Abstract

BACKGROUND

   Obesity, a growing health concern, often leads to metabolic
   disturbances, systemic inflammation, and vascular dysfunction. Emerging
   evidence suggests that adipose tissue-derived extracellular vesicles
   (adiposomes) may propagate obesity-related complications. However,
   their lipid composition and effect on cardiometabolic state remain
   unclear.

METHODS

   This study examined the lipid composition of adiposomes in 122
   participants (75 in obesity group, 47 in lean group) and its connection
   to cardiometabolic risk. Adiposomes were isolated via
   ultracentrifugation and characterized using nanoparticle tracking and
   comprehensive lipidomic analysis by mass spectrometry. Cardiometabolic
   assessments included anthropometry, body composition, glucose-insulin
   homeostasis, lipid profiles, inflammatory markers, and vascular
   function.

RESULTS

   Compared with lean controls, individuals with obesity exhibited
   elevated adiposome release and shifts in lipid composition, including
   higher ceramides, free fatty acids, and acylcarnitines, along with
   reduced levels of phospholipids and sphingomyelins. These alterations
   strongly correlated with increased BMI, insulin resistance, systemic
   inflammation, and impaired vascular function. Pathway enrichment
   analyses highlight dysregulation in glycerophospholipid and
   sphingolipid metabolism, bile secretion, proinflammatory pathways, and
   vascular contractility. Machine-learning models utilizing adiposome
   lipid data accurately classified obesity and predicted cardiometabolic
   conditions, such as diabetes, hypertension, dyslipidemia, and liver
   steatosis, achieving accuracy above 85%.

CONCLUSION

   Obesity profoundly remodels the adiposome lipid landscape, linking
   lipid changes to inflammation, metabolic dysfunction, and vascular
   impairment. These findings underscore adiposome lipids as biomarkers
   for obesity and related cardiometabolic disorders, supporting
   personalized interventions and offering therapeutic value in risk
   stratification and treatment.

FUNDING

   This project was supported by NIH grants R01HL161386, R00HL140049,
   P30DK020595 (PI: AMM), R01DK104927, and P30DK020595 as well as by a VA
   Merit Award (1I01BX003382, PI: BTL).

   Keywords: Inflammation, Metabolism, Vascular biology

   Keywords: Adipose tissue, Lipidomics, Obesity
     __________________________________________________________________

   [46]graphic file with name jciinsight-10-191872-g045.jpg
     __________________________________________________________________

   This study reveals obesity-related changes in adiposome lipids that
   promote inflammation, metabolic dysfunction, and vascular issues,
   highlighting new diagnostic and therapeutic possibilities.

Introduction

   Obesity has become a critical health concern worldwide due to its rapid
   and sustained rise. According to the most recent data from the CDC,
   obesity has reached critical levels in the United States ([47]1). This
   escalating trend correlates with an increased risk of developing a
   range of chronic disorders, including cardiovascular diseases and
   metabolic syndrome ([48]2). However, the precise molecular
   underpinnings by which obesity catalyzes or exacerbates these diseases
   have not been completely delineated. One well-documented hallmark of
   obesity is the expansion of adipose tissue, a highly active endocrine
   organ secreting numerous hormones that regulate critical processes such
   as metabolism, insulin sensitivity, appetite control, and vascular
   function ([49]3). In obesity, however, dysfunctional adipose tissue and
   dysregulated secretome contribute to local and systemic inflammation,
   insulin resistance, and other metabolic derangements ([50]4–[51]6).
   Recent findings from in vitro and preclinical studies emphasize the
   significance of extracellular vesicles (EVs) derived from adipose
   tissues, currently known as adiposomes, in various metabolic and
   cardiovascular disturbances ([52]7, [53]8). However, corresponding
   findings in human adiposomes and their role in metabolic and vascular
   dysfunction are lacking.

   Over the past decade, EVs have emerged as significant players in
   metabolic regulation and interorgan communication. Once considered mere
   byproducts of cellular processes, EVs — encompassing exosomes (~30–150
   nm in diameter) and microvesicles (~100–1000 nm) — are now appreciated
   as sophisticated carriers of lipids, proteins, and nucleic acids
   ([54]9). Numerous studies initially focused on proteins and miRNAs,
   identifying cargo that can modulate inflammatory signaling and insulin
   pathways. However, recent attention has shifted to the lipid
   composition, given the pivotal role of bioactive lipids in cell
   signaling, inflammation, and metabolism ([55]10). Among the lipid
   species commonly identified in EVs are phospholipids and sphingolipids,
   which serve as key structural components of the vesicle membrane and
   influence insulin sensitivity and inflammation ([56]11, [57]12).
   Growing evidence suggests that lipids are not merely structural
   elements of EVs but are also functionally active molecules influencing
   cell signaling and homeostasis ([58]13).

   Adipose tissue releases large quantities of adiposomes that encapsulate
   numerous bioactive components ([59]14–[60]16). Our previous studies
   demonstrated that adiposomes isolated from individuals with obesity
   instigate endothelial dysfunction by impairing caveolae integrity and
   disrupting nitric oxide (NO) signaling in endothelial cells ([61]17,
   [62]18). In a recent pilot study, we observed higher levels of
   ceramides (Cer) and lower phospholipids levels in adiposomes isolated
   from participants with obesity compared with lean controls ([63]19).
   Building on this data, we conducted a comprehensive investigation into
   the relationship between adiposome lipid profiles and obesity. This
   study analyzed adiposome lipidomics in a cohort of 122 individuals with
   obesity and those who are lean, integrating clinical and physiological
   data and applying machine learning models to elucidate their
   associations with obesity and related complications, as outlined in
   [64]Figure 1. Understanding these molecular intricacies holds
   substantial implications for developing novel therapeutic targets,
   enhancing diagnostic biomarker discovery, and devising strategies for
   mitigating obesity-related cardiometabolic complications.

Figure 1. A schematic illustration of the study design comparing individuals
from the lean (n = 47) and obese (n = 75) groups.

   [65]Figure 1
   [66]Open in a new tab

Results

Clinical characteristics of the study participants.

   All clinical characteristics of the participants with obesity (n = 75)
   and those who are lean (n = 47) were summarized in [67]Table 1. The
   average age for both groups was not statistically different. However,
   the average BMI was approximately 70% higher in the participants with
   obesity than the lean controls. Similarly, total fat and visceral fat
   percentages measured using dual-energy x-ray absorptiometry (DEXA)
   scanning were 85% and 93% higher, respectively, in participants with
   obesity than in lean controls. Waist circumference (WC) was
   significantly higher in the obese group (~38%, P < 0.001). Metabolic
   parameters, including fasting plasma glucose, insulin levels,
   hemoglobin A1c (HbA1c), and the insulin resistance index, homeostatic
   model assessment for insulin resistance (HOMA-IR), were significantly
   higher in individuals with obesity. Lipid profile measurements,
   including triglycerides, total cholesterol level, and LDL, were
   markedly elevated in those with obesity compared with lean controls,
   while the HDL levels were significantly lower in the obese group. Liver
   function was impaired in the obese group, as evidenced by significant
   reductions in plasma bilirubin (22%) and total albumin (12%) and a 15%
   increase in alkaline phosphatase. Liver enzymes, including alanine
   transaminase (ALT) and aspartate transaminase (AST) tended to be higher
   in the obese group than in the lean controls, though these differences
   were not statistically significant. Systemic inflammation biomarkers,
   such as interleukin 6 (IL-6) and C-reactive protein (CRP), were
   markedly higher in the obese group. Additional differences included
   significantly higher leptin (105% higher) and lower adiponectin (84%
   lower) in individuals with obesity than in lean controls.

Table 1. Clinical characteristics of participants in the lean and obese
groups.

   [68]graphic file with name jciinsight-10-191872-g038.jpg
   [69]Open in a new tab

   Next, we used brachial flow-mediated dilation (FMD), a noninvasive
   ultrasound technique, to evaluate endothelial function, where higher
   values suggest better vascular health. Individuals with obesity
   exhibited a 60% lower brachial FMD than lean controls (P < 0.001).
   Similarly, arteriolar flow-induced dilation (FID), measured ex vivo in
   adipose tissue-isolated arterioles, was reduced by 45% in participants
   with obesity compared with lean controls (P < 0.001), suggesting
   impaired peripheral arteriolar vasodilation. Additionally, individuals
   with obesity show a 48% reduction in NO bioavailability compared with
   lean controls, indicating impaired endothelial function.

