ORIGINAL ARTICLE
Mass Spectrometric Analysis of Exhaled Breath for the Identification of Volatile Organic Compound Biomarkers in Esophageal and Gastric Adenocarcinoma Sacheen Kumar, MRCS, Juzheng Huang, PhD, Nima Abbassi-Ghadi, MRCS, Hugh A. Mackenzie, MRCS, Kirill A. Veselkov, PhD, Jonathan M. Hoare, PhD, FRCP,y Laurence B. Lovat, PhD, FRCP,z Patrik Sˇpaneˇl, PhD,§ David Smith, PhD, FRS,ô and George B. Hanna, PhD, FRCS
Objective: The present study assessed whether exhaled breath analysis using Selected Ion Flow Tube Mass Spectrometry could distinguish esophageal and gastric adenocarcinoma from noncancer controls. Background: The majority of patients with upper gastrointestinal cancer present with advanced disease, resulting in poor long-term survival rates. Novel methods are needed to diagnose potentially curable upper gastrointestinal malignancies. Methods: A Profile-3 Selected Ion Flow Tube Mass Spectrometry instrument was used for analysis of volatile organic compounds (VOCs) within exhaled breath samples. All study participants had undergone upper gastrointestinal endoscopy on the day of breath sampling. Receiver operating characteristic analysis and a diagnostic risk prediction model were used to assess the discriminatory accuracy of the identified VOCs. Results: Exhaled breath samples were analyzed from 81 patients with esophageal (N ¼ 48) or gastric adenocarcinoma (N ¼ 33) and 129 controls including Barrett’s metaplasia (N ¼ 16), benign upper gastrointestinal diseases (N ¼ 62), or a normal upper gastrointestinal tract (N ¼ 51). Twelve VOCs—pentanoic acid, hexanoic acid, phenol, methyl phenol, ethyl phenol, butanal, pentanal, hexanal, heptanal, octanal, nonanal, and decanal—were present at significantly higher concentrations (P < 0.05) in the cancer groups than in the noncancer controls. The area under the ROC curve using these significant VOCs to discriminate esophageal and gastric adenocarcinoma from those with normal upper gastrointestinal tracts was 0.97 and 0.98, respectively. The area under the ROC curve for the model and validation subsets of the diagnostic prediction model was 0.92 0.01 and 0.87 0.03, respectively. Conclusions: Distinct exhaled breath VOC profiles can distinguish patients with esophageal and gastric adenocarcinoma from noncancer controls.
From the Department of Surgery and Cancer, Imperial College London, St Mary’s Hospital, London, UK; yDepartment of Medicine, Imperial College London, St Mary’s Hospital, London, UK; zDepartment of Surgery and Interventional Science, National Medical Laser Centre, University College London, London, UK; §J. Heyrovsky Institute of Physical Chemistry, Academy of Sciences of the Czech Republic, Prague, Czech Republic; and ôInstitute for Science and Technology in Medicine, Keele University, Guy Hilton Research Centre, Hartshill, UK. Disclosure: Supported by the Imperial National Institute for Health Research Biomedical Research Centre, Rosetrees Trust and The Stoneygate Trust (M261CD1), and the Imperial College Junior Fellowship Research Program. The authors declare no conflicts of interest. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.annalsofsurgery.com). Reprints: George B Hanna, PhD, FRCS, Department of Surgery and Cancer, Imperial College London, 10th Floor QEQM Wing, St Mary’s Hospital, London W2 1NY, UK. E-mail: [email protected]. Copyright ß 2015 Wolters Kluwer Health, Inc. All rights reserved. ISSN: 0003-4932/14/26105-0821 DOI: 10.1097/SLA.0000000000001101
Annals of Surgery Volume 262, Number 6, December 2015
Keywords: breath analysis, esophageal cancer, gastric cancer, mass spectrometry, volatile organic compounds
(Ann Surg 2015;262:981–990)
E
sophageal and gastric malignancies account for 15% of cancerrelated deaths globally.1 Only one-third of patients diagnosed with esophagogastric (EG) cancer are considered suitable for treatment with curative intent.2 Between 2003 and 2009, the 5-year relative survival rate in the United States for esophageal cancer was 17.9% and for gastric cancer was 27.7%.3 The lack of ‘‘alarm’’ symptoms until the disease is at a more advanced stage contribute to these poor statistics. However, diagnosis of these cancers at a treatable stage is associated with a significant survival benefit. Treatment of early esophageal adenocarcinoma has a reported overall 5-year survival rate of 83% to 96%.4 However, in Western countries, an endoscopy-based screening approach is not justified given the low incidence of significant pathology and associated health care costs [estimated at $88,000 for detecting each upper gastrointestinal (UGI) malignancy].5 Alarm symptoms also have low predictive value for underlying malignancy, despite forming a major part of urgent referral guidance for primary care physicians.6 Thus, there remains an important clinical need to develop novel methods for early disease detection in esophageal and gastric cancer. Breath analysis is a diagnostic modality with few routine applications in clinical practice. Notable examples include the measurement of 13C urea for Helicobacter pylori bacterium, hydrogen breath testing for small bowel bacterial overgrowth, and exhaled nitric oxide in asthma.7– 9 However, human breath is also a complex gaseous biological sample containing more than 250 volatile organic compounds (VOCs).10 In recent years, the role of VOCs in cancer has been investigated using both innovative and conventional methods. Sonoda et al11 demonstrated that canine olfaction of VOCs within exhaled breath and stool samples could discriminate patients with colorectal cancer from controls with good accuracy. The study reinforced the theory that cancer-related VOCs exist, and investigation of these compounds can be undertaken using modern chemical analytical techniques. Several studies employing Gas-Chromatography Mass-Spectrometry (GC-MS) have identified potential VOCs of interest within exhaled breath in cancer.12–15 However, GC-MS is an off-line analytical technique that requires breath adsorption devices (eg, Solid Phase Micro-extraction) and column calibration for desired analytes. GC-MS is also less preferable to quantify some important trace compounds that are present in exhaled breath and has lengthy processing times. Technological advances have resulted in the development of novel chemical analytical techniques including proton transfer reaction mass spectrometry and Selected Ion Flow Tube Mass Spectrometry (SIFT-MS), both of which offer real-time quantification of VOCs.16,17 SIFT-MS is a valuable method for www.annalsofsurgery.com | 981
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Kumar et al
Annals of Surgery Volume 262, Number 6, December 2015
ambient gas analysis that utilizes a soft ionization method, which can accurately analyze the trace organic compounds present in air and other complex matrices including exhaled breath and the headspace of blood, urine, and cell and bacterial cultures. The samples can be analyzed in real time without modifying or disturbing the medium, which can occur, for example, when collecting and concentrating the gas phase trace organic compounds onto surfaces for subsequent analysis, such as in GC-MS. It is particularly suited for analysis in the clinical environment given there is no need for any sample preconcentration steps. SIFT-MS has previously been employed to investigate exhaled breath in coeliac disease, chronic renal failure, and for therapeutic monitoring in community-acquired pneumonia.18– 20 We have previously developed and validated our sampling procedures for VOCs within biofluids and exhaled breath.21–23 The aims of the current study were (1) to identify and quantify exhaled breath VOCs using SIFT-MS in patients with esophageal or gastric adenocarcinoma and compare them to noncancer controls and (2) to construct a VOC-based risk prediction model to distinguish patients with esophageal and gastric adenocarcinoma from noncancer controls.
