Stepwise functional connectivity reveals altered sensory‐multimodal integration in medication‐naïve adults with attention deficit hyperactivity disorder

Abstract

Neuroimaging studies indicate that children with attention‐deficit/hyperactivity disorder (ADHD) present alterations in several functional networks of the sensation‐to‐cognition spectrum. These alterations include functional overconnectivity within sensory regions and underconnectivity between sensory regions and neural hubs supporting higher order cognitive functions. Today, it is unknown whether this same pattern of alterations persists in adult patients with ADHD who had never been medicated for their condition. The aim of the present study was to assess whether medication‐naïve adults with ADHD presented alterations in functional networks of the sensation‐to‐cognition spectrum. Thirty‐one medication‐naïve adults with ADHD and twenty‐two healthy adults underwent resting‐state functional magnetic resonance imaging (rs‐fMRI). Stepwise functional connectivity (SFC) was used to characterize the pattern of functional connectivity between sensory seed regions and the rest of the brain at direct, short, intermediate, and long functional connectivity distances, thus covering the continuum from the sensory input to the neural hubs supporting higher order cognitive functions. As compared to controls, adults with ADHD presented increased SFC degree within primary sensory regions and decreased SFC degree between sensory seeds and higher order integration nodes. In addition, they exhibited decreased connectivity degree between sensory seeds and regions of the default‐mode network. Consistently, the higher the score in clinical severity scales the lower connectivity degree between seed regions and the default mode network.

1. INTRODUCTION

Attention‐deficit/hyperactivity disorder (ADHD) is a neurodevelopmental disorder characterized by excessive levels of inattention, impulsivity, and hyperactivity (American Psychiatric Association, 2013). Approximately 35% of children with ADHD still fulfill DSM‐IV diagnostic criteria for ADHD in adult life (Biederman, Petty, Evans, Small, & Faraone, 2010).

Neuroimaging research on ADHD has typically focused on networks supporting higher order functions, with little attention to findings in primary sensory regions. For instance, several studies report a weaker segregation between cognitive control networks and default mode network both in children and adults with ADHD (Cao et al., 2009; Castellanos et al., 2008; Hoekzema et al., 2014; Lin et al., 2018; Sun et al., 2012), although only a few studies highlight the need to clarify how sensory regions interact with higher order association networks in ADHD (Castellanos and Proal, 2012; Lin et al., 2018). In response to this, recently developed analysis tools such as the stepwise functional connectivity (SFC) approach (Sepulcre, Sabuncu, Yeo, Liu, & Johnson, 2012) allow evaluating the presence of abnormalities in multilevel information processing systems from early sensory to higher order cognitive circuits in the brain.

Making use of an SFC protocol, Carmona et al. (2015) provided evidence that the information flow between primary sensory cortices and higher order association nodes might be disrupted in children with ADHD. Compared with controls, children with ADHD presented increased interconnectivity within primary sensory cortices at initial steps of the sensation to cognition continuum (Carmona et al., 2015). At the final steps of the sensation to cognition continuum, children presented decreased SFC degree with executive processing areas and increased SFC degree with DMN areas. These studies indicate atypical connectivity transitions between sensory and higher order large‐scale functional networks, thus potentially compromising the flow of information across the sensation‐to‐cognition continuum. However, whether this pattern is also present in the adult form of the disorder is still unknown.

The aim of the present study was to test whether medication‐naïve adult patients with ADHD show impaired connectivity between primary sensory and higher order cognitive circuits. For this purpose, we applied a SFC protocol aimed to detect which parts of the brain are connected with primary sensory regions not only through direct paths (i.e., one‐step functional distance, which would be the standard functional connectivity analysis), but also through indirect connections that involve a varying number of “link‐step” distances or “relay stations.” Hence, in contrast to other standard methods such as functional connectivity strength evaluation, SFC allows measuring functional connectivity between any pair of brain locations that are connected by any finite number of relay stations.

Based on the assumption that adult ADHD may share similarities with childhood ADHD in terms of functional connectivity alterations, we predicted that our sample of medication‐naïve adults with ADHD would show increased SFC within sensory regions, as well as decreased connectivity between sensory seeds and networks supporting executive functions, and increased connectivity between sensory seeds and key nodes of the DMN.

2. METHODS

2.1. Participants

A total of 31 adults with combined ADHD and 22 healthy controls were recruited (see demographics in Table 1). We ensured both sexes were well represented in both groups (12 women in the ADHD group and 16 women in the HC group). The ADHD patients were selected by a specialized team of psychiatrists and psychologists from Vall d'Hebron Hospital in Barcelona (Spain), where they were evaluated. All patients met DSM‐V criteria (American Psychiatric Association, 2013) for ADHD combined subtype and were medication naïve.

