Fig 1.
Mapping cerebral blood perfusion across participants.
Cerebral blood perfusion is estimated using arterial spin labeling (ASL) from the HCP Lifespan studies, including 627 participants (337 females) from the HCP-D (development) and 678 participants (381 females) from the HCP-A (aging). Each dot corresponds to a participant’s mean brain (grayordinates) blood perfusion level (male: blue, female: red) (See S1 Fig for the mean perfusion trajectory within white matter mask). Sex-stratified generalized additive models for location, scale and shape (GAMLSS) [76] are used to model age-related changes in blood perfusion.
Fig 2.
Areal organization of blood perfusion.
(a) Normative blood perfusion map on lateral and medial views of the inflated and 2D flat cortical surfaces (fsLR). Borders and areal names of the multi-modal Glasser parcellation are overlaid on the flat surface [86]. The figure highlights the non-uniform distribution of perfusion scores across the cortex, with areas that have sharp gradient compared to their underlying perfusion score levels (e.g., LIPv and MT/MST) (see S1 and S2 Tables). (b) Subcortical perfusion is shown on a T2-weighted group average template. Quantification of perfusion scores according to the Tian-S4 subcortical parcellation is shown in S7 Fig [87]. A complete list of subcortical parcels is provided in S3 Table. (c) Correlations between perfusion (y-axis) and measurements of energy consumption (x-axis). Left: Correlation with glucose consumption (; r = 0.74,
) [30]. Middle: Correlation with oxygen consumption (
; r = 0.62,
) [30]. Right: Correlation with resting-state functional MRI connectivity strength (a measure of “hubness”) (r = 0.52,
; see Methods for more details). (d) The transcriptomic signature of cortical layer IV [88] is positively correlated with perfusion (r = 0.52,
). We do not find correspondence between blood perfusion and transcriptomic signature of supragranular (I–III) and infragranular (V–VI) cortical layers (S8 Fig).
Fig 3.
Blood perfusion co-localizes with molecular and cellular features.
(a) The top ten GO biological processes linked with gene sets correlated with perfusion [104]. Terms are sorted based on their associated category-scores. For a complete list of genes associated with a biological pathway, category-scores and statistics, see S4 Table. (b) Cell type enrichment analysis of the cortical perfusion score map [23]. For list of genes associated with a cell type, category scores and statistics, see S5 Table. Red lines indicate statistically significant associations (after FDR-correction). (c) Comparison of perfusion with maps of neurotransmitter receptors (from PET imaging) and neuropeptide receptors (from transcriptomics) [18,106]. Colors indicate directions of association (red: positive, blue: negative) and asterisks indicate significant associations. See S9 Fig for results obtained using a multivariate model with dominance analysis. Methodological details about each neurotransmitter receptor PET map are provided in S6 Table.
Fig 4.
Blood perfusion across development.
(a) Whole-brain perfusion (y-axis), stratified by age (x-axis) and biological sex (male: blue, female: red). S7 Table shows statistical comparisons between males and females in each age bin. (b) Age-effect perfusion map estimated using data from 627 participants in the HCP-Development dataset (5–22 years). Linear age coefficients () are derived from a linear model (
), applied at each vertex/voxel i. Subcortical age-effect values are shown in S11A Fig. Sex-effect and coefficient significance maps are provided in S12 Fig. S13 Fig shows vertex/voxel-wise Spearman correlation maps of blood perfusion and age, stratified by biological sex. See S14 and S15 Figs for sex-stratified non-linear modeling of perfusion changes. (c) Mean age-effect on cerebral perfusion, stratified into unimodal and transmodal cortex (unimodal: visual and somatomotor networks, shown in green; transmodal: frontoparietal and default mode networks, shown in yellow; canonical intrinsic functional networks from Yeo et al. [124]). The age-effect difference (
) is significant using both t-tests (t = 12.97,
) and non-parametric spin tests (
). (d) Correlation between cerebral blood perfusion changes (y-axis) and cortical thickness changes (x-axis) (r = 0.33,
). (e) Correlation between cerebral blood perfusion changes (y-axis) and developmental cortical expansion (x-axis) (r = −0.38,
) [125]. In (c), (d) and (e) maps are parcellated according to Schaefer-400 atlas [126].
