Biological age, not chronological age, captures one’s physical and functional ability and is a determinant of healthspan (years lived in good health). We took an integrated approach encompassing high-resolution metabolomics combined with a panel of SASP and proinflammatory markers in the serum, to define a molecular index for biological aging. Walking ability, reflects an integrated assessment of cardiovascular fitness, muscle strength, and neurological and joint function and is currently the single best predictor in humans of hospitalization, functional decline, disability, surgical complications, institutionalization and death24–27. The ‘SOLVE-IT’ cohort consisted of 196 total participants, 98 individuals above 75 years old that showed good walking ability (walk up a flight of stairs and walk for 15 minutes without resting). These individuals were physically active and hence classified as “healthy” agers. The remaining 98 were classified as “rapid” agers as they displayed poor walking ability despite being chronologically younger than the healthy agers28 (Fig. 1a). Additionally, multiple measures such as gait, function, mental status, strength, activity and comorbidity index were included in our study. Phenotypic age is an effective predictor of overall health risks and it is strongly associated with the chronological age29. However, one of the unique features of the ‘SOLVE-IT’ study is that the biologically-aged individuals are readily distinguishable from chronologically-aged individuals. In this study cohort, frailty, comorbidities, impaired cognitive ability (defined by poor Montreal Cognitive Assessment scores) and higher body mass index (BMI) are negatively associated with chronological age and more prevalent in rapid agers. Such an inverse association of biological age and chronological age, as demonstrated in this cohort, offers an advantage in delineating specific signatures of healthy aging (Fig. 1b). It is to be noted that incidence of declining organ functions such as heart, kidney, and liver failures, as well as, cancer were alike in both rapid and healthy agers (Fig. 1b), thereby strengthening the unique appropriateness of the cohort for identification of markers associated with functional aging.
To define the molecular fingerprint associated with biological aging, we performed high-resolution metabolomics by Ultrahigh Performance Liquid Chromatography-Tandem Mass Spectroscopy (UPLC-MS/MS). We used serum samples for our study since it is minimally invasive, affordable, and has reduced overall risk for patients. In addition, several recent studies with heterochronic parabiosis, as well as blood/serum transfusions in animal models suggest that systemic circulating factors in blood, drives the aging phenotype30. A total of 1327 serum metabolites were identified that belonged to 9 different super pathways as defined by KEGG analysis. Majority of the identified metabolites were lipids (32%), followed by xenobiotics (17%) and amino acids (16%). The identified metabolite super pathways were further summarized into sub pathways. Long and medium-chain acylcarnitines (40), fatty acid dicarboxylate (31), sphingomyelins (29), diacylglycerols (29) and lysophospholipids (25) were predominant groups among lipids. Leucine, Isoleucine and Valine Metabolism (33), Arginine and Proline Metabolism (23), Tryptophan Metabolism (23) and Methionine, Cysteine, SAM, and Taurine Metabolism (22) were some of the sub pathway profiles for amino acid metabolism that were identified (Fig. 1c). The inter-relationship between the identified metabolites revealed 1481 highly correlated (Spearman’s ρ > 0.8) ‘metabolite pairs’. Over 200,000 significant correlations were observed among metabolite pairs, with 33,763 lipid-lipid pairs and 23,761 lipid-amino acid pairs. Interestingly, choline derivatives of long and very long-chain free fatty acids (16–22 carbons) displayed remarkably high correlations. The identified fatty acylcholines (36) had a minimum ρ > 0.606; for example, palmitoylcholine (16:0) and steroylcholine (18:0) showed an exceptionally high correlation ρ = 0.94 (Fig. 1d), suggesting a controlled acylcholine synthesis irrespective of heterogenous traits. Overall, our metabolomic analysis from this cohort generated a rich set of metabolite data that included several metabolic pathways.
Differential metabolome pattern associated with biological age.
