Dynamic lipidome alterations associated with human health, disease and ageing

The study addresses how the human plasma lipidome varies across individuals and changes over time during health and disease, including ageing, insulin resistance (IR), and acute inflammation such as respiratory viral infections (RVI). Lipids are chemically diverse molecules central to membrane structure, signalling and energy storage, and dyslipidaemia is implicated in numerous diseases (metabolic syndrome, type 2 diabetes, cardiovascular, neurodegenerative diseases). Despite advances in genomics and proteomics, lipidomics has been less explored because of lipid chemical complexity. The authors aim to longitudinally profile a broad set of complex lipids in humans to uncover individualized lipid signatures, associations with clinical and immunological measures, dynamics during acute and chronic perturbations, and to identify lipid subclasses (e.g., small vs large TAGs; ester- vs ether-linked PEs) with distinct physiological roles relevant to health and disease.

Prior work has established lipid diversity and roles in homeostasis and inflammation (eicosanoids, endocannabinoids), and linked dyslipidaemia to multiple diseases. High-throughput omics have mainly focused on sequencing and proteomics, with fewer comprehensive lipidomics studies due to analytical challenges. Mass spectrometry-based lipidomics with differential mobility spectrometry has enabled quantitation across lipid classes. Previous studies in related cohorts identified molecular signatures for metabolic and inflammatory states and exercise, but detailed longitudinal lipid dynamics, subclass-specific behaviours (e.g., ether-linked phospholipids), and personalized lipidome features over years remained underexplored. The authors build on this by integrating lipidomics with clinical labs, microbiome and cytokines over extended follow-up.

• Cohort and sampling: 112 adult participants (insulin resistant or insulin sensitive) were followed longitudinally for 2–9 years (average 3.2 years), with a median of ~10 timepoints per participant. Total 1,539 plasma samples were analyzed, collected quarterly during health and densely during acute events (3–7 collections over ~3 weeks) such as RVI or stress. Clinical labs (50 analytes), medical records, and 62 cytokines/chemokines/growth factors were measured at each timepoint. IR/IS classification used steady-state plasma glucose (SSPG) assays in 69 participants (IR if SSPG >150 mg/dL, IS if ≤150 mg/dL).
• Lipidomics platform: Targeted quantitative lipidomics using a Sciex QTRAP 5500 triple-quadrupole with differential mobility separation (Lipidyzer). The assay targeted >1,000 lipids across 16 subclasses (FFA, TAG, DAG, CE, PC, LPC, PE, PE-O, PE-P, LPE, SM, PI, CER, HCER, LCER, DCER). Small and large TAGs were analysed separately (≤48 vs ≥49 total FA carbons). A mix of 54 deuterated internal standards across nine subclasses was spiked for quantification; additional classes (e.g., PI) were normalized via correlation to available spikes. Samples were randomized for extraction and acquisition. Data conversion (wiff→mzML), extraction, and identity matching followed Lipidyzer workflow with modifications.
• Data quality and filtering: After QC and filtering, an average of 778 lipids per sample and 846 lipid species overall were quantified across >1,600 runs (including 104 QC samples). QC median CVs ranged 6.5% (small TAGs) to 20.7% (DAGs). For downstream analyses, 736 lipid species were retained (QC CV <20% and biosample CV > QC CV). Intraparticipant variance was generally lower than interparticipant variance.
• Normalization/imputation: Internal standards used for absolute/estimated concentrations; additional method-3 lipids normalized by correlated spikes and regression-based abundance estimation. Cytokines batch-corrected (ComBat). Missing values imputed via class-wise KNN for truncated distributions.
• Statistical analyses: Weighted gene correlation network analysis (WGCNA) to define lipid modules from healthy samples; associations to 50 clinical labs controlled for sex, age, ethnicity, BMI. Linear mixed models for SSPG associations (controlling participants, sex, ethnicity, age, BMI). Infection models identified RVI-associated lipid changes with mixed models (same covariates). Longitudinal infection/vaccination dynamics assessed and clustered; OmicsLonDA generalized additive mixed models identified IR-vs-IS differential time-intervals during events. Ageing analysis used within-individual Δage linear models (log-lipid vs years since baseline) controlling for BMI and storage time (and sex/IR where indicated). Cytokine–lipid associations used linear mixed-effects models across 1,180 samples (random intercept for participant; covariates BMI, sex, ethnicity) with FDR control. Enrichment analyses (Fisher’s exact) assessed subclass and FA composition enrichments (SFA/MUFA/PUFA, omega status, specific FAs) within positive/negative coefficients.

