Estimating the heritability of structural and functional brain connectivity in families affected by Attention Deficit Hyperactivity Disorder
Associated Data
Abstract
Importance.
Despite its high heritability, few risk genes have been identified for ADHD. Brain-based phenotypes could aid gene discovery. Here we focus on the myriad of structural and functional connections that support cognition. Disruption of such connectivity is a key pathophysiological mechanism for ADHD, and identifying heritable phenotypes within these connections could provide candidates for genomic studies.
Objective:
To identify the structural and functional connections that are heritable and pertinent to ADHD.
Methods:
This objective is attained by studying extended, multigenerational families, enriched for ADHD. Structural connectivity was defined by diffusion tensor imaging (DTI) of white matter tract microstructure and functional connectivity through resting state functional magnetic resonance imaging (rsfMRI). Heritability and association with ADHD symptoms were estimated on 24 extended, multigenerational families enriched for ADHD (305 members with clinical phenotyping, 213 with DTI, 193 with rsfMRI data). Findings were confirmed on 52 nuclear families (132 members with clinical phenotypes, 119 with DTI, 84 with rsfMRI).
Findings.
Microstructural properties of white matter tracts connecting ipsilateral cortical regions and the corpus callosum were significantly heritable (ranging from h2=0.69 (SE 0.13, p=0.0000002) for radial diffusivity of the right superior longitudinal fasciculus to h2=0.45 (SE 0.06, p=0.0009) for fractional anisotropy of the right inferior fronto-occipital fasciculus). Association with ADHD symptoms was found several tracts, most strongly for the right superior longitudinal fasciculus (t=−3.05, p<0.003). Heritable patterns of functional connectivity were detected within the default mode, cognitive control and ventral attention networks (mean h2 =0.34[SE0.15]). In all cases subregions within each network showed heritable functional connectivity with the rest of that network. More symptoms of hyperactivity/impulsivity (t=2.63, p=.008) and inattention (t=−2.34, p=0.02) were associated with decreased functional connectivity within the DMN. Phenotypic and genetic correlations were found between the ventral attention network and a major association white matter tract (the inferior fronto-occipital fasciculus).
Discussion:
Using multigenerational, extended and nuclear families, we identify the features of structural and functional connectivity that are both significantly heritable and associated with ADHD. Additionally, we find shared genetic factors account for some phenotypic correlations between functional and structural connections. Such work helps prioritize the facets of the brain’s connectivity for future genomic studies.
Introduction.
There has been limited progress in identifying the specific genes contributing to the established high heritability of ADHD 1,2. The use of heritable brain-based phenotypes pertinent to the disorder might accelerate progress in part as they lie closer to genes than the more distal clinical phenotype 3,4. Here we focus on the myriad structural and functional connections within the brain that support multiple cognitive, motor and affective processes 5–8. We do so as ADHD is increasingly viewed as the product of anomalous connectivity or ‘miswiring’ that results in disruption to large scale brain systems, producing symptoms 8–10. Additionally, such a focus addresses a gap in our knowledge. Although the heritability and association with ADHD of grey matter morphology has been extensively investigated 11–13, less is known of which aspects of connectivity are both heritable and pertinent to ADHD. Such a study would complement estimates of the heritability of structural and functional connectivity among healthy twins and families 14–23, and among families affected by bipolar affective disorder and schizophrenia 24–26.
Structural and functional connectivity can be studied on many levels. Here we use magnetic resonance imaging to define in vivo the microstructural properties of major white matter tracts.. We take this approach as ADHD and its core cognitive deficits have been associated with anomalies in the white matter tracts connecting different cerebral cortical regions 6. We also define functional connectivity, through the coordinated patterns of neural activity, or intrinsic networks, that emerge spontaneously when a subject is not engaged in task oriented behavior 27,28. ADHD has been conceptualized as an imbalance between these intrinsic networks, particularly the default mode network (DMN) –prominent during internally directed thought- and the networks supporting cognitive control and attention 29,30.
