Fulfilling the Promise of Breathomics: Considerations for the Discovery and Validation of Exhaled Volatile Biomarkers

Abstract

The exhaled breath represents an ideal matrix for noninvasive biomarker discovery, and exhaled metabolomics have the potential to be clinically useful in the era of precision medicine. In this concise translational review, we specifically address volatile organic compounds in the breath, with a view toward fulfilling the promise of these as actionable biomarkers, in particular, for lung diseases. We review the literature paying attention to seminal work linked to key milestones in breath research; discuss potential applications for breath biomarkers across disease areas and healthcare systems, including the perspectives of industry; and outline critical aspects of study design that will need to be considered for any pivotal research going forward if breath analysis is to provide robust validated biomarkers that meet the requirements for future clinical implementation.

The exhaled breath gives access to a multifaceted and (almost) limitless reservoir that comprises biological materials arising from the airway and beyond. As such, the breath potentially represents the ideal medium for biomarker discovery.

Breath comprises three main components that may provide biomarkers: 1) the gas phase, which, in addition to the major constituents (nitrogen, oxygen, and carbon dioxide), comprises a few trace inorganic compounds (e.g., hydrogen, carbon monoxide, ammonia, and nitric oxide) and thousands of trace volatile organic compounds (VOCs) (1); 2) water vapor, which can be collected as exhaled breath condensate and traps both volatile and nonvolatile water-soluble molecules (2); and 3) exhaled breath particles, which require a specific sampling maneuver but provide access to larger molecules arising from airway lining fluid such as surfactant and albumin (3).

In this concise translational review, we specifically address VOCs in the breath with a view toward fulfilling the promise of these as ideal biomarkers, in particular, for lung diseases. We will review the literature paying attention to seminal research linked to key milestones; discuss potential applications for breath biomarkers across disease areas and healthcare systems, including the perspectives of industry; and outline critical aspects of study design that will need to be considered for any pivotal research going forward if breath analysis is to provide robust validated biomarkers that meet the requirements for future clinical implementation (Table 1).

Table 1.

