Typical and disrupted brain circuitry for conscious awareness in full-term and preterm infants

Abstract One of the great frontiers of consciousness science is understanding how early consciousness arises in the development of the human infant. The reciprocal relationship between the default mode network and fronto-parietal networks—the dorsal attention and executive control network—is thought to facilitate integration of information across the brain and its availability for a wide set of conscious mental operations. It remains unknown whether the brain mechanism of conscious awareness is instantiated in infants from birth. To address this gap, we investigated the development of the default mode and fronto-parietal networks and of their reciprocal relationship in neonates. To understand the effect of early neonate age on these networks, we also assessed neonates born prematurely or before term-equivalent age. We used the Developing Human Connectome Project, a unique Open Science dataset which provides a large sample of neonatal functional MRI data with high temporal and spatial resolution. Resting state functional MRI data for full-term neonates (n = 282, age 41.2 weeks ± 12 days) and preterm neonates scanned at term-equivalent age (n = 73, 40.9 weeks ± 14.5 days), or before term-equivalent age (n = 73, 34.6 weeks ± 13.4 days), were obtained from the Developing Human Connectome Project, and for a reference adult group (n = 176, 22–36 years), from the Human Connectome Project. For the first time, we show that the reciprocal relationship between the default mode and dorsal attention network was present at full-term birth or term-equivalent age. Although different from the adult networks, the default mode, dorsal attention and executive control networks were present as distinct networks at full-term birth or term-equivalent age, but premature birth was associated with network disruption. By contrast, neonates before term-equivalent age showed dramatic underdevelopment of high-order networks. Only the dorsal attention network was present as a distinct network and the reciprocal network relationship was not yet formed. Our results suggest that, at full-term birth or by term-equivalent age, infants possess key features of the neural circuitry that enables integration of information across diverse sensory and high-order functional modules, giving rise to conscious awareness. Conversely, they suggest that this brain infrastructure is not present before infants reach term-equivalent age. These findings improve understanding of the ontogeny of high-order network dynamics that support conscious awareness and of their disruption by premature birth.

phase-encoding-direction-induced distortion; 2) Motion correction: realigns the timeseries to correct for subject motion by using a 6 DOF FLIRT (Oxford Centre for Functional MRI of the Brain Fs Linear Registration Tool) registration of each frame to the single-band reference image; 3) Aligns the original EPI data (rs-fMRI data) to Montreal Neurological Institute (MNI) template space: EPI to T1w from FLIRT BBR, fine tuning of EPI to T1w with BBR-register, nonlinear T1w to MNI template; 4) Intensity normalization to mean of 10000 and bias field removal; 5) Temporal high-pass filter: 150s high-pass cut-off; 6) Denoising: removes artefactual or "bad" components using ICA-FIX to automatically. Detailed pre-processing procedure can be found in Glasser et al. 5 . Additionally, we performed a temporal low-pass filter (0.08 Hz low-pass cut-off) on the denoised rs-fMRI data and removed the first five volumes.
Supplementary Fig. 1 provides a schematic of the processing steps for HCP fMRI data. As the selection of the subset of HCP had controlled head motion (i.e., exclusion of participants that had any fMRI run in which more than 50% of TRs had greater than 0.25mm framewise displacement), and adults generally have smaller maximal head motion than neonates, 6 we did not apply the same scrubbing method used in the dHCP dataset to adult data. To assess the effect of this, we compared the mean framewise displacement (FD) value in adults and neonates before/after the scrubbing procedure. 7,8 The FD value indexes the movement of the head from one volume to the next and is defined as the sum of the absolute values of the differential realignment estimates (the six realignment parameters). It has been widely used to index head movement and exclude subjects of high motion. [9][10][11] Independent-samples t-tests showed that adults had significantly lower head motion compared to neonates before (t (491.45) = -12.49, p < 0.001) and even after scrubbing (t (580.56) = -9.92, p < 0.001; Supplementary Fig. 4).

Network definition
We first aligned these ROIs with 40-week dHCP T1w template. 4 This involved: 1) trimming the dura from the 40-week dHCP T1w template, and the cerebellum from both the 40-week dHCP T1w template and MNI T1w template; 2) aligning the 40-week dHCP T1w template to MNI T1w template using non-linear registration (ANTs SyN) (See Supplementary Fig. 2 for the registration accuracy between them); 3) applying the warp file generated in the last step to the ROIs in MNI space with 40-week dHCP T1w template as a reference. In the next step, we needed to align these ROIs in 40-week dHCP T1w template space with neonate native space.
We inverted the func-to-template warp provided by dHCP group and applied this inverted warp to ROIs in the 40-week dHCP T1w template space. Thus, we obtained ROIs in each neonate native functional space ( Supplementary Fig. 3). For adults, as the denoised HCP data had been aligned to MNI space, we used these ROIs in MNI space directly.

