Integrated multi‐omic analyses uncover the effects of aging on cell‐type regulation in glucose‐responsive tissues

Abstract Aging significantly influences cellular activity and metabolism in glucose‐responsive tissues, yet a comprehensive evaluation of the impacts of aging and associated cell‐type responses has been lacking. This study integrates transcriptomic, methylomic, single‐cell RNA sequencing, and metabolomic data to investigate aging‐related regulations in adipose and muscle tissues. Through coexpression network analysis of the adipose tissue, we identified aging‐associated network modules specific to certain cell types, including adipocytes and immune cells. Aging upregulates the metabolic functions of lysosomes and downregulates the branched‐chain amino acids (BCAAs) degradation pathway. Additionally, aging‐associated changes in cell proportions, methylation profiles, and single‐cell expressions were observed in the adipose. In the muscle tissue, aging was found to repress the metabolic processes of glycolysis and oxidative phosphorylation, along with reduced gene activity of fast‐twitch type II muscle fibers. Metabolomic profiling linked aging‐related alterations in plasma metabolites to gene expression in glucose‐responsive tissues, particularly in tRNA modifications, BCAA metabolism, and sex hormone signaling. Together, our multi‐omic analyses provide a comprehensive understanding of the impacts of aging on glucose‐responsive tissues and identify potential plasma biomarkers for these effects.


| INTRODUC TI ON
Aging is a complex and multifactorial process characterized by the gradual accumulation of metabolic changes (Lopez-Otin et al., 2023).One of the most prominent aging-associated disorders is the impairment of nutrient-sensing and physiological functions of glucose-responsive tissues (Ou et al., 2022;Wilkinson et al., 2018).Adipose and muscle are the two major organs responsible for nutrient availability and energy metabolism: Adipose tissue plays a crucial role in storing lipids for energy homeostasis, whereas skeletal muscle accounts for approximately 80% of postprandial glucose disposal (DeFronzo & Tripathy, 2009;Kershaw & Flier, 2004).Understanding the gene-regulatory networks involved in the rewiring of metabolism in these glucose-responsive tissues during aging is critical to identify molecular regulators of the aging process and prioritize treatment strategies.
Adipose and muscle tissues are complex and heterogeneous, consisting of diverse cell types that are differentially impacted by the aging process.Aging is accompanied by a decrease in fat depot size and impairment in the differentiation of preadipocytes into adipocytes, along with elevated expression of antiadipogenic genes (Kirkland et al., 2002).In aged mice, visceral adipose tissue exhibits increased pro-inflammatory macrophages, upregulated TNFα and IL-6, and downregulated PPARγ expressions (Wu et al., 2007).Aging also significantly impacts skeletal muscles, leading to sarcopenia, characterized by a decline in tissue mass and function (Sousa-Victor et al., 2015).
Although prior studies have highlighted the significance of cell-type regulations in adipose and muscle tissues, a comprehensive understanding of aging-related cell-type specificity remains elusive.
Currently, single-cell studies are widely employed to investigate cell type-specific gene activities in glucose-responsive tissues (De Micheli et al., 2020;Emont et al., 2022;Kedlian et al., 2024).However, these experiments are hindered by high costs and gene dropouts, limiting their application in large-scale aging studies.To address these limitations, we previously developed the Multiscale Embedded Gene Co-expression Network Analysis (MEGENA) approach.This method enables the direct inference of cell type-specific high-resolution network modules from bulk transcriptome data sets of human populations (Song & Zhang, 2015;Xu et al., 2022).Additionally, we have established the African American Genetics of Metabolism and Expression (AAGMEx) cohort, which includes comprehensive physiological and multi-omic profiling of adipose and muscle tissues (Sajuthi et al., 2016;Sharma et al., 2016).By combining our MEGENA approach with the AAGMEx cohort, we aim to investigate cell type-specific gene activities associated with aging in glucose-responsive tissues.
Previous studies have identified certain plasma metabolites as potential indicators of the aging process (Chaleckis et al., 2016;Saito et al., 2016).However, the precise mechanistic connections between these biomarkers and the gene-regulatory networks in glucoseresponsive tissues during aging remain incompletely understood.
Leveraging the well-curated AAGMEx cohort, we hypothesize that systematic omics data profiling can enable the detection of circulating metabolites as biomarkers reflecting inter-tissue communication.Therefore, this study focuses on two principal objectives: (1) elucidating the aging-related cell-type networks within adipose and muscle tissues through an integrated examination of transcriptome, methylome, and single-cell analyses and (2) identifying plasma metabolites as dependable biomarkers for intertissue communication during the aging process.By addressing these objectives, we aim to provide a comprehensive understanding of the influence of aging on cell-type regulations and intercellular interactions within glucoseresponsive tissues.

| Study design and analysis pipeline
We conducted a multi-omic analysis on the AAGMEx cohort, which comprised 222 healthy participants (123 males and 99 females), spanning an age range of 18-61 years (Table S1).The AAGMEx cohort offers comprehensive participant data, including demographics such as age and sex, along with physiological metrics like body mass index (BMI) and glucometabolic measures (e.g., insulin sensitivity indices: S I and Matsuda Index).The multi-omic data set from the AAGMEx cohort encompasses transcriptomic data derived from microarrays, methylomic data obtained through reduced representation bisulfite sequencing (RRBS), and metabolomic data generated via untargeted liquid chromatography-mass spectrometry (LC-MS).
For the purpose of independent cohort validation, we utilized adipose and muscle tissue samples from the GTEx consortium, which includes 604 individuals aged between 21 and 70 years.The transcriptome of GTEx data set was acquired from postmortem tissues and sequenced using RNA sequencing (RNA-seq) technology.
Our analysis workflow began with the application of MEGENA to adipose and muscle tissues, aiming to identify co-expression networks and aging-responsive modules (Figure S1).Subsequently, we integrated cell-type markers from single-cell studies to characterize agingassociated cell type-specific modules.To validate the transcriptome findings, we employed cell proportion deconvolution and conducted multi-omic analyses on methylation and single-cell data.Additionally, we performed metabolome profiling of blood plasma from the AAGMEx cohort to investigate aging-related interactions between circulating metabolites and gene activities in glucose-responsive tissues.
Through pairwise module-module and metabolite-gene correlations across plasma, adipose, and muscle tissues, our analyses revealed tissue-tissue communications associated with the aging process.

