Breathomics: Review of Sample Collection and Analysis, Data Modeling and Clinical Applications

Abstract

Metabolomics research is rapidly gaining momentum in disease diagnosis, on top of other Omics technologies. Breathomics, as a branch of metabolomics is developing in various frontiers, for early and noninvasive monitoring of disease. This review starts with a brief introduction to metabolomics and breathomics. A number of important technical issues in exhaled breath collection and factors affecting the sampling procedures are presented. We review the recent progress in metabolomics approaches and a summary of their applications on the respiratory and non-respiratory diseases investigated by breath analysis. Recent reports on breathomics studies retrieved from Scopus and Pubmed were reviewed in this work. We conclude that analyzing breath metabolites (both volatile and nonvolatile) is valuable in disease diagnoses, and therefore believe that breathomics will turn into a promising noninvasive discipline in biomarker discovery and early disease detection in personalized medicine. The problem of wide variations in the reported metabolite concentrations from breathomics studies should be tackled by developing more accurate analytical methods and sophisticated numerical analytical alogorithms.

Keywords:

• Biomarker
• exhaled breath
• metabolomics
• personalized medicine

Introduction

A metabolome is defined as the global collection of small-molecule analytes (<1000–2000 Da) found within a biological sample, where the sample can be derived from a cellular organelles to cells, biofluids, tissues, organs, or even an entire organism. Metabolomics refers to studying metabolomes with regards to chemical processes involving metabolites, substrates, intermediates, and reaction products. Omics approaches connect more detailed molecular biology concepts to the general aspects of systems biology. Metablomics is an essential pillar of molecular medicine and the field is growing rapidly. According to the Human Metabolome Database (http://www.hmdb.ca) today, human body harbors at least 3396 “detected and quantified” metabolites. Given the large number of detected, expected or predicted metabolites (n = 114,265), we clearly lack information on the existence, let alone the function of these metabolites in human physiology. Regarding the most abundant or well-known metabolites, information exists mostly on their canonical function.

The workflow of metabolomics investigations include; sample collection, preparation, analysis, data handling, biomarker discovery and pathway discovery.^[1] Different biological samples including blood, plasma, feces, urine, cells, herbal extracts, exhaled breath (EB) and exhaled breath condensate (EBC) have been used in metabolomics studies. Compounds from endogenous or exogenous sources can penetrate from blood stream to the lung lining fluid, and depending on their solubility and volatility, appear in EB or EBC. On the other hand, the compounds from inhaled air can penetrate from lungs to the blood stream. Both parent compounds and their metabolites might go through penetration. EBC refers to the sample collected by trapping the exhaled aerosols originating from lung lining fluid and contains mainly nonvolatile analytes. It is obvious that some concentrations of volatile compounds could be found in EBC, especially those are soluble in aqueous solutions. EB refers to gaseous samples collected from exhaled breath and contains mainly volatile analytes and very small concentrations of nonvolatile analytes, especially those with lower boiling points. Metabolomic studies usually employ nuclear magnetic resonance (NMR), mass spectrometry (MS), and less commonly fluorescence spectroscopy (FS) to analyze the metabolic profiles in biological samples. Low sensitivity of NMR is a limiting factor, only allowing for detection of metabolites with high concentrations. However, the main advantage of NMR is that no sample pretreatment is required in this technique, while MS requires some pretreatments which may result in longer analysis time and loss of some metabolites in the pretreatment processes. In comparison though, MS is a more sensitive approach. FS is straightforward for analytes possessing native fluorescence but requires derivatization for other analytes. Near infrared (NIR) is another potential spectroscopic method that could be used in metabolomics investigations, and it has shown promising results in similar areas.^[2] Combining MS with different separation instruments such as gas chromatography (GC), high performance liquid chromatography (HPLC), capillary electrophoresis (CE), supercritical fluid chromatography (SFC), and ultra high performance liquid chromatography (UPLC) can provide complementary analytical data, since these techniques provide different and somewhat unique separation bases.

To provide a better overview of the findings, one may gather data from NMR, GC-MS, HPLC-MS, CE-MS, SFC-MS, UPLC-MS, and FS. Due to the high complexity and multidimensionality of data generated from untargeted metabolomics experiments, multivariate statistical tools such as principal component analysis (PCA), partial least squares analysis (PLS), and artificial neural networks (ANN) need to be applied in order to begin to understand the complex relationship between metabolic profiles and the disease or condition under study. Analysis of complex samples results in the complexity of obtained data. EB or EBC are interesting biological samples for detection of human diseases, especially in those of the respiratory system. On the other hand, the possibility of direct injection of EB or EBC samples to selected-ion flow-tube mass spectrometry (SIFT-MS),^[3] proton-transfer-reaction time of flight mass spectrometry (PTR-MS),^[4] atmospheric pressure ionization mass spectrometry (API-MS),^[5] LC-MS/MS,^[6–8] or CE^[9] shows the feasibility of analyzing metabolites in breath analyses.

Breathomics in general refers to multidimensional analyses including identification and quantification of volatile compounds in EB (also called volatolomics) and volatile or nonvolatile compounds soluble in aqueous solutions in EBC. Breathomics is a growing research area in experimental and clinical investigations of respiratory and systemic diseases and provides useful information in personalized medicine, since its final clinical aim is to optimize treatments for patients by considering individual breath characteristics.

There are inter-connections between different omics approaches and also environment, diet etc. as schematically illustrated in Figure 1.

Figure 1. Inter-connection between different “omics” approaches, environment and diet.

Several articles have reviewed the increasing number of studies in breathomics. Lourenço and Turner^[10] reviewed methodological considerations for analysis of volatile organic compounds (VOCs) in breath samples. Kuo et al.^[11] introduced the human breathomics database (HBDB) by gathering available information on VOCs of healthy people and patients in a number of diseases. HBDB provides a useful resource for identifying potential biomarkers from breath of the patients in order to conduct complementary investigations. Metabolomics of chronic obstructive pulmonary disease (COPD), asthma, and cystic fibrosis (CF) using different biological fluids was recently reviewed by Nobakht et al.^[12] Maniscalco et al.^[13] published a comprehensive review on metabolomics of EBC in chronic respiratory diseases. Neerincx et al.^[14] reviewed 12 reports on breathomics of VOCs in pediatric asthma covering the investigations published from 2010 to 2017. Finamore et al.^[15] reviewed 89 reports on the breathomics of VOCs in respiratory diseases, mesothelioma, pulmonary embolism, idiopathic pulmonary fibrosis, and gastroesophageal reflux disease (GERD). Azim et al.^[16] reviewed 66 reports on the assessment of VOCs in adult asthma and concluded that most of the studies are limited by a lack of standardization and external validations. Scheepers and Cocker^[17] provided a comprehensive review on the collection, analysis and interpretation of xenobiotics from occupational and environmental exposures along with their biomarkers in different biological fluids including EBC. Besides, Rahimpour et al.^[18] summarized the analytical techniques performed on EB/EBC samples with their main focus on drugs and biomarkers. Furthermore, Bruderer et al.^[19] reviewed the methodologies developed for on-line breath analysis with emphasis on the use of these technologies in diagnosing respiratory diseases, potential niche applications, and the promise of breath analysis for personalized medicine. However, there is still an obvious gap in reviewing the analytical techniques and sample preparation methods performed on EBC samples, which will be covered here. In this review, the latest articles published in breath analysis, focusing on breathomics are discussed.

EB and EBC collection

A breath sample is categorized into three fractions, including gaseous molecules, volatiles, and condensates. These fractions can be collected noninvasively through bags (for gaseous fraction), adsorbent traps (volatiles), and cold condensation chambers (condensates). The commonly used methods for sample collection for gaseous breath include (i) containers/bags: a simple and most commonly used sampling method where samples are collected in bags of varying volumes (typically 1–10 L) directly via a valve system to prevent rebreathing;^[20] (ii) reservoir/buffered sampling: These methods have been used to separate “upper” from “lower” respiratory tract air by retaining a portion of the breath from late expiration. Relatively simple solutions include the BioVOC and RTubeVOC which vent the initial breath and retain a fixed volume of end-tidal air (127 mL and 65 mL, respectively).^[21] A Breath Collection Apparatus (BCA) provides an integrated solution that also preferentially targets the breath from the end of expiration; and (iii) respiratory-cycle targeting: Another method to target late-expiratory air is to track respiration and sample breath based on the readouts from pressure or CO₂ sensors.^[22,23]

Such methods allow for a more personalized approach and accommodate very large differences in lung volume associated with age, gender, ethnicity and disease. Gaseous breath can be analyzed online or offline, with specific methods for capture and analysis for each method. Online measurement allows for rapid, near patient sample analysis and eliminates the need for sample storage. For offline measurement, the sample must be captured, stored and transported to the analytical instrument.^[24] Examples of analytical methods suitable for online analysis include forms of ion-mobility spectrometry (IMS), such as Selected Flow Tube-MS (SIFT-MS) and proton transfer reaction-MS (PTR-MS), which do not require sample pre-concentration and separation. Offline capturing of volatile compounds in gaseous phase includes their trapping onto a small, portable sorbent tube, solid phase micro extraction (SPME) fibers, or needle-trap devices.^[25,26]

Although numerous gaseous breath tests have been validated for research purposes,^[27] few tests are currently approved for clinical use. Furthermore, some technical difficulties in chemical analysis of breath are: (i) the large numbers of VOCs (possibly 100 or more) found in breath, necessitating separation prior to the analysis (e.g., GC/MS), and (ii) the very low concentration of the compounds, which are below the limits of sensitivity of currently available instruments. The latter issue can be circumvented by the use of a breath collecting apparatus that concentrates the breath into a suitable sample for analysis.

