Main

Aging is a complex and multifactorial process of physiological changes strongly associated with various human diseases, including cardiovascular diseases (CVDs), diabetes, neurodegeneration and cancer1. The alterations of molecules (including transcripts, proteins, metabolites and cytokines) are critically important to understand the underlying mechanism of aging and discover potential therapeutic targets for aging-related diseases. Recently, the development of high-throughput omics technologies has enabled researchers to study molecular changes at the system level2. A growing number of studies have comprehensively explored the molecular changes that occur during aging using omics profiling3,4, and most focus on linear changes5. It is widely recognized that the occurrence of aging-related diseases does not follow a proportional increase with age. Instead, the risk of these diseases accelerates at specific points throughout the human lifespan6. For example, in the United States, the prevalence of CVDs (encompassing atherosclerosis, stroke and myocardial infarction) is approximately 40% between the ages of 40 and 59, increases to about 75% between 60 and 79 and reaches approximately 86% in individuals older than 80 years7. Similarly, also in the United States, the prevalence of neurodegenerative diseases, such as Parkinson’s disease and Alzheimer’s disease, exhibits an upward trend as well as human aging progresses, with distinct turning points occurring around the ages of 40 and 65, respectively8,9,10. Some studies also found that brain aging followed an accelerated decline in flies11 and chimpanzees12 that lived past middle age and advanced age.

The observation of a nonlinear increase in the prevalence of aging-related diseases implies that the process of human aging is not a simple linear trend. Consequently, investigating the nonlinear changes in molecules will likely reveal previously unreported molecular signatures and mechanistic insights. Some studies examined the nonlinear alterations of molecules during human aging13. For instance, nonlinear changes in RNA and protein expression related to aging have been documented14,15,16. Moreover, certain DNA methylation sites have exhibited nonlinear changes in methylation intensity during aging, following a power law pattern17. Li et al.18 identified the 30s and 50s as transitional periods during women’s aging. Although aging patterns are thought to reflect the underlying biological mechanisms, the comprehensive landscape of nonlinear changes of different types of molecules during aging remains largely unexplored. Remarkably, the global monitoring of nonlinear changing molecular profiles throughout human aging has yet to be fully used to extract basic insights into the biology of aging.

In the present study, we conducted a comprehensive deep multi-omics profiling on a longitudinal human cohort comprising 108 individuals aged from 25 years to 75 years. The cohort was followed over a span of several years (median, 1.7 years), with the longest monitoring period for a single participant reaching 6.8 years (2,471 days). Various types of omics data were collected from the participants’ biological samples, including transcriptomics, proteomics, metabolomics, cytokines, clinical laboratory tests, lipidomics, stool microbiome, skin microbiome, oral microbiome and nasal microbiome. The investigation explored the changes occurring across different omics profiles during human aging. Remarkably, many molecular markers and biological pathways exhibited a nonlinear pattern throughout the aging process, thereby providing valuable insight into periods of dramatic alterations during human aging.

Results

Most of the molecules change nonlinearly during aging

We collected longitudinal biological samples from 108 participants over several years, with a median tracking period of 1.7 years and a maximum period of 6.8 years, and conducted multi-omics profiling on the samples. The participants were sampled every 3–6 months while healthy and had diverse ethnic backgrounds and ages ranging from 25 years to 75 years (median, 55.7 years). The participants’ body mass index (BMI) ranged from 19.1 kg m−2 to 40.8 kg m−2 (median, 28.2 kg m−2). Among the participants, 51.9% were female (Fig. 1a and Extended Data Fig. 1a–d). For each visit, we collected blood, stool, skin swab, oral swab and nasal swab samples. In total, 5,405 biological samples (including 1,440 blood samples, 926 stool samples, 1,116 skin swab samples, 1,001 oral swab samples and 922 nasal swab samples) were collected. The biological samples were used for multi-omics data acquisition (including transcriptomics from peripheral blood mononuclear cells (PBMCs), proteomics from plasma, metabolomics from plasma, cytokines from plasma, clinical laboratory tests from plasma, lipidomics from plasma, stool microbiome, skin microbiome, oral microbiome and nasal microbiome; Methods). In total, 135,239 biological features (including 10,346 transcripts, 302 proteins, 814 metabolites, 66 cytokines, 51 clinical laboratory tests, 846 lipids, 52,460 gut microbiome taxons, 8,947 skin microbiome taxons, 8,947 oral microbiome taxons and 52,460 nasal microbiome taxons) were acquired, resulting in 246,507,456,400 data points (Fig. 1b and Extended Data Fig. 1e,f). The average sampling period and number of samples for each participant were 626 days and 47 samples, respectively. Notably, one participant was deeply monitored for 6.8 years (2,471 days), during which 367 samples were collected (Fig. 1c). Overall, this extensive and longitudinal multi-omics dataset enables us to examine the molecular changes that occur during the human aging process. The detailed characteristics of all participants are provided in the Supplementary Data. For each participant, the omics data were aggregated and averaged across all healthy samples to represent the individual’s mean value, as detailed in the Methods section. Compared to cross-sectional cohorts, which have only a one-time point sample from each participant, our longitudinal dataset, which includes multiple time point samples from each participant, is more robust for detecting complex aging-related changes in molecules and functions. This is because analysis of multi-time point samples can detect participants’ baseline and robustly evaluate individuals’ longitudinal molecular changes.

