Abstract
Background
Intrinsic capacity (IC) decline is inherently correlated with aging, yet distinguishing specific IC-related biomarkers from general physiological aging markers remains a significant challenge. We aimed to identify multi-omics signatures associated with IC decline after adjustment for relevant covariates and to explore the functional pathways potentially involved in IC decline.
Methods
We analyzed 110 fecal (metagenomics) and 121 serum (untargeted metabolomics) samples from older adults at Beijing Hospital. Multivariable models were applied adjusting for age, sex, Charlson Comorbidity Index (CCI), fish intake, and fruit intake frequency. Differential analyses and network-based mediation approaches were used to assess microbiome–metabolome–IC associations.
Results
After multivariable adjustment, 57 bacterial species and 56 serum metabolites were associated with IC status. The normal IC group showed enrichment of multiple taxa, including Lactobacillus zeae and Paenibacillus glucanolyticus. IC decline was associated with concurrent alterations in amino acid and carnitine-related metabolic pathways, including changes in L-serine, Cysteine, N6,N6,N6-trimethyl-L-lysine, and carnitine C5-OH. Network-based mediation analysis identified overlapping associations among senescence-related metabolites (N1,N8-diacetylspermidine), dietary-derived microbial products (3-(3-hydroxyphenyl)-3-hydroxypropanoic acid), and secondary bile acids (3-epideoxycholic acid), suggesting a structured microbiome–metabolome architecture linked to IC variation.
Conclusions
This study identifies a multi-omics signature associated with IC decline after adjustment for major demographic, clinical, and dietary factors. The findings reveal concurrent alterations in circulating metabolites related to nutrient and carnitine metabolism, alongside compositional and functional differences in the gut microbiome. Together, these parallel findings characterize a multi-omics profile associated with functional decline. These results provide hypotheses for future validation in longitudinal studies.
Keywords: Intrinsic capacity, Gut microbiota, Metabolomics, Metagenomics
Significance of this study
What is already known on this subject?
Alterations in gut microbial composition and serum metabolome have been reported in aging-related syndromes.
What are the new findings?
-
•
This study identifies distinct gut microbial and serum metabolic signatures associated with intrinsic capacity (IC) decline after adjustment for age, sex, comorbidity burden, and dietary factors.
-
•
The serum metabolic signature associated with IC decline includes concurrent differences in amino acid-related, carnitine-related, and lipid-associated metabolites, indicating changes across multiple circulating metabolite classes rather than an alteration confined to a single pathway.
-
•
Network-based mediation analysis indicates overlapping associations between microbial taxa and circulating metabolites, suggesting a complex microbiome–metabolome structure related to IC-associated metabolic variation.
How might it impact on clinical practice in the foreseeable future?
These findings identify candidate multi-omics features that may eventually complement routine clinical assessment of IC if validated in future prospective studies. Such studies should test whether lower circulating N6,N6,N6-trimethyl-L-lysine and short- and medium-chain acylcarnitines precede IC decline and whether alterations in carnitine metabolism and fatty acid β-oxidation may contribute to its progression. They should also determine whether these metabolites, combined with selected microbial taxa and microbial functional features, provide incremental predictive value beyond established clinical factors.
1. Introduction
Aging is shaped by the complex interaction of biological, environmental, and behavioral factors. To address this complexity, the World Health Organization (WHO) introduced intrinsic capacity (IC) within its healthy aging framework. IC encompasses an individual’s physical and mental abilities, integrating well-being, environmental influences, and risk factors, offering a comprehensive understanding of the aging process [1]. While previous studies have extensively explored the relationship between aging, gut microbiota, and metabolites [[2], [3], [4], [5]], large-scale, multidimensional investigations specifically focusing on IC remain limited. Moreover, since IC decline is inherently correlated with aging, distinguishing specific IC-related biomarkers from general markers of physiological aging remains a significant challenge.
The human gut microbiota, in particular, plays a critical role in maintaining homeostasis and influencing metabolic health, immune function, and cognitive resilience. In addition to reduced alpha diversity, certain taxa such as Clostridiales and Coriobacteriaceae increase, while others, including Lachnospiraceae, Ruminococcaceae, Erysipelotrichaceae, Prevotella spp., and potentially probiotic species like Faecalibacterium prausnitzii, decrease. These shifts in gut microbial composition have been linked to age-related changes [6,7], suggesting that specific microbial taxa or their metabolites could serve as potential indicators of IC. Recently, metabolic dysregulation, such as alterations in acyl carnitine levels and the pentose phosphate pathway, has also been associated with IC decline [8]. However, most existing studies have not fully accounted for the confounding effects of age, leaving it unclear whether these microbial and metabolic shifts are drivers of IC decline or merely byproducts of aging.
Thus, integrating microbiome and metabolomic profiling with rigorous age-adjustment strategies could provide an unprecedented opportunity to quantitatively assess IC. In this study, we employed a multi-omics approach combined with advanced machine learning feature selection to identify biomarkers associated with IC decline after adjustment for relevant covariates. Identifying such biomarkers may contribute to a more objective characterization of IC, provide insights into the biological processes potentially underlying healthy aging, and generate hypotheses for future validation and targeted intervention studies.
