INTRODUCTION

Chronic and excessive prenatal alcohol exposure (PAE) disrupts fetal development, leading to facial and neurological abnormalities in children¹, with cognitive impairments associated with the brain structural and functional alterations. Neuroimaging techniques may be more sensitive to detect the subtle changes induced by low to moderate levels of PAE than neurobehavioral outcome measures², providing an opportunity to quantitatively characterize the effect of prenatal alcohol consumption on brain development. In this study, we applied a method that integrates dynamic functional connectivity (DFC) using the multiplex network approach to enhance the sensitivity to subtle alterations of functional connectivity on participants subject to PAE.

METHODS

Subjects: Ninety subjects were recruited to participate this study, including 24 exposed to no alcohol during pregnancy (nPAE, M/F: 15/9, age: 7.30±0.41), 27 exposed to alcohol during trimester 1 only (PAE T1, M/F: 8/19, age:7.30±0.34), and 39 exposed to alcohol throughout pregnancy (PAE T1-T3, M/F: 22/17, age: 7.27±0.21).
Data acquisition: Resting-state functional MRI (rs-fMRI): TR/TE = 1500/33ms, flip angle=85°, voxel size=2.45×2.45×2.5 mm³ , multiband-factor=3, total time points = 206.
Generation of FC: Rs-fMRI data were preprocessed as follows: motion correction, confound regression, brain extract and normalization, band filtering, and volume censoring³. AAL parcellation was employed to extract region-wise time series. Five fMRI data segments (20 time-points each) were consistently obtained from individual subject, yielding 5 DFC networks and one static FC for each subject.
Proposed multiplex DFC network analysis: We aim to model the diffusion speed on multiplex networks and the main procedure to achieve this are as follows: (1) Five DFC networks were concatenated to construct the supra-adjacency matrix at individual-level⁴; (2) The associated Supra-Laplacian matrix was then computed from the supra-adjacency matrix; (3) Eigenvalue decomposition was applied to the supra-Laplacian matrix to extract the second smallest eigenvalue, λ₂. According to literature⁵, is proportion to the network diffusion speed, and τ=1/λ₂ represents diffusion time; (4) A range of diffusion coefficients (D_x=0.1-30) was employed to estimate λ₂, as proxies for measuring the diffusion speed associated with D_x; (5) For each subject, the maximum and minimum among the 5 λ₂ values computed from the 5 DFC networks were denoted as λ_2max and λ_2min, respectively; (6) The overall mean of the λ_2max and λ_2min were then averaged across all the three groups.
Statistical analysis: The comparisons of λ₂ between groups were analyzed using ANCOVA by controlling for age and gender, followed by multiple comparisons correction with false discovery rate (FDR). As for the statistical analyses collapsed across the predefined range of D_x values, yielding repeated measures of network diffusion speed. The LME analyses were performed as follows: (1) Group effect only; (2) Group + Gender; (3) Group + Age.

RESULTS

Figure 1 shows that greater λ₂ were obtained from the multiplex networks than the overall mean of λ_2max calculated from the single network when D_x exceeds 13.
According to Figure 1, 18 D_x values (D_x=13-30) were chosen as the predefined range, yielding 18 measures of λ₂. When comparing λ₂ between groups, lower λ₂ values were detected in the PAE T1-T3 group with than the nPAE group after correcting for multiple comparisons (p< .05) (Figure 2). However, no significant differences of λ₂ were detected between the PAE T1 and the nPAE groups.
Significantly decreased λ₂ was detected in the PAE T1-T3 group compared with the nPAE group across the predefined range of D_x values (Figure 3). Compared to the nPAE group, the reductions of λ₂ identified in the PAE T1 group were only marginally significant.
When considering the effect of gender, significantly lower λ₂ were detected in the PAE T1-T3 and the PAE T1 groups compared with the nPAE group (Figure 4). In contrast, such significant differences were not observed from the female category.
No significant age effect on λ₂ was detected between the PAE groups and the control group at any age category.

DISCUSSION

We used a novel approach, multiplex DFC network analysis, for quantifying brain network changes. Our results demonstrated its capability in characterizing subtle brain network changes in children with low-moderate PAE. The multiplex networks generally had a greater λ₂ than the single networks; this implies that network diffusion speed might be underestimated when using single networks alone. The lower λ₂ values of the PAE T1-T3 group indicates that the information processing speed could have been compromised due to PAE. Further, the marginally significant effect from PAE T1 likely suggests that brain disruptions might be less severe than PAE T1-T3. In line with literature⁶, the distinct findings between male and female groups revealed that PAE may induce gender-dependent brain disruptions and males are more susceptible to PAE.

Acknowledgements

No acknowledgement found.