Fig. 1: Most molecules and microbes undergo nonlinear changes during human aging.

a, The demographics of the 108 participants in the study are presented. b, Sample collection and multi-omics data acquisition of the cohort. Four types of biological samples were collected, and 10 types of omics data were acquired. c, Collection time range and sample numbers for each participant. The top x axis represents the collection range for each participant (read line), and the bottom x axis represents the sample number for each participant (bar plot). Bars are color-coded by omics type. d, Significantly changed molecules and microbes during aging were detected using the Spearman correlation approach (P < 0.05). The P values were not adjusted (Methods). Dots are color-coded by omics type. e, Differential expressional molecules/microbes in different age ranges compared to baseline (25–40 years old, two-sided Wilcoxon test, P < 0.05). The P values were not adjusted (Methods). f, The linear changing molecules comprised only a small part of dysregulated molecules in at least one age range. g, Heatmap depicting the nonlinear changing molecules and microbes during human aging.

We included samples only from healthy visits and adjusted for confounding factors (for example, BMI, sex, insulin resistance/insulin sensitivity (IRIS) and ethnicity; Extended Data Fig. 1a–d), allowing us to discern the molecules and microbes genuinely associated with aging (Methods). Two common and traditional approaches, linear regression and Spearman correlation, were first used to identify the linear changing molecules during human aging5. The linear regression method is commonly used for linear changing molecules. As expected, both approaches have very high consistent results for each type of omics data (Supplementary Fig. 1a). For convenience, the Spearman correlation approach was used in the analysis. Interestingly, only a small portion of all the molecules and microbes (749 out of 11,305, 6.6%; only genus level was used for microbiome data; Methods) linearly changed during human aging (Fig. 1d and Supplementary Fig. 1b), consistent with our previous studies5 (Methods). Next, we examined nonlinear effects by categorizing all participants into distinct age stages according to their ages and investigated the dysregulated molecules within each age stage compared to the baseline (25–40 years old; Methods). Interestingly, using this approach, 81.03% of molecules (9,106 out of 11,305) exhibited changes in at least one age stage compared to the baseline (Fig. 1e and Extended Data Fig. 2a). Remarkably, the percentage of linear changing molecules was relatively small compared to the overall dysregulated molecules during aging (mean, 16.2%) (Fig. 1f and Extended Data Fig. 2b). To corroborate our findings, we employed a permutation approach to calculate permutated P values, which yielded consistent results (Methods). The heatmap depicting all dysregulated molecules also clearly illustrates pronounced nonlinear changes (Fig. 1g). Taken together, these findings strongly suggest that a substantial number of molecules and microbes undergo nonlinear changes throughout human aging.