Imaging of the brain’s structural and functional connectivity provides a multitude of phenotypes and it is important to prioritize these for future genomic study. We take the strategy of first identifying the subset of phenotypes that is highly heritable. Such highly heritable phenotypes boost the chances of detecting underlying genes. Further prioritization can then be made on the strength of association with ADHD symptoms. We estimate connectivity within multigenerational, extended families in which a high proportion of members are affected by ADHD. This affords an efficient strategy to define both heritability and association with ADHD symptoms. We confirm initial heritability estimates in a separate cohort of nuclear families, also affected by ADHD. This approach also meets three further aims. First, we can define the heritability of all of the major intrinsic networks, broadening the prior focus on the DMN 19,21. Second, we can determine which heritable connectivity features are also pertinent to the symptoms of ADHD. Finally, our family based design allows us to answer the question: do components of the structural and functional connections share genetic determinants, or are they genetically distinct?
Methods.
Participants.
Inclusion criteria for the extended families were (1) the presence of second, third or higher degree relatives; (2) a diagnosis of ADHD in at least 25% of family members (around ten times the adult ADHD and four times the childhood ADHD prevalence rates 31,32). For nuclear families, the main inclusion criterion was at least two first-degree relatives (sibling or parent-child), at least one with ADHD.
The diagnosis of adult ADHD used the Conners’ Adult ADHD Diagnostic Interview for DSM-IV™, a clinician administered structured interview that establishes the number of symptoms of inattention and hyperactivity-impulsivity (0 to 9 for each category). The interview ascertains both current, adult symptoms of ADHD and the childhood history of these symptoms. We leveraged the family design to obtain collateral confirmation of childhood symptoms when possible. Presence of other psychiatric diagnoses was established through the Structured Clinical Interview for DSM Axis I Disorders. For children, the parental Diagnostic Interview for Children and Adolescents Interviews was used 33. Interviews were conducted by two experienced clinicians (PS and WS; inter-rater reliabilities, kappa >0.9), and neurological assessment by a physician (PS), Exclusion criteria were an IQ <80 (determined with Wechsler intelligence scales), neurological disorders affecting brain structure, current substance dependence, or psychotic disorders. Twenty-one of the 24 extended families and 42 of the 52 nuclear families were white, non-Hispanic. The institutional review board of the National Human Genome Research Institute approved the research protocol, and written informed consent was obtained from adult participants and parents; children gave written assent.
Neuroimaging.
Diffusion tensor imaging (DTI) data were collected on a 3-T HDx MRI system (GE Healthcare, Milwaukee, WI) with a single-shot dual-spin-echo echo-planar imaging sequence. Imaging parameters, preprocessing and tensor fitting are described in the Supplemental Material. The same acquisition parameters were used throughout except that 60 volumes were acquired for children to shorten scan time, compared to 80 in adults. Quality control measures included the re-acquisition of corrupted data in real time 34, visual inspection and removal of corrupted data. Participants were excluded if they had >10% of corrupted volumes; a trait value exceeding the sample mean by three or more standard deviations; or a mean overall tract fractional anisotropy of less than 0.25. Overall, 332 of the original 363 DTI data sets were retained. Although there were no significant correlations between head motion parameters and tract measures, we nonetheless considered motion as a covariate.
DTI-TK software registered the diffusion tensors into a common template space 35,36. We considered all of the eleven tracts measured by this software: the bilateral uncinate, inferior fronto-occipital, superior longitudinal, inferior longitudinal, corticospinal fasciculi, and the corpus callosum. For each tract, fractional anisotropy (FA), a summary metric of overall tract diffusion properties was defined, along with axial diffusivity (AD), and radial diffusivity (RD) which are proxies for the flow of water along the axis and the radius of the axon respectively.