Building on Existing Strengths and Tackling Roadblocks in Breath Analysis

Step	Description	Impact on Breath Analysis	Next Steps
Diagnostic definition	The reference standard used to characterize the subject groups or health states between which a test should differentiate	VOC analysis is focused on detecting biological signals. Clinical disease definitions often are imperfect representations of biology, making them imperfect gold standards	Expand understanding of biological origins of VOCs. Acknowledge potential impact on biomarker selection and test performance of imperfect gold standard in diagnostic and monitoring studies
Biological variability	Main biological variability types are: 1.Disease related, 2.Within-person homeokinesis of disease signal, and 3.Within-disease differences	Test focuses on detecting target condition-linked signal (1). Types 2 & 3 biological variability complicate this detection by reducing signal-to-noise ratio. Impact will be largest on biomarker signals relying on detection of concentration differences rather than unique biomarkers	Expand understanding of biological origins of VOCs and capacity of in vitro model systems enabling VOC analysis. Expand capabilities of VOC probe approaches that amplify signal and reduce noise of biological variability
Physiological variability	Variability between and within individuals with respect to their breathing patterns, lung perfusion, vital capacity, and dead space ventilation alter VOC concentrations on breath	Differences may reflect the presence or absence of disease but equally may confound the signal or decrease the signal-to-noise ratio. Impact will be largest on biomarker signals relying on detection of concentration differences rather than unique biomarkers	Expand understanding of impact of lung physiology on breath biomarkers. By increasing understanding of these processes, monitoring tests could be developed and biomarker signals can be corrected. In an ideal scenario, the equivalent of a housekeeping gene is identified, enabling standardization of VOC signals
Sampling	Method of collecting a breath sample, including the volume, sampled fraction, and speed of collection; this includes repeatability and reproducibility of technique as well as preferential sampling of specific chemical groups	Breath collection parameters will directly impact (relative) VOC concentration. Increased volume will increase total VOC concentration if sorbent tubes are used. Selection of different fractions impacts relative concentrations of different VOCs. Collecting over longer periods of time may reduce variability between breath samples Sampling methods could enrich for specific chemical attributes (hydrophilic/hydrophobic) but also may contribute specific VOCs to the sampled mixture. Different technologies will vary with respect to their reproducibility and repeatability. This impacts the signal-to-noise ratio and ability to detect biomarkers directly	Increased emphasis must be placed on understanding performance of different breath collection approaches and their suitability to sample VOCs Studies of diagnostic accuracy should include repeat samples with the same setup to assess repeatability and use different collection devices in the same subject to assess reproducibility By understanding the VOC of interest and/or using an eVOC approach, collection techniques can be optimized for the target molecule
Storage and transportation	Type of VOC storage method used if not analyzing sample directly	Different storage and transport methods result in differences in VOC stability (bag vs. sorbent tubes) and will be selective toward different chemical attributes of VOCs. Storage methods can add compounds to the VOC mixture, and VOCs may degrade over time in storage vessels. Storage temperature and transportation conditions (e.g., vibration, shocks) may impact compound stability. Storage methods must align well with the analytical method to minimize impact on study results	On identification of compounds of interest for a specific application, the impact of storage on their stability needs to be assessed. This should result in recommendations with respect to maximum storage time and transport conditions
VOC analysis	Method to convert VOCs into analyzable data signals; this is generally done by means of analytical methods such as GC-MS or sensor-based approaches (eNose)	All analytical methods for VOCs are known to suffer from drift over time and require some type of calibration. For GC-MS, this can be done using chemical standards. As a representative breath mixture isn’t readily available, calibration of eNose devices is particularly challenging. Second, analytical techniques are known to have between device differences. Calibration approaches can be used to map between devices but this inevitably results in some signal loss	Studies should integrate means of tracking drift over time to assess the potential confounding effect this has on study results and translatability to clinical practice. This will likely require development of novel methodology that is specifically geared toward sensor technologies. Care should be taken during study design and conduct that samples from cases and controls are not separated in time or run consistently on different devices, as this severely increases the risk of false-positive findings. Where possible inter- and intradevice repeatability should be assessed as part of any study. On identifying (a set of) target compounds of interest, calibration mixtures can be developed, and sensors can be attuned to specific VOCs
Data analysis	Methodology used to convert raw output from VOC analysis into classifiers	Challenges particularly exist if a wide range of compounds are evaluated or if a pattern recognition (eNose) approach is used. Those challenges are common to all omics approaches and mostly revolve around this risk of false-positive and false-negative findings. Resource is wasted if wrong decisions are made on the basis of erroneous results	All studies should be appropriately powered and have separate training and validation sets. Site, device, and time biases should be part of standard reporting

Definition of abbreviations: eVOC = exogenous volatile organic compound; GC = gas chromatography; MS = mass spectrometry; VOC = volatile organic compound.

The History of Breath Research and Landmark Discoveries

In the wake of the Human Genome Project in the 1990s, postgenomic technologies such as metabolomics paved the way for breath VOC profiling. As a result of such technological key advances, alongside international collaboration initiatives, breath analysis has become accessible for clinical studies aimed at the discovery of disease-specific “breathprints” (Figure 1).

Figure 1.

Timeline depicting key advances toward clinical application of breath volatiles. ERS = European Respiratory Society; FDA = Food and Drug Administration; GC = gas chromatography; MS = mass spectrometry; PTR = proton-transfer reaction; SIFT = selected-ion flow tube; VOCs = volatile organic compounds. Created with BioRender.com.

Breath research is conducted across a wide number of conditions (4), of which chronic respiratory diseases is an obvious theme, considering the direct involvement of the lungs. In the following narrative synopsis, we aim to illustrate how breath researchers are improving their understanding of clinical utility and the requirements for robust biomarkers, and approaching potential confounders.