EBC comprises another fraction of exhaled breath. Various commercial instruments have been developed for EBC collection. These devices work by rapidly cooling exhaled air with subsequent condensation of water vapor, as well as the sedimentation of aerosol particles onto a cold surface. The most common instruments according to the literature are: (i) the Ecoscreen (Erich Jaeger, Hoechberg, Germany) which is a stationary system, (ii) the RTubeTM (Respiratory Research, Charlottesville, VA, USA), which is a portable device, and (iii) Turbo-Deccs (Transportable Unit; ItalChill, Parma, Italy). In addition, various lab-made instruments have been described in the literature.^[28,29] In pediatrics, exhalation maneuvers could be performed in children starting at the age of 6, and for younger children full facial mask and tidal breathing could be performed.^[30] With comparing sampling procedure for EBC and EB, sample collection for EB and its preservation for long periods are very difficult due to its gaseous nature. So, it is rather impossible to establish a biobank for these types of samples. Furthermore, the main pitfalls in sample collection for both EB/EBC are confounding factors on the concentration of the analytes such as diet, medications, smoking, physical activity and the ambient air of the sample collection area. The possible solutions for the latter factor include using purified high pressure air or subtracting inspired concentration of each analyte from expiratory concentration which requires further validations. The next critical point is to use a standard breath collection procedure, as this step has significant effect on the analysis results. Storage of the samples, especially for VOCs is another important point which should be considered in breathomics studies.^[14] In the case of patients under mechanical ventilation, the waste of the ventilator -which is the condensed form of lung lining fluid could also be collected and analyzed.^[31]

Factors affecting breath sampling

Nasal or oral inhalation

There are important differences between two breathing methods (i.e., nasal inhalation-oral exhalation and oral inhalation-oral exhalation) which may influence the concentration of analytes presented in EBC. During nasal inhalation, inhaled air is humidified in the upper airways and mediators formed in the nose and the sinuses enter the lower airways during nasal inhalation.^[32] This is an important issue to be noted, especially in patients with ongoing upper airway diseases. Measurement with oral inhalation-oral exhalation is preferred, so that only mouth-conditioned exhaled air enters the collection system.^[33]

Oral inhalation in EBC sampling may be performed with or without the use of a nose-clip. By using a nose-clip, subjects are forced to inhale through their mouths. In most studies, sample collection is performed by oral inhalation-oral exhalation.^[34]

Contaminations from salvia and nasal cavity

Many of the biomarkers in EBC are also found in significant concentrations in the salvia. For example, eicosanoids are present at high concentrations in saliva.^[35] Therefore, it is crucial to avoid EBC contamination with saliva. This can be achieved by measuring amylase concentrations^[32] or measurement of sample viscosity. Moreover, some inflammatory mediators formed in the nose and paranasal sinuses (e.g., leukotrienes and prostaglandins) can enter the oral expiratory air via the posterior nasopharynx. Therefore, it is crucial to exclude nasal contamination of EBC samples by measuring biomarker concentrations in three different experimental situations; (i) inspiring and expiring without a nose-clip, (ii) inhaling and exhaling with a nose-clip, and (iii) exhaling against a resistance to minimize nasal contamination.^[36] However, the potential for nasal/salivary contamination is different for each biomarker, and needs to be addressed in each individual case.

Temperature

Environmental and breath temperature may influence EBC results beyond the effects on volume yield. For example, there is a progressive increase in hydrogen peroxide (H₂O₂) and malondialdehyde concentrations and conductivity as the cooling temperature increases.^[37] Czebe et al. showed decreased pH with decreasing temperature^[38] and another study revealed that concentration of ammonia is lower in the ice form of condensate rather than the liquid form.^[34] So, the condensing temperature is important for those mediators/biomarkers that are thermally unstable, such as leukotrienes and purines. It is thus better for condensation temperature to be reported, to facilitate the comparison of data across laboratories.

Ambient air

EBC samples should not be left at room temperature or ambient air after collection. Ambient air contains molecules which may influence biomarker/compound concentrations in EBC through several possible mechanisms. For example, these molecules can (i) directly contribute to EBC, (ii) react and, therefore induce the degradation or formation of substances trapped in EBC, and (iii) promote inflammatory and biochemical changes in the airway that are subsequently reflected by changes in EBC composition. For instance, it has been shown that atmospheric NO reduces exhaled H₂O₂ levels.^[39]

Condenser system

Ahmadzai et al.^[40] listed a comparison of biomarkers measured in EBC that were collected using different devices. The design, inner coating materials used in the collecting device and temperature fluctuations over a condensate collection maneuver influence the concentration of biomarkers.^[41] Uneven condensing temperatures may not only be present between different EBC collection systems, but also in the case when the same condenser is used under different airflow or with different cooling temperatures. Moreover, variability in EBC pH between the various collection systems due to differences in the concentration of absorbed volatile salivary contaminants during collection (e.g., CO₂, ammonia and acetic acid) can differentially influence the concentrations of biomarkers or nonvolatile compounds.^[28] Finally, to provide consistent data, the EBC collection system needs to be standardized with respect to collection temperature, coating material and flow design, as well as other parameters such as ventilatory patterns, tidal volumes, breathing frequencies, exhaled particles, etc.

Dilution

The airway lining fluid component of EBC is highly diluted by condensing vapor phase water.^[42] The concentration of EBC compounds depends on the efficiency of water vapor condensation, which is influenced by the temperature and surface area of collection. Without a reliable measure of a dilution factor, the precise concentration of compounds in the epithelial lining fluid (ELF) cannot be accurately determined. For this reason, a number of normalization factors have been proposed for nonvolatile substances in the literature. It would be preferable to use non-inflammatory reference indicators as internal standards that remain relatively unchanged in the respiratory fluid. Such indicators are ideally similar in concentration to those in the plasma and diffuse through the cell membranes, but are not produced in the alveoli or airways.^[43] Urea as a candidate molecule for this purpose is small, readily diffusible and equally distributed throughout the body. Furthermore, urea is not metabolized in the lungs. Thus, it would be possible to calculate an appropriate dilution factor for the nonvolatile compounds by measuring urea concentration in EBC, and estimating the interstitial urea.^[44] Measurement of urea concentration is based on degradation of urea to NH₄⁺ by urease. However, urea concentration in EBC is relatively low, and serum values are more variable than those of the total electrolytes. Also, there may be a possible bacterial degradation of urea in infected lungs. Therefore, estimates of dilution with urea are consequently not accurate.^[43]

Other alternative ways are total nonvolatile cations, or conductivity measurement of EBC. Effros et al.^[43] have reported condensate electrolyte concentrations as a means to calculate the dilution of respiratory fluid droplets in the condensate. In one study in 2002,^[45] the effect of dilution was minimized by dividing condensate solute concentrations by the sum of the concentrations of the nonvolatile cations (Na⁺ + K⁺) in EBC. In a following work in 2003,^[43] the same group reported that conductivity can be used to estimate airway electrolyte concentrations and the dilution factor, under the condition that the ammonium ions are first removed by lyophilization. The observations in these studies suggest that the conductivity of lyophilized samples can be used as an inexpensive, simple, and reliable measure for estimating dilution of nonvolatile, hydrophilic biomarkers in EBC. Another reported correction for dilution is expressing concentrations of EBC components per time or per duration of EBC collection. However, this strategy has a significant drawback, because of the dependency of the collected volume of EBC on the volume of exhaled air in a given period of time. ^[46] Although, dilution factor calculations are essential for interpretation of the obtained results from EBC studies, none of the above suggested strategies has proven to be a golden standard and the dilution factor calculations were omitted in most of the published articles on EBC.

In the last decade, Reinhold and Knobloch^[47] and Rosias^[48] have recommended to standardize the concentration of any EBC component in relation to a defined volume (100 L) of exhaled air. Comparing different EBC components as exhaled per 100 L exhalate led to a decrease in the variability and to an increase in the repeatability and reproducibility of the obtained data.

In two cases, the dilution factor calculations are not necessary: (i) when multiple interactive or biologically related biomarkers are measured concurrently and their ratios are considered,^[49] and (ii) when a confident assay for a substance is needed, it serves as an on-off indicator of an abnormality. Examples include the presence of Mycobacterium tuberculosis DNA, gastric pepsin, rhinovirus RNA, and anthrax toxin.^[50]

It should be mentioned that the dilution factors are irrelevant for VOCs. The important factors affecting the VOC levels include water solubility, gas-liquid partition coefficients, temperature of the source fluid (airway lining fluid), temperature of the condenser, pH of the source fluid and EBC, and the opportunity to react within (and/or be captured by) the EBC matrix.^[50] Adsorption on the surface of the collection bag and/or containers is the next factor affecting the concentration of VOCs.

