Exhaled breath analysis for the early detection of lung cancer: recent developments and future prospects.

In lung cancer, the prognosis and treatment options depend directly on tumor size and its spread at the time of diagnosis. There is therefore a constant search for methods that will allow early detection of cancerous lung nodules. With advancing imaging technology and implantation of screening routines in high-risk populations by low-dose computerized tomography, a significant increase in the number of diagnosed small peripheral lesions can be expected. While early detection of small cancerous lesions carries the benefit of wider treatment options and better prognosis, the process of obtaining a biopsy to confirm a cancerous tissue is not free of complications and bears inconveniences and stress to the patient. This review discusses the potential use of exhaled breath analysis as a simple, noninvasive tool for early detection of lung cancer and characterization of suspicious lung nodules.

Introduction

Lung cancer (LC) is the second most prevalent cancer in adult men and women around the world, but it is also the deadliest in both sexes.1 The prognosis and treatment options of LC patients depend directly on tumor size and its spread at the time of diagnosis; survival time decreases significantly as the disease is more progressive at detection.2 Screening and early detection through low-dose computerized tomography (LDCT) reduces LC-related mortality by 20%.3,4 Therefore, screening programs are supported now in the US and also recommended by the European societies.5 However, there are still obstacles in the implementation of the screening programs, mainly due to lack of infrastructure to handle high volume of patients with pulmonary findings. Therefore, the unmet current clinical need is to determine the relevant noninvasive biomarkers that may support this effort.6 Here, we review the use of exhaled breath analysis as a potential candidate for such important purpose.

There is a longstanding interest in the potential of using exhaled breath analysis for detection of various diseases, apart from LC detection, which is reviewed in detail in this article. In a most recent publication, Nakhleh et al7 used an artificially intelligent nanoarray sensor in the diagnosis and classification of 17 different diseases from breath samples of 1404 subjects, with 86% accuracy. The same system was able to detect LC in patients with lung nodules,8 detect response to therapy and early recurrence9 and further specify sub-histology of LC10 and even genetic mutation.11 Interestingly, Peng et al12 showed that this system was able to discriminate four types of primary cancers (lung, breast, colorectal and prostate). This review focuses on the ongoing research of exhaled breath analysis for the early detection of LC.

Nature of volatile organic compounds (VOCs)

In addition to oxygen, nitrogen and carbon dioxide, the exhaled breath contains low concentrations of various VOCs. Most VOCs are believed to reflect endogenous metabolic processes at the tissue level, such as inflammation and oxidative stress. However, breath analysis also detects exogenous VOCs, which reflect exposure to carcinogens such as cigarette smoke, pollution and radiation. These VOCs are typically highly reactive substances that damage DNA and proteins, a process that over time might promote cancerous changes in various tissues. When cancer develops, various inflammatory processes as well as gene and protein changes take place; these metabolic changes can create a unique VOC profile that is potentially reflected in the body fluids. As various VOCs reach the lungs through blood stream, they are then excreted by diffusion across the pulmonary alveolar membrane and exhaled through the breath.10,13,14

Although almost 3000 VOCs have been reported in scientific literature, their majority can be divided into few main groups: 1) hydrocarbons – produced mainly by peroxidation of polyunsaturated fatty acids; 2) alcohols – absorbed through the gastrointestinal tract and metabolized mainly in the liver; 3) aldehydes – arise from metabolized alcohols/lipid peroxidation/tobacco smoke/dietary sources; 4) ketones – produced from fatty acids by the liver and influenced by increased fat or protein intake in diet and 5) aromatic and nitrile VOCs – typically considered as arriving from exogenous pollutants such as tobacco smoke and pollution.15

A wide range of modified nanomaterials has been used in exhaled breath analysis, such as nanoparticles (NPs), silicon nanowires (SiNWs) and carbon nanotubes (CNTs). Two main approaches are used in the detection and characterization of VOCs. The first is selective detection of specific VOCs, with gas chromatography being the most commonly used technique. This approach is based on using highly selective receptors designed for pre-identified and selected VOCs. The main limitation of this approach, as will be discussed later, is that currently there are no VOCs that are identified as unique for a specific disease. The second approach is cross-reactive nanomaterial-based sensor arrays, which allows pattern recognition rather than specific VOC identification. There are few techniques that use this approach; the main one is based on electrical conductivity – the electronic nose (eNose). Another technique is the use of colorimetric sensors, which chemically react and change colors upon exposure to different VOCs.15,16 The following sections give a further insight to studies conducted in each of the described approaches. Table 1 presents a selection of studies reporting breath test performance for LC detection.