   Of the 75 participants with obesity, 34 were hypertensive (~45%), 42
   were insulin resistant or had diabetes (56%), 38 were dyslipidemic with
   disrupted lipid profile (~51%), and 46 had steatotic liver (~61%).
   Diagnosis of these conditions relied on measurements conducted during
   study visits and a review of patient medical records. Finally, based on
   medication information collected from participants, 29 were taking
   antihypertensive medications (including angiotensin-converting enzyme
   inhibitors, angiotensin II receptor blockers, β blockers, calcium
   channel blockers, and diuretics), 28 were on diabetic medications
   (including metformin, dipeptidyl peptidase-4 inhibitors, glucagon-like
   peptide-1 agonists, sodium-glucose cotransporter 2 inhibitors,
   sulfonylureas, or insulin), and 14 participants were on statin therapy.

Obesity-driven alterations in adiposome lipids.

   To study obesity-driven alterations in adiposome lipids and their role
   in cardiometabolic dysfunction, human adiposomes were extracted from
   the visceral adipose tissue (VAT) and characterized by transmission
   electron microscopy (TEM) and nanoparticle-tracking analysis (NTA)
   ([70]Figure 2, A–C). These analyses revealed adiposome diameters
   ranging from 50 to 350 nm. The average concentration of adiposomes was
   significantly higher in samples from patients with obesity compared
   with those from lean controls (9.0 × 10¹¹ particles/mL versus 4.5 ×
   10¹¹ particles/mL; P < 0.001) ([71]Figure 2D). Protein analysis of the
   adiposomes confirmed the presence of conventional EVs biomarkers,
   including tetraspanins CD9, CD81, and CD63, while showing no evidence
   of lipoprotein contamination (Apolipoprotein B; APOB) ([72]Figure 2E).
   Additionally, the adiposomes expressed adipocyte-specific proteins such
   as PPARγ, adiponectin, and FA-binding protein 4 (FABP4) ([73]Figure
   2F), further confirming their adipocytic origin.

Figure 2. Analysis of adiposomes and lipid alterations in individuals in the
obese and lean groups.

   [74]Figure 2
   [75]Open in a new tab

   (A) A representative image of the AT biopsy sample. (B) TEM image of
   adiposome particles. Scale bar: 200 nm. (C) Nanoparticle tracking
   analysis displaying adiposome particles in the samples. Original
   magnification ×20. (D) Quantification of adiposome concentration in
   obese and lean groups (Mann-Whitney U test). (E and F) Western blot
   analysis for EV biomarkers and adipocytic proteins. (G) Heatmap of
   adiposome lipid species across lean and obese groups. (H) Percentage
   composition of major lipid classes in the control and obese groups.

   Comprehensive lipidomics analysis identified 595 lipid species spanning
   19 major lipid classes ([76]Figure 2G). Among these, 266 lipids
   exhibited significant alterations in the adiposomes of participants
   with obesity compared with lean controls, with 54 upregulated and 212
   downregulated (FDR < 0.05) after adjusting for age, sex, and
   race/ethnicity. Comparing the percentage composition of lipid classes
   ([77]Figure 2H), the most considerably elevated lipids in the obese
   group belonged to the classes of FA, acylcarnitines (Acar), Cer,
   hexosyl ceramides (HexCer), cholesterol esters (CE), and
   diacylglycerols (DG). In contrast, lipids considerably enriched in the
   control group included phospholipids (phosphatidylcholine [PC]),
   phosphatidylethanolamine (PE), lysophosphatidylcholine (LPC),
   phosphatidic acid (PA), triglycerides (TG), SM, and fatty acyl esters
   of hydroxy FA (FAHFA). Principal Component Analysis (PCA) revealed a
   clear separation of the samples into 2 distinct groups in the PCA plot
   ([78]Figure 3A). The volcano plot analysis was then used to visualize
   the differentially abundant lipids in the isolated adiposomes from
   participants who are lean and those with obesity ([79]Figure 3B).
   Notably, Cer-NS d18:1/23:0 exhibited a significant increase, while
   FAHFA 18:0/20:2 demonstrated a significant decrease in the obese group
   compared with the control group.

Figure 3. Differential adiposome lipids in individuals in the obese and lean
groups.

   [80]Figure 3
   [81]Open in a new tab

   (A) Principal Component Analysis (PCA) plot of adiposome lipids in
   obese and lean groups. (B) Volcano plot illustrates the differential
   lipid expression (log[2] fold change) between obese and lean groups
   (–log[10] P value; unpaired 2-tailed t test with multiple testing
   correction; Benjamini-Hochberg FDR).

   To assess whether specific lipid species were associated with BMI in
   this population, we conducted a linear regression analysis ([82]Figure
   4A). Nineteen lipid species, primarily in the Cer, FA, CE, and Acar
   lipid classes, demonstrated positive β values, indicating a positive
   correlation with BMI. Conversely, negative β values were found for 23
   lipid species, predominantly from phospholipid, sphingomyelin (SM),
   FAHFA, and TG lipid classes, suggesting an inverse relationship with
   the BMI. To evaluate the model’s goodness of fit, we used 3 key
   metrics: adjusted R-squared, root mean squared error (RMSE), and the P
   value of the regression model. The adjusted R-squared value was 0.85,
   indicating that the adiposome lipidomics data explained 85% of the
   variability in BMI.

Figure 4. Regression analysis and association of adiposome lipids with
obesity status.

   [83]Figure 4
   [84]Open in a new tab

   (A) Regression plot showing the association between lipid species and
   obesity status, arranged by their β values (linear regression) and
   their associated P values. (B) Horizontal lollipop chart shows the
   differential lipid abundance from class 2 and 3 individuals with
   obesity using log[2] fold change to represent the magnitude of change.
   For statistical significance, an unpaired 2-tailed t test with multiple
   testing correction (Benjamini-Hochberg FDR) was used.

   The RMSE value of 4.9 represents the average error magnitude between
   the observed BMI values and those predicted by the regression model.
   The P value for the overall regression model was found to be 5.47 ×
   10^–6, indicating that the independent variables (lipid species)
   included in the model are valid predictors of BMI. In summary, the
   linear regression analysis identified 42 specific lipid species
   associated with changes in BMI, providing valuable insights into the
   relationship between adiposome lipid composition and BMI.

   Furthermore, our results also identified a group of 34 lipids,
   predominantly of the phospholipid and SM lipid categories, that were
   significantly less abundant in adiposomes from individuals with class 3
   obesity (BMI ≥ 40 kg/m^2) compared with those from class 2 (BMI,
   35–39.9 kg/m^2). This difference remained significant after adjusting
   for potential confounding variables such as age, sex, and
   race/ethnicity ([85]Figure 4B; FDR < 0.05). This differential human
   lipid profile may have important implications for understanding the
   functional differences in adipose tissue biology and the associated
   metabolic dysregulation in higher BMI categories.

Unique adiposome lipid signatures differentiate individuals with obesity and
those who are lean.

   In this study, adiposome lipid profiling revealed marked differences in
   various lipid species between individuals with obesity and lean healthy
   controls. Notably, LPCs and PCs, key phospholipid components, were
   significantly lower in adiposomes from individuals with obesity,
   indicating an overall reduction in phospholipid species. For instance,
   LPC 20:1, LPC 20:2, PC 36:4, and PC 18:2/20:4 all displayed
   significantly lower levels (P < 0.01). Ether-linked phospholipids
   (EtherPE 18:0e/20:4 and EtherPC 16:0e/20:4) were enriched in lean
   participants, further underscoring disrupted phospholipid metabolism in
   obesity ([86]Figure 5, A–E). Ceramides, SM, TG, ACar, FA, and FAHFAs
   also showed differential abundance. Ceramides exhibited notable shifts:
   certain species (e.g., Cer NDS d22:0/18:1) were significantly higher in
   individuals with obesity, while lean participants showed higher levels
   of others (e.g., Cer AP t42:0) ([87]Figure 6, A–E). Interestingly,
   elevated Cer in lean individuals often contained more carbons and
   unsaturated bonds. SM likewise differed, with SM d14:3/28:2 and SM
   d42:1 enriched in lean participants, whereas SM d33:1 and SM d39:3 were
   higher in the obese group ([88]Figure 6, F–H). Overall, total Cer
   levels were significantly higher in obesity, while total SM levels were
   lower, suggesting enhanced SM hydrolysis of Cer, a process known to
   drive inflammation and insulin resistance ([89]20).