Table 1.

Demographic and clinical data of the ADHD and control samples. Three controls did not complete the attention‐deficit/hyperactivity disorder (ADHD) rating scale. Independent sample t‐tests or chi‐square were used to compare groups

Characteristic	ADHD (N = 31)	Controls (N = 22)	Stat (df)	p‐value
	Mean (SD)	Mean (SD)
Age (range 19 to 52)	35.4 (9.9)	30.4 (5.8)	t(51) = −2.11	.040
ADHD rating scale	32.3 (9.8)	6.0 (5.8)	t(48) = −11.89	< .001
Sex (number of women)	16	12	χ2(1) = 0 .044	n.s.
Number scanned with replacement headcoil	14	6	χ2 (1) = 1.75	n.s.
Framewise displacement	.054 (.036)	.036 (.030)	t(51) = −1.92	.061

Standard ADHD scales were administered to both groups, including the Conners Adult ADHD Rating Scale (CAARS) (Conners, Sitarenios, & Parker, 1998), the Wender Utah Rating Scale (WURS) (Ward & Wender, 1993), and the ADHD Rating Scale (DuPaul, Power, Anastopoulos, & Reid, 1998). All ADHD scores were significantly higher in the ADHD sample (see demographic data in Table 1).

Exclusion criteria included comorbidity with other psychiatric diseases or personality disorders, assessed by the structured Clinical Interview for Axis I (SCID‐I) (First, Spitzer, Gibbon, & Williams, 1997) and Axis II disorders (First, Spitzer, Gibbon, Williams, & Benjamin, 1994). Participants with substance abuse disorder, including those who consumed tobacco and cannabis within the last 6 months, were also excluded. Participants with an estimated WAIS‐III IQ (Wechsler, 1997) lower than 80 were excluded. The study was approved by the Hospital de Vall d'Hebron Ethics Committee, and informed consent was obtained from all participants before taking part in the study.

2.2. fMRI image acquisition and preprocessing

Images were acquired using a Philips Achieva 3T scanner. T1‐weighted images were obtained using a FSPGR sequence (TR: 8.2 ms, TE: 3.7 ms, FA: 88, voxel size: 0.94 × 0.94, slice thickness: 1.00 mm, gap: 0 mm, matrix size: 256 × 256 × 180). An EPI‐T2* sequence was used to obtain the resting‐state functional volumes in a single run that lasted 5.8 min (116 time points, TR: 3000 ms, TE: 35 ms, FA: 90, in‐plane voxel size: 1.80 × 1.80 mm, slice thickness: 3.0 mm, gap: 1.0 mm, matrix size: 128 × 128). Due to a technical problem, 11 participants (evenly distributed between groups; χ² = 1.753; p = .186) were scanned using a different radiofrequency head coil (16 channels instead of 8 channels), which was considered in the analyses. Participants were instructed to remain still and awake with their eyes open during the functional run.

Functional MRI data were preprocessed with the software packages SPM12 (Welcome Department of Imaging Neuroscience, London, UK) and AFNI (Scientific and Statistical Computing Core, National Institute of Mental Health, Bethesda, MD). After removing the first three volumes, functional images were realigned to the mean image to correct for motion‐related artifacts, despiked with 3dDespike AFNI tool (c1 = 2.5, c2 = 4), normalized to MNI standard space, and spatially smoothed with a 6‐mm full‐width‐at‐half‐maximum Gaussian kernel. All functional images were downsampled to 4‐mm voxels to facilitate computational calculations (Sepulcre et al., 2012). Finally, nuisance covariates (six rigid body realignment parameters, mean white matter, mean cerebrospinal fluid, and mean whole brain intensity signals) were regressed out to minimize the effects of movement.

Given the impact of in‐scanner motion on functional connectivity analyses (Ciric et al., 2017; Di Martino et al., 2014; Power et al., 2014), participants with a mean framewise displacement (FD) over 0.2 mm as measured by the MCFLIRT tool (Jenkinson, Bannister, Brady, & Smith, 2002) were discarded. Additionally, we plotted a resting state functional connectivity quality control index (RSFC‐QC;Power et al., 2014) to assess the effect of motion in functional connectivity as a function of node distance. The data quality control showed no relationship between functional connectivity estimates and node distance, thus pointing to a reduced effect of head motion artifacts in our data (Figure S1).

