The research population included residents aged ≥ 65 years from Shanghai, China, who had joined China’s national free physical examination program. A total of 380 subjects had a plasma sample available at baseline. The study participants have been described in our previous study [14]. The design was a nested case–control study. Among the 380 subjects, 332 were normal older adults and 48 were patients with sarcopenia; we obtained 48 normal control (NC) subjects matched by age and sex using propensity score matching from the non-sarcopenia subjects. We employed nearest neighbor matching without replacement in a 1:1 manner. We used a caliper of 0.02 standard deviation of the logit of the propensity score. This study was approved by the Ethics Committee of Shanghai University of Medicine and Health Sciences. All participants voluntarily joined this study, provided written informed consent, and completed questionnaires that provided demographic information including age, sex, lifestyle factors, and medical history. Details of measurement methods have been described in our previous cross-sectional study [15].

Sarcopenia was defined according to the Asian Working Group for Sarcopenia (AWGS) criteria [16], in which a person who has low muscle mass, low muscle strength, and/or low physical performance was identified as having sarcopenia. Low muscle mass was classified as relative skeletal muscle mass index (ASM/ht2) less than 7.0 kg/m2 and 5.7 kg/m2 in males and females, respectively; low muscle strength was defined as grip strength < 28 kg or < 18 kg for males and females, respectively; and low physical performance was defined as walking speed < 1.0 m/s for both males and females.

Muscle mass was measured using a direct segmental multi-frequency bioelectrical impedance analysis (BIA) (In-Body720; Biospace Co., Ltd., Seoul, Korea). Muscle strength was assessed by grip strength, measured using a dynamometer (GRIP-D; Takei Ltd, Niigata, Japan). Usual walking speed (m/s) on a 4-m course was used as an objective measure of physical performance. Details of measurement methods have been described in our previous cross-sectional study [15].

Each plasma sample was collected from the study subjects on an empty stomach in the morning and was then separated and stored in freezers at − 80 °C until the metabolomics assay. We thawed the samples at room temperature. First, 150 μL of plasma was added to a new Eppendorf tube, and 10 μL of L-2-chlorophenylalanine (0.3 mg/ml) with methanol dissolved in the tube was used as the internal standard. Next, a 450-μL mixture of methanol/acetonitrile (2/1) was added and vortexed for 1 min. The whole samples were extracted by ultrasonication for 10 min and stored at − 20 °C for 30 min. The extract was centrifuged for 10 min (4 °C, 13,000 RPM). A total of 200 μL of supernatant was dried in a freeze concentration centrifugal dryer, resolubilized by 300 μL of methanol/water (1/4), vortexed for 30 s, and extracted by ultrasonication for 3 min. After vigorous mixing, samples were centrifuged at 4 °C (13,000 rpm) for 10 min, and 150 μL of supernatants were filtered through 0.22-μm microfilters and transferred to LC vials. The vials were left at − 80 °C and then analyzed by LC–MS.

A total of 150 μL of sample was added to a 1.5-mL Eppendorf tube with 20 μL of 2 chloro-l-phenylalanine (0.3 mg/mL) dissolved in methanol as an internal standard, and the tube was vortexed for 10 s. Subsequently, 450 μL of an ice-cold mixture of methanol and acetonitrile (2/1, v/v) was added, and the mixtures were vortexed for 30 s, ultrasonicated in an ice water bath for 10 min, and stored at − 20 °C for 30 min. The extract was centrifuged at 13,000 rpm and 4 °C for 10 min. In a freeze concentration centrifugal dryer, 200 μL of supernatant was dried in a glass bottle. Then, 80 μL of 15 mg/mL methoxylamine hydrochloride in pyridine was subsequently added. The resultant mixture was vortexed vigorously for 2 min and incubated at 37 °C for 90 min. Then, 50 μL of BSTFA (with 1% TMCS) and 20 μL of n-hexane were added into the mixture, which was vortexed vigorously for 2 min and derivatized at 70 °C for 60 min. The samples were placed at ambient temperature for 30 min before GC–MS analysis.