Figure 5. Differential phospholipid profiles in adiposomes of individuals in
the obese and lean groups.

   [90]Figure 5
   [91]Open in a new tab

   (A) Bar plot of fold changes in the top 69 phospholipid and
   lysophospholipid species between obese and control (lean) groups and
   their significance (–log[10] q values). (B–E) Box plots showing
   examples of the significantly altered phospholipid species between the
   obese and lean groups. For statistical significance, an unpaired
   2-tailed t test with multiple testing correction (Benjamini-Hochberg
   FDR) was used. MPI, mean peak intensity.

Figure 6. Ceramide and sphingomyelin alterations in adiposomes of individuals
in the obese and lean groups.

   [92]Figure 6
   [93]Open in a new tab

   (A) Bar plot of fold changes in Cer and HexCer between obese and
   control groups and their significance (–log[10] q values). (B–E) Box
   plots of key Cer showing significant differences between groups. (F)
   Bar plot of fold changes in the expression levels of SM in obese and
   lean groups. (G and H) Box plots of selected SM species show
   significant differences between the control and obese groups. For
   statistical significance, an unpaired 2-tailed t test with multiple
   testing correction (Benjamini-Hochberg FDR) was used.

   TGs were generally more abundant in lean individuals. Specific TGs
   (e.g., TG 15:0/18:2/18:2 and TG 52:0) were depleted in participants
   with obesity, suggesting possible impairments in TG synthesis or
   elevated lipolysis ([94]Figure 7, A–C). Concurrently, higher levels of
   free FA in the obese group indicated reduced TG storage and increased
   FA intermediates. Most measured FAs, including FA 16:0, FA 17:1, and FA
   18:0, were more abundant in obesity ([95]Figure 7, D and E). FAHFAs,
   which possess antiinflammatory and insulin-sensitizing properties
   ([96]21), were lower in individuals with obesity (e.g., FAHFA 18:1/18:0
   and FAHFA 16:0/22:3) ([97]Figure 7, F and G). This reduction may
   reflect diminished protective lipid species in obesity. Conversely,
   ACar, which are critical for transporting FAs into mitochondria for
   β-oxidation, were significantly elevated in participants with obesity
   (e.g., ACar 16:0 and ACar 18:1), suggesting perturbations in lipid
   metabolism, mitochondrial function, and inflammation ([98]Figure 7, H
   and I).

Figure 7. Differential profiles of triglycerides, fatty acids, FAHFAs, and
acylcarnitines.

   [99]Figure 7
   [100]Open in a new tab

   (A, D, F, and H) Bar plot of fold changes in TG species (A), fatty acid
   species (D), FAHFAs (F), and ACar species (H) between obese and control
   groups and their significance (–log[10] q values [FDR]). (B, C, E, G,
   and I) Box plots illustrate altered TG species (B and C), FA 17:1 (E),
   FAHFA 18:1/18:0 (G), and ACar 16:0 (I) in the lean and obese groups.
   For statistical significance, an unpaired 2-tailed t test with multiple
   testing correction (Benjamini-Hochberg FDR) was used.

   Overall, these findings reveal significant obesity-related alterations
   in the adiposome lipid landscape, including increased levels of Cer,
   free FAs, and ACar, coupled with lower phospholipids, SM, certain TGs,
   and FAHFAs. These lipidomic alterations indicate a substantial
   metabolic remodeling of adiposome function, likely contributing to
   inflammation, insulin resistance, and other obesity-related
   dysfunctions. Understanding these lipidomic changes offers insights
   into obesity pathophysiology and points to potential therapeutic
   targets to restore healthy lipid balance.

Sex and racial/ethnic effects on adiposome lipidomics in individuals with
obesity.

   To study the effect of sex on adiposome lipid profiles, the current
   study included 45 female participants with obesity and 30 male
   participants with obesity. Among clinical measures, the total fat
   percentage was higher in females (53.1 ± 6.4) than in males (46.8 ±
   6.9; P = 0.02), while hemoglobin was lower in females (12.3 ± 1.2)
   compared with males (13.5 ± 0.8; P = 0.0008). Adiposome lipidomics;
   certain lipid species, including SM d41:1, PI 18:1/18:2; and PC 19:0
   lipid species were higher in females (log[2] fold change [FC] = 1.4,
   1.3, 1.1), though these differences showed only borderline significance
   after multiple-testing correction. Participants represented 3
   racial/ethnic groups: African American (AA, n = 42), Hispanic (H, n =
   20), and Non-Hispanic White (NHW, n = 13). No significant differences
   were observed in clinical or physiological measures across these
   groups. However, after correction for multiple testing, 18 adiposome
   lipid species showed significant differences, primarily between AA and
   the other 2 groups. Phospholipids, including PC, PE, and ether PC, were
   lower in AA than H and NHW, while PA were consistently higher in AA and
   H than in NHW. Two sphingomyelin (SM) species were lower in AA versus
   NHW; SM d37:1 also differed between AA and H, and SM d39:1 differed
   between H and NHW. Two Cer species (Cer-NDS d18:0/16:0 and Cer-NS
   d18:1/26:0) were significantly higher in AA and H than in NHW. Three
   FAs (FA 16:1, FA 18:1, and FA 18:0) were also higher in AA. However,
   total Cer and SM levels did not differ significantly among the 3 groups
   ([101]Table 2).

Table 2. Differential lipidomics in adiposomes from African American,
Hispanic, and White participants.

   [102]graphic file with name jciinsight-10-191872-g039.jpg
   [103]Open in a new tab

Adiposome lipidomic differences across individuals with diabetes,
hypertension, and dyslipidemia.

   Our data reveal differential abundances of specific lipid species in
   diabetic, hypertensive, and dyslipidemic participants with obesity
   compared with metabolically healthy counterparts with obesity.
   Individuals with diabetes showed lower levels of FAHFA 18:0/22:0 (P =
   0.00025) and several phospholipid and TG species, whereas ACar (ACar
   18:1 and 16:1) were significantly higher in participants with diabetes
   than in those without it ([104]Figure 8, A–C). This shift toward
   increased ACar levels may reflect altered FA oxidation or mitochondrial
   dysfunction in diabetes. Adiposome lipid profiles also differed between
   obese individuals with and without hypertension. Cer-NDS d23:0/18:2 and
   other Cer were elevated in the hypertensive group, while SM d38:2 was
   significantly lower, suggesting a shift toward Cer production that
   could promote inflammation and vascular dysfunction. Linoleic acid (FA
   18:2) and specific TGs (TG 51:4, TG 15:0/18:2/18:2, TG 17:0/18:1/18:2)
   were also higher in the hypertensive group, possibly reflecting altered
   lipid metabolism or impaired lipid storage ([105]Figure 8, D–F). In
   dyslipidemic versus nondyslipidemic individuals with obesity, there was
   a substantial increase in Cer-NDS d16:0/24:0 (P = 9.8 × 10^–8) and DG
   18:2/22:6, paired with reduced levels of phospholipids, such as
   phosphatidylserine (PS 15:0/18:1) and SM (SM d40:2, SM d33:1)
   ([106]Figure 9, A–C). This pattern indicates a lipid imbalance favoring
   Cer and DG accumulation, potentially exacerbating cardiometabolic risk.

Figure 8. Lipidomic alterations in diabetic and hypertensive individuals.

   [107]Figure 8
   [108]Open in a new tab

   (A and D) Stacked bar plots show the distribution of lipid classes
   (log[2] FC and FDR values are provided in parentheses) in individuals
   with diabetes versus those without diabetes (A) and those who are
   hypertensive versus those who are nonhypertensive (D). (B, C, E, and F)
   Boxplots for lipids altered between diabetic versus nondiabetic groups
   (B and C) and hypertensive versus nonhypertensive groups (E and F). For
   statistical significance, an unpaired 2-tailed t test with multiple
   testing correction (Benjamini-Hochberg FDR) was used.

Figure 9. Lipidomic alterations in dyslipidemic individuals.

   [109]Figure 9
   [110]Open in a new tab

   (A) Stacked bar plots show the distribution of lipid classes (log[2] FC
   and FDR values in parentheses) in dyslipidemic versus nondyslipidemic
   individuals. (B and C) Box plots for lipids altered between
   dyslipidemic and nondyslipidemic groups. For statistical significance,
   an unpaired 2-tailed t test with multiple testing correction
   (Benjamini-Hochberg FDR) was used.