Data analyses
Hierarchical clustering analyses. We captured the structure of the three networks in different groups with hierarchical clustering analysis, 12,13 which has proven informative in prior studies. 14,15 This hierarchical clustering algorithm builds up an entire cluster tree in which neighbouring regions are joined if their similarity is maximal among all pairs of neighbouring regions. Here, we used the time-course extracted from the 19 ROIs as input to access the hierarchical relationship among the ROIs. For the neonate data, we first calculated initial pairwise distance between ROIs using one minus the linear correlation between the scrubbed time-courses extracted from the 19 ROIs at the individual level. For adults, the initial pairwise distance between ROIs was calculated using one minus the linear correlation between the timecourses of 1195 time points extracted from the 19 ROIs at the individual level. Then, we averaged the pairwise distances between ROIs within each group to get the group-level pairwise distances, which were submitted to hierarchical clustering analysis to create a hierarchical cluster tree of the 19 ROIs for each group respectively. The cophenetic correlation coefficient was used to create a dendrogram for each group. The length of each C link in the dendrogram represents the distance between regions/clusters.

Multidimensional scaling analysis.
Non-metric multidimensional scaling (MDS) was also used to facilitate visualizing the similarity of ROIs functional response for adults and neonate groups. The non-metric MDS performs non-metric multidimensional scaling on the dissimilarity matrix of item−item (i.e., ROI−ROI dissimilarity matrix) to compute a configuration. 16 Then, the Euclidean distances between items (i.e., ROIs) in the configuration were obtained. The difference between the monotonic transformed dissimilarities in the item−item (i.e., ROI−ROI) matrix and the Euclidean distances between items (i.e., ROIs) in this configuration were minimized and items (ROIs) were represented in a low-dimensional space (i.e., a 2-D space). The ROI−ROI dissimilarity matrix (one minus the linear correlation between the time-courses) for each group from the hierarchical clustering analysis was submitted to non-metric MDS analysis implemented in MATLAB.

Comparison of head motion in neonates and adults. Independent-sample t-tests were applied
to compare the head motion in the adults and neonates before/after the scrubbing procedure.
We found that adults had significantly lower head motion compared to neonates before (t (491.45) = -12.49, p < 0.001) and after (t (580.56) = -9.92, p < 0.001) scrubbing procedure ( Supplementary Fig. 4). A paired-t test was applied to detect the difference in head motion in neonates before and after the scrubbing procedure. Results showed that neonates had significantly lower head motion after scrubbing relative to before scrubbing (t (427) = -9.69, p < 0.001) ( Supplementary Fig. 4). Bonferroni correction for multiple comparisons was applied to statistical results.

Supplementary Figure 4. Head motion in neonates and adults.
The red dot indicates the mean framewise displacement value in each group. Independent-samples t-tests were applied to detect the difference between the adults and neonates. A paired-t test was applied to detect the difference between the neonates before and after scrubbing procedure. Abbreviations: FD, framewise displacement; ** = p < 0.005.

Comparison of head motion in neonate groups after scrubbing and adults.
We conducted a one-way ANOVA to compare the difference in head motion between the adults and neonate groups after the scrubbing procedure and found a significant main effect of group (F (3, 600) = 16.46, p < 0.001). Independent-sample t-tests were applied to compare head motion between every two groups. We found that that adults had significantly lower head motion relative to the were applied to detect the difference between every two groups. Abbreviations: FD, framewise displacement; TEA, term-equivalent age; ** = p < 0.005.

High-order networks functional connectivity in adults.
In adults, a 2 × 3 repeated measure showed a significant main effect of type of FC (F (1, 175) = 2323.00, p < 0.001), which was driven by higher overall connectivity for the within-relative to between-network (t (175) = 48.11, p < 0.001). We also found a main effect of network (F (1.93, 337.773) = 37.50, p < 0.001), that was driven by lower overall connectivity for the DMN relative to the DAN (t(175) = -8.72, p < 0.001) and ECN (t (175) = -3.71, p < 0.001) and lower overall connectivity for the

Comparison of DMN-frontoparietal functional connectivity in neonates and adults.
To investigate between-network connectivity in neonates relative to adults, we created a GLM that controlled for head motion, and compared each neonate group to the adult group.  (Fig. 7B). Similarly, to full-term neonates, these results demonstrated that the DMN was more functionally differentiated from DAN and ECN, suggesting a stronger reciprocal relationship in adults relative to preterm neonates scanned at TEA. We also compared between-network connectivity in preterm neonates scanned before TEA and adults using a GLM that controlled for head motion, although we did not observe a reciprocal relationship between DMN and frontoparietal network in that neonate group. We

Supplementary Tables
Supplementary Table 1 He et al. 31