2.3. SFC analysis

The SFC analysis allows computing the number of functional paths between previously defined seed regions and every other voxel in the brain at successive numbers of relay stations or “link‐step” distances. Intermediate voxels work as relay stations or “link‐steps” that range between 0 (direct, one link‐step connection) and 6 (seven link‐step connections) before stabilizing (Sepulcre et al., 2012) (see Figure 2). Based on the number of relay stations, the degree of functional connectivity can be classified as direct (one link‐step, and thus zero relay stations), short (two and three link‐steps), medium (four and five link‐steps), and long (six and seven link‐steps).

Figure 2.

(a) Surface projections of the two‐sample t‐test results of uneven link‐step distances for the between‐groups' contrast between adult ADHD patients and the control sample. Images display Cohen's D effect sizes, and the positive (hot) and negative (winter) color maps range from the absolute value corresponding to an uncorrected p < .05 to an absolute value of 1. Values greater than 1 and lower than −1 are collapsed to 1 and − 1, respectively. Subplot (b) indicates to which large‐scale resting‐state functional networks the significant voxels belong according to the parcellation of Yeo et al. (2011). Left hemispheres are displayed. Abbreviation: ADHD, attention‐deficit/hyperactivity disorder [Color figure can be viewed at http://wileyonlinelibrary.com]

For each processed brain, we computed the whole‐brain connectivity matrix by calculating the Pearson's correlation coefficient R for each pair of voxels. From this point onwards, only positive correlations were considered given the challenge of interpreting negative correlations after global signal regression in functional connectivity studies (Murphy, Birn, Handwerker, Jones, & Bandettini, 2009; Van Dijk et al., 2010). Correlation matrices were then filtered to contain only correlations surviving false discovery rate (FDR) correction (q < 0.001). The resulting matrix was then binarized. As a result, we obtained an unweighted “one link‐step” matrix for each subject containing 1's for each pair of voxels whose signals were significantly correlated and zeros otherwise.

In parallel, a set of three masks including three bilateral sensory seed regions was designed, each encompassing eight voxels (each voxel was 4 mm isotropic) forming a cube. Following Sepulcre et al. (2012), the MNI coordinates for the most anterior, lateral, and superior voxel of each cube was [8, −76, 10] in the primary visual cortex, [56, −12, 10] in the auditory cortex, and [0, −28, 66] in the somatosensory cortex. To assess the degree of combined SFC of all sensory seeds irrespective of modality, a fourth mask was built combining information from all three primary sensory regions.

Each n‐step map encoded the number of n‐step connections (SFC values) between every voxel in the brain and the voxels within the mask including the three bilateral seed regions. At each link step, SFC maps were standardized to Z‐scores by subtracting the mean and dividing by its SD to yield SFC values. Henceforth, we refer to these Z‐score values as the SFC values. A more detailed description of the SFC method can be found in Sepulcre et al. (2012) and Sepulcre (2014).

2.4. Statistical analysis

All the statistical analyses were performed using SPM12. Groups were homogeneous for mean FD, head coil, sex, and IQ but differed in age (t[51] = −2.11; p = .04).

For each of the seven SFC maps, general linear models were fit for each group separately. These models included age (mean centered to zero), FD (mean centered to zero), and head coil as covariates of no interest. Statistical inference was performed over the intercept of the models to identify SFC values significantly greater than zero. Since this analysis was performed for exploratory purposes only, we displayed clusters of at least 10 contiguous voxels surviving an uncorrected p < .01.

General linear models were also fit to compare the seven SFC maps between groups. These models included group as a variable of interest and age, sex, FD, and head coil as variables of no interest (age and FD were mean centered to zero). Even though FD, sex, and head coil did not differ significantly between groups, they were included as a precaution. For each model, we tested the effect of group through a t‐test on the value of its estimate.

In addition to group differences, we tested the association between symptom severity—ADHD Rating Scale—and functional connectivity profiles in each step within the group of ADHD patients. In this correlation analysis, we included head coil, FD, sex, and age as nuisance covariates.

As supplementary analyses, we explored if group differences were consistent across gender (see Figure 3) and across the three main sensory modalities (see Figure 4). In addition, we measured the predictive power of the group differences using the software PRoNTo (Schrouff et al., 2013) (see Figure 5 and Table S1).

Figure 3.