   Medications also contributed to these adiposome lipidomic differences.
   Among 12 patients who are hypertensive treated with calcium channel
   blockers (amlodipine, nifedipine, verapamil) alone or in combination
   with β blockers (carvedilol, metoprolol), adiposomes contained
   significantly higher TG 17:0/18:0/18:1 (log[2] FC = 3.28; P = 0.0004)
   and SM d37:1 (log[2] FC = 2.06; P = 0.007) than in other individuals
   with obesity. These patients further displayed elevated SM d34:2, TG
   56:8, and TG 16:0/18:2/22:6 relative to hypertensive patients not
   receiving these medications. Moreover, statin therapy was associated
   with lower FA 18:1 levels (log[2] FC = –2.72; P = 0.0006). In contrast,
   diabetic medications did not produce notable changes in adiposome lipid
   profiles, and the single participant receiving omega-3 supplementation
   did not provide sufficient data for analysis. These findings illustrate
   that comorbidities like diabetes, hypertension, and dyslipidemia
   correspond to distinct adiposome lipid signatures and show that certain
   medications further modify these profiles. Such insights underscore the
   need for individualized therapeutic strategies in managing
   obesity-related cardiometabolic disorders.

Adiposome lipid signature is differentially associated with cardiometabolic
risk.

   A total of 186 adiposome lipids showed significant associations (FDR <
   0.05) with cardiometabolic risk features after adjusting for age, sex,
   and race/ethnicity. These included 100 phospholipids, 28 TGs, 23 Cers,
   19 SMs, 11 FAs and FAHFAs, 3 ACar, and 2 DG. Lipids were summed up
   within each class to examine their relationship to cardiometabolic risk
   factors, and regression analyses were performed. Total Cer, ACar, and
   FAs were elevated in participants with obesity compared with lean
   controls, whereas total TGs, phospholipids (PC/LPC), and FAHFAs were
   lower in those with obesity ([111]Figure 10, A–F). Regression results
   indicated that total Cer, ACar, DG, and FA were strongly linked to BMI,
   fat percentage, glucose-lipid metabolism, vascular function, liver
   function, and systemic inflammation. Measures of adiposity (BMI, fat
   percentage, VAT mass) showed positive correlations with proinflammatory
   lipids such as ACar (r > 0.4, P < 0.001) and Cer (r > 0.5, P < 0.01)
   while correlating negatively with phospholipids (r < –0.5, P < 0.001)
   and FAHFAs (r < –0.5, P < 0.001). HOMA-IR, an indicator of insulin
   resistance, was positively associated with Cer and free FAs but
   negatively correlated with phospholipids. IL-6 and CRP similarly
   correlated positively with Cer, ACar, and FAs while showing negative
   relationships with FAHFAs, highlighting the potential involvement of
   these lipids in the inflammation process ([112]Figure 10G).

Figure 10. Total lipid classes and correlation with clinical parameters.

   [113]Figure 10
   [114]Open in a new tab

   (A–F) Box plots comparing total lipid class between obese and lean
   groups. (G) Correlation heatmap displaying the relationships between
   lipid classes and clinical parameters in individuals in the obese and
   lean groups. For statistical significance, an unpaired 2-tailed t test
   with multiple testing correction (Benjamini-Hochberg FDR) was used.

   Within the obese group, PC, Cer, and FAs were significantly associated
   with BMI, fasting insulin, leptin, HDL, and arterial vasoreactivity
   ([115]Figure 11, A–J). Leptin correlated positively with Cer, DG, and
   FAs but negatively with EtherPC, FAHFAs, and PC/LPC. Adiponectin was
   positively linked to FAHFAs and SM yet negatively correlated with Cer,
   underscoring its protective metabolic role. Regarding vascular health,
   both FMD and arteriolar FID positively correlated with FAHFAs and
   PC/LPC but were negatively related to Cer, FAs, and ACar. Importantly,
   lipids found to be less abundant in adiposomes from patients with
   obesity, such as FAHFAs, SM, and EtherPC, tended to correlate inversely
   with adiposity and inflammation markers, suggesting that their
   depletion in obesity exacerbates metabolic dysfunction
   ([116]Supplemental Tables 1 and 2; supplemental material available
   online with this article;
   [117]https://doi.org/10.1172/jci.insight.191872DS1). Overall, the data
   indicate a striking shift toward proinflammatory lipid profiles
   alongside diminished protective lipids, linking these changes to
   obesity-related inflammation, insulin resistance, and vascular
   impairments.

Figure 11. Correlation of lipidomic profiles with clinical parameters in
participants with obesity.

   [118]Figure 11
   [119]Open in a new tab

   (A–J) Scatter plots of lipid levels with cardiometabolic markers in the
   obese group. Pearson correlation test was used.

Machine learning identifies adiposome lipids as key indicators of
cardiometabolic diseases.

   Next, we leveraged these observed differential lipid profiles to
   develop predictive models (decision tree and random forest) to identify
   cardiometabolic diseases among individuals with obesity, including
   diabetes, hypertension, dyslipidemia, and hepatic steatosis, as well as
   risk factors such as systemic inflammation and vascular dysfunction.
   First, we used these predictive models to distinguish between control
   and obese groups based on adiposome lipidomic features. The decision
   tree model achieved 100% accuracy in predicting obese versus control
   subjects depending on key lipidomic features (FAHFA 18:0/20:2 and FA
   16:0; [120]Figure 12A). The confusion matrix from the random forest
   model highlights perfect classification accuracy, with 99% of obese and
   89% of control samples correctly identified ([121]Figure 12B). The most
   critical lipids contributing to the model’s accuracy include Cer NS
   d15:1/15:0, Cer NS d18:1/23:0, FAHFA 18:0/20:2, and FA 17:1
   ([122]Figure 12C). These findings demonstrate that these lipidomic
   markers are highly effective in accurately classifying the samples
   based on obesity status, as evidenced by the confusion matrix/chart and
   the receiver operating characteristic (ROC) curve ([123]Figure 12, D
   and E). Overall, the results reflect the importance of adiposome
   lipidomic profiling in differentiating between control and obese groups
   with high accuracy.

Figure 12. Predictive modeling of obesity status using lipidomic data.

   [124]Figure 12
   [125]Open in a new tab

   (A) A decision tree. Visualization highlights key lipids separating
   obese from lean individuals. (B) Random forest confusion matrices
   display predictive performance for obesity. (C) Feature importance
   plots rank the contribution of individual lipid species for these
   classifications. (D) Corresponding table and bar plot of classification
   accuracy for obesity. (E) ROC curves demonstrate model performance for
   obesity. Model performance was evaluated using accuracy, sensitivity,
   specificity, and ROC AUC.

   Adiposome lipidomic profiles effectively differentiate diabetic from
   nondiabetic and hypertensive from nonhypertensive individuals with
   obesity. In diabetes classification, a random forest model correctly
   classified 63% of diabetic and 88% of nondiabetic individuals, with
   Cer-NS d15:1/15:0, Cer-NS d18:1/23:0, and FAHFA 18:0/20:2 emerging as
   leading predictors alongside phosphocholines, CE, DG, and TG. A
   decision tree approach performed even better, correctly identifying 94%
   of diabetic and 79% of nondiabetic participants, yielding an overall
   accuracy of approximately 87% ([126]Figure 13, A–C). For hypertension,
   the random forest model achieved 53% and 73% accuracy for hypertensive
   and nonhypertensive groups, respectively, while the decision tree’s
   accuracy reached 92%. Top contributors in distinguishing hypertensive
   versus nonhypertensive individuals included Cer-NS d15:1/15:0, Cer-NS
   d18:1/23:0, and FAHFA 18:0/20:2, underscoring the importance of Cer and
   FAHFAs ([127]Figure 13, E–G). ROC curves further illustrated robust
   discriminative power for both diabetes and hypertension ([128]Figure
   13, D and H).

Figure 13. Predictive modeling of diabetes and hypertension.

   [129]Figure 13
   [130]Open in a new tab

   (A and E) Random forest confusion matrices display predictive
   performance for diabetes (A) and hypertension (E). (B and F) Feature
   importance plots rank the contribution of individual lipid species for
   these classifications. (C and G) Corresponding tables and bar plots of
   classification accuracy for diabetes (C) and hypertension (G). (D and
   H) ROC curves demonstrate model performance for diabetes (D) and
   hypertension (H). Model performance was evaluated using accuracy,
   sensitivity, specificity, and ROC AUC.