In 2009, breath VOC profiling was applied to classify patients with chronic airway diseases versus healthy control subjects (5). Soon after this, the same concept was extended toward differentiation of disease phenotypes, airway inflammation, asthma diagnosis, disease control, and prediction of acute exacerbations (6–8). Differentially expressed volatiles in neutrophilic and eosinophilic asthma were explored in external populations (9, 10) and in vitro (11). In terms of stability, it was reported that some key breath metabolites have physiological circadian variability, whereas others were only variable in individuals with asthma. This finding highlights the importance of controlling for timing of sampling and provides potential mechanistic insight into the disease (12) and should be taken into account when designing new studies on the application of exhaled VOCs for monitoring of airway diseases (13).

Notably, breath biomarker discovery also provides opportunities for gaining mechanistic insight into the physiology and biological processes of the diseases of interest. For instance, the distinct metabolism of infective microbes and their production of microbial VOCs, can be studied and validated in vitro (14–20). To unravel the physiology behind disease-related exhaled VOCs, there have been attempts to classify them with respect to their “fat-to-blood” and “blood-to-air” partition coefficients. These partition coefficients provide an estimation of the relative concentrations of VOCs in alveolar breath (21). For the exploration of biological mechanisms, differential breath profiles can be induced using VOC-based isotope probes (22, 23). Recent advances have also identified the potential utility of using an externally administered volatile substrate as a probe. These exogenous VOC probes may be specifically engineered to be metabolized by a pathway of interest (24); and, therefore, the resulting metabolites have the potential to elicit more specific breath signals for the disease of interest. However, this approach relies on an a priori understanding of the underlying biology and requires clinical trials in human subjects.

Over the past 10 years, there has been an increase in the number of large-scale, multisite, clinical initiatives that integrate breath analysis into clinical phenotyping studies, such as the U-BIOPRED consortium, the UK EMBER trial (25), and the interoperability framework introduced by Gisler and colleagues (26). For example, Wilde and colleagues published a comprehensive method optimization (27), followed by a large cardiorespiratory breath biomarker study reporting several unique signatures of cardiorespiratory exacerbations (28). The analysis pipeline was optimized and focused on the discovery of markers to stratify patients with acute breathlessness. The published methodological considerations within these platforms and recommendations relating to quality assurance procedures are valuable in informing future study designs (29). Another large-scale international collaboration is the Human Breath Atlas, a project that aims to build a large-scale breath biobank at the epidemiological level, through which biomolecules in breaths may be reliably detected and mapped to health status in the future (30).

The Promise: The Exhaled Volatiles and the Potential Clinical Applications

The complex relationships between genes, molecular pathways, and phenotypes make it challenging to understand diseases, their causes, and potential treatments from genetics alone. With the maturation of the omics fields, there is an opportunity for the development of clinically useful biomarkers, which can drive advances in precision medicine. Exhaled VOCs reflect metabolic activity (31, 32) and, therefore, may provide a noninvasive source of biomarkers to diagnose, phenotype and predict diseases and treatment responses. Because of this potential, the application of VOC-based breath analysis for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection screening and diagnosis has been investigated (33–35). Longitudinal sampling also allows disease monitoring and prediction of disease remission and progression, resistance to therapies, and adverse effects. Nevertheless, understanding the links between the origins of the breath metabolites and the underlying biological process and clinical phenotypes holds the key for future clinical application.

Although some breath VOCs are tertiary and quaternary metabolites of known metabolic pathways, the origins of most remain unknown. It is likely that many of these biomolecules arise from metabolism in the lungs, generated by regulatory processes or microorganisms. However, VOCs from distant organs may also diffuse into the alveoli from the systemic circulation. Exhaled volatiles originating within the host (endogenous) are thought to be reflective of underlying metabolic processes and, therefore, are potential biomarkers of interest. However, breath VOCs can be also from diet, drugs, or environment (exogenous) (36, 37).

