Accepted Manuscript

Detection of Lung Cancer and EGFR Mutation by Electronic Nose System

Dekel Shlomi, Manal Abud, Ori Liran, Jair Bar, Naomi Gai-Mor, Maya Ilouze, Amir
Onn, Alon Ben-Nun, Hossam Haick, Nir Peled

PII: S1556-0864(17)30573-7
DOI: 10.1016/j.jtho.2017.06.073
Reference: JTHO 635

To appear in: Journal of Thoracic Oncology

Received Date: 19 September 2016

Revised Date: 17 June 2017
Accepted Date: 26 June 2017

Please cite this article as: Shlomi D, Abud M, Liran O, Bar J, Gai-Mor N, Ilouze M, Onn A, Ben-Nun A,
Haick H, Peled N, Detection of Lung Cancer and EGFR Mutation by Electronic Nose System, Journal of
Thoracic Oncology (2017), doi: 10.1016/j.jtho.2017.06.073.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT
Detection of Lung Cancer and EGFR Mutation by Electronic Nose System 1
2
3
Running title - EGFR Detection by Breath Analysis 4
5
Dekel Shlomi1, Manal Abud2*, Ori Liran3*, Jair Bar4, Naomi Gai-Mor3, Maya Ilouze3, 6
Amir Onn4, Alon Ben-Nun6, Hossam Haick2 and Nir Peled3, 5

PT
7

1 Pulmonary Clinic, Dan Petah-Tiqwa District, Clalit Health Services, Israel. 8

2 Department of Chemical Engineering and Russell Berrie Nanotechnology Institute, 9

RI
Technion – Israel Institute of Technology, Haifa, Israel. 10
3 Thoracic Cancer Research and Detection Center, Sheba Medical Center, Tel 11

SC
Hashomer, Israel. 12
4 Thoracic Oncology Unit, Institute of Oncology, Sheba Medical Center, Tel Hashomer, 13
Israel. 14
5 Thoracic Cancer Unit, Davidoff Cancer Center, Rabin Medical Center, Petah-Tiqwa, 15

U
Israel 16
6 Department of General Thoracic Surgery, Chaim Sheba Medical Center, Tel 17
AN
Hashomer, Israel 18
19
*equal contribution 20
M

21

Address for Correspondence: 22

D

Prof. Nir Peled, MD PhD FCCP 23

TE

Thoracic Cancer Service, Davidoff Cancer Center, 24

Kaplan St, Petach Tiqwa, Israel 49100 25

EP

Tel/Fax: +972 (0)3 9377958; Cell: +972 (0)52 6669172 26

nirp@post.tau.ac.il 27
C

28
AC

29

30

31

32

33

34

1
 ACCEPTED MANUSCRIPT
1

2
Conflict of interest statement: 3
4
Prof. Peled reports grants and personal fees from The European FP7 LCAOS 5
program, grants from The Israel Cancer Association, during the conduct of the study; 6

PT
In addition, Prof. Peled has a patent Volatile Organic Compounds for Detecting Cell 7
Dysplasia and Genetic Alterations Associated with Lung Cancer issued, and a patent 8

RI
Breath Analysis of Pulmonary Nodules issued. 9
10

SC
Prof. Hossam Haick reports grants from FP7 Program, during the conduct of the 11
study; In addition, Prof. Haick has a patent Volatile Organic Compounds for 12
Detecting Cell Dysplasia and Genetic Alterations Associated with Lung Cancer, 13

U
Issued, and a patent Breath Analysis of Pulmonary Nodules pending. 14
AN
15
Dr. Dekel Shlomi and Dr. Maya Ilouze declare that their salaries and travel meetings 16
M

were supported by the FP-7 Grant (Tel - Aviv University, A Nanoscale Artificial 17
Nose to Easily Detect Volatile Biomarkers at Early Stages of Lung Cancer and 18
D