We used Orthogonal partial least square-discriminant analysis (OPLS-DA), a multivariate supervised classification method with 7-fold cross validation consisting of 200 iterations in each round to distinguish metabolic differences between healthy and rapid agers. OPLS-DA analysis produced a model with R2 (cumulative) = 0.76, Q2 (cumulative) = 0.40 with a predictive power of 95.9%, fisher p-value = 1.45 × 10− 45 and a root mean square cross validation error RMSE = 0.386. The model separated healthy and rapid agers demonstrating a clear difference in the metabolome associated with biological age (Fig. 2a). To identify metabolites that are associated with healthy and rapid agers, we used receiver operative characteristic (ROC) analysis based on a logistic regression model. We noted that higher levels of some of the metabolites were predictive of healthy agers, Area Under the Curve (AUC) value for healthy aging ROC curve, ROC_AUCHA>0.5) and some were predictive of rapid agers (AUC value for rapid aging ROC curve, ROC_AUCRA>0.5). Therefore, we combined these ROC_AUC scores to create a single variable AUCcomb (see methods). Metabolites with AUC values more than 0.5 were considered indicators of healthy agers while, less than 0.5 were associated with rapid agers. There were (331) metabolites that significantly distinguished healthy and rapid agers, with (125) metabolites as predictors of healthy aging and (206) as predictors of rapid aging (q-value < 0.05) (Fig. 2b). Eicosenoylcarnitine (C20:1), an acylcarnitine was one of the most influential metabolites in discriminating healthy agers from rapid agers (AUCcomb=0.72). Beside acylcarnitines, healthy agers were also distinguished by beta-cryptoxanthin (AUCcomb=0.70), a precursor of vitamin A, important for general growth, development and immune response. Gut microbiome-metabolite, 1H-indole-7-acetic acid (AUCcomb=0.70) was also elevated in healthy agers. In contrast, dicarboxylic fatty acids (DCA) such as pimelate (C7) (AUCcomb=0.30), suberate (C8), sebacate (C10) (AUCcomb=0.33) and undecanedioate (C11) (AUCcomb=0.31) were elevated in rapid agers. Similarly, glutamate and mannose also elevated in rapid agers (AUCcomb=0.33) (Fig. 2b, source data).
To understand biological pathways associated with either healthy or rapid agers, we analyzed metabolites that significantly separated the two groups at p-value < 0.05, using Ingenuity Pathway Analysis (IPA). We found that Hypoxia-inducible factor 1-alpha (HIF1α) signaling, 4-hydroxyproline degradation, adenosine nucleotides degradation, and citrulline metabolism were mainly associated with accelerated biological age (Fig. 2c). Consistent with our results, other studies have shown that upregulated HIF1α signaling impairs mitochondrial biogenesis and accelerates aging. Sirtuins are the main class of enzymes that destabilize HIF1α thereby promoting mitochondrial health during aging31. Similar to these reports, healthy agers positively associated with sirtuin signaling pathways, generally linked to longevity (supplementary Fig. 1). Choline degradation, carnitine metabolism, gamma-glutamyl cycle were some of the other pathways associated with healthy agers (supplementary Fig. 1). In addition to general metabolic pathways, IPA was used to provide disease and biofunction predictions. Disease/injury associated metabolites were related to gastrointestinal disease, skeletal and muscular disorders, organismal injury and neurological diseases (Fig. 2D). This data suggests that certain co-morbidities may exhibit metabolic profiles which may indicate underlying conditions, although they were not clinically observed in rapid agers at the time of sample collection. Collectively, these results show that metabolites and associated pathways can differentiate healthy and rapid biological agers.
Balance of fatty acid oxidation pathways predicts healthy agers.
