Shared and Specific Neural Correlates of Attention Deficit Hyperactivity Disorder and Autism Spectrum Disorder: A Meta-Analysis of 243 Task-Based Functional MRI Studies

Abstract

Objective:

To investigate shared and specific neural correlates of cognitive functions in attention deficit hyperactivity disorder (ADHD) and autism spectrum disorder (ASD), the authors performed a comprehensive meta-analysis and considered a balanced set of neuropsychological tasks across the two disorders.

Methods:

A broad set of electronic databases was searched up to December 4, 2022, for task-based functional MRI studies investigating differences between individuals with ADHD or ASD and typically developing control subjects. Spatial coordinates of brain loci differing significantly between case and control subjects were extracted. To avoid potential diagnosis-driven selection bias of cognitive tasks, the tasks were grouped according to the Research Domain Criteria framework, and stratified sampling was used to match cognitive component profiles. Activation likelihood estimation was used for the meta-analysis.

Results:

After screening 20,756 potentially relevant references, a meta-analysis of 243 studies was performed, which included 3,084 participants with ADHD (676 females), 2,654 participants with ASD (292 females), and 6,795 control subjects (1,909 females). ASD and ADHD showed shared greater activations in the lingual and rectal gyri and shared lower activations in regions including the middle frontal gyrus, the parahippocampal gyrus, and the insula. By contrast, there were ASD-specific greater and lower activations in regions including the left middle temporal gyrus and the left middle frontal gyrus, respectively, and ADHD-specific greater and lower activations in the amygdala and the global pallidus, respectively.

Conclusions:

Although ASD and ADHD showed both shared and disorder-specific standardized neural activations, disorder-specific activations were more prominent than shared ones. Functional brain differences between ADHD and ASD are more likely to reflect diagnosis-related pathophysiology than bias from the selection of specific neuropsychological tasks.

Attention deficit hyperactivity disorder (ADHD) and autism spectrum disorder (ASD) are the two most commonly diagnosed neurodevelopmental disorders. ADHD is characterized by impaired attention, hyperactivity-impulsivity, or both that are inconsistent with developmental level (1). The core features of ASD are difficulties in social communication along with restricted interest and repetitive behaviors (1). Although ADHD and ASD diagnostic criteria are eminently different, people with ADHD often present ASD symptoms and vice versa (2, 3). In addition to the clinical overlap, ADHD and ASD share genetic underpinning, including rare variants and single-nucleotide polymorphisms (4, 5).

To clarify these clinical observations and genetic findings, numerous task-based functional MRI (fMRI) studies have been conducted to identify the neural correlates of ADHD and ASD core symptom dimensions. Reflecting distinct diagnostic criteria, the conceptualization of the pathophysiology of ADHD and ASD has driven the selection of tasks in most of the available task-based fMRI studies (6, 7). Specifically, task-based fMRI studies of ADHD have typically used attention and inhibition tasks (8), whereas studies of ASD have frequently relied on emotional face recognition tasks (7). Meta-analytic synthesis of task-based fMRI studies of ADHD has found atypical activation in the right dorsolateral prefrontal cortex, the left putamen, and the globus pallidus during attention tasks, with the insula and the anterior cingulate cortex (ACC) implicated in inhibition (9). In relation to ASD, meta-analytic evidence has found that atypical activation in the superior temporal gyrus and the fusiform gyrus is associated with difficulties in social interaction (7, 10–12). Restricted interest and repetitive behaviors have often been interpreted to reflect deficiencies in cognitive flexibility, which have been associated with the inferior parietal gyrus in task-based fMRI studies (13). Although these analyses provide insights into the neural correlates likely to be associated with the typical symptoms of each disorder, they are limited by their a priori selection of specific tasks for ADHD and ASD.