Clustering reveals nonlinear multi-omics changes during aging

Next, we assessed whether the multi-omics data collected from the longitudinal cohort could serve as reliable indicators of the aging process. Our analysis revealed a substantial correlation between a significant proportion of the omics data and the ages of the participants (Fig. 2a). Particularly noteworthy was the observation that, among all the omics data examined, metabolomics, cytokine and oral microbiome data displayed the strongest association with age (Fig. 2a and Extended Data Fig. 3a–c). Partial least squares (PLS) regression was further used to compare the strength of the age effect across different omics data types. The results are consistent with the results presented above in Fig. 2a (Methods). These findings suggest the potential utility of these datasets as indicators of the aging process while acknowledging that further research is needed for validation4. As the omics data are not accurately matched across all the samples, we then smoothed the omics data using our previously published approach19 (Methods and Supplementary Fig. 2a–c). Next, to reveal the specific patterns of molecules that change during human aging, we then grouped all the molecules with similar trajectories using an unsupervised fuzzy c-means clustering approach19 (Methods, Fig. 3b and Supplementary Fig. 2d,e). We identified 11 clusters of molecular trajectories that changed during aging, which ranged in size from 638 to 1,580 molecules/microbes (Supplementary Fig. 2f and Supplementary Data). We found that most molecular patterns exhibit nonlinear changes, indicating that aging is not a linear process (Fig. 2b). Among the 11 identified clusters, three distinct clusters (2, 4 and 5) displayed compelling, straightforward and easily understandable patterns that spanned the entire lifespan (Fig. 2c). Most molecules within these three clusters primarily consist of transcripts (Supplementary Fig. 2f), which is expected because transcripts dominate the multi-omics data (8,556 out of 11,305, 75.7%). Cluster 4 exhibits a relatively stable pattern until approximately 60 years of age, after which it shows a rapid decrease (Fig. 2c). Conversely, clusters 2 and 5 display fluctuations before 60 years of age, followed by a sharp increase and an upper inflection point at approximately 55–60 years of age (Fig. 2c). We also attempted to observe this pattern of molecular change during aging individually. The participant with the longest follow-up period of 6.8 years (Fig. 1c) approached the age of 60 years (range, 59.5–66.3 years; Extended Data Fig. 1g), and it was not possible to identify obvious patterns in this short time window (Supplementary Fig. 2g). Tracking individuals longitudinally over longer periods (decades) will be required to observe these trajectories at an individual level.

Fig. 2: Clustering reveals nonlinear changes in multi-omics profiling during human aging.

a, Spearman correlation (cor) between the first principal component and ages for each type of omics data. The shaded area around the regression line represents the 95% confidence interval. b, The heatmap shows the molecular trajectories in 11 clusters during human aging. The right stacked bar plots show the percentages of different kinds of omics data, and the right box plots show the correlation distribution between features and ages (n = 108 participants). c, Three notable clusters of molecules that exhibit clear and straightforward nonlinear changes during human aging. The top stacked bar plots show the percentages of different kinds of omics data, and the top box plots show the correlation distribution between features and ages (n = 108 participants). The box plot shows the median (line), interquartile range (IQR) (box) and whiskers extending to 1.5 × IQR. Bars and lines are color-coded by omics type. Abs, absolute.

Fig. 3: Functional analysis of nonlinear changing molecules in each cluster.

a, Pathway enrichment and module analysis for each transcriptome cluster. The left panel is the heatmap for the pathways that undergo nonlinear changes across aging. The right panel is the pathway similarity network (Methods) (n = 108 participants). b, Pathway enrichment for metabolomics in each cluster. Enriched pathways and related metabolites are illustrated (Benjamini–Hochberg-adjusted P < 0.05). c, Four clinical laboratory tests that change during human aging: blood urea nitrogen, serum/plasma glucose, mean corpuscular hemoglobin and red cell distribution width (n = 108 participants). The box plot shows the median (line), interquartile range (IQR) (box) and whiskers extending to 1.5 × IQR.

Although confounders, including sex, were corrected before analysis (Methods), we acknowledge that the age range for menopause in females is typically between 45 years and 55 years of age20, which is very close to the major transition points in all three clusters (Fig. 2c). Therefore, we conducted further investigation into whether the menopausal status of females in the dataset contributed to the observed transition point at approximately 55 years of age (Fig. 2c) by performing separate clustering analyses on the male and female datasets. Surprisingly, both the male and female datasets exhibited similar clusters, as illustrated in Extended Data Fig. 4a. This suggests that the transition point observed at approximately 55 years of age is not solely attributed to female menopause but, rather, represents a common phenomenon in the aging process of both sexes. This result is consistent with previous studies14,15, further supporting the notion that this transition point is a major characteristic feature of human aging. Moreover, to investigate the possibility that the transcriptomics data might skew the results toward transcriptomic changes as age-related factors, we conducted two additional clustering analyses—one focusing solely on transcriptomic data and another excluding it. Interestingly, both analyses yielded nearly identical three-cluster configurations, as observed using the complete omics dataset (Extended Data Fig. 4b). This reinforces the robustness of the identified clusters and confirms that they are consistent across various omics platforms, not just driven by transcriptomic data.