Table 1

A selection of studies reporting breath test performance for LC detection

Reference	Participants	Technique	Results
			Sensitivity (%)	Specificity (%)	Accuracy (%)
GC-MS
Gordon et al, 198518	12 PLC patients17 HCs	TD-GC-MS	NR	NR	93 (3 VOCs)100 (22 VOCs)
Phillips et al, 199919	60 PLC patients48 HCs	TD-GC-MS	71.7	66.7	69.4
Phillips et al, 200320	178 bronchoscopy patients(67 PLC patients)41 HCs	TD-GC-MS	85.1	80.5	83.3
Poli et al, 200521	36 NSCLC patients110 controls (25 mild-to-moderate COPD patients, 35 smokers with no COPD and 50 nonsmokers)	SPME/TD-GC-MS	72.2	93.6	88.4
Phillips et al, 200723	193 PLC patients211 HCs	TD-GC-MS	84.6	80.0	NR
Phillips et al, 200824	193 PLC patients211 HCs	TD-GC-MS	84.5	81.0
Bajtarevic et al, 200925,#	65 PLC patients31 HCs	SPME/GC-MS	527180	100 (4 VOCs)100 (15 VOCs)100 (21 VOCs)	NR
Ligor et al, 200926	65 PLC patients31 HCs	SPME/GC-MS	51	100	NR
Poli et al, 201022	40 PLC patients38 HCs	SPME/GC-MS	90	92.1	91.0
Wang et al, 201230	88 PLC patients155 controls (70 benign lung disease patients and 85 HCs)	SPME/GC-MS	96.5	97.5	97.1
Phillips et al, 201538	96 PLC patients205 controls	GC-MS	74.0	70.7	NR
Schallschmidt et al, 201632	37 PLC patients23 HCs	SPME/GC-MS	80	90	NR
eNose
Di Natale et al, 200339	35 PLC patients18 controls 9 after surgical therapy	LibraNose	NR	NR	100
Machado et al, 200540	First phase: 14 bronchogenic carcinoma patients and 45 HCsSecond phase: 14 PLC patients and 62 HCs	Cyranose 320	71.4	91.9	NR
Mazzone et al, 200746	49 PLC patients63 other lung diseases patients21 controls	Colorimetric sensors	73.3	72.4	NR
Dragonieri et al, 200942	10 PLC patients10 HCs10 PLC patients10 COPD patients	Cyranose 320			90
D’Amico et al, 201050	28 PLC patients36 HCs28 PLC patients28 other lung disease patients	QMS sensors	85	100	NR
Santonico et al, 201255	20 PLC patients10 controls	QMS sensors	85	85	85
Hubers et al, 201456	20 PLC patients31 COPD controlsValidation: 18 PLC patients and 8 HCs	Cyranose 320	80	48
McWilliams et al, 201543	25 PLC patients166 controls	Cyranose 320	NR	NR	80
Gasparri et al, 201644	70 PLC patients76 HCs	QMS sensors	81	91	NR
Nisreen et al, 201657	149 PLC patients56 controls (COPD/asthma)	SiNW FET	87	82	84
Both eNose and GC-MS
Peled et al, 20128	53 PLC patients19 benign nodules	NaNose, SPME/GC-MS		96	88
Broza et al, 201351	12 PLC patients5 benign nodules	NaNose, SPME/GC-MS	100	80	94.1
Capuano et al, 201554	20 PLC patients10 other lung diseases patients	NaNose, SPME/GC-MS	NR	NR	90 (NaNose)76(SPME/GC-MS)

Note:

Data is given for 3 separated analyses. Studies were included if sensitivity, specificity or accuracy was specified in their text.

Abbreviations: COPD, chronic obstructive pulmonary disease; eNose, electronic nose; FET, field effect transistors; GC-MS, gas chromatography–mass spectrometry; HC, healthy control; LC, lung cancer; NaNose, Nanoscale Artificial Nose; NSCLC, non-small cell lung cancer; NR, not reported; PLC, primary lung cancer; QMS, quadrupole mass spectrometry; SiNW, silicon nanowire; SPME, solid phase microextraction; TD, thermal desorption; VOC, volatile organic compound.

