cortical and cerebellar levels, that resemble those seen in the never affected, supporting the model of remission as ‘convergence/normalization’. Subcortical activity may reflect childhood history rather than adult status.

Other studies are more in line with the concept of neural compensatory processes underpinning remission, with the recruitment of novel regions during cognitive control. One study used resting state fMRI data to extract a network that closely resembles the cognitive control network (Francx et al., 2015a). Participants had been followed clinically from ages 11 to 18, during which time many showed symptomatic improvement although very few attained remission (only 19 of the 129 [15%] with childhood ADHD had remitted by the study’s endpoint). Thus, the study examines a relatively early phase of symptomatic improvement rather than remission. It found that the degree of improvement in symptoms of hyperactivityimpulsivity was associated with increased connectivity within frontal regions of the cognitive control network. Considered categorically, those who showed any degree of symptomatic improvement had significantly higher connectivity than the never affected, suggesting that stronger integration of anterior regions of the executive control network might underpin improvement. The study thus provides support for the concept that remission is tied to the emergence of ‘atypical’ but beneficial brain activation, here localizing to the cognitive control network, defined using resting state fMRI.

A final set of studies used electrophysiological mapping and found that remitters show activation patterns during cognitive challenge that fell between those seen in persistent ADHD and never affected controls (Cheung et al., 2016; Michelini et al., 2016). Two tasks involving cognitive control were performed by 23 remitters and 87 individuals with persistent ADHD, at a mean age of 19 years, and brain activity was measured using EEG. Indices of cognitive control were extracted, such as parietal activation occurring shortly after a no-go trial (the nogo-P3 amplitude), and conflict monitoring activity following an incongruent stimulus in the flanker test (the N2 signal). In most of these ‘cognitive’ electrophysiological indices, the remitters fell between the persistent ADHD and never affected groups and did not differ significantly from either. As will be discussed later, remitters in this study did however resemble the never affected on measures reflecting ‘bottom-up’ rather than ‘top-down’ processes.

In summary, studies directly probing ‘top-down’ cognitive processes find that remitters closely resemble the never affected, in line with the concept of remission as ‘normalization’. Other studies find that remitters exhibit unique, possibly compensatory features within the cognitive control network. Electrophysiological studies find remitters show neural activation that lies between that seen in persisters and never affected. It is easy to see how these different findings could arise from the differences in imaging modalities (fMRI vs. MEG/EEG), paradigms (e.g. task-free vs task-driven) and analytic approaches. But can we partly reconcile these seemingly disparate findings? One explanation might reside in the age range of the participants in each study. Notably, studies of remission in adults find neural activation during ‘top-down’ processes does not differ from never affected controls. By contrast, studies of remission in adolescence that this age group occupies an intermediate position between persisters and the never affected. Thus, it is possible that that remitters in the adolescent studies were ‘on the way’ to normalization, which they might attain if followed into adulthood.

If remission is underpinned by the ‘normalization’ of neural networks underpinning ‘top-down’ processes, such as cognitive control, then we might expect that the structural backbone of these networks, made up of white matter tracts, to show a similar pattern. White matter tracts connecting regions of the attentional and cognitive control network include the superior longitudinal, the inferior frontooccipital and some divisions of the uncinate fasciculi. Diffusion tensor imaging (DTI) can provide measures of the microstructure of these white matter tracts, such as fractional anisotropy and more focused measures of diffusion parallel (axial diffusivity) and perpendicular (radial diffusivity) to axons (Basser and Pierpaoli, 1996). Microstructural anomalies in white matter tracts have been reported in some of these tracts in adolescents and adults with ADHD (Casey et al., 2007; Chuang et al., 2013; Dramsdahl et al., 2012; Helpern et al., 2011; Konrad et al., 2010; Onnink et al., 2015). Here we focus here on studies that examine white matter tract microstructure in adult remitters.