2. Methods
2.1. Subjects and design
Participants were recruited from a community in Beijing between February 13 and February 17, 2023. A total of 231 biological samples (110 fecal samples and 121 blood samples) were obtained during the clinical examination. We collected demographic and clinical data, including age, sex, body mass index (BMI), and dietary information. The inclusion criteria were: (a) age >60 years; (b) completion of the Intrinsic Capacity (IC) assessment; and (c) provision of written informed consent. The exclusion criteria were: (a) history of severe gastrointestinal diseases (e.g., inflammatory bowel disease, cancer, or advanced adenoma); (b) use of immunosuppressive agents or antibiotics within the past 30 days, or probiotic products within the past 14 days; and (c) presence of metabolic disorders, such as diabetes mellitus, hyperlipidemia, chronic bowel disease, chronic diarrhea, or constipation. Participants were provided with fecal collection kits for self-collection, while blood samples were collected by trained medical personnel.
An IC score was derived based on measures reflecting the five domains proposed by Beard et al. [9], as detailed in the Supplementary Methods. The Short Physical Performance Battery (SPPB) scoring criteria are provided in Supplementary Table S1. The total IC score was calculated by summing the scores of the five domains, ranging from 0 (worst) to 10 (best). Participants were classified into two groups: "IC decline" (total score ≤8) and "normal" (or non-IC decline, total score 9–10).
2.2. Covariate selection
To minimize confounding in downstream biomarker analyses, host and lifestyle variables were first evaluated for their associations with gut microbial community structure using multivariable PERMANOVA and linear models of alpha diversity (Supplementary Tables S2–S4). Variables significantly associated with β-diversity (Fruit and Charlson Comorbidity Index (CCI), P < 0.05), alongside those exhibiting a strong marginal trend (Fish, P = 0.072), were considered potential ecological confounders. In addition, established host-related factors known to heavily influence the gut microbiome, including age and biological sex, were considered a priori covariates regardless of statistical significance. To prevent severe multicollinearity among highly redundant clinical parameters, the number of medications was excluded from all subsequent multivariable models due to its strong biological and statistical collinearity with the CCI, despite its significance in univariable analysis. Accordingly, age, sex, CCI, fruit intake, and fish intake were included as fixed-effect covariates in all MaAsLin2 analyses, and the machine-learning predictive workflows. Because fruit and fish intake frequencies may conceptually overlap with nutritional status captured by the vitality domain, sensitivity analyses were additionally performed using a reduced covariate model adjusted for age, sex, and CCI only.
2.3. Biomarker identification
Microbiota and metabolite analyses were performed as described in the Supplementary Methods. Feature selection was performed within training sets using a repeated subsampling strategy. In each of the 10 iterations, data were stratified into training (70%) and testing (30%) sets. A combined Random Forest (RF) and Least Absolute Shrinkage and Selection Operator (LASSO) workflow was employed. First, RF models (500 trees) were constructed using both microbial/metabolite features and the finalized set of clinical covariates (age, sex, CCI, fruit, and fish intake) to quantify variable importance based on the Mean Decrease Gini index. These covariates were included to account for potential confounding, but were excluded from the candidate list; the top 20 ranked features were retained. These candidates were further refined using LASSO logistic regression with 5-fold cross-validation, with the same set of predefined covariates forcibly included (unpenalized). Features selected in ≥30% of the iterations were defined as core biomarkers.
Based on the identified core biomarkers, two logistic regression models were developed: one incorporating only the core biomarkers and another adjusting for the finalized covariates (biomarkers + covariates). The performance of these models was evaluated using stratified 10-fold cross-validation. Receiver operating characteristic (ROC) curves were generated based on the cross-validated predicted probabilities, and the area under the curve (AUC) was calculated. The statistical significance of the difference in AUC between the two models was assessed using the DeLong test.
2.4. Bidirectional mediation analysis
Bidirectional mediation analysis was conducted to investigate potential mediation patterns linking gut microbiota, circulating metabolites, and IC. Mediation analyses were performed using the R package mediation (v4.5.0) [10]. For each exposure–mediator–outcome triad, two regression models were fitted adjusting for age, sex, CCI, fish intake, and fruit intake. A linear regression model was used for the continuous mediator, whereas a logistic regression model was used for the binary IC outcome. The Average Causal Mediation Effect (ACME), Average Direct Effect (ADE), and Total Effect were estimated using nonparametric bootstrap with 1,000 simulations. Given the cross-sectional nature of the study, analyses were conducted in both directions by alternately specifying microbiota and metabolites as exposures and mediators, respectively, to identify association-consistent mediation patterns rather than infer causal directionality.
2.5. Statistical analysis
Statistical analyses were performed using R software (version 4.4.0). Continuous variables were compared using the unpaired Student’s t-test or Wilcoxon rank-sum test, while categorical data were analyzed using the chi-square test or Fisher’s exact test. Data visualization and statistical operations employed R packages such as ggplot2, vegan, aPCoA, pheatmap, corrplot, Maaslin2, limma, randomForest, glmnet, pROC, mediation, and mixOmics. To examine associations between clinical metadata and microbial and metabolites while adjusting for covariates. Multivariate Association with Linear Models (MaAsLin2) was applied to assess associations between gut bacterial species and IC domains. For serum metabolites, differential associations with IC domains were evaluated using the limma linear modeling framework. Univariate linear regression was used to quantify the variation in IC domains explained by gut bacterial species and metabolites. For multi-omics integration, we employed mixOmics and performed partial Spearman correlation analysis to identify linkages between differential metabolites and microbial species. Statistical significance was defined as a P-value < 0.05 or a false discovery rate (FDR) corrected q-value < 0.1.