Study Design Considerations

Breath-Specific Confounders

Although all the factors that can influence a breath measurement are important, breath-specific confounders (e.g., materials used or air supply) can present a barrier to the discovery of reproducible biomarkers and, therefore, require careful consideration in experimental design and analysis (38).

Endogenous Factors

Distinguishing between endogenous and exogenous VOCs is confounded by the significant overlap of compounds produced in the body and those produced and present in our environment. Furthermore, the endogenous chemical composition of human breath represents metabolic contributions from the host-microbial “superorganism,” with an unknown proportion of exhaled VOCs originating from host microbial communities, such as from the gut (39, 40), mouth (41), and upper and lower respiratory tract (14), as well as their interaction and the effect of any antimicrobial interventions (42) (Figure 2).

Figure 2.

Concept map of breath-specific confounders influencing the composition of volatile organic compounds.

Endogenous VOCs can be subcategorized into those originating directly from the lung and systemic VOCs. Sources associated with the lung include mucus, the airway wall, and lung parenchyma. Systemic VOCs transition from blood to breath during gas exchange, having been produced in distant organs. VOCs are products of healthy metabolism released by cells throughout the body. Diseases manifest as perturbations from this natural metabolic flux, with certain groups of VOCs increasing or decreasing in concentration. For example, branched hydrocarbons (often referred to as methyl-substituted alkanes) represent an abundant chemical class in breath and are chemical endpoints of lipid peroxidation, becoming upregulated as a result of oxidant stress, as observed in diabetes mellitus (43), preeclampsia (44), Alzheimer’s disease (45), lung and breast cancer (46, 47), and aging (48).

The relative concentration of a specific chemical or chemical class is not only a consequence of the rate of production but also dependent on partition properties (e.g., blood:breath ratio, blood-fat partition coefficient) determined by functional group (21, 49); host physiology (e.g., exhalation rate, exercise, posture) (50); and biotransformation, for example, in the liver (51, 52).

Exogenous Factors

The sources of exogenous VOCs include indoor and outdoor air, drugs, diet, personal care products, smoking habits, and sampling materials. Some of the most abundant endogenous VOCs have overlapping exogenous sources, such as acetone (a commonly used solvent) and isoprene (widely emitted from plants and the most abundant biological VOC in the atmosphere) (53). However, confounding factors influencing VOCs that are exclusively exogenous are still important to consider (54). Beauchamp highlighted this importance when outlining the utility of breath analysis for pharmacokinetic and exposure studies, using limonene as an exemplar (55). Limonene is an exogenous VOC with multiple sources, including indoor emissions, hygiene products, and food. Therefore, the concentration of limonene in breath is not only related to its inhaled concentration but also to gut absorption and liver metabolism, and a simple subtraction of its concentration in the air at the time of sampling would fail to reflect all these aspects (24).

Methods for Minimizing Confounders

The solutions for overcoming the barriers for breath-specific confounders include improved data-processing strategies, sampling standardization, confounder-specific studies, and complementary in vitro studies. Statistical algorithms are often used for identifying confounding factors present in breath data. Signals correlating with known biases such as batch effects or medication can be removed through normalization (56, 57). Specific statistical methods used are often study dependent (e.g., sample size, number of features, duration of study), but for a rigorous assessment of confounding factors, more than one approach may be necessary. Although an optimal method to identify, minimize, and adjust for exogenous components has yet to be found, tactics include filtering the inspired air, measuring the VOCs in the room at the time of sampling, and asking the patient not to eat or drink for a set period before the measurement. Studies quantifying the magnitude of impact of potential confounders are essential for assessing the relevance of controlling such factors in discovery studies and identifying sources of variation. Where possible, clinical studies should seek to identify the impact of relevant confounders. Although, for biomarker discovery studies, more rigid controls may be appropriate, in validation studies, it may not be, especially if this requires a trial design that is too far removed from the desired implementation scenario (and thus undermine the generalizability of the results).