Next, we sought to identify specific metabolite signatures that can serve as potential indicators for healthy aging. Acylcarnitines, especially long chain acylcarnitines and dicarboxylic acids (DCAs) were the two important classes that were identified both in the OPLS-DA and ROC analysis (Fig. 3a). Acylcarnitines play a major role in regulating lipid metabolism by shuttling fatty acids into mitochondria. Under physiological conditions, oxidation of long- and medium-chain fatty acids is primarily carried out by mitochondrial β-oxidation (Fig. 3b). We observed 9 acylcarnitines, mostly long chain forms with ROC_AUCHA values 0.6–0.72 (Fig. 3a). An alternate, subsidiary pathway to β-oxidation is ω-oxidation occurring in endoplasmic reticulum (ER) microsomes32. ω-oxidation of fatty acids generates dicarboxylic fatty acids (DCAs) and is an alternate pathway used when mitochondrial β-oxidation is impaired (Fig. 3b). Our analysis detected 3 important DCAs: pimelate, undecanediote and suberate with ROC_AUCRA values 0.6–0.7. In healthy agers, the acylcarnitines levels were higher than in rapid agers suggesting that β-oxidation is predominantly active in the former. On the other hand, rapid agers showed higher levels of DCAs compared to healthy agers indicating increased ω-oxidation (Fig. 3a). The predictive power obtained from the difference in acylcarnitines and DCAs improved by 0.06 compared to the best individual predictor (eicosenoylcarnitine) (Fig. 3c, d). These results suggest a balance between β-oxidation and ω-oxidation pathways can potentially influence biological age.
Identification of Healthy Aging Metabolic (HAM) index
Aging is a complex process that cannot be realized through one metabolic pathway or a metabolite class. Indeed, several reports have implicated the roles of multiple pathways affecting the aging process. For example, dysregulation of the carnitine shuttle and vitamin E pathways have been associated with frailty33 whereas, tryptophan metabolism, particularly, kynurenine pathway is implicated in age-related chronic inflammation and memory impairment34. Therefore, we hypothesized that a combination of metabolites from different pathways could possibly be a better molecular indicator for biological age, rather than a single metabolite/pathway. The combinatorial approach can overcome the moderate predictive power presented by the individual metabolites. In order to identify a panel of metabolites that are better predictors of healthy biological aging we chose all known endogenous metabolites from OPLS-DA analysis with Variable Importance of the Projection (VIP) score > 1 and fit a linear regression model. We used LASSO regression method with 10-fold cross validation with 1000 bootstrapping steps in each validation. LASSO regression model eliminates the collinear variables and retains only the significant variables (p < 0.05). The final model retained a panel of 18 metabolites with a Pearson’s r = 0.74 and p < 0.0001 (Fig. 3e). The model consisted of metabolites primarily related to fatty acid metabolism, the TCA cycle, amino acids and glutathione metabolism and strongly predicted healthy biological agers. Importantly, based on the model values we derived a healthy aging indicator “Healthy Aging Metabolic (HAM) Index”. The HAM index was significantly different between the healthy agers and rapid agers (p < 0.0001) and showed a ROC_AUCHA value of 0.95 in identifying healthy agers (Fig. 3f, g). Compared to some of the biological age indices such frailty index, gait speed, MOCA score (supplementary Fig. 2), the HAM index outperformed these indices in predicting healthy agers from rapid agers (Fig. 3h). This predictive power from a combination of metabolites points out that several different pathways are involved in maintaining a healthy biological age.
SASP markers associated with rapid agers.
Circulating factors such as senescence-associated secretory phenotype (SASP) and pro-inflammatory markers can reflect the state of aging cells. Senescence-associated ‘secretome’ could be a valuable marker for aging and age-associated diseases. With this in mind, we wanted to investigate age-associated proinflammatory markers in the context of biological age. To test this we measured multi-analyte SASP markers in serum using Luminex High Performance Assay. The list of SASP and proinflammatory markers examined were based on evidence from several studies (supplementary table 1). Interestingly, Cystatin C and CCL-2/MCP-1 were found to be elevated in rapid agers compared to healthy agers (Fig. 4a). Serum Cystatin C belongs to the family of Cystatin protease inhibitors and consistent with our analysis, has been shown to increase significantly with age, even in the absence of clinical risk factors for renal dysfunction35. Similarly, monocyte chemoattractant protein 1 (MCP1) is a key chemokine that is important for infiltration of macrophages and monocytes. These data suggest that a subset of SASP factors track with increased biological age independent of chronological age.
