Abstract Age-associated deterioration of physiological functions occur at heterogeneous rates across individual organs. A granular evaluation of systemic metabolic mediators of aging in a healthy human cohort (n = 225) identified prominent increases in circulating uremic toxins that were recapitulated in mice, on which we further characterized the aging phenome across five peripheral organs. Our multi-omics analyses connected systemic aging profiles primarily to kidney metabolism, uncovering a metabolic association between localized glucosylceramide (GluCer) accretion and renal functional decline. Elevated GluCers were also associated with higher risk of deaths in an independent cohort of aged individuals (n = 271). We report GluCer-mTOR signaling commencing at late middle-age that disrupts mitophagy and undermines mitochondrial respiration in kidney. Conserved between human and mice, GluCer-mediated renal dysfunction is female-biased and modulated by intracellular purines. Our work provides molecular basis for the sexually disparate effects of mTOR inhibition on mammalian lifespan, possibly ascribed to the evolutionary cost of female reproduction. 1. Introduction Medical advancements have significantly improved the quality of life and substantially extended the average human lifespan. Concurrently, aging-related diseases have emerged as major public health concerns. The United Nations projected that worldwide, the number of persons aged 65 years or above will reach 1.5 billion in 2050 [[48]1]. In China, the percentage increase in older persons aged 65 years or over between 2019 and 2050 is estimated at 19.9 % [[49]1], which greatly elevates the risk of various non-infectious, age-dependent chronic diseases such as cardiovascular disease (CVD), chronic obstructive pulmonary disease, chronic kidney disease (CKD), dementia and diabetes [[50]2]. Aging represents the greatest common risk factor for chronic diseases [[51]3], while age-related functional decline of different organs denote commonalities underlying various chronic diseases. The trajectory of age-related deterioration in function varies for individual organs, and biological aging clocks specific to distinct organ systems have been established [[52]4]. Chronic diseases impact the overall senescence of body systems in disparate ways. For instance, patients with CKD were found to exhibit the oldest body ages amongst 16 categories of chronic diseases [[53]4]. A multi-organ characterization of aging phenomes coupled with the identification of systemic metabolic mediators of aging is, therefore, expected to unravel inter-organ crosstalk that coordinates the overall aging process of the organism as an entity. Lipids, as integral structural components of mammalian systems, partake in various biochemical signaling processes that underpin cellular and tissue function [[54]5]. Preceding studies have implicated lipids in the pathogenesis of various age-related diseases such as CVD [[55]6,[56]7], diabetes [[57]8,[58]9], Alzheimer's disease (AD) [[59][10], [60][11], [61][12]] and CKD [[62]13,[63]14]. As examples, changes in ether lipids [[64]10], including plasmalogen lipids [[65]11], and ceramides (Cer) [[66]10,[67]12] have been reported in AD. Peripheral signatures from two large clinical studies of AD have identified ether lipids and sphingolipids to be associated with prevalent AD [[68]10]. On the other hand, lower levels of lysophosphatidylcholines (LPCs) were associated with renal failure in CKD [[69]13], while changes in phosphatidylcholines and triacylglycerols were correlated with higher CKD risk [[70]14]. On top of lipids, polar metabolites constitute an added dimension of the biological phenome that together confer the closest readout of the cellular phenotype. Metabolomic profiling of aging cohorts has identified several metabolites from, for examples, the categories of carbohydrates, amino acids and nucleotides, which were associated with the aging phenotype [[71]15,[72]16]. Omics-driven approaches are useful in offering a granular view of the metabolic landscape that facilitates the discovery of novel functional metabolites under different biological context [[73]17]. Analytical challenges such as limited metabolite identification and imprecise quantification, however, can substantially impede the broader application of metabolomics in aging research and