Resting state fMRI was acquired using a gradient-echo echo-planar (EPI) series with whole-brain coverage. Participants were instructed to lie still for 5 minutes, 5 seconds and gaze at a fixation point. Preprocessing used the AFNI software package 37. The first 3 EPI volumes were removed, as well as any volumes that showed motion of > 0.2 mm, and volumes that contained more than 10% of voxels considered to be outliers. Following such ‘scrubbing’, the remaining subjects had a mean of 278s of usable data (SD 34s) and a lower limit of 180 seconds was set. In total 277 (193 extended, 84 nuclear) of the 340 originally acquired rsFMRI scans were retained. The amount of usable data was associated with hyperactivity/impulsivity symptoms (at t=1.68, p=0.1), but did not vary with inattention (t=1.05, p=0.29), age or sex (all p>0.1). The EPI volumes were registered to the individual’s T1-weighted anatomical image and to a MNI template38. Activation in white matter and lateral ventricle masks were removed using ANATICOR and the time derivative of the motion parameters were also regressed out 39.
We used Independent Component Analysis to decompose the BOLD signal into spatially distinct maps and their time courses40. Each independent component is a spatial map of functionally connected regions -an intrinsic network- that shows the strength of the contribution of every voxel to the intrinsic network. These intrinsic networks closely resemble, and are named for large-scale brain networks that support cognition. We focused on the seven major intrinsic networks described by Yeo and colleagues: the default mode, dorsal and ventral attention, cognitive control, affective, visual and somatomotor networks 41. Finally, dual regression created subject-specific spatial maps for each network 42. In each map, the value of each voxel shows the strength of the functional connection between that voxel and the rest of the network for that subject.
Analysis.
Heritability was estimated using Sequential Oligogenic Linkage Analysis Routine (SOLAR)43. It uses a variance component method to estimate the proportion of phenotypic variance due to additive genetic factors (i.e. narrow sense heritability)- see Supplemental Material. Inverse normalization was applied to phenotypes as heritability estimates in SOLAR are sensitive to skewed distributions. SOLAR also estimated the phenotypic correlation between heritable traits and the underlying genetic and environmental correlations, applying a false discovery rate (FDR) of 0.05 to correct for multiple testing 44.
ADHD symptom counts were regressed against each heritable trait, using a mixed-effects model with family identity as the random term. For adults, we were primarily interested in current symptoms, as we earlier found white matter tracts anomalies to be associated with current, not childhood, symptoms in unrelated adults 10. Sex, age, age2, movement, and movement2 were considered as covariates and retained if p <0.1.
We estimated the heritability of 33 white matter tracts properties in the extended families (3 properties for 11 tracts). A Bonferroni correction was applied and heritability declared significant at p<.05/33=0.0015. Confirmation of heritability in the nuclear family cohort was taken at a nominal p<0.05. In resting state analyses in extended families, the probability of false positive spatial clusters was estimated using a non-parametric approach (permutation) setting a voxel-wise p<0.05 with a cluster-corrected alpha level p<0.002 (see supplemental material for details). Confirmation of the heritability in the nuclear families was taken at a at nominal p<0.05 at the voxel level. No cluster extent correction was applied in the nuclear families as we were testing heritability within a region of interest initially defined by the extended families. In testing for associations between heritable connectivity measures and ADHD symptoms, results within each modality were adjusted for multiple testing using Bonferroni correction.
We modeled age–related change in the connectivity measures using linear mixed models- Supplemental Material. We also examined whether heritability estimates were similar in ‘youth’ (<=21years) and ‘adult’ (>21 years) groups. Finally, we determined if associations between ADHD symptoms and the connectivity measures differed between these age groups.
Results
Within the extended families, 115 of 305 (38%) relatives had ADHD. In the nuclear families, 78 of 132 individuals (59%) were affected – Table 1, Supplemental Figures 1,2.