In contrast to diagnosis-driven research constrained by diagnostic criteria, the Research Domain Criteria (RDoC) framework organizes mental health research in functional domains, including cognitive systems, positive and negative valence systems, social processes, arousal and regulatory systems, and sensorimotor systems. In keeping with the RDoC framework, studies have begun to incorporate neuropsychological tasks related to transdiagnostic symptom domains to determine whether neural correlates are distinct or shared in ADHD and ASD (14, 15). Specifically, attention and reward-processing tasks have been administered to individuals with ASD (16, 17), and social tasks have been used to study individuals with ADHD (18, 19). Specific studies have found lower activation in the dorsolateral prefrontal cortex during attention tasks in individuals with ASD (20) and atypical processing in the fusiform gyrus evoked by watching emotional faces in individuals with ADHD (21). These results suggest that some of the neural bases of ADHD symptoms among people with ASD are similar to those of people with ADHD and vice versa. However, results from task-based fMRI studies have been mixed. Thus, the extent to which brain-behavior relationships (i.e., brain regions underpinning behavioral symptoms and cognition) in ADHD and ASD are shared or distinct remains unclear. Gaining insight into these relationships would inform our understanding of the pathophysiology of these two common disorders, with implications for their conceptualization, future diagnostic classifications, and management strategies. Indeed, this understanding would be relevant to the ongoing debate as to whether the current diagnostic categories for neurodevelopmental disorders should be lumped together or split apart (22).

A meta-analysis by Lukito et al. (23) addressed the specificity of the neural underpinnings of cognitive control in ADHD and ASD, examining shared and distinct functional abnormalities in individuals with one or the other of these two conditions. The authors found differences in neural activation during cognitive control tasks between individuals with ADHD and those with ASD. Lower-than-typical activation in the ACC and the middle frontal gyrus and greater-than-typical activation in the precuneus and the fusiform gyrus were specific to ASD. By contrast, no atypical activation was specific to ADHD compared with ASD. Although the work by Lukito et al. provided important novel insight by comparing ADHD with ASD, it focused on only one of six RDoC cognitive domains. There are additional deficits in other cognitive domains that overlap clinically among individuals with ADHD and those with ASD. However, the task-based fMRI literature on these developmental disorders is strongly biased in the selection of neuropsychological tasks. Hence, based on available evidence synthesis in the field, we cannot validly infer commonalities and differences. A more comprehensive synthesis of the neural profiles of individuals with ASD and those with ADHD across studies with common tasks in a sample with ADHD and one with ASD would further our understanding of the pathophysiology of these disorders.

To address this gap, we performed a meta-analysis of task-based fMRI studies that included either or both individuals with ADHD and those with ASD using data sets that were balanced to include a similar number of studies per domain and task across the two disorders. Given the exploratory nature of the meta-analysis, we did not formulate any a priori hypotheses.

METHODS

The study protocol was preregistered on PROSPERO (CRD42021283877).

Search

The search involved a combination of terms for the diagnosis and terms for the imaging modality. The terms for the diagnosis included either “ADHD” or “autism” and related terms, such as “hyperkinetic” and “Asperger,” as well as related terms using wildcards. The terms for the modality included “neuroimaging” and “functional magnetic imaging” but did not specify “task” because this would have missed relevant studies. Specific search terms and syntax are reported in Figure S1 in the online supplement. PubMed (MEDLINE), Ovid databases (PsycInfo, Embase and Embase Classic, and Ovid MEDLINE), and Web of Knowledge (Web of Science [Science Citation Index Expanded], Biological Abstracts, BIOSIS, and Food Science and Technology Abstracts) were searched without language restriction from inception to December 4, 2022. We also hand-searched references from included studies and reviews to identify any eligible study that was missed in the electronic search.

Inclusion and Exclusion Criteria, Screening, and Data Extraction

All articles included in the meta-analysis had to meet the following inclusion criteria: it was an original study; used task-based fMRI; assessed differences in blood-oxygen-level-dependent signals in an ADHD group versus a typically developing control group or in ASD versus control groups; recruited children or adults with a categorical diagnosis of ADHD or ASD (and their equivalent constructs), according to DSM-III to DSM-5 or ICD-10 or an earlier version; and included individuals of any age and both sexes, recruited from clinical settings of the community, with or without ongoing or previous pharmacological treatment for ADHD- or ASD-related problems.

Articles meeting any of the following criteria were excluded: compared only ADHD versus ASD; recruited participants whose symptoms partially remitted and no longer fulfilled the diagnostic criteria; assessed only ADHD symptoms, without establishing a categorical diagnosis; recruited participants with a diagnosis of minimal brain dysfunction or with the deficit in attention, motor control, and perception syndrome (24); and assessed only ASD symptoms without establishing a formal diagnosis (e.g., diagnoses based only on the Autism Spectrum Quotient or the Autism Diagnostic Observation Schedule).