Next, we examined the metabolomic profiles associated with CCL-2/MCP-1 and Cystatin C. As shown in Fig. 4c and source data, CCL-2/MCP-1 was associated with 57 metabolites (FDR-correlated p < 0.2) with majority (56%) contributed by lipids, particularly, lysophospholipids (11), diacylglycerols (6) and phosphatidyl glycerol (6). In contrast to CCL-2/MCP-1, Cystatin C showed positive association with metabolites that belonged to both lipids (175) and amino acids (150) (Fig. 4d, source data). A panel of serum metabolites that displayed remarkable correlation with Cystatin C is shown in Fig. 4e. Interestingly, the levels of CCL-2/MCP-1 and Cystatin C levels were random among the age groups and did not influence one another as suggested from Spearman’s ρ with a p-value > 0.05.
In addition to Cystatin C and CCL-2/MCP-1, C-reactive protein (CRP) and interleukin-6 (IL-6) were also elevated in rapid agers (Fig. 4b). Tryptophan (4), Tyrosine (4) metabolism, fibrinogen cleavage peptide (7), vitamin E metabolism, androgenic steroids (7) showed positive association with CRP whereas ceramides (5) were negatively associated with CRP. Cleaved fibrinogen products, as well as, higher levels of Tryptophan metabolism products such as kynurenine and indole-3-carboxylates suggest chronic inflammation in rapid agers (supplementary table 2). Metabolites associated with IL-6, particularly DSGEGDFXAEGGGVR, Fibrinopeptide B (1–13), ADSGEGDFXAEGGGVR and Fibrinopeptide A (3–16) also support the prevalence of a low-grade inflammation with accelerated biological age36.
In order to understand the relationship between metabolites/metabolic pathways and circulating secretory factors in aging, we looked for common metabolites that were associated with secretory factors and biological age groups. Five classes of metabolites- acylcarnitine, oleoyl/linoleoyl glycerol phosphocholine, carotene diol, γ-glutamyl glutamine and nicotinamide that were strongly associated with healthy aging were found to be negatively associated with Cystatin C, MCP-1, CRP, as well as, IL-6 (supplementary Fig. 3). Similarly, dicarboxylic acids (DCAs), key products of ω-fatty oxidation that were strongly associated with rapid agers were positively associated with MCP-1 and IL-6 (supplementary Fig. 4). An enrichment of correlated metabolites with these secretory factors suggest a synergistic interaction between senescence-associated secretome, metabolism and biological aging.
Metabolites display sexual dimorphism independent of biological age.
Several factors can influence biological age, one such is gender. It is well known that life expectancies for women are usually higher than men37,38. Here we probed the metabolome of our study cohort to understand the implications of sexual dimorphism in biological aging. Consistent with the previous reports39, there was a clear metabolomic difference between the males and females in the SOLVE-IT cohort (Fig. 5a). Among females, sphingomyelins were significantly increased compared to males. Similarly, the levels of one of the major endocannabinoids, arachidanoyl glycerol and its precursor stearoyl arachidanoyl glycerol were higher in females (Fig. 5b). On the other hand, as expected in males, the male hormone, androgen-derived metabolites such as androstenediol disulfate (1), androstenediol monosulfate, androstenediol disulfate (2) were found to be elevated. Examining the SASP and proinflammatory factors, it was clear that matrix metalloproteinase (MMP-1) and plasminogen activator inhibitor-1 (PAI-1) were significantly increased in females compared to males. Taking our previous results into consideration (Fig. 4a), we observed that irrespective of the gender, both CCL-2/MCP-1, and Cystatin C were unaltered suggesting biological aging-associated senescence may not be influenced by gender. Interestingly, we did not see any significant differences among the proinflammatory markers in both the sexes (Fig. 5d).