3. Results
3.1. Baseline characteristics of the study population
Fecal samples were collected from a total of 66 subjects clinically diagnosed with IC decline (average age 71.4 ± 6.5 years; 13 males, 53 females) and 44 normal controls (average age 67.2 ± 6.4 years; 10 males, 34 females). Additionally, blood samples were obtained from 121 subjects for serum metabolomics analysis, comprising 73 patients with IC decline and 48 normal controls. The detailed study workflow is illustrated in Fig. 1. Compared to the control group, the IC decline group showed significant differences in age and specific IC domains, including depression (PHQ-9), cognition (MMSE), and vision. No significant differences were observed in other clinical characteristics or dietary intake between the two groups (Supplementary Table S5 and Supplementary Table S6).
Fig. 1.
Schematic diagram of the study protocol. The workflow included the collection of clinical and dietary information, assessment of intrinsic capacity (IC), and biological sampling from all participants. Blood samples were collected for untargeted metabolomics analysis, and fecal samples were obtained for metagenomic sequencing. Abbreviations: LC-MS, liquid chromatography-mass spectrometry; MNA-SF, Mini Nutritional Assessment-Short Form; PHQ-9, Patient Health Questionnaire-9; MMSE, Mini-Mental State Examination; SPPB, Short Physical Performance Battery.
3.2. Gut microbial community and diversity
At the phylum level, Firmicutes and Bacteroidetes were dominant in all fecal samples (Fig. 2a). The top 10 genera are shown in Fig. 2b; notably, Veillonella, Megamonas, and Bifidobacterium showed significant differences in abundance between the two groups. At the species level, Phocaeicola vulgatus, Bacteroides uniformis, and Phocaeicola dorei were the three most dominant species (Fig. 2c). Regarding alpha diversity, the Shannon index was lower in the IC decline group compared to controls, although the difference was not statistically significant (Fig. 2d). Similarly, no significant differences were observed in the Simpson and ACE indices (Supplementary Fig. S1A–B and Supplementary Table S7). Furthermore, PCoA and PCA revealed no significant structural differences in microbiome composition between the two groups (PERMANOVA: P = 0.296 and P = 0.314, respectively; Fig. 2e, Supplementary Fig. S1C, and Supplementary Table S8). Despite the lack of global separation, 57 bacterial species were differentially abundant (Supplementary Tables S9–S10). Notably, 22 bacterial species, including Bifidobacterium pseudocatenulatum, Bifidobacterium catenulatum, Bacteroides ovatus, and Legionella anisa, were significantly enriched in the IC decline group (increased abundance). In contrast, 35 species, including Lactobacillus zeae, Actinomyces naeslundii, Tannerella forsythia, Paenibacillus glucanolyticus, and Paenalcaligenes hominis were enriched in the normal group (decreased abundance in IC decline) (Fig. 2f).
Fig. 2.
Gut microbiota composition differences between the IC decline and normal groups based on metagenomic data. (a–c) Relative abundance of the bacterial community in both groups at the phylum (a), top 10 genus (b), and top 10 species (c) levels. (d) Comparison of species-level alpha diversity (Shannon index) between the two groups. (e) Principal Coordinate Analysis (PCoA) based on Bray-Curtis dissimilarity. Statistical significance was assessed using PERMANOVA (Adonis) (R2 = 0.010, P = 0.296). (f) Differential abundance of bacterial species in the normal (green, n = 44) and IC decline (yellow, n = 66) groups. The relative abundance is plotted on a log10 scale on the y-axis. The relative abundance is plotted on a log10 scale. Differential abundance was identified using MaAsLin2 with linear mixed models, adjusting for age, sex, CCI, fruit intake frequency, and fish intake frequency. Species with unadjusted P < 0.05 are shown; given the modest sample size, no features reached the FDR-adjusted q-value threshold of 0.25, and results should therefore be interpreted as exploratory.
3.3. Metagenomic sequencing revealed potential biomarkers associated with IC
Among the clinical covariates, age explained the largest proportion of variance (>7%) in IC status, which was statistically significant (Fig. 3a). To characterize domain-specific associations, we performed incremental variance analyses adjusting for age, sex, CCI, fish intake, and fruit intake frequency. These analyses revealed that specific bacterial species explained significant proportions of variance in distinct IC subdomains. Notably, Caulobacter sp. FWC26 showed the broadest associations, explaining 13.8% of variance in the composite IC score, 11.9% in locomotion (SPPB), 10.2% in vitality (MNA-SF), 9.7% in psychological function (PHQ-9), and 6.6% in cognitive function (MMSE), all after covariate adjustment (P < 0.05). Actinomyces naeslundii accounted for 5.7% of variance in the composite IC score, while Bifidobacterium pseudocatenulatum and B. catenulatum each explained approximately 5.1–5.2% of variance specifically in psychological function (PHQ-9), suggesting domain-preferential rather than uniform associations (Fig. 3b and Supplementary Table S11).
Fig. 3.
Microbial characteristics associated with IC decline. (a) Variance in IC status explained by covariates, diet, and the microbiome. Significance is indicated by *P < 0.05. (b) Incremental variance in IC subdomains explained by individual bacterial species after adjustment for age, sex, CCI, fish intake, and fruit intake frequency. Only species with at least one significant association (P < 0.05) are displayed. Color intensity reflects the proportion of incremental variance explained (%). (c) Partial Spearman correlations between individual bacterial species and continuous IC domain scores, adjusted for age, sex, CCI, fish intake, and fruit intake frequency. Only species with at least one significant association are shown. Color represents the direction and magnitude of the partial correlation coefficient (ρ); asterisks indicate P < 0.05. (d) Receiver operating characteristic (ROC) analysis evaluating the performance of core species in discriminating IC decline from normal status. AUC values are displayed for models with and without clinical covariates adjustment. (e) KEGG pathway enrichment analysis of significantly differential KO genes. Red and green bars indicate pathways enriched in the normal and IC decline groups, respectively.