Nonlinear changes in function and disease risk during aging

To gain further insight into the biological functions associated with the nonlinear changing molecules within the three identified clusters, we conducted separate functional analyses for transcriptomics, proteomics and metabolomics datasets for all three clusters. In brief, we constructed a similarity network using enriched pathways from various databases (Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome) and identified modules to eliminate redundant annotations. We then used all modules from different databases to reduce redundancy further using the same approach and define the final functional modules (Methods, Extended Data Fig. 4c and Supplementary Data). We identified some functional modules that were reported in previous studies, but we defined their more accurate patterns of change during human aging. Additionally, we also found previously unreported potential functional modules during human aging (Supplementary Data). For instance, in cluster 2, we identified a transcriptomic module associated with GTPase activity (adjusted P = 1.64 × 10−6) and histone modification (adjusted P = 6.36 × 10−7) (Fig. 3a). Because we lack epigenomic data in this study, our findings should be validated through additional experiments in the future. GTPase activity is closely correlated with programmed cell death (apoptosis), and some previous studies showed that this activity increases during aging21. Additionally, histone modifications have been demonstrated to increase during human aging22. In cluster 4, we identified one transcriptomics module associated with oxidative stress; this module includes antioxidant activity, oxygen carrier activity, oxygen binding and peroxidase activity (adjusted P = 0.029) (Fig. 3a). Previous studies demonstrated that oxidative stress and many reactive oxygen species (ROS) are positively associated with increased inflammation in relation to aging23. In cluster 5, the first transcriptomics module is associated with mRNA stability, which includes mRNA destabilization (adjusted P = 0.0032), mRNA processing (adjusted P = 3.2 × 10−4), positive regulation of the mRNA catabolic process (adjusted P = 1.46 × 10−4) and positive regulation of the mRNA metabolic process (adjusted P = 0.00177) (Fig. 3a). Previous studies showed that mRNA turnover is associated with aging24. The second module is associated with autophagy (Fig. 3a), which increases during human aging, as demonstrated in previous studies25.

In addition, we also identified certain modules in the clusters that suggest a nonlinear increase in several disease risks during human aging. For instance, in cluster 2, where components increase gradually and then rapidly after age 60, the phenylalanine metabolism pathway (adjusted P = 4.95 × 10−4) was identified (Fig. 3b). Previous studies showed that aging is associated with a progressive increase in plasma phenylalanine levels concomitant with cardiac dysfunction, and dysregulated phenylalanine catabolism is a factor that triggers deviations from healthy cardiac aging trajectories26. Additionally, C-X-C motif chemokine 5 (CXCL5 or ENA78) from proteomics data, which has higher concentrations in atherosclerosis27, is also detected in cluster 2 (Supplementary Data). The clinical laboratory test blood urea nitrogen, which provides important information about kidney function, is also detected in cluster 2 (Fig. 3c). This indicates that kidney function nonlinearly decreases during aging. Furthermore, the clinical laboratory test for serum/plasma glucose, a marker of type 2 diabetes (T2D), falls within cluster 2. This is consistent with and supported by many previous studies demonstrating that aging is a major risk factor for T2D28. Collectively, these findings suggest a nonlinear escalation in the risk of cardiovascular and kidney diseases and T2D with advancing age, particularly after the age of 60 years (Fig. 2c).

The identified modules in cluster 4 also indicate a nonlinear increase in disease risks. For instance, the unsaturated fatty acids biosynthesis pathway (adjusted P = 4.71 × 10−7) is decreased in cluster 4. Studies have shown that unsaturated fatty acids are helpful in reducing CVD risk and maintaining brain function29,30. The pathway of alpha-linolenic acid and linolenic acid metabolism (adjusted P = 1.32 × 10−4) can reduce aging-associated diseases, such as CVD31. We also detected the caffeine metabolism pathway (adjusted P = 7.34 × 10−5) in cluster 4, which suggests that the ability to metabolize caffeine decreases during aging. Additionally, the cytokine MCP1 (chemokine (C-C motif) ligand 2 (CCL2)), a member of the CC chemokine family, plays an important immune regulatory role and is also in cluster 4 (Supplementary Data). These findings further support previous observations and highlight the nonlinear increase in age-related disease risk as individuals age.

Cluster 5 comprises the clinical tests of mean corpuscular hemoglobin and red cell distribution width (Fig. 3c). These tests assess the average hemoglobin content per red blood cell and the variability in the size and volume of red blood cells, respectively. These findings align with the aforementioned transcriptomic data, which suggest a nonlinear reduction in the oxygen-carrying capacity associated with the aging process.