| Aging regulates cell type-specific modules of adipose tissue
We first explored the impacts of aging on gene expression within adipose tissue of the AAGMEx cohort.As linear regression analysis indicated a moderately significant correlation (p = 0.06) between age and BMI (Figure S2a), we normalized gene expression levels by regressing out the covariate effects of BMI and sex.Through Spearman correlation analysis on these normalized datasets, we identified 1994 genes associated with aging in adipose tissue, exhibiting a median absolute correlation coefficient of 0.23 and a False Discovery Rate (FDR) below 0.05 (Table S2).Subsequent multivariate regression analysis affirmed the aging association for 925 (89%) upregulated genes and 841 (88%) downregulated genes (FDR < 0.05) (Figure S2b).To corroborate these findings within the AAGMEx cohort, Spearman correlation analysis was applied to adipose tissue data from the GTEx consortium, identifying 1501 genes associated with aging, including 263 genes common to both cohorts (Figure 1a).Among these shared genes, 248 (94%) exhibited consistent aging correlation directions across both cohorts, while merely 15 (6%) displayed reversed directions, underscoring a general agreement in aging regulatory patterns (Figure 1b).
Notably, the 248 consistently associated genes demonstrated stronger aging correlations in the AAGMEx cohort compared to the GTEx cohort, as evidenced by higher absolute Spearman correlation coefficients (Figure 1c).This indicates that despite its relatively smaller size, the AAGMEx cohort is particularly adept at detecting genes associated with aging.
Next, we constructed a MEGENA co-expression network of the AAGMEx adipose tissue using covariates-adjusted gene expressions.We identified 549 hierarchical modules, among which 75 modules were enriched for the aging-upregulated genes and 52 modules were enriched for the aging-downregulated genes (Fisher's exact test [FET], FDR < 0.05) (Table S3).The eigengenes (the first principal component of module expression) of top-ranked modules, including M3 and M14, showed different trends of correlation with age (Figure 1d).By identifying cell-type marker gene signatures from the single nucleus RNA-sequencing (snRNA-Seq) data set of adipose tissue (n = 57,599 nuclei) (Emont et al., 2022), we performed an enrichment test of network modules for the marker genes of each cell type in the adipose.We found that 34 aging-associated modules (size >50 gene members) of the adipose were significantly enriched for the marker gene signatures of 12 different cell types (FET FDR < 0.05) (Figure 1e and Table S4).The major aging-associated modules M3 and M14 were enriched for the maker gene signatures of immune cells and adipocytes, respectively.The modules from immune cells and adipose stem and progenitor cells (ASPCs) showed positive correlations with age, while the modules from the adipocytes negatively correlated with age.
To corroborate the network findings derived from the AAGMEx cohort, we performed MEGENA on the adipose tissue data from the GTEx consortium.The network preservation analysis showed that the principal cell type-specific modules identified within the AAGMEx cohort are conserved within the GTEx dataset (Figure 1f).
For example, the AAGMEx module M3, characterized by its enrichment in immune cells, aligns with the GTEx module M406, with both modules exhibiting a positive correlation with age.Similarly, the adipocyte-specific module AAGMEx M14 and its counterpart in GTEx module M11 were negatively correlated with age.These results underscored the consistency of cell type-specific aging responses across two independent cohorts.

| Aging-associated modules regulate metabolic pathways of adipose tissue
To gain a deeper understanding of the impacts of aging, we annotated the aging-associated genes and network modules in the AAGMEx adipose tissue using the KEGG pathway database.Our analysis revealed that genes upregulated by aging were enriched in hematopoietic cell development and lysosome pathways (Figure 2a).
Conversely, genes downregulated by aging were enriched in the degradation of branched-chain amino acids (BCAAs; including valine, leucine, and isoleucine), propanoate, and fatty acids.These metabolic pathways were also prominent in the aging-associated cell type-specific modules.Notably, the immune cell module M3 was positively correlated with age and stood out as the most enriched for the lysosome pathway (Figure 2b).Within M3, 24 genes from the lysosome pathway were upregulated by age, including key hub genes HEXB and GBA encoding lysosomal enzymes beta-hexosaminidase and beta-glucocerebrosidase, respectively (Figure 2c).On the other hand, the adipocyte module M14 was involved in BCAA degradation.In M14, seven genes involved in the BCAA degradation pathway were downregulated by age, such as PCCA and PCCB which encode the enzyme propionyl-CoA carboxylase, a mitochondrial enzyme critical for BCAA catabolism (Figure 2d).
In previous studies, we have conducted extensive research on glucometabolic regulations within the AAGMEx cohort (Das & Das, 2024;Sharma et al., 2020;Xu et al., 2023).To further explore the relationship between aging and glucose response, we employed Spearman correlation analysis adjusted for age and sex to pinpoint the genes associated with BMI and insulin sensitivity (S I ).
Our findings revealed that in adipose tissue, 72% and 43% of agingassociated genes were correlated with BMI and S I , respectively (Figure S2c).Additionally, through network module enrichment analysis, we discovered several aging-associated modules enriched with BMI-and S I -associated genes.Notably, modules M3 and M14, unique to immune cells and adipocytes, respectively, were enriched with BMI-and S I -associated genes (Figure S2d).This finding indicates that these cell-specific modules likely play dual roles in regulating both aging and glucose metabolism processes.