In one DTI study, 43 young adults with remitted ADHD showed fractional anisotropy in major white matter tracts that did not differ significantly from a never-affected comparison group (Sudre et al., 2015). By contrast, 32 adults with persistent ADHD had reduced fractional anisotropy in the bilateral uncinate and the right inferior fronto-occipital fasciculi. Findings held when those regularly taking psychostimulants were excluded and held when parameters reflecting residual motion were entered as covariates. Such findings are in keeping with the model of remission as ‘normalization’, as reduced fractional anisotropy in these regions has been reported in studies of childhood ADHD (van Ewijk et al., 2012). Similar findings emerged in a DTI study of adolescents who showed symptomatic improvement (Francx et al., 2015b). In this study, remitters did not differ from controls in the microstructural properties of some tracts linked with cognitive control network- such as the superior longitudinal fasciculus - but did differ significantly from those with persistent ADHD. Combined these DTI studies find similar white matter microstructure in remitters and the never affected, compatible with the model of remission as ‘convergence/normalization’.

The finding of typical white matter microstructure in remission receives more mixed support from a cohort of children followed into their 40s (Cortese et al., 2013). DTI data were acquired on 15 of these older adults with persistent ADHD, 25 with remitted ADHD and 66 never affected controls. When data were analyzed on the basis of childhood diagnosis, combining those with persistent and remitted ADHD, decreases relative to never affected controls were found in the fractional anisotropy of several projection and cortico-cortical association tracts. When data were parsed by adult outcome in a contrast of the remitted vs persistent ADHD groups, no significant differences were found. Nonetheless some qualitative differences emerged between the outcome groups. Thus, the remitted group showed less extensive changes in fractional anisotropy than the persistent group, compatible with some degree of convergence towards more typical microstructure in tandem with clinical improvement.

A further DTI study contrasted never affected controls against 30 young adults at a mean age of 22, all of whom had remitted from ADHD diagnosed around age 16 (Bode et al., 2015). It found increased fractional anisotropy in the remitted group driven by changes in radial diffusivity in the left forceps minor, which connects lateral and medial surfaces of the frontal cortex via the genu of the corpus callosum. The finding of increased fractional anisotropy in remitters is at odds with other DTI studies and may reflect cohort characteristics. For example, this study found remission in all patients between the ages of 16 and 22, whereas most studies find much lower rates of syndromic remission during this age window. The lack of a group with persistent ADHD in this study also means it is unclear if these changes reflect a fixed anomaly which would be shared with a persistent ADHD group, or a compensatory process unique to the remitted group.

Overall, the structural connectivity studies give mixed support to concept of remission as tied to typical white matter microstructural features. While such ‘normalization’ was found at the level of entire tracts, more focal anomalies, detected using ‘voxel’ rather than tract level measures, were detected among both those who remit and persist, likely reflecting fixed childhood deficits.

Finally, we consider neuroanatomic imaging, a modality that has been in use for decades. One study prospectively acquired neuroanatomic images in tandem with clinical assessments from childhood into adulthood on a cohort of 92 children with ADHD and 184 never affected controls (Shaw et al., 2013). A link with adult symptom outcome emerged in the cingulate cortex bilaterally, the right inferior parietal, right dorsolateral prefrontal and left sensorimotor cortex. In these regions, increasing severity of adult inattention but not hyperactivity-impulsivity was linked with higher rates of cortical thinning during adolescence. The net effect of these differences was that those whose ADHD symptoms improved showed a significant convergence during adolescence towards typical cortical dimensions, rectifying early anomalies. By contrast, those with persisting symptoms had childhood cortical anomalies that persisted into adulthood. Notably, many of these regions where cortical trajectories were tied to outcome - the dorsolateral prefrontal cortex and parietal lobule - are components of the cognitive control network. The findings held when those with comorbid disorders or on regular psychostimulant medication were excluded. A link to parietal trajectories was also reported in an earlier report on this cohort into the cortical correlates of early adolescent outcome (mean age of 15 years) with a shorter follow-up period of 6 years (Shaw et al., 2006). In short, a prospective study that yoked clinical and neuroanatomic observations provides evidence that ‘normalization’ of the cortical anatomy of the cognitive control network may occur in tandem with symptom remission.