Biological influences

In addition to the above mentioned factors, the amount of biological variability that is present even in healthy subjects under physiological conditions needs to be evaluated before interpretation of data. This physiological variability includes intra-subject and inter-subject components. Inter-subject variability is determined by the homogeneity or non-homogeneity of a selected group. Intra-subject variability includes, for example, circadian influences, effect of diet, effect of respiration (tidal volumes, breathing frequency, airflow rates and the ratio between dead space volume and alveolar volume) or age-dependent effects.^[51]

Analytical advancements

Biological samples have complicated matrices and should be analyzed by sensitive techniques. Since the constituent substances in such samples have diverse physicochemical properties, several analytical techniques are sometimes needed to expand metabolite coverage. Different analytical instruments such as GC, HPLC, CE, SFC, NMR, and FT-IR have been used in metabolomics studies.

NMR

NMR is the most common method used in the analysis of different compounds in biological samples. In this method, nuclei are influenced by an external field and their transitions in magnetic energy levels are recorded. NMR spectroscopy allows for the visualization of single atoms and molecules in various media in solution as well as in solid state. NMR can be used for identification and quantification of different compounds with unbiased results. NMR results are repeatable, reproducible and accurate, and detection limits of micro to nano molar ranges can be obtained. Due to these advantages, NMR spectroscopy is used in metabolomics for studying different biological samples. In 2014, proton nuclear magnetic resonance spectroscopy (H-NMR) was used in profiling the EBC of patients with COPD.^[52] For this purpose, the collected EBC samples were mixed with sodium phosphate and sodium azide and the mixture was transferred for H-NMR analysis. To compare the results, EBC samples of healthy individuals were also analyzed. Using this method, acetone, lysine, valine, tyrosine, proline, serine, propionate, acetate, and lactate were quantified. The results showed that assessment of these compounds is informative for determining the pathological state of COPD. Metabolomics of EBC discriminated COPD patients from controls with an overall accuracy of 86%. Compared to controls, EBC from COPD featured significantly lower levels of acetone, valine and lysine, and significantly higher levels of lactate, acetate, propionate, serine, proline and tyrosine. H-NMR was applied to EBC samples obtained from the asthmatic patients.^[53] In this study, EBC samples were collected at different temperatures to follow the changes in the type and concentration of the substances in EBC. The experiments were done on EBC samples which were collected with two different devices collecting the EBCs at –23 and –4.8 °C. The results verified that the EBC collection device temperature affects the level and the type of substances detected in the samples. The samples were initially analyzed and assessed the within-day, between-day, and technical repeatabilities. The results showed that relative standard deviation (RSD) was ≤11.2%. In this method the samples were classified into three groups and analyzed together, disregarding the condensing temperature. The results showed that there are strong correlative models (95%) with high average-quality parameters for spectral profiling and targeted profiling.

Maniscalco et al.^[54] analyzed biological samples of asthmatic and COPD patients including saliva, serum, EBC, and urine with H-NMR and compared the compounds profiles of these samples. The results showed that H-NMR discriminates asthma and COPD patients with high sensitivity and specificity, and this may help the clinicians to decrease the number of erroneous diagnoses. This research group used H-NMR for surveying the differences between groups of subjects occupationally exposed to levels of airborne inhalable dust, formaldehyde, phenol, and VOCs below regulatory limits.^[55] The results demonstrated that NMR-based metabolomics of EBC is a promising strategy with potential to be successfully used in occupational health to discover subjects exposed to very low levels of airborne chemicals. In another work performed by D’Amato et al., the EBC profiling was performed in the EBC samples of bronchiectatic patients using NMR.^[56]

MS

MS is another analytical technique widely used in metabolomics. MS can analyze low concentrations of different substances with high resolution. In contrast to NMR, MS provides low linear range and its sensitivity is strongly related to the ionization efficiency of the compounds. However, coupling MS with separation systems such as LC, GC, and CE improves the separation and identification of new biomarkers. HPLC-MS technique was applied on EBC samples for metabolomics analysis.^[57] In this study, after removal of the proteins with precipitation, the samples were extracted with different sorbents during solid phase extraction (SPE). MS analysis led to detection of fatty acids, amino acids, fatty aldehydes, fatty amides, oxoanionic, imidazoles, sphingoid bases compounds, hydroxy acids and aliphatic acyclic acids.

Metabolomics profiles of saliva, EBC, and plasma was studied using HPLC-MS to detect air toxics.^[58] The samples were collected from individuals before and after a 2-hour road traffic exposure. The analysis yielded 7,110, 6,019, and 7,747 reproducible features in plasma, EBC, and saliva, respectively. Correlations were moderate-to-strong (R = 0.41 − 0.80) across all pairwise comparisons of feature intensity within profiles, with the strongest correlation being between EBC and saliva. EBC metabolomics analysis by UPLC-MS was used for early detection of CF and acute pulmonary exacerbations.^[59]

UPLC-MS was used to profile the EBC metabolites of clinically stable CF patients, CF patients who required hospitalization, 5 CF patients in recovery step, and 4 CF patients who were clinically stable at the time of EBC collection. The result showed that 3 discriminant metabolic features identified as 1) pyroglutamic acid, 2) short chain carboxylic acids and 3) hydroxyeicosatetraenoic acids differentiated the CF patients from healthy people. Pre-APE samples were distinguished from stable CF samples. In another study, EBC samples of patients suffering from CF were analyzed by ion mobility-MS,^[4] finding three metabolites discriminating patient EBCs from healthy subjects from both sources.

GC-time of flight (TOF)-MS can also be employed in metabolomics analysis of EBC samples. This method was applied for profiling volatile and nonvolatile compounds.^[60] The collected samples were extracted with Liquid–liquid extraction (LLE) and SPE methods. The results showed that LLE provided the widest information on the composition of EBC with tentative identification of 51 compounds. Among the identified compounds were fatty acids and other derivatives such as methyl esters and amide derivatives, which are not considered as volatile components. LLE provided an extraction recovery in the range of 77% in a single extraction. The method repeatability was examined on EBCs from 50 healthy individuals reporting a within-day variability below 7%. Other representative family of compounds found in EBC was that of prenol lipids such as terpenes, which are volatile. Castro and his group used EBCs to develop a screening tool for lung cancer discrimination between two groups with and without risk factor.^[61,62]

The analyses of amino acid content in EBC samples has attracted more attention in identification of several diseases. For this purpose, Konieczna et al.^[63] developed a HPLC tandem mass spectrometry (MS/MS) method for simultaneous determination of 16 amino acids in EBC samples. Validation parameters such as precision (RSDs ≤ 11%), accuracy (RSDs ≤ 10%), recovery (52.2-108.2%) and linear range (0.25–500 ng/mL) were investigated and it was found that the developed method was robust and reliable for the determination of the selected amino acids.

Other methods

Gas sensing using tetrahertz radiation is a new opportunity to follow up VOCs. In this technique, heavy gas molecules display rotational resonance at microwave frequencies and light gas molecules in the mid-infrared frequencies.^[64] Electronic nose technology provides another opportunity for breathomics studies.^[65,66] Sniffphone is a newly developed combination of nanoparticle based gas sensors with smart phones which provides a point of care tool for disease diagnosis.^[67] Nakhleh et al.^[68] reviewed the progresses made on using gold nanoparticles for breath analysis. Arrays of mono-layer-capped gold nanoparticle sensors provided fast, high-throughput and cost-effective methods for exhaled breath analysis. The produced results were used for numerical analyses in order to find out the pattern of variations of VOCs in different diseases.^[68] Tang et al.^[69] reviewed the applications of nanostructured metal sulfide materials for sensing of VOCs and inorganic gases such as CO₂, NH₃, H₂S, SO₂, CH₄, H₂ and nitric oxide. Recent advances in the role of eNoses in phenotyping of COPD were reviewed by Searlata et al.^[70] Mirzaei et al.^[71] review and compare devices used for Δ9-tetrahydrocannabinol detection in EB such as field asymmetric ion mobility spectrometry, semiconductor-enriched single-walled carbon nanotube chemiresistors, LC-MS/MS, microfluidic-based artificial olfaction, and optical-based gas sensing. They reported that amongst the examined platforms, from the portability and usability perspectives, optical-based, single-walled carbon nanotube chemiresistors, and artificial olfaction are the most promising, whereas field asymmetric ion mobility spectrometry and LC-MS/MS are best suited for highly sensitive and selective testing with toxicology lab confirmation.

FS is a simple and low cost method with minimum sample pretreatment, requiring a short analysis time for metabolomics studies. FS has been used for metabolomics investigations in plasma ^[72] and urine;^[73] however, to the best of our knowledge, there is no report on the application of FS for EBC analysis.