| Aging regulates the proportion of different cell types and DNA methylation levels within adipose tissue
To assess the impact of aging on cell-type proportions, we employed bulk gene expression deconvolution to estimate the relative abundances of distinct cell types within the adipose tissue of the AAGMEx cohort.For this purpose, we utilized BisqueRNA, a method that relies on marker gene expression profiles derived from singlecell experiment (Emont et al., 2022;Jew et al., 2020).Correlation analysis revealed that aging differentially affects the proportions of various cell types (Table S5).Notably, the proportion of adipocytes exhibited a negative correlation with age (Figure 3a), whereas the proportions of ASPCs and immune cells, including macrophages, positively correlated with age.These deconvolution results align with the observed changes in gene expression patterns of cell typespecific modules, further highlighting the association between aging and alterations in cell-type proportions.
Given the well-established association between aging and epigenetic changes across various human tissue (Jones et al., 2015), we analyzed the gene methylation profiles with 1 kb tiling windows across the genome in adipose tissue of AAGMEx cohort.
Using Spearman correlation analysis, we identified 2925 genes exhibiting aging-related methylation patterns within a 1 kb window of their promoters (p < 0.05, Table S6).Among these, 428 genes demonstrated concurrent aging-associated changes in both expression and DNA methylation levels, including 130 cell typespecific marker genes.Specifically, 19, 20, and 35 marker genes were identified for adipocytes, ASPCs, and macrophages, respectively (Figure 3b and Table S7).IRS1, a gene crucial for insulin and IGF-1-related metabolic actions as well as adipocyte differentiation, displayed age-associated downregulation in expression along with increased methylation levels (Figure 3c).Similarly, ESR1, F I G U R E 1 Aging-associated genes and network modules in adipose tissue.(a) Venn diagram illustrating the overlap of aging-related genes in the AAGMEx and GTEx cohorts.(b) Scatterplot comparing age correlation coefficients of aging-related genes across AAGMEx and GTEx cohorts.The x-and y-axes represent coefficients calculated using Spearman correlation in each cohort.(c) Distribution of Spearman coefficients for genes commonly associated with aging in both AAGMEx and GTEx cohorts.(d) Sunburst plot visualizing hierarchy of MEGENA network modules enriched for aging-associated genes.The inner layers of the plot represent the parent modules of the outer layers.Color intensity indicates the enrichment significance, as measured by −log 10 (FDR).Positive values signify enrichments of agingupregulated genes, whereas negative values represent aging-downregulated genes.Modules emphasized for downstream characterization are marked with red asterisks.(e) Heatmap depicting the cell-type specificity of age-related network modules in adipose tissue.The top bar plot shows the correlation coefficients between module eigengenes and age.ASPC, adipose stem and progenitor cells; LEC, lymphatic endothelial cell; SMC, smooth muscle cell.(f) Heatmap displaying the significance of Fisher's exact test for the preservation of top 20 AAGMEx modules in both cohorts.The accompanying bar plots indicate the correlation coefficients between module eigengenes and age.
encoding an estrogen receptor inversely associated with adiposity and positively associated with insulin sensitivity, also exhibited age-related downregulation in expression along with elevated methylation.Beyond these genes, we also identified 2497 genes that solely exhibited aging-associated methylation changes.For instance, DIO3, a gene regulating thyroid hormone metabolism, displayed elevated methylation levels with age (r = 0.47, p = 2e-13), while its expression remained relatively unchanged.The genes that were associated with both aging and DNA methylation showed significant enrichment in 45 network modules, including the immune cell-specific M3 module and the adipocyte-specific M14 module (Figure 3d).It is worth noting that module M3 contained 149 genes (constituting 35% of the total genes), which were simultaneously influenced by aging and methylation.This underscores the intricate interplay between gene expression and methylation during the aging process in adipose tissue.
We conducted a detailed analysis of aging-related gene expressions across different cell types using single-nucleus RNA-sequencing (snRNA-seq) profiles from 13 individuals (mean age: 47 years, SD: ±15) (Emont et al., 2022).After adjusting for covariates such as body mass index (BMI) and sex, our pseudobulk analysis identified 45 genes significantly associated with aging (p < 0.05), which also exhibited consistent age-related patterns in bulk RNA-sequencing data from the AAGMEx adipose tissue cohort (Table S8).Notably, among these genes, ANXA1, a key regulator of immune responses and glucocorticoid-mediated effect, demonstrated upregulation with aging in adipocytes (Figure S3a