In summary, studies of cortical/cerebellar activity during ‘top-down processes’, white matter microstructure and a prospective anatomic imaging study are all compatible with the concept of remission as the endpoint of processes that ‘normalize’ early neural anomalies. Subcortical activity during ‘top-down’ processes may show anomalies that have been fixed since childhood. Finally, evidence suggestive of compensation/neural reorganization comes from a report of novel patterns of intrinsic connectivity in remitters within the cognitive control network.

b. Remission and the neural substrates of ‘bottom-up’ cognitive processes

While ADHD is usually linked with deficits in executive function, anomalies in ‘bottom-up’ cognitive processes including vigilance, the preparation of responses and reward processing have also been implicated in the onset of ADHD (Castellanos et al., 2006; Huang-Pollock et al., 2012; Jackson and MacKillop, 2016; Kofler et al., 2013; Sergeant, 2005). Might the clinical outcome of ADHD be linked to changes in these ‘bottom-up’ processes? There is limited support at the behavioral level, with most studies finding little predictive value for ‘bottom-up’ behavioral measures (van Lieshout et al., 2013). Here, however, we focus on neural processes.

The first study, mentioned earlier, used EEG and ERP to characterize vigilance and response preparation during two cognitive tasks (Cheung et al., 2016; Michelini et al., 2016). Electrophysiological measures of ‘bottom-up’ processes were extracted, including the contingent negative variation (the cerebral potential that follows a warning stimulus preparing the individual to respond to an imperative stimulus) and indices of error processing. Remitters did not differ from the never affected group on these measures of preparation-vigilance whereas those with persistent ADHD showed consistent deficits. Such findings are compatible with the concept of normalization of ‘bottom-up’ processes playing a role in remission. The findings stand in contrast to the results of ‘top-down’ cognitive indices obtained in the same experiment, in which the remitters generally occupied a position intermediate to the never affected and persistent groups.

The neural substrate of response preparation - a ‘bottom-up’ process- has also been parsed using fMRI which allows a more detailed characterization of subcortical processes than EEG and MEG. One such study found atypical hypo-activation in thalamo-striato-prefrontal cortical regions during response preparation in both remitters (N=19) and persisters (N=16) (Clerkin et al., 2013). Only one difference emerged that reflected adult outcome: the remitted group showed increased functional connectivity between the thalamus and the prefrontal cortex during the preparation of responses to a cue, compared to both persisters and the never affected. This is compatible with a compensatory model.

Anomalous reward processing has frequently been implicated in ADHD and one group asked if it also relates to adult outcomes (Wetterling et al., 2015). Using fMRI, the study reported similar brain activation patterns between remitters and never affected. For example, both groups showed coordinated activity of the medial prefrontal and striatal regions during the processing of denied rewards. By contrast, the persistent group showed an absence of such coordinated activity. While all groups showed decreased activation following punishment, this localized to slightly different areas of the prefrontal cortex for remitters and the never affected, compatible with compensatory activity in remitters.

Do the findings from other imaging modalities support the concept of ‘bottom-up’ contributors to remission? Pertinent findings come from a study mentioned earlier that imaged a large group of adult males (mean age 41 years) with a history of childhood ADHD (diagnosed at age 8)(Proal et al., 2011). There was a globally thinner cortex in both the adult persistent and remitted ADHD groups compared to controls, particularly in parietal, temporal and frontopolar regions and the precuneus. The ADHD persisters and remitters mostly did not differ significantly from one another, compatible with fixed ‘trait’ deficits possibly carried forward from childhood. However, there were less prominent changes that did reflect outcome in brain regions that support ‘bottom up’ processes. Thus, in the limbic cortex and the thalamus, there was a trend toward lower grey matter density in the persistent compared to both the remitted and never affected groups, who did not differ significantly from one another. This finding is compatible with’normalization/convergence’ model in regions supporting ‘bottom-up’ processes.

