Dynamic spatiotemporal patterns of brain connectivity reorganize across development

Late human development is characterized by the maturation of high-level functional processes, which rely on reshaping of white matter connections, as well as synaptic density. However, the relationship between the whole-brain dynamics and the underlying white matter networks in neurodevelopment is largely unknown. In this study, we focused on how the structural connectome shapes the emerging dynamics of cerebral development between the ages of 6 and 33 years, using functional and diffusion magnetic resonance imaging combined into a spatiotemporal connectivity framework. We defined two new measures of brain dynamics, namely the system diversity and the spatiotemporal diversity, which quantify the level of integration/segregation between functional systems and the level of temporal self-similarity of the functional patterns of brain dynamics, respectively. We observed a global increase in system diversity and a global decrease and local refinement in spatiotemporal diversity values with age. In support of these findings, we further found an increase in the usage of long-range and inter-system white matter connectivity and a decrease in the usage of short-range connectivity with age. These findings suggest that dynamic functional patterns in the brain progressively become more integrative and temporally self-similar with age. These functional changes are supported by a greater involvement of long-range and inter-system axonal pathways.

(2019). Supporting information for "Dynamic spatio-temporal patterns of brain connectivity reorganize across development." Network Neuroscience. Advance publication. https://doi.org/10.1162/netn_a_00111 had two datapoints in the group as not to be influenced by non-independent results, 2) the correlation analysis was performed for the 69 subjects (87 scans). To account for longitudinal datapoints we averaged the subjects with multiple datapoints. For the independent analysis on the longitudinal data please see below the section on Repeated Measure Correlation.

Information about the CCs
We calculated for every subject scan information about CCs yielded from the multilayer network. For the number of CCs each subject is part of, we found no correlation with age (r = -0.05, p-value =0.63). In terms of the spatial and temporal span of the CCs per subject we also found no significant differences with age. Spatial size (r = 0.16, p-value = 0.131) and Temporal size (r = 0.12, p-value = 0.27).

Datasets
The spatio-temporal network was constructed according to (1). Briefly, as a functional dataset we used the individual subject BOLD timeseries for our 69 subjects (see above for details on subject age). The signal was pre-processed, z-scored and converted to binarised point-process at threshold of two standard deviation(1). As a structural dataset, we used the template structural connectivity matrix from the original DSI cohort of 68 subjects (1). We considered a robust structural connection if at least 50% of subjects possessed it. This approach and the Vohryzek, J., Griffa, A., Mullier, E., Friedrichs-Maeder, C., Sandini, C., Schaer, M., Eliez, S., & Hagmann, P.
(2019). Supporting information for "Dynamic spatio-temporal patterns of brain connectivity reorganize across development." Network Neuroscience. Advance publication. https://doi.org/10.1162/netn_a_00111 DSI dataset were chosen to minimise false negative and positive connections which are ultimately introduced through the white-matter tracts reconstruction (2).

Metadata
For the analysis on connection length, we used the original structural dataset of 69 subjects.
We used the averaged connection length matrix across the subjects in order to label every edge in the multi-layer network with a length measure. To normalise for inter-subject head variability, we corrected the short and long edge-length threshold by a reference head size. To do so, we computed average InterCranial Volume (ICV) for a group of adult subjects (68 original subjects (1)) and used it to multiply thresholds for each subject as followed !"#$%&' )*+ ,-".' /01"2 )*+ 3 . Furthermore, for the inter-system edge analysis, we attributed to each cortical region a label with the associated functional system(3).

Motion Correction
In order to address motion-related confounds in the BOLD signal, we used quality measures, namely the Framewise Displacement (FD) and DV, and novel scrubbing approach in the network space. Firstly, we computed FD from six motion signals (three translations and three rotations). The rotational signal was converted from radians to millimetres (mm) as a displacement on a 50 mm radius sphere(4). On the other hand, Derivative of Variance (DV) was defined as the root mean square of the differentiated BOLD signal prior to pre-processing steps that might alter the signal such as nuisance parameters regression, linear detrending and bandpass filtering. We set the thresholds to FD=0.4mm and DV=25 and whenever corrupted timepoints exceeded 10% of the subject's recording timepoints for either FD or DV, the subject Vohryzek, J., Griffa, A., Mullier, E., Friedrichs-Maeder, C., Sandini, C., Schaer, M., Eliez, S., & Hagmann, P.
(2019). Supporting information for "Dynamic spatio-temporal patterns of brain connectivity reorganize across development." Network Neuroscience. Advance publication. https://doi.org/10.1162/netn_a_00111 was excluded. To be more rigorous, a novel scrubbing technique was performed in the graph space where we discarded all activations within the multilayer graph with spatial span of less than six regions and temporal span of less than two timepoints. Furthermore, we excluded all connected components implicated in corrupted timepoints and so eliminated any functional activations that might have arisen several points before and after the corrupted timepoints due

Repeated Measure Correlation
In addition to our main analysis we wanted to quantify intra-subject variability differences with age. To do so, we performed a repeated measure correlation (rmcorr). This method has been introduced to measure the intra-subject linear strength by removing the inter-subject variability and fitting the best parallel linear line to the individual subjects for varying Vohryzek, J., Griffa, A., Mullier, E., Friedrichs-Maeder, C., Sandini, C., Schaer, M., Eliez, S., & Hagmann, P. (2019). Supporting information for "Dynamic spatio-temporal patterns of brain connectivity reorganize across development." Network Neuroscience. Advance publication. https://doi.org/10.1162/netn_a_00111 intercepts (5). For a subgroup of 17 subjects with repeated data acquisition with age. We found no significant results for the probability of short and long fibre usage with age (rmcorr = 0.173, p-value = 0.48 and rmcorr = 0.143, p-value = 0.55) and significant increase for the intra-subject variability for the proportion of inter-system edges with age (rmcorr = 0.569, p-value = 0.01).

Test-Retest Analysis
In order to address the reliability of the measures across different scanning sessions, we performed the same analysis on HCP dataset of 97 subjects. The resting-state fMRI dataset of each HCP subject consists of four runs of approximately 15 minutes each, with two sessions of two runs. Each session included oblique axial acquisitions alternated between phase encoding in right-to-left (RL) direction for run1 and phase encoding in left-to-right (LR) direction for run2. For every run, we calculated the nodal map of SD and STD from all the CCs of the 97 subjects. We compared all the sessions between each other using Pearson's correlation. The values can be found in Table 1.
(2019). Supporting information for "Dynamic spatio-temporal patterns of brain connectivity reorganize across development." Network Neuroscience. Advance publication. https://doi.org/10.1162/netn_a_00111 session were relatively high (minimum Pearson's correlation value = 0.9339), indicating a high reproducibility of SD and STD measures across different fMRI acquisitions, over the same set of subjects. To further verify the reproducibility of both measures, we plotted the SD and STD values on a standard cortical surface for all the four runs, as demonstrated in Figure   6 and 7. The cortical patterns obtained from different fMRI session were visually very similar.
(2019). Supporting information for "Dynamic spatio-temporal patterns of brain connectivity reorganize across development." Network Neuroscience. Advance publication. https://doi.org/10.1162/netn_a_00111 Correlation between different scanning runs. S1 and S2 represents the scanning sessions with RL being right-to-left and LR left-to-right phase encoding.