We have shown that this technique for breath collection paired with metabolomic VOC analysis by GC-ToF-MS is able to classify COPD from healthy controls with moderate accuracy. However we were able to demonstrate much greater accuracy when looking at sub-groups of clinical interest such as smokers with COPD versus asymptomatic smokers, and COPD subjects with sputum eosinophilia, or those liable to suffer frequent exacerbations. If validated in prospective cohorts, this technique may provide a non-invasive method for phenotyping COPD in the future, with clinical applications for example in personalised therapeutics and prognosis.
Recent data have also shown discrimination between COPD and healthy controls using GC-MS, but based upon analysis of single breath samples . Current work in metabolomic analysis of exhaled VOCs is exploratory and aimed at discovering potential novel biomarkers, so it is critical that the sensitivity of the detection system be optimised. Each of our breath samples contains the VOCs absorbed from 3 litres of late-expiratory air, typically representing 50 to 100 breaths per subject. Paired with GC-ToF-MS, we have a highly sensitive methodology for both absorption and detection of VOCs that may well be present in minute concentrations in a single breath. It is of interest that the classification model developed in the Van Berkel study had higher accuracy for discriminating COPD versus controls than ours. Although the breath collection methodologies differed, the analytical techniques, based on GC-ToF-MS, were similar. It may be that the large difference in demographics between the COPD and healthy cohorts in that study (the COPD group were 21 years older, and 76% were current smokers) contributed to the differences seen. Likewise the close matching of current smoking status in our cohorts may have contributed to the moderate success of our classification model, as active smoking is clearly likely to be a dominant confounder in exhaled breath analysis where data from both smokers and non-smokers are analysed together. Indeed when we looked only at smokers with and without COPD (thus neutralising this confounding effect in the analysis), we found our model classified disease with far greater accuracy. The scatter plot (Figure 2) shows the COPD data to be clustered in one region of (but still within) the data from the healthy controls. This is consistent with the concept of COPD as a disease of accelerated lung aging , and it is perhaps not surprising that the metabolomic profile of patients with this disease clusters at one extreme of normality, rather than apart from it.
Sputum eosinophilia predicts steroid-responsiveness in COPD , but the test is labour-intensive, time consuming, and samples are not obtainable in a significant minority of patients . Whilst our technique for breath collection and analysis is currently relatively high-cost and labour-intensive, if a set of candidate biomarkers were validated for predicting steroid responsiveness, work would focus on developing and producing small, user-friendly point-of-care sensors specifically for this purpose. We have previously shown that breath VOC profiles can also predict sputum eosinophilia in asthma . Unsurprisingly, given the differences in demographics and disease processes between the studies, the specific VOCs used in the models were not the same. The identification of VOCs patterns specific to sputum inflammatory profile, and phenotypes such as “frequent exacerbators” may not only provide biomarkers for clinical use, but also could potentially provide new insights in disease pathophysiology.
Whilst absolute control of environmental VOCs is practically impossible, it is desirable to reduce background levels as much as is practicable. We use a VOC-filter in our circuit, an equilibration time of at least five minutes, and collect samples in the same room for this purpose. Even so, exogenous VOCs may be differentially handled by the airways in health and disease, and alteration in exhaled concentrations may therefore be relevant. Examples include: low-molecular weight molecules being absorbed by excessive airway secretions more readily than heavier molecules; the systemic circulation, acting as a reservoir, re-releasing environmental VOCs into the air at rates determined by cardiac output and lung circulation ; airway and alveolar inflammatory processes metabolising inhaled VOCs for example by oxidation; and the air trapping seen in obstructive lung disease altering the washout time for gas-phase molecules compared to healthy lungs. One approach to correcting for “exogenous” VOCs is to subtract the content of a contemporaneous environmental sample from the expired sample , but this oversimplifies the metabolic and physiological impact of the airways and circulation.
Our findings require validation in an independent group of subjects then definitive compound identification and calibration curves determined by injection of known standards. The origins of these compounds are as yet unknown, but hypotheses can be generated. For example six of the 11 compounds discriminating COPD from health were aldehydes, and all had strong loading onto the first principal component, an interesting finding in line with the findings of Van Berkel et al.. It may be that the metabolic upregulation in the mucosa of COPD patients removes aldehydes from the air; it is known for example that the aldehyde scavengers N-acetylcysteine and glutathione monoethyl ester completely remove unsaturated (but not saturated) aldehydes from a cigarette smoke extract . There may also be an effect of ICS on suppressing exhaled aldehyde levels, supported by the contribution of undecanal to both models. Further, it may be instructive in future studies to compare these putative exhaled markers of inflammation to existing disease-relevant breath biomarkers such as leukotriene B4 and other eicosanoids .