In the first stage of screening, two authors (H.T. and J.F.) independently screened titles and abstracts for inclusion. A final list was agreed on, with discrepancies resolved by consensus between the two authors. When consensus was not reached, a third senior author arbitrated (Y.Y.A. or T.I.). If any doubt about inclusion remained, the article proceeded to the next stage. The full-text version of the articles that passed the first stage of screening was assessed for eligibility by two researchers independently. Discrepancies were resolved by consensus between the two authors, and if needed, a third researcher acted as an arbitrator. Two researchers independently extracted demographic, clinical, and neuroimaging data. Where required, we contacted the corresponding author via e-mail with questions regarding study eligibility or with a request for additional data that were needed for us to include their study in the meta-analysis.

Labeling of Neuropsychological Tasks

Two authors (H.T. and F.J.) extracted information on the neuropsychological tasks used in each study. Another author (Y.Y.A.) assigned the neuropsychological tasks to subconstructs in the corresponding latest version of RDoC (25). Another author (T.I.) confirmed the classification. Each domain contains corresponding constructs and subconstructs, as shown in Table 1. For the meta-analysis, we extracted coordinates of loci in which the blood-oxygen-level-dependent response differed significantly between the clinical and control groups.

TABLE 1.

Details of psychological tasks and direction of findings of the included studies^a

	Number of Experiments
RDoC: Domain and Construct	ASD > Control	Control > ASD	ADHD > Control	Control > ADHD
Negative valence systems
Acute threat (“fear”)	0	0	1	2
Potential threat (anxiety)	0	0	0	0
Sustained threat	0	1	0	1
Loss	0	0	1	2
Frustrative nonreward	0	0	0	0
Positive valence systems
Reward responsiveness	0	3	11	11
Reward learning	1	1	2	2
Reward valuation	2	3	10	9
Cognitive systems
Attention	9	9	11	10
Perception	22	30	4	10
Declarative memory	2	9	0	1
Language	9	11	1	0
Cognitive control	18	14	44	74
Working memory	4	7	18	29
Social processes
Affiliation and attachment	0	2	0	0
Perception and understanding of self	2	7	0	0
Perception and understanding of others	13	32	2	4
Social communication	18	33	2	2
Arousal and regulatory systems
Arousal	1	1	1	0
Circadian rhythms	0	0	0	0
Sleep-wakefulness	0	0	0	0
Sensorimotor systems
Motor actions	5	6	0	4
Agency and ownership	0	0	0	0
Habit sensorimotor	0	0	0	0
Innate motor patterns	1	1	0	1

^aADHD=attention deficit hyperactivity disorder; ASD=autism spectrum disorder; RDoC=Research Domain Criteria.

Stratified Sampling

Balanced subsampling is a sampling method to create a subsample that ensures matching in key parameters when the whole sample differs in the distribution of key parameters between groups. We adopted a stratified subsampling method to balance the distribution of neuropsychological tasks between ADHD and ASD. To conduct stratified subsampling (diagnosis and direction of effect), we first focused on subconstructs that contained two or more experiments in each of four categories: that is, ASD greater than control, ASD lower than control, ADHD greater than control, and ADHD lower than control. Studies that satisfied this criterion proceeded to stratified subsampling at the subconstruct level. Studies that did not proceed to stratified subsampling at the subconstruct level were aggregated at the construct level. Constructs with at least two studies in all four categories proceeded to stratified subsampling, and the other constructs were collapsed to the corresponding domain. The domains that did not have at least two studies in all four categories did not proceed to stratified subsampling.

Analysis

Next, we used a random balanced subsampling approach, ensuring that each sample contained an equal number of experiments from each domain. For each disorder (ADHD and ASD) and direction (greater and lower than control), we calculated the revised activation likelihood estimation (ALE) for 10,000 of these subsamples to approximate the true underlying effect. ALE identifies brain regions in which study results converge, taking into account the spatial uncertainty in the results of each study (26–29). Thus, we excluded studies that did not report significant group differences, because they do not contribute to spatial convergence. Results were thresholded at a p value of 0.05 using cluster-wise family-wise error rate correction, as per current standard (30). For each data set, the algorithm outputs a binary brain map indicating voxels for which significant convergence was found. We repeated the subsampling procedures 10,000 times, yielding 10,000 binary convergence maps for each direction and each diagnostic group. We then computed the average of all binary convergence maps per direction and diagnostic group to estimate the voxel-wise proportion of significant convergence over all subsamples. These procedures produced four maps (i.e., two directions for each of the two diagnostic groups), with each map yielding the probability of finding significant convergence in a balanced setting. We then performed conjunction analyses for each direction (i.e., greater and lower activations) across the two diagnostic groups. This was done by overlaying the greater- and lower-activation probability maps of each disorder and taking the minimum. In the resulting maps, areas with values greater than zero were interpreted as being shared across disorders. Studies that recruited both individuals with ADHD and individuals with ASD were included in both the ADHD and ASD analyses.

