GC-MS-based metabolomics of volatile organic compounds in exhaled breath: applications in health and disease. A review

Exhaled breath analysis, with particular emphasis on volatile organic compounds, represents a growing area of clinical research due to its obvious advantages over other diagnostic tests. Numerous pathologies have been extensively investigated for the identification of specific biomarkers in exhalates through metabolomics. However, the transference of breath tests to clinics remains limited, mainly due to deficiency in methodological standardization. Critical steps include the selection of breath sample types, collection devices, and enrichment techniques. GC-MS is the reference analytical technique for the analysis of volatile organic compounds in exhalates, especially during the biomarker discovery phase in metabolomics. This review comprehensively examines and compares metabolomic studies focusing on cancer, lung diseases, and infectious diseases. In addition to delving into the experimental designs reported, it also provides a critical discussion of the methodological aspects, ranging from the experimental design and sample collection to the identification of potential pathology-specific biomarkers.


Introduction 1.Volatile organic compounds
Volatile organic compounds (VOCs) are small molecules (MW <500 Da) with low boiling points and high vapor pressures at ambient temperature.The profile of VOCs released by an organism is called the volatilome, reflecting the metabolic state and playing essential ecological and regulatory roles (Mansurova et al., 2018;Netzker et al., 2020;Sidorova et al., 2021).In humans, VOCs are released through breath, skin, feces, urine, sweat, and saliva, among others (Drabińska et al., 2021), and their origin can be endogenous and exogenous (Pleil et al., 2013).Microorganism-derived VOCs, which include symbionts, commensals, and pathogens, should be considered endogenous since they play significant roles in human health (De Vos et al., 2022).

Breath test along the history
The origin of the breath test can be traced back to ancient Greece.Hippocrates of Kos (460-370 BC) described specific types of odors associated with physiological imbalance, such as fetor hepaticus for liver dysfunction, fetor oris for halitosis, the fruity and sweet odor of patients with uncontrolled diabetes, the urine-like smell of kidney failure, and the putrid stench of lung abscess.Paracelsus, in the 16th century, further emphasized the link between "bad" breath and pathology (Fortes et al., 2017).
In the 18th century, Antoine Lavoisier discovered the role of oxygen in combustion and understood the respiratory physiology in animals (Karamanou and Androutsos, 2013).That was the origin of capnography and modern biochemistry.The sensitive detection of VOCs became possible with the introduction of colorimetry in the mid-19th century.Ethanol was isolated from breath by Francis E. Anstie, and acetone was found increased in the breath of diabetes mellitus patients by A. Nebelthau (Phillips, 1992).
Discoveries made in the 20th century are (Amann et al., 2014): mercaptans were detected in the breath of severe liver disease patients by Davidson (1949), connecting them to the fetor hepaticus described by Hippocrates of Kos; acetonitrile was detected in the breath of smokers by McKee et al. (1962); methanol was found in human breath (Eriksen and Kulkarni, 1963); volatile fatty acids were reported in patients with cirrhosis (Chen et al., 1970); ammonia was measured spectrometrically by Hunt and Williams (1977); and dimethyl-and trimethylamine were detected in the breath of end-stage renal disease patients (Simenhoff et al., 1977).
The turning point came when Pauling et al. (1971) published a pioneering study using gas-liquid partition chromatography to analyze body fluids and breath to investigate the influence of diet on human microbiota and health.This study detected 250 VOCs in human breath, offering promising prospects for further research in the field.

Breath test and clinical applications
Breath samples are particularly valuable for VOCs analysis.The gaseous fraction contains over 1,000 VOCs, with acetone and isoprene being the most abundant (Kuo et al., 2020;Drabińska et al., 2021).
Breath tests aim to distinguish between healthy and pathological states by analyzing exhaled breath VOC profiles, identifying pathology-specific compounds and elucidating their biochemical origin.Compared to routine diagnostic methods, they offer several advantages: they are non-invasive, cost-effective, and fast and easy to perform, have an unlimited sample size, and can be safely and repeatedly collected (Sharma et al., 2023).Despite their simplicity, to date, just a few tests are used in clinical practice, such as the fractional exhaled nitric oxide (FeNO) test for asthma diagnosis, the 13 C-urea breath test for Helicobacter pylori infection, the hydrogen/methane test to detect lactose and/or fructose intolerance, also to detect small intestine bacterial overgrowth, standard capnography based on monitoring CO 2 partial pressure levels during anesthesia and intensive care, and the alcohol breath test used by the police (Simrén and Stotzer, 2006;Buszewski et al., 2007Buszewski et al., , 2013)).
Although many studies propose potential biomarkers for various pathologies, the expected clinical application of breath tests has not progressed as expected (Buszewski et al., 2007;Sharma et al., 2023).Additionally, the link between potential biomarkers and specific pathologies is not clear (Haick et al., 2014;Zou et al., 2022).

Major sources of endogenous VOCs
Oxidative stress (OS) and cytochrome P450 (CYP) enzymes are the main sources of endogenous VOCs.OS damages cellular components, such as phospholipids, proteins, and DNA, thus being involved in the development of many pathological conditions such as cancer, inflammation, and aging.Lipid peroxidation, especially of polyunsaturated fatty acids (PUFAs), is a significant source of VOCs.The breakdown of lipid peroxides produces a wide range of compounds, such as alkanes, alkenes, alcohols, aldehydes, carboxylic acids, esters, epoxides, and furans (Calenic et al., 2015;Ratcliffe et al., 2020).
CYP enzymes participate in reactive oxygen species (ROS) generation and lipid peroxidation, affecting the oxidation-reduction balance and OS, therefore also contributing to VOC generation.CYP enzymes are found in various tissues, with higher levels in the liver and enterocytes (Murray et al., 2009;Veith and Moorthy, 2018;Behrendorff, 2021).

Exhaled breath sampling
The average human expiratory volume is 500 mL, comprising three portions: dead space air, air from the airways and alveoli, and alveolar breath.Capnography can monitor the respiratory cycle, as the CO 2 level shows different trends in each portion.Consequently, breath samples can be classified into three types: mixed breath (containing the three portions), late expiratory breath (excluding dead space air), and alveolar breath (containing only the last portion of the expiration) (Beauchamp and Miekisch, 2020).
There are two types of breath analysis: online and offline.Online provides fast results and allows the volatilome to be monitored with minimal sample manipulation.Nevertheless, offline analysis (storing the sample for subsequent analysis) is the most widely used, as it enables sampling at different locations (Sola-Martínez et al., 2022).Sampling, transport, and storage are critical in the offline analysis of gas samples, since the samples may suffer from possible losses, adsorption, and artifact formation (Alonso and Sanchez, 2013).Therefore, the correct choice of sampling methodology is crucial.
Breath sampling methods can be categorized according to the type of sample collected.Devices employed for mixed breath include containers and bags with a valve system to prevent re-breathing, such as Tedlar ® and Mylar bags, sorbent tubes, canisters, sampling tubes/bulbs, and the Pneumopipe device (Pennazza et al., 2014;White and Fowler, 2019).Although these devices are simpler to use, they may lead to losses, diffusion, adsorption onto the sampling device material, and potential contamination, especially with reactive VOCs (Miekisch et al., 2012;Tang et al., 2015;Beale et al., 2016).In particular, Tedlar ® bags emit contaminants like N,N-dimethylacetamide, phenol, carbonyl sulfide, and carbon disulfide.To preserve sample integrity, storage time should be minimized, and analysis is recommended within 10 h (Beauchamp et al., 2008;Mochalski et al., 2009).
For collecting late expiratory or alveolar breaths, traditional sampling devices present some adaptations, such as a T-shaped mouthpiece, a spirometer system, and CO 2 and pressure sensors (Alonso and Sanchez, 2013;Tang et al., 2015).CO 2 sensors are commonly used for alveolar breath sampling because CO 2 concentrations are highest and constant in the alveolar phase (Lawal et al., 2017).Various devices are available for this purpose, either collecting a final fixed volume, based on the Haldane-Priestly approach, or using CO 2 and pressure sensors: BioVOC ® , RTubeVOC, QuinTron AlveoSampler, ReCIVA, the adaptive breath sampler (ABS), breath collection apparatus (BCA), and SOFIA sampler (Phillips, 1997;Basanta et al., 2007;Beale et al., 2016;White and Fowler, 2019).

Analytical platforms: GC-MS
The concentrations of VOCs in exhalates range from partsper-million (ppmv) to parts-per-trillion (pptv), requiring highly sensitive analytical techniques to detect these compounds.Analytical platforms used for online and offline analysis include laser spectrometry, selected ion flow tube-mass spectrometry (SIFT-MS), proton transfer reaction-mass spectrometry (PTR-MS), secondary electrospray ionizationmass spectrometry (SESI-MS), ion molecule reaction-mass spectrometry (IMR-MS), and ion mobility spectrometry (IMS).These techniques perform fast analysis and present high sensitivity, although they involve high costs and/or require skilled technicians.An alternative method, also emerging as a point-of-care tool, electronic noses (E-nose) combine selective electronic sensors, offering rapid analysis and affordability.Basically, E-noses are used to detect patterns between the samples, which are further resolved through statistical methods and machine learning.Other online approaches such as optical/laser absorption spectroscopy-based methods detect small molecules with narrow adsorption lines, commonly used for acetone analysis.Additionally, compound identification is limited, and no accepted standards ensure interoperability/normalization of methodologies.A promising approach utilizes nanomaterialbased VOC/gas sensors, which offers a wider dynamic detection range and high selectivity; however, some challenges include receptor immobilization compromising functionality, potentially irreversible reactions between VOCs and the receptor (due to high selectivity), and a reduced likelihood of VOC-receptor interaction due to the small surface area of nanoscale elements (Buszewski et al., 2013;Bruderer et al., 2019;Wojnowski et al., 2019;Sharma et al., 2023).
GC-MS is a mature technique that is considered the "gold standard" for VOC analysis in exhaled breath (De Lacy Costello et al., 2014;Drabińska et al., 2021).It offers high sensitivity and reproducibility, and the ability to identify and elucidate unknown compounds, especially with high-resolution instruments (Sola-Martínez et al., 2022;Sharma et al., 2023).In addition to requiring an offline approach such as a pre-concentration step, GC-MS applicability may be hampered by its high costs, complex and time-consuming sampling, requirement for standardization and trained personnel, and inapplicability for online analysis (Xu et al., 2016).Nonetheless, its application to the clinical setting is valuable due to its capabilities in biomarker discovery.