   Adiposome lipids also distinguished dyslipidemic from nondyslipidemic
   individuals with obesity: the random forest model correctly classified
   56% of dyslipidemic and 71% of nondyslipidemic individuals, whereas the
   decision tree model accurately identified 97% and 88%, respectively,
   yielding an overall accuracy of 92%. Cer NS d15:1/15:0, FA 20:0, and
   LPC 22:4 lipid species were among the key lipid predictors ([131]Figure
   14, A–D). For liver steatosis, random forest correctly classified 72%
   of steatotic and 71% of nonsteatotic individuals, while the decision
   tree model achieved ~88% and ~98% accuracy in these groups,
   respectively. Cer-NS d15:1/15:0, FAHFA 18:0/20:2, and Cer-NS d18:1/23:0
   emerged as the top contributing lipids ([132]Figure 14, E–H).

Figure 14. Predictive modeling of dyslipidemia and steatosis.

   [133]Figure 14
   [134]Open in a new tab

   (A and E) Random forest confusion matrices display predictive
   performance for dyslipidemia (A) and steatosis (E). (B and F) Feature
   importance plots rank the contribution of individual lipid species for
   these classifications. (C and G) Corresponding tables and bar plots of
   classification accuracy for dyslipidemia (C) and steatosis (G). (D and
   H) ROC curves demonstrate model performance for dyslipidemia (D) and
   steatosis (H). Model performance was evaluated using accuracy,
   sensitivity, specificity, and ROC AUC.

   Individuals with obesity were classified as having vascular dysfunction
   if their brachial FMD or arteriolar FID fell below the average values
   of lean, healthy controls (6.8% for brachial FMD and 76.1% for
   arteriolar FID). A random forest model accurately predicted 98% of
   those with impaired vascular function and 89% with normal function. Its
   top predictors included Cer-NS d15:1/15:0, FAHFA 18:0/20:2, and FA
   17:1. A decision tree model achieved 100% accuracy in impaired function
   and 79% in normal function, with an AUC of 0.98 and overall accuracy of
   92% ([135]Figure 15, A–D). Individuals with obesity were also
   classified as having high inflammation if their circulating levels of
   IL-6 and CRP were higher than the average in the lean, healthy controls
   (5.4 pg/mL for IL-6 and 0.67 mg/dL for CRP). The random forest model
   correctly predicted 73% of those with low inflammation and 89% with
   high inflammation. Its top predictors were Cer-NS d15:1/15:0, FAHFA
   18:0/20:2, and FAHFA 16:0/22:3. The decision tree model accurately
   classified 65% of individuals with obesity with low inflammation and
   98% of those with high inflammation, yielding an AUC of 0.92 and an
   overall accuracy of 86.7% ([136]Figure 15, E–H). Collectively, these
   results underscore the value of adiposome lipid profiling combined with
   machine-learning algorithms in classifying obesity-related
   comorbidities and highlight the potential of adiposome lipidomics as a
   powerful tool for identifying metabolic dysfunction and guiding
   personalized interventions in individuals with obesity.

Figure 15. Predictive modeling of vascular dysfunction and inflammation.

   [137]Figure 15
   [138]Open in a new tab

   (A and E) Random forest confusion matrices display predictive
   performance for vascular dysfunction (A) and inflammation (E). (B and
   F) Feature importance plots rank the contribution of individual lipid
   species for these classifications. (C and G) Corresponding tables and
   bar plots of classification accuracy for vascular dysfunction (C) and
   inflammation (G). (D and H) ROC curves demonstrate model performance
   for vascular dysfunction (D) and inflammation (H). Model performance
   was evaluated using accuracy, sensitivity, specificity, and ROC AUC.

Pathway enrichment analysis.

   A lipid pathway enrichment analysis (LIPEA) was performed to link each
   identified lipid to relevant metabolic or signaling pathways
   ([139]Figure 16A). Among the top enriched pathways were
   glycerophospholipid and sphingolipid metabolism, arachidonic acid and
   cholesterol metabolism, inositol phosphate, bile secretion, steroid
   biosynthesis, and advanced glycation end (AGE) products/receptor of AGE
   (RAGE) signaling. Inflammatory and vascular pathways, including
   vascular smooth muscle cell contraction and TRP channel regulation,
   were also identified. Additional enriched pathways involved
   ferroptosis, autophagy, and key signaling cascades (e.g., PI3K/AKT,
   TNF, AMPK, and mTOR), underscoring adiposome lipids’ roles in immune,
   vascular, and metabolic regulation. KEGG mapping highlighted
   significant lipid perturbations in cholesterol metabolism,
   prostaglandin, and leukotriene pathways, eicosanoid synthesis, and
   lipid biosynthesis ([140]Figure 16B). Overall, these findings indicate
   a distinct clustering of lipid alterations concentrated in
   inflammation, energy metabolism, and membrane homeostasis pathways.

Figure 16. Pathway enrichment analysis of differential adiposome lipid
species.

   [141]Figure 16
   [142]Open in a new tab

   (A) A Sankey plot visualizes the connection between lipids (labeled
   with KEGG codes) and enriched KEGG pathways. Each molecule is linked to
   its associated pathways, with the statistical significance shown in the
   bubble plot. (B) Bar chart of KEGG enrichment analysis showing the top
   biological processes related to altered lipids.

Discussion

   Adiposomes are increasingly recognized as key mediators of interorgan
   communication, and we have previously reported their role in signaling
   pathways between adipocytes and vascular cells ([143]17, [144]18,
   [145]22, [146]23). While the protein and nucleic acid content of EVs,
   including adiposomes, has been extensively studied, their lipid
   composition has received less attention, particularly in the context of
   cardiometabolic health. To address this gap, we performed comprehensive
   lipidomics profiling of adiposomes isolated from the VAT of
   participants with obesity and those who are lean, alongside a detailed
   assessment of cardiometabolic risk factors. First, we observed that the
   concentration of adiposomes was nearly twice as high in samples from
   patients with obesity compared with those from lean controls,
   corroborating findings from prior studies by our group and others
   ([147]17, [148]19, [149]24). Such elevated EV release may modify
   paracrine and endocrine signaling, potentially contributing to
   metabolic dysregulation. This study suggests that the adiposome
   lipidomic landscape could be a biomarker or mediator of obesity-related
   inflammation and insulin resistance.

   Our comprehensive lipidomics analysis identified 266 lipid species that
   differed significantly between obese and lean adiposomes. In
   particular, the PC, LPC, and other phospholipids (PE, ether-linked PE
   [EtherPE], EtherPC) were significantly lower in individuals with
   obesity. These findings align with previous studies that reported
   similar reductions in PC/LPC species in the plasma of individuals with
   obesity with or without diabetes compared with lean controls
   ([150]25–[151]27). Phospholipids are crucial for membrane integrity,
   cellular signaling, and lipid transport, and their depletion may
   compromise adiposome stability or alter their biological interactions
   ([152]28). Phospholipid insufficiency has also been linked with
   impaired insulin signaling and a higher risk of cardiovascular disease,
   possibly due to compromised membrane fluidity and diminished capacity
   for NO production ([153]29–[154]34). Indeed, our investigation revealed
   a significant inverse relationship between total and individual
   phospholipid levels and BMI, VAT mass, fasting plasma insulin, HOMA-IR,
   and circulating leptin concentrations. Conversely, a positive
   correlation was identified between adiposome phospholipid levels and
   circulating NO, brachial FMD, and arteriolar FID.