| Aging regulates cell type-specific network modules of muscle tissue
We further conducted a detailed investigation of the impact of aging on gene expression in muscle tissue from the AAGMEx cohort.After adjusting for BMI and sex covariates, we employed Spearman correlation analysis and identified 2497 genes significantly associated with aging (FDR < 0.05), which exhibited a median absolute correlation coefficient of 0.23 (Table S9).Using a multivariate regression model, we confirmed the association of 1101 (94%) aging-upregulated genes and 1203 (91%) aging downregulated genes (Figure S4a).To validate our findings in an independent cohort, we performed Spearman correlation analysis on the GTEx data set.This analysis revealed 2331 aging-associated genes in muscle tissue, including 397 genes commonly associated with aging in both cohorts (Figure 4a).Among the shared agingassociated genes, 365 (92%) genes exhibited the same direction of aging correlation, whereas only 32 (8%) genes showed the opposite direction (Figure 4b).Consistent with our observations in adipose tissue, the 365 conserved genes demonstrated a stronger correlation with aging in the AAGMEx cohort compared with the GTEx cohort (Figure 4c).
Using the MEGENA pipeline, we identified 521 hierarchical modules from the co-expression network in muscle tissue.Among these, 49 modules were enriched for aging-upregulated genes, whereas 66 modules were enriched for aging-downregulated genes (FET FDR < 0.05) (Table S10).The top-ranked aging-associated modules included M3, which positively correlated with aging, and M9, which was negatively regulated by aging (Figure 4d).To gain further insights, we analyzed single-cell transcriptome profiles (n = 22,000 cells) of human muscle tissue and identified network modules specific to different cell types (De Micheli et al., 2020).The enrichment test revealed that 26 aging-associated modules preferentially expressed marker genes of 10 cell types.Modules including M3 were enriched for marker genes from endothelial cells and immune cells, whereas modules including M9 were primarily derived from muscle fiber cells (Figure 4e and Table S11).Additionally, network preservation analysis demonstrated that major cell type-specific modules in the AAGMEx cohort, such as M3 and M9, were also conserved in the GTEx cohort (Figure 4f).
We further explored the impact of BMI and insulin sensitivity (S I ) on aging-associated genes in the muscle tissue.We found that 10% and 6% of aging-related genes were correlated with BMI and S I , respectively (Figure S4b).Several aging-associated modules were enriched for the BMI-and S I -linked genes.Notably, module M9, which was negatively correlated with age and unique to muscle fiber cells, was enriched for BMI-and S I -associated genes (Figure S4c), suggesting its role in both aging process and glucose response.

| Aging-associated modules participate in metabolic regulation of muscle tissue
By annotating the network modules using KEGG pathways, we discovered that the pathogenic infection pathway was enriched among the aging-upregulated genes.GO database annotation further broadened our understanding of the gene functions of these upregulated genes, revealing categories such as "RNA splicing" and "actin filament organization" among the top 10 ranked GO processes.On the other hand, the aging-downregulated genes impacted several crucial metabolic pathways in the KEGG database, including oxidative phosphorylation, glycolysis, and gluconeogenesis (Figure 5a).
Among the aging-associated modules involved in metabolism, M6 was enriched for glycolysis and gluconeogenesis.It comprised 961 genes, including 12 genes of the glycolysis and gluconeogenesis pathway downregulated by age (Figure 5b).For instance, PFKM encodes 6-phosphofructokinase, which catalyzes the phosphorylation of fructose 6-phosphate to fructose 1,6-bisphosphate, representing the major rate-limiting step of glycolysis (Schöneberg et al., 2013) (Figure S5a).Additionally, M9 was enriched for the oxidative phosphorylation pathway and contained 837 module genes.In M9, 32 genes of the oxidative phosphorylation pathway were downregulated by age, including hub genes CYC1, COX5B, and UQCRH (Figure S5b).These genes encode enzymes of the mitochondrial respiratory chain, crucial for oxidative phosphorylation functions.

| Aging regulates proportion and gene expression of different muscle fiber types
Skeletal muscles encompass two distinct muscle fiber types: slowtwitch (type I) and fast-twitch (type II).Previous studies have demonstrated that aging leads to preferential loss and atrophy of glycolytic, fast-twitch type II muscle fiber, indicating differential regulation of muscle fiber types by aging (Akasaki et al., 2014).
To further investigate this, we conducted deconvolution analysis on muscle tissues from the AAGMEx and GTEx cohorts, utilizing snRNA-Seq expression profiles (Jew et al., 2020;Perez et al., 2022).Our analysis revealed an increase in the proportion of slow-twitch type I muscle fibers and a decrease in the proportion of fast-twitch type II muscle fibers with age in both cohorts (Figure 5c).Consistent with this observation, we found that MYH7 and MYH1, markers of slowt-witch type I and fast-twitch type II fibers, respectively (Schiaffino & Reggiani, 2011), correlated with age in opposite directions in the AAGMEx cohort (Figure 5d).This   Periasamy & Kalyanasundaram, 2007).At the network level, the type I fiber marker genes (MYH7 and ATP2A2) were co-expressed in M23, which was enriched for aging-upregulated genes (FET FDR = 1.2e-24), whereas the type II fiber marker genes (MYH1 and ATP2A1) were co-expressed in M26, which was enriched for agingdownregulated genes (FET FDR = 3.6e-07) (Figure S5c).These results suggest distinct proportions and gene activities of muscle fiber types in response to the aging process.