Reactive oxygen species are chemically reactive species containing oxygen that are produced naturally during oxygen metabolism and are now known to possess important roles in cell signaling and homeostasis. Reactive oxygen species levels are elevated drastically under oxidative stress and can significantly damage cellular structures. These compounds have short half-lives and their determination in biological fluids is challenging. In this line, EBC can be a good candidate for analyzing oxidative products such as malondialdehyde. Different analytical methods can be used to quantify such oxidative products in EBC samples, as shown before for malondialdehyde.^[74–76] Lačná et al. determined aldehyde metabolites in EBC samples with high sensitivity using CE with laser-induced fluorescence detection. In the project, one dialdehyde (malondialdehyde), eight saturated monoaldehydes, and an unsaturated aldehyde were investigated, as the main aldehydic metabolites in EBC.^[77] Under optimal conditions, target analytes could be well separated, and sensitive, repeatable, and reliable determination of the analytes was possible. The method was performed on EBC samples of smoker and nonsmoker men and women. The results showed that the concentrations of the compounds in EBC samples of male smokers were dramatically higher (∼eight folds) than those of male nonsmokers. However there was no significant difference between the smoker males and females.

Numerical and informatics methods in metabolomics research

Metabolomics (encompassing breathomics) analysis usually generates multi-dimensional data. Current metabolomics exploits classic analytical chemistry combined with cheminformatics and bioinformatics, to enable large-scale data analysis. Chemometrics is a branch of science dealing with data-driven means to extract information from chemical systems.^[78] In the majority of cases, the goal of metabolomics analysis in general and breathomics in particular is to separate metabolome snapshots from different samples or help identify biomarkers for various diseases. This feat is possible through classical statistics in parallel with multidimensional data analysis tools. The latter tools are essential in light of the large data that can be acquired using current technologies such as MS.^[79] In this regard, it is of utmost importance to recognize these tools and meticulously use them based on need and the area of application. Classic statistical tools such as t-test, Mann-Whitney test and one-way analysis of variance (ANOVA) have been extensively used and reviewed in the past. Therefore, in this section, we will focus on common multidimensional data analysis and multivariate statistical approaches used in metabolomics research. In this regard, robust data analysis pipelines are well established in transcriptomics and to a great extent in proteomics. Scientists in these areas have tried to deal with big data challenges over time. Such workflows have been also applied in metabolomics.

Usually a routine pipeline for biomarker discovery involves the following steps: automated detection of metabolites in the raw files,^[80] preprocessing (subtraction of background noise, peak detection and quantification) and cleaning of raw profiling data (removal of contaminants or missing values or imputation) and normalization. The data are then subjected to statistical analysis and selection of the most statistically significant and biologically relevant metabolites through feature selection, clustering, classification and pathway/network analyses. Finally, making the data sets publicly available is crucial, so that the community can reanalyze and reevaluate the quality of experimental outputs. In this line, Metabolomics Standards Initiative (MSI) ^[81] is a community driven effort that sets global consensus for minimum reporting standards. Open-access online repositories such as MetaboLights^[82] and Metabolomics WorkBench allow for experimental data deposition, which in turn enables transparency and knowledge transfer. Furthermore, specific reporitories such as Human Breathomics Database can also be used for data/study deposition.^[11]

Multivariate data analysis

Once the data is cleaned, univariate analysis methods are applied to compare groups. Such methods include for example t-test, ANOVA, Kruskal–Wallis, etc. Different types of adjustments for multiple testing such as Bonferroni correction or Benjamini-Hochberg correction are also required to minimize the false discovery rates.^[83]

However, to comprehensively analyze a large data set, multivariate analysis methods are essential. Multivariate statistics encompasses methods that can be employed for simultaneous observation and analysis of data containing two or more variables. Through these methods, the dimensionality of data is reduced, improving their interpretation and visualization. Multivariate statistics can help to separate the signal from noise in high dimensional data. In addition, they can be useful in selection of variables such as biomarkers separating the samples or populations.^[84,85]

Typical computational techniques used in metabolomics include but are not limited to PCA, clustering (e.g., hierarchical and k-means), and classification (e.g., Canonical Variate Analysis (CVA) and Support Vector Machine (SVM)). These analysis tools are mainly used for overall assessment of different biological systems or conditions, and/or quantifying the contribution of each metabolite to the differences between these systems and conditions. The top features separating the biosystems can thus be potential biomarkers.

Multivariate data analysis usually begins with unsupervised exploration of data. If unsupervised techniques do not yield satisfactory results, supervised methods such as classifiers can be used. After classification, internal cross validation strategies such as bootstrapping, split-half, or leave-one-out are used. Finally, external validation is performed by evaluating the performance of the model, using data that were not used in developing/training the model. The ultimate validation would be to test the model on a newly recruited population.^[86]

Exploratory data analysis by unsupervised tools

Exploratory data analysis tools are divided into projection methods and partitional clustering techniques. PCA is the most extensively used projection method in omics.^[66] PCA is an exploratory data analysis tool and a dimension reduction technique that is routinely used for data quality control, i.e., reproducibility between replicates or adjacency or relatedness of populations/samples. Furthermore, PCA is generally employed in omics studies including metabolomics for dimension reduction, to basically showcase separation and trends/patterns in the data.^[87] PCA output is a set of principal components. The first principal component has the largest possible variance and explains much of the variability in the data. By visualizing the samples in the PCA space, one can determine the variables most contributing to separation of samples within each component, most of the time even by naked eye. The other components explain less and less of the variation in the data in a step-wise manner. PCA loading can help differentiate the features (here metabolites) that have the most discriminatory power in separation of samples.

Canonical correlation analysis (CCA) is also gaining popularity in metabolomics. CCA is related to PCA, the difference is that while PCA works by defining coordinate systems that describe the variance in an individual data set, CCA defines the cross-variance between two data sets.^[88] CCA is an approach that is used to find the relationship between two multivariate sets of variables quantified for the same set of samples.^[89] Therefore, CCA is believed to be an extension of bivariate correlations. Among two vectors with random variables, CCA will discover linear combinations (called canonical variate) within the vectors that have maximum correlation with each other. The statistically significant canonical variates are then presented in a CCA score plot. Each point on such a plot corresponds to the combined information from the two vectors, whereas the shape of the cloud is indicative of the correlation between the two blocks of data. The relation between the two blocks of data is strong, only if the cloud of points forms a coherent scatter along a line and vice versa. CCA readout is the canonical coefficients, which show the contribution of the original parameters to the correlation. For example, Smolinska et al.^[90] used CCA to investigate the relationship between volatile metabolites in breath and gut microbiome in active and quiescent Crohn's disease (CD). In this study, CCA analysis identified 18 volatile metabolites significantly correlating with 19 fecal bacterial taxa (R = 0.91; p value = 3.5 × 10^–4) in active disease, while in the quiescent disease 17 metabolites were correlating with 17 bacterial taxa (R = 0.96; p value = 2.8 × 10^–4).

In this context, Factor Analysis also has special applications.^[91] Factor analysis is very similar to PCA, but these two methods are not identical. The aim of both methods is to reduce the dimensionality of the data. Unlike PCA, factor analysis is specifically tasked with the objective to determine a set of unobservable factors from the observed variables. This method helps to find a presumed small number of underlying variables by investigating a large number of observed variables.

Clustering methods comprise another type of exploratory data analysis tools. Basically, clustering is tasked with assigning group membership to samples in a data set and partitioning the data into clusters with homogenous members based only on quantitative data. The number of clusters are gradually reduced to reach a single cluster. In other words, the samples with similar features are assigned to the same cluster without a priori assumptions. This makes clustering an unbiased approach. Clusters are built in a way that samples within a cluster have the highest association with each other, while being only weakly associated with samples in other clusters.^[92] A proximity measure is used to partition the dataset into clusters in such methods. For example, in hierarchical clustering, distances are used to assess the similarity of samples and to organize them into an ordered grouping or a dendrogram. K-means uses a similar proximity measure, but the number of clusters is pre-defined by the user. Euclidean distance and Mahalanobis are the most commonly used distances. Clustering is routinely used in metabolomics for data exploration. ^[91]

As an example, de Vries et al.^[66] applied a breathomics approach for phenotyping chronic airway disease. First, they used PCA to reduce the dimensionality of the data prior to clustering. Such dimension reduction would reduce the risk of overfitting. Subsequent unsupervised hierarchical clustering resulted in five significant combined asthma and COPD clusters that differed regarding ethnicity, systemic eosinophilia and neutrophilia, body mass index (BMI), exhaled nitric oxide fraction, atopy and exacerbation rate. The importance of these results is that these phenotypes lead to clusters that cannot be determined by diagnosis and are only reflected in clinical/inflammatory characteristics.

Supervised methods

The second major and fundamental part of the multivariate analysis is covered by classification methods, which aim at finding mathematical models that are capable of assigning membership of each population/sample to a proper class in the data set. In these analyses, each sample is assumed to belong to a relevant class. Such methods can be applied to any types of high dimensional data resulting from proteomics, spectroscopic, chromatographic and metabolomics analysis. In the case of metabolomics for example, the samples would be the quantified metabolomes and the data set would be the whole data obtained from the measurements. Supervised methods can also be used on relative abundance data as well as absolute quantification data.