the UGI tract, liver disease, small bowel/colonic pathology, any nonUGI cancer, and those with any signs/symptoms of acute infection. Approval for the study was obtained from the institutional ethics review committee. Fully informed, written consent was obtained from all participants before enrolment in the study. Demographic and clinical data were collected using a standardized proforma and archived in a linked, anonymized database. The study enrolled 210 consecutive patients (81 cancer cases and 129 controls) during the recruitment period. There were 48 patients with esophageal adenocarcinoma (stage I—19%, stage II—27%, and stage III—54%) and 33 patients with gastric adenocarcinoma (stage I—24%, stage II—21%, and stage III—55%). The control groups had 16 patients with Barrett’s metaplasia, 62 patients classified as benign controls, and 51 patients were included in the normal UGI tract group. Patients within the benign group had the following diagnoses: esophagitis (N ¼ 15), esophageal stricture (N ¼ 5), esophageal candidiasis (N ¼ 2), gastritis (N ¼ 20), gastric ulcer (N ¼ 6), gastrojejunostomy-induced gastritis (N ¼ 2), duodenitis (N ¼ 9), and duodenal ulcer (N ¼ 3). Table 1 outlines the demographic and clinical data for the patient groups. Details of the clinical staging of the cancer groups are provided in Table 2.
METHODS Exhaled Breath Sample Collection Study Population Eligible patients referred for UGI endoscopy were recruited through Imperial College Healthcare NHS Trust between November 2011 and August 2013. Exhaled breath samples were collected when patients were attending for one of the following investigations— staging laparoscopy and esophago-gastro-duodenoscopy (EGD), UGI endoscopic ultrasound, or EGD only. The EGD component of the investigations and histological confirmation formed the reference standard for classification of patients within the study groups. Patients with histologically confirmed invasive esophageal or gastric adenocarcinoma (before the commencement of any treatment) and considered for an ‘‘intention-to-cure’’ treatment pathway were included in the cancer cohort. Patients with gastroesophageal junctional tumors were all classified within the esophageal cancer group. The noncancer controls were classified according to endoscopy findings and biopsy results into Barrett’s metaplasia, benign disease, and those with a normal UGI tract and negative rapid urease test. Exclusion criteria included patients with squamous cell carcinoma of
Breath sampling and analysis procedures were conducted using a validated method.21 All patients were fasted for a minimum of 6 hours before breath sample collection. Mixed alveolar breath samples were collected before the patients’ scheduled clinical investigations in secure double thickness (2 25 mm) Nalophan (Kalle UK Ltd., Witham, United Kingdom) bags via a 1-mL Luer lok syringe (Terumo Europe, Leuven, Belgium). Breath sampling bags were subject to strict quality control protocols, including every Nalophan bag having a fixed volume of 2 L and being washed with dry synthetic air (BOC Ltd, Guildford, United Kingdom) before its use. Initially, all participants were requested to perform a single deep nasal inhalation (as close to total lung capacity as possible), followed by complete exhalation via their mouth. This exhaled breath was not collected for analysis but served the purpose of demonstrating the breath maneuver required for sampling to the subject. Participants were then requested to repeat this procedure using the same breath maneuver directly into the Nalophan bag via the aperture of the 1-mL Luer lok syringe barrel. All breath samples were analyzed within an
TABLE 1. Patient Characteristics Across Study Groups
Age (IQR), yrz Sex (male:female) Smoking Nonsmoker Ex-smoker Smoker Alcohol intake Within guidance Excess Diabetes mellitus Hypertension/IHD Respiratory disorder Renal disease H. Pylori status
Gastric Adenocarcinoma (n ¼ 33)
Esophageal Adenocarcinoma (n ¼ 48)
Barrett’s Metaplasia (n ¼ 16)
Benign Controls (n ¼ 62)
Normal Upper Gastrointestinal Tract (n ¼ 51)
58 (50.5–70.5) 24:9
63.5 (55.3–72.8) 40:8
67 (58.8–72.8) 11:5
64.5 (50–72) 38:24
61 (50–73) 30:21
14 (42.4) 15 (45.5) 4 (12.1)
21 (43.8) 17 (35.4) 10 (20.8)
8 (50.0) 4 (25.0) 4 (25.0)
32 (51.6) 17 (27.4) 13 (21.0)
33 (64.7) 11 (21.6) 7 (13.7)
32 (97.0) 1 (3.0) 6 (18.2) 13 (39.4) 4 (12.1) 2 (6.1) 4 (12.1)
43 (89.6) 5 (10.4) 4 (8.3) 23 (47.9) 5 (10.4) 2 (4.2) 0 (0.0)
15 (93.8) 1 (6.3) 2 (12.5) 7 (43.8) 2 (12.5) 1 (6.3) 0 (0.0)
58 (93.6) 4 (6.5) 8 (12.9) 22 (35.5) 8 (12.9) 2 (3.2) 4 (6.5)
48 3 13 23 5 0 0
P 0.47y 0.07 0.31
0.76 (94.1) (5.9) (25.5) (45.1) (9.8) (0.0) (0.0)
0.18 0.72 0.99 0.55 0.07
Percentages are given within parentheses. 2 x test, except. yKruskal-Wallis Test. zValues are median (IQR), otherwise n (%).