Sample pre-concentration strategies
Exhaled breath samples, especially mixed breath, require enrichment before offline analysis due to low VOC concentration and high water vapor content.Pre-concentration methods usually include two consecutive steps, consisting of trapping VOCs in sorbents followed by their release via thermal desorption.Three main techniques especially suited for GC are used (Figure 2): solidphase microextraction (SMPE), thermal desorption tubes (TD), and needle-trap devices (NTDs) (Lawal et al., 2017;Sola-Martínez et al., 2022).SPME (Figure 2A) was first applied to human breath by Grote and Pawliszyn (1997).Equilibrium is established during sampling based on analyte and sorbent physicochemical properties within the fiber (Beauchamp and Miekisch, 2020).The fiber, coated usually with polydimethylsiloxane (PDMS), Carboxen (Car), or divinylbenzene (DVB), can also have a combination of coatings (Car and/or DVB embedded into PDMS) for a wider chemical species extraction (Trujillo-Rodríguez et al., 2020).Moreover, derivatization reactions can be performed by doping the fiber to increase the affinity of the analyte to the coating (Vas and Vékey, 2004).
TD (Figure 2B) allows longer periods of storage and ease of transport without affecting the sample.The device, composed of a stainless steel or a glass tube, contains sorbent materials like organic polymers (e.g., Tenax TA), graphitized carbon (e.g., Carbopack X) or carbon molecular sieves (e.g., Carboxen).TD can have single-or multi-bed sorbents, with the latter covering a wider range of analytes, but compromising reproducibility due to analyte-sorbent interactions (Lawal et al., 2017;Beauchamp and Miekisch, 2020;Sola-Martínez et al., 2022).
NTD (Figure 2C) is less common but shares similarities with SPME and TD.It uses a needle-shaped device filled with sorbent materials to capture compounds by drawing breath through the needle.Similar to SPME, NTD requires a small sample volume, although the sensitivity is volume dependent as in TD (Trefz et al., 2012).Storage and transportation are also similar to that of TD (Lawal et al., 2017).

Metabolomics
Metabolomics has gained significant attention in clinical research, providing insights into the pathological pathways of various pathologies.These studies can be broadly categorized into two approaches: untargeted and targeted.Untargeted metabolomics is the non-biased approach, which aims to study as many metabolites as possible to discover changes among the groups of samples, while targeted metabolomics focuses on specific metabolites, offering better sensitivity and specificity.Combining both approaches allows for hypothesis generation (untargeted) and the validation of findings (targeted).Workflows and methodologies for both approaches have subtle differences (Patti et al., 2012).

Objectives and literature search
This review aims to identify potential VOC biomarkers that are consistent across different pathologies and to consolidate and discuss the methodologies employed for exhaled breath sampling and analysis.To achieve this, a literature search was conducted, focusing on studies that analyzed human exhaled breath by GC-MS published since 2012.The search strategy utilized specific keywords, such as "volatile organic compounds," "exhaled breath" or "breath test," "gas chromatography," and "mass spectrometry."The databases employed were Scopus and Web of Science.Initially, 377 articles were obtained, which were then narrowed down to 152 after title and abstract evaluation, and the articles were sorted according to the pathology studied.Finally, 70 articles focusing on 10 pathologies of significant interest were included in this review, and categorized in: cancer (such as lung, gastric, colorectal, and breast cancers), other pulmonary pathologies (comprising asthma, COPD, OSA, and cystic fibrosis), and infectious pathologies (encompassing community-acquired pneumonia (CAP)/hospitalacquired pneumonia (HAP)/ventilator-associated pneumonia (VAP) and COVID-19).

VOCs in exhaled breath in health and pathology
In the following sections, selected studies for each pathology are discussed, along with the identified candidate VOCs reported as pathology-specific biomarkers.Figure 3 illustrates the distribution of studies, showing that lung cancer has been the most extensively studied, followed by asthma, COPD, and CAP/HAP/VAP.
Low-dose computed tomography (LDCT) is the main screening tool, although it exhibits a high false-positive rate (Nooreldeen and Bach, 2021).Lung tissue biopsy, the gold standard procedure for diagnosis, determines malignancy, histological type, and TNM (tumor, nodule, and metastases) stage.However, this procedure is highly invasive and can lead to complications, such as pneumothorax and pneumonia (Zhang et al., 2020b).Indeed, the development of rapid and non-invasive early diagnostic tests is urgently required, and breath tests offer promising alternatives.Among the pathologies studied in exhaled breath, LC is the most prevalent, as this pathology is directly related to the respiratory tract.
Twenty-three studies focusing on potential biomarkers for LC are summarized in Supplementary Tables S1-S3, referring to metabolomic methodology, group comparisons, and VOC biomarkers, respectively.A Chinese group performed two untargeted studies on the same data, comparing LC patients and healthy controls (HCs).The first study (Zou et al., 2021) developed a prediction model based on the whole breath profile (308 peaks), achieving 85.0% accuracy, 83.0% sensitivity, and 85.0% specificity.Twenty-two discriminative VOCs were annotated, styrene being also found downregulated in LC patients who responded partially to treatment or remained stable (Supplementary Table S3), along with two other VOCs (dodecane, 4-methyl and α-phellandrene) (Nardi-Agmon et al., 2016).The second study (Zou et al., 2022) selected 31 VOCs as biomarkers in the univariate analysis (UVA), which showed 0.787 AUC in the multivariate analysis (MVA) after crossvalidation.Additionally, eight VOCs were found to be involved in a total of 18 metabolic pathways, of which 11 were significantly altered.
A Polish group compared LC patients with HCs.Buszewski et al. (2012) divided both groups according to smoking habits, identifying 12 significant VOCs between non-smokers, 7 being upregulated when compared to active smokers.Rudnicka et al. (2014) measured 43 VOCs and developed a model with 88 features, yielding 0.970 AUC, 74.0% sensitivity, and 73.0%specificity, with dimethyl sulfide as the main discriminating VOC.Ligor et al. (2015) applied machine learning algorithms, and the final model formed by eight compounds showed an value (e.i.0.650) AUC.In a subsequent study (Rudnicka et al., 2019), the model containing seven VOCs selected from the UVA presented an improved performance, showing 86.4% sensitivity and specificity in the test

FIGURE 3
Pie plot depicting the proportion of studies for each pathology included in this review.CAP, community-acquired pneumonia; COPD, chronic obstructive pulmonary disease; HAP, hospitalacquired pneumonia; OSA, obstructive sleep apnea; VAP, ventilator-associated pneumonia.
group.Twelve VOCs were found in common between these four abovementioned studies (Supplementary Table S3).Schallschmidt et al. (2016) focused on 24 VOCs previously selected as potential LC biomarkers, 20 being also reported in other studies (Supplementary Table S3).In the UVA, 11 and 7 VOCs were significantly altered between LC patients and HCs (non-smokers and active smokers, respectively), 8 VOCs seemingly unrelated to smoking.Moreover, four models were constructed with different subsets of the targeted VOCs, achieving the highest sensitivity (92.0%) with a subset of four VOCs, and the highest specificity (96.0%) with seven VOCs.Ethanol and octane were two target VOCs proposed as potential biomarkers in other studies (Supplementary Table S3).Sakumura et al. (2017) reported ethanol, along with other four VOCs, in a study classifying LC and HCs using a support vector machine (SVM) algorithm, achieving 89.0% accuracy, a 94.4% truepositive ratio, and a 89.7% true-negative ratio when combining different subsets of five VOCs.Furthermore, the distance to the SVM classification boundary provided information on the cancer stage, with early-stage LC located closer to the boundary than advanced-stage LC.
Two research groups from China and Greece conducted several studies comparing LC patients, pulmonary non-malignant disease (PNMD) patients, and HCs.The Chinese group conducted 4 studies, sharing 27 VOCs (Supplementary Table S3).Wang et al. (2012) found 23 significant VOCs with AUCs >0.6, unrelated to smoking, as potential biomarkers, of which five VOCs were significant between squamous carcinoma and adenocarcinoma LC patients.The discrimination model for LC, PNMD, and HCs could correctly classify 96.5% of LCs.Zou et al. (2014) selected five VOCs as LCspecific biomarkers, achieving AUCs ranging from 0.672 to 1 in a validation cohort, with hexadecanal being the most discriminative.Additionally, Chen et al. (2021) annotated 19 VOCs that could discriminate LC from PNMD, as well as 20 VOCs that differentiated LC from HCs with AUCs of 0.809 and 0.987, respectively.Moreover, LC patients could be distinguished by histology (NSCLC and SCLC) using 20 VOCs value (e.i.0.939) AUC and stage (early and advanced) with 19 VOCs value (e.i.0.827) AUC.The Greek group used both targeted and untargeted approaches on the same data set.The targeted study (Koureas et al., 2020) included 19 VOCs, of which 17 VOCs were also found in other studies (Supplementary Table S3).In the UVA, seven VOCs showed significance when comparing LC, PNMD, and HCs, although no single VOC was altered between LC and PNMD.However, LC and HCs were correctly classified by either including 19 VOCs, nine VOCs selected in the UVA (LC vs. HCs), or a subset of VOCs identified by feature selection (FS) (AUCs 0.769-0.970).In the untargeted study (Koureas et al., 2021), 29 features were considered for the analysis, 18 features (12 VOCs annotated) showing significance between LC and HCs, and only 2 (1 VOC annotated) among LC and PNMD.Moreover, LC and HCs were correctly classified using either 29 features or a subset of eight features identified by FS (AUCs 0.940 and 0.960, respectively).In the case of LC and PNMD, three VOCs achieved 75.0%discrimination accuracy value (e.i.0.820) AUC.Among the features/VOCs from both approaches, three VOCs (one from the targeted and two from the untargeted) achieved an accuracy of 72.0% in discriminating LC and PNMD value (e.i.0.780) AUC.However, the VOC from the targeted study was detected in extremely low frequencies.Another targeted study focusing on 21 VOCs identified four upregulated VOCs in LC compared to PNMD (Corradi et al., 2015), of which two (hexane and ethylbenzene) were also included in the targeted study by Koureas et al. (2020) (Supplementary Table S3), showing elevated increased levels in adenocarcinoma LC (hexane) and in advancedstage LC (ethylbenzene) patients.
Furthermore, both untargeted and targeted approaches were performed in the same study comparing LC, COPD, asthmatic patients, and HCs.Monedeiro et al. (2021) built an RF model with the 12 most important VOCs from the untargeted analysis, achieving an overall accuracy of 85.7%.In the following targeted approach, 29 VOCs were preselected, of which 9 were used to build the classification model that provided 91.0% overall accuracy.Additionally, Callol-Sanchez et al. (2017) identified nonanoic acid significantly altered in LC patients compared to both COPD patients and HCs in a targeted study, and Muñoz-Lucas et al. (2020) found elevated levels of propionic acid in LC patients with COPD, mainly detected in advanced-stage LC.