Table 1:
FAMILY | Total size/ ADHD N(%) | Age (years): Mean (SD) | Min (years) | Max (years) | DTI: N (%) | rsMFRI N (%) |
---|---|---|---|---|---|---|
AA | 25 /7 [28%] | 31.2 (17) | 9.8 | 64.2 | 16 (64%) | 15 (60%) |
BB | 22 /9 [41%] | 34.5 (20.6) | 4.5 | 65.5 | 17 (77%) | 15 (68%) |
CC | 21 /9 [43%] | 31.3 (20.4) | 5.7 | 70.9 | 16 (76%) | 11 (52%) |
DD | 19 /7 [37%] | 29.8 (20.7) | 5.4 | 68.9 | 11 (57%) | 11 (57%) |
EE | 19 /8 [42%] | 25.3 (18.5) | 4.8 | 58.9 | 16 (84%) | 15 (78%) |
FF | 18 /6 [33%] | 32.8 (21.2) | 7.7 | 70.1 | 16 (88%) | 12 (66%) |
GG | 17 /4 [24%] | 37.1 (23) | 10.3 | 77.4 | 14 (82%) | 14 (82%) |
HH | 17 /7 [41%] | 15 (10.2) | 4.7 | 35.3 | 10 (58%) | 9 (52%) |
II | 16 /5 [31%] | 18.3 (13.5) | 6.6 | 44.3 | 11 (68%) | 9 (56%) |
JJ | 15 /6 [40%] | 32.2 (13.6) | 22.6 | 41.9 | 2 (13%) | 0 (0%) |
KK | 13 /4 [31%] | 32.1 (21.4) | 6.0 | 65.9 | 9 (69%) | 8 (61%) |
LL | 13 /4 [31%] | 28 (19.1) | 5.4 | 71.1 | 9 (69%) | 9 (69%) |
MM | 13 /8 [62%] | 28.8 (18.4) | 6.2 | 48.7 | 6 (46%) | 7 (53%) |
NN | 12 /7 [58%] | 40 (23.5) | 18.2 | 85.8 | 9 (75%) | 9 (75%) |
OO | 10 /3 [30%] | 29.3 (17.2) | 8.3 | 49.5 | 8 (80%) | 7 (70%) |
PP | 8 /3 [38%] | 24 (16) | 7.6 | 43.1 | 6 (75%) | 5 (62%) |
8 /2 [25%] | 32.5 (17.5) | 7.3 | 62.3 | 6 (75%) | 6 (75%) | |
RR | 8 /2 [25%] | 39.1 (26.9) | 10.0 | 77.6 | 7 (87%) | 7 (87%) |
SS | 8 /2 [25%] | 28.3 (24) | 4.9 | 71.3 | 5 (62%) | 4 (50%) |
TT | 6 /4 [67%] | 40.6 (16.3) | 26.9 | 62.7 | 5 (83%) | 5 (83%) |
UU | 5 /2 [40%] | 50.5 (21.7) | 19.6 | 68.2 | 3 (60%) | 4 (80%) |
VV | 4 /3 [75%] | 35.6 (22.1) | 12.8 | 55.1 | 4 (100%) | 4 (100%) |
WW | 4 /1 [25%] | 21.5 (13.8) | 11.0 | 41.3 | 4 (100%) | 4 (100%) |
ZZ | 4 /2 [50%] | 25.4 (16) | 7.0 | 37.1 | 3 (75%) | 3 (75%) |
Total | 305 /115 [38%] | 30.4 (19.7) | 213 (69%) | 193 (63%) |
Structural connectivity.
Fourteen of 33 white matter tract properties emerged as significantly heritable in the 213 relatives from extended families - Figure 1, Supplemental Table 3. Estimates ranged from h2=0.69 (SE 0.13, p=0.0000002) for radial diffusivity of the right superior longitudinal fasciculus to h2=0.46 (SE 0.15, p=0.0009) for fractional anisotropy of the right inferior fronto-occipital fasciculus. Twelve of these tract properties were further confirmed as significantly heritable (at p<0.05) in 119 individuals from nuclear families.
