Metabolite profile of COVID-19 revealed by UPLC-MS/MS-based widely targeted metabolomics

Abstract

The metabolic characteristics of COVID-19 disease are still largely unknown. Here, 44 patients with COVID-19 (31 mild COVID-19 patients and 13 severe COVID-19 patients), 42 healthy controls (HC), and 42 patients with community-acquired pneumonia (CAP), were involved in the study to assess their serum metabolomic profiles. We used widely targeted metabolomics based on an ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS). The differentially expressed metabolites in the plasma of mild and severe COVID-19 patients, CAP patients, and HC subjects were screened, and the main metabolic pathways involved were analyzed. Multiple mature machine learning algorithms confirmed that the metabolites performed excellently in discriminating COVID-19 groups from CAP and HC subjects, with an area under the curve (AUC) of 1. The specific dysregulation of AMP, dGMP, sn-glycero-3-phosphocholine, and carnitine was observed in the severe COVID-19 group. Moreover, random forest analysis suggested that these metabolites could discriminate between severe COVID-19 patients and mild COVID-19 patients, with an AUC of 0.921. This study may broaden our understanding of pathophysiological mechanisms of COVID-19 and may offer an experimental basis for developing novel treatment strategies against it.

Keywords: COVID-19; UPLC-MS/MS; machine learning; potential biomarkers; widely targeted metabolites.

Figure 1

Identification of differentially expressed metabolites between COVID-19 patients and healthy controls. (A) Score plots of three-dimension principal component analysis (3D-PCA) discriminating between the metabolic profiles of healthy controls (HC) and COVID-19 patients (CP). (B) Orthogonal partial least-squares discriminant analysis (OPLS-DA) score plot of HC and CP. (C) Volcanic map of differentially expressed metabolites between HC and CP. Red dots represent upregulated metabolites, gray dots represent unchanged metabolites, and green dots represent downregulated metabolites. (D) The top 20 differentially expressed metabolites ranked by the value importance plot (VIP). (E) The top 20 differentially expressed metabolites ranked by the log₂ fold change (Log₂ FC).

Figure 2

Identification of differentially expressed metabolites between COVID-19 patients and community-acquired pneumonia. (A) Score plots of three-dimension principal component analysis (3D-PCA) discriminating between the metabolic profiles of community-acquired pneumonia (CAP) and COVID-19 patients (CP). (B) Orthogonal partial least-squares discriminant analysis (OPLS-DA) score plot of CAP and CP. (C) Volcanic map of differentially expressed metabolites between CAP and CP. Red dots represent upregulated metabolites, gray dots represent unchanged metabolites, and green dots represent downregulated metabolites. (D) The top 20 differentially expressed metabolites ranked by the value importance plot (VIP). (E) The top 20 differentially expressed metabolites ranked by the log2 fold change (Log2 FC).

Figure 3

Pathway enrichment analysis of differentially expressed metabolites between COVID-19 patients and healthy controls. (A) Kyoto Encyclopedia of Genes and Genomes (KEGG) classification of differentially expressed metabolites between healthy controls (HC) and COVID-19 patients (CP). (B) KEGG pathway enrichment analysis of differentially expressed metabolites. (C) Metabolite Set Enrichment Analysis (MSEA) between HC and CP.

Figure 4

Pathway enrichment analysis of differentially expressed metabolites between COVID-19 patients and community-acquired pneumonia. (A) Kyoto Encyclopedia of Genes and Genomes (KEGG) classification of differentially expressed metabolites between community-acquired pneumonia (CAP) and COVID-19 patients (CP). (B) KEGG pathway enrichment analysis of differentially expressed metabolites. (C) Metabolite Set Enrichment Analysis (MSEA) between CAP and CP.

Figure 5

K-mean clustering analysis and pathway enrichment analysis. (A) K-means clustering analysis of differentially expressed metabolites across different groups. (B) KEGG enrichment analysis of metabolites in subcluster 2. (C) KEGG enrichment analysis of metabolites in subcluster 5. (D) KEGG enrichment analysis of metabolites in subcluster 6. HC, healthy control; CAP, community-acquired pneumonia; CP, COVID-19 patients.

Figure 6

Predictive value of differentially expressed metabolites in predicting COVID-19 patients. (A) Venn diagram showing differentially expressed metabolites between two groups: healthy controls (HC) vs. COVID-19 patients (CP) and community-acquired pneumonia (CAP) vs. CP. (B) KEGG enrichment analysis of common differentially expressed metabolites. (C) Metabolite Set Enrichment Analysis (MSEA) of common differentially expressed metabolites. (D) Pearson correlation of 92 differentially metabolites between the two groups. Only the index of metabolites in the database is shown, and detailed metabolic information and classification are listed in Table S12 . (E) Receiver operating characteristic (ROC) curves of the two classifiers based on cross-validation in distinguishing COVID-19 patients (CP) from community-acquired pneumonia (CAP). (F) The ROC of the two classifiers based on the cross-validation in distinguishing CP from healthy controls (HC). LR, logistic regression; RF, random forest.

Figure 7

Identification of metabolites specific to severe COVID-19 patients. (A) Venn diagram showing differentially expressed metabolites between three groups: healthy controls (HC) vs. COVID-19 patients (CP), community-acquired pneumonia (CAP) vs. CP, and mild COVID-19 (MC) patients vs. severe COVID-19 (SC). (B) KEGG classification of common differentially expressed metabolites. KEGG enrichment pathway analysis (C) and Metabolite Set Enrichment Analysis (MSEA) (D) of 21 common differentially expressed metabolites.

Figure 8

Metabolite performance in predicting severe COVID-19. (A) Kruskal test revealing significant differences between all tested groups: healthy controls (HC), community-acquired pneumonia (CAP), mild COVID-19 (MC), and severe COVID-19 (SC) patients. All metabolite profiles were first log2 transformed and then normalized using Z-score standardization. (B) Pearson correlation analysis of differential metabolites. (C) The top ten ranking metabolites are indicated by random forest. (D) Predictive performance of biomarker metabolites evaluated by ROC analysis. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.