• Scale and assay: Quantified 846 lipid species across 16 subclasses; retained 736 high-quality lipids. TAGs (especially large TAGs) and SMs were abundant; subclasses spanned >4 orders of magnitude. QC CVs low (6.5–20.7%).
• Personalization: Many lipids (notably TAGs, SMs, HCERs, CEs) showed high interindividual variance, with >50% variance attributable to participant for some species; FFAs showed lower participant-specific variation. t-SNE using 100 most personalized lipids largely clustered by individual across years, indicating stable personal lipid signatures.
• Lipid modules vs clinical labs (healthy): WGCNA modules M1 and M5 (enriched for CER, PE, small and large TAGs) positively associated with T2D-relevant measures (A1C, fasting glucose, fasting insulin) and inflammation (hsCRP, WBC) and negatively with HDL, indicating adverse health associations. M3 (PE-P/PE-O enriched) associated with higher HDL and lower fasting insulin (healthier signature). M7 (some FFAs, LPCs) correlated with lower CRP and A1C. Microbiome associations included negative correlations of specific TAGs with Oscillospiraceae and of (L)PE/PC/CE with Clostridiaceae.
• Insulin resistance: In 69 SSPG-tested participants, 424/736 lipids significantly associated with SSPG (BH FDR <5%) after covariate control. TAGs and DAGs, and subsets of CERs, were positively associated with SSPG (higher in IR). Enrichment revealed ether-linked PE (PE-P), in contrast to PE overall, associated with lower SSPG, suggesting antioxidant/signalling roles protective against IR-associated inflammation. IR status modified lipid–clinical lab correlations: numerous measures (e.g., LDL/HDL ratio with PE-P/PE-O positive and LPE negative in IR; opposite or different directions for A1C–SM, SSPG–CER/PI, IgM–PE-P/PE-O, monocyte–PE-P/PE-O, eosinophil–TAG, WBC–PI) differed between IR and IS.
• Respiratory viral infection (RVI) and vaccination: Across 72 RVI episodes (390 samples), 210 lipids significantly changed (FDR <10%) across multiple subclasses. Enrichments included ether-linked PEs and TAGs containing saturated FAs (e.g., FA(12:0), FA(16:0)). Four trajectory clusters revealed coordinated dynamics: small TAGs dropped sharply early with rapid recovery (blue cluster) correlating with total lipids (cholesterol, LDL); LPC/large TAG/ether-PE cluster decreased early with delayed recovery and inverse correlations with CRP/neutrophils; PCs (green) increased; FFA cluster showed slow decline reaching nadir in recovery, linked to immune parameters. IR vs IS showed distinct dynamics: early FFAs higher and mid-late PCs more elevated in IR; TAGs and some PEs higher in IS mid-to-late. Post-vaccination patterns differed from infection; e.g., fewer TAG elevations in IS and LCER upregulation in IR.
• Ageing: Within-individual Δage models showed increases over 5 years in several subclasses: CERs (LCER, HCER, DCER), SMs, LPCs, CEs; TAG increases attenuated after BMI control. Species-level enrichment indicated ageing-associated shift toward higher SFAs and MUFAs and reduced PUFAs, including declines in omega-3 FAs (DHA FA(22:6) in TAG; EPA FA(20:5) in PE) and linoleic acid (FA(18:2)); arachidonic acid (FA(20:4)) tended to increase under looser filters, consistent with proinflammatory trends. Large vs small TAGs showed distinct ageing patterns. Strong sex dimorphism in ageing for multiple subclasses (e.g., small TAGs higher with age in men, lower in women). IR accelerated ageing signatures: larger ageing coefficients for HCER, LCER, SM, CE in IR vs IS; DAGs negatively associated with ageing in IR (controlling BMI); PI and PE exhibited opposite ageing effects in IR vs IS.
• Cytokine–lipid networks: Across 1,180 samples, 1,245 significant associations linked 580 lipids to 40 cytokines/growth factors. Leptin and GM-CSF showed many positive associations with TAGs (and also with DAG, PC, PE). IL-6 and IL-10 showed negative associations with subsets of TAGs, indicating functional heterogeneity of TAG species. LPCs showed mixed associations: positive with growth factors (EGF, VEGF, BDNF) and inflammatory mediators (sCD40L, IL-1α, resistin), and negative with CRP/SSPG in other contexts, reflecting pleiotropic roles. Enrichment analyses indicated SFA-rich and small TAGs enriched among positive leptin–TAG associations; large TAGs negatively associated with IL-6/IL-10; TAGs containing FA(22:5) showed hubs of negative associations with several cytokines. PCs containing linoleic acid were negatively associated with IFNγ-inducible chemokines CXCL9/CXCL10, suggesting potential modulation of immune cell recruitment.

