Metabolites of the gut microbiota may serve as precise diagnostic markers for sarcopenia in the elderly

This cross-sectional study cohort was composed of two groups with a total of 63 participants that were recruited from Haikou, China. Diagnosis of the sarcopenia was based on the Asian Working Group for Sarcopenia (AWGS) in 2019 criteria (Chen L. K. et al., 2020). Specifically, sarcopenia was diagnosed by meeting one of the following two conditions: low appendicular skeletal muscle mass (ASM) and low Physical performance, or low ASM and low muscle strength. Low muscle strength was classified as handgrip strength of less than 28 kg for males and 18 kg for females. The criteria for low physical performance include a 4-metre walk speed of fewer than 1.0 m/s, a score of 9 on the Short Physical Performance Battery (SPPB), or a time of more than 12 s on the 5-time chair stand test. It was defined as low muscle mass when the ASM value was less than 7 kg/m² for men and less than 5.7 kg/m² for women. All subjects were tested according to the above criteria and subsequently divided into control (CON group, n = 31) and sarcopenia groups (SAR group, n = 32) according to diagnostic criteria. The grip strength of participants’ dominant hand was tested with the dynamometer (CAMRY EH101, China). ASM measurement using bioelectrical impedance analysis (Body composition analyzer CP10A, Runcobo, China).

No participants used antibiotics within a month before collecting fecal samples. All participants fasted for 12 h before the collection of blood samples for the clinical examination of the Liver function tests (TP, ALB, PA, GLB, A/G, CRP), Cholesterol tests (TC, TG, HDL, LDL), Kidney function tests (SCR, UA, BUN), HGB and GLU. On the same day of clinical examination, fecal samples were collected from each participant. After determining the weight of the fecal samples, the sample protector (CW0592M, CWBIO, China) was applied at a one-part fecal sample to five-part sample protectors ratio, and the samples remained at −20°C until further steps. All 63 samples were used for both Whole-genome shotgun (WGS) sequencing and Metabolomics measurements.

The DNA Fecal Mini Kit Produced by QIAamp® (Qiagen, Hilden, Germany) was employed to perform DNA extraction. The integrity of the DNA was evaluated through 0.8% agarose gel electrophoresis, while the spectrophotometric measurement was employed to determine the OD 260/280 ratio. All the DNA samples were used to shotgun metagenomic sequencing utilizing an Illumina HiSeq 2,500 device (Illumina, CA, United States) at Novogene Company (Beijing, China), employing stringent protocols to ensure accurate and comprehensive analysis. Libraries were generated with a targeted fragment length of approximately 300 bp. Paired-end reads were then generated, with 100 bp in both the forward and reverse directions. The reads underwent thorough quality control analysis using FastQC to ensure data quality. Subsequently, the reads were aligned to the human genome in order to eliminate any host DNA fragments.

Species annotation of metagenome samples was performed using metaphan3 software (v3.0.14), and the relative abundance of each species among all samples was calculated. Meanwhile, we used HumanN3 software (v3.0.1) for qualitative and quantitative analysis of metabolic pathways in metagenome samples based on the UniRef90 database.

For CAZyme annotation, we first performed an overlapping assembly of the shotgun reads using MEGAHIT (v1.0) and default parameters. Subsequently, genes were predicted by Metagenemark software based on the assembled contigs. Based on the dbcan database, CAZyme genes were annotated out by default parameters. Meanwhile, the ensemble of genes was de-redundantly adopted by the cd-hit software. Based on bowtie software, we built library comparisons for the redundant post-existing genes and calculated the abundance of functional genes in each sample.