   Elevated Cer levels, particularly those with moderate chain lengths and
   saturated FA, were a hallmark of obese adiposomes, correlating with
   adiposity, insulin resistance, systemic inflammation, and vascular
   dysfunction. Obesity-driven sphingolipid metabolism shifts toward Cer
   accumulation, often depleting SM ([155]35). Indeed, we observed lower
   SM levels, suggesting upregulated sphingomyelinase (SMase) activity
   converting SM into Cer. Ceramides are linked to insulin resistance,
   inflammation, and endothelial dysfunction by disrupting insulin
   receptor signaling and promoting mitochondrial reactive oxygen species
   ([156]36). They also trigger proinflammatory cascades, worsening
   metabolic and vascular dysfunction ([157]37). Our findings align with
   studies showing elevated plasma Cer in obesity, insulin resistance, and
   diabetes ([158]38). However, this study identifies ceramide enrichment
   specifically within adiposomes, highlighting their potential role in
   mediating obesity-related metabolic and vascular complications. Several
   studies have previously reported that acid SMase (aSMase) and neutral
   SMase (nSMase) are elevated in adipose tissue from obese versus lean
   groups, resulting in increased ceramide generation, inflammation, and
   insulin resistance. These enzymatic alterations have been linked to
   impaired lipid metabolism and cardiometabolic diseases
   ([159]39–[160]41). These findings highlight the importance of SMase
   activity, suggesting that future studies measuring aSMase and nSMase
   activities could provide valuable insights into their role in shaping
   the lipid profiles identified in this study.

   Notably, Cer in lean individuals were predominantly longer (≥ 38
   carbons) and/or unsaturated, such as Cer 42:0, Cer 43:2, and Cer 26:2,
   while those in individuals with obesity were shorter (≤ 26 carbons)
   and/or saturated, such as Cer 18:0/22:0 and Cer 15:1/15:0. Prior
   studies link plasma Cer with C14–C26 acyl chains to obesity, metabolic
   disorders, and vascular dysfunction, whereas ultra–long chain Cer (>26
   carbons) are essential for cellular integrity and skin barrier function
   ([161]42, [162]43). C16 and C18 ceramides are particularly associated
   with obesity, type 2 diabetes, and liver steatosis, with studies
   showing that disrupting C18 Cer synthesis improves insulin sensitivity
   and glucose tolerance in high-fat diet models ([163]44, [164]45). The
   mechanism behind the abundance of short unsaturated Cer is not known.
   However, some plausible explanations for this phenomenon could be
   related to the increased activity of the enzymes involved in the
   initial steps of Cer synthesis, such as serine palmitoyltransferase,
   favoring the incorporation of saturated FAs (like palmitate) into Cer
   backbones. Additionally, metabolic overload may impair FA elongation
   and desaturation, leading to an imbalance favoring short-chain
   saturated Cer ([165]46). Further studies are needed to elucidate these
   mechanisms.

   Our previous studies show that adiposomes from individuals with obesity
   induce structural and functional changes in endothelial cells,
   including caveolae disruption, impaired endothelial NO synthase (eNOS)
   signaling, and compromised membrane integrity ([166]17, [167]18,
   [168]22, [169]23). Building on this, our work links dysregulated
   adiposome Cer to vascular dysfunction, marked by reduced brachial
   artery dilation, impaired arteriolar vasoreactivity, elevated systolic
   blood pressure, and lower NO bioavailability. Additionally, adiposome
   Cer levels strongly correlated with circulating inflammatory markers
   CRP and IL-6, highlighting their role in systemic inflammation and
   vascular impairment.

   Adiposomes from the obese group also displayed higher free FAs and
   ACar. Elevated ACar can reflect incomplete mitochondrial FA oxidation,
   contributing to lipotoxic intermediates that worsen insulin sensitivity
   and prompt inflammatory pathways ([170]47). Conversely, ACar
   accumulation could be interpreted as an indicator of increased FA flux.
   Future functional studies are required to directly investigate the role
   of ACar-rich adiposomes in modulating mitochondrial oxidation capacity,
   providing a more mechanistic understanding of their effect on metabolic
   regulation ([171]48, [172]49). Similarly, excess-free FAs from
   dysfunctional adipocytes exacerbate ectopic fat deposition, insulin
   resistance, and pancreatic β cell dysfunction ([173]50). In this study,
   excess FAs and ACar correlated positively with BMI, higher fat mass,
   insulin resistance, systemic inflammation, and vascular dysfunction,
   reinforcing their potential role as markers or mediators of
   adiposity-driven metabolic stress. Conversely, lipids known for their
   insulin-sensitizing and antiinflammatory properties, such as FAHFAs,
   were significantly lower in individuals with obesity. Overall, these
   lipidomic changes converge into a state that promotes metabolic stress
   and vasculopathy, underlining the possibility that adiposomes are
   conduits through which pathological lipids traffic between local
   adipose depots and systemic circulation.

   Examining the link between adiposome lipidomic alterations and
   metabolic disease states, individuals with obesity who have diabetes or
   insulin resistance exhibited higher ACar and lower phospholipids and
   FAHFAs compared with those who have obesity but are metabolically
   healthy. This pattern aligns with studies associating impaired
   mitochondrial FA oxidation and FAHFA depletion with poorer glycemic
   control ([174]51, [175]52). Similarly, hypertensive individuals with
   obesity exhibited elevated Cer and reduced SM, suggesting an
   upregulated SMase pathway that not only yields Cer but may directly
   modulate vascular reactivity. Significant alterations were observed in
   adiposome Cer, DG, and specific free FAs among individuals with hepatic
   steatosis. This highlights the potential role of adiposomes in the
   interplay between adipose tissue and the liver in obesity, where
   excessive lipolysis and free FA release may contribute to hepatic lipid
   accumulation and the development of fatty liver ([176]53). Regression
   analyses in the current study further support these relationships
   between adiposome lipids and metabolic diseases/risks, as revealed by
   the strong relationships between adiposome lipids and cardiometabolic
   indices, including BMI, WC, HOMA-IR, systemic inflammation, and
   vascular function (brachial FMD and arteriolar FID). Ceramides, free
   FAs, and ACar were positively correlated with adiposity and insulin
   resistance but were inversely related to endothelial function. At the
   same time, SM and FAHFAs displayed protective associations linked to
   improved glycemic control and vascular health. Prior research has
   linked circulating lipids such as monoacylglycerols, DG, free FAs, and
   cholesteryl esters to cardiometabolic risk, emphasizing the roles of
   specific lipids in obesity, insulin resistance, inflammation, and
   vascular health ([177]54). However, our study investigates these
   relationships within adiposomes, offering comprehensive insights into
   their contribution to metabolic and vascular dysfunction.

   Obese adiposomes had lower TG levels, regardless of whether they were
   on lipid-lowering medication, which may indicate reduced lipogenesis or
   increased lipolysis. In obesity, lower lipogenesis might reflect an
   impaired ability to synthesize and store TGs due to associated
   metabolic stress or hormonal imbalances ([178]55), while enhanced
   lipolysis may suggest an increased breakdown of stored TGs driven by
   elevated adipose tissue inflammation, insulin resistance, or heightened
   catecholamine activity ([179]56–[180]58). Higher adiposome TG levels
   were also observed in hypertensive compared with nonhypertensive
   individuals with obesity, potentially influenced by antihypertensive
   medications, particularly calcium channel blockers (e.g., amlodipine)
   and β blockers ([181]59). These drugs can alter lipid metabolism by
   affecting adipocyte hormonal and signaling pathways. Our analysis
   confirmed a significant effect of these medications on adiposome TG and
   SM levels. To better understand the paradox of having lower TG in obese
   adiposomes, future studies should include comprehensive analyses of TG
   content in the corresponding adipose tissue, as well as the activity
   and expression levels of key enzymes in lipid metabolism, such as
   lipoprotein lipase, hormone-sensitive lipase, or adipose TG lipase.
   Such studies may provide valuable insights into the role of adiposome
   lipids in obesity and related metabolic disorders.

   One of the unique aspects of this study was applying machine learning
   algorithms to classify obesity, cardiometabolic diseases, and
   clinically relevant risk states (vascular dysfunction, systemic
   inflammation) using adiposome lipidomic profiles. Our models accurately
   distinguished individuals with obesity from those who are lean, often
   surpassing 90% accuracy with strong sensitivity and specificity.
   Notably, certain Cer species (Cer-NS d15:1/15:0, Cer-NS d18:1/23:0) and
   FAHFAs (FAHFA 18:0/20:2) emerged repeatedly among the top
   discriminators, underscoring their utility as robust biomarkers. These
   models were similarly effective at stratifying participants with
   obesity who have diabetes, hypertension, dyslipidemia, and liver
   steatosis versus those who do not. Although the random forest models
   tended to have slightly lower sensitivity in some comparisons (e.g.,
   ~63% for diabetic versus nondiabetic classification), the decision
   trees often performed exceptionally well (up to ~94% accuracy for
   diabetic and ~91% for hypertensive classification). These data
   collectively argue that adiposome lipid profiles could help identify
   individuals at the greatest risk of obesity-related complications,
   informing targeted intervention strategies.