982 | www.annalsofsurgery.com
ß
2015 Wolters Kluwer Health, Inc. All rights reserved.
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Annals of Surgery Volume 262, Number 6, December 2015
Exhaled Breath Analysis in Esophageal and Gastric Cancer
TABLE 2. Clinical Staging of the Gastric and Esophageal Adenocarcinoma Groups
groups were included in a binary logistic regression model. Receiver operating characteristic (ROC) curves are constructed by plotting the sensitivity against 1specificity for various thresholds or variables of a diagnostic test. The area under the curve (AUC) measures the discrimination power of the test, which is the ability of the test to correctly classify subjects into the appropriate group. The accuracies of the models were assessed using the AUC of the ROC curves.24 Details of ROC construction are given in Supplemental Material, available at http://links.lww.com/SLA/A705. Potential confounding factors across the study groups were also evaluated by employing the Kruskal-Wallis test for age and x2 test to assess for differences in sex, ischemic heart disease, diabetes mellitus, respiratory disorders, renal disease, medications, H. pylori status, smoking status, and alcohol intake. Linear regression models to assess any influence of patient demographic factors, UGI factors, and medications on VOC concentrations were also created. Each individual VOC was used as a dependent variable and each potential confounding factor and cancer disease status as the independent variables. Age was continuous, whereas the remaining independent variables were categorical; medications, H. pylori status, and cancer disease status were binary, either present or absent, and smoking status and alcohol intake were ternary (smoker, ex-smoker, nonsmoker, and no alcohol, within guidance, excess of guidance).
Characteristics Site of tumor Lower esophagus Gastroesophageal junction Proximal stomach Distal stomach T Stage T1 T2 T3 T4 Lymph node status N0 N1 N2 N3
Gastric Adenocarcinoma (n ¼ 33)
Esophageal Adenocarcinoma (n ¼ 48)
N/A N/A 17 (51.5%) 16 (48.5%)
23 (47.9%) 25 (52.1%) N/A N/A
4 (12.1%) 11 (33.3%) 12 (36.4%) 6 (18.2%)
11 (22.9%) 7 (14.6%) 29 (60.4%) 1 (2.1%)
10 (30.3%) 11 (33.3%) 12 (36.4%) 0 (0.0)
19 (39.5%) 27 (56.3%) 2 (4.2%) 0 (0%)
N/A indicates not applicable.
hour of collection. Analysis of the exhaled breath samples was conducted by a research team member blinded to the endoscopy result. Additional experiments to investigate any potential contribution of the surrounding environment and release from the Nalophan bag surface to the measured VOC concentrations were also conducted. The intersample reliability and reproducibility of the aforementioned breath sampling methodology has also previously been evaluated.21
Analysis by SIFT-MS SIFT-MS is a real-time mass spectrometric analytical technique particularly suited to exhaled breath analysis; it allows the simultaneous quantification of several VOCs within a gaseous mixture. A Profile-3 SIFT-MS instrument (Instrument Science, Crewe, United Kingdom) was employed to analyze exhaled breath samples using the Multiple Ion Monitoring Mode. In this study, the Nalophan bag containing the breath sample was directly connected to the sample inlet arm of the SIFT-MS instrument and enclosed within an incubator held at 378C for the duration of the analysis. The breath sample automatically flowed into the helium carrier gas at a fixed rate (20 mL/min) through the sampling line, which is held at a constant temperature of 808C. Selected VOCs from exhaled breath samples were analyzed for a total of 60 seconds and the measured concentrations were averaged over this analysis time for each VOC. The principle of SIFT-MS and analytical procedures employed is provided in the Supplemental Material, available at http://links. lww.com/SLA/A705.
Statistical Analysis of VOC Data Statistical analysis was performed using IBM SPSS Statistics 20 (IBM Corp., Armonk, NY). A P value of less than 0.05 was considered significant, and all statistical tests were 2-sided. MannWhitney U tests were used to compare the concentrations of the measured VOCs. The comparisons made were gastric adenocarcinoma versus normal UGI tract, esophageal adenocarcinoma versus normal UGI tract, gastric adenocarcinoma versus noncancer controls (benign controls and normal UGI tract), esophageal adenocarcinoma versus noncancer controls (Barrett’s metaplasia and benign controls and normal UGI tract), gastric adenocarcinoma versus esophageal adenocarcinoma, and finally benign controls versus normal UGI tract. The significantly different VOCs between the aforementioned ß
2015 Wolters Kluwer Health, Inc. All rights reserved.
Risk Prediction Model A diagnostic prediction model was created for EG adenocarcinoma versus noncancer controls using binary logistic regression analysis on the basis of subset of VOCs statistically relevant for discrimination. The cancer group included both esophageal and gastric adenocarcinoma; this combined group was tested as these cancers are considered comparable subtypes that have frequently been grouped together in neoadjuvant chemotherapy trials.25 The noncancer group included the Barrett’s metaplasia, benign disease, and the normal UGI tract controls. Step 1, we performed a 10-fold cross-validation procedure for all VOCs. Using this analysis, 10% of randomly selected patients are left out at a time and the model is constructed on the basis of the remaining 90% of the data. As part of this model building process, we employed analysis of variance to select the VOCs relevant for discrimination between the 2 groups on the basis of the P value threshold of 0.05. The same procedure is iteratively applied until the ‘‘cancer’’ versus ‘‘noncancer’’ status of each patient has been predicted. Using a 10-fold cross-validation analysis results in 10 separate models being constructed, and a frequency histogram demonstrating the number of times a given VOC has been selected for discrimination using analysis of variance is produced. Step 2, the data were then split through a random number generation process; 2/3 of data were used to create the prediction model (model subset) and 1/3 to test the model (validation subset). The statistically significant VOCs were entered into a logistic regression analysis in a stepwise fashion. The most significant predictors were used to create the diagnostic prediction model. The accuracy of the resultant models was assessed using the area under the ROC curve, sensitivity and specificity measures. To estimate variability, we performed a Monte Carlo simulation by repeating the aforementioned procedure 10 times, with the resultant AUC presented as a mean standard deviation.