Next, we sought to understand the impact of gender in biological age-associated metabolic signatures (Fig. 5e). For this, we compared the ROC curves of males versus females in predicting healthy and rapid agers. This enables a direct comparison between the sexes by factoring age groups but without compromising the statistical power of the analysis. The analysis identified four groups of metabolites: Cluster I- metabolites that were elevated in male rapid agers, but lower in female rapid agers (Male AUCcomb<0.5 and female AUCcomb> 0.5, upper left quadrant); Cluster II- metabolites elevated in healthy agers of both the sexes (Male AUCcomb > 0.5 and Female AUCcomb >0.5, upper right quadrant); Cluster III- metabolites elevated in male healthy and female rapid agers (Male AUCcomb>0.5 and female AUCcomb<0.5, lower right quadrant) and Cluster IV- metabolites elevated in rapid agers of both the sexes (Male AUCcomb < 0.5 and Female AUCcomb<0.5, lower left quadrant). A total of 113 metabolites were identified, with 57 metabolites mapping to Cluster 1 (male rapid agers, linked to female healthy agers). Six acetylated metabolites namely, N2-acetyl, N6-methyllysine, N2-acetyl, N6, N6-dimethyllysine, N-acetylcitrulline, N-acetyl phenylalanine, N-acetylarginine, N-acetyl-3-methylhistidine belonged to Cluster I. Likewise, very-long-chain (C > 22) acylcarnitines such as nervonoylcarnitine, cerotoylcarnitine, behenoylcarnitine, ximenoylcarnitine correlated remarkably well with female healthy agers but not with male healthy agers. However, long chain acylcarnitines (18 ≤ C ≤ 22) were associated with both male and female healthy agers. This suggests that even though acylcarnitines are universal markers of aging33, gender can have a significant impact. Similarly, some male healthy aging-associated metabolites (24) were identified with female rapid agers. For example, oxidized methionine, methionine sulfone showed the highest difference in the AUCcomb values with 0.70 in males and 0.39 in females. A few amino acid metabolites such as ornithine, 5-(galactosylhydroxy)-L-lysine, C-glycosyltryptophan, 3-methyl glutaryl carnitine, cystathionine, dimethylguanidino valeric acid (DMGV), hydantoin-5-propionate and hydroxyasparagine were also identified with healthy agers in males but not with females.
We identified 16 metabolites that were associated with healthy agers in both males and females but with significant differences in their power of association (Cluster II). The top metabolites in this group were long chain acylcarnitines like octadecanedioylcarnitine, octadecenedioylcarnitine, which are strong predictors of male healthy agers but weak predictors of female healthy agers. Similarly, gamma-glutamylcitrulline and S-methyl methionine could predict the male healthy agers with an AUCcomb value of 0.80 but this value was reduced to 0.61 and 0.63 respectively, in female healthy agers. On the other hand, levels of two major glycolytic metabolites; glucose and pyruvate were strong predictors of male rapid agers but their predictive power was significantly lower for female rapid agers (Cluster IV) (Fig. 5e, source data). Overall, these results suggest a decisive role of sexual dimorphism in the metabolism associated with aging. Of note, none of the biomarkers from the healthy aging metabolomic (HAM) index was affected by gender differences, pointing to the robustness of the HAM index in predicting healthy agers.
Metabolites associated with race.