undermine the validity of age-related metabolite markers [[74]18]. In addition, age-associated metabolic perturbations in humans might not be fully recapitulated in mouse models [[75]19], rendering a combinatorial approach necessary in deciphering age-related metabolic interventions to ameliorate pathological aging in humans. In this study, we first utilized precise, quantitative metabolomic approaches [[76]18] to establish an array of systemic, age-related metabolite alterations in the human plasma of a cross-sectional cohort. These systemic metabolic mediators were recapitulated during the aging process in mice, which enabled us to leverage on the murine model to uncover the contributory roles of different organ systems towards the systemic metabolic perturbations observed in human aging. Using high-coverage, quantitative lipidomic profiling [[77]20] of metabolic changes in five major peripheral organs and/or tissues in mice, our findings untangle metabolic signaling pathways crucial to preserving kidney function, which contribute positively to human health and unveil new intervention targets to promote healthy aging. 2. Results 2.1. Overlapping metabolic features between human systemic aging and compromised renal function Using a robust, non-targeted strategy for accurate quantitation and precise profiling of 480 metabolites developed in-house [[78]18], we investigated the aging-associated plasma metabolomes in a cross-sectional cohort of 225 ostensibly healthy individuals aged between 20 and 88 years ([79]Table 1). Linear regression analysis uncovered an array of 280 plasma metabolites associated with aging. Over 70 % (200/280) of these age-associated metabolites were positively correlated with increasing age, which predominantly consisted of acylcarnitines, long-chain fatty acids and dipeptides ([80]Fig. S1A). The age-associated accretions in plasma acylcarnitines, which were inversely associated with estimated glomerular filtration rate (eGFR) [[81]21], indicates declined renal function with aging. The kidney plays a key role in carnitine biosynthesis from lysine and methionine, and in carnitine excretion into the urine and plasma [[82]22]. Pathway enrichment analysis of age-related metabolites uncovered several pathways related to urea metabolism, including “urea cycle and metabolism of amino groups”, “biomarkers for urea disorders” and “urea cycle and associated pathways” ([83]Fig. S1B). The involvement of urea metabolism suggests perturbations in hepatic and/or renal function with aging [[84]23]. A gross comparison between the age-associated plasma metabolite changes reported here with preceding metabolome perturbations in CKD revealed significant overlap, such as aberrant levels of short and medium-chain acylcarnitines [[85]14,[86][24], [87][25], [88][26]], gut microbiota metabolites [[89]26], tryptophan metabolites [[90]27], purine metabolites [[91]28,[92]29], acetylated amino acids [[93][30], [94][31], [95][32]], bile acids [[96]26], LPCs [[97]26,[98]33] and lysophosphatidylethanolamines (LPEs) [[99]26,[100]34] ([101]Table S1). Table 1. Clinical characteristics of normative aging cohort. Characteristic Statistics, N = 225 Sex (female), n (%) 114 (50.7 %) Age (years) Mean (SD) 52.56(18.00) Min-Max 20.00–88.00 Median(Q1,Q3) 52.00(38.00,68.00) BMI (kg/m2) Mean (SD) 23.73(3.64) Min-Max 14.61–36.41 Median(Q1,Q3) 23.89(21.35,25.77) Unknown 57 (25.3 %) History of hypertension, yes (%) 39 (17.3 %) Unknown 55 (24.4 %) History of coronary heart disease, yes (%) 13 (5.8 %) Unknown 55 (24.4 %) History of diabetes, yes (%) 12 (5.3 %) Unknown 55 (24.4 %) History of dyslipidemia, yes (%) 39 (17.3 %) Unknown 55 (24.4 %) History of myocardial infarction, yes (%) 0 (0 %) Unknown 55 (24.4 %) History of drinking alcohol, n (%) 27 (12 %) Unknown 55 (24.4 %) SBP (mmHg) Mean (SD) 123.03(16.70) Min-Max 82.00–173.00 Median(Q1,Q3) 120.00(111.00,133.50) Unknown 58 (25.8 %) DBP (mmHg) Mean (SD) 72.87(11.61) Min-Max 3.00–102.00 Median(Q1,Q3) 72.00(65.00,78.75) Unknown 59 (26.2 %) TG (mmol/L) Mean (SD) 1.25(0.83) Min-Max 0.24–6.48 Median(Q1,Q3) 1.07(0.79,1.45) Unknown 57 (25.3 %) TC (mmol/L) Mean (SD) 4.73(0.84) Min-Max 2.88–7.11 Median(Q1,Q3) 4.65(4.10,5.26) Unknown 57 (25.3 %) HDL (mmol/L) Mean (SD) 1.41(0.35) Min-Max 0.78–2.70 Median(Q1,Q3) 1.38(1.17,1.61) Unknown 57 (25.3 %) LDL (mmol/L) Mean (SD) 2.81(0.73) Min-Max 1.27–5.08 