To further characterize the directionality of these associations, partial Spearman correlation analyses adjusting for the same covariates revealed that Actinomyces naeslundii, Lactobacillus zeae, Curtobacterium sp. BH.2.1.1, and Rhodobacter sp. CZR27, which were observed at lower abundance in the IC decline group, were positively correlated with the composite IC score (ρ = 0.21–0.34, P < 0.05). By contrast, Legionella anisa, Enterobacter sp. N18.03635, and B. pseudocatenulatum, which were more abundant in the IC decline group, showed negative correlations with the IC score (ρ = −0.19 to −0.30, P < 0.05). Domain-specific patterns were also observed: Brevibacterium linens was positively correlated with MMSE, whereas Paenibacillus glucanolyticus and Flavobacterium pallidum were negatively correlated with SPPB (Fig. 3c).
To assess diagnostic performance, we identified 15 core bacterial species using a combined Random Forest and LASSO feature selection strategy, with age, sex, CCI, fish intake, and fruit intake frequency included as fixed covariates throughout the selection process to isolate microbial signals independent of these confounders (Supplementary Table S12). Logistic regression models were then constructed to discriminate IC decline from normal status using 10-fold cross-validation. The model based solely on these 15 core species achieved a CV-AUC of 0.685. Incorporating the five covariates into the model further improved predictive performance, yielding a CV-AUC of 0.770 (DeLong test, P = 0.028; Fig. 3d), suggesting that microbial markers and host clinical characteristics provide complementary predictive information.
Functional annotation of the metagenome identified 7 significant metabolic pathways based on differential KO genes. Notably, four differential KO genes mapped to the glycolysis/gluconeogenesis pathway were enriched in the normal group (Fig. 3e). In contrast, the majority of differential KO genes were enriched in the IC decline group, involving pathways such as cationic antimicrobial peptide (CAMP) resistance, glutathione metabolism, and taurine and hypotaurine metabolism.
3.4. Metabolic profiling of IC decline and non-IC decline individuals
PLS-DA and OPLS-DA analyses revealed distinct metabolic profiles between the IC decline and control groups (Fig. 4a–b); however, as these methods do not adjust for covariates, differential metabolite identification was performed using linear models with adjustment for age, sex, CCI, fish intake, and fruit intake frequency. A total of 56 metabolites were significantly differentially abundant between groups (P < 0.05; Fig. 4c and Supplementary Table S13). Specifically, 17 metabolites were elevated in the IC decline group, including L-Serine, L-Isoserine, His-Ala-Tyr-Tyr-Leu, 3-(3-Hydroxyphenyl)-3-hydroxypropanoic acid, 3-Epideoxycholic acid, Ile-Arg-Phe, and Leu-Gln-Asn-Arg. In contrast, 39 metabolites were reduced in the IC decline group, including FAHFA (8:0/10:0), Phe-Met, S-methyl-5-thio-D-ribofuranose, succinic acid, methylmalonic acid, kynurenic acid, N6,N6,N6-trimethyl-L-lysine, and multiple acylcarnitine species (carnitine C5-OH, C6:0, C10:1, C10:2), suggesting impaired fatty acid β-oxidation in individuals with IC decline. Sensitivity analyses excluding fruit and fish intake frequencies produced highly concordant effect estimates, with all primary bacterial and metabolite candidates retaining the same direction of association and Spearman correlations of 0.989 and 0.966, respectively. To characterize domain-specific associations, incremental variance analyses adjusting for the same covariates revealed that specific metabolites explained significant proportions of variance in distinct IC subdomains. Tyr-Phe-Val-Arg explained 7.4% of variance in the composite IC score, while His-Ala-Tyr-Tyr-Leu and Leu-Gln-Asn-Arg each explained approximately 7.0% of IC score variance. Carnitine C5-OH showed preferential association with cognitive function (MMSE, 6.3%), and kynurenic acid and 3-(pyrazol-1-yl)-L-alanine were specifically associated with vitality (MNA-SF, 6.0–7.2%), demonstrating domain-preferential rather than uniform metabolic associations with IC (Fig. 4d and Supplementary Table S14). Domain-specific patterns were also observed: Carnitine C5–OH was positively correlated with MMSE, whereas N1,N8–diacetylspermidine was negatively correlated with SPPB (Supplementary Fig. S2).
Fig. 4.
Aberrant metabolic profiles associated with IC decline. (a–b) Score plots of partial least-squares discriminant analysis (PLS-DA) and orthogonal partial least-squares discriminant analysis (OPLS-DA) showing group separation. (c) Volcano plot showing differential metabolites between the IC decline and normal groups. Red and blue points represent metabolites significantly up- and down-regulated in the IC decline group, respectively (P < 0.05). Differential metabolites were defined as those with nominal P < 0.05 and |log2 fold change| > 0.3. (d) Variance in IC components explained by individual metabolites. The height of each bar represents the proportion of variance explained in in multivariable linear regression models, calculated after adjusting for age, sex, CCI, and dietary intake. Only metabolites with significant associations (P < 0.05) are shown. (e) Receiver operating characteristic (ROC) curves evaluating the predictive performance of identified metabolites in discriminating IC decline from normal IC status. AUC values are displayed for models with and without covariate adjustment. (f) Microbe–metabolite interaction network constructed using MixOmics. Nodes represent metabolites (squares) and microbial species (circles). Edge colors indicate positive (orange) or negative (blue) correlations, and edge thickness denotes correlation strength.
