Proteomic, metabolomic and lipidomic profiles in community acquired pneumonia for differentiating viral and bacterial infections

Abstract

Community-acquired pneumonia (CAP) has a significant impact on public health, especially in light of the recent SARS-CoV-2 pandemic. To enhance disease characterization and improve understanding of the underlying mechanisms, a comprehensive analysis of the plasma lipidome, metabolome and proteome was conducted in patients with viral and bacterial CAP infections, including those induced by SARS-CoV-2. Lipidomic, metabolomic and proteomic profiling were conducted on plasma samples of 69 patients suffering either from viral or bacterial CAP. Lipid and metabolite analyses were LC-MS-based, while proteomic analyses were performed using multiple panels of the Olink platform. Statistical methods, machine learning and pathway analyses were conducted investigating differences between the infection types. Through comparison of the bacterial and viral pathogen groups, distinct signatures were observed in the plasma profiles. Notably, linoleic acid-derived inflammation signaling metabolites (EpOME and DiHOME) were increased in viral CAP compared to bacterial CAP. Similarly, proteins involved in cellular immune response and apoptosis (LAG-3 and TRAIL) showed elevated levels in viral CAP, while bacterial CAP exhibited notable elevation in pattern-recognizing receptors (CLEC4D and EN-RAGE). Additionally, within the lipidomic profile at baseline, several lipids displayed notable differences between viral and bacterial pneumonia, including bile acids (GCA, TCA, TCDCA), various tri- and diglycerides (TGs and DGs), and several phosphatidylcholines (PCs). These findings hold promise for facilitating the differential diagnosis of viral and bacterial pulmonary infections based on the systemic lipidome, metabolome and proteome, enabling timely treatment decisions. Additionally, they highlight potential targets for drug research, advancing therapeutic interventions in CAP. By providing valuable insights into the molecular characterization of CAP, this study contributes to the improvement of understanding the disease and, ultimately, the development of effective treatment strategies.

Fig. 1

Metabolomic and lipidomic analyses – 1 (a): Volcano plot with differentially abundant lipids and metabolites. Vertical cut-offs indicate a p-Value below 0.05. Horizontal cut-offs indicate a mean fold change of 20% towards bacterial infection (right) or viral infection (left); 1 (b): Group comparison of selected lipids and metabolites with results of unpaired t-Tests across sampling instances (Day after admission on top. The exact sampling time points in relation to the baseline of patient admission can be obtained from Figure S1). Lipids are presented in an arbitrary unit; 1 (c): Scatter plot of the first two PC of the whole lipidomics/metabolomics dataset, color-coded for pathogen type and severity; 1 (d): Scatter plot of the first two latent components of PLS-DA, color-coded for pathogen type and severity; 1 (e): Ten most important features each differentiating between viral and bacterial infection within the first latent component of the PLS-DA, scaled to 100%.; 1 (f): Lipid subnetwork identified by lipid network enrichment.

Fig. 2

Oxylipin formation from linolenic acid – Scatterplot representing individual sample concentrations, with a horizontal line representing the median. While lipoxygenases-derived linolenic acid metabolites displayed moderately increased effect sizes in the viral infection group, larger and significant changes were observed in the CYP2A2-derived 12,13-EpOME, 12,13-DiHOME and 9,10-DiHOME. Here, Student’s t-Tests with p-Values unadjusted for multiple testing compare the difference in means of viral and bacterial pneumonia. Notably, 9,10-DiHOME remains significant after FDR adjustment, as shown in Table S1.

Fig. 3

Proteomic analyses – 3 (a): Volcano plot with differentially abundant proteins. Vertical cut-offs indicate a p-Value below 0.05. Horizontal cut-offs indicate a mean fold change of 20% towards bacterial infection (right) or viral infection (left); 3 (b): Group comparison of selected proteins with results of unpaired t-Tests across sampling instances (Day after admission on top. The exact sampling time points in relation to the baseline of patient admission can be obtained from Figure S1). Proteins are presented in arbitrary units; 3 (c): Scatter plot of the first two PC of the whole proteomics dataset, color-coded for pathogen type and severity; 3 (d): Scatter plot of the first two latent components of PLS-DA, color-coded for pathogen type and severity; 3 (e): Ten most important features each differentiating between viral and bacterial infection within the first latent component of the PLS-DA, scaled to 100%.; 3 (f): Top 10 over-expressed pathways for each group resulting from PADOG gene set analysis.

Fig. 4

Co-expression of significantly altered lipids, metabolites and proteins – Hierarchical agglomerative clustering of significantly altered lipids, metabolites and proteins. Rows of the heatmap are ordered according to pathogen type. The dendrogram was cut to reveal the two sub-clusters with least within-cluster distance.