The longitudinal, multi-omic analysis reveals that the plasma lipidome is highly individualized yet dynamically responsive to health-to-disease transitions. Distinct lipid subclasses, particularly ether-linked PEs and small versus large TAGs, display unique behaviour across conditions. Associations of TAG-, CER-, and DAG-rich modules with glycaemic and inflammatory clinical measures underscore mechanistic links to metabolic dysfunction and IR. Ether-linked PEs associate with healthier clinical profiles and lower SSPG and are depleted during early infection, supporting roles in antioxidation and inflammation resolution. During RVI, rapid small TAG depletion with recovery and coordinated shifts in LPCs and ether-PEs align with energy mobilization and inflammatory phase transitions. Ageing is characterized by a proinflammatory lipid shift (increased CER/SM/CE/LPC; higher SFA/MUFA, reduced PUFA and omega-3), with sex-specific patterns and apparent acceleration in IR, suggesting that IR perturbs and potentially accelerates lipid ageing trajectories. Extensive cytokine–lipid networks delineate immunoregulatory roles for lipid subclasses and FA configurations, differentiating pro- from anti-inflammatory associations within TAGs and highlighting pleiotropy in LPC biology. Collectively, these findings explain how lipid dynamics map onto clinical and immunological phenotypes, advancing understanding of lipid roles in immune homeostasis, metabolic health, and ageing, and indicating that conventional lipid panels miss clinically relevant subclass/species-level information.

This study delivers a deep, longitudinal resource of the human plasma lipidome across health, IR, acute infection, ageing and cytokine networks. Key contributions include: (1) evidence for strong personalization of lipid profiles; (2) identification of lipid modules linked to clinical measures; (3) comprehensive IR-associated lipid alterations (elevated TAGs/DAGs/CERs; ether-PEs linked to healthier status); (4) mapping of acute RVI lipid trajectories with subclass-specific dynamics and IR–IS differences; (5) characterization of ageing-associated lipid shifts toward proinflammatory compositions, with sex dimorphism and acceleration in IR; and (6) extensive cytokine–lipid associations revealing immunoregulatory specificity by subclass and FA composition. These insights suggest small vs large TAGs and ether-linked PEs as promising biomarkers and potential intervention targets. Future research should validate findings in broader, diverse cohorts; integrate mechanistic studies to test causality (e.g., dietary supplementation with ether-PEs or omega-3s, modulation of lipid-converting enzymes); develop predictive models; expand lipid coverage (e.g., PC-O, PC-P) with improved standards; and assess sex- and age-tailored interventions.

• Cohort composition and generalizability: Bias toward middle-aged, highly educated participants from northern California may limit generalizability; some analyses involved small subgroups (e.g., outlier cases).
• Analytical coverage and identification: The targeted pipeline (>1,100 species) sometimes lacks exact structural resolution (e.g., double-bond positions). Some classes (e.g., PI) lacked class-specific internal standards, limiting absolute quantification accuracy; several plasma lipids (PC-O, PC-P) were not monitored.
• Extraction/material effects: Biphasic extraction using non-glass materials elevated baseline FFAs, potentially compressing ratios and reducing sensitivity to subtle FFA differences.
• Modelling approach: Primarily descriptive associations without predictive validation; correlations do not imply causation. Multiple covariates (e.g., BMI) were controlled, which may obscure intertwined biology (e.g., BMI–inflammation–ageing). Ageing analyses excluded certain long-interval samples and a participant with many samples to reduce bias.
• Data processing: Some classes (PA, PS/LPS, PG/LPG) were excluded due to missingness or DMS separation limits; one PI species removed due to mass mismatch. Imputation and batch correction, while standard, can introduce assumptions.