Alterations in Serum Polyunsaturated Fatty Acids and Eicosanoids in Patients with Mild to Moderate Chronic Obstructive Pulmonary Disease (COPD)

Abstract

Smoking is a major risk factor for several diseases including chronic obstructive pulmonary disease (COPD). To better understand the systemic effects of cigarette smoke exposure and mild to moderate COPD-and to support future biomarker development-we profiled the serum lipidomes of healthy smokers, smokers with mild to moderate COPD (GOLD stages 1 and 2), former smokers, and never-smokers (n = 40 per group) (ClinicalTrials.gov registration: NCT01780298). Serum lipidome profiling was conducted with untargeted and targeted mass spectrometry-based lipidomics. Guided by weighted lipid co-expression network analysis, we identified three main trends comparing smokers, especially those with COPD, with non-smokers: a general increase in glycero(phospho)lipids, including triglycerols; changes in fatty acid desaturation (decrease in ω-3 polyunsaturated fatty acids, and an increase in monounsaturated fatty acids); and an imbalance in eicosanoids (increase in 11,12- and 14,15-DHETs (dihydroxyeicosatrienoic acids), and a decrease in 9- and 13-HODEs (hydroxyoctadecadienoic acids)). The lipidome profiles supported classification of study subjects as smokers or non-smokers, but were not sufficient to distinguish between smokers with and without COPD. Overall, our study yielded further insights into the complex interplay between smoke exposure, lung disease, and systemic alterations in serum lipid profiles.

Keywords: biomarker; chronic obstructive pulmonary disease; clinical study; lipidomics; tobacco smoke.

Conflict of interest statement

The authors except for Terhi Vihervaar and Kim Ekroos are employees of Philip Morris International. Terhi Vihervaara and Kim Ekroos are employees of Zora Biosciences. Philip Morris International is the sole source of funding and sponsor of this study.

Figures

Figure 1

The serum lipidome reflects smoking and chronic obstructive pulmonary disease (COPD) status. (A) Measured lipid classes and their median concentrations; (B) differentially abundant lipid classes. The heatmap shows lipid classes (y-axis) with significant difference in abundance in any of the group comparisons (x-axis). The log 2-fold change is color-coded (see color key). Statistical significance is marked: ×, FDR adjusted p-value <0.20, *, FDR adjusted p-value <0.05. Sex, age group, and body mass index were used as covariates in the statistical model; (C) differentially abundant lipid species. Heatmap as in Figure 1B, but for individual lipids. The lipids are grouped by class (color bar on right). See the abbreviations section for lipid nomenclature. Triacylglycerols (TAGs) are summarized by their total composition, e.g., TAG 54:3 total (total fatty acid chain length: number of double bonds; the possible individual fatty acid compositions are in parentheses).

Figure 2

Lipidomics data allow for classification of the study groups. (A) Performance characteristics of elastic net logistic regressions for five classification tasks: all smokers (S) vs. all non-smokers (N), CS vs. NS, COPD vs. NS, COPD vs. CS, and FS vs. NS. The classifier was assessed by repeated (n = 100) three-fold cross-validation and four performance metrics are shown (mean ± SEM): Matthew’s correlation coefficient (MCC), area under the receiver operator curve (ROC), sensitivity (Sens), and specificity (Spec); (B) average classification predictions of the S vs. N classifier for each individual sample of the four study groups. The cross-validation (CV) predictions for each sample were averaged over the repeats (n = 100) and represented as a boxplot for each study group (black line = median). By default, a sample with a predicted score >0.5 (red, dashed line) was classified as a smoker (S), otherwise as a non-smoker (N); (C) final classification prediction scores for the full S vs. N model. Other details are as in Figure 2B; (D) lipids and their coefficients in the S vs. N classifier. An increase in lipids with a negative coefficient (red) tilted the classification of a sample toward smoker; an increase in lipids with a positive coefficient (blue) tilted the classification of a sample toward non-smoker. See abbreviations section for lipid nomenclature.

Figure 3

Weighted lipid co-expression network reveals structural and functional lipid modules. (A) Lipid modules detected in the weighted lipid co-expression network. Cluster dendrogram shows the hierarchical clustering of the topological overlap metric. Identified modules are color coded and numbered from 0 to 13. The module features are summarized for the main clusters discussed in the text. See Figure S1 for the complete annotation of the modules; (B) associations between the lipid modules (M0–M13) and the fatty acid composition and lipid classes of their members. See abbreviations section for lipid nomenclature.

Figure 4

Lipid modules show differences among the study groups. (A) Association of the lipid clusters with study group comparisons. Note that this analysis is based on the module eigenlipids (MEs), which are numbered according to their respective modules; (B) lipids with 20:5 and 22:6 fatty acids were significantly less abundant in the COPD group. The heatmap is as in Figure 1B, but for the concentrations of the different (conjugated) fatty acids.

Figure 5

Lipid modules correlate with clinical parameters, blood markers, and sputum protein profiles. (A) Correlation between module eigenlipids and clinical measurements: body mass index (BMI), glucose, triacylglycerols, cholesterol measured in blood by clinical chemistry, lipid modifier use, and the lung function parameters forced FEV1 % Pred., FEV1/FVC best ratio, and transfer factor for carbon monoxide (TLCO) % Pred. Pearson correlation coefficients are shown (red: positive, blue: negative). Significant correlations with a FDR adjusted p-value <0.05 are marked (“*”); (B) correlation between module eigenlipids and plasma markers. The Pearson correlation coefficients are color-coded and statistically significant correlations with FDR adjusted p-values <0.05 are marked. Only markers with at least one significant correlation are included. For the correlation, the measured marker values/concentrations were log-transformed. See Table S3 for the group comparison results for these markers. C3, complement factor C3; CRP, C-reactive protein; VDBP, vitamin D-binding protein; FRTN, ferritin; AAT, alpha-1-antitrypsin; (C) Correlation between module eigenlipids and protein expression profiles in induced sputum [8]. Protein labels are the official symbols of the respective genes (www.genenames.org) [30]. The Pearson correlation coefficients are color-coded and statistically significant correlations with FDR adjusted p-values <0.05 are marked (“*”). Only module eigenlipids with at least one significant correlation are included.