Lipidome atlas of the adult human brain

The human brain is one of the most complex biological structures. Contemporary research often proceeds along two tracks: molecular studies of cells and brain structures (e.g., single-cell gene expression) and macroscale studies of structural and functional organization (e.g., brain imaging). Layers potentially bridging these domains, such as the brain lipidome, have received less focus. Lipids are major components of brain tissue (78% of axon myelin dry weight; 35–40% of neuron-rich gray matter). Historical and modern studies have detailed lipid classes (phosphatidylcholines, phosphatidylethanolamines, ceramides, cholesterol) and variation in class content and fatty acid composition across brain structures and cell types (e.g., sphingolipid enrichment in myelin, neutral lipid deposits in astrocytes, distinct synaptic membrane profiles). Despite progress, a systematic link between lipidome variation and the brain’s structural and functional organization across many regions remains underexplored. This study constructs a comprehensive lipidome map across 75 adult human brain regions and integrates lipid composition with myelin content, cell-type composition, functional connectivity, and information-processing hierarchy.

Prior work has characterized lipidome variation across species and brain components, identifying differences among regions, cell types, and neuronal projections in humans, macaques, and mice. Lipids influence membrane geometry, fluidity, rafts, energy metabolism, cell differentiation, signaling, inflammatory responses, and protein complex modulation. Altered brain lipidomes are linked to disorders including autism, schizophrenia, Alzheimer’s disease, and alcohol-related damage. Rodent studies show aging-related lipidome changes and regional and cell-type-specific lipid differences. However, most human brain lipidome studies sampled fewer than ten regions, limiting association with structural and functional networks. This study addresses that limitation by profiling 75 human regions (and 38 macaque regions) using untargeted HRMS and targeted MRM, integrating lipid data with sMRI-derived myelin content, rs-fMRI functional connectivity, and mRNA-based cell-type markers.

Sampling: Lipidomes were measured in 75 anatomically and functionally distinct regions from four adult neurotypical humans; 38 of these regions were profiled in three adult macaques raised in standardized environments. Regions spanned 11 anatomical structures, with 56 neocortical areas subdivided into primary, secondary, associative, and limbic cortices, plus basal ganglia, thalamus, hypothalamus, midbrain, white matter, and cerebellar gray/white matter. Mass spectrometry: Two complementary modalities were used: untargeted high-resolution mass spectrometry (HRMS) and targeted LC-MS/MS (MRM). HRMS annotated 419 lipid species across 21 classes (LIPID MAPS subclasses, with defined exceptions), aided by ion fragmentation and retention behavior; MRM targeted 216 lipids, with 169 shared across techniques. Macaque HRMS covered 394 of the 419 human HRMS lipids; MRM covered all 216 targeted lipids. Data processing and normalization: Lipid intensities (peak areas) were log10-transformed, normalized to internal standards and sample wet weight, then per-individual mean-normalized to reduce inter-individual differences and focus on regional relative representation. XCMS was used for peak detection, alignment, and imputation; stringent filters removed features with high blank signals or poor QC performance in MRM. Structural and functional measures: Myelin content per region was estimated from T1w/T2w sMRI signals (Human Connectome Project and macaque datasets), aligned to MNI space. Functional connectivity (FC) was derived from rs-fMRI for 59 regions; hierarchical information processing (HR) levels were assigned for 56 regions (primary, secondary, associative, limbic) based on published parcellations. Lipid categorization: ANOVA identified lipids with significant regional variation (BH-adjusted p < 0.01). A linear model related lipid intensities to sMRI-derived myelin content: significantly positive correlations defined myelin+ lipids; negative correlations defined myelin− lipids; remaining regionally varying lipids were “unexplained.” Lipids without significant regional differences were split by within-region variability into “housekeeping” (low variability) and “variable.” Cell-type association: RNA-seq data from 35 regions in the same four individuals were used to construct mRNA profiles for six main brain cell types (oligodendrocytes, microglia, astrocytes, OPCs, inhibitory and excitatory neurons) using 15 marker genes. Lipids were assigned to cell types based on Pearson correlation (R > 0.5) between lipid intensity and marker expression profiles. Imaging validation: MALDI imaging and ToF-SIMS were performed on selected human brain sections to spatially visualize lipid distributions and head-group ions across gray and white matter, corroborating category assignments. Additional validation used sorted Thy1-ChR2-YFP+ mouse pyramidal neurons vs unlabeled cells. Comparative analysis: Cross-species comparisons evaluated consistency of lipid categories and regional profiles between humans and macaques; mouse datasets were used to assess conservation of cell-type-specific lipid classes.