To assess diagnostic performance, we identified 19 core metabolites using a combined Random Forest and LASSO feature selection strategy, with age, sex, CCI, fish intake, and fruit intake frequency included as fixed covariates throughout to isolate metabolic signals independent of these confounders (Supplementary Table S15). Logistic regression models were evaluated by 10-fold cross-validation. The model based solely on these 19 core metabolites achieved a CV-AUC of 0.778. Incorporating the five covariates further improved predictive performance to a CV-AUC of 0.875 (DeLong test, P = 0.008; Fig. 4e), suggesting that metabolic markers and host clinical characteristics provide complementary predictive information.
3.5. Integrative multi-omics signatures of IC decline and non-IC decline individuals
To explore the interplay between the gut microbiome and metabolome, we constructed a microbe–metabolite interaction network using MixOmics (Fig. 4f) and a correlation heatmap. Strong correlations were observed between differential microbial species and altered metabolites, with 22 significant microbe–metabolite associations identified (Fig. 5a and Supplementary Table S16). The majority were positive correlations (17 of 22), indicating co-directional abundance patterns between specific bacterial taxa and metabolites. Leu-Gln-Asn-Arg showed the broadest connectivity, being positively associated with seven microbial species including Pantoea dispersa (ρ = 0.30), Enterobacter sp. N18.03635 (ρ = 0.27), Propionimicrobium sp. Marseille-P3275 (ρ = 0.26), and Spirosoma rigui (ρ = 0.25), all of which are enriched in the IC decline group. Among negative associations, Georgenia sp. Z443 was inversely correlated with carnitine C6:0 (ρ = −0.25), and Paenalcaligenes hominis showed a negative association with N6,N6,N6-trimethyl-L-lysine (ρ = −0.21), both of which are reduced in IC decline. Pathway enrichment analysis identified significant enrichment of pathways related to carnitine metabolism, thiamine metabolism, energy metabolism, and amino acid metabolism and its derivatives (Fig. 5b).
Fig. 5.
Schematic overview of the potential biological mechanisms underlying IC decline. (a) Covariate-adjusted Spearman correlation heatmap between significantly altered microbial species and metabolites. Red and blue indicate positive and negative correlations, respectively. Significance levels are denoted as *P < 0.05. (b) Top 25 enriched metabolic pathways identified using RaMP-DB. Dot size reflects the number of metabolites mapped to each pathway, and color indicates the significance level (P-value). (c) Summary of multi-omics features associated with IC decline. Serum metabolomic findings and microbiome-derived functional annotations are displayed in separate compartments according to their analytical origins. Carnitine synthesis enrichment was derived from serum metabolite-set analysis using RaMP-DB. K25026, K01689, and K01803 represent metagenome-derived microbial functional annotations related to glycolysis/gluconeogenesis and should not be interpreted as host enzyme activity.
Finally, covariate-adjusted bidirectional mediation analyses identified potential association-consistent patterns linking the gut microbiome, circulating metabolome, and IC decline (Supplementary Fig. S3, and Supplementary Table S17). In models specifying microbial taxa as exposures and metabolites as mediators, three broad metabolite classes, including carnitine-related metabolites, amino acid-derived metabolites, and lipid- or bile acid-related metabolites, were recurrently involved in statistical mediation patterns linking multiple microbial taxa with IC outcomes. In the alternative model specification, in which metabolites were specified as exposures and microbial taxa as mediators, similar metabolite classes were also recurrently observed, suggesting a consistent covariation structure across the microbiome, metabolome, and IC axis. Overall, these findings indicate that energy related, amino acid related, and lipid related metabolic modules may jointly reflect the integrated microbial and metabolic landscape associated with IC decline, rather than implying a specific causal direction.
In summary, biological processes including amino acid metabolism disorders (specifically cysteine and methionine), impaired carnitine synthesis, and energy metabolism dysfunction appear to be closely associated with IC decline (Fig. 5c). A detailed summary of significant pathways and their related genes is provided in Supplementary Table S18.
4. Discussion
By integrating multi-omics data, this study characterizes distinct microbiota and serum metabolite signatures associated with IC decline after adjusting for age, sex, CCI, fish intake, and fruit intake frequency. Our findings suggest that compositional differences in microbial and metabolic profiles remain evident after accounting for major demographic, clinical, and dietary factors. At the metabolite level, the observed differences involved amino acid metabolism, carnitine-related metabolism, and lipid-associated processes, indicating that IC decline is associated with alterations across multiple metabolic systems. Together, the parallel microbial and metabolic findings provide a basis for future studies to test whether host–microbiome interactions contribute to IC decline [11,12].
A broad range of aging-related syndromes (e.g., obesity, type 2 diabetes, insulin resistance, frailty, and sarcopenia) are associated with changes in the gut microbiome [13,14]. Previous studies have indicated that Coprobacillus, Dialister significantly increased and Faecalibacterium, Paraprevotella, and Sutterella significantly decreased in these conditions. However, most existing studies have not fully accounted for confounding variables, and strategies like age-sex matching often result in reduced sample utilization. To maximize statistical power, we utilized the full dataset and rigorously corrected for covariates in our analysis. Using this approach, we identified 57 bacterial species with marked differential abundance. Notably, we observed a significant enrichment of beneficial taxa, such as Lactobacillus zeae and Paenibacillus glucanolyticus in the normal group. These microbes play pivotal roles in maintaining gut homeostasis, primarily by degrading dietary fibers into short-chain fatty acids (SCFAs) and preventing pathogen colonization [15,16]. Although B. catenulatum and B. pseudocatenulatum are generally considered beneficial commensals, their enrichment in the IC decline group may reflect context-dependent ecological shifts within the gut microbial community. Such patterns have been observed in other aging- or disease-related conditions, where taxa traditionally regarded as beneficial do not necessarily correspond to improved host functional status. This may be related to changes in available substrates, altered microbial competition, or reorganization of the gut ecological niche under conditions of reduced intrinsic capacity.