Gastric cancer
Gastric cancer (GaC) is among the five deadliest cancers in 2020, according to the World Health Organization (WHO) (Cancer, 2023).The 5-year relative survival rate is 72% when localized and decreases to 6% when distant at the time of diagnosis (2012-2018) (American Cancer Society, 2017).The main risk factors for GaC, which is predominantly sporadic (90%), include smoking, high meat intake, alcohol consumption, obesity, and Helicobacter pylori infection (Conti et al., 2023).Persistent H. pylori infection causes chronic inflammation, leading to precursor lesions associated with GaC: atrophy, metaplasia, dysplasia, and carcinoma (Conti et al., 2023).
The gold standard diagnostic technique is upper endoscopy, followed by a biopsy, although it is invasive and requires specialists (Hamashima, 2016).While high-incidence countries have implemented screening programs, low-incidence countries require cost-effective alternatives (Herrera-Pariente et al., 2021).Serum biomarkers, which include carcinoembryonic antigen (CEA), alpha-fetoprotein (AFP), and carbohydrate antigens (CA19-9 or CA72-4), have been used for early diagnosis, but their lack of specificity results in low positive rates and the inability to detect precancerous lesions (Feng et al., 2017).
Five studies focusing on biomarkers for GaC, all employing an untargeted metabolomics approach, are included in Supplementary Tables S1-S3.Two studies, conducted in China and Latvia, compared GaC patients, peptic ulcer disease (PUD) patients, and controls.Xu et al. (2013) identified three upregulated VOCs in GaC and four VOCs upregulated in PUD compared to HCs, with one VOC (furfural) shared among comparisons.Likewise, Amal et al. (2013) found four VOCs upregulated in GaC of which two were also upregulated in PUD.However, no single discriminating VOC between GaC and PUD was identified in any of the studies.Only one VOC was found to be common to different geographical areas, 6-methyl-5-hepten-2-one (Supplementary Table S3).Tong et al. (2017) reported 11 candidate GaC biomarkers comparing GaC patients with PUD, gastritis patients, and HCs, using UVA and MVA.One VOC, nonanal, was also found by Amal et al. (2013) to be significantly altered between GaC, PUD, and an additional group stratified based on the operative link on gastric intestinal metaplasia (OLGIM), which classifies patients according to the presence/absence and stage of precancerous lesions.Among the multiple comparisons, eight VOCs showed alterations among groups, seven of which were upregulated in GaC compared to OLGIM, only one VOC being altered between GaC and OLGIM III-IV, and three VOCs in PUD compared to OLGIM.
Lastly, Bhandari et al. (2023) explored the correlation between the fecal microbiome and exhaled breath VOCs.Two VOCs (1octanol and dioctyl ether) were significantly altered and exclusively present in GaC.Moreover, 14 VOCs from GaC patients were correlated with 33 fecal bacterial taxa, and 7 VOCs from HCs were correlated with 17 bacterial taxa, with no common VOCs between groups.

Colorectal cancer
Colorectal cancer (CRC) ranks among the most common cancers worldwide (Ferlay et al., 2021).The 5-year relative survival drops from 65% to 15.6% when diagnosed at later stages (2013)(2014)(2015)(2016)(2017)(2018)(2019), which represents 23% of cases, as early symptoms are not pathology-specific (Colorectal Cancer-Cancer Stat Facts, 2023).The most applied screening tools are the fecal immunochemical test (FIT) and colonoscopy (Helsingen and Kalager, 2022).The FIT test is based on the measurement of the amount of hemoglobin in feces, and one-third of stage I cancers are missed (Niedermaier et al., 2020).Colonoscopy, while effective, is invasive, time-consuming, and expensive and is performed with conscious sedation, carrying the risk of colonic perforation and major bleeding (Qaseem et al., 2019;Helsingen and Kalager, 2022).Other CRC screening tests such as the guaiac-based fecal occult blood test (gFOBT), sigmoidoscopy, fecal biomarker panel test, and computed tomography (CT) colonography have several limitations, such as false-positive results, invasiveness, and high cost (Qaseem et al., 2019).
Five studies focusing on potential biomarkers for CRC are summarized in Supplementary Tables S1-S3, all employing an untargeted metabolomics approach.The studies that analyzed mixed breath sampled the same cohort of CRC patients.The first study (Altomare et al., 2013) compared CRC and HCs by selecting a pattern of 15 VOCs by UVA to construct the probabilistic neural network (PNN) model, which yielded 76.0% accuracy in the validation cohort.In a subsequent study (Altomare et al., 2015), the data were reprocessed, and 32 of 52 CRC patients were resampled after cancer removal.The PNN model was constructed with 31 VOCs selected by UVA, yielding 97.5% and 97.7% accuracies discriminating pre-and post-surgery CRC patients, and post-surgery CRC and HCs, respectively.Additionally, 11 VOCs shared with the previous study could discriminate pre-and post-surgery CRC patients with 98.8% accuracy.These results demonstrate the metabolic change in exhaled VOC patterns due to cancer cell metabolism and suggest that metabolism does not return to the pre-cancer state after cancer removal.
Another study by the aforementioned research group investigated potential biomarkers of cancer stages (early/I-II or advanced/III-IV) (Altomare et al., 2020).Fifteen VOCs were selected by UVA comparing CRC and HCs, to build a model that included age class (>65 vs. ≤65 year olds).Fourteen identified VOCs could discriminate CRC from HCs, with a 93.0%overall positive predictive value (PPV) after cross-validation, whereas eight and five VOCs could discriminate early-CRC from HCs with an 86.0%PPV and advanced-CRC from HCs with a 91.0%PPV.Three common VOCs between UVA and MVA, namely, ethylbenzene, methylbenzene, and tetradecane, were quantified to establish the threshold concentration values.However, none of these compounds were reported in other studies.Nevertheless, five out of the 15 VOCs were common with previous studies (Altomare et al., 2013(Altomare et al., , 2015)), and three were reported as significantly altered between CRC and HCs: 4methyloctane and ethanol (research group from Latvia) (Amal et al., 2016) and dodecane (research group from China) (Wang et al., 2014a) (Supplementary Table S3).
Likewise, Amal et al. (2016) found four significantly altered VOCs between CRC and HCs, which were identified by UVA and subsequently quantified: 4-methyloctane and ethanol were downregulated, whereas acetone and ethyl acetate were upregulated.Likewise, Wang et al. (2014a) found nine potential biomarkers (eight upregulated and one downregulated) for CRC patients with adenocarcinoma by MVA.

Breast cancer
Breast cancer (BC) is the most diagnosed type of cancer and the fifth cause of cancer-related mortality (Cancer;Ferlay et al., 2021).While the 5-year relative survival rate stands at 90.8%, it drops dramatically to 31% when diagnosed at a distant stage (2012-2019) (Female Breast Cancer, 2023).The current gold standard screening methods include annual mammography and clinical breast examination for women over the age of 40.Unfortunately, physical breast examinations, even when performed by a physician, fail to reduce mortality (Barba et al., 2021).Regarding mammography, the sensitivity is compromised by breast density (Boyd et al., 2007), and the procedure requires X-ray examination and may lead to overdiagnosis, resulting in unnecessary procedures and treatments (Løberg et al., 2015).Alternative screening approaches, such as digital breast tomosynthesis (DBT), ultrasonography, magnetic resonance imaging (MRI), and positron emission tomography/computed tomography (PET/CT), are hampered by high costs, discomfort, the requirement for trained technicians, and radiation exposure (Barba et al., 2021).Therefore, there is an urgent requirement for innovative screening tools that can overcome these drawbacks, and breath tests show promise as a potential approach.
Four studies focusing on potential BC biomarkers are summarized in Supplementary Tables S1-S3.These studies were conducted in the same geographical area (China).The targeted study by Li et al. (2014) focused on four aldehydes and their potential to discriminate between BC patients, breast nonmalignant disease (BNMD) patients, and HCs.All the targeted aldehydes were significantly upregulated in BC, while hexanal was upregulated in BNMD, both compared to HC.Furthermore, nonanal was increased in BC when compared to BNMD.The combination of these VOCs showed 91.7% sensitivity and 95.8% specificity (0.934 AUC) in discriminating early-stage BC from HCs, and the predictive model achieved 80.4% correct classification after leave-one-out cross-validation (LOOCV).Hexanal was also identified as a potential biomarker in a different study (Supplementary Table S3).
Two untargeted studies compared BC with HCs and BNMD.Barash et al. (2015) identified 23 VOCs by UVA, 21 of which showed significant differences between HCs and patients with breast lesions [BC, BNMD, and an additional group of patients with ductal carcinoma in situ (DCIS)], and four VOCs were significant between BC and DCIS.The MVA revealed 14 VOCs that could discriminate BC from HC and BNMD, and from DCIS, yielding 72.0% and 81.0%accuracies after LOOCV, respectively.Additionally, two of these 14 VOCs were consistent with findings from other studies (Supplementary Table S3).Wang et al. (2014b) annotated 28 potential biomarkers, of which 21, 6, and 8 VOCs were significantly altered in BC when compared separately to HCs, BNMD (cyclomastopathy and mammary gland fibroma), and DCIS, respectively.Among these, three VOCs, namely, cyclohexanone, 1,4-dimethoxy-2,3-butanediol, and 2,5,6trimethyloctane, were upregulated in BC compared to both HCs and BNMD.Only cyclohexanone was again reported by Zhang et al. (2020a) (Supplementary Table S3).