Follow-Up Analysis

To better understand the results of the main “subsampling ALE,” we ran multiple follow-up analyses. In the first analysis, we calculated the relative contribution of each RDoC subconstruct, construct, or domain to significant clusters, to assess specificity of the construct or domain to the cluster. To calculate the relative contribution, we first calculated the proportion of the modeled activation value per voxel in the significant clusters per experiment per sub-sample. We then averaged the proportion over all experiments belonging to a certain domain, which showed the relative contribution of each domain to each individual voxel of the cluster on a subsample level. Next, we averaged over subsamples, which showed the contribution on a voxel and construct or domain level. Lastly, we took the average over all voxels of the cluster, weighted by the probability of finding convergence, increasing the impact of frequently activated voxels on the final contribution.

The second analysis we ran was a meta-regression to assess the effect of age on the resulting clusters. This was necessary because we included studies with samples of children, adolescents, and adults. We limited the meta-regression to clusters surviving a 50% density-based thresholding procedure to limit inferences to the most stable results. We then fitted a linear regression model between the mean age in each experiment and a cluster’s total contribution based on modeled activation to the remaining clusters.

As a third follow-up analysis, we ran meta-analytic connectivity modeling and functional decoding on the clusters surviving the 50% density-based thresholding procedure (27, 29–31). Meta-analytic connectivity modeling is a meta-analytic approach to brain connectivity in which large neuroimaging databases are used to find regions in the brain showing significant convergence with a particular seed region. Here, the seed regions were the clusters found in the main analysis, and we used BrainMap as our database of choice (32, 33). Functional decoding uses the behavioral domain and paradigm class metadata categories from the BrainMap database to allow for functional characterization of resulting clusters (34, 35). This approach uses both forward and reverse inference. Forward inference assesses the probability of detecting activity in a certain brain region when a specific task is performed or cognitive process is present. Conversely, reverse inference reveals the probability of a specific task or cognitive process being present when there is activation in a particular brain region. Together, meta-analytic connectivity modeling and functional decoding allow for data-driven conclusions about the clusters resulting from the main “subsampling ALE.”

RESULTS

Included Studies

The initial search produced 32,069 hits (see Figure S1 in the online supplement for the PRISMA flowchart). After deduplication, 20,756 references were screened, and 370 studies were deemed potentially eligible for meta-analysis (see Table S1 and the list of references in the online supplement). From these, 243 studies, which included a total of 3,084 participants with ADHD (676 females), 2,654 participants with ASD (292 females), and 6,795 control subjects (1,909 females), reported data that could be meta-analyzed, including 113 studies that recruited individuals with ADHD, 122 that recruited individuals with ASD, and eight that recruited both individuals with ASD and individuals with ADHD. The included studies used a wide variety of psychological tasks during the scan. For example, among cognitive systems tasks, go/no-go and n-back tasks were used most frequently. These two tasks were used in seven and two ASD studies, respectively, and both of these tasks were used in 10 ADHD studies. ASD studies used visual and spatial tasks more frequently, and ADHD studies focused mainly on inhibition tasks. All the studies that recruited both individuals with ASD and those with ADHD had direct comparisons between the two diagnoses. Four studies showed only differences in brain activation between ASD and ADHD, and the other four studies showed both shared and different brain activations. In general, previous studies showed more differences than shared activations between ASD and ADHD. Although some empirical results were similar to the results of the meta-analysis, such as greater activation in the middle frontal gyrus and inferior frontal gyrus in ASD, most empirical findings were not concordant with the meta-analysis. Details of the included subconstructs, constructs, and domains are shown in Figure 1 (for details of the contrasts, see Table S2 in the online supplement). The remaining studies were not included because they did not report peak coordinates of group differences or did not show significant between-group differences.

