Spatial and spectral trajectories in typical neurodevelopment from childhood to middle age

Abstract

Detailed characterization of typical human neurodevelopment is key if we are to understand the nature of mental and neurological pathology. While research on the cellular processes of neurodevelopment has made great advances, in vivo human imaging is crucial to understand our uniquely human capabilities, as well as the pathologies that affect them. Using magnetoencephalography data in the largest normative sample currently available (324 participants aged 6-45 years), we assess the developmental trajectory of resting-state oscillatory power and functional connectivity from childhood to middle age. The maturational course of power, indicative of local processing, was found to both increase and decrease in a spectrally dependent fashion. Using the strength of phase-synchrony between parcellated regions, we found significant linear and nonlinear (quadratic and logarithmic) trajectories to be characterized in a spatially heterogeneous frequency-specific manner, such as a superior frontal region with linear and nonlinear trajectories in theta and gamma band respectively. Assessment of global efficiency revealed similar significant nonlinear trajectories across all frequency bands. Our results link with the development of human cognitive abilities; they also highlight the complexity of neurodevelopment and provide quantitative parameters for replication and a robust footing from which clinical research may map pathological deviations from these typical trajectories.

Keywords: Functional connectivity; MEG; Maturational trajectories; Neurodevelopment; Phase synchronisation; Power spectral density; Resting state.

Figure 1.

Grand-averaged functional connectivity (wPLI). Within each frequency quadrant, connections falling within the 95th percentile are plotted in the circle plots and glass brains. The circle plots are arranged anatomically such that left hemisphere regions appear on the left of the plot. Regional connections (edges) within the glass brains are scaled within band, with thicker edges indicating a greater connection strength. The spheres indicating individual regions (nodes) are scaled by connectivity strength, with larger spheres indicating greater connectivity values to connected regions, within the 95th percentile of connections. Adjacency matrix labels indicate the lobular segregation of the matrix, in the order of frontal (F), temporal (T), subcortical (S), parietal (P), and occipital (O). Note that adjacency matrix color axes are scaled independently by band.

Figure 2.

Typical neurodevelopmental trajectories assessed with wPLI-S. Each quadrant (band) contains the following sections: Section (i) plots curve types that were found to be statistically significant on a regional basis. Any regions in gray indicate regions with nonsignificant fits. Section (ii) presents the standard deviation (SD) of the model, indicating to what extent the data change over the developmental course. Section (iii) plots axial brain images indicating the gradient between age and wPLI-strength. Hotter colors indicate a steeply increasing gradient (e.g., the upward portion of a quadratic or log fit) and cooler colors indicate a steeply decreasing gradient; note the change in color axes with frequency band. Regions in gray are regions where the best fit was found not to be significant. Arrows on the first brain in (iii) correspond, in color, to the graphs plotted in (iv) and (v), plotting the relationship between age and wPLI-strength, with the colored line indicating the model that best characterizes that regional relationship. See Figures S7 and S8 (Hunt et al., 2019) for videos depicting gradient change from 6 to 45 years of age. Video S7 presents gradients only for significantly characterized curves, and S8 presents all gradients. Table S10 (Hunt et al., 2019) details the best fitting model coefficients for each region and frequency band. We also present a figure focused on subcortical trajectories in the supplementary material, Figure S1 (Hunt et al., 2019).

Figure 3.

Typical neurodevelopmental trajectories assessed with global efficiency (GE). Each plot presents the statistically significant curve fits following our cross-validation analysis. Theta band GE is significantly characterized by a linearly increasing fit, whereas the alpha and gamma bands were characterized by a quadratic curve. The beta band was significantly characterized by a logarithmic curve.

Figure 4.

The relationship between age and power spectral density. A–D present regional correlations between PSD and age for each frequency band. Hotter colors indicate a positive correlation and cooler colors indicate the opposite relationship. E–H plot the correlation between PSD and age for exemplar regions (indicated by white arrows in A–D). See Figure S3 (Hunt et al., 2019) for the results of an analysis investigating the extent to which these differences arise because of age-related changes in SNR (A–D) or head motion (E–G).

Figure 5.

Assessing the influence of age on signal-to-noise ratio. The upper left graph plots the correlation between a proxy for global signal-to-noise ratio (SNR; Frobenius norm of the lead fields) and age (nonsignificant, p > 0.05). The brain plots expand on this relation by plotting the Pearson correlation between SNR and age on a regional basis. The white vertical bars on color scale indicate threshold for significant correlation (r = ±0.19). See Figure S5A (Hunt et al., 2019) for a greater exploration of this effect.

Figure 6.

Processing pipeline schematic. MEG data were first bandpass filtered, and independent component analysis (ICA) was used to de-noise the data. Following this, data were epoched into 10-s “trials” (for processing efficiency) and beamformed to parcels of the AAL atlas. Functional connectivity (blue): Data were frequency filtered into bands and functional connectivity calculated using the weighted phase lag index (wPLI). wPLI-S and GE were entered into a cross-validation curve/model fitting algorithm, which fit either linear, quadratic, or logarithmic curves to data, quantifying changing graph properties with age. Power spectral density (green): Using atlas-beamformed data, we used Welch’s method to calculate PSD. These values were binned into frequency bands and Pearson’s correlation was used to infer relations between PSD and age. Signal-to-noise (red): Using the lead fields generated to solve the forward problem, we investigated whether any relation between SNR and age existed, using both global and regional SNR/lead field values. Note that figures in brackets appear in the Supporting Information section (Hunt et al., 2019).