The detailed method of BALF sample pretreatment is described in the Supplement 3. All compounds in BALF were analyzed using a Cortecs C18 column (2.1 × 100 mm, Waters) on an Ultimate 3000 UHPLC (Dionex) system coupled with a Q Exactive (Orbitrap) mass spectrometer (Thermo Fisher, CA). Detailed parameters for the untargeted metabolic and lipidomic analyses were set following the protocols of our previously reported study.23 (link) Data-dependent MS/MS acquisition (DDA) of all samples was analyzed using TraceFinderTM (Thermo, CA). Metabolites and lipids were identified based on matching precursor and characteristic fragment masses and then assigned using in-house databases in “screening” mode. Any metabolite feature with more than 20% missing values was removed from the result.24 (link),25 (link) Missing values were estimated by the Bayesian PCA (BPCA) method. Data were normalized by the QC group and then auto-scaled using MetaboAnalystR 3.0.26 (link) The normalized data were used for downstream analysis. Principal component analysis (PCA) and orthogonal projections to latent structures discriminant analysis (OPLS-DA) were conducted using SIMCA v14.1 (Umetrics, Sweden).