Finally, we consider the cerebellum. The cerebellum is a pivotal component of the neural circuitry underlying both ‘top-down’ cognitive functions, such as working memory, response inhibition, attention shifting, and also ‘bottom-up’ processes such as error monitoring, the processing of rewards and temporal information (Durston et al., 2011; Noreika et al., 2013; Strick et al., 2009; Toplak et al., 2006). Two findings point to the cerebellum as a possible contributor to remission in ADHD. First, links with adult outcome emerge from the fMRI study of response inhibition discussed earlier, which found cerebellar activity during successful inhibition to reflect adult outcome (Szekely et al., 2017). Thus, remitters showed typical activation and those with persistent ADHD showed hypo-activation. Secondly, a prospective neuroanatomic imaging study found that a group of children with ADHD showed some trajectories were sensitive to outcome, with convergence to typical dimensions in the anterior hemisphere in remitters and divergence in the posterior hemispheres among persisters. The study however also reported some ‘fixed’ anatomic anomalies in other cerebellar regions- specifically the superior vermis - that persisted regardless of outcome (Mackie et al., 2007). This study complements a voxel based morphometric study that showed similar cerebellum volumes in remitters and the never affected, as opposed to volume loss in persisters (Proal et al., 2011).

In summary, there is emerging evidence, mainly from electrophysiological mapping, that remission may be associated with typical cerebral cortical activity during ‘bottom-up’ cognitive processes. Functional MRI gives a mixed picture, with some studies suggesting subcortical activity during ‘bottom-up’ processing may reflect fixed anomalies carried from childhood, and yet other studies suggest that compensatory processes may be tied to remission, particularly during response preparation and reward processing.

c. Remission and the default mode network

Finally, we consider the role of the default mode network (DMN) in remission. Intrusions of the DMN – prominent during introspective, task-free processing- into task-oriented processing has been implicated in ADHD (Fassbender et al., 2009; Rosenberg et al., 2015; Wen et al., 2012). Might change in the DMN, or its interactions with other networks underpin remission? The first study to address this question examined the DMN in 35 adults with a childhood history of ADHD and 17 controls (Mattfeld et al., 2014). It found anomalies of the DMN to be confined to those with persistent ADHD; this group showed decreased functional connectivity between posterior and anterior midline regions of the DMN. By contrast, the DMN of those who had remitted as adults did not differ from never affected individuals. This study also found that the typical counterbalance between the default mode network and the cognitive control network was lost in both remitted and persistent ADHD. Findings held when subjects taking psychostimulants were excluded. In short, anomalies within the default mode network itself reflected adult outcome, with an essentially typical DMN in remitters. By contrast, the cross-talk between different networks, defined using resting state fMRI, was abnormal, regardless of adult outcome.

A second study examined 205 young adults (Sudre et al., 2017). Using fMRI and MEG during task free processing, stable patterns of connectivity were identified that reflected functional connections within and between the brain’s major networks, including the DMN, cognitive control and attention networks. Several of these stable connectivity patterns were associated with the severity of symptoms of adult inattention (but not hyperactivity-impulsivity) that had persisted from childhood. Specifically, adult inattention persisting from childhood was associated with a partial loss of the balance of connectivity within the DMN and connectivity between the DMN and other networks. At a categorical level, those with persistent ADHD differed significantly in these connectivity patterns from both the remitted and never-affected groups. By contrast, the remitted and never-affected groups were remarkably similar in the patterns of within and between network connectivity. The findings cannot be attributed to motion, as data showing more than minimal motion were excluded and results held when parameters reflecting residual motion were entered in all analyses. The findings also held after controlling for psychostimulant medication histories.