We next examined association between the heritable tract properties and ADHD symptom count. Radial diffusivity of the right superior longitudinal fasciculus was associated with inattention at a corrected level of significance (t=−3.05, P<0.003, Bonferroni adjusted P<0.05). Axial diffusivity of this tract showed a nominally significant association with inattention (t=−2.51, p=0.01). Association was also found between fractional anisotropy of the right inferior fronto-occipital fasciculus and hyperactivity/impulsivity at a nominal level of significance (t=−2.35, p=0.02). Results of DSM-5 diagnostic group contrasts are given in the Supplemental Material. Thus, radial diffusivity of the right superior longitudinal fasciculus emerged as the most robustly heritable and ADHD-associated white matter tract property.
Functional connectivity.
The intrinsic functional networks found to have regions of heritable functional connectivity in the 193 relatives from extended families are shown in Figure 2 and Supplemental Table 4. First, within DMN, functional connectivity between a posterior cingulate region and the remainder of the network was heritable (h2=0.36, SE 0.16, cluster level significance p<0.002). Within the cognitive control network, functional connectivity between its right inferior parietal ‘component and the rest of the network emerged as heritable (h2= 0.32, SE=0.15, cluster level significance p<0.002). For the ventral attention networks, heritability localized to the right superior frontal gyrus (h2= 0.36, SE=0.15, p<0.002). No patterns of heritable functional connectivity were found within the other networks.
These heritability findings were confirmed using rsfMRI data from 84 members of nuclear families- Figure 3, Supplemental Table 5. Functional connectivity between the posterior cingulate region and the rest of the network was found to be heritable in nuclear families. Similarly, the patterns of heritable functional connectivity within the cognitive control, dorsal and ventral attention networks first delineated in extended families were also present in nuclear families. Throughout, the heritable regions in the nuclear families were less extensive than those initially defined in the extended families.
Associations were found between the heritable functional connectivity of the DMN and both hyperactive/impulsive symptoms (t=2.63, p=0.008) and inattentive symptoms (t=2.34, p=0.02). A significant association between hyperactive/impulsive symptoms and the functional connectivity patterns within the ventral attention network was also found (t=2.76, p=0.006).
Considering developmental trends, the fractional anisotropy of most white matter tract showed a childhood and adolescent increase, which stabilized and then decreased in adulthood – (see Supplemental Material pages 10–16). We did not detect age related change in connectivity between the heritable regions and the rest of each network. Heritability estimates were mostly similar in younger (<=21years) and adult (>21 years) groups. Also, associations between symptoms and connectivity measures did not vary significantly between these age groups. Heritability estimates were robust to the exclusion of those on psychostimulants, and the exclusion of those on any psychotropic medication– Supplemental Tables 6 and 7.
Phenotypic and genetic correlations.
Among white matter properties, 383 of 492 possible phenotypic trait pairs were significantly correlated (applying a FDR, q<0.05). Correlated traits clustered more by diffusivity property than by tract location – see Figure 3 and Supplemental Figures 3,4. Functional connectivity was defined by an approach that provides independent components and thus phenotypic correlations were neither expected nor found.
Genetic correlations were found within and across modalities. Within the twelve heritable white matter properties, shared heritability was found for 58 of the 66 possible pairs (FDR, q<0.05). Genetic correlations were present between the cognitive control and dorsal attention networks at trend level only (rhog=.41 , p=.06).
Finally, we tested for cross-modal correlations. Heritable components of the ventral attention network showed both phenotypic and genetic correlations with the inferior fronto-occipital fasciculus. Specifically, the ventral attention network showed phenotypic (rhop=−.11, p=.02) and genetic correlations (rhog=−.45, p=.02) with radial diffusivity of the right interior fronto-occipital fasciculus. This implies that this phenotypic correlation is partly genetically determined. Some cross-modal correlations were purely phenotypic, including a correlation between the heritable aspects of the DMN and the right superior longitudinal fasciculus (rhop=−.12, p=.03).