Results of the regression analysis using the ADHD rating scale score as a predictor of stepwise functional connectivity. Subplots (a), (c), and (e) depict positive correlations and subplots (b), (d), and (f) depict negative correlations. Top surface images show the correlation coefficient R at three, five, and seven link‐step distances, and the color maps range from r = 0.25 absolute value (which corresponds to the minimum [bilateral] significant correlation at p < .05) to r = 0.6 absolute value. Subplots (c) and (d) show the scatter plots for the positive and negative correlations, respectively. Bottom surface images ( e and f) indicate to which large‐scale resting‐state functional networks the significant voxels belong according to the parcellation of Yeo et al. (2011). Left hemispheres are displayed. Abbreviations: ADHD, attention‐deficit/hyperactivity disorder; SFC, stepwise functional connectivity [Color figure can be viewed at http://wileyonlinelibrary.com]

To correct for multiple comparisons, we used a Monte‐Carlo simulation implemented in the AFNI 3dClustSim function (Forman et al., 1995; accessed July 18, 2018). This method was used to obtain an experiment Family Wise Error (FWE) corrected type 1 error probability of 0.05 (α^FWE = .05). To achieve it, we needed a map FWE corrected type 1 error probability of .0071 (because we have 7 maps and 0.05/7 = 0.0071). For this purpose, we thresholded the statistical maps with a minimum cluster size of 174 contiguous 4 mm³voxels surviving the uncorrected p < .05.

2.5. Cortical and network visualization

To facilitate interpretation of results in the context of large‐scale functional networks, the percentage of significant voxels that overlapped with each of the seven cortical resting‐state functional networks described by Yeo et al. (2011) was calculated for each analysis using MATLAB 2019a.

Surface projections of SFC maps were performed via a Matlab in‐house script that uses nearest neighbor (for Yeo's atlas) or linear (for the quantitative maps) interpolation and the surface normals to project cortical voxels onto the surface. The surfaces employed were the left and right “Q1‐Q6_R440.#.midthickness.164k_fs_LR.surf.gii” of the software Connectome Workbench (Marcus et al., 2011). To avoid redundancy, we only present stepwise connectivity profiles of the left hemisphere with uneven step numbers in the main document.

3. RESULTS

3.1. SFC maps in patients with ADHD and healthy controls

The combined SFC maps for adult patients with ADHD and healthy controls are shown in Figure 1a,b, respectively. As expected, at short link‐step distances (one and three link‐step maps), both groups exhibited functional connectivity between the sensory seeds and primary sensory regions. At longer link‐step distances (five and seven link‐step maps), functional connectivity between sensory seed regions and frontoparietal areas was established in both groups. In turn, healthy controls showed connectivity with medial frontal areas and the precuneus at longer link‐step distances, which was not observed in the ADHD group.

Figure 1.

Surface projections of the one‐sample t‐test results of uneven link‐step distances for (a) the adult sample of ADHD patients and (b) the control sample of healthy adults. Each image represents Cohen's D ranging from the value that corresponds with a p < .01 (.6 for the control sample and .47 for the ADHD sample) and a value of 1. Cohen's D effect sizes greater than 1 are collapsed to 1. Left hemispheres are displayed. Abbreviation: ADHD, attention‐deficit/hyperactivity disorder [Color figure can be viewed at http://wileyonlinelibrary.com]

3.2. Between‐group differences

Between group comparisons are shown in Table 2 and map projections of between‐group differences are presented in Figure 2a. All results were corrected for multiple comparisons (see Methods, Statistical analysis).

Table 2.

Results of the stepwise functional connectivity (SFC) analyses, including between‐subject comparisons (adult patients with attention‐deficit/hyperactivity disorder [ADHD] vs. healthy controls) for each SFC map at different functional distances (one‐step to seven‐steps). Results were corrected for multiple comparisons by means of a Monte‐Carlo simulation

Between‐subject comparison	Peak MNI coordinates			Number of voxels	Highest T‐score	Cluster‐level p‐value
	x	y	z
One‐step
Three‐steps
ADHD > controls
L calcarine	−4	−64	16	248	3.25	<.001
Five‐steps
Controls > ADHD
L medial orbitofrontal gyrus	−4	58	−17	185	3.62	.001
ADHD > controls
L calcarine	−4	−68	16	398	3.47	<.001
Seven‐steps
ADHD > controls
R lingual gyrus	16	−61	−9	283	3.17	<.001

Abbreviations: L, left; R, right.

The analysis revealed functional connectivity differences starting from the three‐step connectivity maps, with ADHD patients exhibiting increased seed region connectivity with the left calcarine sulcus compared to controls. These functional connectivity differences were maintained until the seven link‐step maps, which exhibited between‐group differences that peaked in the right lingual gyrus.