RESULTS Patient Characteristics A total of 210 patients were recruited during the study period. The 81 patients with EG adenocarcinoma and 129 noncancer controls had median ages of 62 years (range, 53–71 years) and 64 years (range, 51–72 years), respectively. There were no statistically www.annalsofsurgery.com | 983
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Kumar et al
Annals of Surgery Volume 262, Number 6, December 2015
significant differences in age, sex, or potential confounding factors across the groups except for H. pylori status, which (as expected) was higher in gastric cancer and benign groups. With regard to medications, aspirin was the only drug that demonstrated a statistical difference across the groups (Supplemental Digital Content, Table 1, available at http://links.lww.com/SLA/A705). Any potential influence of all medications that demonstrated a P value of less than 0.2 was further assessed with linear regression analysis.
The analytical information, including chemical formula, precursor ions, m/z ratio, and characteristic product ions of the investigated VOCs, is given in the Supplemental Digital Content, Table 2, available at http://links.lww.com/SLA/A705. The heatmap demonstrated higher concentrations in the region of the fatty acids, phenols, and aldehydes within both cancer groups (Fig. 1). Univariate MannWhitney U analysis revealed 12 VOCs at significantly higher concentrations in both the gastric and esophageal adenocarcinoma groups than in noncancer controls. These were pentanoic acid, hexanoic acid, phenol, methyl phenol, ethyl phenol, butanal, pentanal, hexanal, heptanal, octanal, nonanal, and decanal (Fig. 1). In addition to these VOCs, butyric acid was significantly higher in the esophageal adenocarcinoma group than in the noncancer controls. There was no significant difference in the abundant breath VOCs
Volatile Organic Compound Analysis A total of 29 VOCs from 8 major chemical groups present within exhaled breath (including alcohols, phenols, ketones, fatty acids, aldehydes, sulfur-containing compounds, nitrogen compounds, and ether) were unambiguously identified and quantified.
FIGURE 1. A, Heatmap of the investigated VOCs from all patients across the study groups. Each column in the heatmap represents 1 patient (n ¼ 210). In the heatmap, the study groups include the normal upper gastrointestinal tract, the benign disease controls, Barrett’s metaplasia, esophageal cancer, and gastric cancer groups. Each value in the unit is the measured concentration (ppbv) of the specific VOC. Within the color scheme, the 25th lower quartile, median, and the 75th upper quartile of the measured concentrations for specific VOC are defined as pure green, yellow, and red, respectively. B–G, Box & Whisker plots of the measured concentrations (ppbv) of hexanoic acid, phenol, methyl phenol, ethyl phenol, nonanal, and decanal in the exhaled breath of the study groups. Normal represents the normal UGI tract group; benign represents the benign disease group; Barrett’s is the Barrett’s metaplasia group; EC is the esophageal cancer group, and GC is the gastric cancer group. 984 | www.annalsofsurgery.com
ß
2015 Wolters Kluwer Health, Inc. All rights reserved.
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
ß
2015 Wolters Kluwer Health, Inc. All rights reserved.
36 12 14 10 8 12 4 2 5 5 5 5 10 233 262 452 69
Median
IQR
23–61 9–18 10–22 6–25 3–10 7–21 2–7 2–7 3–10 2–7 3–9 2–10 5–17 157–333 146–573 275–687 49–100
40 12 13 9 7 8 4 3 6 4 3 3 8 245 305 478 67
Median 29–55 8–19 9–30 6–18 4–13 4–14 2–7 2–7 3–11 2–6 1–7 1–8 5–15 180–344 133–588 266–635 49–109
IQR
Esophageal Cancer
40 9 6 5 3 4 3 2 5 2 2 1 2 258 389 549 64
Median 24–43 6–14 5–11 1–11 2–5 2–7 2–4 1–4 1–6 1–4 1–5 0–3 0–5 171–317 157–674 346–705 41–79
IQR
Barrett’s Metaplasia
33 9 7 5 4 5 3 2 4 3 2 2 3 248 341 367 58
Median 23–43 6–17 5–10 2–8 2–6 2–8 1–6 0–3 2–7 1–5 1–3 1–4 2–5 187–291 233–596 190–570 41–89
IQR
Benign Controls
Median and interquartile range (IQR) are measured by parts-per-billion by volume (ppbv). The noncancer controls comprise the normal upper gastrointestinal tract, benign and Barrett’s metaplasia groups. yThe noncancer controls comprises normal upper gastrointestinal tract and benign groups. zThe carcinoma group comprises gastric adenocarcinoma and esophageal adenocarcinoma.