FIGURE 1. Distribution of experiments across Research Domain Criteria (RDoC) domains and constructs based on greater or lower activation in autism spectrum disorder (ASD) and attention deficit disorder (ADHD) groups relative to a typically developing control group^a.

^a Percentages of experiments are visualized based on the directionality of the group comparison and categorized by RDoC domains and constructs. Panel A shows the percentages of experiments reporting greater activations in the ASD group compared with the control group, and panel B represents the percentages of experiments reporting lower activations in the ASD group. Panel C represents the percentages of experiments showing greater activations in the ADHD group relative to the control group, and panel D shows the percentages of experiments showing lower activations in the ADHD group. Additional details are provided in Table S2 in the online supplement.

Convergence of Results Across Tasks and Across ADHD and ASD Groups

Figure 2 shows the brain regions with greater and lower activation that were shared across the ADHD and ASD groups (for details, see Table S2 in the online supplement). In comparison to the control group, both the ADHD and ASD groups exhibited right-lateralized, greater activations in brain areas that included the lingual gyrus (Montreal Neurological Institute coordinates: x=2, y=−90, z=−2; cluster size [k]=53) and the rectal gyrus (x=6, y=44, z=−26; k=51). The regions with lower activation, however, were localized in the left hemisphere, particularly in the middle frontal gyrus (x=−40, y=34, z=26; k=107) and the superior temporal gyrus (x=−54, y=−30, z=−4; k=27) (Figure 2; for details, see Table S2 in the online supplement).

FIGURE 2. Shared greater and lower brain activations in autism spectrum disorder (ASD) and attention deficit hyperactivity disorder (ADHD) groups relative to a typically developing control group^a.

^a In panel A, a conjunction analysis of activation likelihood estimation (ALE) meta-analyses with a stratified sampling of psychological tasks identified brain regions showing shared greater activations in ASD and ADHD groups relative to the control group. In panel B, a conjunction analysis of ALE meta-analyses identified brain regions exhibiting shared lower activations in ASD and ADHD groups relative to the control group. The colors of boundaries represent the corresponding resting-state networks. Additional details are provided in Table S2 in the online supplement.

Convergence of Results Across Tasks in ASD-Specific and ADHD-Specific Activations

Compared with the control group, the ASD group exhibited greater activations in several cortical regions, including the left middle temporal gyrus (x=−46, y=−66, z=2; k=620) and the inferior parietal lobule (x=−40, y=−52, z=42; k=591), and subcortically in the right hippocampus (x=30, y=−36, z=4; k=105) and the left putamen (x=−26, y=−10, z=8; k=87). In contrast, the ASD group showed lower activations in cortical regions, such as the left middle frontal gyrus (x=−44, y=26, z=32; k=812) and the right middle temporal gyrus (x=52, y=−36, z=4; k=525), as well as in subcortical regions, including the left amygdala (x=−24, y=0, z=−12; k=760) and the right hippocampus (x=24, y=−4, z=−20; k=148). Figure 3 provides a visual representation of the greater and lower activations (for details, see Table S2 in the online supplement; for the unthresholded brain activation map, see Figure S2 in the online supplement).

FIGURE 3. Brain activations in the autism spectrum disorder (ASD) group relative to a typically developing control group^a.

^a In panel A, activation likelihood estimation (ALE) meta-analysis identified brain regions showing greater activations in the ASD group relative to the control group. In panel B, ALE meta-analysis identified brain regions showing lower activations in the ASD group relative to the control group. The colors of boundaries represent the corresponding resting-state networks. For the top five largest clusters, the contributions of each Research Domain Criteria subconstruct, construct, or domain were visualized (darker blue indicates higher contribution). Additional details are provided in Table S2 in the online supplement.

In comparison to the control group, the ADHD group demonstrated greater activations in cortical regions, such as the right insula (x=34, y=14, z=8; k=627) and the posterior cingulate cortex (x=6, y=−50, z=10; k=203), and in subcortical regions, including the right amygdala (x=−22, y=−2, z=−14; k=826) and the putamen (x=20, y=2, z=2; k=406). ADHD-related lower activations were distributed in cortical regions, such as the right middle temporal gyrus (x=60, y=−8, z=−12; k=417) and the left inferior frontal gyrus (x=−46, y=15, z=0; k=318), and in subcortical regions, including the right global pallidus (x=20, y=−8, z=−2; k=731) and the left thalamus (x=−8, y=−16, z=0; k=308). Figure 4 illustrates the greater and lower activations (for details, see Table S2 in the online supplement; for the unthresholded brain activation map, see Figure S2 in the online supplement).

