Introduction

Monitoring and preserving pulmonary function represents a critical public health priority, especially for native high-altitude Tibetan populations who undergo lifelong chronic hypoxic stress that profoundly reshapes respiratory physiological adaptation (Moore et al. 1998, Beall 2007, West 2012). Although environmental and genetic determinants of lung function in high-altitude residents have been increasingly documented (Simonson et al. 2010, Bigham et al. 2013), the systemic metabolic mechanisms underlying individual differences in ventilatory capacity remain poorly characterized. Given the unique evolutionary and physiological adaptive features of native Tibetans, identifying metabolic signatures linked to pulmonary function may advance our understanding of hypoxia-related respiratory regulation and provide candidate biomarkers for evaluating ventilatory adaptation status (Bigham et al. 2010).

Untargeted metabolomics enables high-throughput profiling of circulating small-molecule metabolites and has become a powerful tool for uncovering molecular correlates of complex physiological phenotypes (Minno et al. 2021). Nevertheless, metabolome-wide association studies (MWAS) encounter common analytical challenges, including ultra-high dimensionality, residual clinical confounding, and unstable variable selection (Bictash et al. 2010). Notably, complex physiological traits such as maximal voluntary ventilation (MVV) are typically regulated by weak and dispersed metabolic signals, which frequently results in null outputs from conventional regularization screening. Traditional single-step screening strategies are therefore prone to false-positive associations or complete feature loss, highlighting the necessity of standardized multi-stage screening frameworks, predefined fallback rules, and rigorous stability validation in metabolomic analyses.

Regularized regression approaches, particularly elastic net regression, effectively handle high-dimensional metabolomic data by balancing variable selection and multicollinearity tolerance (Hong et al. 2023). Compared with single-penalty regularization methods, elastic net (α = 0.5) provides more stable feature retention for complex metabolomic datasets (Friedman et al. 2010). Furthermore, bootstrap resampling and independent dataset validation have been widely recommended to evaluate screening robustness and model generalizability, which substantially improves the reliability of MWAS findings (Zucchini 2000). Despite these methodological advances, metabolomic signatures specifically associated with MVV, a key indicator of overall ventilatory reserve, have not been systematically explored in native high-altitude Tibetans.

To fill this research gap, the present study conducted a comprehensive serum metabolome-wide analysis in 168 native high-altitude Tibetan adults. We adopted a rigorous multi-stage analytical framework integrating tiered univariate filtering, elastic net regularization, and a predefined fallback selection strategy to address potential feature loss during regularized screening. Bootstrap stability evaluation, ten-fold internal cross-validation, and independent external validation were sequentially performed to quantify screening repeatability and model performance after full adjustment for clinical collinearity and confounding factors. This study aimed to identify serum metabolites independently correlated with MVV and characterize potential metabolic correlates of pulmonary ventilatory function under chronic high-altitude hypoxia.