A central role for the DMN in remission might also imply that its underlying structural basis would similarly reflect outcome. Indeed, the only prospective neuroanatomic study discussed earlier found cortical dimensions of the posterior cingulate, a key hub of the DMN, converge to the typical range in tandem with symptom improvement (Buckner et al., 2008; Shaw et al., 2013). What of the white matter tracts underlying the DMN? These include the cingulate bundles, connecting the posterior to anterior cingulate, and to the parieto-temporal cortex (Teipel et al., 2010). However, studies of adult remission have either not examined the cingulate bundle or found its microstructure did not reflect outcome (Bode et al., 2015; Cortese et al., 2013; Sudre et al., 2015).

A preliminary picture emerges: the default mode network per se appears atypical among those with ADHD symptoms persisting from childhood but has a more typical architecture among those who have remitted. Interactions between the DMN and task positive networks, when defined using methods with exquisite temporal resolution, if not by fMRI, show a similar pattern; namely remitters do not differ from the never affected. Further examination of the structural connections of the DMN are warranted.

Conclusion.

We provide an overview of the cognitive and developmental models that might explain remission from childhood ADHD. Many studies converge to find that neural features in adult who have remitted from childhood ADHD are essentially indistinguishable from those who were never affected. Functionally, this was found for prefrontal cortical function under cognitive challenge, cortical electrophysiological activity during response preparation/vigilance and intrinsic functional connectivity centered on the default mode network. A similar theme held for some white matter tract microstructural properties and the anatomic development of fronto-parietal and posterior cingulate regions. By contrast, others report atypical neural features- particularly posterior cerebral structure and thalamo-striatal function- in both remitters and persisters. These findings are more compatible with the model of ‘fixed anomalies’ that reflect the onset of childhood ADHD and persist into adulthood regardless of outcome. There is overall less evidence for neural re-organization, compensating for ADHD symptoms, as a mechanism that leads to remission. These initial comments are based on our synthesis of the literature. Quantitative meta-analyses were not feasible as data are too sparse. For example, the task-based imaging studies all used different paradigms and thus are not amenable to meta-analysis. In the studies of resting state, only one study reported on the entire brain, and the other two studies reported only on networks of interest (the cognitive control network in one study and the DMN in the other). Thus again, a quantitative analysis is not feasible, this may become possible as more data are reported.

Some insights pertinent to these models of remission might be gained from contrasts of studies into ADHD at different life stages. For example, studies of preschool children likely include highly symptomatic participants, given the high threshold for diagnosis in this age range. By contrast, studies of school age children with ADHD will likely include some who are already showing improvement, particularly in hyperactivity-impulsivity (Faraone et al., 2006), and adolescent studies will likely include many who are firmly on a remitting trajectory (but not yet attained categorical remission). In this regard, a recent multi-site, age-stratified mega-analysis of 3242 participants is particularly informative (Hoogman et al., 2017). In this study, children under 12 showed significant, albeit modest reductions in the volumes of subcortical structures; volume loss was marginal in the adolescent age group and was non-significant in adult groups. A similar pattern can be discerned from key individual studies. For example, a large study of pre-school age children found that ADHD-related volume reductions were associated with medium to large effect sizes as opposed to the smaller effects seen in studies of adolescents (Greven et al., 2015; Rosch et al., 2018). In short, if we accept the premise that the proportion of ADHD subjects showing symptomatic improvement (but not yet attaining remission) will be higher in adolescent compared to younger groups, then the cross-sectional studies are perhaps compatible with the concept of early remission as convergence towards typical dimensions.