Related Genetic Mutations). 19

20
TE

All other authors declare that there is no actual or potential conflict of interest, 21
financial and personal or any relationships with other people or organizations that 22
EP

could inappropriately influence or bias this work. 23

24
C

Funding 25
AC

The European FP7 LCAOS program and the Israel Cancer Association 26

27

28
Keywords: lung cancer; pulmonary nodule; volatile organic compound; sensor; 29
electronic nose. 30
31

32

2
 ACCEPTED MANUSCRIPT
Abstract 1

Background: Early detection of lung cancer (LC) has been well established as a 2
significant key point in patient survival and prognosis. New highly sensitive 3
nanoarray sensors for exhaled Volatile Organic Compounds (VOCs) have been 4
developed and coupled with powerful statistical programs may be used when diseases 5
such as LC are suspected. Detection of genetic aberration mutation by nanoarray 6

PT
sensors is the next target. 7
Methods: Breath samples were taken from patients who were evaluated for suspicious 8

RI
pulmonary lesions. Patients were classified as those with benign nodules, LC patients 9
with or without the EGFR mutation, and according to their smoking status. 'Breath- 10

SC
prints' were recognized by nanomaterial based sensor array (NA-NOSE), and pattern 11
recognition methods were used. 12
Results: A total of 119 patients participated in this study, 30 patients with benign 13

U
nodules and 89 LC patients (n=16 with early and n=73 with advanced disease). . LC 14
AN
patients who harbor the EGFR mutation (n=19) could be discriminated from wild- 15
type (n=34) with an accuracy of 83%, a sensitivity of 79% and a specificity of 85%. 16
M

Discrimination of early LC from benign nodules had 87% accuracy and positive and 17
negative predictive values (PPV and NPV) of 87.7 and 87.5% respectively. Moderate 18
D

discrimination (accuracy of 76%) was found between LC of heavy smokers and never 19
or distant past light smokers. 20
TE

Conclusion: Breath analysis could discriminate LC patients who harbor the EGFR 21
mutation from wild type and those with benign pulmonary nodules from early LC 22
EP

patients. Positive 'breath-print' for the EGFR mutation may be used in treatment 23
decisions if tissue sampling does not provide adequate material for definitive mutation 24
C

analysis. 25
26
AC

27

3
 ACCEPTED MANUSCRIPT
Introduction 1
Lung cancer (LC) is the leading cause of cancer-related death for both men and 2
women world-wide1. Unfortunately, only 15% of new LC cases are being diagnosed 3
2
at a localized stage. The National Lung Screening Trial (NLST) demonstrated that 4
LC screening by low-dose computed tomography (LDCT) scans reduced LC mortality 5
rates by 20%, but among abnormal results, 96% were false positive.3 Lung nodules 6

PT
detected by computed tomography (CT) may lead to invasive procedures while 7
patients with pulmonary nodules may assume that they have cancer and suffer fear of 8
death and fear of chemotherapy toxicity.4,5 The need to distinguish between benign

RI
9
and malignant pulmonary nodules calls for the development of diagnostic tools such 10

SC
as biomarkers, ideally by non-invasive tools. Exhaled volatile organic compounds 11
(VOCs) are organic compounds that have a high vapor pressure at room-temperature. 12
They are generated either by various cellular biochemical processes in the body, or by 13

U
absorption from the environment through ingestion, inhalation or through skin 14
AN
contact.6 Some of the VOCs may be involved or produced in biochemical processes 15
7,8,9
such as oxidative stress, or in cytochrome p450, carbohydrate or lipid metabolism. 16
M

Organic compounds with low solubility in the blood generally diffuse via gas 17
exchange occurring in the lung alveoli while the more soluble ones have a higher 18
chance of diffusing throughout the bronchial tree.10,11 Breath samples might contain
D

19
up to 3000 different VOCs, mostly at low concentrations (ppt to ppb).12 Using a 20
TE

colorimetric sensor array, Mazzone and colleagues demonstrated a sensitivity of 21

73.3% and specificity of 72.4% in distinguishing NSCLC patients from those with 22
EP

other pulmonary diseases and control participants.13 Recently, using mass 23

spectrometry, Schumer and colleagues found that 4 volatile carbonyl compounds 24
C

could be sensitive markers for lung cancer while specificity was proportional to the 25
number of the elevated markers.14 Nanoarray sensors allow the identification of a 26
AC

large number of VOCs through precise monitoring of changes in electrical resistance. 27