Median(Q1,Q3) 2.70(2.27,3.29) Unknown 57 (25.3 %) CREA (umol/L) Mean (SD) 64.85(14.72) Min-Max 35.30–116.00 Median(Q1,Q3) 64.15(54.50,72.35) Unknown 53 (23.6 %) BUN (mmol/L) Mean (SD) 4.93(1.36) Min-Max 2.40–9.50 Median(Q1,Q3) 4.70(3.90,5.83) Unknown 53 (23.6 %) UA (umol/L) Mean (SD) 323.23(76.17) Min-Max 142.20–573.90 Median(Q1,Q3) 321.00(269.53 378.23) Unknown 53 (23.6 %) RI (mU/L) Mean (SD) 7.68(6.84) Min-Max 1.90–81.50 Median(Q1,Q3) 6.80(5.20,8.60) Unknown 80 (35.6 %) GLU (mmol/L) Mean (SD) 5.32(0.73) Min-Max 3.92–8.60 Median(Q1,Q3) 5.16(4.88,5.57) Unknown 57 (25.3 %) [102]Open in a new tab BMI: body mass index, SBP: systolic blood pressure, DBP: diastolic blood pressure, Hcy: homocysteine, HSCRP: hypersensitive-c-reactive-protein, TG: triglycerides, TC: total cholesterol, HDL: high-density lipoprotein, LDL: low-density lipoprotein, CREA: creatinine, BUN: urea nitrogen, UA: uric acid, RI: regular insulin, GLU: glucose. Standard linear regression analyses might overlook undulating metabolite alterations across aging. To overcome this, we next conducted unsupervised hierarchical clustering to group age-associated plasma metabolites that possessed similar trajectories of changes. Nine metabolite trajectories were obtained ([103]Table S2, [104]Figs. S1C–D), demonstrating the non-linear nature of several age-associated metabolite alterations. Classification of metabolite trajectory allows a clear visualization of temporal fluctuations in metabolite levels with age ([105]Figs. S1D–E), and metabolites with common trajectory (i.e. within the same cluster) possibly indicate co-regulation across aging. We defined the metabolic nature of individual clusters based on the dominant metabolite class. Metabolites relevant to renal function were distributed amongst clusters 2–4, which predominantly comprised lipids, uremic toxins, N-acetyl-amino acids (N-acetyl-AAs) and acylcarnitines. Clusters 2 and 3 contained metabolites downstream of tryptophan and tyrosine metabolism, such as kynurenic acid, l-kynurenine and P-cresol-sulfate, which denote uremic toxins normally excreted by the kidneys. Uremic toxins accumulate under impaired kidney function and can inflict damages on multiple organs [[106]29]. Cluster 3 also included several N-acetyl-AAs generated from catabolism of N-acetylated proteins. Under normal circumstances, N-acetyl-AAs are deacetylated and reabsorbed in the kidneys via amino acid salvage pathway, and their accumulation illustrates compromised renal salvage function [[107]35]. A simple overview of age-related metabolic alterations thus underscores dysregulated renal function as a key aspect of systemic aging in humans. 2.2. DE-SWAN analysis identifies bursts in differential metabolites across aging Since metabolic reactions are intertwined and often buffered by compensatory mechanisms, aging-associated metabolic phenotypes might be masked until resiliency mechanisms fall apart. Instead of a gradual, linear process, we therefore expect systemic aging to occur in waves from a metabolic perspective. Utilizing differential expression-sliding window analysis (DE-SWAN) [[108]36] designed to select quantitative changes in phenotype throughout life, we uncovered three metabolic crests at ages 25, 56, and 72, which corresponded approximately to life stages of young, middle-aged, and old-aged in human ([109]Fig. 1A). DE-SWAN algorithm analyzes metabolite changes in sliding windows (in increments of 1 year) of 20 years, and compares two groups in parcels of 10 years (e.g., 35–45 years compared to 45–55 years) throughout all ages examined [[110]36]. DE-SWAN analysis unmasked the sequential effects of aging on the systemic metabolome and revealed several metabolite clusters that were altered in waves across aging ([111]Fig. 1B). As anticipated, DE-SWAN analysis additionally identified 18, 20, and 25 metabolites specifically altered at ages 25, 56 and 72, respectively, which were not revealed by linear regression analysis ([112]Fig. 1C). To obtain metabolic representation of these age-related crests, we conducted pathway enrichment analysis based on metabolites within individual clusters ([113]Fig. 1D–[114]Table S3). DE-SWAN analysis detected temporal dysregulation in metabolic pathways otherwise masked in linear modeling. For example, linear regression analysis indicated a general downregulation in “nicotinamide