Discussion
Several facets of structural and functional connectivity emerged as significantly heritable within extended families affected by ADHD. A separate cohort of nuclear families confirmed this heritability. For white matter tracts, heritability was found for microstructural features of the association (superior longitudinal fasciculi, inferior fronto-occipital, uncinate fasciculi) and commissural (corpus callosum) but not projection tracts (corticospinal tract). Heritable patterns of functional connectivity were also noted within the default mode, cognitive control and attention networks. Association with ADHD symptom severity emerged primarily for the heritable facets of the right superior longitudinal fasciculus and the DMN. Finally, cross-modal phenotypic and genetic correlations were found. A white matter tract, the inferior fronto-occipital fasciculus was phenotypically correlated, and shared common genetic determinants with the ventral attention network.
Microstructural properties of the right superior longitudinal fasciculus were both heritable and associated with ADHD, following adjustment for multiple comparisons. Meta-analyses find this tract is compromised in ADHD 6, associated with working memory and sustained attention in children with and without ADHD 45–48, and lesions of the tract are tied to deficits in these cognitive domains 49,50. Thus, the tract is a promising candidate phenotype for future genetic studies.
Our estimates of heritability of white matter tracts are consistent with prior studies of extended families and twins, either unaffected or heavily affected by psychiatric disorder 14–18,20,24–26. However, different tracts appear to be associated with different psychiatric disorders. For example, while heritable measures of the corpus callosum have been associated with bipolar affective disorder 24, we link ADHD symptoms with heritable properties of the right superior longitudinal fasciculus.
Using families enriched for ADHD, we confirm the heritability of the DMN that was first reported among extended families not ascertained for mental illness 19. We also show an association between these heritable DMN components and ADHD symptoms, implying a partly genetic determination of the atypical DMN activity that can disrupt goal-directed activity and drive ADHD symptoms. Further, we find that other intrinsic networks closely associated with ADHD- the cognitive control and ventral attention networks- had heritable patterns of functional connectivity.
We detected genetic and phenotypic correlations spanning our measures of structural and functional connectivity. We find that genetic factors contribute to the phenotypic correlation between the functional connectivity within the ventral attention network and a major white matter tract, the inferior fronto-occipital fasciculus. The inferior fronto-occipital fasciculus connects dorsal parieto- and basal temporo-occipital regions to the dorsolateral prefrontal and orbitofrontal cortex 51,52, making it a plausible substrate for the physical connections within the ventral attention network. The ventral attention system is right lateralized and it is thus of note that we find the right inferior fronto-occipital fasciculus to be correlated 53. Altered microstructural properties of this tract have been reported in adult ADHD 5,54, consonant with its role in cognitive skills which are often impaired in ADHD, such as visuospatial integration and attention set–shifting 55–58. Such cross-modal phenotypes that share genetic determinants may provide well-constrained phenotypes, ideal for future genomic studies.
There are several limitations. First, while the study’s cross-sectional design limits inferences about developmental trends, we found that heritability estimates and the associations between symptoms and connectivity measures did not differ by age group. However, a longitudinal study is needed to fully characterize possible interactions between age and heritability in ADHD. Second, many but not all findings in the extended families were confirmed in the nuclear families. The lack of confirmation could arise from differences in age composition between the extended and nuclear families and the smaller overall size of the nuclear family cohort. Finally, although we estimated movement during the resting state scan, we did not enquire whether the participant was able to sustain gaze throughout.
Demonstrating heritability and associations with a disorder is an initial but vital stage for the use of the structural and functional connectivity as phenotypes. The next step is to ask which genes drive this heritability and confer risk for ADHD.
Supplementary Material
Figure S1
Suppl Material
eTable1-5; supplemental methods; efigure 1a & 1b
More supplemental materials
Acknowledgements
Funded by the intramural research program of the NHGRI and NIMH.
Footnotes
All authors declare no conflict of interest.