A Pilot Study on Proteomic Predictors of Mortality in Stable COPD

Abstract

Chronic Obstructive Pulmonary Disease (COPD) is the third leading cause of global mortality. Despite clinical predictors (age, severity, comorbidities, etc.) being established, proteomics offers comprehensive biological profiling to obtain deeper insights into COPD pathophysiology and survival prognoses. This pilot study aimed to identify proteomic footprints that could be potentially useful in predicting mortality in stable COPD patients. Plasma samples from 40 patients were subjected to both blind (liquid chromatography-mass spectrometry) and hypothesis-driven (multiplex immunoassays) proteomic analyses supported by artificial intelligence (AI) before a 4-year clinical follow-up. Among the 34 patients whose survival status was confirmed (mean age 69 ± 9 years, 29.5% women, FEV₁ 42 ± 15.3% ref.), 32% were dead in the fourth year. The analysis identified 363 proteins/peptides, with 31 showing significant differences between the survivors and non-survivors. These proteins predominantly belonged to different aspects of the immune response (12 proteins), hemostasis (9), and proinflammatory cytokines (5). The predictive modeling achieved excellent accuracy for mortality (90%) but a weaker performance for days of survival (Q² 0.18), improving mildly with AI-mediated blind selection of proteins (accuracy of 95%, Q² of 0.52). Further stratification by protein groups highlighted the predictive value for mortality of either hemostasis or pro-inflammatory markers alone (accuracies of 95 and 89%, respectively). Therefore, stable COPD patients' proteomic footprints can effectively forecast 4-year mortality, emphasizing the role of inflammatory, immune, and cardiovascular events. Future applications may enhance the prognostic precision and guide preventive interventions.

Keywords: COPD; hemostasis; immunity; mortality; prognosis; proteomic fingerprint.

Figure 1

Protein–protein interaction network of proteins significantly related to mortality at four years. STRING-generated network where each node lists the gene symbol or the Ig fraction of the identified proteins. Square nodes represent DAPs identified by the qualitative analysis whereas circle nodes represent the DAPs identified by quantitative analysis. The border color represents the stats value. The fill color represents the assigned functional group: ‘hemostasis’ (purple), ‘cytokine’ (red), ‘complement cascade’ (green), ‘immune adaptive’ (blue), ‘other immune-related pathways’ (brown), and ‘orphan’ (black). A0A075B6K4 (IGLV3-10), A0A0C4DH24 (IGKV6-21), P01717 (IGLV3-25), and P01817 (IGHV2-5) were manually added since they are not included in the STRING database. Abbreviations: Δ%, percent change; MCC, Matthews correlation coefficient (also called the phi coefficient, φ or r_φ).

Figure 2

Protein–protein interaction network of proteins that AI selected for (a) categorical and (b) continuous mortality models. STRING-generated network where each node lists the gene name/Ig fraction of the AI-selected proteins. Square nodes represent qualitative variables, while circle nodes represent quantitative variables. The border color represents the presence/abundance at 4 years in non-survivor (red) and survivor (blue) patients. The fill color represents the assigned functional group: ‘hemostasis’ (purple), ‘cytokine’ (red), ‘complement cascade’ (green), ‘adaptive immunity’ (blue), ‘other immune-related pathways’ (brown), and ‘orphan’ (black). Ig fractions A0A0C4DH24 (IGKV6-21) and P01721 (IGLV6-57) were manually added since they are not included in the STRING database.

Figure 3

Venn diagram of protein variables used for modeling by selection method. Proteins included in the mortality modeling were selected based on a conventional univariable analysis (t-test/Bernard test) or AI selection. The text color represents the assigned functional group: ‘hemostasis’ (purple), ‘cytokine’ (red), ‘complement cascade’ (green), ‘adaptive immunity’ (blue), ‘other immune-related pathways’ (brown), and ‘orphan’ (black). Protein lists are sorted by group and name. ■: qualitative variable (present/absent), +: higher abundance in non-survivors.