FIGURE 4. Brain activations in the attention deficit hyperactivity disorder (ADHD) group relative to a typically developing control group^a.

^a In panel A, activation likelihood estimation (ALE) meta-analysis identified brain regions showing greater activations in the ADHD group relative to the control group. In panel B, ALE meta-analysis identified brain regions showing lower activations in the ADHD group relative to the control group. The colors of boundaries represent the corresponding resting-state networks. For the top five largest clusters, the contributions of each Research Domain Criteria subconstruct, construct, or domain were visualized (darker blue indicates higher contribution). Additional details are provided in Table S2 in the online supplement.

Contribution of RDoC Subconstructs, Constructs, and Domains to ASD- and ADHD-Specific Activations

With regard to ASD, the left middle temporal gyrus (x=−46, y=−66, z=2; k=620) and the right superior temporal pole (x=32, y=8, z=−28; k=111) clusters that showed greater activation in the ASD group than in the control group were predominantly associated with social processes and attention, respectively, and the cluster in the left middle frontal gyrus (x=−44, y=26, z=32; k=289) that showed lower activation in ASD consisted of a combination of positive valence, social processes, cognitive systems, visual perception, and response inhibition (Figure 3; for details, see Table S2 in the online supplement).

With regard to ADHD, the left amygdala cluster (x=−22, y=−2, z=−14; k=760) that showed greater activation in the ADHD group than in the control group was predominantly associated with cognitive systems, and the cluster in the right middle temporal gyrus (x=60, y=−8, z=−12; k=417) showing lower activation in ADHD was predominantly associated with both visual perception and response inhibition (Figure 4; for details, see Table S2 in the online supplement).

Meta-Regression

To examine age-dependent effects, we performed a meta-regression of age. As shown in Table S3 in the online supplement, no brain regions exhibited a significant association with age (all p values >0.06).

Functional Decoding and Meta-Analytic Connectivity Modeling

The ASD-related greater activation in the right inferior frontal gyrus (x=22, y=46, z=22) was associated with domains related to the emotional aspect of disgust and the cognitive aspect of music (see Figure S3A in the online supplement). The corresponding coactivated pattern comprised the ACC, the left and right inferior frontal gyri, and the left and right insular cortices (see Figure S3B in the online supplement). The second ASD-related greater activation, observed in the ACC (x=−2, y=36, z=24), was associated with domains related to the emotional aspects of valence and reward and gain (see Figure S4A in the online supplement). The corresponding coactivated pattern contained the left and right insular cortices, the posterior cingulate cortex, and the thalamus (see Figure S4B in the online supplement). The ASD-related lower activation in the ACC (x=0, y=38, z=20) was associated with the emotional aspect of reward and gain (see Figure S5A in the online supplement). The functional coactivated pattern of this cluster contained the left and right insular cortices, the posterior cingulate cortex, the thalamus, and the nucleus accumbens (see Figure S5B in the online supplement).

The ADHD-related greater activation in the left midbrain (x=−4, y=−34, z=−16) was associated with the emotional aspects of positive valence, fear, and reward and gain, the cognitive aspects of spatial cognition and reasoning, and the perceptual aspect of gustation (see Figure S6A in the online supplement). The coactivated pattern of this cluster involved the left and right insular cortices, the ACC, the thalamus, the right putamen, the left globus pallidus, and the left and right cerebellum (see Figure S6B in the online supplement). The second ADHD-related greater activation, observed in the left globus pallidus (x=−16, y=−4, z=−5), was associated with the interoceptive aspect of sexuality, the emotional aspects of disgust and reward and gain, the perceptual aspect of ol-faction, and the cognitive aspect of reasoning (see Figure S7A in the online supplement). The coactivated pattern of this cluster contained the left and right insular cortices, the ACC, the thalamus, the putamen, and the nucleus accumbens (see Figure S7B in the online supplement). The ADHD-related lower activation in the left superior temporal gyrus (x=−58, y=−22, z=6) was associated with domains related to the perceptual aspect of audition, the cognitive aspects of music, phonology, and speech, and the action aspect of speech (see Figure S8A in the online supplement). The corresponding coactivated pattern of this cluster comprised the left and right insular cortices, the ACC, the left and right superior temporal gyri, the thalamus, the putamen, and the cerebellum (see Figure S8B in the online supplement).