Gas chromatography–mass spectrometry (GC-MS)

In the GC-MS method, exhaled breath is collected and temporarily stored in designated containers (e.g., in inert bags or sorption tubes). Then, a helium stream is used to carry the sample though a long, heated capillary column. The VOCs are separated based on their chemical properties, consecutively ionized and separated by their mass/charge (m/z) ratio and are identified by a spectral library in the GC software.17

Back in 1985, Gordon et al examined breath samples of 12 LC patients and 17 healthy volunteers. A total of 22 VOCs that showed the largest difference between the two groups were recognized. In further analysis, three VOCs were chosen according to their peak and occurrence in the subjects, including acetone, methyl ethyl ketone and n-propanol. Using these three VOCs, the authors accurately classified 93% of the samples.18

Phillips et al collected breath samples from 108 patients with an abnormal chest radiograph who were scheduled for bronchoscopy. LC was confirmed histologically in 60 patients. Analyzing the breath samples, a combination of 22 VOCs discriminated between patients with and without LC, regardless of stage. For stage 1 LC, the 22 VOCs had 100% sensitivity and 81.3% specificity.19 In a different study, the same authors reported identifying primary LC with a sensitivity of 89.6% and a specificity of 82.9% using nine VOCs that were chosen after comparing primary LC patients with healthy controls. The authors noted that patients with primary LC had breath test findings that were consistent with the accelerated catabolism of alkanes and monomethylated alkanes.20

Poli et al chose 13 VOCs to compare three groups, patients with non-small cell lung cancer (NSCLC), subjects with mild–moderate chronic obstructive pulmonary disease (COPD) and asymptomatic smokers as controls. None of the selected VOCs alone distinguished the NSCLC patients from the other study groups, but overall VOC concentrations were highly discriminant (>70%).21 In a later study, the same authors measured exhaled aldehydes as a biomarker of NSCLC; the levels of all aldehydes in the study were found to be increased in NSCLC patients compared to those in healthy controls, with a discrimination power of >90%.22

Phillips et al performed two further studies on breath samples from 193 LC patients and 211 controls. In the earlier study, LC was predicted with 84.6% sensitivity and 80.0% specificity by using a model of 16 VOCs. The authors calculated that if applied to a population with 2% prevalence of LC, a screening breath test would have a negative predictive value of 0.985 and a positive predictive value of 0.163.23 In a later study, the authors used the same database to develop a nonlinear method of multivariate analysis, which accurately identified LC patients using 30 VOCs with 84.5% sensitivity and 81.0% specificity.24

Bajtarevic et al used four VOCs that were found in exhaled breath samples of LC patients but not in those of healthy controls to detect LC patients with a sensitivity of 52% and a specificity of 100%. Using 15 and 21 VOCs, they reached a sensitivity of 71% and 80%, respectively, with 100% specificity in both cases.25 Ligor et al26 used eight VOCs to detect LC patients with 51% sensitivity and 100% specificity. Fuchs et al examined aldehyde concentrations as biomarkers of LC. Four compounds were found to be significantly higher in LC patients than in smokers and healthy controls.27 Song et al28 recognized two VOCs as significantly higher in the breath of LC patients compared with healthy controls. Ulanowska et al29 reported increased concentrations of 11 VOCs in the breath of 137 LC patients compared with 143 healthy nonsmokers; three VOCs (pentanal, hexanal and nonane) were identified only in the breath of LC patients. Wang et al30 used 23 VOCs to discriminate LC patients from controls with 96.5% sensitivity and 97.5% specificity.

The GC-MS method was used in even a wider context in a study by Peng et al. In this study, breath samples were collected not only from patients with LC but also from patients with colon, breast and prostate cancers, as well as from healthy controls. According to the results, the system could successfully differentiate between “cancerous” and “healthy” breath and, furthermore, between the breath of patients having different cancer types.12

As opposed to all the above-cited studies, in a study comparing LC patients with smokers and healthy controls, Kischkel et al noted that differences in exhalation profiles in cancer and noncancer patients did not persist if physiology and confounding variables were taken into account. The authors concluded that exhaled substance concentrations may depend on a variety of parameters other than the disease under investigation, with the main confounding variables identified as smoking history, inspired substance concentrations, age and sex.31 In a recent article, Schallschmidt et al reached a similar conclusion when examining 24 VOCs that were suggested as potential cancer markers in previous studies; the authors concluded that the currently available methodology is not able to reliably discriminate cancer patients and healthy controls by using VOCs. Observational studies often tend to note significant differences in the levels of certain oxygenated VOCs, but without the resolution required for practical application, due to low and variable VOC concentrations.32

Indeed, the GC-MS method has few major weaknesses. A major one is that it requires preparations and expert personnel to handle the samples and is also expensive.33 Another issue is that the method of breath collection affects the concentration of VOCs in the sample. It had been claimed that alveolar breath has the highest levels of exhaled components and the lowest concentrations of contaminants;34 however, the ability of obtaining alveolar breath samples does not depend only on the collection device itself but also on the collaboration and physical capability of the participant. Finally, another raised question is the stability of compounds and the external effects of different storage bags. Most studies either maintain very short sample storage time to minimize its effect on the breath sample or use tubes that are suitable for long-term storing. Yet, the effects of storage method and equipment are still an issue in allowing a reliable interpretation of the breath analysis.35–37

Recognizing the weakness of the method but not abandoning the utility of breath analysis in the detection of LC, Phillips et al39 showed that breath biomarkers increased the sensitivity, specificity and positive and negative predictive values of chest computed tomography (CT) for LC when the tests were combined in series or parallel.