Patients with ADHD showed decreased functional connectivity in the five link‐step distance map in the left medial orbitofrontal gyrus compared to controls. These differences, however, faded in the seven link‐step map. Network plots showed that the functional connectivity decreases observed in controls versus ADHD patients in five‐step maps largely overlapped with the DMN (77.84%) and, to a lesser extent, the frontoparietal network (13.51%) and limbic circuits (8.65%, see Figure 2b).

Results from the supplementary analysis indicated that group differences were consistent across gender and sensory modalities (see Figures 3 and 4). They also indicated a predictive power of the main group differences of 55.6, 64.08, and 62.46% for three, five, and seven link‐steps distances, respectively (see Figure 5 and Table S1).

3.3. Association with ADHD symptom severity

Table 3 and Figure 3 show the results of the regression analysis with the symptom severity scales.

Table 3.

Results of the stepwise functional connectivity (SFC) analyses, including positive and negative associations of the attention‐deficit/hyperactivity disorder (ADHD) rating scale in adult ADHD patients with each of the SFC maps at different functional distances (one‐step to seven‐steps). Results were corrected for multiple comparisons by means of a Monte‐Carlo simulation

ADHD rating scale	Peak MNI coordinates			Number of voxels	Highest R	Cluster‐level p‐value
	x	y	z
One‐step
Three‐steps
Positive association
L middle frontal gyrus	−36	3	59	556	0.68	<.001
Negative association
R superior frontal gyrus	20	31	53	496	0.64	<.001
Five‐steps
Positive association
R superior temporal gyrus	56	−35	21	1,179	0.62	<.001
Negative association
L superior frontal gyrus	−28	65	15	859	0.71	<.001
R precuneus	12	−58	35	184	0.59	.001
Seven‐steps
Positive association
R superior temporal gyrus	56	−35	21	1,185	0.61	<.001
Negative association
L middle frontal gyrus	−40	23	46	287	0.7	<.001
R superior frontal gyrus	−28	65	15	441	0.65	<.001
R precuneus	12	−58	35	206	0.61	.001

Abbreviations: L, left; R, right.

As observed in Figure 3, symptom severity was positively associated with the degree of functional connectivity in the left middle frontal gyrus and the right superior temporal gyrus, regions that largely overlapped with the sensory‐motor network (56.65% in three link‐step, 59.88% in five link‐step, and 60.42% in seven link‐step maps) and the dorsal attention network (28.24% in three‐step, 26.38% in five link‐step, and 25.57% in seven link‐step maps). With regard to negative correlations, we found that the higher the ADHD rating scale score the lower the degree of functional connectivity in the bilateral superior frontal gyrus, clusters that largely overlapped with the DMN (68.15% in three link‐step, 64.14% in five link‐step, and 62.96% in seven link‐step maps) and the frontoparietal network (21.17% in three link‐step, 25.50% in five link‐step, and 25.05% in seven link‐step maps). The associations between ADHD symptom severity scores and mean SFC values in each significant cluster per link‐step distance map are illustrated in Figure 3c, d.

4. DISCUSSION

The present study aimed to elucidate how primary sensory regions interact with higher order association networks in adult ADHD, a disorder that is typically approached with a strong focus on higher order cognitive functions in neuroimaging research. Hence, we used SFC to assess multilevel information processing between early sensory and higher order cognitive circuits in the brain of medication‐naïve adults with ADHD compared to healthy adults. Our results partially align with a previous SFC study on children with ADHD (Carmona et al., 2015), suggesting that the increased functional connectivity within sensory regions may persist in adulthood. However, sensorial integration into the DMN was lower in adults with ADHD compared with controls, the reverse pattern of that found in children with ADHD (hyper‐connectivity) (Carmona et al., 2015). Thus, deviations from typical SFC patterns in adult ADHD only partially resembled those observed in children with ADHD in previous studies. The correlations between SFC values and symptom severity in the adult ADHD sample corroborated the between‐groups findings. In particular, ADHD rating scale scores were positively associated with increased functional connectivity within the somatosensory‐motor network and between seed regions and the dorsal attention network, and inversely associated with functional connectivity between sensory seed regions and the DMN and the frontoparietal network at short, medium, and long functional distances. We discuss each of these findings below.