Classification methods are divided into various types: linear and nonlinear, probabilistic and distance-based, pure classification methods and class-modeling methods. While the best linear boundary is used for class discrimination in linear classification methods, as the name suggests, the best curve or nonlinear boundary is used in non-linear methods. The classification in probabilistic methods are based on estimates of probability distributions, while distances between samples are used in distance-based methods. In pure classification methods, the hyperspace is separated into as many regions as the number of classes. However, in class-modeling methods, the focus is on modeling the analogies among the objects of a class.^[93]

CVAs and SVMs are two types of pure classifiers. In CVA, the samples are separated into classes by obtaining the minimal class variance and maximal between-class variance. Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA) and Partial Least Squares-Discriminant Analysis (PLS-DA) are also similar or related to CVA. For example, Multilevel PLS was used for data reduction in a breathomics study to assess the effects of treatment and withdrawal with beclomethasone/formoterol inhalation in patients with COPD.^[94] PLS-DA as a supervised modeling approach is also gaining popularity in analysis of omics data, both in exploratory data overview and discovery of specific features in the data contributing to a condition under study.^[95] For example, in a recent tool called ProTargetMiner,^[96] PLS-DA modeling was used to contrast one proteome against others, to highlight the features that discriminate one treatment against others in the loading plot. Since this method can pull out the features that have the largest discriminatory power, such a paradigm can also be used in biomarker discovery in metabolomics and other omics studies. PLS-DA produces an R² value representing the goodness of the model fit and a Q2 value, which corresponds to the model predictive power. The latter is calculated by an internal cross-validation, which is described in detail in the above article. For more detailed explanation, see Umetrics documentation.^[97]

SVM on the other hand, is tasked with classifying the data in space with a separator described by a hyperplane. Both linear and non-linear combinations of functions parametrized by support vectors can be used to express the hyperplane. This is why SVMs are suited and commonly used for non-linear separations. Other non-linear techniques include k-nearest neighbors (k-NN),^[94] Artificial Neural Networks and Random Forests.^[95] For example Daniel and Thangavel^[98] analyzed the breath of 49 gastric cancer and 30 gastric ulcer patients to distinguish the normal, suspected, and positive cases using back-propagation neural network. This analysis produced an accuracy of 93%, sensitivity of 94.38%, and specificity of 89.93%.

The class-modeling methods or so called one-class classifiers such as Soft-Independent Modeling of Class Analogy (SIMCA) can also have specific applications in metabolomics.^[99] In these methods, each class is modeled separately. These classifiers are usually employed when a reference class exists, for example in food authentication and toxicology.

Different classifiers can have various performances in various studies or applications. Therefore, they are sometimes tested simultaneously. For example, in a breathomics study, Arasaradnam et al.^[100] investigated exhaled VOC analysis to detect hepatic encephalopathy (HE). Breath samples from 22 HE patients and 20 control subjects were captured and analyzed using an uvFAIMS portable e-nose. Data were analyzed using 4 different classifiers including sparse logistic regression, Random Forests, SVM and Gaussian Process, showing the difference in the performance of these techniques.

In this section, we provided a brief introduction to the multivariate statistical tools in metabolomics and breathomics. This account is far from comprehensive. For a complete overview of the field, the reader is directed to other reviews.^{[78,79,83–85,93,101–104]} Saccenti et al.^[83] have interesting reflections on univariate and multivariate analysis of metabolomics data.

The great impact of databases on breathomics research

Breathomics research is accelerating especially by introduction of databases such as Human Breathomics Database (HBDB),^[11] METLIN^[105] and MassBank.^[106] HBDB for example contains a total of 913 VOCs in relation to human exhaled breath researches reported in 2,766 publications. The HBDB is the most comprehensive HBDB of VOCs in human breath to date. This along with the high number of publications in the field underlines the necessity of developing and advancing statistical tools in coping with this progress.^[79] Automation of metabolomics data analysis is also an emerging topic. For example, different clustering approaches have been developed such as DBSCAN and OPTICS and very recently VOCCluster.^[107]

Considerations and limitations

In the end, one should never forget the importance of method validation.^[108] Although many molecular fingerprints and biomarkers have been identified in the literature using the above computational techniques, only a fraction of the reported findings have been validated and proven to be useful. The large nature of data obtained in metabolomics requires good experimental design and identification of biological and technical issues that might give rise to bias, leading to false positive and negative discoveries. An efficient experimental design must consider different components such as formulation of hypothesis, proper and adequate definition of phenotype, power calculations, sample size calculations, optimization/standardization of protocols, quality control, multiple testing correction methods and plans for validations of findings.

A common problem in omics data analysis is the extremely large number of variables compared to the small number of samples. In ideal case, for accurate estimation of the covariance matrices, the number of samples must at least be equal to the number of variables. Otherwise, false discoveries arise due to overfitting of the diagnostic algorithm.^[101] Overfitting can be circumvented by adequate sample size or more often by proper correction for multiple comparisons.^[86,109,110] Other challenges in chemometrics include dealing with noisy data, data outliers, missing data, alignment of batch data and merging of multiomics datasets. These challenges must be prioritized as a research frontier in metabolomics in the coming years.

Reports on metabolomics analysis of EBC

Here we will review recent studies based on metabolomics analysis of EBC in lung diseases as well as non-respiratory disorders and animal studies.

Lung diseases

The complexity of lung disease is due to various mechanisms that are not present in all patients at a given time or in the same patient at different time points.^[111] Phenotyping of the patients using specific biomarkers is required in clinical practice for characterizing, prognosis and treatment of patients suffering from respiratory diseases. This is in line with the universal efforts toward personalized medicine. For such a complex disease, a single biomarker cannot be used and a panel of biomarkers is required for an accurate molecular level explanation of the disease.^[112] It is also important to identify the biomarkers that are normally present in EBC from those that appear later when one gets ill. In this line, Kazeminasab et al.^[113] discuss the significant role of exhaled breath biomarkers in COPD. They reported that reactive oxygen species production has the potential to damage lipids, proteins and nucleic acids as three classes of biomarkers in the EBC which act as indicators of biological or pathological processes in the lungs. Therefore, analysis of pulmonary biomarkers provides new insight into modern medicine that could be useful in aiding risk assessment, disease prevention, early and accurate diagnosis and monitoring treatment effectiveness in COPD.

Kononikhin et al.^[114] used a high resolution nano HPLC-MS method on EBC samples of newborns (20–40 days of life). This method was used to identify respiratory diseases by analyzing the EBC of 119 subjects and 164 metabolites were successfully characterized. To define respiratory disease-specific features, label-free semiquantitative proteomics and metabolomics data were obtained from 24 intubated neonates, through an easy, fast, and noninvasive method. The obtained differential biomarkers could be used in identification of respiratory deficiencies and diseases.

Similar to the above study, the composition of amino acids was investigated in EBC samples of children suffering from asthma.^[115] The results showed that amino acids levels especially glycine and methionine in children with a positive history of allergies can be an additional prognostic or diagnostic criterion for asthma. In another research amino acids and other asthma-related molecules were assessed in EBC and saliva samples using HPLC-quadruple-TOF (Q-TOF)-MS.^[116] According to this study, amino acids and several small molecules can be quantified in EBC samples, the concentration of which could be used for asthma diagnosis. On the other hand, 77 metabolites were identified in EBC samples obtained from lipids, spices, herbs, plants, food, herbicides, dipeptides, and polycyclic Aromatic Hydrocarbon (PAH) degradants. Interestingly, many of metabolites were undetectable in saliva samples and could only be found in EBC.

Table 1 lists the published works on breathomics studies of lung diseases based on both EB and EBC samples.

Table 1. Metabolomics studies on respiratory diseases using EB or EBC (breathomics) samples.

