Multimodal combination of GC × GC-HRTOFMS and SIFT-MS for asthma phenotyping using exhaled breath

Abstract

Chronic inflammatory lung diseases impact more than 300 million of people worldwide. Because they are not curable, these diseases have a high impact on both the quality of life of patients and the healthcare budget. The stability of patient condition relies mostly on constant treatment adaptation and lung function monitoring. However, due to the variety of inflammation phenotypes, almost one third of the patients receive an ineffective treatment. To improve phenotyping, we evaluated the complementarity of two techniques for exhaled breath analysis: full resolving comprehensive two-dimensional gas chromatography coupled to high-resolution time-of-flight mass spectrometry (GC × GC-HRTOFMS) and rapid screening selected ion flow tube MS (SIFT-MS). GC × GC-HRTOFMS has a high resolving power and offers a full overview of sample composition, providing deep insights on the ongoing biology. SIFT-MS is usually used for targeted analyses, allowing rapid classification of samples in defined groups. In this study, we used SIFT-MS in a possible untargeted full-scan mode, where it provides pattern-based classification capacity. We analyzed the exhaled breath of 50 asthmatic patients. Both techniques provided good classification accuracy (around 75%), similar to the efficiency of other clinical tools routinely used for asthma phenotyping. Moreover, our study provides useful information regarding the complementarity of the two techniques.

Figure 1

Study analytical design and patient population. Breath samples were collected using gas sampling bags. For GC × GC-HRTOFMS analyses, the bags were transferred onto thermal desorption tubes prior to injection. For SIFT-MS, the bags were directly deflated into the instrument. Data were analyzed using identical processing workflows and outcomes of GC × GC-HRTOFMS and SIFT-MS analyses were judged against each other to serve as a reference in the field. We observed that both approaches offered similar classification capacities.

Figure 2

Unsupervised PCA plot for SIFT-MS (left) and GC × GC-HRTOFMS (right) illustrating the absence of outliers or particular clustering trends.

Figure 3

Supervised classification model outcomes using the most significant features (above 0.0015 MDA) using Random Forest algorithm (Top: GC × GC-HRTOFMS; Bottom SIFT-MS). Receiver operating characteristic curves (left) show AUROC values of 73% and 87% for GC × GC-HRTOFMS and SIFT-MS, respectively. PCA score plots (right) depict the apparent differentiation between the eosinophilic phenotype and the others.

Figure 4

Data processing workflow for SIFT-MS and GC × GC-HRTOFMS. The same design was applied for both techniques, but specific pre-processing was conducted.