PO99 - VOC Breath Signatures with Diagnostic, Subtype Stratification, and Prognostic potential in Interstitial Lung Diseases
Eliza Tsitoura (Greece)1; Eirini Vasarmidi (Greece)1; Ioanna Argyriou (Greece)1; Matthew Kerr (UK)2; Liam Grimmett (UK)2; Jessica Herbert (UK)2; Owen Birch (UK)2; Tilly Woodland (UK)2; Huw Davies (UK)2; Athina Trachalaki (UK)1 3; Marilena Lourou (Greece)1; Lutz Wollin (Germany)4; Katerina M. Antoniou (Greece)1;
1 - Department of Respiratory Medicine, Laboratory of Molecular and Cellular Pneumonology, School of Medicine, University of Crete, Heraklion, Greece; 2 - Owlstone Medical Ltd, Cambridge, UK; 3 - National Heart and Lung Institute, Imperial College London, London, UK; 4 - Boehinger Ingelheim Pharma GmbH & Co KG, Biberach, Germany;
Keywords: exhaled breath; fibrotic ILDs; biomarkers;
Select the theme: Fibrotic Pulmonary Interstitial Fibrosis
Type: Original Papers
Presentation: Poster Presentation

Introduction: The diagnosis and prognostication of interstitial lung disease (ILD) currently rely on invasive procedures and radiological patterns that frequently fail to capture underlying molecular heterogeneity.

Objectives: Volatile organic compounds (VOCs) in exhaled breath represent a promising, non-invasive alternative for molecular profiling.

Materials and Methods: This single-center, longitudinal study analysed 172 breath samples collected over 12 months from 82 patients with ILD (IPF, cHP, RA-ILD, SSc-ILD, diverse other ILDs) and 13 non-ILD controls. Samples were analysed via untargeted thermal desorption gas chromatography-mass spectrometry (TD-GC-MS). A dataset of 380 VOCs among the 1,339 features identified, fulfilling “on-breath” criteria were included in the analysis. Support Vector Machine (SVM) models with a radial basis function (RBF) kernel were utilized for diagnostic classification, while unsupervised clustering was employed to identify prognostic "breathotypes." VOCs were further mapped to known ILD pathogenic pathways.

Results: A 4-VOC signature distinguished ILD from controls with 100% sensitivity and 62% specificity. A 22-VOC panel achieved a balanced accuracy of 76% across ILD subtypes, showing high performance for RA-ILD (100%) and SSc-ILD (89%), but lower accuracy for cHP (36%). Mapping of an 18-VOC ILD-specific signature identified three primary drivers of pathogenesis: oxidative stress (elevated pentane, p = 0.003), dysbiosis (elevated acetoin, p = 0.034), and vascular dysfunction (decreased isoprene, p = 0.039). Four distinct breathotypes were identified, including a high-risk "progressor" endotype characterized by a 53% progression rate, an 11.7% decline in DLCO, and elevated mitochondrial ROS in airway macrophages and monocyte-like cells. Disease progression was associated with declining 1-hexadecanol (p = 0.035), suggesting antioxidant depletion, and static dimethyl selenide levels (p = 0.002), indicative of increased selenoprotein biosynthesis.

Conclusions: VOC analysis facilitates non-invasive classification, molecular subtyping, and risk stratification in ILD. These breath-based biomarkers reflect key redox, microbiome, and immune-mediated pathways, providing a viable framework for clinical trial enrichment and prognostic potential for disease trajectory.