External Validation of Potential Breath Biomarkers for Asthma: A Step Forward Toward the Clinical Implementation of Breath Analysis

Asthma is a chronic respiratory disease characterized by airway inflammation with a high prevalence in both children and adults worldwide. Asthma has high rates of underdiagnosis and overdiagnosis, either of which could have a negative impact on the quality of life of patients and economic burden within health care and the wider society. The discovery of accurate biomarkers could enhance the process of confirming the diagnosis of asthma and help in the choice of an appropriate treatment (1). Breathomics, a branch of omics focused on exhaled breath research, has aroused great interest, and breath analysis has been emerged as a potential tool for the diagnosis and monitoring of many diseases, such as asthma. Human exhaled breath contains hundreds of volatile organic compounds (VOCs) that have shown promise as asthma biomarkers; exhaled breath collection can be performed in a noninvasive and cost-effective approach, and advances in high-throughput mass spectrometry (MS)–based technologies enable the successful detection of most VOCs in exhaled breath samples. Nevertheless, breath analysis has still not been introduced into daily clinical practice and has not yet overcome the biomarker discovery phase (2).

Technical standards established by scientific consensus (3) and the recent publication of rigorous systematic reviews (4–6), which have carefully assessed the studies focused on the search for asthma biomarkers in exhaled breath, have highlighted the challenges that urgently need to be faced to ensure the effective progress of breath analysis. Thus, issues addressed by well-established practices in other omics, such as metabolomics, were highlighted as gaps in breathomics research (4, 5). In the past few years, the scientific community has taken into account these recommendations, and several key milestones have been achieved; for instance, the recruitment of large cohorts (7, 8), the implementation of batch-effect-correction algorithms (9), the development of a workflow for the preprocessing of raw data from gas chromatography-MS (GC-MS) that involves functions from three R packages (10), and a mechanistic study to understand the interconnection between lipid peroxidation and VOCs that are detected in human exhaled breath (11), among others. However, the lack of external validation has remained a major bottleneck for the advancement of breathomics. For example when considering breath analysis by MS-based technologies, only Schleich and colleagues (7) had performed external validation of putative asthma biomarkers (specifically for neutrophilic and eosinophilic asthma) on a new and independent cohort, and very few studies had tested an asthma-related VOC profile on an independent set (validation set) that was not used for model training (8, 12). This issue of the Journal goes a long way to address this problem, including a concise translational review (13) and two original research articles with multicenter external validations (14, 15).

In this issue of the Journal, Brinkman and colleagues (pp. 1079–1090) show an updated review of achievements and pitfalls in breath analysis research, with particular emphasis on the identification of VOCs in exhaled breath that have real potential to be robust, validated biomarkers that can be used in the diagnosis, phenotyping, and treatment monitoring of prevalent diseases such as asthma (13). Knowledge of the metabolic sources of exhaled VOCs is essential for the correct choice of biomarkers. Therefore, Brinkman and colleagues (13) suggest complementing in vivo studies (VOC analysis in exhaled breath) with in vitro studies to clarify the metabolism of putative biomarkers. In addition, a good biomarker must be reproducibly measured in different laboratories. Therefore, minimizing confounding factors that can influence in VOC analysis is crucial to allow for multicenter comparisons. In this review, apart from traditional claims such as standardization of sampling and analytical methodologies, other strategies for reducing breath-specific confounders are mentioned. For example, suggestions are put forward, such as carrying out exhaustive studies to evaluate the influence of confounding factors on the composition of exhaled breath or the application of algorithms during data preprocessing and data analysis stages to identify the relevant confounders and to remove or reduce their effect on breath outcomes.

Also in this issue of the Journal, Shahbazi Khamas and colleagues (pp. 1091–1100) report a breathomics study that is focused on asthma control classification in a pediatric population (6 and 17 yr old) (14). In this study, they identified a VOC profile (acetophenone, ethylbenzene, and styrene) in exhaled breath that discriminates between children with controlled and uncontrolled asthma from the Systems Pharmacology Approach to Uncontrolled Pediatric Asthma (or, SysPharmaPediA) cohort (n = 100). VOC measurement was performed by GC-MS, and data modeling was conducted by sparse partial least square discriminant analysis. Not only were the VOC profile and model constructed validated with good performance on a test set within the SysPharmaPediA cohort, but they were also validated in two independent cohorts, U-BIOPRED (n = 49) and PANDA (n = 47). As a result, a rigorous process was performed to reduce the risk of model overfitting. In addition, the batch effect was corrected using bioinformatics tools, and the potential bias associated to other factors such as second-hand smoking was evaluated. However, as mentioned by the authors, further research is needed to establish the origin of the selected VOCs and their relationship with human metabolism.

Finally, in this issue of the Journal, Peltrini and colleagues (pp. 1101–1112) have found a VOC profile (19 VOCs) associated with eosinophilic airway inflammation in adults with asthma (15). This comprehensive study is an example of the combination of in vitro and in vivo approaches. Thus, first, these 19 VOCs were selected by VOC analysis of sputum samples (headspace analysis) from subjects with severe asthma (n = 36), and subsequently, the VOC profile was validated on the exhaled breath samples of subjects with asthma from two independent cohorts, the EMBER cohort (n = 65) and the U-BIOPRED cohort (n = 42). The models that were based on this VOC profile obtained excellent performance in the replication cohorts. Samples were analyzed by different types of GC-MS, and elastic net regression was used for model building. Elastic net regression, such as least absolute shrinkage and selection operator logistic regression, is a robust statistical method that allows the generation of models where the weight of each variable is easy to interpret, which is important for the clinical applicability of potential biomarkers (8). In addition, Peltrini and colleagues (15) have tried to draw parallels between the 19 selected VOCs and human metabolic pathways, as well as between the findings of previous studies. Furthermore, in an online supplement, the methods conducted in each step of breath analysis are described in detail, which is welcome, as it is not often found in breathomics studies and may bring consistency to the field.

The recommendations and findings shown in these papers paves the way for advancing breathomics toward clinical translation. However, despite these positive steps, it is essential to conduct further research on the weaknesses of this technique. Further external validations in larger cohorts of subjects, exhaustive studies focused on the assessment of metabolic coverage linked to exhaled VOCs, studies with well-controlled sampling conditions (monitoring of the human respiratory cycle during breath sampling, assessment of the influence of environmental air, etc.), the common use of chemical standards for more accurate quantification and identification of VOCs, the widespread use of open-source workflows for data preprocessing and of robust algorithm for data analysis, and the establishment of large breathomics databases are strongly recommended. Applied research testing and validating standards for collection in practice and linking results to specific clinical features or decision problems will also be vital before routine use. These efforts for the sake of transparency and reproducibility will help further international cooperation, as in the studies of Shahbazi Khamas and colleagues (14) and Peltrini and colleagues (15), to ensure the progress of breath analysis.