| Aging regulates metabolome network in blood plasma
We conducted a thorough investigation to assess the impact of gene activities in adipose and muscle tissues on the levels of circulating metabolites during aging.Utilizing LC-MS, we profiled the metabolome of blood plasma from the AAGMEx cohort and successfully quantified 1124 metabolites in the plasma.After adjusting for BMI and sex, we identified 267 metabolites that were positively correlated with age and 61 metabolites that negatively correlated with age using the Spearman correlation test (FDR < 0.05).Subsequently, a multivariate linear regression model confirmed 221 (83%) aging-upregulated metabolites and 43 (70%) aging-downregulated metabolites (FDR < 0.05) (Table S12 and Figure S6).This observed difference might be attributed to the fact that the Spearman correlation excels at measuring the strength and direction of monotonic relationships between variables without assuming linearity.Furthermore, using MEGENA for metabolic network construction, we identified seven modules enriched for aging-upregulated metabolites and five modules enriched for agingdownregulated metabolites (Figure 6a and Table S13).Among these, the top-ranked modules included hierarchical modules M4, M18, and M70, which positively correlated with age and were involved in amino acid metabolism pathways (Figure 6b).On the other hand, M11 emerged as a top aging-downregulated module, playing a crucial role in androgenic steroid metabolism.
To identify the interactions between plasma components and glucose-responsive tissues, we conducted module-wise correlation analyses involving the aging-associated modules derived from plasma metabolome data and the transcriptome data of adipose and muscle tissues.The Spearman correlation analysis unveiled significant associations between the top-ranking aging-associated modules found in plasma (e.g., M18, M70, M11) and the corresponding aging-associated gene modules in adipose and muscle tissues (Figure S7a,b).Moreover, we also identified substantial correlations between the high-ranking aging-associated modules within adipose and muscle tissues themselves (Figure S7c).Significant correlations were observed between the adipose immune cell module M3 and the muscle fiber modules M23 and M26.Specifically, a positive correlation was found between adipose M3 and type I muscle fiber module M23 (r = 0.27, FDR = 0.004), while a negative correlation existed between adipose M3 and type II muscle fiber module M26 (r = −0.21,FDR = 0.04).
Given that adipose tissue is a crucial endocrine organ contributing to a pro-inflammatory environment, and that adipose-derived factors regulate mitochondrial function in skeletal muscle (Parra-Peralbo et al., 2021;Seldin et al., 2018), the observed associations imply inter-tissue communication influenced by the aging process within the human body.

| Plasma metabolites reflect aging-associated gene activities of glucose-responsive tissues
To gain a deeper understanding of the roles of circulating biomarkers during the aging process, we conducted a thorough analysis of metabolite networks and aging-related modules.We found that module M18 contained 89 metabolites and was positively correlated with age (r = 0.4, FDR = 1.5e-10).Among the hub metabolites of M18, three compounds, namely N1-methylinosine, N2,N2-dimethylguanosine, and N6-carbamoylthreonyladenosine, are involved in posttranscriptional chemical modifications of the tRNA maturation process (Figure 6c).The corresponding biosynthetic enzyme-encoding genes of the three compounds, including TRMT5, TRMT1, and YRDC (de Crecy-Lagard et al., 2019), also positively correlated with age in adipose or muscle tissues (Figure 6d and Figure S8).Correlation analysis further revealed that the plasma level of N1-methylinosine positively correlated with the biosynthetic gene TRMT5 in both adipose (r = 0.13, p = 0.05) and muscle (r = 0.19, p = 0.005) tissues.
Module M70, a sub-module of M18, contained 39 metabolites.A key compound within M70 was 2-methylmalonylcarnitine, a metabolite upregulated during aging and a byproduct of BCAA catabolism (Figure 6c,d).Consistent with these findings, the propionyl-CoA carboxylase (encoded by PCCA and PCC) responsible for BCAA degradation was downregulated with aging in both adipose and muscle tissues.
The plasma levels of 2-methylmalonylcarnitine negatively correlated with PCCA (r = −0.18,p = 0.008) and PCCB (r = −0.20,p = 0.003) expression in adipose tissue.Furthermore, we found that the α-subunit encoding gene of succinyl-CoA ligase (SUCLG1) was downregulated with aging in both adipose and muscle tissues (Figure S8).This supports the hypothesis that the accumulation of 2-methylmalonylcarnitine in plasma may also be attributed to the reduced function of succinyl-CoA ligase, as previous report (Van Hove et al., 2010).
The primary aging-downregulated module, M11, encompassed 34 metabolites and exhibited a negative correlation with age (r = −0.4,FDR = 1e-06).The hub compounds of M11 primarily consisted of sex hormone intermediates, including dehydroepiandrosterone sulfate (DHEA-S) (Figure 6c,d).These sex hormone intermediates were downregulated with aging in both males and females.Notably, the plasma level of DHEA-S positively correlated with the estrogen receptor ESR1 in adipose tissue (r = 0.24, p = 0.003) but not in muscle tissue (r = 0.005, p = 0.48).This suggests that estrogen signaling plays a crucial role in the deposition and metabolism of adipose tissue in both sexes (Cooke et al., 2017).Consistent with this, the expression of the estrogen receptor-encoding gene ESR1 decreased with age in adipose tissue, while the expression of the androgen receptorencoding gene AR remained relatively unchanged (Figure S8).