Using pattern recognition methods and classification techniques, the combined 28
response from all sensors establishes a volatolomic signature. Diseases such as cancer 29
lead to a shift in the composition of these organic compounds, presenting a unique 30
'breath-print' in exhaled air.10,11,15 31
In recent years a growing body of evidence has shown that cancer patients can be 32
distinguished from other patients or healthy individuals by a unique VOC pattern 33
which is identified by a combination of very sensitive chemical nanoarray sensors and 34
4
 ACCEPTED MANUSCRIPT
powerful computerized analysis learning models.13,16–25 In a study of 14 patients with 1
bronchogenic carcinoma and 45 controls without cancer, nanoarray sensors had a 2
sensitivity of 71.4% and a specificity of 91.9% for detecting LC.20 In another study 3
specific 'breath-prints' of Non-Small Cell Lung Carcinoma (NSCLC) patients could 4
be distinguished from Chronic Obstructive Pulmonary Disease (COPD) patients and 5
healthy controls.19 Also, nanoarray sensors successfully distinguished between 6

PT
patients with head and neck cancer, LC and healthy controls.23 A recent cross- 7
sectional study which included the usage of a chemiresistor-based alkane sensor, 8

RI
demonstrated that the peak output and the time to peak values for LC patients were 9
26
statistically different than COPD and healthy control patients. Lung cancer was 10

SC
diagnosed with sensitivity of 83.3% and a specificity of 88%. In another interesting 11
study performed by Peralbo-Molina et al., 5 different metabolites (monopalmitin, 12
benzyl alcohol, monostearin, 2,4-diphenyl-4-methyl-2-E-pentene and p-cresol) were 13

U
used to form three-panels.27 LC patients were differentiated from healthy controls 14
AN
and patients with risk factors with sensitivity above 77.9%, specificity above 67.5% 15
and an AUC above 77.5%. 16
M

The next challenge is to discriminate histologically between different types of 17

cancers in the same organ, or between different stages of cancer. In our previous 18
D

study we used gold nanoparticle (GNP) sensors to distinguish VOCs from cell lines of 19
healthy individuals and different types of LCs.24 We could discriminate significantly 20
TE

between lung carcinoma and healthy cells, between NSCLC and Small Cell Lung 21
Carcinoma (SCLC) cells, and between adenocarcinoma and squamous cell carcinoma 22
EP

with 90-96% accuracy. In another study by our group we evaluated patients with 23
pulmonary nodules from the University of Colorado Cancer Center (Aurora, CO) and 24
Denver Veterans Affairs Medical Center (Denver, CO).25 The nanoarray sensors
C

25
distinguished successfully between benign versus malignant nodules, between adeno- 26
AC

and squamous-cell carcinomas, and between early stage and advanced disease with an 27
accuracy of 88%. 28
A more interesting aspiration in cancer detection is to identify genetic aberrations 29
that could alter treatments. In our previous pilot study of cell-lines we identified the 30
volatile fingerprints that are associated with EGFR and KRAS mutations as well as 31
EML4-ALK rearrangement genes.28 Unique fingerprints of KRAS and TP53 were 32
29
also found in bronchial epithelial cell-lines by Davis and colleagues. Furthermore, in 33

5
 ACCEPTED MANUSCRIPT
a clinical study using ion mobility spectrometry, Handa and colleagues were able to 1
discriminate LC patients with or without EGFR mutation.30 2
In this study we aimed to evaluate the potential of exhaled breath analysis in the 3
diagnosis of the EGFR mutation in LC patients. Based on our previous studies with 4
cancer patients we selected a unique set of 40 sensors. We explored the system in 5
discriminating LC patients who harbor the EGFR mutation from wild-type as well as 6

PT
reexamining the system's ability to identify early LC from benign pulmonary nodules. 7
Furthermore, we used our system in order to identify different 'breath-prints' between 8

RI
LC patients with a history of heavy recent smoking and never or distant past light 9
smokers. 10

SC
11
Methods 12
The study was conducted at the Sheba Medical Center (SMC), Israel, a tertiary 13