Butyric acid Pentanoic acid Hexanoic acid Phenol Methyl phenol Ethyl phenol Butanal Pentanal Hexanal Heptanal Octanal Nonanal Decanal Methanol Acetone Ammonia Isoprene
Gastric Cancer
31 10 7 4 3 4 2 2 3 2 1 1 2 248 338 330 64
Median 22–43 8–15 3–10 3–7 2–5 2–6 1–4 0–3 2–5 1–4 1–2 0–2 1–3 204–315 190–789 277–575 51–83
IQR
Normal Upper Gastrointestinal Tract
TABLE 3. Concentrations of the Statistically Significant and Abundant VOCs Across the Study Groups
0.007 0.02
Mass Spectrometric Analysis of Exhaled Breath for the Identification of Volatile Organic Compound Biomarkers in Esophageal and Gastric Adenocarcinoma Sacheen Kumar, MRCS, Juzheng Huang, PhD, Nima Abbassi-Ghadi, MRCS, Hugh A. Mackenzie, MRCS, Kirill A. Veselkov, PhD, Jonathan M. Hoare, PhD, FRCP,y Laurence B. Lovat, PhD, FRCP,z Patrik Sˇpaneˇl, PhD,§ David Smith, PhD, FRS,ô and George B. Hanna, PhD, FRCS
Objective: The present study assessed whether exhaled breath analysis using Selected Ion Flow Tube Mass Spectrometry could distinguish esophageal and gastric adenocarcinoma from noncancer controls. Background: The majority of patients with upper gastrointestinal cancer present with advanced disease, resulting in poor long-term survival rates. Novel methods are needed to diagnose potentially curable upper gastrointestinal malignancies. Methods: A Profile-3 Selected Ion Flow Tube Mass Spectrometry instrument was used for analysis of volatile organic compounds (VOCs) within exhaled breath samples. All study participants had undergone upper gastrointestinal endoscopy on the day of breath sampling. Receiver operating characteristic analysis and a diagnostic risk prediction model were used to assess the discriminatory accuracy of the identified VOCs. Results: Exhaled breath samples were analyzed from 81 patients with esophageal (N ¼ 48) or gastric adenocarcinoma (N ¼ 33) and 129 controls including Barrett’s metaplasia (N ¼ 16), benign upper gastrointestinal diseases (N ¼ 62), or a normal upper gastrointestinal tract (N ¼ 51). Twelve VOCs—pentanoic acid, hexanoic acid, phenol, methyl phenol, ethyl phenol, butanal, pentanal, hexanal, heptanal, octanal, nonanal, and decanal—were present at significantly higher concentrations (P < 0.05) in the cancer groups than in the noncancer controls. The area under the ROC curve using these significant VOCs to discriminate esophageal and gastric adenocarcinoma from those with normal upper gastrointestinal tracts was 0.97 and 0.98, respectively. The area under the ROC curve for the model and validation subsets of the diagnostic prediction model was 0.92 0.01 and 0.87 0.03, respectively. Conclusions: Distinct exhaled breath VOC profiles can distinguish patients with esophageal and gastric adenocarcinoma from noncancer controls.
From the Department of Surgery and Cancer, Imperial College London, St Mary’s Hospital, London, UK; yDepartment of Medicine, Imperial College London, St Mary’s Hospital, London, UK; zDepartment of Surgery and Interventional Science, National Medical Laser Centre, University College London, London, UK; §J. Heyrovsky Institute of Physical Chemistry, Academy of Sciences of the Czech Republic, Prague, Czech Republic; and ôInstitute for Science and Technology in Medicine, Keele University, Guy Hilton Research Centre, Hartshill, UK. Disclosure: Supported by the Imperial National Institute for Health Research Biomedical Research Centre, Rosetrees Trust and The Stoneygate Trust (M261CD1), and the Imperial College Junior Fellowship Research Program. The authors declare no conflicts of interest. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.annalsofsurgery.com). Reprints: George B Hanna, PhD, FRCS, Department of Surgery and Cancer, Imperial College London, 10th Floor QEQM Wing, St Mary’s Hospital, London W2 1NY, UK. E-mail: [email protected]. Copyright ß 2015 Wolters Kluwer Health, Inc. All rights reserved. ISSN: 0003-4932/14/26105-0821 DOI: 10.1097/SLA.0000000000001101
Annals of Surgery Volume 262, Number 6, December 2015
Keywords: breath analysis, esophageal cancer, gastric cancer, mass spectrometry, volatile organic compounds
(Ann Surg 2015;262:981–990)
E
sophageal and gastric malignancies account for 15% of cancerrelated deaths globally.1 Only one-third of patients diagnosed with esophagogastric (EG) cancer are considered suitable for treatment with curative intent.2 Between 2003 and 2009, the 5-year relative survival rate in the United States for esophageal cancer was 17.9% and for gastric cancer was 27.7%.3 The lack of ‘‘alarm’’ symptoms until the disease is at a more advanced stage contribute to these poor statistics. However, diagnosis of these cancers at a treatable stage is associated with a significant survival benefit. Treatment of early esophageal adenocarcinoma has a reported overall 5-year survival rate of 83% to 96%.4 However, in Western countries, an endoscopy-based screening approach is not justified given the low incidence of significant pathology and associated health care costs [estimated at $88,000 for detecting each upper gastrointestinal (UGI) malignancy].5 Alarm symptoms also have low predictive value for underlying malignancy, despite forming a major part of urgent referral guidance for primary care physicians.6 Thus, there remains an important clinical need to develop novel methods for early disease detection in esophageal and gastric cancer. Breath analysis is a diagnostic modality with few routine applications in clinical practice. Notable examples include the measurement of 13C urea for Helicobacter pylori bacterium, hydrogen breath testing for small bowel bacterial overgrowth, and exhaled nitric oxide in asthma.7– 9 However, human breath is also a complex gaseous biological sample containing more than 250 volatile organic compounds (VOCs).10 In recent years, the role of VOCs in cancer has been investigated using both innovative and conventional methods. Sonoda et al11 demonstrated that canine olfaction of VOCs within exhaled breath and stool samples could discriminate patients with colorectal cancer from controls with good accuracy. The study reinforced the theory that cancer-related VOCs exist, and investigation of these compounds can be undertaken using modern chemical analytical techniques. Several studies employing Gas-Chromatography Mass-Spectrometry (GC-MS) have identified potential VOCs of interest within exhaled breath in cancer.12–15 However, GC-MS is an off-line analytical technique that requires breath adsorption devices (eg, Solid Phase Micro-extraction) and column calibration for desired analytes. GC-MS is also less preferable to quantify some important trace compounds that are present in exhaled breath and has lengthy processing times. Technological advances have resulted in the development of novel chemical analytical techniques including proton transfer reaction mass spectrometry and Selected Ion Flow Tube Mass Spectrometry (SIFT-MS), both of which offer real-time quantification of VOCs.16,17 SIFT-MS is a valuable method for www.annalsofsurgery.com | 981
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Kumar et al
Annals of Surgery Volume 262, Number 6, December 2015
ambient gas analysis that utilizes a soft ionization method, which can accurately analyze the trace organic compounds present in air and other complex matrices including exhaled breath and the headspace of blood, urine, and cell and bacterial cultures. The samples can be analyzed in real time without modifying or disturbing the medium, which can occur, for example, when collecting and concentrating the gas phase trace organic compounds onto surfaces for subsequent analysis, such as in GC-MS. It is particularly suited for analysis in the clinical environment given there is no need for any sample preconcentration steps. SIFT-MS has previously been employed to investigate exhaled breath in coeliac disease, chronic renal failure, and for therapeutic monitoring in community-acquired pneumonia.18– 20 We have previously developed and validated our sampling procedures for VOCs within biofluids and exhaled breath.21–23 The aims of the current study were (1) to identify and quantify exhaled breath VOCs using SIFT-MS in patients with esophageal or gastric adenocarcinoma and compare them to noncancer controls and (2) to construct a VOC-based risk prediction model to distinguish patients with esophageal and gastric adenocarcinoma from noncancer controls.
