Serum Proteomics Identifies Biomarkers Associated With the Pathogenesis of Idiopathic Pulmonary Fibrosis

Abstract

The heterogeneity of idiopathic pulmonary fibrosis (IPF) limits its diagnosis and treatment. The association between the pathophysiological features and the serum protein signatures of IPF currently remains unclear. The present study analyzed the specific proteins and patterns associated with the clinical parameters of IPF based on a serum proteomic dataset by data-independent acquisition using MS. Differentiated proteins in sera distinguished patients with IPF into three subgroups in signal pathways and overall survival. Aging-associated signatures by weighted gene correlation network analysis coincidently provided clear and direct evidence that aging is a critical risk factor for IPF rather than a single biomarker. Expression of LDHA and CCT6A, which was associated with glucose metabolic reprogramming, was correlated with high serum lactic acid content in patients with IPF. Cross-model analysis and machine learning showed that a combinatorial biomarker accurately distinguished patients with IPF from healthy individuals with an area under the curve of 0.848 (95% CI = 0.684-0.941) and validated from another cohort and ELISA assay. This serum proteomic profile provides rigorous evidence that enables an understanding of the heterogeneity of IPF and protein alterations that could help in its diagnosis and treatment decisions.

Keywords: combinatorial biomarker; indicator panel; machine learning; molecular subtype; serum proteome.

Graphical abstract

Fig. 1

Proteomic features of the IPF subgroups. Molecular subtyping of IPF was based on altered proteomes and their correlations with clinical features. A, cumulative number of proteins identified in serum samples from 30 healthy controls (blue dots) and 30 patients with IPF (red dots). B, numbers of identified proteins in serum samples from 30 healthy controls (blue dots) and 30 Patients with IPF (red dots). C, relative abundance of 2314 serum proteins. Several proteins ranged widely in abundance (black dots). D, consensus clustering analysis of the proteomic profiling identifying three subtypes in the IPF cohort. E, Kaplan–Meier analyses of overall survival (OS) of patients in the S-I (n = 16), S-II (n = 4), and S-III (n = 10) subgroups. (p-values calculated by two-sided log-rank tests). F, heatmap of the overrepresented proteins in the three IPF subtypes. G, proteins differentially expressed in the three IPF subtypes. H, associations between expression of BMP2K, PI16, and ATP5A1 proteins, and overall survival (Kaplan–Meier analysis, p-value from log-rank test, high means IPF/N >median value). I, age with the three IPF proteomic subtypes (p-values calculated by Fisher exact tests). IPF, idiopathic pulmonary fibrosis.

Fig. 2

WGCNA identification of modules of highly correlated genes and assessment of their relationships with clinical variables.A, heatmap of the weighted gene co-expression network. The plot indicates the TOM among all genes analyzed. Genes in columns and their corresponding rows are hierarchically clustered by cluster dendrograms, which are presented along the top and left sides of the plot. B, module-trait relationships between six modules and ten clinical traits. C, heatmap of the change in genes in the module of age. D, heatmap of the age-related genes in the three subgroups. E, associations of HSP90AB and CAMKK1 expression with clinical outcomes in 30 patients with IPF. IPF, idiopathic pulmonary fibrosis; WGCNA, Weighted gene correlation network analysis.

Fig. 3

Aberrantly expressed metabolic enzymes involved in enhanced glycolysis in serum proteomes of patients with IPF.A, pathway schematic showing DEPs (t test, p < 0.05) mapped onto glucose metabolism pathways. B, boxplots showing proteins differentially expressed by Patients with IPF with normal and above-normal levels of serum lactate (p values calculated by t test). C, violin plots of LDHA and LDHB expression in 30 healthy controls (blue dots) and 30 patients with IPF (red dots). D, associations of LDHA expression with clinical outcomes in patients with IPF (p values calculated by log-rank tests). E, violin plots of CCT6A expression in 30 healthy controls (blue dots) and 30 patients with IPF (red dots). F, ELISA validation of CCT6A expression in patients with IPF (p values calculated by t-tests). G, IHC staining showing CCT6A expression in lungs from healthy controls and Patients with IPF. H, IHC staining showing CCT6A expression in the bleomycin model of lung fibrosis in mice. I, representative immunoblots of whole lung lysates of mice incubated with antibodies against CCT6A and GAPDH. J, Western blots of CCT6A expression normalized to β-actin. ⁎p < 0.05, as determined by ANOVA. DEPs, differentially expressed proteins; IHC, immunohistochemistry; IPF, idiopathic pulmonary fibrosis.

Fig. 4

Association of changes in CCT6A expression and high lactic acid concentrations with the fibroblast phenotype.A, representative immunoblots showing CCT6A and α-SMA expression in MCR5 cells transfected with control plasmid and plasmid overexpressing CCT6A. B, Western blots of CCT6A expression normalized to β-actin. ⁎p < 0.05, as determined by ANOVA. C, representative images of α-SMA immunofluorescence staining of MRC5 cells. Original magnification, ×100. Scale bars: 5 μm. D, representative immunoblots showing CCT6A, COLA1, and FN expression in MCR5 cells transfected with control and CCT6A siRNAs. E, Western blots of CCT6A expression normalized to β-actin. ∗p < 0.05, ∗∗p < 0.01, as determined by ANOVA. F, ECAR of control and CCT6A-overexpressing MRC5 cells. G and H, lactate production in the supernatants of MRC5 cells and in the lungs of bleomycin mice. I, pyruvate production in MRC5 cells. J, expression of LDHA mRNA in MRC5 cells overexpressing CCT6A. K, representative immunoblots showing LDHA expression in MRC5 cells overexpressing CCT6A. L, Western blots of LDHA expression normalized to β-actin. ∗p < 0.05, as determined by ANOVA.

Fig. 5

Machine learning–based selection of biomarker combinations for classification of IPF.A, receiver–operating characteristic (ROC) curve for the classification model. Calculation of AUC values in the patient cohort by five-fold cross-validation. Confusion matrix of the four-protein combination in the patient cohort. B, ROC curve for the test model Calculation of AUC values in the public cohort by five-fold cross-validation. Confusion matrix of the four-protein combination in the public cohort. C, associations between the protein combinations and clinical outcomes in 30 patients with IPF of the classification model. D, heatmap of the combination biomarkers in the public cohort (PRIDE project PXD010965). E, ELISA determination of TTR expression in an independent cohort. (p values calculated by t tests). F, correlation between TTR expression and patient age in the study cohort.