U
medical center. Patients with suspicious lung lesions referred for evaluation at our 14
AN
Institute of Pulmonology were recruited for the study. Patients were classified with 15
benign pulmonary nodule according to histology or if nodules were radiographically 16
M

stable over at least two years of follow-up. LC patients were classified as early (stage 17
I and II) and Advanced (stage III and IV) Breath samples from LC patients were 18
D

collected prior to any treatment. Ethical approval was obtained from the ethics 19
committee of the SMC for supervision of human experiments (8663-11-SMC), and 20
TE

the clinical trial was registered at Clinicaltrials.gov (NCT01386203). All patients 21

were provided with a written and verbal explanation about the study prior to their 22
EP

enrollment, and signed a written consent of participation. 23

Breath collection: Breath samples for patients scheduled for biopsy were collected 24
C

before the procedures. In order to minimize the effect of confounding factors on the 25
analysis the volunteers were asked not to ingest coffee and food for at least one hour, 26
AC

alcohol for at least 12 hours and to avoid smoking for at least half an hour prior to the 27
breath collection. Lung washout to reduce exogenous VOCs was obtained by 3 min of 28
inhalation through a mouthpiece with a filter cartridge on the inspiratory port 29
mouthpiece (Eco Medics, Duerten, Switzerland). The patients were asked to inhale 30
to the level of total lung capacity and then to slowly exhale through the mouthpiece 31
against 10–15 cm H2O pressure, to ensure closure of the vellum to eliminate 32
contamination through nasal entrainment. In order to focus on the VOCs originating 33
from the blood, we collected alveolar air. The respiratory dead space air was collected 34

6
 ACCEPTED MANUSCRIPT
by a designated bag of 440 ml, which was then removed, while the remaining alveolar 1
air from the lungs was collected in 750 ml GaSampler collection bags (QuinTron, 2
Milwaukee, WI, USA). Two bags were collected per patient. The air samples were 3
trapped and pre-concentrated by pumping the contents of the bag through a sorbent 4
tube for 7 min (flow rate: 100 ml/min) in two-bed ORBOTM 420 reusable Tenax® 5
TA sorbent tubes for gas and vapor sampling (specially treated; 35/60 mesh; 100/50 6

PT
mg; Sigma-Aldrich, St. Louis, Missouri, USA). The Tenax tubes were stored and 7
transported at 4°C to the laboratory for Nanomaterial-Based Devices, Technion - IIT, 8

RI
Haifa, Israel. Since we used different Tenax tubes with different absorption capacity 9
this group was analyzed separately from patients from our previous studies. 10

SC
Nanoarray sensors: The nanomaterial-based sensor array used in this study 11
contained 40 cross-reactive, chemically diverse chemiresistors based on two types of 12
nanomaterials: (i) organically-stabilized spherical gold nanoparticles (GNPs, core 13

U
diameter: 3-4 nm), and (ii) single-walled carbon nanotubes (SWCNTs) capped with 14
AN
polycyclic aromatic hydrocarbon (PAH) or Hexabenzocoronene (HBC). Previous 15
17,18,31,32 32
articles have described the synthesis of the GNPs and SWCNTs. The GNP 16
M

and SWCNT/PAH or SWCNT/HBC sensors used in this study responded rapidly and 17
reversibly when exposed to typical breath VOCs. Based on our previous researches, 18
D

we chose 40 sensors (26 GNP, 8 SWCNT/PAH and 6 SWCNT/HBC) that were able 19
to distinguish between cancer patients and non-cancer subjects. 20
TE