Efforts in standardization are important for reducing variation due to sampling artifacts that can be introduced directly from the materials used or from inconsistent sampling practice. VOCs are emitted from masks, tubing, bag samplers, and sorbent-based materials; therefore, it is crucial to be consistent in their use across multisite studies. It is also important to monitor their VOC contribution through the collection of routine procedural blanks (e.g., sorbent tubes) and/or to minimize their contribution by degrading with thermal conditioning or surface cleaning. As noted previously, performing relevant in vitro studies alongside the analysis of breath samples collected in vivo (58) can lead to better understanding of the origins of endogenous compounds.

Biological Validation of Breath Volatile Discoveries

Although associations observed in clinical studies can provide “proof-of-concept,” understanding the source of any putative VOC biomarker and whether it is linked to the disease phenotype of interest is important for biological validation and may also help elucidate target pathways for therapeutic interventions.

Endogenous VOCs can be difficult to directly link to metabolic pathways without metabolic flux studies, as they merely represent correlations. Furthermore, VOCs may be part of multiple metabolic processes that result in high intercorrelation, making it difficult to develop disease-specific, single-biomarker assays. There has been progress: The work of Ratcliffe and colleagues established the theoretical products that can result from lipid peroxidation (59), and others have developed systems for sampling breath from mice (60–62), paving the way toward mechanistic understanding.

Biological systems are dynamic; therefore, a deeper understanding is required to interpret the complex associations between in vitro and in vivo findings. Biological validation of biomarkers reduces the risk of false-positive interpretation when a signal is influenced by confounders. In addition, knowledge of biomarker metabolism also reduces uncertainty of whether compounds are directly linked to disease pathogenesis or host response to clinical interventions.

Several research groups have developed in vitro models to further understand the role and function of biogenic VOCs. Examples of cell models include mammalian cell culture with an air–liquid interface (63, 64), culturing microbes in artificial sputum (65), and isolating leukocytes from patient blood or in native sputum headspace (11, 66, 67). In vitro models are also important in establishing thresholds and compound recovery from complex systems such as a ventilator or mouse models (68, 69). It is important to investigate the influence of the surrounding environment, as biological systems, such as facultative anaerobic bacteria, may adapt their nutritional requirements and metabolism.

We acknowledge that these models have limitations, but they also represent an important tool in the evaluation of candidate biomarkers, providing the tools to evaluate pathways associated with VOC production. This provides value in several ways. It will provide assurance to progress biomarkers to (more expensive and complex) validation studies by confirming their association with biological mechanisms that are relevant to the target disease. Further, the inverse is also true: Identifying that differentiating compounds are of nonbiological origin (e.g., hospital cleaning agents) helps minimize false biomarker discovery. Understanding biological pathways can also help develop targeted probing strategies and analytical techniques that increase the signal-to-noise ratio, thereby improving the diagnostic potential of the candidate biomarkers.

Technical Considerations

The technologies underpinning exhaled breath analysis are diverse, and each analytical approach offers a unique set of advantages and disadvantages depending on the clinical application and target analytes (Table 2). The analytical benefits and limitations of specific instrumentations, devices, and methods have already outlined in detail; for example, for offline sample collection methods (70) or chromatographic and spectrometric techniques versus sensor-based platforms (71, 72). The choices regarding analytical method selection are considered herein with a focus on suitability for 1) discovering signatures and understanding the source of discriminatory VOCs and 2) the successful implementation of a future point-of-care device (postdiscovery).

Table 2.