Another demographic feature that can influence aging is race. There is strong evidence supporting racial differences in health and life expectancies40. We wanted to understand the impact of race on biological aging. So, we analyzed the metabolome of African Americans and Caucasians in the ‘SOLVE-IT’ cohort (supplementary Fig. 5). OPLS-DA model showed a marked distinction in the metabolomic profiles of African Americans and Caucasians (Fig. 6a). Top predictive metabolites largely belonged to lipids. Hydroxyproline, N6-methyl lysine and xenobiotics sulfate of piperine metabolite (2), sulfate of piperine metabolite (3) were also among top metabolite predictors (Fig. 6b). Plasmalogens containing poly-unsaturated fatty acids (PUFAs)- arachidonic acid and linoleic acids, as well as, lysoplasmalogens were significantly elevated in the African Americans (Fig. 6c). Plasmalogens are unique glycerophospholipids with a vinyl ether moiety at the sn-1-position of the glycerol backbone and are involved in protecting cells from reactive oxygen species (ROS)-induced damage41. Interestingly, some studies have shown lower systemic F2-Isoprostanes (a validated ROS index) in African American population, suggesting lower oxidative stress42. Our data suggests, that elevated levels of plasmalogens in African Americans may possibly be linked to their reduced oxidative stress status. We did not observe any significant difference among the SASP factors (Fig. 6d. Interestingly, IL-6 and soluble IL-6 receptor (sIL-6r) levels were significantly increased in African Americans and Caucasians, respectively. IL-6 is a clear marker of inflammation, whereas sIL-6r is associated with percent body-fat composition43,44 (Fig. 6e). Tryptophan levels were increased in African American rapid agers but not altered among Caucasians. Similarly, the levels of hydroxy lysine and azelate (nonanedioate; C9), a DCA, were reduced among African American healthy agers but significantly increased in Caucasian healthy agers. On the other hand, N-acetylcarnosine was increased in African American healthy agers but decreased in Caucasian healthy agers. Another significantly important metabolite, bromotryptophan was identified with African American rapid agers and not among Caucasians (Fig. 6e, source data). It has been reported that the levels of serum 6-bromotryptophan is a risk factor for chronic kidney disease (CKD) progression45. Considering this, rapid agers of the African American descent in this study cohort may be at a higher risk for CKD. These findings suggest that age-associated metabolites may be influenced by race.
Metabolites associated with smokers.
Biological aging can be influenced by lifestyle choices such as smoking. Cigarette smoke produces numerous (~ 4000) compounds with varying levels of toxicity and is known to increase the risk of COPD, cardiovascular disease and other age-related diseases. Previous studies have established differences in metabolome of smoked and never-smoked individuals46. However, the effect of smoking on the metabolites associated with biological age remains unexplored. Therefore, we analyzed the effect of smoking on the metabolome of the SOLVE-IT cohort. Our OPLS-DA analysis demonstrated a moderate separation between the smoked and never-smoked individuals (Fig. 7a). This separation was predominantly due to xenobiotics related to benzoates and caffeine metabolism. Some of the major metabolites in this list includes, 3-methyl catechol sulfate(s), 3-ethyl catechol sulfate(s), and caffeine (Fig. 7b). It is important to note that about 91% of the ‘smoked’ population had quit smoking (average years since they quit was 31 years). We did not observe any changes in both SASP (Fig. 7c) and proinflammatory markers (Fig. 7d) suggesting a ‘prior smoking status’ did not induce prolonged low-grade inflammation.
We then compared the impact of smoking on metabolites associated with biological age. Over all 67 metabolites were significantly altered by the smoking status in the context of biological aging. Cyclic AMP (cAMP) was strongly associated with healthy agers among the smokers but not the never-smokers. cAMP is known to slow aging process by binding to sirtuins 1 and 347. Interestingly, it has been reported that cAMP levels increase during smoking48. Therefore, one possible explanation of our data is that the healthy biological aging of some of the smokers in our study cohort may be linked to the increase in their cAMP levels. Similarly, choline, urea, guanidinosuccinate were strongly associated with the healthy agers of smoked population but they were fairly distributed among healthy and rapid agers of non-smokers. Metabolites such as trigonelline, 3-ethylcatechol sulfate and methyl 3-catechol sulfate were elevated among healthy agers of ‘never smoked’ population but were not observed in the ‘smoked’ population (Fig. 7e, source data). These metabolites are usually from food; but, it is postulated that smoking can enhance its biological conversion49,50. Our results indicate that trigonelline, 3-ethylcatechol sulfate and methyl 3-catechol sulfate may not be affected long-term by smoking. Taken together, our results suggest that smoking status possibly affects some metabolites long-term but their effect on biological age is moderate.