5. INCREASED SFC IN VISUAL CORTICES

Our results indicate increased functional connectivity between primary sensory areas and the visual cortex in adults with ADHD compared to controls. These findings are in line with Carmona et al. (2015) observations in children with ADHD, suggesting a similar pattern of deviations in children and adults with ADHD at intermediate and long functional distances. However, while children with ADHD showed hyperconnectivity within a small area of the lateral occipital cortex at short link‐step distances, the differences cover almost all of the bilateral medial occipital cortices in adults with ADHD (including V1, V2, V3, which are part of both the dorsal and the ventral visual circuitry), and are present at short, intermediate, and long link‐step distances.

We believe that increased connectivity at medium and long functional distances may reflect the general visual network hyperconnectivity frequently described in children with ADHD (Cao et al., 2006; Kessler, Angstadt, Welsh, & Sripada, 2014; Marcos‐Vidal et al., 2018; Wang et al., 2009). Our results suggest that at least part of the sensory information in adults with ADHD keeps reverberating within the visual loops, decreasing the information flow between sensory regions and neural hubs supporting higher order cognitive functions. In addition, our observations support the notion that existent models of ADHD would benefit from incorporating alterations in primary sensory areas, which are often ignored or taken as a false positive (Castellanos and Proal, 2012) but may nonetheless significantly alter multilevel information processing in ADHD.

6. DECREASED SFC IN DMN REGIONS

While the intravisual loops were more functionally connected in adults with ADHD compared to controls, the circuits connecting sensory cortices with areas associated with higher order cognitive functions were weaker. At medium functional distance (five link‐step distance), adults with ADHD exhibited reduced degree of functional connectivity with, predominantly, the DMN, and with the frontoparietal and limbic networks to a lesser extent. Such differences do not necessarily imply a less direct connection between sensory regions and the DMN. Since our measurements of between‐region functional connections included a number of relay stations (which define functional distance), these differences could be due to: (a) a weaker connection between the sensory cortices and the relay stations, or (b) a weaker connection between the relay stations and areas associated with higher order cognitive functions. Since the direct functional connectivity between primary sensory cortices and the relay stations was not weaker, the most likely compromised loops are those connecting relay stations (attentional or secondary sensory cortices) with the DMN.

These results are in line with studies pointing at deficits in DMN interconnectivity in adult ADHD, for example, decreased functional connectivity between the anterior cingulate cortex and the precuneus/posterior cingulate cortex (Castellanos et al., 2008), decreased network homogeneity in the DMN (Uddin et al., 2008), and distributed hypo‐connectivity within the DMN (Sripada et al., 2014). Additionally, children with ADHD have been found to exhibit decreased short and long‐range functional connectivity density in regions of the DMN (Tomasi & Volkow, 2012) and increased local functional connectivity in the boundaries of the DMN (Marcos‐Vidal et al., 2018). In adults with ADHD weaker segregation has been found between the DMN and cognitive control networks (Lin, Cocchi, et al., 2018). Altogether, these results point to a lack of integration in DMN regions and a lack of segregation between DMN and task positive networks.

As a part of the DMN, the medial prefrontal cortex (MPFC) was particularly affected in our sample. ADHD‐associated alterations in the MPFC have been reported in a wide variety of studies, including altered functional connectivity with other DMN nodes (Castellanos et al., 2008; Uddin et al., 2008), reduced deactivation while completing a task (Peterson et al., 2009), and slower cortical thinning in children with higher symptom severity (Shaw et al., 2011). Since the MPFC plays a key role within the DMN, the sensory hypoconnectivity with the MPFC observed in the present data set could be pointing at alterations in DMN‐associated functions such as mind‐wandering (Fox et al., 2005), which could underlie attentional deficits in ADHD.

Our findings contrast with those in Carmona et al. (2015), which found greater SFC in the DMN in children with ADHD. In general, the DMN undergoes intense maturational changes with age as it transitions from sparse within‐network functional connectivity in typically developing children to a more robustly interconnected network in neurotypical adults (Fair et al., 2008). On a speculative note, the hypoconnectivity and hyperconnectivity profiles observed at long functional distances in adults and children with ADHD, respectively, could be ascribed to altered DMN consolidation in early ages, yielding to abnormal functional connectivity profiles when compared to age‐matched controls. However, a longitudinal study would be needed in order to establish conclusions on the evolution of default mode SFC throughout the lifespan of ADHD patients from childhood into adulthood compared to controls.