• Broad regional variation: In humans, 391/419 (93%) HRMS lipids varied significantly among regions (ANOVA, BH-adjusted p < 0.01). In macaques, 279/394 (71%) varied. • Myelin association: Lipidome PC1 correlated strongly with sMRI-derived myelin content (humans R = 0.78; macaques R = 0.77; p < 0.0001). Across individual lipids: 65% (N = 274) positively correlated with myelin (myelin+), 17% (N = 71) negatively correlated (myelin−), totaling 82% myelin-associated; 11% (N = 46) unexplained by myelin; 5% (N = 20) housekeeping; 2% (N = 8) variable. • Technique and species concordance: Normalized lipid profiles correlated between HRMS and MRM and between humans and macaques (Mann–Whitney U p < 0.0001). Category assignments were consistent across species for 84% (HRMS) and 88% (MRM) of overlapping lipids; myelin+ showed highest concordance. • Biochemical differences by category: Myelin+ lipids were enriched for specific classes (e.g., lysophosphatidylcholines, phosphatidylethanolamines), exhibited higher predicted membrane fluidity, and tended toward polyunsaturated chains in PE/PE-P; myelin− showed opposite trends, with sphingolipid enrichment. Housekeeping lipids were enriched for free fatty acids, often short and unsaturated. • Cell-type associations: 353/419 (84%) lipids correlated with at least one cell-type marker profile. Myelin+ lipids predominantly aligned with oligodendrocyte markers; myelin− with inhibitory/excitatory neuron markers. Unexplained lipids associated with astrocytes, OPCs, and inhibitory neurons. • Hierarchical processing (HR): Lipidome PC2 correlated with HR levels (R = 0.58, p = 3.4×10⁻⁶), with piriform cortex as an outlier. Sixty lipids correlated significantly with HR (BH p < 0.05): 26 positively (HR+), 34 negatively (HR−). HR+ lipids were overrepresented in the myelin category, enriched in phosphatidylcholines, and contained more PUFA residues (notably omega-3 DHA). HR− lipids were enriched in sphingomyelins, had saturated/oligo-unsaturated residues (≤4 double bonds), and showed omega-6 enrichment. • Functional connectivity (FC): The first PC of FC distances correlated with lipidome PC2 (R = 0.35, p = 0.006). 76 lipids showed positive correlations with FC (Mantel nominal p < 0.005). Myelin+ lipids exhibited significantly stronger FC correlations than other categories (Mann–Whitney U p = 1.5×10⁻⁴). Sulfatides, hexosylceramides, and diacylglycerols (enriched in myelin+) had higher FC correlations (BH-corrected p < 0.05). • Validation: Findings were supported by macaque brain analyses, targeted MRM, MALDI and ToF-SIMS imaging, and enrichment of neuron-assigned lipids in sorted Thy1-ChR2-YFP+ mouse neurons (Wilcoxon p = 0.0077).

This work bridges molecular lipid composition with macroscale brain structure and function. The strong alignment between lipidome variation and myelin content confirms myelin’s dominant biochemical signature across regions. Beyond myelination, lipid categories show distinct class distributions, chain lengths, and unsaturation, with predicted membrane fluidity differences consistent with white vs gray matter properties. Integrating mRNA-based cell-type markers reveals robust correspondence between lipid profiles and cellular composition, particularly oligodendrocyte-associated myelin+ lipids and neuron-associated myelin− lipids. Importantly, lipidome gradients relate to brain functional architectures: hierarchical information processing and resting-state connectivity. HR-associated lipids segregate by biochemical class and omega-3/omega-6 composition, implying underlying gradients in astrocyte and neuronal representation or subtype composition along processing hierarchies. Myelin+ lipids’ stronger association with FC suggests that lipid composition of myelinated axonal tracts underpins functional network coupling, consistent with known relationships between intracortical myelin and functional connectivity. Together, these results demonstrate that the brain lipidome reflects and can help infer structural and functional organization.

The study provides a comprehensive atlas of the adult human brain lipidome across 75 regions, integrating lipid composition with myelin content, cell-type composition, information-processing hierarchy, and functional connectivity. Most lipids exhibit myelin-associated regional patterns, with distinct biochemical and cell-type signatures. Myelin-independent lipid variation relates to processing hierarchy, while myelin-associated lipids align with resting-state network connectivity. This atlas establishes a baseline for future investigations into lipid alterations in neurological and psychiatric conditions and highlights lipids as informative markers linking microscale cellular composition to macroscale brain architecture. Future research should expand sample sizes and demographic diversity, incorporate additional lipid classes (e.g., cardiolipins, gangliosides), enhance spatial resolution and subcortical/white matter coverage, and integrate single-cell and spatial multi-omics to refine cell subtype associations.

• Sample size: Only four human individuals were analyzed, potentially limiting generalizability and power to detect region-specific effects. • Category misassignment risk: Underpowered to identify alterations confined to single regions may misclassify some lipids as housekeeping or variable. • Lipid class coverage: Important classes such as cardiolipins and gangliosides were not included. • Cell-type assignment: Indirect deconvolution using mRNA markers cannot resolve cellular subtypes with correlated expression and may leave some lipids unassigned. • Regional bias: Sampling emphasized neocortex, with less detailed coverage of subcortical structures, axonal tracts, and infracortical white matter. • Spatial resolution: LC-MS lacked intra-regional spatial detail; imaging was applied only to selected regions.