DISCUSSION

In this systematic review and meta-analysis focused on finding shared and distinct neural correlates in ADHD and ASD in task-based fMRI studies, we addressed, for the first time, bias related to diagnosis-driven selection of neuropsychological tasks by using a stratified sampling of psychological tasks. We identified both shared and disorder-specific convergence of results across tasks and across the two disorders. Overall, disorder-specific abnormalities were more prominent than shared ones. Specifically, relative to the typically developing control group, we found shared greater activations in the lingual and rectal gyri and lower activations in the middle frontal gyrus and superior temporal gyrus, regardless of task. By contrast, greater activations in the middle temporal gyrus and the ACC and lower activations in the middle frontal gyrus and middle temporal gyrus represented convergence of results across tasks specific to ASD, whereas greater activations in the insula and the posterior cingulate cortex and lower activations in the middle temporal gyrus and the inferior frontal gyrus represented convergence of results across tasks specific to ADHD.

Importantly, our findings are consistent with results of studies using other MRI modalities. For example, abnormalities in the superior temporal gyrus, the inferior frontal gyrus, and the middle frontal gyrus have been reported in both ASD and ADHD samples using voxel-based morphometry (31, 32) and resting-state fMRI (33, 34). The consistency with our results suggests that atypical activation in these brain regions may underlie atypical processing during psychological tasks, rather than being its consequence. However, cross-sectional studies cannot address causality. Future research will need to examine causal relations between atypical morphometry and activation during tasks and in the resting state, as well as in relation to symptoms.

Intriguingly, some of the shared abnormalities in, for example, the superior temporal gyrus, have been observed in other mental disorders, such as schizophrenia and bipolar disorder (35–37). Similarly, MRI studies have shown abnormal structure and metabolism in the middle frontal gyrus in schizophrenia and bipolar disorder (38, 39). Task-based fMRI studies have shown that the superior temporal gyrus and the middle frontal gyrus are crucial for attention and social cognition and for working memory, respectively (40). Given that these cognitive components are impaired nonspecifically across psychiatric disorders, abnormalities in these brain regions may reflect transdiagnostic vulnerability to impaired social functioning, rather than the neural basis of specific symptoms of one diagnosis.

Among the disorder-specific activations, both greater and lower activations in the ACC were observed in the ASD group. Meta-analyses of task-based fMRI studies of ASD have repeatedly reported atypical activation in the ACC in social and nonsocial tasks (41) and in reward processing tasks (8). Atypical ACC activation in ASD across neuropsychological tasks is in line with our results and suggests that the ACC represents a hub in the pathophysiology of ASD. In ADHD, we found greater activation in the insula. A previous meta-analysis of fMRI studies showed that the insula is a site of action of methylphenidate (42), suggesting that this region is a potential pathophysiological hub in ADHD. In addition, our meta-analysis showed the largest cluster with lower activation in the globus pallidus. Intriguingly, structural neuroimaging studies of ADHD have repeatedly shown abnormal volumes in the globus pallidus (32, 43). These findings suggest that our strategy of stratified sampling of neuropsychological tasks may have minimized the effects of tasks and emphasized the brain regions where abnormalities across tasks could be observed.

Our findings of disorder-specific abnormalities are at odds with the RDoC framework that posits brain-behavior relationships as being largely independent of clinical diagnoses (44). We consider two potential explanations for this unexpected difference. First, although we assigned psychological tasks to RDoC constructs, selection bias factors may have remained. For example, stimuli in ASD studies were typically voices or eyes, whereas ADHD studies used letters or colors (45, 46). Although the cues differed, they were assigned to cognitive systems as long as the contrast focused mainly on cognitive systems. Thus, selection bias may not have been fully eliminated despite our systematic efforts. Second, individuals with ASD and those with ADHD are often treated with different medications, which may produce different secondary symptoms. Because we prioritized stratified sampling to minimize task selection bias, we did not exclude studies in which participants were on medication. Differences in those factors might have influenced the results. Nevertheless, the available data suggest that what we detected reflects real differences in neural activations between the two diagnoses.