salvaging” and “nicotinamide metabolism” across aging, but DE-SWAN revealed that these pathways were upregulated at 25 years then downregulated at 56 years instead ([115]Fig. 1D). Enhanced nicotinamide salving in young individuals and its subsequent decline at middle age are in agreement with the reported effect of increasing NAD + bioavailability on improving healthspan [[116]37]. Importantly, pathways concerning the regulation of organic and/or inorganic ion and amino acid transport mediated by solute carrier proteins (SLC), such as “SLC-mediated transmembrane transport” and “transport of inorganic cations/anions and amino acids/oligopeptides”, were downregulated with aging, particularly at 72 years ([117]Fig. 1D). Enrichment of these pathways was primarily ascribed to reductions in plasma isoleucine and leucine levels with aging. Indeed, appreciable reduction in plasma isoleucine was previously observed in patients with acute kidney injuries, attributed to abated activity of SLC6a19 neutral amino acid uniporter in proximal renal tubular cells - an early cellular response to kidney injuries resulting from ischemia [[118]38]. As one of the most energy-demanding organs in the human body [[119]39], the kidneys consume substantial amounts of ATPs in executing waste removal and nutrient reabsorption from blood, and in maintaining systemic electrolyte and fluid balance, which makes the kidneys particularly vulnerable to metabolic disturbances associated with aging. Indeed, amongst the top ten metabolites ranked by statistical significance in the old-aged cluster at 72 years, seven metabolites (i.e. tryptophan 2-C-mannoside, N-acetyl-l-alanine, pseudouridine, sn2 LysoPE(22:6), l-cystine, N-acetyl-l-aspartic acid, symmetric dimethylarginine) ([120]Table S3) were reported to significantly correlate with eGFR [[121]34]. To further highlight the decline in renal function within individual age-associated metabolic crests defined by DE-SWAN, we manually selected ten metabolites implicated in CKD pathogenesis based on published literature [[122]26,[123]34,[124]40], which included symmetric dimethylarginine, l-kynurenine, citrulline, tryptophan 2-C-mannoside, succinyladenosine, pseudouridine, p-cresol glucuronide, p-cresol sulfate, creatinine and uric acid. The number of CKD-associated metabolites significantly altered in each of the three metabolic crests was compared, and the number of fluctuating CKD-associated metabolites was evidently highest at old-age (72 years), but statistical significance emerged as early as middle-age (56 years) ([125]Fig. 1E and F). Fig. 1. [126]Fig. 1 [127]Open in a new tab (A) Schematic diagram illustrating the study design. Baseline blood samples were collected from CKD patients. Plasma and RBCs were isolated from the whole blood samples from 45 to 117 patients, respectively. The clinical follow-up period lasted for approximately 7 years ≈ 86 months, during which 11/45 patients in the plasma cohort and 37/117 patients in the RBC cohort recorded newly onset cardio-cerebrovascular events. RBCs from a subset of 40 patients (20 events) were also subjected to proteomics analysis. (B) Quantitative lipidomics was carried out to analyze the whole-lipidome of plasma (red rim) and polar lipidome of RBCs (blue rim). Lipid species were classified according to major lipid classes defined on the circumference, and the number of species detected in each lipid class was summarized on the upper left corner. The x-axis illustrates increasing double bond number (starting from left to right) while the y-axis denotes increasing carbon atom number (radiating from the interior to exterior region). The number of lipids for each specific combination of carbon atom numbers and double bond numbers was defined by the color of each dot. RBC: red blood cells; PC: phosphatidylcholines; PE: phosphatidylethanolamines; PG: phosphatidylglycerols; PI: phosphatidylinositols; PS: phosphatidylserines; SM: sphingomyelins; TG: triacylglycerols; CAR: acylcarnitines; CE: cholesteryl esters; Cer: ceramides; DG: diacylglycerols; FA: free fatty acids; Gb3: globotriaosylceramide; GlcCer: glucosylceramides; GM3: monosialo-dihexosyl gangliosides; LacCer: lactosylceramides; LPA: lyso-PA; LPC: lyso-PC; LPE: lyso-PE; LPI: lyso-PI; LPS: lyso-PS; PA: phosphatidic acids. (For interpretation of the references to color in this figure legend, the reader is