Nanoarray analysis

eNose

In this method, an exposure of the sensors to a mixture of VOCs changes their electrical resistance. The data are transformed by statistical or structural algorithms to identify various volatile patterns, leading to a production of a “breath print”. This method is easy to use and provides highly sensitive results but does not allow quantitative results or identification of unknown substances.33 As mentioned earlier, different eNose technologies were used in the research for LC detection40–41; all provided similar success in the correct recognition of the disease with respect to control groups.

Di Natale et al reported 100% of classification of LC patients versus healthy controls by using eight quartz micro-balance (QMB) sensors, and 94% of the reference was correctly classified.39 Machado et al used a group composed of patients with bronchogenic carcinoma and healthy controls to identify discriminant breath patterns between cancer and noncancer participants; the data were used to create and apply a cancer prediction model prospectively in a separate group of 76 individuals (14 individuals with LC). In the validation study, the eNose had 71.4% sensitivity and 91.9% specificity for detecting LC; positive and negative predictive values were 66.6% and 93.4%, respectively.40

Dragonieri et al42 reported discriminating 85% of NSCLC patients from COPD patients and 90% of NSCLC patients from healthy controls. McWilliams et al43 reported distinguishing 25 LC patients from 166 high-risk smokers without cancer with a better than 80% classification accuracy. Gasparri et al examined the breath of 70 LC patients and 76 healthy controls and showed 81% sensitivity and 91% specificity in differentiating the two groups by breath prints. The largest sensitivity was observed for patients with early-stage LC with respect to late-stage disease.44

As specified elsewhere in the text, the group of Haick and Peled has reported series of studies using the NaNose system, with high sensitivity and specificity to detect LC as well as other types of cancer. Interestingly, the NaNose was able to discriminate between subtypes of LC (small cell lung cancer [SCLC], adenocarcinoma and squamous cell carcinoma)45 and mutation status11 and to track response to therapy and early recurrence.9

Colorimetric sensors

This is a slightly different method of nanoarray analysis that uses sensors composed of chemically sensitive compounds that change their colors depending on the chemicals with which they come into contact. Mazzone et al used a prediction model that was developed from breath samples of individuals with LC and other lung diseases (e.g., COPD, sarcoidosis) and healthy controls. Applying the model on another group of participants with a similar disease profile, it was able to predict the presence of LC with 73.3% sensitivity and 72.4% specificity.46 In a later study, the same authors used the colorimetric sensors to distinguish LC patients from healthy controls with moderate success (C-statistic 0.811). Furthermore, individuals with different histologies could be accurately distinguished from one another (C-statistic 0.864 for adenocarcinoma versus squamous cell carcinoma).47

Potential use of breath analysis as part of screening programs for LC

Screening for LC by LDCT in high-risk populations is becoming more of general practice in the US and Europe. Populations at high risk are considered as individuals aged 55–74 years, who have at least 30 pack-years and who have smoked up to the last 15 years.5,48 With advancing imaging technology and implantation of screening routines, a significant increase in the number of diagnosed lung lesions can be expected; consequently, more patients will need to go through evaluation of the findings in order to define whether they are malignant and post a threat to patient’s life.49 There is therefore a need for advanced tools that will assist to determine the nature of the detected lesion in as minimally invasive manner as possible. D’amico et al51 examined breath samples of 56 patients who were about to go through bronchoscopy due to anomalous chest X-ray. Following the procedure, 28 patients were diagnosed with LC. The use of gas sensor arrays allowed a successful discrimination of 79.3% between cancer and noncancer breath samples. Peled et al analyzed breath samples of 72 patients who were going through investigations for pulmonary nodules by both GC-MS and chemical nanoarrays. A total of 53 of the nodules turned out to be malignant; 19 were benign. The GC-MS analysis identified significantly higher concentration of 1-octene in LC patients’ breath. The chemical nanoarrays significantly distinguished benign versus malignant nodules, with 86% ± 4% sensitivity, 96% ± 4% specificity, and 88% ± 3% accuracy.8