A Summary of Advantages and Limitations of the Most Common VOC Analysis Methods and Sampling Techniques with Best Practice Methodology Suggestions and Clinical Considerations

Type of Method	Advantages	Disadvantages	Best Practice Methodology	Considerations for Clinical Application
Analytical method Offline methods
GC-MS	Quantitative, well established identification of a wide range of compounds; accurate detection; reliable; robust	Offline; no structural confirmation of compounds; not suitable for temperature-sensitive compounds; frequent maintenance needed; peak shifts occur because of column aging	Preconcentrating samples; calibration of sample sets using authentic standards; sampling background; and potentially excluding background related peaks	Samples need to be sent to a centralized lab for campaigns. The Human Breath Peppermint Consortium has developed analytical methods for cross-lab calibration of VOCs
GC×GC	Increased peak capacity; comprehensive analysis of multiple classes of analytes; improved resolution and detectability limits; small sample size	Offline; long sample-processing time; advanced data analysis needed	Periodic calibration needed; best practice methodology, as per GC-MS	As per GC-MS, clinical programs require input from analytics that are trained in relevant chemometrics for GCxGC
GC-IMS	Positive and negative ionization allowing detection of a wide variety of compounds; short sample running time; isothermal temperature; can be used onsite for direct sampling; potentially point of care	Ion mobility spectral libraries are still evolving	As per manufacturer guidance for specific devices	Suitable for point of care and likely most useful once a candidate set of VOCs have been identified from discovery campaigns
GC-Orbitrap	Increased annotation of a wide variety of compounds; robust identification; and quantification and potential structural elucidation	Offline; advanced data analysis still under development	N/A	Lab-grade technique is not suitable for large-scale clinical programs.
Online methods
SIFT-MS	Online ambient ionization; quantitative; high sensitivity; ability to analyze high-humidity samples; low maintenance	Background subtraction needed to eliminate ambient compounds	N/A	This is suitable only for the monitoring of known targets
eNose	Low cost; portable	Low precision at low concentrations; cannot be used in discovery because of low sensitivity	N/A	This is suitable for point-of-care studies. Sensor array elements need to be carefully optimized for the VOC of interest
SESI-MS	Online ambient ionization; real-time analysis; can detect a very wide range of compounds; allows detection of positively and negatively charged ions; low limits of detection are achievable without any sample preconcentration	Background subtraction needed to eliminate ambient compounds	Minimize environmental background	This is highly sensitive and can be used for discovery
PTR-ToF-MS	Online ambient ionization; sensitive to very low concentrations	Limitations in accurate quantitative analysis; nonlinear response of ToF-MS limits quantitative abilities at low concentrations; incomplete database	N/A	This is suitable only for the monitoring of known targets
Sampling techniques
Sampling bags	Easy to use; inexpensive; widely used	Release of contaminants; high background signals; chemical stability of VOC may be compromised with storage; single use only; high permeation of CO2	Fill the bag to at least 80%. Use tubing for direct bag filling to avoid contamination Fill at 3 L/min. Store protected from light at room temperature for no more than 48 h	Despite potential limitation, this is advantageous for large-scale clinical programs. Impact of Tedlar on VOC biomarkers of interest should be defined in non-clinical analytical studies before clinical campaigns
ReCIVA TD tubes	Highly sensitive and lower detection limits of volatiles; suitable for temperature used in GC analysis; versatile for different analytical methods; allows the selective concentration of target analytes	Sample concentration needed; long sampling time (around 15 min); preconditioning of TD tubes required before use	Blanks need to be used Condition newly packed tubes for at least 2 h Store in refrigerator	Significant training is required for staff to deliver campaigns with ReCIVA Cost of equipment may be prohibitive for multicenter campaigns
BCA TD tubes	Highly sensitive and lower detection limits of volatiles	Preconditioning of TD tubes required before use	Two separate sorbent traps capture samples of alveolar breath and room air	Cost of equipment may be prohibitive for multicenter campaigns

Definition of abbreviations: BCA = breath collection apparatus; GC = gas chromatography; GC×GC = multidimensional gas chromatography; IMS = ion mobility spectrometry; MS = mass spectrometry; N/A = not applicable; PTR = proton-transfer reaction; SESI = secondary electrospray ionization; SIFT = selected-ion flow tube; TD = thermal desorption; ToF = time of flight; VOC = volatile organic compound.