F I G R E 6
The network modules of plasma metabolites and their interactions with adipose and muscle genes.We identified three representative plasma compounds that serve as potential biomarkers of the aging process: enhanced tRNA modifications, accumulation of BCAA byproducts, and reduced sex hormone molecules.tRNAs, which are transcribed by RNA polymerase III and involved in protein biosynthesis, exhibit increased levels with aging and potentially indicate mitochondrial dysfunction (Dluzen et al., 2018).
Enhanced tRNA modifications might reflect aging-related changes in mitochondrial activity and cellular metabolism.BCAAs include leucine, isoleucine, and valine that can activate mTORC1 signaling (Wolfson et al., 2016).Dietary BCAA restriction promotes metabolic health and lifespan extension in mice (Richardson et al., 2021).Consistent with this, we observed repressed BCAA degradation in adipose tissue and accumulation of BCAA byproducts in plasma during aging.Furthermore, we found that aging is associated with reduced sex hormone compounds and downregulated expression of ESR1 in adipose tissue.Previous studies have shown that ESR1 deletion leads to increased adiposity, fibrosis, insulin resistance, and glucose intolerance in both male and female mice (Heine et al., 2000).Therefore, the decreased plasma levels of sex hormone compounds may reflect the progressive age-related decline in hormone production and action.
We also found that network modules of aging-associated cell types (immune cells, adipocytes, and muscle fiber cells) were enriched for BMI-associated genes, indicating that these cell types were regulated by obesity.The association of specific cell types in adipose and muscle tissues with obesity and diabetes was reported in recent studies.For example, deconvolution analysis of adipose tissues showed that the proportion of adipocytes and immune cells was correlated with BMI (Emont et al., 2022); single-cell experiment identified fibro-adipogenic progenitor cells associated with extracellular matrix remodeling in muscle tissues of diabetic patients (Farup et al., 2021); Population studies revealed that BCAA metabolism was associated with incidence of diabetes (Chai et al., 2022).Since aging-associated cell types are crucial for energy metabolism, they may play dual roles in the physiological changes of aging and obesity.
Targeting aging-vulnerable cell types, such as lifestyle modifications through the exercise program and calorie restriction, may provide therapeutic clues to alleviate aging-associated metabolic disorders (Pataky et al., 2021).Our study suggests that the aging process and a constellation of major chronic diseases, including obesity, diabetes, and cardiovascular diseases, likely share common metabolic, inflammatory, and epigenetic drivers.These conditions may be prevented potentially by modulating common targets with single interventions.
Although we conducted independent cohort validations to confirm aging-associated regulations, the issue of noise and the absence of epigenomic and metabolomic data in the GTEx data set impede a comprehensive validation.Notably, the AAGMEx cohort was specifically designed for metabolism and aging studies, whereas the GTEx datasets originated from postmortem human tissues sourced primarily from European ancestry individuals.This difference may have influenced the level of gene expression (Ferreira et al., 2018).
Consequently, while the GTEx cohort was larger, we observed less uniform and weaker aging associations compared to the AAGMEx cohort.This suggests that factors such as sample collection, sequencing, and analysis processes play a significant role.In our analysis of snRNA-seq data for cell type-specific gene regulations in adipose tissue, the donor size (n = 13) was relatively small for the pseudobulk analysis.To establish a more robust correlation, it is crucial to include more individuals from different aging stages.Additionally, while we used Spearman correlation to assess the strength and direction of monotonic relationships between variables, it is possible that complex nonlinear changes in aging regulations on metabolic rates and gene expressions may have been overlooked.Furthermore, the observed correlation between plasma metabolome data and transcriptome data from adipose and muscle tissues does not prove causality.

| Transcriptome profiling of adipose and muscle tissues
Transcriptome profiling in adipose and muscle tissues in the AAGMEx cohort was conducted according to previous publications (GEO ID #GSE95674 and #GSE95675 in super series #GSE95676) (Sajuthi et al., 2016;Sharma et al., 2016).Briefly, total RNA was extracted from adipose and muscle tissues using miRNeasy Mini Kit (Qiagen) and Ultraspec RNA total RNA extraction reagent (Biotecx laboratories), respectively.The extracted RNA was subjected to quality control using ultraviolet spectrophotometry (Nanodrop, Thermo

| DNA methylome profiling of adipose tissue
Epigenome-wide profiling of DNA methylation in adipose was conducted and described in the previous study (Sharma et al., 2020).
Genomic DNA was isolated from 100 mg frozen subcutaneous adipose biopsies using the Qiagen DNeasy tissue kit.Epigenome-wide profiling of DNA methylation levels was conducted via reduced representation bisulfite sequencing (RRBS) by Diagenode RRBS service (Diagenode, Liege, Belgium).RRBS libraries were prepared and sequenced on a HiSeq3000 using 50 base pair single-read sequencing to obtain at least 30 million reads/sample.The adapter-trimmed sequence reads were aligned to the Homo sapiens reference genome (GRCh37, hg19) using Bismark v0.20.0 (Krueger & Andrews, 2011).
MethylKit was utilized to filter and normalize the CpG data set (Akalin et al., 2012).We retained CpGs with at least 10X coverage, present in at least 75% of the samples, resulting in 1,073,614 CpGs.
As previous study indicated that the average methylation level of a methylation region is likely to be robust to measurement errors (Orozco et al., 2018), methylation levels for 205,566 CpG regions were extracted using MethylKit, with 1000-bp tiling windows across the genome, covering more than 10× in at least 75% of the samples.The R package "EnsDb.Hsapiens.v86"was used to extract 1 kb promoter regions of human genome and identify promoter-located methylation of CpG regions.

| Metabolome profiling of blood plasma
EDTA-plasma samples were obtained from the AAGMEx cohort following overnight fasting and stored at −80°C until analysis.Untargeted liquid chromatography-mass spectrometry was applied for metabolome profiling using the DiscoveryHD4 panel (Metabolon, Morrisville, NC).The resulting data was normalized by median-scaling, log-transformed, and missing values were imputed to the lowest measured value.Metabolites with missing data in more than 75% of the samples were excluded, leaving 1124 metabolites of good quality for downstream analyses.