the UGI tract, liver disease, small bowel/colonic pathology, any nonUGI cancer, and those with any signs/symptoms of acute infection. Approval for the study was obtained from the institutional ethics review committee. Fully informed, written consent was obtained from all participants before enrolment in the study. Demographic and clinical data were collected using a standardized proforma and archived in a linked, anonymized database. The study enrolled 210 consecutive patients (81 cancer cases and 129 controls) during the recruitment period. There were 48 patients with esophageal adenocarcinoma (stage I—19%, stage II—27%, and stage III—54%) and 33 patients with gastric adenocarcinoma (stage I—24%, stage II—21%, and stage III—55%). The control groups had 16 patients with Barrett’s metaplasia, 62 patients classified as benign controls, and 51 patients were included in the normal UGI tract group. Patients within the benign group had the following diagnoses: esophagitis (N ¼ 15), esophageal stricture (N ¼ 5), esophageal candidiasis (N ¼ 2), gastritis (N ¼ 20), gastric ulcer (N ¼ 6), gastrojejunostomy-induced gastritis (N ¼ 2), duodenitis (N ¼ 9), and duodenal ulcer (N ¼ 3). Table 1 outlines the demographic and clinical data for the patient groups. Details of the clinical staging of the cancer groups are provided in Table 2.
METHODS Exhaled Breath Sample Collection Study Population Eligible patients referred for UGI endoscopy were recruited through Imperial College Healthcare NHS Trust between November 2011 and August 2013. Exhaled breath samples were collected when patients were attending for one of the following investigations— staging laparoscopy and esophago-gastro-duodenoscopy (EGD), UGI endoscopic ultrasound, or EGD only. The EGD component of the investigations and histological confirmation formed the reference standard for classification of patients within the study groups. Patients with histologically confirmed invasive esophageal or gastric adenocarcinoma (before the commencement of any treatment) and considered for an ‘‘intention-to-cure’’ treatment pathway were included in the cancer cohort. Patients with gastroesophageal junctional tumors were all classified within the esophageal cancer group. The noncancer controls were classified according to endoscopy findings and biopsy results into Barrett’s metaplasia, benign disease, and those with a normal UGI tract and negative rapid urease test. Exclusion criteria included patients with squamous cell carcinoma of
Breath sampling and analysis procedures were conducted using a validated method.21 All patients were fasted for a minimum of 6 hours before breath sample collection. Mixed alveolar breath samples were collected before the patients’ scheduled clinical investigations in secure double thickness (2 25 mm) Nalophan (Kalle UK Ltd., Witham, United Kingdom) bags via a 1-mL Luer lok syringe (Terumo Europe, Leuven, Belgium). Breath sampling bags were subject to strict quality control protocols, including every Nalophan bag having a fixed volume of 2 L and being washed with dry synthetic air (BOC Ltd, Guildford, United Kingdom) before its use. Initially, all participants were requested to perform a single deep nasal inhalation (as close to total lung capacity as possible), followed by complete exhalation via their mouth. This exhaled breath was not collected for analysis but served the purpose of demonstrating the breath maneuver required for sampling to the subject. Participants were then requested to repeat this procedure using the same breath maneuver directly into the Nalophan bag via the aperture of the 1-mL Luer lok syringe barrel. All breath samples were analyzed within an
TABLE 1. Patient Characteristics Across Study Groups
Age (IQR), yrz Sex (male:female) Smoking Nonsmoker Ex-smoker Smoker Alcohol intake Within guidance Excess Diabetes mellitus Hypertension/IHD Respiratory disorder Renal disease H. Pylori status
Gastric Adenocarcinoma (n ¼ 33)
Esophageal Adenocarcinoma (n ¼ 48)
Barrett’s Metaplasia (n ¼ 16)
Benign Controls (n ¼ 62)
Normal Upper Gastrointestinal Tract (n ¼ 51)
58 (50.5–70.5) 24:9
63.5 (55.3–72.8) 40:8
67 (58.8–72.8) 11:5
64.5 (50–72) 38:24
61 (50–73) 30:21
14 (42.4) 15 (45.5) 4 (12.1)
21 (43.8) 17 (35.4) 10 (20.8)
8 (50.0) 4 (25.0) 4 (25.0)
32 (51.6) 17 (27.4) 13 (21.0)
33 (64.7) 11 (21.6) 7 (13.7)
32 (97.0) 1 (3.0) 6 (18.2) 13 (39.4) 4 (12.1) 2 (6.1) 4 (12.1)
43 (89.6) 5 (10.4) 4 (8.3) 23 (47.9) 5 (10.4) 2 (4.2) 0 (0.0)
15 (93.8) 1 (6.3) 2 (12.5) 7 (43.8) 2 (12.5) 1 (6.3) 0 (0.0)
58 (93.6) 4 (6.5) 8 (12.9) 22 (35.5) 8 (12.9) 2 (3.2) 4 (6.5)
48 3 13 23 5 0 0
P 0.47y 0.07 0.31
0.76 (94.1) (5.9) (25.5) (45.1) (9.8) (0.0) (0.0)
0.18 0.72 0.99 0.55 0.07
Percentages are given within parentheses. 2 x test, except. yKruskal-Wallis Test. zValues are median (IQR), otherwise n (%).
