Integrated Proteomic and Metabolomic Profiling for Developing Novel Plasma-Based Diagnostic Models of Sarcopenia

Abstract

Background: Sarcopenia is a progressive, age-related condition characterized by a decline in skeletal muscle mass, strength and performance. Diagnosis remains challenging because current consensus criteria are difficult to scale and existing biomarkers lack accuracy. This study aimed to develop high-performance plasma-based diagnostic models for sarcopenia by integrating proteomic and metabolomic profiles.

Methods: Participants were selected from the West China Health and Aging Trend study. Sarcopenia was defined according to the 2019 Asian Working Group for Sarcopenia (AWGS) criteria. Two independent 1:1 age- and sex-matched cohorts were constructed: a discovery cohort (40 sarcopenic, 40 non-sarcopenic) and a validation cohort (30 sarcopenic, 30 non-sarcopenic). Fasting plasma samples were profiled using the Olink Explore 384 Inflammation Panel and liquid chromatography-mass spectrometry-based untargeted metabolomics. Gaussian naïve Bayes classifiers were trained for single-omics models, and logistic regression was used to construct combined models in the discovery cohort and evaluate performance in the validation cohort.

Results: Baseline age and sex were similar in sarcopenic and non-sarcopenic groups (discovery: median 72.0 vs. 71.5 years, p = 0.714; validation: 71.0 vs. 71.5 years, p = 0.594; women: 52.5% and 53.3%). The sarcopenic group had lower skeletal muscle index, grip strength and gait speed (all p < 0.05). Sixty-five proteins and 268 metabolites differed between groups. A 7-protein Gaussian naïve Bayes model achieved AUCs of 0.743 (95% CI 0.718-0.767) in discovery and 0.698 (0.561-0.834) in validation; the metabolomic model yielded 0.828 (0.808-0.849) and 0.751 (0.617-0.885). Combined Model 1 integrated the probabilistic outputs of the proteomic (7 proteins) and metabolomic (7 metabolites) models and reached AUCs of 0.951 (0.937-0.965) and 0.823 (0.717-0.930), outperforming single-omics models (discovery: both p < 0.001; validation: vs. proteomic p < 0.05; vs. metabolomic p = 0.147). Combined Model 2 incorporated only the top two biomarkers from each platform (CCL13, FGF2, N-hexadecanoylpyrrolidine and 1-(cyclohexylmethyl)proline), achieving AUCs of 0.853 (0.828-0.878) in discovery and 0.911 (0.839-0.983) in validation and remained superior to single-omics models (discovery: both p < 0.001; validation: both p < 0.05). Its validation performance was comparable to Combined Model 1 (p = 0.124), with sensitivity 86.7%, specificity 80.0%, precision 81.2% and F1-score 0.839.

Conclusions: We have developed high-performance plasma-based diagnostic models for sarcopenia by integrating inflammatory proteomic and metabolomic signatures. A four-biomarker model (Combined Model 2) demonstrated excellent diagnostic performance and may provide a promising clinically scalable approach for the early detection of sarcopenia.

Keywords: combined model; diagnosis; machine learning; metabolomic; proteomic; sarcopenia.