   Our machine learning analyses classified participants as having normal
   or impaired vascular function based on brachial FMD or arteriolar FID
   thresholds and low or high inflammation defined by IL-6 and CRP levels.
   Classification accuracy often exceeded 85% and sometimes reached
   98%–100% in the impaired vascular function group, underscoring the
   strong predictive power of adiposome lipid composition. Similarly,
   individuals with high inflammation were identified with high accuracy.
   Notably, Cer-NS d15:1/15:0 repeatedly emerged among the top features,
   suggesting a key role in endothelial dysfunction and inflammatory
   signaling. These findings highlight the translational potential of
   adiposome lipidomics. With standardized protocols and larger validation
   cohorts, it may be possible to develop clinically validated models to
   predict or monitor insulin resistance, vascular dysfunction, and
   inflammation in obesity, enabling personalized therapies and earlier
   interventions.

   Finally, our LIPEA/KEGG-based pathway enrichment results revealed
   perturbations in glycerophospholipid, sphingolipid, and FA metabolism,
   along with pathways including arachidonic acid, inositol phosphate, and
   steroid biosynthesis, among others. Notably, the involvement of
   AGE/RAGE signaling and vascular smooth muscle contraction hints that
   these lipidomic changes may converge on pathways known to modulate
   endothelial homeostasis and vasoreactivity. Moreover, detecting
   ferroptosis, autophagy, and central signaling cascades (TNF, PI3K/Akt,
   mTOR, AMPK) underscores the broad cellular ramifications of adiposome
   lipids. Ceramides, for instance, are recognized triggers of autophagy
   and apoptosis in endothelial cells, while specific phospholipids might
   influence ferroptosis by modulating lipid peroxidation
   ([182]60–[183]62). This network-level view supports a scenario in which
   adiposome-mediated lipid transport from hypertrophic or inflamed
   adipose tissue exerts far-reaching systemic effects, orchestrating
   multiple mechanistic arms of metabolic and vascular pathology.

   In summary, adiposomes loaded with bioactive lipids may carry signals
   from dysfunctional adipose tissue to distant organs. Elevated Cer and
   free FAs in these vesicles can drive ectopic fat deposition,
   inflammatory cytokine release, and disrupted insulin signaling, while
   the depletion of protective lipids (FAHFAs, phospholipids, SM) dampens
   antiinflammatory capacity ([184]63, [185]64). Over time, this imbalance
   can lead to chronic inflammation, hepatic steatosis, and vascular
   remodeling that promotes hypertension and atherosclerosis. Although it
   remains unclear whether adiposome changes drive or simply reflect
   metabolic decline, our data suggest that adiposomes act as integrative
   agents in obesity pathophysiology. These findings align with growing
   evidence that obesity increases EV secretion of proinflammatory,
   insulin-desensitizing cargo. By providing a high-resolution lipidomic
   profile in participants who are lean or have obesity, we extend these
   observations and show that machine learning classifiers can accurately
   distinguish multiple metabolic phenotypes from adiposome lipid
   patterns. Unlike many studies on plasma-derived EVs, we examined
   adiposomes directly from adipose tissue biopsies, giving a
   tissue-specific view of vesicle composition. Strong links between
   adiposome lipids, cardiometabolic indices, and vascular function
   underscore a pivotal role for local adipose-derived vesicles in
   obesity, expanding the concept of adipo-organ crosstalk.

   Adiposomes carry diverse lipids reflective of adipose tissue function
   and metabolic states. They uniquely contain high neutral lipid levels,
   notably TG, DG, and CE, aligning with adipocytes’ role in lipid
   storage. Adiposomes are also rich in phospholipids like PC, PE, and PS,
   with distinct fatty acyl chain profiles compared with EVs from other
   cells. Additionally, bioactive lipids such as FA, FAHFA, Cer, and SM
   highlight adiposomes’ roles in metabolic regulation and inflammation
   ([186]63, [187]65). EV lipidomes differ by cellular origin: MSC-derived
   EVs contain lower neutral lipids but higher cholesterol, PS, and Cer
   ([188]66). Hepatocyte-derived EVs emphasize phospholipids, complex
   cholesterol metabolites, and higher sphingolipids ([189]67), while
   neuronal-derived EVs feature glycolipids like gangliosides, high
   cholesterol, and SM levels but lower neutral lipids ([190]68). Platelet
   and plasma EVs are enriched in PS, essential for coagulation, yet show
   lower neutral lipids compared with adiposomes ([191]69). Muscle-derived
   EVs transport unique Cer and glycerolipids, affecting adipocyte
   function ([192]70). Future studies will systematically compare our
   adiposome lipidome with datasets from various tissues available in
   repositories like Vesiclepedia ([193]http://www.exocarta.org/) and
   ExoCarta ([194]http://www.microvesicles.org/) to identify unique and
   shared lipid signatures across sources. Due to their complexity, such
   evaluations extend beyond this study but remain essential to our future
   research.

   This study has many strengths, including (a) comprehensive phenotyping,
   for which we integrated extensive clinical, biochemical, vascular, and
   lipidomic data, enabling robust correlations and machine learning
   analyses. (b) For high-resolution lipidomics, our approach encompassed
   595 lipid species spanning 19 classes, capturing a broad scope of lipid
   perturbations. (c) Direct adiposome isolation used adipose tissue
   biopsies to reduce confounding from EVs of nonadipose origin and
   confirm adipocyte-specific markers (PPARγ, FABP4). (d) Comparisons
   across comorbidities categorized participants with obesity by diabetes,
   hypertension, dyslipidemia, and liver steatosis status, providing
   valuable insight into differential lipid signatures linked to each
   condition. However, this study has limitations, including (a)
   cross-sectional design, since our findings reveal associations rather
   than causal relationships. A longitudinal approach could clarify
   whether the lipid shifts precede or follow the onset of insulin
   resistance and vascular dysfunction. (b) Sample size and diversity is
   another limitation. Although the total sample was sizable (n = 122),
   subgroups with particular conditions (e.g., dyslipidemic versus
   nondyslipidemic) were smaller, potentially limiting statistical power
   for certain comparisons. More ethnically and geographically diverse
   cohorts are also needed to generalize the results. (c) Medication
   effects is a limitation because many participants with obesity were on
   antihypertensive, antidiabetic, or lipid-lowering therapy. While our
   analyses tracked and accounted for these medications, isolating their
   full effect on lipidomic profiles remains challenging. (d) Functional
   studies is a limitation because, although we identified potential
   mechanistic pathways, our data remains largely correlative. Future
   functional experiments (e.g., adiposome transfer to cultured cells or
   by injecting animals) are needed to confirm the direct biological
   actions of specific lipid species. (e) While optimized to broadly
   detect multiple lipid classes, our untargeted lipidomics approach had
   limited sensitivity for complex sphingolipids such as gangliosides.
   Therefore, efforts are currently underway to optimize a targeted
   sphingolipidomic method to comprehensively assess these species in
   future studies. (f) The current analysis utilized 27 internal
   standards, ensuring that at least 2 distinct standards represented each
   lipid class. Five of these standards showed statistically significant
   differences between the lean and obese groups. These changes were
   likely due to matrix effects, such as ion suppression or coelution with
   endogenous lipids ([195]71). To maintain reliable quantification, we
   excluded these standards from further analysis, ensuring the stability
   of our results and minimizing the risk of skewed data. (g) Finally, we
   did not include an overweight category in our analysis. Including this
   group would have provided a more nuanced understanding of how lipid
   profiles and adiposome content change across the full spectrum of body
   weight. Despite these limitations, the study offers a compelling
   portrait of how adiposome lipid composition is reshaped in obesity,
   pointing to potential diagnostic and therapeutic strategies in
   obesity-related diseases.

Conclusion and future perspectives.