| RNA-seq analysis of the GTEx data set
We obtained raw RNA-seq data from human adipose and muscle tissues sourced from the GTEx (v8) database.For each tissue, we performed data normalization using the trimmed mean of M-values (TMM) method to account for differences in sequencing library sizes (Robinson et al., 2010).The resulting normalized gene expression values were subjected to a log2 transformation.We also filtered out genes with low expression levels, defined as having log2 counts per million (CPM) value of ≤1 in over 75% of the total samples.We employed a linear model to examine correlations between gene expression and various traits (age, gender, BMI, and ethnicity), as well as covariates potentially contributing to batch effect and expression variation, including "SMCENTER" (collection sites), "SMRIN" (RNA integrity), "SMTSISCH" (ischemic time), "SMEXNCRT" (exonic rate), "SMRRNART" (rRNA rate), and "SMNTERRT" (intergenic rate).The beta coefficient and p-value associated with age were extracted from the linear regression model and used for comparison analysis with the AAGMEx cohort.

| MEGENA co-expression network analysis
Before network construction, we normalized the transcripts from adipose and muscle tissues and metabolites from blood plasma using a multivariate regression model to remove the covariate effects of data set can be downloaded through GitHub (https:// github.com/ pengu ab/ AAGMEx_ Aging_ Network).

R E FE R E N C E S
Akalin, A., Kormaksson, M., Li, S., Garrett-Bakelman, F. E., Figueroa, M. E., Melnick, A., & Mason, C. E. (2012).methylKit: A comprehensive R package for the analysis of genome-wide DNA methylation profiles.Genome Biology, 13(10), R87.https:// doi.org/ 10. 1186/ gb-2012-13-10-r87 Metabolic pathways of aging-associated network modules in adipose tissue.(a) KEGG pathway enrichment analysis of aging-related genes in adipose tissue of the AAGMEx cohort.The top 10 most significantly enriched pathways are presented for genes exhibiting positive (indicated by red) and negative (indicated by blue) age correlations.(b) MEGENA module enrichment of lysosome and BCAA pathway genes.The x-axis represents the correlation coefficient between module eigengene and age, whereas the y-axis indicates the significance of enrichment.(c, d) Network visualizations of module M3 for the lysosome pathway (c) and M14 for BCAA degradation (d).Nodes represent genes, and links indicate co-expression relationships.Node size correlates with network connectivity, and pink and blue colors represent positive and negative age correlations, respectively.The names of age-associated genes from each pathway are labeled above the nodes, with red color indicating network hub genes.
,b).MAP3K8, which controls the production of the pro-inflammatory cytokine TNF-alpha (TNF) during immune response, displayed a positive correlation with aging in macrophages 0.29 p:9.0e−06Adipose stem and progenitor cell (ASPC) r:0.20 p:2.5e−03Macrophage r:0.33 p:5.4e−07Age mRNA Level IRS1 r:−0.33 p:4.3e−07Age Methylation Level r:0.20 p:2.6e−03Chr2:U R E 3 Aging-associated cell-type proportion and DNA methylation profile changes in adipose tissue.(a) Proportions of adipocytes, ASPC, and macrophages across different ages in the adipose tissue of the AAGMEx cohort.Cell-type proportions were calculated using deconvolution analysis based on single-nuclei transcriptome profiles.The black line in the point plot represents the slope of the linear regression model.(b) Volcano plots depicting gene expression and methylation alterations of aging-associated marker genes in various adipose tissue cell types.Each dot represents an aging-associated marker gene for a specific cell type.The x-axis indicates the age correlation of gene expressions, whereas the y-axis shows the correlation significance.Red dots indicate marker genes that also exhibit aging-associated DNA methylation changes in the promoter region within a 1 kb window.(c) Aging-associated expression and DNA methylation levels of representative adipose marker genes.Each panel displays the aging correlation of gene expressions on the left dot plot and the aging correlation of methylations in the promoter region on the right dot plot.(d) Enrichment of methylated aging-associated genes in the adipose network modules.The correlation coefficients between module eigengenes and age are plotted on the x-axis, whereas the enrichment significance, represented by −log 10 (FDR), is shown on the y-axis.F I G U R E 4 Aging-associated genes and network modules in muscle tissue.(a) Venn diagram depicting aging-associated genes in muscle tissue from the AAGMEx and GTEx cohorts.(b) Point plots exhibiting the concordance of age correlation coefficients for aging-associated genes across both the AAGMEx and GTEx cohorts.The x-and y-axes represent coefficients calculated using Spearman correlation in each cohort.(c) Distribution of Spearman coefficients for aging-associated genes in both the AAGMEx and GTEx cohorts.(d) Sunburst plot visualizing hierarchy of MEGENA network modules enriched for aging-associated genes.The inner layers of the plot represent the parent modules of the outer layers.Color intensity indicates the enrichment significance, as measured by −log 10 (FDR).Positive values signify enrichments of aging-upregulated genes, whereas negative values represent aging-downregulated genes.Red asterisks highlight modules for downstream analysis.(e) Cell-type specificity of age-associated network modules in muscle tissue.The heatmap illustrates the enrichment of cell-type markers within the network modules.The bar plot at the top displays the correlation coefficients between module eigengenes and age.(f) Preservation of aging-associated modules in the AAGMEx and GTEx cohorts.The heatmap shows the significance of Fisher's exact test for the top 20 preserved AAGMEx modules.The bar plots indicate the correlation coefficients between module eigengenes and age.
0.56, p = 0.05).ITK, a crucial gene in immune cell module M3 within adipose tissue, regulates T cell development and differentiation and showed a positive correlation with age in T cells in our snRNA-seq pseudobulk analysis (r = 0.65, p = 0.02).These findings further support the role of these aging-associated genes in adipose tissue development across different cell types.
finding was corroborated by a parallel marker gene set from the sarcoendoplasmic reticulum calcium transport ATPase, including ATP2A2 (marker of type I fiber) and ATP2A1 (marker of type II fiber) F I G U R E 5 Aging-associated metabolic pathways and muscle fiber types in muscle tissue.(a) KEGG pathway enrichment analysis of aging-associated genes in muscle tissue of the AAGMEx cohort.The top 10 most significantly enriched pathways are presented for genes exhibiting positive (indicated by red) and negative (indicated by blue) age correlations.(b) MEGENA module enrichment for glycolysis and oxidative phosphorylation genes.The x-axis represents the correlation coefficient between module eigengene and age, whereas the y-axis indicates enrichment significance.(c) Proportions of slow and fast muscle fibers with age in the adipose tissue of the AAGMEx cohort.Celltype proportions were calculated using deconvolution analysis based on single-nuclei transcriptome profiles.The black line in the point plot represents the slope of the linear regression model.(d) Dot plots depicting age correlations with marker gene expressions in different muscle fiber types.MYH7 and ATP2A2 for slow-twitch type I fibers; MYH1 and ATP2A1 for fast-twitch type II fibers.Red and blue colors represent female and male samples, respectively.The black line indicates the slope of the linear regression model.
(a) Sunburst plot visualizing hierarchy of MEGENA network modules enriched for aging-associated metabolites.The inner layers of the plot represent the parent modules of the outer layers.Color intensity indicates the enrichment significance, as measured by −log 10 (FDR).Positive values signify enrichments of aging-upregulated genes, whereas negative values represent aging downregulated genes.(b) Functional annotations of network modules enriched for age-associated metabolites.(c) Network plots of top-ranked aging-associated modules.Nodes represent metabolites, and lines indicate co-regulation relationships.Hub metabolite names are labeled above the nodes.Node color (pink/blue) and metabolite names indicate positive/negative age correlations.(d) Examples of age-associated metabolites and their biosynthesis or pathway genes in adipose and muscle tissues.Dot plots depict correlations between age and metabolite/gene expression levels.Red and blue colors represent female and male samples, respectively.The black line represents the slope of the linear regression model.