Asthma
Asthma is a chronic and heterogeneous lung pathology characterized by inflammation and airway obstruction, manifesting with variable symptoms that include cough, wheezing, shortness of breath, and chest tightness (Asthma, 2023;Asthma-Diagnosis, 2023).This pathology places a significant economic burden on healthcare systems, affecting approximately 292 million people worldwide, typically being developed during childhood.Asthma's impact on patients' quality of life and the risk of premature death are major concerns (The Global Asthma Report, 2022).
Diagnosis relies on spirometry, bronchoprovocation tests, peak expiratory flow tests, allergy skin or blood tests, and FeNO tests (Asthma-Diagnosis, 2023).Patients may experience a loss of pathology control and acute exacerbations of symptoms, leading to significant morbidity and a progressive loss of lung function (Castillo et al., 2017).Moreover, the heterogeneity of asthma, concerning severity and response to treatment, is a consequence of the underlying pathophysiological mechanisms.Patients can be classified into different phenotypes based on observable characteristics (steroid response, obesity, allergies, etc.), or endotypes based on the underlying cellular and molecular mechanisms (Kuruvilla et al., 2019).In this context, breath tests offer a non-invasive and easy-to-perform approach for early diagnosis and exacerbation prediction, especially suitable for children, and could be also used to define phenotypes and endotypes by analyzing the profile of endogenous VOCs, which reflects the inflammatory state of the bronchia and underlying molecular mechanisms involved, allowing a significant improvement in treatment effectiveness.
Eleven untargeted studies focused on asthma are included in Tables 1, 2; Supplementary Table S4.Several studies were conducted by the same research group focusing on asthmatic children.Van Vliet et al. (2016, 2017) studied the loss of asthma control and exacerbation episodes over a period of 1 year.In the first study (Van Vliet et al., 2016), a combination of 15 VOCs (10 annotated) showed 86.0% accuracy in classifying persistently controlled and uncontrolled asthma, although no association was found between different exhaled inflammatory markers [FeNO, exhaled breath condensate (EBC), and VOCs] and asthma control.Subsequently (Van Vliet et al., 2017), in a larger cohort of asthmatic children, the combination of seven VOCs used to construct the RF model could predict 88.0% of asthma exacerbation episodes within 14 days.These two studies shared only two VOCs: 1,2-dimethylcyclohexane and 2methylfuran (Table 2).Additionally, Robroeks et al. (2013) annotated 30 VOCs related to asthma exacerbation, and the models combining six and seven VOCs could correctly classify 96.0% of baseline and exacerbation samples taken from the same patient (100% sensitivity and 93.0%specificity) and 91.0% of patients who would have future exacerbations or not, respectively.These results suggest that the profile of VOCs can identify exacerbations and could be used to predict which patients will suffer these episodes.Additionally, Smolinska et al. (2014) studied a cohort of wheezing children with HCs between the ages of 2 and 4 years until the age of 6 years, to find potential biomarkers for preclinical asthma.A total of 17 VOCs (13 annotated) were selected by comparing asthmatic children with HCs and with transient wheezers, which could correctly classify 80.0% of the wheezing children at inclusion, differentiating those who would develop asthma from those who were transient wheezers.Notably, three VOCs reported in these studies (2-methylfuran, 3methylfuran, and m-cymene) were also identified by Monedeiro et al. (2021) when comparing LC, COPD asthmatic patients, and HCs.In this study, the model built with 12 VOCs from the untargeted data presented 85.7% overall accuracy, and another with 9 of the 29 targeted VOCs provided 91.0% overall accuracy.
Likewise, two other studies included a cohort of asthmatic children, in this case, compared to HC. Gahleitner et al. (2013) identified a panel of eight candidate VOCs, all of which were upregulated in asthmatic children.Moreover, Caldeira et al. (2012) built a model with the full data set of metabolites ( 134), yielding a classification rate of 98.0% (96.0%sensitivity and 95.0% specificity).Among these metabolites, six alkanes were related to allergic asthma and four aldehydes and one alkene to HC.The new model that included nine alkanes and aldehydes showed a classification rate of 96.0% (98.0%sensitivity and 93.0%specificity).One VOC from the latter study, decane, was also reported by Sola-Martínez et al. (2021).In this study, a population of women 3 months postpartum was recruited and divided into asthmatics with other coexisting atopic diseases (A-AD) and non-asthmatics, and the latter were further divided into those with and without other atopic diseases (NA-AD and NA-NAD, respectively).Several models were built to compare the different groups, selecting a total of nine VOCs, which could discriminate between asthmatic and non-asthmatic patients, even in the validation cohort (AUCs 0.670-0.900,71.0%-100% sensitivity, and 60.0%-70.0%specificity), although the accuracy decreased when asthmatic patients were compared to the nonasthmatic groups separately (AUCs 0.680-0.810for NA-AD and 0.603-0.750for NA-NAD).
Furthermore, two articles studied asthma phenotypes and endotypes.Schleich et al. (2019) conducted a study on a group of asthmatic patients classified by inflammatory subtypes.From all binary comparisons, 12 VOCs were selected, of which eight were identified as candidate biomarkers.Among them, two VOCs (hexane and 2-hexanone), along with 1-propanol, were selected from the comparison between eosinophilic and paucigranulocytic value (e.i.0.680) AUC.Meanwhile, the comparison between neutrophilic and paucigranulocytic yielded two VOCs (3-tetradecene and pentadecene) in the discovery phase and another two (undecane and nonanal) in the replication value (e.i.0.850 and 0.700, respectively) AUC.Furthermore, when comparing neutrophilic to eosinophilic, three VOCs (3,7-dimethylnonane, 1-propanol, and nonanal) were identified in the discovery phase value (e.i.0.920) AUC, although only nonanal, along with hexane, showed the best classification performance in the replication phase value (e.i.0.710) AUC.As a result, two (hexane and 2-hexanone) and three (nonanal, 1-propanol, and hexane) VOCs could discriminate eosinophilic and neutrophilic asthma from other phenotypes value (e.i.0.720 and 0.730, respectively) AUC.Moreover, Meyer et al. (2014), besides building a model based on 16 VOCs that could discriminate asthmatic patients from HC (100% sensitivity and 91.1% specificity), performed a cluster analysis that included clinical, medication features, and four VOCs that were only present in asthmatic patients, to identify different asthma endotypes.As a result, seven clusters were formed, two with non-allergic asthma and five with allergic asthma.Some clusters presented high clinical similarity but different profiles of VOCs, as well as similar profiles and different clinical symptoms.Although no common VOCs were found between these two studies, eight VOCs were shared with others (Table 2).

COPD
COPD is characterized by chronic respiratory symptoms, such as dyspnea, cough, production of sputum, and/or exacerbations, caused by abnormalities in the airways (bronchitis and bronchiolitis) and/or the alveoli (emphysema), resulting in persistent and progressive airflow obstruction.The causes of the pathology are environmental exposures (tobacco smoking, toxic particles, and gases) and/or genetic risk factors.According to the WHO, 3.23 million people died from COPD in 2019, with 90% of deaths (under the age of 70) occurring in low-and middleincome countries.COPD often coexists with chronic pathologies, such as lung infections and cancer, heart problems, depression, and anxiety (GOLDCOPD, 2023).
COPD diagnosis relies on spirometry, with weak specificity.Additional tests, lung imaging and arterial blood gas tests, can help assess pathology severity.The symptoms develop slowly, and even though COPD is not curable, different treatments can be applied.However, under-or misdiagnosis can lead to lack/incorrect treatment (GOLDCOPD, 2023), and most patients are diagnosed when the lung damage is irreversible (Fazleen and Wilkinson, 2020).Detecting early or pre-COPD cases, where clinical signs are absent or airflow obstruction is not evident in spirometry, can be challenging.Breath tests offer a valuable tool for identifying these cases that diagnostic tests may miss.
Four studies compared COPD patients with HCs, two being performed in the same geographical area (UK), in different research groups.Phillips et al. (2012) applied different machine learning methods, which included a step of FS, to compare the whole group of COPD with HCs, active with former smokers within the COPD group, and COPD with HCs (non-smokers).Of the automatically generated VOCs in the three comparisons (12, 13, and 10, respectively), six overlapped.Likewise, two of these six shared VOCs were reported by Gaida et al. (2016), and another six VOCs were reported in different studies (Table 2).Moreover, Basanta et al. (2012) built a classification model containing 11 VOCs after data reduction (UVA and PCA), with an accuracy of 70.0%.The groups were further divided and compared by smoking status, improving the performance of the model, especially when active smokers were compared (91.0%accuracy).Furthermore, four VOCs were correlated with sputum eosinophils ≥1%, one VOC with sputum eosinophils ≥2%, and four VOCs with exacerbation episodes (≥2/year).The prediction models showed an accuracy of 75.0% and 88.0% for sputum eosinophils ≥1% and sputum eosinophils ≥2%, respectively, and 83.0% for exacerbations, after LOOCV.Of these 11 VOCs, 3 aldehydes (decanal, hexanal, and nonanal) were shared with Jareño-Esteban et al. ( 2017) and Phillips et al. (2012) (Table 2).Additionally, Gaida et al. (2016) studied two cohorts of COPD patients and HCs from different locations, which were split by smoking habits.Overall, 14 VOCs showed potential as COPD biomarkers, with 4 being reported also by Phillips et al. (2012) (Table 2).
The study by Monedeiro et al. (2021) was previously mentioned in the LC/asthma section, with untargeted and targeted analyses to build classification models that distinguish COPD, LC, asthma patients, and HCs, yielding 85.7% and 91% overall accuracy (untargeted and targeted, respectively).Two of these VOCs (isopropyl alcohol and limonene) were shared with Cazzola et al. (2015), and two additional VOCs were common with other studies (Table 2).
The remaining studies focused on COPD exacerbations.Pizzini et al. (2018) applied UVA and post hoc analysis between pairwise combinations, resulting in 12 significant VOCs.Additionally, four VOCs were classified as discriminative for acute exacerbation (A) COPD, two VOCs were classified as discriminative for stable (S) COPD, and two VOCs as associated with COPD.The RF model containing these 12 VOCs could classify COPD patients value (e.i.0.970, 78.0% sensitivity, and 91.0%specificity) AUC.Meanwhile, van Velzen et al. (2019) sampled the same cohort of COPD patients before (baseline), during, and after (recovery) an exacerbation episode.The UVA between the Clinical COPD Questionnaire (CCQ) symptom scores and VOCs resulted in 10 discriminative compounds.The subsequent MVA discriminated between baseline and exacerbation and between exacerbation and recovery with accuracies of 71.0% and 75.0%, respectively.