982 | www.annalsofsurgery.com
ß
2015 Wolters Kluwer Health, Inc. All rights reserved.
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Annals of Surgery Volume 262, Number 6, December 2015
Exhaled Breath Analysis in Esophageal and Gastric Cancer
TABLE 2. Clinical Staging of the Gastric and Esophageal Adenocarcinoma Groups
groups were included in a binary logistic regression model. Receiver operating characteristic (ROC) curves are constructed by plotting the sensitivity against 1specificity for various thresholds or variables of a diagnostic test. The area under the curve (AUC) measures the discrimination power of the test, which is the ability of the test to correctly classify subjects into the appropriate group. The accuracies of the models were assessed using the AUC of the ROC curves.24 Details of ROC construction are given in Supplemental Material, available at http://links.lww.com/SLA/A705. Potential confounding factors across the study groups were also evaluated by employing the Kruskal-Wallis test for age and x2 test to assess for differences in sex, ischemic heart disease, diabetes mellitus, respiratory disorders, renal disease, medications, H. pylori status, smoking status, and alcohol intake. Linear regression models to assess any influence of patient demographic factors, UGI factors, and medications on VOC concentrations were also created. Each individual VOC was used as a dependent variable and each potential confounding factor and cancer disease status as the independent variables. Age was continuous, whereas the remaining independent variables were categorical; medications, H. pylori status, and cancer disease status were binary, either present or absent, and smoking status and alcohol intake were ternary (smoker, ex-smoker, nonsmoker, and no alcohol, within guidance, excess of guidance).
Characteristics Site of tumor Lower esophagus Gastroesophageal junction Proximal stomach Distal stomach T Stage T1 T2 T3 T4 Lymph node status N0 N1 N2 N3
Gastric Adenocarcinoma (n ¼ 33)
Esophageal Adenocarcinoma (n ¼ 48)
N/A N/A 17 (51.5%) 16 (48.5%)
23 (47.9%) 25 (52.1%) N/A N/A
4 (12.1%) 11 (33.3%) 12 (36.4%) 6 (18.2%)
11 (22.9%) 7 (14.6%) 29 (60.4%) 1 (2.1%)
10 (30.3%) 11 (33.3%) 12 (36.4%) 0 (0.0)
19 (39.5%) 27 (56.3%) 2 (4.2%) 0 (0%)
N/A indicates not applicable.
hour of collection. Analysis of the exhaled breath samples was conducted by a research team member blinded to the endoscopy result. Additional experiments to investigate any potential contribution of the surrounding environment and release from the Nalophan bag surface to the measured VOC concentrations were also conducted. The intersample reliability and reproducibility of the aforementioned breath sampling methodology has also previously been evaluated.21
Analysis by SIFT-MS SIFT-MS is a real-time mass spectrometric analytical technique particularly suited to exhaled breath analysis; it allows the simultaneous quantification of several VOCs within a gaseous mixture. A Profile-3 SIFT-MS instrument (Instrument Science, Crewe, United Kingdom) was employed to analyze exhaled breath samples using the Multiple Ion Monitoring Mode. In this study, the Nalophan bag containing the breath sample was directly connected to the sample inlet arm of the SIFT-MS instrument and enclosed within an incubator held at 378C for the duration of the analysis. The breath sample automatically flowed into the helium carrier gas at a fixed rate (20 mL/min) through the sampling line, which is held at a constant temperature of 808C. Selected VOCs from exhaled breath samples were analyzed for a total of 60 seconds and the measured concentrations were averaged over this analysis time for each VOC. The principle of SIFT-MS and analytical procedures employed is provided in the Supplemental Material, available at http://links. lww.com/SLA/A705.
Statistical Analysis of VOC Data Statistical analysis was performed using IBM SPSS Statistics 20 (IBM Corp., Armonk, NY). A P value of less than 0.05 was considered significant, and all statistical tests were 2-sided. MannWhitney U tests were used to compare the concentrations of the measured VOCs. The comparisons made were gastric adenocarcinoma versus normal UGI tract, esophageal adenocarcinoma versus normal UGI tract, gastric adenocarcinoma versus noncancer controls (benign controls and normal UGI tract), esophageal adenocarcinoma versus noncancer controls (Barrett’s metaplasia and benign controls and normal UGI tract), gastric adenocarcinoma versus esophageal adenocarcinoma, and finally benign controls versus normal UGI tract. The significantly different VOCs between the aforementioned ß
2015 Wolters Kluwer Health, Inc. All rights reserved.
Risk Prediction Model A diagnostic prediction model was created for EG adenocarcinoma versus noncancer controls using binary logistic regression analysis on the basis of subset of VOCs statistically relevant for discrimination. The cancer group included both esophageal and gastric adenocarcinoma; this combined group was tested as these cancers are considered comparable subtypes that have frequently been grouped together in neoadjuvant chemotherapy trials.25 The noncancer group included the Barrett’s metaplasia, benign disease, and the normal UGI tract controls. Step 1, we performed a 10-fold cross-validation procedure for all VOCs. Using this analysis, 10% of randomly selected patients are left out at a time and the model is constructed on the basis of the remaining 90% of the data. As part of this model building process, we employed analysis of variance to select the VOCs relevant for discrimination between the 2 groups on the basis of the P value threshold of 0.05. The same procedure is iteratively applied until the ‘‘cancer’’ versus ‘‘noncancer’’ status of each patient has been predicted. Using a 10-fold cross-validation analysis results in 10 separate models being constructed, and a frequency histogram demonstrating the number of times a given VOC has been selected for discrimination using analysis of variance is produced. Step 2, the data were then split through a random number generation process; 2/3 of data were used to create the prediction model (model subset) and 1/3 to test the model (validation subset). The statistically significant VOCs were entered into a logistic regression analysis in a stepwise fashion. The most significant predictors were used to create the diagnostic prediction model. The accuracy of the resultant models was assessed using the area under the ROC curve, sensitivity and specificity measures. To estimate variability, we performed a Monte Carlo simulation by repeating the aforementioned procedure 10 times, with the resultant AUC presented as a mean standard deviation.