Disease/Case	Number of participants (case/control)	Metabolomic approach	Potential biomarkers	Analytical method	Sensitivity	Specificity	EB or EBC	References
Acute respiratory distress syndrome	49/52	Metabolic profiling	Octane, acetaldehyde and 3-methylheptane	GC-MS	95.0%	–	EB	^[^117^]
Asthma COPD	20/NR 32/NR	Metabolic profiling	Ethanol, methanol, formate and acetone/acetoin	1H-NMR	92.3%	95.0%	EBC	^[^54^]
Asthma	55/55	Metabolic profiling	Succinate, hippurate, pyruvate, methanol, trimethylamin, acetate, urocanic acid, ethanol, and etc.	NMR	> 90.0%	> 90.0%	EBC	^[^53^]
Asthma	11/12	Target analysis	1-(Methylsulfanyl) propane, ethylbenzene, 1,4-dichlorobenzene, 4-isopropenyl-1-methylcyclohexene, 2-octenal, octadecyne, 1-isopropyl-3- methylbenzene, and 1,7-dimethylnaphtalene	GC-MS	–	–	EB	^[^30^]
Asthma	25/11	Metabolic profiling	Acetylated and oxidized compounds	NMR	–	–	EBC	^[^118^]
Asthma	Adult: 28/50 Children: 33/50	Target analysis	Leukotriene B4, D4, and E4	GC-MS	–	–	EBC	^[^119^]
Asthma	63/57	Target analysis	Branched hydrocarbons (C13H28 and C11H24), carbon disulfide (CS2), 1-penten-2-on, butanoic acid, 3-(1-methylethyl)-benzene, unsaturated hydrocarbon (C15H26), benzoic acid, and p-xylene	GC-MS	89.0%	95.0%	EB	^[^120^]
COPD	37/25	Metabolic profiling	Acetone, valine, lysine, lactate, acetate, propionate, serine, proline, and tyrosine	1H-NMR	> 69.0%	> 84.0%	EBC	^[^52^]
COPD	15/20	Target analysis	Acetate, 1-methylimidazole, 2-propanol and isobutyrate	NMR	–	–	EBC	^[^121^]
Cystic fibrosis	210/NR	Metabolic profiling and target analysis	Lactic acid and 4-hydroxycyclohexylcarboxylic acid	UPLC-MS	Pediatrics, 85.7%, adults, 89.5%	Pediatrics, 88.4%, adults, 84.1%	EBC	^[^122^]
Cystic fibrosis	4/5	Metabolic profiling and target analysis	Pyroglutamic acid, 1-pyrroline-4-hydroxy-2-carboxylate, N-acryloylglycine, pyrroline hydroxycarboxylic acid, dimethylmalonic acid, 2-acetolactate, glutaric acid, monoethyl malonic acid, ethylmalonic acid, methylsuccinic acid, glycyl-glycine, asparagine, ureidopropionic acid, N-carbamoylsarcosine, 5-, 8-, 9-, 11-, 12-, 15-, 16-, 17-, 18-, 19-, and 20-HETE; 5,6-; 8,9-; 11,12-; 14,15-EET	TM-DART-TWIMS-TOF MS	100.0%	100.0%	EBC	^[^4^]
Cystic fibrosis	48/57	Target analysis	C5-C16 hydrocarbons, N-methyl-2-methylpropylamine	GC-TOF-MS	58.0 to 100.0%	91.0 to 100.0%	EB	^[^123^]
Cystic fibrosis	17/NR	Target analysis	4-Hydroxycyclohexylcarboxylic acid, pyroglutamic acid, lactic acid and pyroglutamic acid	UPLC-QTOF-MS	77.8%	88.2%	EBC	^[^59^]
Cystic fibrosis	29/31	Target analysis	Ethanol, acetate, 2-propanol, and acetone	NMR	91.0%	96.0%	EBC	^[^124^]
Lung cancer	68/256	Metabolic profiling	Eucalyptol, indole, benzoic acid 4-ethoxy-ethy ester, benzoic acid methyl ester, 3,5-di-t-butyl-4-hydroxy-cinnamic acid, octinoxate, cresol, and etc.	GC-TOF/MS	77.9%	67.5%	EBC	^[^62^]
Lung cancer	48/61	Metabolic profiling	Lipids such as monoacylglycerols and squalene	GC-TOF-MS	UA, 83.7% DA, 77.5	UA, 83.3% DA, 89.8%	EBC	^[^61^]
Lung cancer	12/17	Metabolic profiling	VOCs	GC-MS	–	–	EB	^[^125^]
Lung cancer	8/1	Metabolic profiling	VOCs	GC-MS	–	–	EB	^[^126^]
Lung cancer	108//50	Metabolic profiling	Pentane and isoprene	GC-MS	100.0%	81.3%	EB	^[^127^]
Lung cancer	40/56	Metabolic profiling	VOCs	GC-MS	–	–	EB	^[^128^]
Lung cancer	43/41	Metabolic profiling	VOCs	GC-MS	95.3%	85.4%	EB	^[^129^]
Lung cancer	33/NR	Metabolic profiling	VOCs	GC-MS	100.0%	80.0%	EB	^[^130^]
Lung cancer	65/31	Target analysis	VOCs	PTR-MS/GC-MSc	80.0%	100.0%	EB	^[^131^]
Lung cancer	107/88	Target analysis	VOCs	FT-ICR-MS	83.0%	74.0%	EB	^[^132^]
Lung cancer	29/44	Target analysis	VOCs	GC-MS	82.2%	82.4%	EB	^[^133^]
Lung cancer	5/5	Target analysis	VOCs	SAW sensor	–	–	EB	^[^134^]
Lung cancer	71/67	Target analysis	H2O2	GC-MS	65.2%	71.2%	EB	^[^135^]
Lung cancer	11/30	Target analysis	Thiobarbituric-acid-reactive substances (TBARS) and conjugated dienes (CD)	GC-MS	–	–	EB	^[^136^]
Lung cancer	10/10	Target analysis	VOCs	GC-MS	–	–	EB	^[^137^]
Lung cancer	22/22	Target analysis	VOCs	PTR-MS	–	–	EB	^[^138^]
Lung cancer	22/21	Target analysis	VOCs	GC-MS	–	–	EB	^[^139^]
Lung cancer	97/88	Target analysis	VOCs	FT-ICR-MS	89.8%	81.3%	EB	^[^140^]
Lung cancer	12/12	Target analysis	VOCs	GC-MS	8.3–75.0%	91.7–100.0%	EB	^[^141^]
Lung cancer	28/10	Target analysis	VOCs	GC-MS	–	–	EB	^[^142^]
Lung cancer	80/38	Target analysis	VOCs	IMS	70.0%	89.7%	EB	^[^143^]
Lung cancer	31/31	Target analysis	VOCs	GC-MS	–	–	EB	^[^144^]
Lung cancer	34/85	Target analysis	VOCs	FT-ICR-MS and GC-MS	96.0%	64.0%	EB	^[^145^]
Lung cancer	65/31	Target analysis	VOCs	GC-MS	51.0%	100.0%	EB	^[^131^]
Lung cancer	123/361	Target analysis	VOCs	GC-MS	63.5%	72.4%	EB	^[^146^]
Lung cancer	13/25	Target analysis	VOCs	GC × GC-FID	–	–	EB	^[^147^]
Lung cancer	10/10	Target analysis	VOCs	GC-MS	–	–	EB	^[^148^]
Lung cancer	53/19	Target analysis	VOCs	GC-MS	86.0%	96.0%	EB	^[^149^]
Lung cancer	62/40	Target analysis	VOCs	GC-MS	–	–	EB	^[^150^]
Lung cancer	30/22	Target analysis	VOCs	GC-MS	–	–	EB	^[^151^]
Lung cancer	5/41	Target analysis	VOCs	GC-MS	89.6%	82.9%	EB	^[^152^]
Lung cancer	193/211	Target analysis	VOCs	GC-MS	84.5%	81.0%	EB	^[^153^]
Lung cancer	35/50	Target analysis	VOCs	GC-MS	72.2%	93.6%	EB	^[^154^]
Lung cancer	40/38	Target analysis	VOCs	GC-MS	–	–	EB	^[^155^]
Lung cancer	23/30	Target analysis	VOCs	GC-TOF-MS	–	–	EB	^[^156^]
Lung cancer	107/29	Target analysis	VOCs	GC-MS	–	–	EB	^[^157^]
Lung cancer	37/23	Target analysis	VOCs	GC-MS	92.0%	96.0%	EB	^[^158^]
Lung cancer	156/194	Target analysis	Volatile carbonyl compounds	Silicon chip-MS	93.6%	85.6%	EB	^[^159^]
Lung cancer	34/187	Target analysis	Exhaled carbonyl compounds (ECCs)	Silicon chip-MS	94.0%	–	EB	^[^160^]
Lung cancer	52/58	Target analysis	H2O2	TDLS	–	–	EB	^[^161^]
Lung cancer	43/41	Target analysis	VOCs	GC-MS	95.3%	85.4%	EB	^[^129^]
Lung cancer	137/143	Target analysis	VOCs	GC-MS	–	–		^[^162^]
Lung cancer	88/85	Target analysis	VOCs	GC-MS	96.5%	97.5%	EB	^[^163^]
Lung cancer	17/170	Target analysis	VOCs	PTR-MS	54.0%	99.0%	EB	^[^164^]
Non-cystic fibrosis bronchiectatic patients	40/19	Target analysis	Acetate, acetoin, lactate, and citrate	NMR	–	–	EBC	^[^56^]
PLCH	15/20	Target analysis	Acetate, 1-methylimidazole, 2-propanol, and isobutyrate	NMR	–	–	EBC	^[^121^]
Pneumonia	46/47	Target analysis	VOCs	GC-MS	–	–	EB	^[^165^]
Pneumonia	46/NR	Metabolic profiling	VOCs	TD-GC-TOF-MS	98.0%	97.0%	EB	^[^166^]

Non-respiratory diseases

Multi-omics approaches including volatomics for biomarker discovery and target validation in different biological specimens for amylotrophic lateral sclerosis (ALS) were reviewed by Mitropoulos et al.^[167] Bannaga et al.^[168] reviewed VOCs of breath, urine and stool samples of patients with inflammatory bowel diseases and discussed their sensitivity and specificity.

