Lipid Ratios for Diagnosis and Prognosis of Pulmonary Hypertension

Abstract

Rationale: Pulmonary hypertension (PH) poses a significant health threat. Current biomarkers for PH lack specificity and have poor prognostic capabilities. Objectives: To develop better biomarkers for PH that are useful for patient identification and management. Methods: An explorative analysis was conducted of a broad spectrum of metabolites in patients with PH, healthy control subjects, and diseased control subjects in training and validation cohorts, together with in vitro studies on human pulmonary arteries. Measurements and Main Results: High-resolution mass spectrometry was performed in 233 subjects coupled with machine learning analysis. Histologic and gene expression analysis was conducted, with a focus on lipid metabolism in human pulmonary arteries of idiopathic pulmonary arterial hypertension lungs and assessment of the acute effects of extrinsic fatty acids (FAs). We enrolled a training cohort of 74 patients with PH, 30 diseased control subjects without PH, and 65 healthy control subjects, as well as an independent validation cohort of 64 subjects. Among other metabolites, FAs were significantly increased. Machine learning showed a high diagnostic potential for PH. In addition, we developed fully explainable lipid ratios with exceptional diagnostic accuracy for PH (areas under the curve of 0.89 in the training cohort and 0.90 in the external validation cohort), outperforming machine learning results. These ratios were also prognostic and complemented established clinical markers and scores, significantly increasing their hazard ratios for mortality risk. Idiopathic pulmonary arterial hypertension lungs showed lipid accumulation and altered expression of lipid homeostasis-related genes. In human pulmonary artery smooth muscle and endothelial cells, FAs caused excessive proliferation and barrier dysfunction, respectively. Conclusions: Our metabolomics approach suggests that lipid alterations in PH provide diagnostic and prognostic information, complementing established markers. These alterations may reflect pathologic changes in the pulmonary arteries of patients with PH.

Keywords: biomarker; blood-based test; metabolomics; prognosis; pulmonary hypertension.

Figure 1.

Study overview for all cohorts. (A) Schematic workflow of metabolomics measurement and computational analysis (created using BioRender.com). (B) Schematic overview of group distribution of all included patients (n = 233) in the training and validation cohorts. (C) Scatterplot of BMI versus age (shape of symbols by sex) in the training cohort with distribution histograms per PH group and per HC and DC group. BMI = body mass index; DC = diseased control subject; HC = healthy subject; HILIC = hydrophilic interaction liquid chromatography; HRMS = high-resolution mass spectrometry; PH = pulmonary hypertension.

Figure 2.

Pulmonary hypertension (PH) is associated with a strong metabolome shift in the training cohort. (A) Independent PC analysis score plot representing the metabolome profile of each subject as a dot. The proximity of the dots indicates the similarity of the subjects’ metabolomes. Clear group separation by PH is visible along the first component. (B) Loading plot corresponding to score plot in A. Each dot represents the contribution of the metabolite to the group separation observed in the score plot. FAs (yellow circles) strongly drive the group separation and are increased in patients with PH. (C) Orthogonal projections to latent structures discriminant analysis maximizes the group difference from PH to HC and DC, and the resulting score plot represents, as in A, the metabolome of each subject. Similarly, proximity indicates similarity, and ellipses mark the 95% confidence interval of the groups. The difference between the metabolome of the PH and HC and DC groups was significant (Q² > 50%; P < 0.001 in 1,000 random permutations). (D) Volcano plot of univariate analysis showing significant (P_BH < 0.05, gray horizontal line) and strong (absolute contrast ratio > 0.25, gray vertical lines) increase in FAs (yellow triangles). For all methods (A–D), 164 identified metabolites from the training cohort samples (n = 169) were used (drift-corrected, log₁₀-transformed data). Colors as in B. FA = fatty acid; Orth = orthogonal; P_BH = Benjamini-Hochberg P value; PC = principal component; sig. = significant(ly).

Figure 3.