Aside from these three distinct clusters (Fig. 2c), we also conducted pathway enrichment analysis across all other eight clusters, which displayed highly nonlinear trajectories, employing the same method (Fig. 2b and Supplementary Data). Notably, cluster 11 exhibited a consistent increase up until the age of 50, followed by a decline until the age of 56, after which no substantial changes were observed up to the age of 75. A particular transcriptomics module related to DNA repair was identified, encompassing three pathways: positive regulation of double-strand break repair (adjusted P = 0.042), peptidyl−lysine acetylation (adjusted P = 1.36 × 10−5) and histone acetylation (adjusted P = 3.45 × 10−4) (Extended Data Fig. 4d). These three pathways are critical in genomic stability, gene expression and metabolic balances, with their levels diminishing across the human lifespan32,33,34. Our findings reveal a nonlinear alteration across the human lifespan in these pathways, indicating an enhancement in DNA repair capabilities before the age of 50, a marked reduction between the ages of 50 and 56 and stabilization after that until the age of 75. The pathway enrichment results for all clusters are detailed in the Supplementary Data.

Altogether, the comprehensive functional analysis offers valuable insights into the nonlinear changes observed in molecular profiles and their correlations with biological functions and disease risks across the human lifespan. Our findings reveal that individuals aged 60 and older exhibit increased susceptibility to CVD, kidney issues and T2D. These results carry important implications for both the diagnosis and prevention of these diseases. Notably, many clinically actionable markers were identified, which have the potential for improved healthcare management and enhanced overall well-being of the aging population.

Uncovering waves of aging-related molecules during aging

Although the trajectory clustering approaches described above effectively identify nonlinear changing molecules and microbes that exhibit clear and compelling patterns throughout human aging, it may not be as effective in capturing substantial changes that occur at specific chronological aging periods. In such cases, alternative analytical approaches may be necessary to detect and characterize these dynamics. To gain a comprehensive understanding of changes in multi-omics profiling during human aging, we used a modified version of the DE-SWAN algorithm14, as described in the Methods section. This algorithm identifies dysregulated molecules and microbes throughout the human lifespan by analyzing molecule levels within 20-year windows and comparing two groups in 10-year parcels while sliding the window incrementally from young to old ages14. Using this approach and multiomics data, we detected changes at specific stages of lifespan and uncovered the sequential effects of aging. Our analysis revealed thousands of molecules exhibiting changing patterns throughout aging, forming distinct waves, as illustrated in Fig. 3a. Notably, we observed two prominent crests occurring around the ages of 45 and 65, respectively (Fig. 4a). Notably, too, these crests were consistent with findings from a previous study that included only proteomics data14. Specifically, crest 2 aligns with our previous trajectory clustering result, indicating a turning point at approximately 60 years of age (Fig. 2c).

Fig. 4: Waves of molecules and microbes during aging.

a, Number of molecules and microbes differentially expressed during aging. Two local crests at the ages of 44 years and 60 years were identified. b,c, The same waves were detected using different q value (b) and window (c) cutoffs. d, The number of molecules/microbes differentially expressed for different types of omics data during human aging.

To demonstrate the significance of the two crests, we employed different q value cutoffs and sliding window parameters, which consistently revealed the same detectable waves (Fig. 4b,c and Supplementary Fig. 4a,b). Furthermore, when we permuted the ages of individuals, the crests disappeared (Supplementary Figs. 3a and 4c) (Methods). These observations highlight the robustness of the two major waves of aging-related molecular changes across the human lifespan. Although we already accounted for confounders before our statistical analysis, we took additional steps to explore their impact. Specifically, we investigated whether confounders, such as insulin sensitivity, sex and ethnicity, differed between the two crests across various age ranges. As anticipated, these confounders did not show significant differences across other age brackets (Supplementary Fig. 4d). This further supports our conclusion that the observed differences in the two crests are attributable to age rather than other confounding variables.

The identified crests represent notable milestones in the aging process and suggest specific age ranges where substantial molecular alterations occur. Therefore, we investigated the age-related waves for each type of omics data. Interestingly, most types of omics data exhibited two distinct crests that were highly robust (Fig. 3b and Supplementary Fig. 4). Notably, the proteomics data displayed two age-related crests at ages around 40 years and 60 years. Only a small overlap was observed between our dataset and the results from the previous study (1,305 proteins versus 302 proteins, with only 75 proteins overlapping). The observed pattern in our study was largely consistent with the previous findings14. However, our finding that many types of omics data, including transcriptomics, proteomics, metabolomics, cytokine, gut microbiome, skin microbiome and nasal microbiome, exhibit these waves, often with a similar pattern as the proteomics data (Fig. 4d), supports the hypothesis that aging-related changes are not limited to a specific omics layer but, rather, involve a coordinated and systemic alteration across multiple molecular components. Identifying consistent crests across different omics data underscores the robustness and reliability of these molecular milestones in the aging process.