The fecal sample, preserved in −80°C, was thawed in ice. A solution of 400 μL (Methanol: Water = 7:3, V/V), was introduced to a 20 mg sample, and mixed using vortexing for a duration of 3 min. The sample underwent sonication in an ice bath for a period of 10 min, succeeded by vortexing for 1 min. Subsequently, it was placed in a − 20°C environment for 30 min. Following that, the sample was centrifuged at 12000 rpm for 10 min at 4°C. And the sediment was discarded, the resulting supernatant was centrifuged at 12000 rpm for a duration of 3 min at 4°C. Subsequently, 200 μL aliquots of the supernatant were transferred for analysis using Liquid Chromatograph Mass Spectrometer (LC–MS). All samples were acquired using the LC–MS in strict adherence to the prescribed machine protocols and procedures. The analytical conditions were according to the following parameters, Ultra Performance Liquid Chromatography: column, Waters ACQUITY UPLC HSS T3 C18 (1.8 μm, 2.1*100 mm); column temperature, 40°C; flow rate, 0.4 mL/min; injection volume, 2 μL; solvent system, water (0.1% formic acid): acetonitrile (0.1% formic acid); gradient program, 95:5 V/V at 0 min, 10:90 V/V at 11.0 min, 10:90 V/V at 12.0 min, 95:5 V/V at 12.1 min, 95:5 V/V at 14.0 min.

The random forest algorithm was applied to train sample classifiers for distinguishing health conditions employing Significantly different microbiomes and metabolites in SAR and the healthy group, respectively. We used the “ranger” R package (v0.12.1) to build the random forest technique using the default hyperparameters except for 5,000 trees in each classification job. A fivefold cross-validation method assessed the RF models’ prediction performance. We further confirmed the performance via the 50–50 training and testing divides. The findings showed the accuracy in the 50% holdout test set and compared the ultimate performance (accuracy) across several cross-validation methods.

We developed disease classifiers utilizing a range of simplified sets of microbiological characteristics and evaluated their performance to ascertain the optimal number of features that might optimize the model performance. The goal is to track the rise or peak in prediction accuracy as more characteristics are incorporated into a classification model.

The statistical analyses were conducted utilizing the R (v4.2.2) software. The data are reported as mean ± Standard Error of Mean (SEM). Differential abundances of the genera were assessed utilizing both the Wilcoxon rank-sum test and the Kruskal-Wallis test. Genera with a p-value less than 0.05 were deemed significantly different. Principal coordinate analysis (PCoA) was carried out utilizing the “ade4” package to explore the multivariate patterns in the data. Line charts, box plots, volcano plots, and bar charts were generated using the “ggplot2” package for data visualization. Additionally, the construction of the heatmap was accomplished using the “pheatmap” package. The p-values underwent filtering and correction through the implementation of the “DESeq2” package within the bubble diagram. The inference of microbial ecological networks was accomplished by employing Spearman’s rank correlation coefficient using the metagenomic sequencing data. Traceability of metabolites using Metorigin software1 (Yu et al., 2022). These networks were subsequently visualized in Cytoscape (Version 3.10.0) to facilitate meticulous analysis and interpretation.

In this study, 63 participants were recruited and separated into two groups; group one, the healthy control group (CON, n = 31), and group two, the sarcopenia patient group (SAR, n = 32) (Figure 1A). The basic characteristics of these subjects, age, sex, height, and sleep quality, were recorded in Supplementary Table S1. The age distribution of elderly people with sarcopenia was 62–88, and healthy elderly people were 60–79. Meanwhile, we recorded the subjects’ basic motor abilities according to the Strength, assistance with walking, rising from a chair, climbing stairs, and falls (SARC-F) questionnaire.

Obviously, in individuals with sarcopenia, there was a notable reduction in both body mass index (BMI) and calf circumference compared to the healthy group (Figure 1B). From a more detailed perspective, participants with sarcopenia exhibited significant decreases in appendicular skeletal muscle mass (ASM), handgrip strength, 4-metre walking speed, short physical performance battery (SPPB), and SARC-F scores, while the 5-time chair stand test showed a significant increase compared to the control group. The changes in these indexes reveal a comprehensive profile of muscle mass, muscle strength, and physical performance in the poor physical quality of patients with sarcopenia. These underlying physical differences directly map to changes in clinical serum indexes. First, we checked the kidney-related indexes in patients with sarcopenia. There were no significant differences in Serum creatinine (SCR), Uric Acid (UA), and Blood Urea nitrogen (BUN) between CON and SAR groups (Figure 1C). C-reactive protein (CRP), hemoglobin (HGB), total protein (TP) and blood glucose (GLU) in SAR were also within healthy limits (Figures 1D,E). In contrast, prealbumin (PA) was significantly lower in SAR (Figure 1E). In addition, we found a slight increase in serum globulin (GLB) and a slight decrease in serum Albumin (ALB) and albumin to globulin ratio (A/G) in patients with sarcopenia, although the absence of the significant difference between SAR and CON.