Karyakin et al.^[169] used EBC and whole blood samples to determine glucose in diabetic patients. They used a biosensor based on glucose oxidase for blood and EBC samples. This procedure was performed on the samples collected in a 24-hour period. The results showed that EBC glucose levels correlate positively with blood glucose levels, therefore offering the prospect of a noninvasive approach to monitor diabetes. Furthermore, glucose metabolism to lactate was studied and glucose consumption in EBC was verified by observing the accumulation of lactate in living cells.^[170]

VOCs of children suffering from type 1 diabetes mellitus (N = 53) were investigated in comparison with a matched healthy control group (N = 60).^[26] As shown in Figure 2, higher concentrations of isoprene, pentanol, dimethylsulfide, ethanol and isopropanol were observed in case group. In patient group, exhaled isopropanol was related to HbA_1c (R = 0.57). In addition, exhaled isopropanol and pentanol levels were correlated with blood levels of cholesterol and LDL-cholesterol. Controversially, the research group did not observe the elevated exhaled acetone in diabetic patients whereas, previous investigations reported the elevated exhaled acetone (e.g., Ref. 170–172). Trefz et al.^[26] discussed this controversy by considering a reduction of acetone to isopropanol by nicotinamide adenine dinucleotide dependent redox reaction.

Figure 2. Box plots of exhaled concentrations of acetone, isoprene, pentanal and DMS (A) as well as limonene, ethanol and isopropanol (B). Black box plots: healthy controls; red box plots: T1DM patients; * and # indicate statistically significant differences with p < 0.001 and p = 0.002, respectively. ©Nature Publishing Group. Reproduced with permission Nature Publishing Group.^[26] Permission to reuse must be obtained from the rightsholder.

In another study, amino acid analysis were performed on EBC samples collected from children with leukemia. The differences in the levels of amino acids in patients were assessed, and alterations in the levels of amino acids were found to be specific to leukemia.^[63] The established procedure verifies the possibility of using amino acid profiles as leukemia biomarkers. In a review article, Robles and Priefer^[173] introduced hydrogen breath test as a relatively low cost, available, noninvasive and popular technology to aid in the diagnosis of many gastroenterological diseases with special focus on lactose intolerance. They reported that in various studies, the diagnosis of a positive breath test for lactose intolerance was either based on a rise in >20 ppm from baseline for hydrogen or >10 ppm for methane.

VOCs of exhaled breath of patients with mild- to- moderate kidney disease (N = 48) or patients with a functional renal transplant KTx (N = 8) were compared with those of a matched healthy control group (N = 60). As shown in Figure 3, the levels of ammonia (a key molecule in the urea cycle removing nitrogen from protein metabolism and also a product of bacteria in saliva, respiratory and gastrointestinal tract), ethanol, isoprene (a by-product of cholesterol biosynthesis), pentanal and heptanal (markers of oxidative stress) in patient group were higher than those of healthy group and a reversed pattern was observed for methylamine (product of adrenaline deamination by monoamine oxidase).^[174] Table 2 lists the published works on breathomics studies of non-respiratory diseases.

Figure 3. EBC levels of the biomarkers in healthy control (blue) and chronic kidney disease patients (red). ©Kluwer Academic. Reproduced by permission of Kluwer Academic.^[175] Permission to reuse must be obtained from the rightsholder.

Table 2. Metabolomics studies on non-respiratory diseases using EB or EBC (breathomics) samples.

Disease/case	Number of participants (case/control)	Metabolomic approach	Potential biomarkers	Analytical method	Sensitivity	Specificity	EB or EBC	References
Air pollution exposure	4/1	Metabolic profiling	2,3-Benzofuran, methylbenzaldehyde, n-nitrosopyrrolidine, 2,3-dimethylphenol, 2,3-dimethylphenol,octanal, pentylbenzene), lauric acid, …	ESI-MS	–	–	EBC	^[^58^]
Breast cancer	36/7	Target analysis	Tert-dodecanethiol, 2-ethylhexanethiol, 2-mercaptobenzyl alcohol, 2-mercaptobenzoazole, calixarene, and cubic Pt NPs benzylmercaptan	nanoscale artificial NOSE (TMNA-NOSE)	–	–	EB	^[^175^]
Breast cancer	169/82	Target analysis	VOCs	GC-MS	81.0 to 88.0%	76.0 to 96.0%	EB	^[^176^]
Breast cancer	39/24	Target analysis	VOCs	GC-MS	91.7%	95.8%	EB	^[^177^]
Breast cancer	22/22	Target analysis	VOCs	GC-MS	–	–	EB	^[^151^]
Breast cancer	201/42	Target analysis	VOCs	GC-MS	94.0%	73.8%	EB	^[^178^]
Breast cancer	85/45	Target analysis	VOCs	GC-MS	–	–	EB	^[^179^]
CD	135/110	Target analysis	VOCs	GC-TOF-MS	81.0%	80.0%	EB	^[^180^]
Chronic kidney disease	56/60	Target analysis	Ammonia↑, ethanol↑, isoprene↑, pentanal↑, heptanal↑, methylamine↓	PTR-ToF-MS	–	–	EB	^[^181^]
Colorectal cancer	37/41	Target analysis	VOCs	GC-MS	86.0%	83.0%	EB	^[^182^]
Colorectal cancer	87/122	Target analysis	VOCs	GC-MS	94.0%	91.0%	EB	^[^183^]
Colorectal cancer	26/22	Target analysis	VOCs	GC-MS	–	–	EB	^[^151^]
Colorectal cancer	20/20	Target analysis	VOCs	GC-MS	–	–	EB	^[^184^]
Diabetes mellitus type 2	48/39	Target analysis	Isopropanol, 2,3,4-trimethylhexane, 2,6,8-trimethyldecane, tridecane and undecane	GC-MS	97.9%	100.0%	EB	^[^185^]
Diabetes mellitus type 1	53/60	Target analysis	VOCs	PTR-ToF-MS	–	–	EB	^[^26^]
Esophagogastric cancer	20/10	Target analysis	Ethane	TDLS	–	–	EB	^[^186^]
Esophagogastric cancer	18/17	Target analysis	VOCs	SIFT-MS	–	–	EB	^[^187^]
Esophagogastric cancer	81/129	Target analysis	VOCs	SIFT-MS	86.7-100.0%	81.2-91.2%	EB	^[^188^]
Gastric cancer	69/61	Target analysis	VOCs	GC-MS	–	–	EB	^[^189^]
Gastric cancer	95/30	Target analysis	VOCs	GC-MS	73.0%	98.0%	EB	^[^190^]
Gastric cancer	37/93	Target analysis	VOCs	GC-MS	84.0%	87.0%	EB	^[^191^]
Head and neck cancer	62/19	Target analysis	Ethanol, 2-propenenitrile, and undecane	GC-MS	77.0-100.0%	75.0-91%	EB	^[^192^]
Head and neck cancer Lung cancer	22/40 25/40	Target analysis	VOCs	GC-MS NA-NOSE	100.0%	100.0%	EB	^[^193^]
Head-and-neck cancer	22/40	Target analysis	4,6-Dimethyl-dodecane, 5-Methyl-3-hexanone, 2,2-dimethyl-decane, limonene, 2,2,3-erimethyl, exobicyclo[2.2.1]heptane, and 2,2-dimethyl-propanoic acid	Nanoscale Artificial Nose (NA-NOSE)	100.0%	92.0%	EB	^[^193^]
IBD	62/55	Target analysis	VOCs	SIFT-MS	92.3%	97.1%	EB	^[^194^]
IBD (CD)	68/NR	Target analysis	VOCs	GC-MS	–	–	EB	^[^90^]
IBD (UC and CD)	30/9	Target analysis	VOCs	eNose and GC-IMS	67.0% 87.0%	89.0% 89.0%	EB	^[^195^]
IBD (UC and CD)	33/102 34/102	Target analysis	VOCs	IMR-MS	94.0% for both	71.0% 76.0%	EB	^[^196^]
IBD (UC and CD)	51/14 136/14	Target analysis	VOCs	SIFT-MS	–	–	EB	^[^197^]
IBD (UC and CD)	29/22 25/22	Target analysis	VOCs	FAIMS	74.0%	75.0%	EB	^[^198^]
IBD (UC and CD)	20/18 18/18	Target analysis	VOCs	SIFT-MS	88.9-94.4%	90-94.4%	EB	^[^199^]
IBD (UC)	72/22	Target analysis	VOCs	GC-TOF-MS	92.0%	77.0%	EB	^[^200^]
IPAA, IBD (UC and CD)	30/53 11/53 24/53	Target analysis	VOCs	IFT-MS	–	–	EB	^[^201^]
Laryngeal cancer	21/10	Target analysis	VOCs	GC-MS	–	–	EB	^[^202^]
Liver cancer	57/36	Target analysis	VOCs	GC-MS	86.7%	91.7%	EB	^[^203^]
Lung, breast, colorectal, and prostate cancers	155/22	Metabolic profiling	VOCs	GC-MS	–	–	EB	^[^151^]
MS pateints	34/17	Metabolic profiling	VOCs	GC-MS	85.3%	70.6%	EB	^[^204^]
Menstrual cycle/with and without oral contraception	12/12	Target analysis	VOCs	PTR-ToF-MS	–	–	EB	^[^205^]
Normal volunteers	–	Metabolic profiling	Pentane and isoprene	GC-MS	–	–	EB	^[^206^]
Normal volunteers	12	Metabolic profiling	VOCs	Ion-trap	–	–	EB	^[^207^]
Occupation related disease	20/10	Metabolic profiling	Phenol, formaldehyde and volatile organic compounds, fatty acids and alcohols	NMR	–	–	EBC	^[^55^]
Oral cavity cancer	52/26	Target analysis	VOCs	GC-MS	–	–	EB	^[^208^]
Oral cavity cancer	14/21	Target analysis	Volatile sulfur compounds (VSCs)	Portable GC			EB	^[^209^]
Ovarian cancer	151/48	Target analysis	VOCs	GC-MS	79.0%	100.0%	EB	^[^210^]
Thyroid cancer	64/32	Target analysis	VOCs	GC-MS	92.0%	82.0%	EB	^[^211^]