The discrepancies between our study and the previous study on SFC in children with ADHD (Carmona et al., 2015) could also be related to medication status. While Carmona et al. (2015) analyzed brain activity in a mixed sample of medicated and medication‐naïve children, our present study comprised medication‐naïve adults. As reported by Carmona et al. (2015), medication status can influence SFC profiles (see Figure 5 in Carmona et al., 2015). Moreover, atomoxetine has been found to strengthen the anticorrelation between the DMN and task‐positive networks (Lin, Tseng, Lai, Matsuo, & Gau, 2015), and ADHD medication in general has been proposed to normalize DMN activity (Pereira & de Castro‐Manglano, 2017). All in all, medication status seems a relevant factor to consider in future studies specifically addressing DMN functional connectivity in ADHD across the lifespan.

7. ASSOCIATIONS WITH CLINICAL SCALES

At short, medium, and long functional distance, ADHD symptom scores were positively correlated with the degree of functional connectivity between seed regions and regions of the somatosensory‐motor and dorsal attention networks. The positive correlation with somatosensory‐motor regions is coherent with the increased SFC within sensory cortices in ADHD patients compared to controls. Indeed, children with ADHD also show increased degree centrality, that is, increased number of direct connections with other nodes in the somatosensory cortex (Di Martino et al., 2013). Anatomical studies also point to alterations in somatosensory‐motor cortices that persist into adulthood (Marcos‐Vidal et al., 2018). For instance, studies in children with ADHD report decreased gray matter volume in the somatosensory, motor, and premotor cortices (Carmona et al., 2005), while studies in adults with ADHD found increased cortical thickness in the presupplementary motor area and the somatosensory cortex (Duerden, Tannock, & Dockstader, 2012).

We also found a positive correlation between ADHD symptom severity and SFC in the dorsal attentional network. This finding dovetails with studies pointing to altered within network connectivity in the dorsal attentional network in adults with ADHD (Sidlauskaite, Sonuga‐Barke, Roeyers, & Wiersema, 2016) and increased connectivity between the dorsal attentional network and regions of the DMN in medication‐naïve children with ADHD (Lin, Lin, et al., 2018). Given the relevance of the dorsal attention network in attentional performance (Vossel, Geng, & Fink, 2014), it would be interesting to test whether the increased SFC between seed regions and nodes of the dorsal attention network predicted attentional performance.

Regarding negative correlations, we found that the higher the scores in ADHD symptom severity, the lower the degree of connectivity between the seed regions and the DMN and the frontoparietal network at short, medium, and long functional distance (three to seven link‐step maps). These findings are aligned with studies pointing to impaired functioning of the frontoparietal network in ADHD (Castellanos and Proal, 2012; Dickstein, Bannon, Xavier Castellanos, & Milham, 2006; Lin et al., 2015; Silk et al., 2005; Silk, Vance, Rinehart, Bradshaw, & Cunnington, 2008) but also highlight the relevance of DMN alterations in adult ADHD.

Altogether, our results are consistent with the view that ADHD is associated with altered information flow between sensory and neural nodes supporting higher order cognitive functions: whereas primary sensory areas seem to be hyperconnected in the first steps, they seem to be under‐connected to brain regions supporting higher order cognitive functions, especially the DMN, at long functional distances (Sepulcre et al., 2012). Our results also point out that, in terms of SFC, the ADHD brain is highly heterogeneous—as indicated by our limited classification accuracy—and suggest that part of this variability might be driven by differences in the severity of ADHD symptoms.

8. LIMITATIONS

The main limitation of our study is the relatively small sample size with a wide age range. This stems from the difficulty in recruiting ADHD patients who reach adulthood without comorbidity with other disorders (including tobacco and alcohol use in the last 6 months) and, importantly, without previous exposure to ADHD medication.

Also, as our sample exclusively comprised patients with the combined subtype, we could not provide a specific account of what precise ADHD phenotypes are associated with the SFC alterations. The question remains whether impaired sensory‐multimodal integration may be affecting attentional control, as well as other symptoms such as hyperactivity and impulsivity. Further research should be able to draw more precise conclusions on what particular phenotypes are linked to which specific SFC alteration.

Regarding methodological concerns, we should specify two. First, computational constraints required us to downsample data to relatively large voxels (4 mm³). As computational power continues to increase, the specific details of our results could be reexamined in the original data and in future data sets acquired at even greater temporal and spatial resolutions. Second, SFC analysis does not provide information on the directionality of the functional connectivity network under study; that is, the alterations observed could be interpreted as affecting sensory‐to‐cognitive and/or cognitive‐to‐sensory information processing. If SFC decrease was affecting sensory‐to‐cognitive functional streams, this would involve a reduced information feed from sensory up to higher level association nodes; if it was affecting cognitive‐to‐sensory functional streams, this would entail lower cognitive control over incoming perceptual information—thus hindering selective attention. Hence, although previous studies using SFC analysis tend to interpret their results in the sensory‐to‐cognitive direction (Carmona et al., 2015; Hong et al., 2019; Sepulcre et al., 2012), the directionality of the observed alterations should be tested in future studies using methods such as Dynamic Causal Modeling (Friston, Li, Daunizeau, & Stephan, 2011).