Sampling in discovery studies typically involves the collection of a prespecified volume (often 0.5–1 L) of breath, with a preconcentration of VOCs onto sorbent material, to maximize the recovery of VOCs at trace concentrations (down to parts per trillion). The sorbent material can consist of up to three sorbent types for capturing a broader range of VOCs that exhibit different absorption characteristics for minimizing selectivity bias. The efficiency of sorbent materials varies and needs to be considered if measuring a predefined VOCs. Characteristics to take into account concern, in particular, the volatility and polarity of a VOC (73). Breath collection is made either by trapping one to five exhalations in a chemically inert bag (made of, e.g., Tedlar, layered foil, or polyterafluoroethylene), followed by active pumping of the VOC-rich air through sorbent material, or by using a sampling device for automatically trapping breath VOCs on the sorbent collection apparatus (e.g., ReCIVA, Owlstone Medical) (74) or breath collection apparatus (Menssana Systems (75). Whatever the method of breath collection, decoupling the sampling procedure from the analytical measurement (referred to as “offline” analysis) provides unique advantages. Sampling can focus on ease of collection and patient acceptability while allowing for analysis with complex, lab-based analytical equipment and time-consuming data processing.

The amount and type of VOCs that are detected in a breath sample can vary greatly depending on whether the collection is of alveolar breath or whole breath. When whole-breath sampling is done, the sample may contain more highly soluble VOCs, whereas alveolar sampling allows the detection of more low-blood soluble VOCs, as gas exchange occurs mainly in the alveolar region (76).

Discovery technologies typically consist of a chromatographic separation coupled with mass spectrometric detection such as gas chromatography (GS)–mass spectrometry (MS). However, with the increasing accessibility of analytical hardware, more advanced technologies such as multidimensional GC and techniques coupled with high-resolution MS are redefining the gold standard in breath analysis (27, 77). By combining GC and MS, these instruments afford structural identification with increased confidence in chemical assignment alongside accurate semiquantification. Rigorous characterization of VOCs is the first step toward the annotation of pathways. Therefore, these platforms provide the high-fidelity data required for the discovery of VOC signatures. However, they are time-consuming, are expensive to operate and maintain, and require extensive analytical validation. Future technologies exploring ambient ionization and GC-independent online analysis of primary metabolites in breath are eagerly awaited but currently suffer from a lack of extensive compound libraries (78).

Consequently, once a breath biomarker or signature has been discovered (and validated), techniques that are most suited for the successful implementation of a point-of-care device include those developed as an all-in-one solution (sampling, detection, and analysis), which often is the case for ion mobility spectrometry (IMS) (72) or eNoses (79). eNoses are relatively cheap to produce and can be designed and trained for the detection or monitoring of specific VOC-based signatures. It is anticipated that the currently suboptimal selectivity and sensitivity of the chemical sensors embedded in eNoses will be optimized to allow the development of “tailor-made” eNoses (80). Techniques such as IMS or GC-IMS represent a middle ground (72, 78). Although they are relatively costly and require dedicated technical support, they are highly sensitive and capable of separating target compounds within complex VOC profiles at a low parts-per-billion level (34).

Deploying a point-of-care test is not without its challenges, as the equipment needs to be cost-effective, needs to be subjected to regulatory approvals, and requires robust calibration methods. To date, no sensor-based technologies have managed to meet such requirements while maintaining sufficient accuracy. Nevertheless, with better understanding of target compounds, this may become a possibility.

Data Preprocessing, Analysis, and Metabolite Reporting

A typical workflow for handling breath data consists of 1) preprocessing, 2) modeling and visualization, and 3) compound identification. Different data preprocessing procedures exist, of which most consist of similar substeps. In this regard, in recent years, some open-source workflows for data preprocessing have been developed (81–84).