The plasma sample preparation along with LC–MS analysis have been described in detail in our previous study [14]. Briefly, LC–MS analysis was performed on a liquid mass spectrometer system consisting of an ACQUITY ultra-performance liquid chromatography (UPLC) I-Class tandem VION IMS QT high-resolution mass spectrometer (Waters Corporation, Milford, USA). The samples were separated on the ACQUITY UPLC BEH C18 column (Waters Corporation; 1.7 μm, 100 × 2.1 mm) at a flow rate of 0.4 ml/min. The column was maintained at 45 °C, the sample chamber was set at 4 °C, and the injection volume was set to 1 μL. The mobile phases were water containing 0.1% formic acid (solution A) and acetonitrile/methanol (2/3, vol/vol) containing 0.1% formic acid (solution B). The gradient was 0–1 min, 30% B; 1–2.5 min, 30–60% B; 2.5–6.5 min, 60–90% B; 6.5–8.5 min, 90–100% B; 8.5–10.7 min, 100% B; 10.7–10.8 min, 100–1% B, 10.8–13 min, 1%B. The ion source was electrospray ionization (ESI), and the sample mass spectrometry signal acquisition was performed in positive and negative ion scanning mode, respectively. Mass spectrometric tuning parameters for LC–MS analysis employed optimized settings as follows: ion source temperature, 150 °C; capillary voltages, 2.5 kV; desolvation gas flow, 900 L/h; declustering potential, 40 V; collision energy, 4 eV; mass scan range, m/z 50–1,000; and scan time, 0.2 s [14].

A DB-5MSf used-silica capillary column (30 m × 0.25 mm × 0.25 μm, Agilent J& W Scientific, Folsom, CA, USA) was utilized to separate the derivatives; the derived samples were analyzed by GC–MS on an Agilent 7890B gas chromatography system coupled to an Agilent 5977A MSD system (Agilent Technologies Inc., CA, USA) [17]. In splitless mode, the injector temperature was held at 260 °C, and the injection volume was set at 1 μL. The initial oven temperature commenced at 60 °C held at 60 °C for 0.5 min and increased to 125 °C at a rate of 8 °C/min, followed by a ramp to 210 °C at a rate of 8 °C/min, further ramping to 270 °C at a rate of 15 °C/min, and ultimately reaching 305 °C at a rate of 20 C/min, where it was held for 5 min. The MS quadrupole and ion source (electron impact) were set at temperatures of 150 and 230 °C, respectively. Applying a collision energy of 70 eV, mass spectrometric data acquisition took place in full-scan mode (m/z 50–500) with a 5-min solvent delay time. Throughout the analytical run, quality control samples (QCs) were injected at regular intervals (every 10 samples) to generate a dataset for assessing repeatability.

The LC–MS data were processed by the software Progenesis QI version 2.3 (Nonlinear, Dynamics, Newcastle, UK) for meaningful data mining, performing peak alignment, picking, normalizing, and correcting the retention time (RT). The resulting matrix of features included information on the mass-to-charge ratio (m/z), RT, and peak intensities. The identification of compounds is based on the precise m/z, secondary fragments, and isotopic distribution, and the Human Metabolome Database (HMDB) (http://www.hmdb.ca/), LIPID MAPS (version 2.3) (http://www.lipidmaps.org/), Metabolite Mass Spectral Database (METLIN) (http://metlin.scripps.edu/), and self-built databases (EMDB) were used for qualitative analysis. The GC–MS data were imported into MS-DIAL software (version 2.74) for peak detection, peak identification, characterization, peak alignment, wave filtering, etc. The LUG database (Untargeted database of GC–MS rom Lumingbio) was used to characterize the metabolites. The three-dimensional matrix includes the following: sample information, the name of the peak of each substance, retention time, retention index, mass-to-charge ratio, and signal intensity. After screening, all peak signal intensities in each sample were segmented and normalized according to the internal standards with RSD > 0.3. Then, redundancy removal and peak merging were conducted to obtain the data matrix.

To understand the differences in metabolic profiles between the control and sarcopenia groups, principal component analysis (PCA) and orthogonal projection to latent structure with discriminant analysis (OPLS-DA) were used as a statistical analysis tool. To assess the OPLS-DA, two parameters, R2Y and Q2, are used. At the same time, the OPLS-DA model was cross validated by a 200-fold permutation test; the permutation test is evaluated by cross-validation, and the correlation coefficients R2 and Q2 of the cross-validation were used to verify whether there was overfitting [18].

Differential metabolites between groups were selected using a multidimensional couple with single-dimensional analysis. The variable importance in projection (VIP) generated in OPLS-DA represented differential metabolites with biological significance. Furthermore, the significance of differential metabolites was further verified by Student’s t-test. Variables with VIP > 1.0 and p < 0.05 were considered to be potential biomarkers of sarcopenia. The predictive performance of the model was assessed by estimating the area under the receiver operating characteristic (ROC) curve (AUC). At the same time, we also analyzed the correlation between the top 20 metabolites we screened and the components of sarcopenia (muscle mass, grip strength, and walking speed).

Baseline sociodemographic characteristics between the control and sarcopenia groups were compared using an independent t-test for numerical variables and the chi-squared test for categorical variables. Data with a normal distribution are expressed as the mean ± SD and categorical variables are expressed as proportions. Statistical analyses were performed using SPSS version 26.0 (SPSS Incorporation, Chicago, IL, USA) The significance standard was p < 0.05.