OSA
OSA is a respiratory disorder with an incidence of 24% in men and 9% in women (30-60 years of age), affecting nearly 1 billion people worldwide (Lv et al., 2023).OSA is characterized by the repeated collapse of the pharynx, leading to episodes of apnea or hypopnea accompanied by decreased oxygen levels and interruptions in sleep.It is associated with poor sleep quality and daytime sleepiness, as well as an increased risk for several metabolic and cardiovascular pathologies (arterial hypertension, diabetes, etc.), and depression (Lévy et al., 2015;Schwarz et al., 2017;Nowak et al., 2021).
Current OSA diagnosis relies on sleep examination (monitoring sleep stages and cycles), mainly through polysomnography, a costly, time-consuming, and inconvenient test.Although home tests are available, these devices are subject to more measurement errors compared to polysomnography (Kapur et al., 2017).Moreover, several nights should be monitored to obtain a more reliable diagnosis (Stöberl et al., 2017).Therefore, breath tests are presented as a potential tool for both the screening and diagnosis of OSA.
Two independent targeted studies focusing on OSA are included in Table 1 and Supplementary Table S4.Bayrakli et al. (2016) studied the levels of acetone and butanol in patients before and after sleep.Although butanol was upregulated in patients compared to HC (after sleep), this VOC was not significantly increased between patients (before vs. after sleep).Conversely, Aoki et al. (2017) focused on 14 VOCs, which included aromatic, alicyclic, chain hydrocarbons, isoprene, acetone, and ethanol, and classified OSA patients into moderate, severe, and most severe in terms of the apnea-hypopnea index (AHI).The UVA yielded four VOCs upregulated in all OSA patients, four VOCs in severe and most severe OSA patients, and three VOCs exclusively in the most severe OSA patients compared to HCs.Furthermore, four of these VOCs (ethylbenzene, p-xylene, phenylacetic acid, and nonane) showed increased levels according to OSA severity, being correlated with the AHI, arousal index, and duration of percutaneous oxygen saturation (SpO 2 ) ≤ 90%.Additionally, the levels of acetone and isoprene decreased after continuous positive airway pressure treatment.Nevertheless, no common VOCs were found between these two studies.

Cystic fibrosis
Cystic fibrosis (CF) is an autosomal recessive genetic pathology caused by a mutation in the cystic fibrosis transmembrane conductance regulator (CFTR) gene.This mutation disrupts the cells' electrolyte transport system, affecting mainly organs with secretory functions, such as as the lungs, pancreas, and reproductive system (Cystic Fibrosis-Causes, 2023).In the lungs, altered sodium absorption results in thick, hardened secretions, increasing the risk of respiratory infections, inflammation, and oxidative stress (Roesch et al., 2018).Pulmonary exacerbations (PEx) are frequent events in the progression of the pathology, potentially leading to permanent lung function loss, reduced quality of life, and decreased survival.PEx treatment includes antibiotics, but delayed symptom onset worsens outcomes (Goss, 2019).The identification of PEx relies on symptomatology, clinical evaluation, and the measurement of changes in forced expiratory volume in one second (FEV1pp) using spirometry devices (Goss, 2019).The use of breath tests to predict PEx in CF is a promising approach.In two independent untargeted studies, CF PEx in children was studied, as shown in Table 1; Supplementary Table S4.van Horck et al. (2021) performed a 1-year observational pilot study, recruiting patients from three different centers.The RF model with the nine most discriminating VOCs could predict 79.0% of patients with stable or upcoming PEx (within 7 days) (79.0%sensitivity and 78.0%specificity).However, no single VOC was found significantly altered when applying UVA between stable and CF PEx patients.Meanwhile, Woollam et al. (2022b) divided the CF patients into CF baseline (not suffering from PEx) and CF PEx.Four VOCs were found to be correlated with FEV1pp at the time of breath collection, of which two VOCs (4-methyl-octane and 3,7dimethyldecane) were further correlated with changes in FEV1pp.Moreover, four VOCs were found to be significantly different between CF baseline and CF PEx patients: 3,7dimethyldecane, durene, and 5-methyltridecane were downregulated, and 2,4,4-trimethyl-1,3-pentanediol 1-isobutyrate was upregulated in PEx patients.Although both studies aimed to identify differential VOCs between CF stable and CF PEx patients, none of the reported were shared.
Current diagnostics rely on clinical, radiological, and microbiological cultures of respiratory samples [endotracheal aspirates, bronchoalveolar lavage (BAL), and protected specimen brush], which present high inter-variability and moderate sensitivity and specificity.The microbiological confirmation can take several days, leading to overtreatment with antibiotics until the specific pathogen is identified (Fernando et al., 2020;Modi and Kovacs, 2020).Therefore, there is an urgent requirement for less invasive and faster diagnostic techniques.
In the case of VAP, van Oort et al. (2017a) presented a protocol for a prospective multicenter study named BreathDx (Molecular Analysis of Exhaled Breath as Diagnostic Test for Ventilator-Associated Pneumonia), aiming to develop a breath test capable of distinguishing suspected VAP patients, with a 99% sensitivity for culture-positive cases.It also aimed to identify unique VOC patterns that could predict specific pathogen infections, holding promise for more efficient VAP diagnosis and treatment.
Seven studies focusing on VAP, and one on CAP/HAP, are summarized in Tables 3-5.To date, two studies have been conducted in relation to BreathDx.Van Oort et al. ( 2022) performed an untargeted study within a group of intubated and ventilated ICU patients with suspected VAP, further divided into culture-positive (CP) and culture-negative (CN) BAL samples.Moreover, two platforms were used to cover a wider range of compounds: GC-MS-1 for more volatile compounds and GC-MS-2 for heavier and cyclic volatile compounds.The discriminative model that included 20 VOCs previously selected by UVA and MVA showed 0.830-0.870AUCs, even when applied to a different set of samples.Furthermore, Ahmed et al. (2023) performed a targeted study focusing on microbial VOCs (mVOCs) previously selected from bacterial species associated with VAP (S. aureus, P. aeruginosa, K. pneumoniae, and Escherichia coli).In the case of CP for S. aureus, two VOCs were upregulated compared to the other patients value (e.i.0.790-0.870)AUC.In the case of CP for P. aeruginosa, two VOCs were downregulated compared to CP for other pathogens, and one of these VOCs (identified as 3-methylbutanal) was common with CP for S. aureus.Moreover, those VAP patients with CP for bacteria known to metabolize tryptophan (E.coli, Klebsiella oxytoca, and H. influenzae) presented increased levels of indole.Despite the fact that both studies followed the same BreathDx protocol, no shared VOCs were identified.However, two VOCs (dimethyl sulfide and tetrahydrofuran) reported by Van Oort et al. (2022) and another two (3-methylbutanal and acetone) by Ahmed et al. (2023) were also found in other studies (Table 5).
Additionally, several research groups participating in BreathDx had previously conducted studies focusing on CAP/HAP/VAP, one aiming at possible biomarkers for CAP/HAP (Van Oort et al., 2017b).In this study, patients were categorized based on their clinical suspicion, namely, probable CAP/HAP patients (high clinical suspicion), possible CAP/HAP patients (low clinical suspicion), colonized patients (without symptoms of pneumonia), and controls.Additionally, the entire patient cohort was divided into CP and CN.In the UVA, probable CAP/HAP patients and those who were CP presented 11 and 52 downregulated VOCs, respectively, and the classification models could discriminate between groups based on their clinical suspicion, and among CP and CN, even after LOOCV value (e.i.0.730 and 0.690, respectively) AUC.While this study differed from the others, since they focused on CAP/HAP, several VOCs were shared, such as acetone, which was described by Ahmed et al. (2023), and 2-methylcyclopentanone, as reported by Fowler et al. (2015) (Table 5).The aforementioned study (Fowler et al., 2015) was performed by another research group involved in BreathDx, where ventilated ICU patients were sampled over their stay at five different time points to identify the VOCs that could be used to predict the risk of developing VAP.The model could separate CP and CN patients (sensitivity 98.0% and specificity 97.0%), and eight VOCs were selected as potential predictors (four downregulated and four upregulated).Several of these VOCs were common in different studies, such as ethanol, which was reported in a total of four independent studies (Table 5).In this regard, Schnabel et al. (2015) constructed an RF model based on 12 VOCs, such as ethanol, which correctly classified 74.2% of VAP and non-VAP patients (75.8% sensitivity, 73.0% specificity, and 0.870 AUC).Furthermore, when searching these VOCs in human and VAPcausing bacteria pathways, ethanol was found to be involved in six distinct pathways.Although ethanol seems to be a promising biomarker, its involvement in VAP development should be further studied, as this VOC participates in many physiological and pathological processes, such as OS, and its origin can be attributed to alcohol consumption.
Additionally, two studies focused on mVOCs previously detected in vitro from different cultures of pathogens associated with VAP.Filipiak et al. (2015) annotated 13 mVOCs in CP for S. aureus and 11 mVOCs in CP for Candida albicans.Considering the possible coexistence of VAP-causing pathogens, the study further aimed to explore and assess differential mVOCs that could potentially be associated with the progression of VAP caused by each pathogen.In this regard, 4-heptanone was found to be possibly related to C. albicans; propane and butane to S. aureus; acetanilide, 2-pentanone, and dimethyl sulfide to E. coli; 3-methyl-1-butene to H. influenzae; 1-undecene to P. aeruginosa; and n-hexane, isobutane, and 2-methyl-1-butene to S. pneumoniae.Likewise, Gao et al. (2016) studied the presence of mVOCs in VAP patients, focusing on A. baumannii.For this purpose, A. baumannii VAP patients, A. baumannii colonized patients, and controls were compared, yielding 19 significant VOCs by UVA, 4 being also detected in in vitro A. baumannii cultures.Moreover, 8 of these VOCs were considered derived from A. baumannii, being able to differentiate A. baumannii VAP patients from colonized patients, as well as from controls (0.880 and 0.890 AUCs, respectively).Both studies reported three VOCs in common.Additionally, one VOC was reported by Filipiak et al. (2015).3-Methylbutanal was also found in the BreathDx study (Ahmed et al., 2023) (Tables 5).