Potential use of breath analysis to monitor response to anti-cancerous treatment

In a study by Broza et al,51 breath samples were collected from patients prior to and after lung resection. GC-MS analysis showed that five VOCs were significantly reduced after LC surgery; a nanomaterial-based sensor array distinguished between pre- and postsurgery states. The traced VOCs were attributed to oxidative stress (three VOCs) and to exogenous carcinogens (two VOCs). The study included only a small number of patients and was rather a proof-of-concept study, yet the results indicate that breath pattern changes can be detected not only when LC is present but also after its resection. But what happens in cases when the patient goes though systemic, rather than surgical, treatments? Currently, monitoring response to anti-cancerous treatment in LC is based upon consecutive CT scans. However, time intervals between scans might be too long to allow early identification of treatment failure. Nardi-Agmon et al9 assessed the use of breath analysis by both GC-MS and nanoarrays in monitoring response to treatment in 39 patients with advanced LC. When one sensor analysis was used, there was 85% success in monitoring disease control. This study suggests that the eNose has also a potential as a quick, simple method to identify lack of response to an anticancer treatment and to prompt treatment reassessment.

Discussion

Diagnosis and evaluation of cancerous disease involved a variety of imaging techniques, depending on the cancer type and available technology. Constant new developments in the imaging field allow earlier detection and better visualization of body tissues. However, the final determination of cancer is made at the tissue level by pathological examination of tissue biopsy taken from a suspected lesion. The process of obtaining a tissue biopsy is not free of complications and bears inconveniences and stress to the patient. There is therefore a need for advanced, minimally invasive diagnosis tools to assist early and reliable recognition of lesions that are suspected to be cancerous.

Currently, no unique tumor marker for LC has been identified. However, further research and development in the field of breath analysis could potentially provide a simple tool to assist early recognition and reliable characterization of suspected lung nodules. To date, the absence of a standardized method for breath analysis among the various researches is a main weakness in assessing the relevance and reliability of the different results obtained. For example, in a review of 10 studies regarding VOCs in LC, 170 different VOCs were detected in total; however, only 17 of them appeared in at least two different studies.52 In addition, as discussed earlier, independent of the various analysis techniques, the un-unified collection and storage of the breath samples constitute some other major challenges in creating a standardized study method and producing reliable results.35–37

A question has risen regarding the difference in breath chemical components between alveolar air and mouth air. While some papers claim that alveolar breath has higher levels of exhaled components,34,53 in 2015, Capuano et al had conducted a comparative analysis of exhaled breath and air sampled from inside the lungs. The study included 30 patients, 20 with LC and 10 with other lung diseases. The GC-MS and an eNose were used in the analysis of air collected from inside both lungs with a bronchoscopy probe, and the samples were compared with regular breath samples collected from the same patients. The authors found that the diagnostic capability of the eNose did not depend on whether the sampled air was taken from the lung on the side of the tumor itself or from the other lung (in both cases, 90% of correct classification between cancer and noncancer samples was obtained). GC-MS analysis of the air samples from the lung and of breath samples demonstrated a substantial preservation of the pattern of VOCs from inside the lung to the exhaled breath.54

The exhaled medium is a feasible approach to support early diagnosis of LC; however, it requires further standardization and clinical implementation. The various results of the GC-MS studies teach us that while it is not possible to identify single specific breath markers for LC, we can reliably say that LC alters the concentrations of a manifold of compounds and modifies the overall chemical composition of the breath. Accepting this important observation brings affront the use of eNose as a useful tool for detecting LC by typical “breath print” rather than by a specific component of breath.

Looking to the future, breath tests are not likely to replace the current imaging modalities and certainly not the need of obtaining actual tissue biopsy to make final and accurate diagnosis (especially as the field of personalized medicine and molecular profiling of tumors is becoming more extensive). However, breath analysis has the potential of constituting a powerful tool to aid and enhance early detection of disease as well as reliable characterization of suspected nodules. Ideally, breath collection and analysis will be done by a simple, easy-to-use tool, which will be available at the doctor’s office. Conceptually, breath sample will be collected from a patient with risk factors for cancer development or upon raised clinical suspicion; further on, consecutive breath samples could be collected to assess patient’s response to treatment and to detect early signs of recurrence.

Conclusion

LC research is a rapidly developing field. Further understanding of carcinogenesis pathways can potentially provide a better understanding of LC biomarkers, with a more specific research in the breath analysis field. Standardization of breath collection process therefore appears crucial in order to allow further advances in breath analysis research. Breath analysis might not easily replace traditional methods of LC detection and histological characterization, but it surely has a promising potential to augment detection techniques as a noninvasive, non-expensive and rather simple tool.