RESULTS Patient Characteristics A total of 210 patients were recruited during the study period. The 81 patients with EG adenocarcinoma and 129 noncancer controls had median ages of 62 years (range, 53–71 years) and 64 years (range, 51–72 years), respectively. There were no statistically www.annalsofsurgery.com | 983
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Kumar et al
Annals of Surgery Volume 262, Number 6, December 2015
significant differences in age, sex, or potential confounding factors across the groups except for H. pylori status, which (as expected) was higher in gastric cancer and benign groups. With regard to medications, aspirin was the only drug that demonstrated a statistical difference across the groups (Supplemental Digital Content, Table 1, available at http://links.lww.com/SLA/A705). Any potential influence of all medications that demonstrated a P value of less than 0.2 was further assessed with linear regression analysis.
The analytical information, including chemical formula, precursor ions, m/z ratio, and characteristic product ions of the investigated VOCs, is given in the Supplemental Digital Content, Table 2, available at http://links.lww.com/SLA/A705. The heatmap demonstrated higher concentrations in the region of the fatty acids, phenols, and aldehydes within both cancer groups (Fig. 1). Univariate MannWhitney U analysis revealed 12 VOCs at significantly higher concentrations in both the gastric and esophageal adenocarcinoma groups than in noncancer controls. These were pentanoic acid, hexanoic acid, phenol, methyl phenol, ethyl phenol, butanal, pentanal, hexanal, heptanal, octanal, nonanal, and decanal (Fig. 1). In addition to these VOCs, butyric acid was significantly higher in the esophageal adenocarcinoma group than in the noncancer controls. There was no significant difference in the abundant breath VOCs
Volatile Organic Compound Analysis A total of 29 VOCs from 8 major chemical groups present within exhaled breath (including alcohols, phenols, ketones, fatty acids, aldehydes, sulfur-containing compounds, nitrogen compounds, and ether) were unambiguously identified and quantified.
FIGURE 1. A, Heatmap of the investigated VOCs from all patients across the study groups. Each column in the heatmap represents 1 patient (n ¼ 210). In the heatmap, the study groups include the normal upper gastrointestinal tract, the benign disease controls, Barrett’s metaplasia, esophageal cancer, and gastric cancer groups. Each value in the unit is the measured concentration (ppbv) of the specific VOC. Within the color scheme, the 25th lower quartile, median, and the 75th upper quartile of the measured concentrations for specific VOC are defined as pure green, yellow, and red, respectively. B–G, Box & Whisker plots of the measured concentrations (ppbv) of hexanoic acid, phenol, methyl phenol, ethyl phenol, nonanal, and decanal in the exhaled breath of the study groups. Normal represents the normal UGI tract group; benign represents the benign disease group; Barrett’s is the Barrett’s metaplasia group; EC is the esophageal cancer group, and GC is the gastric cancer group. 984 | www.annalsofsurgery.com
ß
2015 Wolters Kluwer Health, Inc. All rights reserved.
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
ß
2015 Wolters Kluwer Health, Inc. All rights reserved.
36 12 14 10 8 12 4 2 5 5 5 5 10 233 262 452 69
Median
IQR
23–61 9–18 10–22 6–25 3–10 7–21 2–7 2–7 3–10 2–7 3–9 2–10 5–17 157–333 146–573 275–687 49–100
40 12 13 9 7 8 4 3 6 4 3 3 8 245 305 478 67
Median 29–55 8–19 9–30 6–18 4–13 4–14 2–7 2–7 3–11 2–6 1–7 1–8 5–15 180–344 133–588 266–635 49–109
IQR
Esophageal Cancer
40 9 6 5 3 4 3 2 5 2 2 1 2 258 389 549 64
Median 24–43 6–14 5–11 1–11 2–5 2–7 2–4 1–4 1–6 1–4 1–5 0–3 0–5 171–317 157–674 346–705 41–79
IQR
Barrett’s Metaplasia
33 9 7 5 4 5 3 2 4 3 2 2 3 248 341 367 58
Median 23–43 6–17 5–10 2–8 2–6 2–8 1–6 0–3 2–7 1–5 1–3 1–4 2–5 187–291 233–596 190–570 41–89
IQR
Benign Controls
Median and interquartile range (IQR) are measured by parts-per-billion by volume (ppbv). The noncancer controls comprise the normal upper gastrointestinal tract, benign and Barrett’s metaplasia groups. yThe noncancer controls comprises normal upper gastrointestinal tract and benign groups. zThe carcinoma group comprises gastric adenocarcinoma and esophageal adenocarcinoma.
Butyric acid Pentanoic acid Hexanoic acid Phenol Methyl phenol Ethyl phenol Butanal Pentanal Hexanal Heptanal Octanal Nonanal Decanal Methanol Acetone Ammonia Isoprene
Gastric Cancer
31 10 7 4 3 4 2 2 3 2 1 1 2 248 338 330 64
Median 22–43 8–15 3–10 3–7 2–5 2–6 1–4 0–3 2–5 1–4 1–2 0–2 1–3 204–315 190–789 277–575 51–83
IQR
Normal Upper Gastrointestinal Tract
TABLE 3. Concentrations of the Statistically Significant and Abundant VOCs Across the Study Groups
0.007 0.02