COVID-19
In the past 3 years, COVID-19 has led to approximately 750 million confirmed cases and nearly 7 million deaths worldwide according to the WHO (WHO Coronavirus Dashboard, 2023).Several diagnostic tests were developed to contain the outbreak, such as reverse-transcription polymerase chain reaction (RT-PCR) for SARS-CoV-2 RNA detection in nasopharyngeal or oral swab samples, and antigen tests for spike (S) protein and nucleocapsid (N) protein detection.However, these tests have variable false-negative rates (Kucirka et al., 2020), with antigen tests being less sensitive and specific than RT-PCR (Scohy et al., 2020).Furthermore, these tests require multiple reagents, and in the case of RT-PCR tests, specialized equipment and trained technicians are required.
Despite the vaccination of over 13 million people worldwide (WHO Coronavirus Dashboard, 2023), COVID-19 remains an ongoing public health challenge.The potential emergence of more transmissible variants, changes in clinical symptoms, immune evasion (even in vaccinated individuals), and the possibility of reinfection are significant concerns.Additionally, distinguishing COVID-19 from other upper respiratory infections is crucial for isolation and transmission prevention.Consequently, breath tests, particularly in resource-limited settings, could offer a rapid means of diagnosing COVID-19.
Five studies that focused on COVID-19 are included in Tables 3  and 4. Two studies were conducted within a cohort of hospitalized patients.The targeted study by Berna et al. (2021) was performed in a cohort of pediatric patients.Six of the 84 targeted VOCs were upregulated in COVID-19 patients, which were further validated in an independent cohort.Moreover, the cumulative abundance of these six VOCs was evaluated as a diagnostic strategy (0.920 AUC, 91.0% sensitivity, and 75.0%specificity     infections.These two studies presented one VOC in common, octanal (Table 5).
Conversely, two other studies were conducted on non-hospitalized COVID-19-positive patients.Woollam et al. (2022a) enrolled a cohort undergoing COVID-19 testing due to symptom onset, contact with symptomatic individuals, or mitigation testing.When COVID-19 patients were compared with HC, 41 VOCs were found to be significantly altered, mostly upregulated.Curiously, COVID-19 patients were divided into two subclasses based on their VOC profiles, one of which presented 4 of the 41 VOCs upregulated compared to the other subclass and HCs.Furthermore, the set of 41 VOCs could distinguish among groups with 96.0% accuracy, increasing to 100% when the 16 most significant VOCs were selected.The predictive classification model based on three VOCs (hexyl acetate, cedrene, and 3,5,5-trimethylhexanal) presented 100% sensitivity and 92.0%specificity value (e.i.0.990) AUC.Lastly, 11 COVID-19 patients were sampled after recovery, and 34 VOCs recovered baseline levels, although five were still upregulated.When including this group in the final model, recovered COVID-19 patients clustered with controls and could be distinguished from COVID-19 patients with 90.0% accuracy.Myers et al. (2023) included patients presenting upper respiratory infections from two different ambulatory TABLE 4 (Continued) Summary of group comparisons, statistical approaches, and identified VOCs in the studies focused on pneumonia and COVID-19.CAP, community-acquired pneumonia; CLZ, airway colonized; CN, culture-negative; CP, culture-positive; CTR, controls; FU COVID-19, follow-up samples of COVID-19 patients; HAP, hospital-acquired pneumonia; MVA, multivariate analysis; non-VAC, non-vaccinated; non-VAP, non-ventilator-associated pneumonia; PR CAP/ HAP, probable community-acquired pneumonia/hospital-acquired pneumonia; PS CAP/HAP, possible community-acquired pneumonia/hospital-acquired pneumonia; UVA, univariate analysis; VAC, vaccinated; VAP, ventilator-associated pneumonia; VOCs, volatile organic compounds.
Additionally, Cen et al. (2023) investigated the metabolic reprogramming triggered by the inactivated COVID-19 vaccine, comparing the VOC profiles of COVID-19 vaccinated and unvaccinated subjects.The discriminative model based on nine VOCs (from 21 identified in both UVA and MVA), which included 6-methyl-5-hepten-2-one already found by Myers et al. (2023) (Table 5), exhibited 94.4% overall accuracy, 91.3% sensitivity, and 98.6% specificity value (e.i.0.995) AUC.Furthermore, the examination of the biomarkers' metabolic pathways demonstrated that the protective metabolic regulation induced by the vaccine influences enzymatic activity and microbial metabolism within the lungs, liver, and gastrointestinal tract.

Searching for pathology-specific VOCs in human exhaled breath
The search for potential biomarkers in exhaled breath is challenging due to the substantial variability in the concentration of VOCs.This variability is due to metabolic activity but also depends on lifestyle choices (smoking, exercise, diet, etc.) and/or exposure to exogenous factors, such as pollutants and other environmental compounds, among others.Despite this challenge, numerous studies have focused on identifying specific VOCs associated with a wide range of pathologies.Nevertheless, the results of these studies should be interpreted with caution.In most instances, the origin of these VOCs remains unidentified, which can lead to false discoveries.The VOCs included in Tables 2, 5; Supplementary Table S3, as classified by Drabińska et al. (2021), are illustrated in Figure 4.As noted, the analysis of exhaled breath covers a wide range of chemical species, although the distribution of these is variable across pathologies.Aldehydes are the most abundant, mainly derived from alcohol metabolism in the liver or the reduction of hydroperoxides during lipid peroxidation (Murray et al., 2009;Hakim et al., 2012), although aldehydes can also come from cigarette smoking or tobacco components' detoxification by cytochrome P450 (Furge and Guengerich, 2006;Papaefstathiou et al., 2020).This chemical group is predominant in CAP/HAP/ VAP, COVID-19, and COPD.
Ketones are also strongly represented in CAP/HAP/VAP and COVID-19, mainly resulting from the liver's synthesis of ketone bodies (acetoacetate, acetone, etc.) during conditions like diabetes, fasting, or alcoholism, formed through the metabolism of proteins and/or as secondary products of lipid peroxidation (Vaz and Coon, 1987;Murray et al., 2009).Remarkably, a significant proportion of ethers is observed in COVID-19 and BC, although its origin is commonly attributed to exogenous sources.Other abundant compounds in BC are alcohols, which may come from the gastrointestinal tract or are formed through the hydrocarbon's metabolism or lipid peroxidation (Ortiz De Montellano, 2010;Ratcliffe et al., 2020).
Additionally, hydrocarbons are widely reported in exhalates, primarily saturated, aromatic, branched and cyclic.These compounds are highly represented in LC, asthma, and CRC.Hydrocarbons are mainly produced by lipid peroxidation, in an abnormal metabolic state.Branched-chain hydrocarbons may be of an endogenous origin from bacterial metabolism (Ratcliffe et al., 2020).
The concept of the exposome is gaining popularity, encompassing not only external exposures (chemical agents, radiation, etc.) and associated physiological responses but also internal sources, such as microbiota, and "psychosocial components" (Vineis et al., 2020).Several studies have focused on identifying metabolites related to the exposome, such as the database developed by Neveu et al. (2023), which includes microbial metabolites and is supported by evidence on their origin, and the method developed by González-Domínguez et al. ( 2020) for exposome research.Furthermore, the effect of the exposome on human health has been widely studied (Morales et al., 2022).Nevertheless, many metabolites associated with the exposome overlap with those produced by human cells/tissues, making it a difficult task to establish what can be considered truly endogenous.This issue is especially challenging for VOCs detected in exhaled breath, since the pulmonary tract is closely associated with environmental exposure.
The full list of reported VOCs was used to identify pathologyspecific compounds (Figure 5).The overlapping VOCs may come from exogenous sources (exposome), such as cigarette smoking, environmental pollution, or diet, as well as shared endogenous origins like the ones derived from OS or common VOCs found in breath, such as isoprene and acetone.
Regarding pathology-specific possible biomarkers, several unique VOCs were found (Figure 5), especially in LC, according to the literature reviewed herein.Furthermore, the obtained list of unique pathology-specific VOCs was submitted to searches on KEGG and BioCyc databases with the aim of excluding the VOCs mainly coming from exogenous sources.Upon exclusion of such exogenous VOCs, the final list of pathology-specific possible biomarkers is compiled in Table 6.It is worth mentioning that although these candidate biomarkers might provide useful information, further research is required to establish associations with metabolic alterations in each pathology, as well as to discern between the VOCs that may be related to the exposome and the ones that are truly endogenous.
Furthermore, the correct metabolite identification is a highly important aspect in metabolomics, and different levels can be distinguished based on the reliability of the identification.In this regard, the Metabolomics Standard Initiative (MSI) levels can range from 1 to 4, level 1 being the most rigorous (Sumner et al., 2007).
In the case of LC, four VOCs from Table 6 were reported as potential biomarkers in at least two different studies (Supplementary Table S3).In this regard, 2,3-butanedione and butanal (both MSI level 1) were found to be upregulated.The remaining candidate biomarkers were reported as altered; thus, the trend of their levels should be assessed.Propionic acid (MSI levels 1 and 2) was reported as upregulated and downregulated in two different studies; therefore, it is not an adequate candidate due to the contradictory findings.The remaining metabolites included in Table 6 were reported only once, requiring further study for their use as pathology-specific biomarkers.Additional candidate biomarkers whose endogenous origin has not been established include 2-nonenal (MSI levels 1 and 2), 3methylhexane (MSI levels 1 and 2), butanal (MSI level 1), pentane (MSI level 1), and propylene (MSI levels 1 and 2) for LC, all of which are reported several times and show a trend toward increased levels; acrylonitrile (MSI level 1) for GaC is reported as upregulated in two independent studies; methacrylic acid (MSI level 2) for BC presents decreased levels; and 1,2dimethylcyclohexane (MSI level 2) for asthma and 2methylcyclopentanone (MSI level 2) for CAP/HAP/VAP are reported as downregulated (Tables 2, 5; Supplementary Table S3).

