Background: Patients with congenital heart disease (CHD) are at risk for cognitive impairments¹. Executive function difficulties are among the most prevalent non-cardiac impairments observed in CHD and are associated with increased mental health problems and lower quality of life^2,3. Cerebral MRI studies in neonates with CHD have identified structural network alterations in comparison to healthy neonates^4–6 and first evidence has shown that these alterations seem to persist into adolescence and correlate with cognitive impairments, including executive dysfunction⁷. However, within and between study variability is high and studies are needed to identify those patients which are most vulnerable to brain alterations and neurodevelopmental impairments. Thus, we aimed to investigate if network connectivity measures were associated with the individual patient’s cumulative load of clinical risk factors and with family-environmental factors in a cohort of adolescents with CHD. Further, we investigated associations between brain networks and executive function impairments.
Methods: Adolescents with CHD who underwent infant open-heart surgery, and healthy controls between 10 and 15 years of age were eligible for this study⁸. Diffusion MRI (dMRI) data was acquired on a GE MR750 MRI scanner (GE HealthCare, Milwaukee, WI, USA), with TR/TE=7500/89 ms; matrix=96x96; FOV=280 mm, slice thickness=3.6mm. A total of 35 diffusion-weighted gradient directions were acquired with b=1000 s/mm2 and four non-diffusion weighted images. 3D T1-weighted cerebral MRI data were collected with an inversion-recovery prepared, spoiled gradient echo volume (IR-SPGR) with TI/TE/TR= 600/5/11 ms and a voxel resolution of 1x1x1mm3. For dMRI data, denoising using patch2self⁹, Gibbs ringing correction^10,11, motion and eddy current correction¹², and B1 bias field correction^10,11 was conducted. To reconstruct structural brain networks, single-shell three-tissue constrained spherical deconvolution (SS3T-CSD) was conducted with MRtrix3Tissue^13,14. Probabilistic anatomically constrained tractography was performed to model the white matter fiber orientation distribution with the iFOD2 algorithm and 10 million streamlines were estimated (SIFT2). Dynamic seeding was used and a stopping criteria of streamline estimation was implemented based on the five-tissue-type image (white matter, subcortical and cortical grey matter, cerebrospinal fluid, and pathological tissue), which was generated by segmenting the 3D T1 images with FreeSurfer 5.3. Biologically accurate weights were generated for each subject and were used to normalize for inter-subject connection density. Weighted, undirected structural connectivity networks were formed in 162x162 matrices where edges correspond to the sum of weighted streamlines and nodes correspond to the grey matter regions derived from the AAL3v1 atlas. Structural network metrics were estimated with network-based statistics (NBS) and graph theory using Brain Connectivity Toolbox (BCT). A cumulative clinical risk (CCR) score was built by summing up binary risk factors (neonatal, cardiac, neurologic) based on clinically relevant thresholds. Family-environmental factors (parental education, parental mental health, and family function) were assessed with standardized parent-reported questionnaires. Cognitive data on executive functioning was acquired with a standardized neuropsychological test battery and an age-adjusted executive function summary score (z-score) was built. All analyses were corrected for age and sex.
Results: A total of 53 patients (10% with single-ventricle CHD, mean age 13.6±1.1 years, 64% male sex) and 75 healthy controls (mean age 12.9±11.4 years, 49% male sex) were included in the analyses. Patients demonstrated significantly poorer executive function performance relative to controls (Patients: mean=-1.0±1.1, Controls: mean=-0.1±1.0, p<0.001). Graph theory analysis identified that patients had similar network integration, estimated by global efficiency (β=0.13, p=0.141), but lower network segregation, estimated by mean local efficiency (β=0.20, p=0.027), in comparison to controls. Network-based statistics identified a dense network which had lower network strength in patients compared to controls (p<0.001; Figure 1). In patients, a higher CCR score was associated with lower network segregation (β=-0.40, p=0.004; Figure 2), lower edge strength of a subnetwork including inter-frontal and fronto-parietal-thalamic connections (p<0.001; Figure 3), and with poorer executive functioning. There was no association between any network connectivity measure and family-environmental factors (all p>0.05). Poorer executive functioning was associated with lower network integration (β=0.34, p=0.015) and network segregation (β=0.34, p=0.013.
Discussion and Conclusion: We provided important evidence that structural brain network alterations, which have been described in neonates with CHD, persist into adolescence. Importantly, we demonstrated that patients with CHD, who face a cumulative exposure to multiple clinical risk factors over time, are particularly vulnerable to alterations in structural brain networks and poorer executive functioning. Quantifying the cumulative load of risk early in life may improve the prediction of long-term brain development of the individual patient. This can help to identify and support the most vulnerable patients as early as possible and to guide clinical and surgical decision making.

Acknowledgements

This project was supported by the Swiss National Science Foundation (SNF 32003B_172914) and the University Research Priority Program (URPP) ‘Adaptive Brain Circuits in Development and Learning (AdaBD)’ of the University of Zurich.