Integrating brainstem and cortical functional architectures

The brainstem is a fundamental component of the central nervous system yet it is typically excluded from in vivo human brain mapping efforts, precluding a complete understanding of how the brainstem influences cortical function. Here we use high-resolution 7 Tesla fMRI to derive a functional connectome encompassing cortex as well as 58 brainstem nuclei spanning the midbrain, pons and medulla. We identify a compact set of integrative hubs in the brainstem with widespread connectivity with cerebral cortex. Patterns of connectivity between brainstem and cerebral cortex manifest as multiple emergent phenomena including neurophysiological oscillatory rhythms, patterns of cognitive functional specialization, and the unimodal-transmodal functional hierarchy. This persistent alignment between cortical functional topographies and brainstem nuclei is shaped by the spatial arrangement of multiple neurotransmitter receptors and transporters. We replicate all findings using 3 Tesla data from the same participants. Collectively, we find that multiple organizational features of cortical activity can be traced back to the brainstem.

Figure S3.Brainstem communities when γ = 1.9 | The Louvain community detection algorithm was repeated for γ = 1.9 which identified three stable communities.

Figure S1 .
Figure S1.Neurotransmitter receptor and transporter densities in the brainstem | 18 PET-derived neurotransmitter receptor and transporter density profiles are shown in the brainstem, as well as the median density across all 18 maps.Coronal (posterior view), sagittal, and axial perspectives of brainstem nuclei are shown.A legend of neuromodulatory nuclei in the brainstem is shown in the bottom row.

Figure S2 .
Figure S2.Temporal signal-to-noise ratio and parcel size.(a) Temporal signal-to-noise ratio (tSNR) is calculated as the ratio of the mean of a region's time-series to its standard deviation (prior to demeaning the time-series in the preprocessing pipeline).tSNR is shown for the cortex and brainstem together (top) and the brainstem only (bottom).Cortical tSNR ∈ [5.37, 36.51],brainstem tSNR ∈ [5.38, 15.92].(b) Parcel volume in mm 3 is shown for each brainstem nucleus.(c) Scatter plot showing the relationship between parcel volume and tSNR of brainstem (green; r = −0.45,p = 0.0004) and cortical (pink; r = −0.15,p = 0.003) regions.

Figure S4 .
Figure S4.Brainstem communities when γ = 2.2 | The Louvain community detection algorithm was repeated for γ = 2.2 which identified four stable communities.

Figure S5 .
Figure S5.Cortical communities of brainstem functional connectivity | The Louvain community detection algorithm was applied to a correlation matrix representing how similarly (Spearman's r) two cortical regions are functionally connected with the brainstem above and beyond the dominant pattern of brainstem connectivity.(a) Top: mean and variance of the z-scored rand index across 250 repetitions of the Louvain algorithm at each resolution parameter γ ∈ [0.1, 6.0].Bottom: number of communities identified for each γ.The dashed vertical line exists at γ = 0.6.(b) Cortical region × region correlation matrix representing how similarly cortical regions are functionally connected with the brainstem.Regions are ordered according to the two communities identified at γ = 0.6.(c) Community affiliations for each cortical region.(d) Brainstem weighted degree of the green (left, unimodal) community and the red (right, transmodal) community.Specifically, for each brainstem nucleus, we sum its regressed functional connectivity with all cortical regions in the green/red community.

Figure S6 .
Figure S6.Split-half analysis |The 20 participants included in the present study were randomly divided into two groups of 10 (100 repetitions).Group-average functional connectivity, cortex-to-brainstem weighted degree patterns, and brainstem-to-cortex weighted degree patterns were recalculated within these groups are correlated.

Figure S7 .
Figure S7.Replication using 100 cortical regions | Analyses were repeated using the 100-region Schaefer parcellation [96].(a) Functional connectivity and functional connectivity density distributions.(b) Brainstem-to-cortex weighted degree.(c) Cortexto-brainstem weighted degree.(d) Community affiliations of brainstem nodes under identical parameters as shown in Fig. 3. (e) Correlation between the first gradient of cortex-to-brainstem functional connectivity and the cortical functional hierarchy.

Figure S8 .
Figure S8.Replication using 3 Tesla fMRI data | Analyses were repeated using 3 Tesla fMRI data acquired in the same 20 subjects, under the Schaefer-400 parcellation [96].(a) Functional connectivity.(b) Functional connectivity density distributions.(c) Spearman correlations between 7 Tesla functional connectivity data used in the main analyses and 3 Tesla functional connectivity replication data.

Figure S9 .
Figure S9.Extending analyses to subcortical and diencephalic structures | Functional images were also acquired for the 14 bilateral FreeSurfer subcortical structures (caudate, putamen, pallidum, nucleus accumbens, thalamus, amygdala, hippocampus (not technically subcortex but allocortex)) as well as 8 bilateral diencephalic structures from the Brainstem Navigator (lateral geniculate nucleus, medial geniculate nucleus, subthalamic nuclei subregions 1 & 2).(a) For each brain regions, the centroid coordinate is plotted with colours indicating structure.Grey: 400 neocortical structures; yellow: 14 FreeSurfer subcortical structures; pink: 8 Brainstem Navigator diencephalic regions; green: 58 Brainstem Navigator brainstem nuclei.(b) Left: FreeSurfer subcortical plot of weighted degree of brainstem functional connectivity, representing how much each FreeSurfer subcortical parcel is connected with the brainstem.Each structure is labeled.Right: Brainstem Navigator brainstem and diencephalic centroid coordinates coloured according to their weighted degree of brainstem functional connectivity, representing how much each nucleus is connected with the 58 brainstem structures.The 8 diencephalic nuclei are labeled.(c) Left: region × region similarity matrix representing how similarly two non-neocortical (i.e.brainstem, subcortical, or diencephalic) regions are functionally connected with the cortex.Outlines are placed around the identified communities.Middle: point brain plot of community assignments for brainstem and diencephalic nuclei.Right: FreeSurfer surface plot of community assignments for FreeSurfer subcortical regions.(d) Left: region × region similarity matrix representing how similarly two cortical regions are functionally connected with non-neocortical regions (i.e.brainstem, subcortical, diencephalic).Outlines are shown around the seven Yeo-Krienen resting-state networks (order: control, default mode, dorsal attention, limbic, ventral attention, somato-motor, visual).Middle: coritcal surface plot of the first gradient from diffusion map embedding of how similarly cortical regions are functionally connected with non-neocortical regions.Right: FreeSurfer subcortical weighted degree patterns, calculated as the sum of a FreeSurfer subcortical region's functional connectivity with all negatively-(left) or positively-(right) scored regions of the cortical gradient shown on the left.FreeSurfer subcortical structures were plotted using the enigmatoolbox [62].

Figure S10 .
Figure S10.Community detection performance across resolution parameter γ | Top: mean and variance of the z-scored rand index across 250 repetitions of the Louvain community detection algorithm at each γ (for γ ∈ [0.1, 6.0]).Community detection solutions are considered better quality (i.e. more stable) when the mean of the z-scored rand index is high and the variance is low.Dashed vertical lines are placed at values of γ where the community detection solution is shown in the text (γ = 1.9 shown in Fig. S3, γ = 2.2 shown in Fig.S4, and γ = 2.8 shown in Fig.3).Bottom: the number of communities identified by the algorithm across values of γ.

TABLE S2 .
Neurosynth terms | Terms that overlapped between the Neurosynth database [126] and the Cognitive Atlas [86] were used in the cognitive decoding analysis.