For lipid metabolism, we noted a significant decrease in triglycerides (TG) and a significant increase in high-density lipoprotein cholesterol (HDL-C) in patients with sarcopenia (Figure 1D). Similar results have been reported in other clinical studies of sarcopenia (Wang et al., 2020; Jiang et al., 2023).

We next sought to explore the difference in the intestinal microbiome of patients between CON and SAR. The alpha diversity of the gut microbiome was quantified based on profiles of metagenomics species-level (Figure 2A). There was no significant difference in the Shannon index and Simpson index between the CON and SAR groups. This indicates that there is no significant difference in species richness of the gut microbiota between sarcopenic and healthy populations. However, from the perspective of β-diversity of the microbiome, principal coordinates analysis (PCoA) based on Bray-Curtis distances demonstrated that sample points representing the two groups showed significantly different separation trends (p < 0.05) (Figure 2B). This suggests that the microbiome structure is significantly different between patients with sarcopenia and healthy older adults.

Since we have already observed that the gut microbiome structure was disordered in SAR, we further screened for differences in species between the two groups with the aim of quantifying in more detail the differences in the microbiome between the two groups. The proportional abundance of Bifidobacterium longum, Bifidobacterium pseudocatenulatum, and Bifidobacterium adolescentis was significantly decreased in the SAR group (Figure 2C). In addition, several potentially beneficial species were found to be significantly declining in abundance such as Faecalibacterium prausnitzii, Fusicatenibacter saccharivorans, and Phascolarctobacterium faecium. In contrast, the relative abundance of some non-beneficial species, such as Clostridium sp. CAG 242, Phascolarctobacterium succinatutens, and Clostridium sp. CAG 58, showed a significantly higher abundance in SAR compared to the CON group.

After clarifying the structural and species-level differences in the gut microbiome of sarcopenia, we have further analyzed the functionality of the gut microbiome in both cohorts. The top 20 metabolic pathways with significant differences were screened to gain more insight into the possible functional impact of different microbial structures (Figure 3A). Surprisingly, all the top 20 metabolic pathways were enriched in the SAR group, such as ornithine, methionine, and tyrosine arginine and polyamine biosynthesis, which are involved in the metabolic function of amino acids. In addition, fucose degradation, fucose and rhamnose degradation, β-(1,4)-mannan degradation, and TCA cycle V were also significantly upregulated in SAR compared with CON. However, an increase in the functional pathway of partial sugar degradation does not fully represent a concomitant increase in the utilization of carbohydrates. Therefore, we proceeded to compare the abundance of functional genes for carbohydrate-active enzymes in SAR and CON metagenome samples based on the Carbohydrate-Active enzymes (CAZyme) database. The number of significantly up-regulated CAZyme genes in the CON group (3159) was well above that in the SAR group (1891) (Figure 3B). This implied that patients with sarcopenia have a significantly poorer ability in the intestine to synthesize or break down complex carbohydrates and sugar complexes than healthy people.

These results indicate that the gut microbiome of patients with sarcopenia is structurally and functionally distinct from that of the healthy population. Thus, the microbiota in the gut can screen patients with sarcopenia to some extent. Therefore, we chose the 20 species that were screened with significant differences, as noted above. A supervised classification model to screen for sarcopenia was trained based on a random forest approach. The accuracy performance of the model was further validated by 50–50% training and testing of the segmentation. In three consecutive random sample validations, the classifier could distinguish to some extent between healthy people and patients with sarcopenia at the level of intestinal species, although there was some fluctuation in accuracy from low to high, the area under curve (AUC) = 70.83–88.33% (Figure 3C). Also, we ranked 20 differential species based on their contribution to this classifier based on three random samples (Figure 3D). The top 3 highest average rankings were Fusicatenibacter saccharivorans, Bifidobacterium pseudocatenulatum, and Escherichia coli, while the lowest were Bifidobacterium adolescentis, Holdemania filiformis and Dialister sp. CAG 357.