First, it is vital to remove noise and to transform the raw analytical signal into a computationally‐usable format (81, 84). During the next preprocessing step, the analytical signal needs to be converted from an instrumental response to a numerical value; the abundance of the signal for that compound is proportional to its concentration in breath and can be converted from a relative abundance to an absolute unit using a calibration curve (85). Peak detection or deconvolution can be applied, as well as data alignment. Each feature or compound detected will have an abundance (e.g., integrated peak area) and is assigned one or more unique identifiers (e.g., drift time, sensor output, retention time or mass‐to‐charge ratio, and library match). Data alignment ensures the same breath VOC (or “feature”) is identified across all samples, which is critical for curating the multivariate dataset and ensuring that each feature is unique and comparable across all samples (84).

The generation of a data matrix permits interrogation of the dataset using a range of tools from multivariate statistics, chemometrics, machine learning, and artificial intelligence for the identification of discriminatory biomarkers and signatures. The first step is often data normalization and scaling to overcome the large dynamic range of breath VOC concentrations. For breath data generated using discovery platforms such as multidimensional GC–time-of-flight MS and IMS, the number of variables often far exceeds the number of observations. Therefore, a key step in any discovery workflow is data reduction. Principal component analysis is a typical reduction technique. However, there is a risk of significant data loss where it becomes detrimental to the discriminatory power and wealth of phenotypic information originally afforded by measuring the metabolome. For data modeling, widely used techniques are partial least squares discriminant analysis, multinomial regression analysis, and random forest, among others (83, 84). Methods of visualization ranging from receiver operating characteristic curves, to clustering and network analysis used to support the output of the model building and model validation (81, 83). As metabolomics data typically have a large number of variables derived from much smaller sample sizes, statistical analysis based on one method may lead to overfitting (83, 86). Using multiple methods to derive reproducible selections of compounds, with internal and external validation, are paramount to maximize the validity of findings (86, 87).

As breath is a complex mixture (88), compound identification is often left until the end to avoid the time and expense of attempting to identify and quantify every metabolite detected. Once a concatenated list of discriminatory breath volatiles has been extruded from the data matrix, compounds can be identified by using their analytical identifiers (e.g., drift time, sensor output, retention time or mass-to-charge ratio, library match) or by comparison with a reference standard. The level of confidence in the assignment of a chemical structure and name can be classified on the basis of Metabolomics Standard Initiative identification levels (89).

The Future of Exhaled Volatiles: Bridging the Gaps and Fulfilling the Promise

Critical to the future clinical application of exhaled VOCs, the first step is the identification of reliable and clinically useful biomarkers and the subsequent establishment of feasible analytical technologies and evaluation in the intended populations at clinically relevant settings (Figure 3). Fulfilling the promise of exhaled volatiles in precision medicine requires a collaborative effort with expertise from clinical medicine, biology, analytical chemistry, bioinformatics, engineering and health economics, and industry and regulatory bodies. Study of the discovery, validation, and subsequent integration of fractional exhaled nitric oxide into clinical practice provides a guide to the possibilities and pitfalls that precede widespread clinical adoption of breath tests. Bridging the gaps between the discovery of VOC signatures and their clinical implementation, future research investment must focus on performance (with differing cutoffs by application and confounding clinical factors, including age, atopy, smoking, physiological deadspace, and sampling flow rate), standardization of sampling and analytical methodologies, accurate quantification, addressing environmental contamination, and clinical validation in large patient cohorts. Finally, the accuracy and efficiency for its intended use, the feasibility of application, and the cost-effectiveness of clinical implementation in the healthcare settings should be considered and refined at every stage of breath research.

Figure 3.

Current status of breath sampling and analytical technologies using the technology readiness level (TRL) assessment method (adapted from https://www.gov.uk/government/news/guidance-on-technology-readiness-levels). Clinically validated biomarkers are presented as comparative models for airway disease stratification. Online versus offline methods are color coded as orange versus red, respectively. Fe_NO = fractional exhaled nitric oxide; GC = gas chromatography; IMS = ion mobility spectrometry; MS = mass spectrometry; PTR = proton-transfer reaction; SESI = secondary electrospray ionization; SICRIT = soft ionization by chemical reaction in transfer; SIFT = selected-ion flow tube; TD = thermal desorption; TOF = time of flight.