Reflecting our multivariable-adjusted serum metabolomic analysis, we identified altered circulating profiles of short oligopeptides (di- to pentapeptides) that remained significantly associated with IC decline after adjusting for covariates. These peptides may reflect altered protein and peptide turnover within the systemic circulation, although the underlying biological sources remain to be determined [17,18]. Pathway-level interpretation of these metabolites suggested enrichment in metabolic networks related to amino acid and energy metabolism, including cysteine and methionine metabolism as well as carnitine-related pathways. Within the cysteine and methionine metabolism framework, we observed increased levels of upstream metabolites such as L-serine together with reduced levels of downstream intermediates, suggesting potential alterations in sulfur amino acid metabolic balance. Given the central role of this pathway in redox homeostasis and methylation processes, these findings may indicate shifts in systemic metabolic regulation associated with IC decline [19,20]. In addition, we observed reduced levels of N6,N6,N6-trimethyl-L-lysine (TML) and carnitine-related metabolites such as C5-OH carnitine, suggesting potential alterations in carnitine biosynthesis and fatty acid oxidation–related metabolic processes [21,22]. Overall, while these findings are hypothesis-generating, they suggest concurrent alterations in amino acid and carnitine-related metabolic pathways associated with IC decline [8,23,24]. Further targeted metabolomic quantification and prospective validation will be required to confirm these observations.
IC is a composite construct encompassing cognitive, locomotor, sensory, psychological, and vitality domains, and individuals classified as IC decline may therefore differ substantially in their dominant functional impairments. We observed domain-specific associations, where certain metabolites and microbial taxa showed differential correlations with MMSE, PHQ-9, MNA-SF, and SPPB scores, suggesting partially distinct microbiome–metabolome patterns underlying cognitive and physical function. These findings indicate that cognitive and locomotor components of IC may reflect related but not identical biological processes. Future studies with larger cohorts and continuous modeling of IC subdomains are needed to better resolve domain-specific microbiome–metabolome relationships and their temporal dynamics.
Mediation analysis revealed complex and partially overlapping relationships between gut microbial taxa, circulating metabolites, and IC. Rather than suggesting unidirectional causal pathways, these findings support a network-level organization of the microbiome–metabolome–IC axis, in which multiple metabolites may act as shared mediators linking different microbial features to host functional status. Among the identified metabolic signals, N1,N8-diacetylspermidine emerged as a key senescence-associated metabolite consistently linked to IC decline, suggesting involvement of systemic aging-related metabolic processes [25]. Interestingly, Bacteroides ovatus, identified as a microbial biomarker, showed a positive correlation with N1,N8-diacetylspermidine, a senescence-associated metabolite biomarker in our data, suggesting that this senescence-associated metabolite covaries with a specific microbial taxon in the context of IC decline. In addition, 3-(3-hydroxyphenyl)-3-hydroxypropanoic acid, a microbial-derived product of aromatic amino acid and polyphenol metabolism, indicated a potential link between dietary substrate processing and host metabolic state [26]. Furthermore, 3-epideoxycholic acid, a secondary bile acid produced through microbial biotransformation, suggested that alterations in bile acid metabolism may represent another layer of microbiome-associated metabolic remodeling in IC decline [27]. Together, these results suggest that microbiome-associated metabolic alterations in senescence-related, aromatic amino acid–derived, and bile acid–related pathways may jointly contribute to the metabolic landscape associated with reduced intrinsic capacity.
Our study has several notable strengths. To our knowledge, this is one of the most comprehensive studies integrating multi-omics data with detailed dietary information. To prevent the information loss associated with dichotomizing IC, we investigated the associations between IC (and its specific domains), the microbiota, and metabolites. Crucially, we applied multivariable models adjusting for key demographic, clinical, and dietary covariates, including age, sex, CCI, fish intake, and fruit intake frequency. This approach helped reduce confounding from major demographic and lifestyle factors that may influence both IC and the gut microbiome. Accordingly, the observed associations should be interpreted as associations independent of measured covariates rather than being exclusively driven by chronological aging.
Several limitations should be acknowledged. First, due to the cross-sectional design, causal inference cannot be established, and residual confounding cannot be fully excluded despite multivariable adjustment for age, sex, comorbidity burden, and dietary factors. Although CCI was included as a proxy for overall medication burden because of its collinearity with the number of medications, it may not fully capture the independent effects of polypharmacy on the gut microbiome and serum metabolome, particularly because detailed information on medication classes, doses, and treatment durations was not comprehensively available; therefore, residual confounding from medication exposure cannot be excluded. Information on other potentially relevant factors, including probiotic use and detailed dietary composition, was also limited and may have influenced the observed associations. Future studies with larger sample sizes and more comprehensive phenotypic characterization are needed to validate these findings. Second, given the modest sample size and the absence of an external validation cohort, the identified features should be regarded as discovery-phase candidates. Their generalizability remains uncertain and requires replication in geographically or demographically distinct populations before clinical application can be considered. Third, the underlying biological mechanisms remain experimentally unverified; future in vivo and in vitro studies are warranted to elucidate the specific molecular pathways linking these markers to IC decline.