Some limitations of our study should be mentioned. In addition to the possible residual selection bias noted above, we were unable to subdivide the data sets to perform sensitivity analyses, for example, by separating children and adults or excluding particular domains or constructs. This is because we prioritized stratified sampling, which required at least two studies in each subconstruct or construct. Moreover, although we extracted the medication status of participants, we could not address effects of concurrent pharmacological treatments, because controlling for medication status would have required access to individual participant data, which was well beyond the scope of this work. We did not include studies with only a dimensional measurement of symptom severity. Because the aim of this study was to test the impact of the disorder on neural activation, including participants without a categorical diagnosis would not have been consistent with our research question and would have increased heterogeneity in clinical presentations. However, reporting only symptom severity would not necessarily mean that participants did not meet criteria for a categorical diagnosis. The exclusion of these studies resulted in the reduction of the number of integrated studies. Future quantitative syntheses should include both categorical and dimensional studies, relying on the strengths of both, to overcome heterogeneity and to maximize the number of integrated studies. We also excluded studies that did not report significant differences. Because the primary goal of ALE is to examine spatial convergence rather than effect sizes, studies that did not report significant group differences would not affect this likelihood. We meta-analyzed all studies with an overlapping set of participants as long as the studies adopted different neuropsychological tasks. There were two reasons for this decision. First, it was not possible to completely exclude the possibility of participants overlapping across studies. Second, we assumed that different neuropsychological tasks demand different neural activations. Although studies with known participant overlap indeed showed different neural activation patterns with different tasks, our results may have been biased by unknown confounding factors present in participants who appeared more than once in the analysis. Potential unknown confounding factors include unknown genetic vulnerability or environmental factors as well as known confounding factors such as age, sex, and comorbidities. Finally, although we included studies regardless of participants’ age, sex, or intelligence, we note that participants in the included studies reflected mainly intellectually high-functioning individuals, because performing neuropsychological tasks is more challenging for people with low functioning.

CONCLUSIONS

Pooling data from 243 task-based fMRI studies and using an advanced approach to address task selection bias, we found that individuals with ASD and ADHD shared some brain activation abnormalities, although disorder-specific alterations predominated. Our findings can inform the ongoing clinical debate on whether ADHD and ASD should be merged together as “neurodevelopmental conditions” or should be kept as distinct entities. The frequent co-occurrence of ASD and ADHD as well as the shared and specific abnormalities may support the need for more integrated pathways of care. From a clinical and service organization standpoint, considering neurodevelopmental disorders as a more homogeneous construct, rather than separate disorders, may provide more efficient delivery of care. However, our findings should also encourage clinicians to be mindful of specific cognitive and behavioral features of these disorders, because they may require specific management strategies.

Continuing Medical Education.

You can earn CME credits by reading this article. Three articles in every American Journal of Psychiatry issue comprise a short course for up to 1 AMA PRA Category 1 Credit^™ each. The course consists of reading the article and answering three multiple-choice questions with a single correct answer. CME credit is issued only online. Readers who want credit must subscribe to the AJP Continuing Medical Education Course Program (psychiatryonline.org/cme), select The American Journal of Psychiatry at that site, take the course(s) of their choosing, complete an evaluation form, and submit their answers for CME credit. A certificate for each course will be generated upon successful completion. This activity is sponsored by the American Psychiatric Association.

Examination Questions for “Shared and Specific Neural Correlates of Attention Deficit Hyperactivity Disorder and Autism Spectrum Disorder: A Meta-Analysis of 243 Task-Based Functional MRI Studies”.

1. One factor that has hampered the understanding of shared and specific neural correlates in studies of functional magnetic resonance imaging (fMRI) in ADHD and ASD is:

   1. The imbalance of type of fMRI tasks in studies of ADHD and ASD, respectively
   2. The higher prevalence of ADHD compared to ASD
   3. Higher motion artefacts in ADHD
   4. None of the above
2. The meta-analysis showed that both children with ADHD and those with ASD showed:

   1. Higher activation in the left middle frontal gyrus
   2. Lower activation in the left middle frontal gyrus
   3. Higher activation in the right middle frontal gyrus
   4. Lower activation in the right middle frontal gyrus
3. The meta-analysis showed ADHD-specific hyper-activation in:

   1. The thalamus
   2. The left insula
   3. The right insula
   4. All of the above