Altered development of cortical networks likely occurs in patients with congenital heart disease (CHD) and is hypothesized to arise from multiple etiologies (poor perfusion, genetics, surgical trauma) that likely affect early brain development. A “connectome” approach has the potential to help delineate the relationship between neonatal cortical connectivity development and different etiologies. We applied resting-state functional connectivity MRI (rs-fcMRI) and graph analysis to neonatal CHD patients both pre- and post-operatively as well as normal healthy controls in order to investigate possible effects of CHD and palliative surgery on functional topology.

Complex CHD and healthy term neonates were prospectively recruited from two large CHD surgical centers (Children’s Hospital of Pittsburgh and Children’s Hospital Los Angeles) over a five year period (2009-2014). The CHD cohort comprised pathologies including single physiology, aortic arch obstruction, fetal mixing, conotruncal, aortic valve outflow obstruction and heterotaxy based upon a pre- or postnatal echocardiogram reviewed by a fetal cardiologist. Rs-fcMRI data was obtained from 157 total unsedated neonates using a multichannel coil with TR = 2 s, 150 volumes collected for a total scanning time of 5 minutes. At CHLA, data was acquired on a Philips 3 T Achieva system. At CHP, data was acquired on GE 1.5 T, GE 3 T, and Siemens Skyra 3T systems. Of those 157, 117 met criteria (described below) for acceptable motion and were retained for further analysis. Datasets from 19 CHD participants scanned either on the GE 1.5 T or the Philips 3 T scanner were excluded from the control-CHP comparison (as no control datasets were obtained on these scanners) but were included in the within-CHD pre-operative vs. post-operative comparison.

The data analysis approach closely followed that of Power et al.¹ in order to minimize the risk of spurious findings due to participant motion. Steps included: slice timing correction; affine motion correction; spatial normalization to a neonatal template²; intensity normalization (grand mean = 1000); time course extraction according to a 90-region neonatal parcellation atlas²; volume censoring according to framewise displacement (FD) > 0.2 mm or intensity-related parameter (DVARS) > 30; regressing out of nuisance parameters including global signal, motion parameters, and linear and quadratic drift; and band-pass filtering (0.009 Hz < f < 0.08 Hz). The entire dataset was discarded if there were not at least 100 usable frames. Correlation matrices were constructed as the absolute value of the correlation coefficient between two time courses. Binary unweighted graphs were obtained via either thresholding at R values ranging from 0.2 to 0.6, step 0.05 (cost-dependent analysis) or thresholding at cost values ranging from 0.05 to 0.45, step 0.05 (cost-independent analysis), to either include or exclude the effect of factors which would result in global connectivity changes. Graph metrics were computed using routines in Brain Connectivity Toolbox (Indiana University, Bloomington, IN) and IDL (http://www.ittvis.com, Boulder, CO).

A mixed-effects General Linear Model was used with random effect of participant and fixed effects of CHD status (Control vs. CHD comparison) or postoperative status (Pre-op vs. Post-op comparison) as variable of interest and sex, PCA at birth, PCA at scan, and scanner/site as covariates of no interest. The mixed-effects GLM accounts for correlated errors at different values of cost or correlation threshold. For nodal graph metrics, the False Discovery Rate (IBHLog procedure)^3,4 was used to correct for multiple comparisons across brain regions.

For both the cost-dependent and cost-independent analyses, CHD neonates displayed decreased network segregation. For the cost-dependent analysis, CHD neonates displayed decreased transitivity (p < 0.01) and clustering coefficient (q < 0.05), mainly in the left hemisphere and subcortically (Figure 1). For the cost-independent analysis, CHD neonates displayed decreased transitivity (p < 0.01) and modularity (p < 0.01) and increased participation coefficient (q < 0.05) mainly subcortically and in the medial backbone (Figure 2). However, no differences were found for the post-operative vs. pre-operative comparison, with the exception of increased clustering coefficient in the right lingual gyrus (q < 0.05) for the cost-dependent analysis.Complex CHD results in lasting changes to neonatal functional network topology which are not reversed or ameliorated by the effects of surgery. These functional connectivity changes involve decreases in network segregation at the nodal level (found in the cost-dependent analysis) and changes in community structure (found in the cost-independent analysis) which may reflect brain dysmaturation. Importantly, differences between cost-dependent and cost-independent based functional connectivity analyses may potentially provide imaging biomarkers that may help distinguish in utero etiologies (i.e. poor cardiac perfusion or genetic factors) that may drive poor neurodevelopmental outcomes in the CHD group^5,6.