9. CONCLUSIONS

In this study, we characterized how primary sensory regions interact with networks supporting higher order cognitive functions in adult ADHD by means of an SFC protocol. Furthermore, we ensured that this characterization was biased neither by comorbidities nor by medication. Our results suggest that the brain of adults with ADHD presents an atypical flow of information from short to long functional distances of the sensation to cognition continuum. In particular, SFC in medication‐naïve adults with ADHD was characterized by over‐connectivity within primary sensory regions followed by under‐connectivity between sensory regions and nodes of the DMN. Importantly, this pattern was associated with the severity of ADHD symptoms. These findings highlight the need to draw greater attention to altered multilevel information processing in adult ADHD, with an emphasis on the interaction between primary sensory regions and the DMN.

CONFLICT OF INTERESTS

Dr. Ramos‐Quiroga and Dra. Richarte have served on the speakers' bureau and acted as consultant for Eli Lilly and Co., Janssen‐Cilag and Shire. Dr. Ramos‐Quiroga has also served on the speakers' bureau and acted as consultant for Laboratorios Rubió, Novartis, Lundbeck and Ferrer. Both have received travel awards from Eli Lilly and Co., Janssen‐Cilag, and Shire for participating in psychiatric meetings. The ADHD Program has received unrestricted educational and research support from Eli Lilly and Co., Janssen‐Cilag, Shire, Rovi, and Laboratorios Rubió in the past two years. The rest of the authors declare no conflict of interest.

Supporting information

Figure S1 Quality control (QC) of the Resting State Functional Connectivity‐correlations (RS‐FC). Scatter plot of the mean frame‐wise displacement (FD) over a subject's scan and each pairwise functional connectivity correlation across subjects (Y‐Axis) versus the Euclidian distance between each pair of voxels measured as the Pearson correlation (X‐Axis). Horizontal white dots illustrate the mean's correlation at every distance and black dashed line represents 0.

Figure S2. Spatial correlations between consecutive pairs of Stepwise Functional Connectivity (SFC) maps. The results showed that consecutive pairs of SFC maps from seven steps distance barely change. Therefore, including more SFC maps do not add almost any new information of multimodal integration.

Figure S3. Surface projections of the two‐sample t‐test results of uneven link‐step distances for (A) the female sample and (B) the male sample. Images display Cohen's D effect sizes from 1 to −1. Positive values indicate regions were the Stepwise Functional Connectivity in the ADHD is greater than in controls. Negative values indicate regions were the Stepwise Functional Connectivity in the controls is greater than in ADHD. Left hemispheres are displayed.

Figure S4. Surface projections of the two‐sample t‐test results of uneven link‐step distances from (A) the auditory seed (B), the somatomotor seed, and (C) the visual seed. Images display positive (ADHD>Controls) and negative (Controls>ADHD) D‐Cohen effect sizes. Cohen's D effect sizes greater than 1 and lower than −1 are collapsed to 1 and − 1, respectively. Left hemispheres are displayed.

Figure S5. Predictive power was measured with the PRoNTo toolbox (Schrouff et al., 2013), using a Leave One Out design and a Support Vector Machine model with a kernel matrix as input data. This kernel matrix is a distance matrix that was obtained with the values of SFC of the significant voxels for a determined functional distance. The functions represent the values acquired for the subjects of the control (Black line) and ADHD (Red line) groups respectively and the blue dashed line is the threshold that determine the prediction. Negative values of the function will classify a subject as ADHD and positive values of the function will classify a subject as control. ADHD: Attention Deficit and Hyperactivity Disorder; SFC: Stepwise Functional Connectivity.

Table S1. Table with a summary of the predictive power of SFC. SFC: Stepwise Functional Connectivity; ADHD: Attention Deficit and Hyperactivity Disorder; CTR: Controls

Pretus C, Marcos‐Vidal L, Martínez‐García M, et al. Stepwise functional connectivity reveals altered sensory‐multimodal integration in medication‐naïve adults with attention deficit hyperactivity disorder. Hum Brain Mapp. 2019;40:4645–4656. 10.1002/hbm.24727