Absolute quantification and characterization of oxylipins in lupus nephritis and systemic lupus erythematosus

Abstract

Systemic lupus erythematosus (SLE) is a chronic autoimmune disease with multi-organ inflammation and defect, which is linked to many molecule mediators. Oxylipins as a class of lipid mediator have not been broadly investigated in SLE. Here, we applied targeted mass spectrometry analysis to screen the alteration of oxylipins in serum of 98 SLE patients and 106 healthy controls. The correlation of oxylipins to lupus nephritis (LN) and SLE disease activity, and the biomarkers for SLE classification, were analyzed. Among 128 oxylipins analyzed, 92 were absolutely quantified and 26 were significantly changed. They were mainly generated from the metabolism of several polyunsaturated fatty acids, including arachidonic acid (AA), linoleic acid (LA), docosahexanoic acid (DHA), eicosapentanoic acid (EPA) and dihomo-γ-linolenic acid (DGLA). Several oxylipins, especially those produced from AA, showed different abundance between patients with and without lupus nephritis (LN). The DGLA metabolic activity and DGLA generated PGE1, were significantly associated with SLE disease activity. Random forest-based machine learning identified a 5-oxylipin combination as potential biomarker for SLE classification with high accuracy. Seven individual oxylipin biomarkers were also identified with good performance in distinguishing SLE patients from healthy controls (individual AUC > 0.7). Interestingly, the biomarkers for differentiating SLE patients from healthy controls are distinct from the oxylipins differentially expressed in LN patients vs. non-LN patients. This study provides possibilities for the understanding of SLE characteristics and the development of new tools for SLE classification.

Keywords: SLEDAI; lupus nephritis; oxylipin; polyunsaturated fatty acids; systemic lupus erythematosus.

Figure 1

Extracted ion chromatograms (EIC) for oxylipin quantification. (A) LC-MS/MS chromatogram of a standard mixture of 128 oxylipins. (B) Several example of LC-MS/MS chromatograms of single oxylipin in magnified views.

Figure 2

Distinct oxylipin expression profile in SLE patients. (A) PCA plot based on all 92 quantified oxylipins showed a separation of SLE patients and healthy controls. (B) Euclidean distance between SLE patients and healthy controls in the current cohort based on the global alterations of oxylipin abundance. C-C, distance within control group; S-S, distance within SLE group and C-S, distance between SLE and healthy controls. ***, p<0.001 by rank-sum test. (C) Diagram summarizing the concentration changes of oxylipins in linoleic acid metabolism and alpha-linolenic acid metabolism pathway. Significantly changed metabolites were indicated in rectangle with different colors. The fold changes of the altered metabolites in SLE patients were showed. *, p<0.05; **, p<0.01; ***, p<0.001. and n.s., non-significant. Linoleic acid (LA), γ-linolenic acid, Dihomo-γ-linolenic acid and α-linolenic acid (ALA) were not included in the 128 molecules analyzed. LOX, COX and CYP represent the enzymes that catalyze the formation of oxylipins.

Figure 3

Different oxylipin metabolism between LN and non-LN patients. Box plots of 12 significantly different oxylipins between LN (n=33) and non-LN (n=65) patients. Dots represent outliers in each group. Control group is also included in this plot. The eight oxylipins in dashed box were all produced from AA. *, p-value<0.05; **, p-value<0.01 and ***, p-value<0.001 by student’s t test or rank-sum test.

Figure 4

Clinical relationship of oxylipins with disease activity descriptors. (A) Heatmap of correlation matrix between PUFAs metabolic activity and disease activity descriptors. Significant correlations were labeled. Red, positive correlation; Blue, negative correlation. PLT, platelet count; WBC, leukocyte count. *, p-value<0.05. (B) Correlation of DGLA metabolic activity with SLEDAI score by linear regression. The correlation coefficient (r) and p-value were also showed. (C) DGLA metabolic activity among active patients, inactive patients and healthy controls. *, p-value<0.05 and ***, p-value<0.001 by rank-sum test. (D) Spearman’s correlation between individual oxylipin and disease activity descriptors. Red, positive correlation; Green, negative correlation. *, p-value<0.05.

Figure 5

Random forest-based classification of SLE patients. (A) ROC curve of different oxylipin combinations for SLE classification. Six curves (3, 5, 10, 20, 46 and 92 features respectively) and the corresponding AUC and 95% confidence interval (CI) were shown. (B) Selected frequency of the top 5 most important features in random forest modeling. (C) ROC curve based on the five most important features in (B) Two thirds of samples were used for random forest modeling by directly using the five most important features in (B, D) Confusion matrix for the prediction of the one third hold out samples. The precision and recall value also calculated.

Figure 6

Univariate ROC analysis of individual oxylipins. (A) ROC curves showed all the oxylipins with AUC > 0.7 for the classification of SLE patients. The AUCs for different oxylipins were 0.823 (9-OxoODE), 0.805 (AA), 0.798 (LTB4), 0.75 (5-OxoETE), 0.745 (15-OxoEDE), 0.739 (AdA) and 0.719 (DHA). (B) The sensitivity and specificity of each seven individual oxylipin biomarkers for the detection of SLE.