A total of 13,025 metabolites in fecal samples from two groups of 63 subjects was identified based on positive/negative ion mode. The ratio of metabolite classes differs in the two ionic modes, mainly amino acid and its metabolites (13.72–27.37%), benzene and substituted derivatives (12.5–16.32%) and heterocyclic compounds (10.82–15.62%) (Figure 4A). From the structure of the metabolome, the samples representing the SAR and CON groups did not form a distinctly separated trend based on principal component analysis (PCA) (Figure 4B). Compared to the gut microbiome, the structural differences in the intestinal metabolome were not significant. However, in view of specific metabolite levels, there were still many differential metabolites enriched in the CON and SAR groups, respectively (Figure 4C). There were 731 differential metabolites significantly up-regulated in the CON group (means significantly down-regulated in the SAR group) and 351 differential metabolites significantly up-regulated in the SAR group (Figure 4E). To locate the more critical metabolites, we performed further screening with |Log2FoldChange| > 2 and P-adjust>0.01. Finally, 34 highly significant differential metabolites were screened out (Figure 5A). Some potentially beneficial substances, such as Shikimic acid and SM (d18:0/24:0), were significantly down-regulated in the SAR group. Interestingly, short peptides such as Tyr-Ala, Pro-Gly-Asn, Met-Met-Arg, and His-Glu-Phe-Glu were significantly decreased in the SAR group, which may indicate better utilization of these proteins in healthy populations compared to patients with sarcopenia. Although the separation in the structure of the metabolome was not significant, we still found differences in specific metabolite species between the two groups.

Therefore, the same approach was used to build classification models aimed at these differential metabolites. Excitingly, the accuracy of the classification models established for the differential metabolites was consistently above 97% (AUC = 97.08–98.33%) in three consecutive random sample validations (Figure 5B). And among these metabolites, highest three contributing metabolites were Shikimic acid, Glu-Met-Gln and SM (d18:0/24:0), while lowest three were 4-Bromo-2-iodo-6-nitroaniline, 5’-Hydroxymethyl-5′-desmethylmeloxicam and Phenylalanyl-Glycine (Figure 5C).

As mentioned above, patients with sarcopenia and healthy older adults showed disturbances in the gut microbiome and fecal metabolome, further performing significant differences in some clinical indexes. These results prompted us to explore specific associations between the different omics further. Therefore, a network was generated by the Spearman rank correlation coefficient to show the association between bacterial species, metabolites, and clinical indexes associated with sarcopenia (Figure 6). As we emphasized previously, Bifidobacterium longum and Bifidobacterium pseudocatenulatum were significantly decreased in the intestine of the sarcopenic group, and there may be this mutually reinforcing effect relationship between them. The latter, Bifidobacterium pseudocatenulatum, was then linked to calf circumference and showed a positive correlation. We also observed that Bifidobacterium longum showed a positive correlation with Glu-Met-Gln and Carbobenzoxy-4-fluoro-L-phenylalanine, while both may promote the improvement of TG, AMS, Handgrip strength, BMI, and calf circumference. In contrast, Protoasukamycin, positively associated with Collinsella stercoris, may inhibit TG elevation. In addition, Phascolarctobacterium faecium was found to possibly restrain Phascolarctobacterium succinatutens and promote the production of Shikimic acid. Based on the metabolite traceability by Metorigin, we further clarified that Phascolarctobacterium faecium was able to produce Shikimic acid. Then, elevated shikimic acid was likewise linked to AMS, Handgrip strength, BMI, and calf circumference (positive correlation). Thus, the down-regulation of certain crucial species in the gut of patients with sarcopenia may affect their Clinical indexes related to muscle mass and physical performance through the combined action of multiple metabolites.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: NCBI - PRJNA992232, MetaboLights (MTBLS8149).