   In summary, this study demonstrates that obesity is linked with
   pronounced alterations in adiposome lipid content, including elevated
   Cer, free FAs, and ACar and reduced phospholipids, SM, and FAHFAs.
   These disruptions correlate strongly with systemic inflammation,
   insulin resistance, and endothelial dysfunction, affecting adiposomes
   as active participants in obesity pathogenesis rather than mere
   bystanders. However, while these findings illustrate robust
   associations, they do not establish causation. By leveraging advanced
   lipidomics and machine learning, we identified lipid species and
   overall lipid classes that not only distinguish individuals with
   obesity from those who are lean but also accurately classify
   comorbidities (diabetes, hypertension, dyslipidemia, steatosis) and
   predict impaired vascular function or high inflammation states. Our
   work underscores the potential utility of adiposome-derived lipid
   signatures in improving risk assessment, early diagnosis, and
   eventually personalizing therapies for obesity-related metabolic and
   cardiovascular diseases. Future studies aimed at mechanistically
   unraveling how adiposomes orchestrate intertissue communication,
   validating candidate lipids as functional mediators, and exploring
   interventions that selectively modify adiposome cargo hold promise for
   advancing obesity management. The comprehensive view provided here
   enriches our understanding of the lipid-mediated crosstalk in obesity
   and sets the stage for translational research targeting EV biology for
   cardiometabolic benefit.

   Future work should prioritize mechanistic validation and causality. In
   vitro exposures of endothelial cells or hepatocytes to purified
   adiposomes with specific lipid profiles could determine whether Cer,
   free FAs, or depleted FAHFAs directly impair insulin signaling,
   activate inflammatory pathways, or compromise NO-mediated vasodilation.
   Animal models of diet-induced obesity manipulating adiposome production
   or lipid species would reveal if reversing specific lipid imbalances
   improves metabolic and vascular outcomes. Another promising avenue
   involves targeting adiposome secretion or cargo. Small-molecule
   inhibitors of nSMase may reduce Cer overproduction in adiposomes,
   enhancing insulin sensitivity and vascular function. Conversely,
   boosting protective lipids like FAHFAs could restore immunometabolic
   balance. Such interventions demand a thorough understanding of
   adipocyte-derived EV biogenesis and downstream molecular targets.
   Moreover, our machine-learning data suggest that analyzing adiposome
   lipidomic signatures could guide personalized risk assessments.
   Moreover, integrating lipidomics with proteomics or metabolomics may
   refine biomarker accuracy and uncover additional mechanistic insights
   for obesity management.

   Studies comparing lipids in EV membranes versus their internal contents
   highlight distinct lipid profiles, with EV membranes being enriched in
   cholesterol, SM, and PC. These lipids contribute to enhanced membrane
   stability, rigidity, and cargo sorting. In contrast, the internal lipid
   content, including TGs and cholesteryl esters, reflects cellular
   metabolic states or stress ([196]72). The interplay between membrane
   and internal lipids is crucial for regulating EV function. While the
   current investigation did not differentiate between membrane and
   internal EV lipids, our future studies will specifically address this
   distinction to gain deeper insights into how these lipid profiles
   govern EV biogenesis, cargo composition, and their role in
   intercellular communication, which could be pivotal for advancing
   therapeutic applications targeting EVs. Finally, future studies are
   required to compare the lipidome of s.c. and visceral adiposomes to
   advance our understanding of how distinct adipose depots influence
   vascular and metabolic health.

Methods

Sex as a biological variable.

   In designing this cross-sectional study, we recruited both male and
   female adults undergoing elective surgery, ensuring adequate
   representation in lean and obese groups. Clinical and laboratory
   assessments (e.g., metabolic and vascular parameters) were recorded
   separately for men and women. Data were analyzed for main effects of
   sex and potential interactions with adiposity. Where applicable, post
   hoc analyses explored sex-specific differences in adiposome release,
   lipid composition, and associated cardiometabolic outcomes.

Subject enrollment.

   This study recruited 75 adults with obesity (45 females, 30 males, BMI
   ≥ 30 kg/m^2) undergoing bariatric surgery (sleeve gastrectomy) and 47
   lean adults (26 females, 21 males, BMI < 25 kg/m^2) undergoing elective
   surgery (e.g., hernia repair) at the University of Illinois Hospital in
   Chicago. The race/ethnicity composition of the obese group included AA
   (n = 42), H (n = 20), and NHW (n = 13). In the lean group, the
   distribution was as follows: AA (n = 23), H (n = 15), and NHW (n = 9).
   Sex and race/ethnicity were self-reported. Ages ranged from 19 to 50
   years. Blood samples and physiological measurements were taken 2–3
   weeks before surgery, and VAT was collected during surgery. Exclusion
   criteria included current pregnancy, smoking, previous bariatric
   surgery, and current liver failure, renal failure, heart failure,
   cancer, or autoimmune disease.

Cardiometabolic risk, adiposome isolation and lipid analysis, and vascular
function assessment.

   Detailed methods are provided in the [197]Supplemental Methods.

Statistics.

   All statistical analyses were conducted using SPSS statistical software
   (version 26.0; SPSS Inc.) and RStudio (v.4.4.1). Data are presented as
   the mean ± SD for continuous variables or frequencies and percentages
   for categorical variables. All variables were tested for normality
   using the Shapiro-Wilk test and visualized using histograms and Q-Q
   plots. Based on the results, parametric tests were used for normally
   distributed data, while nonparametric tests were applied for
   nonnormally distributed data. 2-tailed Student’s t tests or 1-way
   ANOVA, adjusted for age, sex, and race/ethnicity, assessed
   significance, and P values were corrected by Benjamini-Hochberg FDR.
   When the assumptions of the t test were not met, the Mann-Whitney U
   test was used. Log[2] FC was calculated as the log[2] of the ratio of
   average expression in the obese group to that in lean controls. Linear
   regression models (adjusted for age, sex, and race/ethnicity) evaluated
   associations between lipids and BMI. The reported estimates (β)
   represent the effect of a 1-SD increase in the independent variables.

   PCA was conducted using RStudio on scaled data (mean 0, SD 1) via the
   prcomp function that uses singular value decomposition (SVD).
   Hierarchical clustering was performed using the Ward method on scaled
   data, and data matrices were visualized using the heatmap function to
   identify clusters. Visualizations (bar plots, box plots, volcano plots)
   were generated with ggplot2. A statistically significant linear
   relationship between continuous variables was tested using a bivariate
   Pearson or Spearman correlation. Lipids with FDR < 0.05 were imported
   into LIPEA for pathway mapping using Fisher’s exact tests.

   IBM SPSS (version 29) was employed to produce both random forest plots
   and decision trees and to configure prediction models. Obesity,
   diabetes, hypertension, dyslipidemia, liver steatosis, vascular
   dysfunction, and high systemic inflammation were used as the response
   variables, and adiposome concentrations of lipid species were used as
   quantitative exploratory variables. For random forest, the sample size
   was 122 for predicting obesity and 75 for predicting cardiometabolic
   diseases in the obese group; the number of trees ranged from 300 to
   500, the maximum tree depth was 20, and imbalanced data were handled by
   adjusting weights inversely proportional to class frequencies. Decision
   trees were constructed via the Classification and Regression Trees
   (CART) algorithm included in SPSS’s Decision Trees module, using a
   minimum parent size of 5, a minimum son size of 2, a maximum tree depth
   of 3, and a Gini splitting criterion. The model was pruned and
   validated with standard cross-validation (10-fold) to minimize
   overfitting. Each prediction model’s performance was then assessed
   using metrics such as accuracy, sensitivity, specificity, and the AUC
   of the ROC, ensuring interpretable results.

Study approval.

   The research was performed in compliance with the Declaration of
   Helsinki and approved by the IRB of the University of Illinois at
   Chicago (protocol no. 2021-1113, approved in October 2021). Informed
   written consent was obtained from all the participants involved in the
   study.

Data availability.

   The datasets used and/or analyzed during the current study are
   available in the [198]Supporting Data Values file and can also be
   obtained from the corresponding author upon reasonable request.

Author contributions

   EM, IM, AM, MMA, MCA, and AMM generated data and figures. IM, MHM, GS,
   FMB, MAM, and CH assisted with patient recruitment and sample
   collection. BTL and AMM led study conceptualization, manuscript review,
   and editing. AMM secured funding and supervised the research. All
   authors reviewed and approved the final manuscript.

Supplementary Material

   Supplemental data
   [199]jciinsight-10-191872-s040.pdf^ (243.2KB, pdf)
   ICMJE disclosure forms
   [200]jciinsight-10-191872-s041.pdf^ (702.7KB, pdf)
   Unedited blot and gel images
   [201]jciinsight-10-191872-s042.pdf^ (161.6KB, pdf)
   Supporting data values
   [202]jciinsight-10-191872-s043.xlsx^ (193.6KB, xlsx)

Acknowledgments