Diagnostic accuracy for PH in training and validation cohorts. (A) ROC plots of random forest (RF) (green) and extreme gradient boosting (XGBoost) (blue) model performance to predict class (PH or HC and DC) in both cohorts with either drift-corrected, log₁₀-transformed data (left and middle) or non–drift–corrected, log₁₀-transformed data (right) on the basis of 153 metabolites. The 95% confidence intervals of the AUCs are given in parentheses. (B) ROC plots of the three best out of 240,570 lipid ratios (LRs) for training and validation cohorts with either drift-corrected, log₁₀-transformed data or log₁₀-transformed data with no drift correction (insets). (A and B) Training cohort, n = 169 (solid line), validation cohort, n = 64 (dashed line); ribbons mark 95% confidence intervals. (C) Model performance metrics of sensitivity, specificity, and balanced accuracy for RF, XGBoost, and the three best LRs when based on either training (circles) or validation (diamonds) cohorts only or all available data (squares). The performance of RF and XGBoost was comparable for all three metabolite subsets with and without drift correction. The performance of the three best LRs was consistently more stable and balanced than that of RF or XGBoost. AUC = area under the curve; FA = fatty acid; LPC = lysophosphatidylcholine; PC = phosphatidylcholine; ROC = receiver operating characteristic.

Figure 4.

Lipid ratio 1 (LR1) predicts survival and improves the prediction of established clinical scores. (A) Cox HR analysis for survival from baseline without age as confounder (left side) and with age (right side; age HR shown in gray). Whiskers mark the 95% confidence intervals, and statistical significance is coded as follows: *P < 0.05, **P < 0.01, and ***P < 0.001. Combining LR1 with FPHR4p or COMPERA (Comparative, Prospective Registry of Newly Initiated Therapies for Pulmonary Hypertension) 2.0 increased the HR compared with either alone. (B–D) Kaplan-Meier curves of survival from baseline by (B) LR1, (C) FPHR4p, and (D) COMPERA 2.0 alone and combined with LR1. All cutoffs defining high and low were optimized using maxstat (https://CRAN.R-project.org/package=maxstat). LR1 was based on log₁₀-transformed data without drift correction. LR1 was combined and scaled from 0 to 1, with equal weighing with either score. FPHR4p was inverted so that higher scores represent higher risk, as in COMPERA 2.0. FA = fatty acid; PC = phosphatidylcholine.

Figure 5.

Presence of lipid accumulation and expression of lipid metabolism genes in idiopathic pulmonary arterial hypertension (IPAH) pulmonary arteries. (A and B) Visualization of endothelium (vWF) and smooth muscle cell layer (SMA [smooth muscle actin]) together with oil red O staining in human donor (A) or IPAH (B) lung serial sections. (C) Gene expression of lipid metabolism–associated genes in laser-capture microdissected human PAs (n = 8–10 patients). Vertical lines represent means with SEMs. Asterisks mark Mann-Whitney test P < 0.05. Cell organelles are adapted from Servier Medical Art (CC BY 2.0 DEED). (D) Immunofluorescence staining of lipid metabolism–associated proteins. Lung sections were labeled with specific antibodies conjugated with SLC27A5 (red), LIPIN2 (red), and DGAT1 (red), with αSMA (green). For control staining, see Figure E10. LCM = laser-capture microdissection; PA = pulmonary artery; vWF = von Willebrand factor.

Figure 6.

Effects of FA treatment on human pulmonary artery smooth muscle cells (hPASMCs) and human pulmonary artery endothelial cells (hPAECs). (A) Representative BODIPY fluorescence staining of hPASMCs and hPAECs in the absence (control [Ctrl]) or presence of extrinsic FAs (scale bar, 50 μm). (B–D) Comparison of cells without (Ctrl, left) and with (FA, right) FA exposure, each dot representing repeated measurements in cells from one donor, connected by a black line, respectively. *P < 0.05 (paired Quade test). (B) PDGF-BB–induced proliferation of primary hPASMCs measured with thymidine incorporation (n = 6). (C) Acetylcholine-induced NO production in primary hPAECs (n = 6) measured as mean fluorescence intensity. (D) Transendothelial electrical resistance, as determined by electrical cell-substrate impedance sensor, showing a significant decrease in hPAECs treated with FA, suggesting endothelial leakage. Representative original curve (left) and individual minimal values (right) (n = 6). (E and F) Summarized data from hPASMCs and hPAECs using the Seahorse XFe24 extracellular flux analyzer. Oxygen consumption rate (OCR) values represent FA-to-Ctrl ratios of individual donors (hPASMCs, n = 9; hPAECs, n = 6). The boxes extend from the 25th to 75th percentiles, the middle line denotes the median, and the whiskers mark the minimum and maximum. The Friedman test was performed for all OCR measures on the FA-to-Ctrl ratios. *P < 0.05 (paired, Wilcoxon post hoc test) for normalized OCR; see also Figure E11. FA = fatty acid; FFA = free fatty acid; PDGF-BB = platelet-derived growth factor, two B subunits.