List of abbreviations used in Tables 1 and 2:

CD: Crohn’s disease; DA: lower breath; FAIMS: field asymmetric ion mobility spectroscopy; FL: Fluorescence; FT-ICR-MS: Fourier transform–ion cyclotron resonance–mass spectrometry; IMS: ion mobility spectrometry; EESI-MS: extractive electrospray ionization mass spectrometry; EET, epoxyeicosatrienoic acid; GC-MS: Gas chromatography-mass spectrometry; GC-TOF-MS: Gas chromatography time-of-flight mass spectrometry; HETE, hydroxyeicosatetraenoic acid; IBD: Inflammatory bowel disease; IMR-MS: Ion molecule reaction mass spectrometry ion mobility spectrometry time-of-flight mass spectrometry, NA, Not applicable; NR; Not reported; PTR-MS: proton transfer reaction–mass spectrometry; SAW: surface acoustic wave; SIFT-MS: selected ion flow tube mass spectrometry; TD-GC-TOF-MS: Thermal-desorption/gas chromatography/time-of-flight mass spectrometry, TM-DART-TWIMS-TOF MS: Transmission-mode direct analysis in real time traveling wave; UC: Ulcerative colitis; UA, upper breath; UPLC-QTOF-MS: Ultraperformance liquid chromatography quadrupole-time-of-flight mass spectrometry.

Animal studies

The determination of breath constituents is of great importance in veterinary.^[46] The discovery of metabolic biomarkers in animals’ EBC may offer valuable insight to their metabolic origin and such results can be cautiously extrapolated or validated in humans for biomarker development. Recently, Carlos et al.^[212] reviewed progress on the detection of canine metabolites. They discussed the potential uses of metabolomics to discover biomarkers of diseases in dogs.

Due to the special anatomy of the respiratory system of cetaceans, EBC samples provide valuable information on body physiology. Aksenov et al.^[213] designed a portable device to trap EBC samples from cetaceans. The schematic representation of the device and the sampling process are presented in Figure 4. The metabolomes of the collected EBC samples from trained and wild dolphins were analyzed by GC-MS and LC-MS. To analyze the metabolites with GC-MS, samples were pre-concentrated by a headspace equipped solid phase micro extraction (SPME) device. However, only a lyophilization step was carried out prior to the LC. After the determination step, results were analyzed by PLS-DA.

Figure 4. Schematic representation of the device for EBC sample collection from cetaceans. ©American Chemical Society. Reproduced by permission of American Chemical Society (https://pubs.acs.org/doi/abs/10.1021/ac5024217).^[213] Permission to reuse must be obtained from the rightsholder.

Results revealed that dolphin EBC contains a mixture of various volatile and nonvolatile compounds in trace concentrations. There was a similarity between the compounds detected from human and dolphins especially regarding the volatile compounds. However, some small amines (1-pentanamine, 5-(hydroxymethyl)-2-pyrrolidinone and bis (2-hydroxypropyl) amine) were reported in dolphins for the first time. Later, Zamuruyev et al.^[214] used the modified version of the device to collect EBC samples from small cetaceans. The fabricated device was used for reproducible EBC collection from bottlenose dolphins (Tursiops truncatus) as small cetaceans. After sampling, the metabolite contents were analyzed by GC-MS and LC-MS for the detection of volatile and nonvolatile compounds, respectively. Before analysis, the metabolites were extracted with polyacrylate based SPME approaches. In comparison with the previously developed device, the improved version could help quantify more than 230 and 2,400 metabolites with GC-MS and LC-MS, respectively.

Pasamontes et al.^[215] determined the metabolite content of wild cetaceans EBCs to investigate a possible relationship between EBC metabolites and pulmonary abnormalities using LC-MS/MS. Results showed meaningful correlations among EBC metabolites and health measures, suggesting possible use of EBC for evaluating health condition of wild cetaceans. Also, Borras et al.^[216] analyzed the EBC samples of bottlenose dolphin (Tursiops truncatus) to monitor their health conditions. The hydrophilic interaction and reverse-phased LC were used to separate and record metabolite peaks to observe any health alterations. Results suggested that various biomarkers and metabolites are related to specific diseases in dolphins.

Cumeras et al.^[217] collected EBCs whales (Eschrichtius robustus) to analyze metabolites important in health condition monitoring of whales. SPME approach was used to extract metabolites from EBC samples, which were then analyzed with GC-MS. About 70 compounds were separated and identified in the whale EBC samples, out of which about 44% were found in healthy human EBCs. In another work, whale hormones were assessed in EBC samples. Testosterone, progesterone, and cortisol concentrations were quantified by enzyme immunoassay kits (EIA) using urea as a normalizing agent. Results reveled that testosterone and cortisol levels were different in individual whales, and could be regarded as potential markers for investigation of adrenal activity of various whales.^[218] Also, Hogg et al.^[219] and Dunstan et al.^[220] evaluated LC-MS technique for determination of progesterone, testosterone and cortisol levels in EBC samples. Compared to LC-MS technique, EIA immunoassay showed better sensitivity with a Limit of Quantitation (LOQ) of 500 pg/mL for all studied hormones.

Asthma is a long-term chronic inflammatory disease of the lung airways which will affect about 400 million people around the world by the year 2025. Annually, about 250 thousand people pass away due to asthma. Hence, it is of great importance reduce asthma incidence, and also to detect this disease at the early stages. New biomarkers emerging from metabolomics studies can result in a decrease in the mortality rate of asthma. Fulcher et al.^[221] utilized NMR to distinguish the differences in the EBC samples of cats before and after asthma induction. Acquired spectra were analyzed with various chemometrics software to find possible biomarkers. Results suggested that acetone levels were increased in EBC samples and about 74% of cats with early asthma contained acetone in their EBCs. Also, the concentration of hydroxyphenyl-containing aromatic compounds was increased, whereas phthalate concentrations were decreased in about 60% of the cats.

Bos et al.^[222] pursued changes in EBC metabolites in two rat groups in which lung injury was induced by intratracheal (IT) and intravenous (IV) lipopolysaccharide (LPS). The mixture of EBC samples were determined by GC-MS. Metabolomic studies showed a decrease in hexanal, 6,10-dimethyl-5,9-undecadien-2-one and pentadecane concentrations for both IV- and IT-induced LPS groups. Based on metabolomic studies, some biomarkers were identified for lung injury and also nonanoic acid was proposed for alveolar neutrophil influx.

Fischer et al.^[223] conducted a study in which the effect of food intake on VOC composition and concentrations of nonvolatile in EBC samples collected from caprine was investigated. GC-MS was used as analytical data acquisition instrument. They showed that the concentration of hydrocarbons and alcohols were different before and after food intake. Also, Küntzel et al.^[224] designed a device for metabolite analysis in bovine EBC samples. They developed a proton transfer reaction TOF-MS(PTR-TOF-MS) device for real-time and direct EBC analysis. Bazzano et al.^[225] determined the metabolite composition of EBC and tracheal wash samples of healthy and asthma-affected horses. After collection, the samples were analyzed with H-NMR, approving eight metabolites of ethanol, methanol, formate, trimethylamine, acetone, acetate, lactate and butanone in EBC samples. Interestingly, the composition of the horse EBC is similar to humans. Neuhaus et al.^[226] compared the metabolites composition of EBC and bronchoalveolar lavage fluid collected from asthma-affected mice. To determine the concentration of metabolites, multi-capillary column ion mobility MS (MCC/IMS) was utilized for noninvasive detection. Histamine, nitric oxide, and arachidonic acid were detected by MCC/IMS.

Conclusions

As conclusion of this subsection, metabolomic studies of animals revealed that there is a significant overlap between human metabolites and those of other animals. The present and future needs of healthcare systems are to employ low cost, easy to use, noninvasive and reliable techniques for disease screening and diagnosis, as well as determining the disease stage. Such techniques would assist clinicians in early detection, accurate diagnosis, successful treatment and follow-up of patients. Metabolomics in general and breathomics in particular provide an opportunity to develop such analytical methods. In this regard breathomics is emerging as a noninvasive and patient-friendly diagnostic tool. In this work, an update on sample collection, metabolomics analysis, advances in analytical techniques, common numerical analyses used in data analysis and some clinical applications were reviewed.

Author contributions

MK and AJ conceived the concept of the review and finalized the manuscript, MA conducted literature search and drafting chemical analysis section, ER conducted literature search and drafting EBC collection section, AAS conducted literature search and drafting numerical analysis section, JS conducted literature search and drafting animal studies section. All authors contributed in final manuscript preparation.

Disclosure statement

The authors declare no conflict of interest.

Additional information

Funding

This work is supported by Pharmaceutical Analysis Research Center, Tabriz University of Medical Sciences, Tabriz, Iran, under grant number of 64109.