Enrichment
In this study, we conducted a comprehensive network analysis of adipose and muscle tissues to investigate aging-associated regulations across different cell types and circulating metabolite biomarkers.Through network modeling, we revealed cell-type-specific modules during aging in adipose tissue, including reduced adipocyte activity and enhanced immune cell activity.By integrating multi-omic data, we revealed that aging has cell-type-specific effects on gene expressions, methylations, and cell proportions in adipose tissue.In muscle tissue, aging downregulated metabolic functions in oxidative phosphorylation, glycolysis, and gluconeogenesis.It also repressed gene expressions in fast-twitch type II muscle fibers while upregulating activities of slowtwitch type I muscle fibers.Furthermore, we observed correlations between plasma metabolites and corresponding gene expression levels in glucose-responsive tissues, indicating coordinated regulations of crosstissue metabolism during the aging process.Overall, our network analyses provided a comprehensive understanding of the cell-type-specific landscape involved in the aging process of adipose and muscle tissues.

Future
work will require further validations, such as immunofluorescence and transgenic experiments, to establish the causal relationship and delve deeper into the dynamic aging-associated expressions and functions of network hub genes in adipose and muscle tissues.In summary, our study offers a comprehensive overview of celltype-specific regulations and metabolic dysfunctions in glucoseresponsive tissues during aging.The findings emphasize the vulnerable cell types and aging-associated regulatory networks in adipose and muscle tissues, which may contribute to the development of age-related metabolic disorders.The connections between plasma metabolites and corresponding gene expressions in glucoseresponsive tissues suggest potential biomarkers and therapeutic targets for treating aging-related disorders.4 | ME THODS 4.1 | The African American Genetics of Metabolism and Expression (AAGMEx) cohort AAGMEx cohort consists of 256 unrelated, nondiabetic individuals who underwent multi-omic profiling of plasma, adipose, and muscle tissues (Sajuthi et al., 2016; Sharma et al., 2016).For | 13 of 16 XU et al. our analysis, we mainly focused on the transcriptome of 222 individuals with complete omics and phenotype data.The cohort consisted of healthy, self-reported African Americans residing in North Carolina, including 123 men and 99 women, with an age range of 18-61 years and a BMI between 18 and 42 kg/m 2 .Abdominal subcutaneous adipose tissue near the umbilicus and vastus lateralis skeletal muscle biopsies were collected from participants after an overnight fast by Bergstrom needle under local anesthesia.The tissues were immediately rinsed in sterile saline, quick-frozen in liquid nitrogen, and stored at −80°C.Fasting blood samples were also drawn for DNA isolation and biochemical analyses.All participants provided written informed consent under protocols approved by the institutional review boards at Wake Forest University School of Medicine.
Scientific) and electrophoresis (Experion nucleic acid analyzer, BioRad Laboratories, Inc.).Genome-wide expression data were generated using HumanHT-12 v4 Expression BeadChip (Illumina, San Diego, CA).The chips were scanned by the Illumina HiScan Reader using Illumina iScan Control Software, and probe-level expressions were extracted using Illumina GenomeStudio V2011.1.The expression levels were log2 transformed, robust multi-array average normalized (RMA, including quantile normalization), and batchcorrected using ComBat.Transcripts that were not significantly expressed (p-value < 0.05) in ≥25% of the samples and transcript probes encompassing common SNPs (based on ReAnnotator, or SnpInProbe annotation, and UCSC SNPv141) were excluded from downstream analyses.