The models of normalization, compensation, or fixed anomalies might also be used to help explain patterns of comorbidity with ADHD across the lifespan. The core symptom dimensions of ADHD- inattention, impulsivity, hyperactivity- are not confined to the disorder, but present in many other disorders; for example, the inattention seen in anxiety disorders, and the impulsivity of conduct disorder and substance misuse. If inattention and its underlying cognitive anomalies remit due to ‘normalization’, then we would expect improvement to extend beyond ADHD to disorders in which inattention is a prominent feature. Conversely, persistent anomalies in the neurocognitive substrates of impulsivity could not only contribute to the persistence of impulsivity in ADHD but also the emergence of substance misuse in adolescence and beyond.

While impressive initial advances have been made, research into remission is still at an early stage and no definitive conclusions can be made: data are limited and there are some inconsistencies. Thus, there is no clear winner between the models. However, perhaps this ‘competition’ is more apparent than real. Just as we need to invoke multiple cognitive or neural pathways to explain the onset of ADHD (Castellanos and Proal, 2012; Durston et al., 2011; Johnson et al., 2009; Sergeant, 2005; Sonuga-Barke, 2002) so there may be multiple pathways to remission. Indeed, perhaps each individual’s pathway to remission might represent the ‘undoing’ of his/her idiosyncratic neurocognitive deficits. Secondly, the models are not mutually exclusive, particularly as several studies find that different developmental processes unfold in different brain regions (Clerkin et al., 2013; Halperin and Schulz, 2006; Szekely et al., 2017).

As mentioned earlier, ADHD is a highly heterogeneous disorder. Part of this heterogeneity is reflected in DSM by the recognition of predominately inattentive, predominately hyperactive-impulsive and combined presentations (previously called subtypes). Only a few studies of adult outcome have considered these different presentations, and the central findings are summarized in Table 3. It appears that different paradigms emphasize different cognitive mechanisms and thus associate with different symptom domains. For example, response inhibition is usually linked with impulsive symptoms; accordingly, imaging studies mostly find that anomalous inhibitory neural processes are more strongly tied to the persistence of hyperactive-impulsive than inattentive symptoms (Schneider et al., 2010; Szekely et al., 2017). By contrast, DMN anomalies have been linked to behavioral attentional lapses, and thus it is notable that DMN anomalies are mostly associated with the persistence of inattentive but not hyperactiveimpulsive symptoms (Sudre et al., 2017). Anomalies in white matter microstructure have been associated with hyperactive-impulsive symptoms in a study of late adolescence (Francx et al., 2015b) and with presence of inattentive symptoms in a study of young adults (Sudre et al., 2015). These different findings might reflect the developmental trajectories of each symptom domain, as symptoms of hyperactivity-impulsivity are usually more prominent in adolescence than adulthood whereas inattention persists more into adulthood (Faraone et al., 2006). Future advances would benefit from a consistent reporting of associations with each symptom domains as well as categorical contrasts.

We alluded earlier to a major limitation of imaging data acquired in adulthood only. When adult remitters resemble a never affected comparison group this could result from either convergence towards more typical structure/function or represents more typical childhood features that are simply carried forward into adulthood. To address this issue, prospective studies are needed that repeatedly acquire imaging in tandem with clinical assessments. This approach has been elegantly used to parse the links between the longitudinal course of ADHD symptoms and cognitive skills (Karalunas et al., 2017) and could readily be extended to the brain. Even stronger inferences about mechanisms underpinning symptom improvement could be drawn from neuroimaging within the context of randomized controlled trials: an exceptionally challenging but rewarding design (Ishii-Takahashi et al., 2015; Schrantee et al., 2016).

The field is well-positioned to tackle the complex issue of explaining remission. Increased awareness of adult ADHD has fostered advances in concepts of remission. Additionally, multi-site collaborations have been formed that can attain the large sample sizes that will be needed to parse the heterogeneous neural mechanisms underlying remission.

Highlights.

• We review neuroimaging studies of adults who have remitted from childhood ADHD

• Most studies find adult remitters show neural features indistinguishable from controls

• Fewer studies find that remission arises due to compensatory neural re-organization

• Some deep brain anomalies may persist from childhood even in those who remit