An ANGPTL8‐AKT2‐mTOR Axis Drives Adipose Senescence and Aging‐Related Functional Decline

Figures

• 

  FIGURE 1. Overview of the development and analytical workflow for the MBA8‐Clock and AIMR models.

• 

  FIGURE 2. Construction, performance, and interpretability of the MBA8‐Clock. (A) Comparative performance of five machine‐learning algorithms, including multilayer perceptron (MLP), random forest (RF), extreme gradient boosting (XGB), light gradient boosting machine (LGB), and logistic regression (LR), evaluated using 100 repeated runs. Model performance was assessed by R 2, root mean square error (RMSE), median absolute error (MED), and mean absolute error (MAE). Bars represent mean values across runs, with error bars indicating standard deviation. (B) Representative comparison between predicted biological age and chronological age for 50 randomly selected participants using the MLP‐based MBA8‐Clock. Lines depict individual trajectories, illustrating the concordance between predicted and actual age estimates. (C) SHapley Additive exPlanations (SHAP) beeswarm plot summarizing the global feature importance of all predictors included in the MBA8‐Clock. Features are ranked by mean absolute SHAP value. Each point represents one individual, with color indicating the relative feature value (low to high). (D) Partial dependence plot illustrating the nonlinear relationship between circulating…

• 

  FIGURE 3. Development, performance evaluation, and interpretability of the AIMR‐Model. (A) Receiver operating characteristic (ROC) curves of five machine learning algorithms, including random forest (RF), extreme gradient boosting (XGB), light gradient boosting machine (LGB), logistic regression (LR), and multilayer perceptron (MLP), for mortality risk prediction. (B) Confusion matrices of RF, XGB, LGB, LR, and MLP models illustrating classification performance in the test set. (C) Bar plots summarizing recall, precision, accuracy, and F1 score for each model, based on 100 repeated runs to assess model stability and robustness. (D) SHapley Additive exPlanations (SHAP) beeswarm plot of the final MBA8‐Clock, displaying the global importance and directional effects of individual features on model output. (E) SHAP interaction value analysis of the top seven contributing features, including Age, ANGPTL8 (A8), creatinine, systolic blood pressure (SBP), γ‐glutamyl transferase (GGT), aspartate aminotransferase (AST), and postprandial blood glucose (PBG), highlighting pairwise feature interactions within the model. (F) Representative SHAP and LIME explanations for a randomly selected participant…

• 

  FIGURE 4. Phenotypic stratification and SHAP‐based characterization of subgroups defined by the MBA8‐Clock and AIMR‐Model. (A) Radar plot comparing standardized clinical values of the top five SHAP‐ranked features, including ANGPTL8, diastolic blood pressure (DBP), creatinine, systolic blood pressure (SBP), and body weight, between two subgroups stratified by the MBA8‐Clock (individuals with age overestimation versus age underestimation). The radar plot illustrates markedly distinct phenotypic profiles between the two subgroups. (B) Radar plot integrating SHAP values of the same two MBA8‐Clock‐defined subgroups, highlighting differences in feature contributions to biological age acceleration or deceleration. (C) Radar plot comparing standardized clinical values of the top five SHAP‐ranked features, including ANGPTL8, γ‐glutamyl transferase (GGT), SBP, age, and creatinine, between two subgroups stratified by the AIMR‐Model (individuals with high predicted mortality risk versus low predicted mortality risk). Distinct phenotypic patterns are observed between the two mortality risk strata. (D) Radar plot integrating SHAP values of the AIMR‐Model‐defined high‐ and low‐risk subgroups,…

• 

  FIGURE 5. Genetic ablation of Angptl8 attenuates aging‐associated phenotypes, preserves musculoskeletal integrity, and improves functional performance in mice. (A) Schematic overview of the experimental design. Homozygous Angptl8 knockout (Angptl8 −/− ) mice were generated on a C57BL/6J background, with age‐ and sex‐matched wild‐type (WT) littermates used as controls. Mice of both genotypes were enrolled at 4 months of age and randomly assigned to two predefined cohorts. One cohort was sacrificed at 22 months of age for comprehensive tissue‐level, histological, and molecular analyzes, whereas the second cohort was longitudinally followed until natural death to assess lifespan and survival outcomes. (B) Bar graph comparing circulating ANGPTL8 protein concentrations in serum from young (Young, n = 9) and aged (Old, n = 9) mice, demonstrating an age‐associated increase in ANGPTL8 levels. Data are presented as mean ± SEM. Statistical significance was determined using an unpaired two‐tailed Student's t‐test. *p < 0.05, **p < 0.01, ***p < 0.001. (C) Kaplan–Meier survival curves comparing overall survival between WT mice (n = 10) and Angptl8 −/− mice (n = 10). Mice were monitored until…

• 

  FIGURE 6. Age‐associated upregulation and adipose‐enriched expression of Angptl8 drive adipocyte senescence in mice and humans. (A) Representative western blot analysis of ANGPTL8 protein expression in liver tissues from young (4‐month‐old) and aged (22‐month‐old) mice. (B) Representative western blot analysis of ANGPTL8 protein expression in epididymal white adipose tissues from young (4‐month‐old) and aged (22‐month‐old) mice. (C) Representative liver tissue sections from young (left) and aged (right) mice, with corresponding spatial localization of ANGPTL8 protein, illustrating age‐dependent changes in its distribution. (D) Quantification of the proportion of hepatocytes expressing Angptl8 in liver tissues from young and aged mice. (E) Quantitative analysis of Angptl8 expression intensity in hepatocytes from young and aged mouse livers. UMAP visualization of single‐cell transcriptomic profiles from abdominal adipose tissue of young human donors. Distinct colors denote annotated cell clusters (legend on the right). Lower panels summarize the proportion of ANGPTL8‐expressing cells and the corresponding expression intensity across clusters. (F, G) UMAP visualization of single‐cell…

• 

  FIGURE 7. ANGPTL8 promotes adipocyte senescence through AKT2–mTOR–S6K signaling. (A) Predicted binding mode of the ANGPTL8–AKT2 complex. ANGPTL8 is shown in pink and AKT2 in blue. Yellow dashed lines indicate putative hydrogen bond interactions between the two proteins. Molecular docking predicted a potential interaction interface between ANGPTL8 and AKT2, with a docking score of approximately −263.76 kcal/mo. (B) Immunoprecipitation (IP) assays in undifferentiated 3 T3‐L1 preadipocytes co‐transfected with HA‐tagged ANGPTL8 and Flag‐tagged AKT2. Cell lysates were immunoprecipitated with anti‐HA or anti‐Flag antibodies, followed by immunoblotting with the indicated antibodies to assess protein–protein interactions. (C) Representative western blot images showing the expression and phosphorylation levels of AKT2, mTOR, and S6K in primary preadipocytes isolated from 4‐month‐old mice. Cells were assigned to five experimental groups: Wild‐type (WT), Angptl8 −/− (Angptl8 KO), Angptl8 overexpression (Angptl8 OE), Angptl8 OE treated with the AKT inhibitor MK‐2206, and Angptl8 OE treated with the mTOR inhibitor rapamycin. (D) Senescence‐associated β‐galactosidase (SA‐β‐gal) staining of epididymal…