5. Conclusion
In conclusion, our multi-omics study identified distinct gut microbial and serum metabolic signatures associated with IC decline after adjustment for age, clinical, and dietary covariates. These findings highlight concurrent alterations in amino acid metabolism, carnitine-related pathways, and lipid-associated metabolic networks in the circulating metabolome, together with differences in gut microbial composition and function. Furthermore, our network-based mediation analysis revealed overlapping associations between microbial taxa and circulating metabolites, indicating a complex microbiome–metabolome structure underlying IC-related metabolic variation. Collectively, these results provide an integrated view of the metabolic landscape associated with IC decline and offer hypotheses for future mechanistic and prospective validation studies.
Authors’ contributions
L.Y. prepared figures and wrote the main manuscript text. L.Y., Z.C., and H.X. did experiments and analysis. K.Y., Z.C., Z.Y., P.J., and Z.J. collected samples. L.J., S.H., and S.J. revised the manuscript. K.Y. and S.J. conceived, designed, and supervised this work. All authors reviewed the manuscript.
Consent for publication
All authors read and approved the final manuscript for publication.
Ethics approval and consent to participate
This cross-sectional study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Beijing Hospital (Approval No. 2022BJYYEC-260-03). All subjects provided informed consent to participate.
Declaration of Generative AI and AI-assisted technologies in the writing process
The authors only used AI to correct the grammar in some sentences. All scientific content, data analyses, interpretations, and conclusions were developed, reviewed, and verified by the authors. The authors take full responsibility for the accuracy and integrity of the final manuscript.
Funding
This work was supported by the National High Level Hospital Clinical Research Funding (BJ-2025-240, BJ-2025-231, BJ-2025-253, BJ-2022-149) and the National Key R&D Program of China (2021YFE0111800).
Availability of data and material
The raw sequencing and metabolomics data reported in this study are available under the China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences project PRJCA030188. The raw sequencing data have been deposited in GSA-Human (accession HRA015392), and the metabolomics data have been deposited in OMIX (accession OMIX013526). Both datasets are available through the CNGB/NGDC data access request process at https://ngdc.cncb.ac.cn/gsa-human/browse/HRA015392 and https://ngdc.cncb.ac.cn/omix/release/OMIX013526.
Data Availability
The data that has been used is confidential.
Data will be made available on request.
Declaration of competing interest
The authors declare that they have no competing interests.
Acknowledgments
Not applicable.
Footnotes
Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.jnha.2026.100945.
Contributor Information
Yuting Kang, Email: kangyuting5049@bjhmoh.cn.
Ji Shen, Email: shenji4350@bjhmoh.cn.
Appendix A. Supplementary data
The following are Supplementary data to this article:
References
- 1.World Health Organization . World Health Organization; Geneva: 2015. World report on ageing and health; p. 2015. [Google Scholar]
- 2.DeJong E.N., Surette M.G., Bowdish D.M.E. The gut microbiota and unhealthy aging: disentangling cause from consequence. Cell Host Microbe. 2020;28(2):180–189. doi: 10.1016/j.chom.2020.07.013. [DOI] [PubMed] [Google Scholar]
- 3.O’Toole P.W., Jeffery I.B. Gut microbiota and aging. Science. 2015;350(6265):1214–1215. doi: 10.1126/science.aac8469. [DOI] [PubMed] [Google Scholar]
- 4.Sun L., Li Z., Hu C., Ding J., Zhou Q., Pang G., et al. Age-dependent changes in the gut microbiota and serum metabolome correlate with renal function and human aging. Aging Cell. 2023;22(12) doi: 10.1111/acel.14028. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Tian H., Ni Z., Lam S.M., Jiang W., Li F., Du J., et al. Precise metabolomics reveals a diversity of aging-associated metabolic features. Small Methods. 2022;6(7) doi: 10.1002/smtd.202200130. [DOI] [PubMed] [Google Scholar]
- 6.Rampelli S., Candela M., Turroni S., Biagi E., Collino S., Franceschi C., et al. Functional metagenomic profiling of intestinal microbiome in extreme ageing. Aging (Albany NY) 2013;5(12):902–912. doi: 10.18632/aging.100623. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Jackson M.A., Jeffery I.B., Beaumont M., Bell J.T., Clark A.G., Ley R.E., et al. Signatures of early frailty in the gut microbiota. Genome Med. 2016;8(1):8. doi: 10.1186/s13073-016-0262-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Pan Y., Li Y., Chhetri J.K., Liu P., Li B., Liu Z., et al. Dysregulation of acyl carnitines, pentose phosphate pathway and arginine and ornithine metabolism are associated with decline in intrinsic capacity in Chinese older adults. Aging Clin Exp Res. 2024;36(1):36. doi: 10.1007/s40520-023-02654-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Beard J.R., Officer A., de Carvalho I.A., Sadana R., Pot A.M., Michel J.P., et al. The World report on ageing and health: a policy framework for healthy ageing. Lancet. 2016;387(10033):2145–2154. doi: 10.1016/s0140-6736(15)00516-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Tingley D., Yamamoto T., Hirose K., Keele L., Imai K. Mediation: R package for causal mediation analysis. J Stat Software. 2014;59(5):1–38. doi: 10.18637/jss.v059.i05. [DOI] [Google Scholar]