Methodologies
The methodologies used for breath sampling and VOCs' preconcentration and separation in the reviewed studies are presented in Tables 1, 3; Supplementary Table S1 and illustrated in Figure 6.

Exhaled breath sampling
Breath samples were categorized in mixed or alveolar breath, as late expiratory breath sampling was not specified in any study, being usually confused with alveolar breath.The lack of distinction may be due to the absence of standardized protocols or guidelines for the collection of late expiratory breath.Devices that discard dead space air may not ensure true alveolar breath sampling; therefore, only those with CO 2 or pressure sensors should be used for this type of sample.
Exhaled breath sampling devices can collect from a few milliliters to 10 L, which depends not only on the device's capacity but also on the fraction of breath sampled, since alveolar breath represents approximately 350 mL of the total expiratory volume.Even though breath samples can be taken from a single expiration or multiple expirations, VOC profiles can vary from breath to breath (Khoubnasabjafari et al., 2022), and the concentrations differ significantly in hypoventilation, hyperventilation, and normal ventilation (Cope et al., 2004).
The selection of sample type depends on the compounds of interest.When studying endogenous VOCs, late expiratory or alveolar breath are preferred, the latter being more convenient due to the higher concentration of VOCs and reduced contamination (Miekisch et al., 2008).However, as depicted in Figure 6A, mixed breath was the most analyzed sample type, probably because the devices used for this type of sample are more affordable and easier to use.However, due to the increasing interest in endogenous VOCs, alveolar breath was analyzed in a significant number of studies, especially in those focused on cancer.
The most common collection device was Tedlar ® bags (Figure 6B), consistent with previous reviews (Lawal et al., 2017;Westphal et al., 2023).While these bags are subject to contamination and have limited sample storage time, they are affordable and reusable, with several cleaning protocols available (Westphal et al., 2023).Additionally, devices such as Bio-VOC ® , breath-gas sampler, and ReCIVA are gaining popularity, although they are not as widely used.
Sampling methodologies are organized per type of pathology in this section as the severity of the pathology may justify specific approaches.

Cancer
In LC, 14 studies analyzed alveolar breath, while the remaining 9 analyzed mixed breath.The alveolar breath samples were collected using various techniques, such as Bio-VOC ® , BCA, Tedlar ® /Mylar bags, or other devices (analytical barrier bag and breath reservoir).For mixed breath samples, Tedlar ® /Mylar bags were predominantly used, and some studies employed self-developed devices and glass bulbs.
In GaC, alveolar breath was the main type, sampled either with Mylar/Tedlar ® bags, GaSampler collection bags (QuinTron), or a custom-built in-house breath sampler.One study used gastight syringes for mixed breath sampling.In CRC, two studies sampled mixed breath using Tedlar ® bags, and the remaining alveolar breath employed ReCIVA ® or other devices [GaSampler Collection Bag (QuinTron) and gas-tight syringes].In BC, all studies collected alveolar breath, using various devices such as gas-tight syringes, Tedlar ® bags, and Bio-VOC ® .

Other pulmonary pathologies
In asthma, all the studies collected mixed breath samples using Tedlar ® bags, except for one that analyzed alveolar breath and used an ABS.In COPD, alveolar breath and mixed breath were selected with equal frequency.The samplers employed were either Tedlar ® bags or stainless steel tubes for mixed breath, and Bio-VOC ® or glass syringes for alveolar breath.Furthermore, in one study, alveolar breath was also collected directly into the pre-concentration device.
In OSA, alveolar and mixed breath samples were sampled with Bio-VOC ® and Tedlar ® bags, respectively, and in CF, mixed breath was collected using Tedlar ® bags.

Infectious pathologies
In CAP/HAP/VAP, one of the targeted studies sampled alveolar breath employing glass syringes, while the remaining studies analyzed mixed breath samples.Since the patients were intubated and mechanically ventilated, most of the sampling was performed directly in sorbent tubes, except for three studies that employed either Tedlar ® bags or a breath-gas sampler.In COVID-19, most of the studies analyzed mixed breath, collecting the sample either using Tedlar ® bags or SamplePro FlexFilm Sample Bags.Only one study sampled alveolar breath employing ReCIVA.

Pre-concentration techniques
TD were the most utilized pre-concentration technique, followed by SPME (Figure 6C), consistent with previous reviews (Lawal et al., 2017;Westphal et al., 2023).The widespread use of TD can be attributed to their suitability for long-term sample storage, ease of transport, and the stability of the entrapped compounds.However, SPME requires smaller sample volumes and is less affected by humidity, offering a similar extraction range to TD. Notably, NTD was used in only a few studies, despite presenting aspects of both SPME and TD.The choice of the sorbent material depends on the chemical nature of the analytes of interest, including polarity and molecular weight (MW).The most used SPME fiber coating material was Car/ PDMS (mainly 75 µm thickness), followed by DVB/Car/PDMS, and lastly, PDMS 100 µm and PDMS/DVB (Tables 1, 3; Supplementary Table S1).These fibers are bipolar, except for PDMS (non-polar), with the latter compromising the extraction of polar metabolites (Vas and Vékey, 2004).The wider use of Car/PDMS fiber could be due to the ability of this coating material to better extract lowmolecular-weight volatiles (MW 30-225 g/mol) compared to PDMS 100 µm (MW 60-275 g/mol) and DVB/PDMS (MW 80-300 g/mol) (Lawal et al., 2017).However, in a recent study (Schulz et al., 2023), DVB/Car/PDMS turned out to be the most adequate for untargeted studies compared to Car/PDMS and PDMS fibers due to the higher number of extracted VOCs and the stronger overall GC-MS signal.
Regarding TD, Tenax TA is the most used sorbent material (Tables 1, 3; Supplementary Table S1).Although this material captures heavy-and less-volatile compounds, its low affinity to water and the broad sampling range (C6-C30) makes it adequate for untargeted analysis (Wilkinson et al., 2020).Similarly, a few studies have opted for Tenax GR.Other sorbent materials encompass carbon black adsorbents, such as Carbograph 1TD, Carbograph 5TD, Carbopack X, Carbopack B, and Carbotrap.Additionally, carbon-based materials like Carboxen were employed.These alternatives offer a narrower range (C3-C20 and C2-C5), although they facilitate the capture of low-molecular-weight and more volatile compounds (Lawal et al., 2017;Westphal et al., 2023).Many studies have employed multi-bed sorbents, with the most prevalent being Carbograph 1TD/Carbopack X.Furthermore, combinations of the aforementioned materials have been used, such as Tenax TA/Carbograph 1TD and Tenax TA/Carbograph 5TD, both of which have been considered for exhaled breath analysis in previous studies (Wilkinson et al., 2020).Additional combinations, such as Tenax/Carbograph/Carboxen and Carbotrap B/Carbopack X, were also employed.