Next, we investigated the roles and functions of dysregulated molecules within two distinct crests. Notably, we found that the two crests related to aging predominantly consisted of the same molecules (Supplementary Fig. 6). To focus on the unique biological functions associated with each crest and eliminate commonly occurring molecules, we removed background molecules present in most stages. To explore the specific biological functions associated with each type of omics data (transcriptomics, proteomics and metabolomics) for both crests, we employed the function annotation approach described above (Methods). In brief, we constructed a similarity network of enriched pathways and identified modules to remove redundant annotations (Supplementary Fig. 6 and Extended Data Fig. 5a,b). Furthermore, we applied the same approach to all modules to reduce redundancy and identify the final functional modules (Methods and Extended Data Fig. 6a). Our analysis revealed significant changes in multiple modules associated with the two crests (Extended Data Fig. 6b–d). To present the results clearly, Fig. 5a displays the top 20 pathways (according to adjusted P value) for each type of omics data, and the Supplementary Data provides a comprehensive list of all identified functional modules.

Fig. 5: Functional analysis of aging-related waves of molecules across the human lifespan.

a, Pathway enrichment and biological functional module analysis for crests 1 and 2. Dots and lines are color-coded by omics type. b, The overlapping of molecules between two crests and three clusters.

Interestingly, the analysis identifies many dysregulated functional modules in crests 1 and 2, indicating a nonlinear risk for aging-related diseases. In particular, several modules associated with CVD were identified in both crest 1 and crest 2 (Fig. 5a), which is consistent with the above results (Fig. 3b). For instance, the dysregulation of platelet degranulation (crest 1: adjusted P = 1.77 × 10−30; crest 2: adjusted P = 1.73 × 10−26)35,36, complement cascade (crest 1: adjusted P = 3.84 × 10−30; crest 2: adjusted P = 2.02 × 10−28), complement and coagulation cascades (crest 1: adjusted P = 1.78 × 10−46; crest 2: adjusted P = 2.02 × 10−28)37,38, protein activation cascade (crest 1: adjusted P = 1.56 × 10−17; crest 2: adjusted P = 1.61 × 10−8) and protease binding (crest 1: adjusted P = 2.7 × 10−6; crest 2: adjusted P = 0.0114)39 have various effects on the cardiovascular system and can contribute to various CVDs. Furthermore, blood coagulation (crest 1: adjusted P = 1.48 × 10−28; crest 2: adjusted P = 9.10 × 10−17) and fibrinolysis (crest 1: adjusted P = 2.11 × 10−15; crest 2: adjusted P = 1.64 × 10−10) were also identified, which are essential processes for maintaining blood fluidity, and dysregulation in these modules can lead to thrombotic and cardiovascular events40,41. We also identified certain dysregulated metabolic modules associated with CVD. For example, aging has been linked to an incremental rise in plasma phenylalanine levels (crest 1: adjusted P = 9.214 × 10−4; crest 2: adjusted P = 0.0453), which can contribute to the development of cardiac hypertrophy, fibrosis and dysfunction26. Branched-chain amino acids (BCAAs), including valine, leucine and isoleucine (crest 1: adjusted P: not significant (NS); crest 2: adjusted P = 0.0173), have also been implicated in CVD development42,43 and T2D, highlighting their relevance in CVD pathophysiology. Furthermore, research suggests that alpha-linolenic acid (ALA) and linoleic acid metabolism (crest 1: adjusted P: NS; crest 2: adjusted P = 0.0217) may be protective against coronary heart disease44,45. Our investigation also identified lipid metabolism modules that are associated with CVD, including high-density lipoprotein (HDL) remodeling (crest 1: adjusted P = 1.073 × 10−8; crest 2: adjusted P = 2.589 × 10−9) and glycerophospholipid metabolism (crest 1: adjusted P: NS; crest 2: adjusted P = 0.0033), which influence various CVDs46,47,48.

In addition, the dysregulation of skin and muscle stability was found to be increased at crest 1 and crest 2, as evidenced by the identification of numerous modules associated with these processes (Fig. 5a,b). This suggests that the aging of skin and muscle is markedly accelerated at crest 1 and crest 2. The extracellular matrix (ECM) provides structural stability, mechanical s