- 11.Amorim J.A., Coppotelli G., Rolo A.P., Palmeira C.M., Ross J.M., Sinclair D.A. Mitochondrial and metabolic dysfunction in ageing and age-related diseases. Nat Rev Endocrinol. 2022;18(4):243–258. doi: 10.1038/s41574-021-00626-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Liu Z., Dai X., Zhang H., Shi R., Hui Y., Jin X., et al. Gut microbiota mediates intermittent-fasting alleviation of diabetes-induced cognitive impairment. Nat Commun. 2020;11(1):855. doi: 10.1038/s41467-020-14676-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Lim M.Y., Hong S., Kim J.-H., Nam Y.-D. Association between gut microbiome and frailty in the older adult population in Korea. J Gerontol Series A. 2021;76(8):1362–1368. doi: 10.1093/gerona/glaa319. [DOI] [PubMed] [Google Scholar]
- 14.Pedersen H.K., Gudmundsdottir V., Nielsen H.B., Hyotylainen T., Nielsen T., Jensen B.A.H., et al. Human gut microbes impact host serum metabolome and insulin sensitivity. Nature. 2016;535(7612):376–381. doi: 10.1038/nature18646. [DOI] [PubMed] [Google Scholar]
- 15.Farsijani S., Cauley J.A., Cawthon P.M., Langsetmo L., Orwoll E.S., Kado D.M., et al. Associations between walking speed and gut microbiome composition in older men from the MrOS study. J Gerontol A Biol Sci Med Sci. 2024;79(4) doi: 10.1093/gerona/glae030. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Singh S.D., Bharali P., Nagamani S. Exploring bacterial metabolites in microbe-human host dialogue and their therapeutic potential in Alzheimer’s diseases. Mol Diversity. 2025;29(5):4213–4229. doi: 10.1007/s11030-024-11028-y. [DOI] [PubMed] [Google Scholar]
- 17.Franceschi C., Garagnani P., Parini P., Giuliani C., Santoro A. Inflammaging: a new immune-metabolic viewpoint for age-related diseases. Nat Rev Endocrinol. 2018;14(10):576–590. doi: 10.1038/s41574-018-0059-4. [DOI] [PubMed] [Google Scholar]
- 18.Arapidi G.P., Urban A.S., Osetrova M.S., Shender V.O., Butenko I.O., Bukato O.N., et al. Non-human peptides revealed in blood reflect the composition of intestinal microbiota. BMC Biol. 2024;22(1):178. doi: 10.1186/s12915-024-01975-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Whiley L., Chappell K.E., D’Hondt E., Lewis M.R., Jiménez B., Snowden S.G., et al. Metabolic phenotyping reveals a reduction in the bioavailability of serotonin and kynurenine pathway metabolites in both the urine and serum of individuals living with Alzheimer’s disease. Alzheimers Res Ther. 2021;13(1):20. doi: 10.1186/s13195-020-00741-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Wang K., Liao M., Zhou N., Bao L., Ma K., Zheng Z., et al. Parabacteroides distasonis alleviates obesity and metabolic dysfunctions via production of succinate and secondary bile acids. Cell Rep. 2019;26(1):222–235.e5. doi: 10.1016/j.celrep.2018.12.028. [DOI] [PubMed] [Google Scholar]
- 21.Upadhyayula P.S., Higgins D.M., Mela A., Banu M., Dovas A., Zandkarimi F., et al. Dietary restriction of cysteine and methionine sensitizes gliomas to ferroptosis and induces alterations in energetic metabolism. Nat Commun. 2023;14(1):1187. doi: 10.1038/s41467-023-36630-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Xia C., Peng P., Zhang W., Xing X., Jin X., Du J., et al. Methionine-SAM metabolism-dependent ubiquinone synthesis is crucial for ROS accumulation in ferroptosis induction. Nat Commun. 2024;15(1):8971. doi: 10.1038/s41467-024-53380-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Rattray N.J.W., Trivedi D.K., Xu Y., Chandola T., Johnson C.H., Marshall A.D., et al. Metabolic dysregulation in vitamin E and carnitine shuttle energy mechanisms associate with human frailty. Nat Commun. 2019;10(1):5027. doi: 10.1038/s41467-019-12716-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Mendoza A., Takemoto Y., Cruzado K.T., Masoud S.S., Nagata A., Tantipanjaporn A., et al. Controlled lipid β-oxidation and carnitine biosynthesis by a vitamin D metabolite. Cell Chem Biol. 2022;29(4):660–669.e12. doi: 10.1016/j.chembiol.2021.08.008. [DOI] [PubMed] [Google Scholar]
- 25.Saiki S., Sasazawa Y., Fujimaki M., Kamagata K., Kaga N., Taka H., et al. A metabolic profile of polyamines in parkinson disease: a promising biomarker. Ann Neurol. 2019;86(2):251–263. doi: 10.1002/ana.25516. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Dodd D., Spitzer M.H., Van Treuren W., Merrill B.D., Hryckowian A.J., Higginbottom S.K., et al. A gut bacterial pathway metabolizes aromatic amino acids into nine circulating metabolites. Nature. 2017;551(7682):648–652. doi: 10.1038/nature24661. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Aliwa B., Horvath A., Traub J., Feldbacher N., Habisch H., Fauler G., et al. Altered gut microbiome, bile acid composition and metabolome in sarcopenia in liver cirrhosis. J Cachexia Sarcopenia Muscle. 2023;14(6):2676–2691. doi: 10.1002/jcsm.13342. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The raw sequencing and metabolomics data reported in this study are available under the China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences project PRJCA030188. The raw sequencing data have been deposited in GSA-Human (accession HRA015392), and the metabolomics data have been deposited in OMIX (accession OMIX013526). Both datasets are available through the CNGB/NGDC data access request process at https://ngdc.cncb.ac.cn/gsa-human/browse/HRA015392 and https://ngdc.cncb.ac.cn/omix/release/OMIX013526.
The data that has been used is confidential.
Data will be made available on request.