GC-MS methods
Considering that the main objective of most studies discussed in the present review was to obtain a snapshot of the VOC content in the breath samples, and also the analysis conditions and the preconcentration techniques.Conditions of the GC-MS method such as the injector mode, chromatographic column, and type of detector employed for analysis are as important as the preceding preconcentration technique.
Regarding GC injector parameters, from the 70 works reviewed, an astonishing 61% do not detail the type of injector or injection employed.Such a number is alarming given the fundamental difference between injecting gaseous and liquid samples.In fact, for gaseous injections choosing an injector glass liner of smaller inner diameter would provide a more efficient transfer of analytes onto the GC column, thus yielding more peak capacity efficiency.Nonetheless, of the 39% of works that mentioned the employment of a splitless/split injector, none mentioned the dimensions of the injector glass liner diameter employed.Out of these split/splitless injections, 70% of the injected samples are in the splitless mode, which would indeed be expected for pre-concentration techniques such as direct TD and SPME.
When it comes to GC columns as presented in Figure 6D (compiling the information from Tables 1, 3; Supplementary Table S1), over 50% of the studies herein reviewed employed 5% diphenyl/95% polydimethylsiloxane phases, here named as X-5 columns (DB-5, RTX-5, SLB-5, HP-5, VF-5, etc.).This type of stationary phase is considered the most versatile owing to the slight polarity imparted by the substitution of 5% of dimethyl groups by diphenyl.This addition also makes this stationary phase suitable for the separation of unsaturated hydrocarbons and aromatic compounds.Conversely, this stationary phase should not be the first choice regarding the analysis of VOCs.In fact, despite its slight polarity, it does not provide sufficient retention and efficiency for the separation of low-molecularweight polar VOCs, such as alcohols, aldehydes, and organic acids.
The second most used types of GC stationary phase are X-624 and X-1, each being employed in 14% of the studies presented in Tables 1,  3; Supplementary Table S1.The X-624 stationary phase (DB-624, VP-624, SLB-624, etc.) consists of polydimethylsiloxane with 6% of the dimethyl substituted by cyanopropyl and phenyl groups.Therefore, as expected, this is a low-polarity phase, though of higher polarity than its X-5 counterparts.A key characteristic of these columns is the thickness of the stationary phase.While most X-5 columns employ stationary phases of 0.25 µm thickness, X-624 columns are coated with, at least, 1.4 µm of the stationary phase.Therefore, in addition to its chemistry allowing for better selectivity, there is also a considerable gain in retention for low-molecular-weight polar VOCs and a wider range of VOC classes could be successfully analyzed in breath samples.In fact, X-624 stationary phases are the most suitable for VOC analyses, as it is the official stationary phase for a variety of Environmental Protection Agency (EPA) methods dealing with VOCs.X-1 columns (DB-1, RTX-1, SLB-1, HP-1, VF-1, etc.) contain 100% polydimethylsiloxane, the most non-polar stationary phase.Similar to X-5 types of phases, this stationary phase does not provide sufficient selectivity for the separation of small and polar VOCs even when employing thicker phases.
Given the importance of aldehydes and alcohols in the studies herein reviewed, as presented in Figure 4, it may be surprising that only two studies employ polyethylene glycol (WAX) stationary phases, as they are highly selective for polar compounds such as alcohols.A plausible explanation in this type of application might be related to the presence of water in the breath sample.WAX columns are particularly sensitive to moisture in the sample, which may lead to the degradation of this stationary phase.
While all other GC columns mentioned here encompass wallcoated open tubular (WCOT) columns, porous layer open tube (PLOT) columns are highly retentive and, therefore, are primarily employed for the analysis of very low boiling point compounds that are gaseous at room temperature, such as sulfides.Applied in 12% of the studies herein reviewed, Q-PLOT columns are non-polar, as they employ 100% divinylbenzene as adsorbent, therefore imparting great selectivity and retention for low-molecular-weight hydrocarbons.
Most of the studies included in Tables 1, 3; Supplementary Table S1 were performed by one-dimensional (1D) GC-MS.Additionally, five studies applied comprehensive bidimensional gas chromatography (GC×GC).The most common GC×GC setup employs an orthogonal mechanism of separation, using two sequential GC columns with stationary phases of different polarities, with a modulator between them.In short, a narrow band eluting from the first dimension (1D) column is collected and focused on the modulator, and then sent to the second dimension (2D) column (which is much shorter than the 1D).In this way, for example, compounds of similar boiling points coeluting on the 1D could be separated according to their polarity differences on the 2D.This technique offers significant advantages: the extended peak capacity improves peak space separation and allows the detection of coeluting compounds that could be missed by conventional 1D-GC.Moreover, given the acquisition speed required by the narrow bands eluting from the 2D in GC×GC, this technique is often hyphenated with MS detectors with rapid MS analyzers such as TOF, providing also higher sensitivity than that obtained by 1D-GC employing single quadruple MS.As an example of the advanced capacities of GC×GC-TOF-MS, Caldeira et al. (2012) could detect eight-fold more compounds, especially alkanes, alkenes, aldehydes, and ketones, and the concentration range achieved was lower than that of a previous study performed with GC-qMS.Four out of the five studies presented the traditional apolar × polar configuration, employing stationary phases like X-624 and X-5 in the first dimension (1D) and a polar polyethylene glycol-based (WAX) phase in the second dimension (2D) (Caldeira et al., 2012;Pesesse et al., 2019;Schleich et al., 2019;Cen et al., 2023).This combination reduces the interaction of water with the polar stationary phase and provides information on both the volatility (1D) and polarity (2D) of the compounds in the sample (Wilde et al., 2019).Interestingly, Berna et al. (2021) employed a polar × polar setup, using a WAX-based column in the 1D and a trifluoropropylmethyl polysiloxane (RTX-200) column in the 2D.The 2D stationary phase offers a unique selectivity for electron-rich molecules and resolves compounds that could not be resolved by the Wax 1D column.The limited use of this technology can be attributed to the high costs of instrumentation, especially for cryo-based modulators that are the most adequate for applications such as breath analysis due to their ability to successfully trap very volatile compounds.Moreover, method optimization in GC×GC is not as straightforward as in 1D-GC, hence requiring specialized personnel from method development to data process and interpret (Pesesse et al., 2019).
Moreover, 20 studies have employed high-resolution mass spectrometers (TOF-MS), of which more than half were included in studies on other pulmonary diseases (Tables 1, 3; Supplementary Table S1).The high-resolution approach offers notable advantages, especially when performing an untargeted study.In this regard, sensitivity and selectivity are improved compared to the low-resolution approach.The spectral libraries used for compound identification include accurate mass, which further allows for the enhanced structural elucidation of unknown compounds.Nonetheless, the use of this equipment is more complex; data processing requires more time and space and the price is higher (Rey-Stolle et al., 2021).

Quality control
In breath analysis, as in any metabolomics study, evaluating the quality of the obtained data is crucial.This is essential not only for obtaining reproducible results but also to ensure that the differences observed between groups are attributable to the composition of the samples rather than analytical/instrumentation variations.Such assurance involves the analysis of blanks and the use of internal standards among other strategies detailed below (Dudzik et al., 2018).
The analysis of blanks allows for the identification of contaminants (e.g., Tedlar ® bag contaminants) and artifacts (e.g., polydimethylsiloxanes), and its elimination from the data matrix, avoiding false discoveries.The main blanks include collecting device blank, air blank, and vial/tube blank (Westphal et al., 2023).In this regard, several studies have reviewed and conducted analyses of ambient air, yet a considerable number of studies do not include this step, such as that of Gao et al. (2016), Aoki et al. (2017), Van Vliet et al. (2017), andSaidi et al. (2020).Moreover, the inclusion of the remaining blanks is not specified in most of the studies.The injection of a standard mixture over the sequence is recommended, although just a few studies include it in their workflow (Schleich et al., 2019;Koureas et al., 2021).
Generally, to ensure quality in the analysis of exhaled breath samples by GC-MS, and therefore reliable findings, several key quality assurance and quality control measures are essential.These measures aim to guarantee the accuracy, reproducibility, and reliability of the entire analytical results and are as follows (Li et al., 2019;Becker, 2020;Westphal et al., 2023): • Calibrants and reference materials: Calibrants and reference materials containing volatile marker compounds are used to establish detector stability, known detection limits, and instrument calibration.These materials should be available consistently during clinical trials and routine applications.• Instrument calibration: Regular instrument calibration using internal standards, such as stable isotope-labeled compounds, enhances QC/QA efforts.It helps in tracking instrument performance and ensuring the accuracy of results.• Blank analysis: To identify and remove contaminants, blank analyses are essential.These blanks include air blanks, system blanks to identify instrument artifacts, and blanks to account for chemical backgrounds originating from sampling materials like Tedlar ® bags.
• Spiking samples with internal standards of known concentrations and different retention times aids in data normalization and enhances data quality.• Inter-laboratory comparisons: For diagnostic purposes, it is important to compare data obtained from different laboratories and methods.It helps assess the reproducibility and relevance of potential biomarkers.
In summary, ensuring quality in exhalation analysis by GC-MS involves a comprehensive approach that encompasses quality control, standardization, instrument calibration, and data management.All these measures must be implemented without any exception in clinical applications where breath analysis holds potential for disease diagnosis and monitoring.

Conclusion
This comprehensive review aims to investigate the potential of GC-MS analysis of VOCs in breath as biomarkers for severe pathologies, such as cancer, pulmonary diseases, and infectious diseases.Critical aspects of the workflow are thoroughly considered and discussed, encompassing the type of exhaled breath, collection devices, pre-concentration techniques, and analysis, as well as the experimental designs, statistical analysis, identification strategies, and proposed potential VOCs biomarkers.
Tedlar ® bags and TD are by far the most extended for collection and pre-concentration, respectively.However, the choice of the type of breath sample was more diverse, spanning between mixed and alveolar breath, a critical consideration when aiming to accurately compare and establish levels of endogenous VOCs.Despite the wealth of studies, the conspicuous lack of standardization in the methodological approach and the scarce absolute quantitation of potential biomarkers delay their transference to clinics.Additionally, relatively small cohorts with only a limited model validation in an independent cohort, along with the lack of consensus in altered findings among different studies hindered the identification of a single pathology-specific VOC.A deeper understanding of the endogenous origin of VOCs is imperative to fully grasp the significance of each VOC in discriminating between healthy and pathological states.
Overall, this review underscores the substantial potential of VOCs as biomarkers in health and pathology.Nonetheless, to fully harness this potential, it is crucial to address the lack of standardization in methodological approaches, include larger and well-defined cohorts, and validate models in independent cohorts.As we delve deeper into the complexities of VOCs in exhaled breath, we are poised to advance personalized and non-invasive diagnostic strategies that can revolutionize the detection and management of the pathology, ultimately benefiting public health.

FIGURE 4
FIGURE 4Bar chart of the chemical classes of VOCs (referred by two or more studies) reported in each pathology.CAP, community-acquired pneumonia; COPD, chronic obstructive pulmonary disease; HAP, hospital-acquired pneumonia; VAP, ventilator-associated pneumonia.
(Continued on following page)Frontiers in Molecular Biosciences frontiersin.org
).Likewise, Ibrahim et al. (2021) identified six VOCs (seven features) that could discriminate COVID-19-positive test patients and COVID-19-negative test patients (0.836 AUC, 68.0% sensitivity, and 85.0% specificity), although the model based on 11 VOCs showed 0.659 AUC, discriminating patients based on clinical suspicion.In both comparisons, only two VOCs, 1-propanol and benzaldehyde, were common, suggesting that the specific metabolic alterations caused by COVID-19 are not necessarily related to symptomatology, especially if the symptoms are shared with other upper respiratory

TABLE 5
VOCs reported in pneumonia and COVID-19 (two or more studies).na, not applicable; ppbv, parts per billion by volume.*Downregulated in culture positive for P. aeruginosa ventilator-associated pneumonia patients and upregulated in culture positive for S. aureus ventilator-associated pneumonia patients; **downregulated in COVID-19 patients compared to follow-up COVID-19 patients and upregulated in COVID-19-positive patients compared to COVID-19-negative patients and healthy controls.

TABLE 5 (
Continued) VOCs reported in pneumonia and COVID-19 (two or more studies).na, not applicable; ppbv, parts per billion by volume.*Downregulated in culture positive for P. aeruginosa ventilator-associated pneumonia patients and upregulated in culture positive for S. aureus ventilator-associated pneumonia patients; **downregulated in COVID-19 patients compared to follow-up COVID-19 patients and upregulated in COVID-19-positive patients compared to COVID-19-negative patients and healthy controls.
• Training and test samples: For the reproducible identification of specific odor patterns, the availability of appropriate sets of training and test samples is essential.These samples aid in electronic nose-based or canine-based identification methods.• Standardization and harmonization: Standardization of breath sampling procedures is crucial to minimize interobserver and intra-observer errors.Researchers involved in breath sampling should undergo certification to ensure uniform and accurate collection processes.Standardization also involves monitoring room air for potential VOC contamination.