Fig 5.
(a) Age-effect perfusion map estimated using data from 678 participants in the HCP-Aging dataset (36–100 years). Linear age coefficients () are derived from a linear model (
), applied at each vertex/voxel i. Subcortical age-effect values are shown in S11B Fig. Sex-effect and coefficient significance maps are provided in S16 Fig. S17 Fig shows vertex/voxel-wise Spearman correlation maps of blood perfusion and age, stratified by biological sex. See S18 Fig for sex-stratified non-linear modeling of perfusion changes. (b) The age-effect pattern aligns with arterial border-zone cortical regions. The border-zone areas are described based on: (1) Left: First principal component of estimated arterial transit time maps (greater values correspond to regions with later arterial transit time; see S19 Fig for analysis details). In the figure, “ATT” stands for arterial transit time. (2) Right: Intersections of major cerebral arterial territories derived from a stroke-based atlas [142,143]. (c) Correlation between cerebral blood perfusion age-related changes (x-axis) and arterial transit time scores (y-axis) (r = −0.71,
). Red dots represent border-zone regions defined according to (2). (d) Using partial least squares (PLS) analysis, we find a significant latent variable that accounts for 82.0%, and 90.6% of the covariance between cortical blood perfusion and biomarkers (S20 Fig) in male and female participants (in both cases:
). The bar plot visualizes the contribution of individual biomarkers to the first latent variable. The significance of each biomarker’s contribution to the overall pattern is assessed by bootstrap resampling (
bootstraps; see S21 Fig). (e) The brain loadings of the first latent variable are shown for both groups. These maps correlate with the arterial transit time score map shown in (b) (male: r = 0.47,
, female: r = 0.52,
). Brain maps in (a), (b) and (e) are shown on the inflated and 2D flat cortical surfaces (fsLR); maps in (b)-right and (e) are parcellated according to the Schaefer-400 functional atlas [126]. Results of PLS analysis after regressing out the linear effect of age and sex from the perfusion maps and the biomarkers are presented in S22 Fig.
Fig 6.
Cerebral blood perfusion and susceptibility to neurodegeneration.
(a) Comparison of perfusion with eight pathology-confirmed disease atrophy maps (voxel-based morphometry; VBM data from Harper et al. [159]), both parcellated using the Schaefer-400 functional atlas [126]. Colors indicate directions of association (red: positive, blue: negative) and asterisks indicate significant associations. (b) Late-onset Alzheimer’s disease (r = −0.43; FDR-corrected, ), (c) TDP-43C (r = −0.61; FDR-corrected,
), (d) and dementia with Lewy bodies (r = −0.37; FDR-corrected,
) show significant negative correlations with the perfusion map, indicating that these disease-specific atrophy patterns co-localize with areas of lower cerebral perfusion.
Fig 7.
Data-driven mapping and functional annotation of vascular territories.
(a) The first two gradients of blood perfusion covariance matrix are shown on the inflated and 2D flat cortical surfaces (fsLR), and subcortical regions are shown on a subcortical mask (from HCP-S release). (b) The arterial territories for posterior cerebral artery, middle cerebral artery and anterior cerebral artery from an independent stroke-based atlas [142,143]. (c) Correlation values between cortical gradient maps and Neurosynth meta-analytic atlas (x-axis: correlations with gradient 1; y-axis: correlations with gradient 2). Here, only functional/cognitive terms with correlation values exceeding thresholds determined by spatial auto-correlation-preserving permutation nulls (
> 0.05) are annotated; for a list of 124 included Neurosynth functional/cognitive terms refer to S8 Table.