The fabricated sensors were mounted on a custom polytetrafluoroethylene (PTFE) 21

circuit board inside a stainless-steel test chamber with a volume of 330 ml which 22
EP

delivers pulses of the breath samples from the thermal desorption device to the 23
chemical nanoarray. The samples underwent thermal desorption (at 250 °C) in an 24
C

auto-sampler thermal desorption system (TD20; Shimadzu Corporation, Japan), and 25

the desorbed samples were temporarily stored in a stainless steel column (150 °C). In 26
AC

parallel, the chamber containing the sensors was kept under vacuum conditions (~30 27
mtorr), until the sample was directed into the chamber, and the remaining volume was 28
filled with N2 (99.999%) until reaching atmospheric pressure. A Keithley data logger 29
device (model 2701 DMM) was used to obtain sequential resistance readings from the 30
sensor array, during 5 minutes in a vacuum prior to exposure (baseline), followed by 5 31
minutes of breath sample filling the chamber, followed by another 5 minutes of sensor 32
recovery, starting with chamber vacuum. The whole system was controlled by a 33
custom-made LabView program, and the chamber was evacuated between exposures. 34
7
 ACCEPTED MANUSCRIPT
An Agilent multifunction switch 34980 was used to measure the resistance of all 1
sensors as a function of time. In order to supervise the sensor’s functionality during 2
the experiment, and also to overcome its response drift, a fixed calibration gas 3
mixture, containing 23.8 ppm isopropyl alcohol, 6.3 ppm trimethylbenzene and 1.2 4
ppm 2-ethylhexanol was exposed to the sensors on a daily basis, and the raw signals 5
of response of the breath samples was normalized by its parallel response to the 6

PT
calibration gas measured the same day. The samples were run blindly and then 7
analyzed at the Technion-IIT. 8

RI
9
Statistical analysis 10

SC
From each sensor response to a breath sample, four sensing features were read 11
out; the normalized change of the sensor’s resistance at the peak, middle and end of 12
the exposure, and the area under curve (resistance over time). Discriminate function 13

U
analysis (DFA) was used to classify the data17. DFA performs a multivariate test of 14
AN
differences between groups, and is used to determine the minimum number of 15
dimensions needed to describe these differences. A distinction is sometimes made 16
M

between descriptive discriminant analysis and predictive discriminant analysis. DFA 17

derives canonical variables (CV; linear combinations of the interval variables) that 18
D

summarize between-class sensors in much the same way that principal components 19
summarize total variation. Given two or more groups of observations with 20
TE

measurements on several interval variables, DFA derives a linear combination of the 21

variables that has the highest possible multiple correlation with the groups. This 22
EP

maximal multiple correlation is called the first canonical correlation (CV1). The 23
coefficients of the linear combination are the canonical coefficients or canonical 24
C

weights. The variable defined by the linear combination is the CV1 or canonical 25
component. The second canonical correlation is obtained by finding the linear 26
AC

combination uncorrelated with the first canonical variable that has the highest 27
possible multiple correlation with the groups. The process of extracting canonical 28
variables can be repeated until the number of canonical variables equals the number 29
of original variables or the number of classes minus one, whichever is smaller. 30
In order to prevent over-fitting of the data, we used the minimal number of sensing 31
features to build a discriminative model, stressing a ratio lower than 1:6 of features to 32
samples. The leave-one-out cross-validation method was then employed to calculate 33
the accuracy of prediction. The Wilcoxon test was used to identify the statistically 34
8
 ACCEPTED MANUSCRIPT
significant prediction models. Statistical analysis was carried out using SAS JMP, 1
Version 10.0 (SAS Institute Inc., Cary, NC, USA, 1989-2005). We used Kruskal- 2
Wallis one-way analysis of variance and parametric chi-square tests using SPSS 3
Statistics for analysis of the clinical characteristics. 4
5
Results 6

PT
Between January 2012 and May 2014, a total of 119 patients were enrolled in the 7
study. Benign nodules were identified in 30 patients while 89 patients were diagnosed 8

RI
with LC of which 16 patients were diagnosed with early LC and 73 patients with 9
advanced LC. Clinical characteristics are summarized in Table 1. As expected, 10

SC
patients with benign pulmonary nodules were younger, smoked less and had smaller 11
nodules compared to LC patients. Of the 30 patients with benign (or likely benign) 12
nodules, 3 were granulomatous, 1 was inflammatory, 3 were non-specific on biopsy 13

U
and 23 did not have biopsy. Of the 89 patients with confirmed lung cancer, 73 had 14
AN
adenocarcinoma, 9 had SCC and 7 had SCLC (all extensive disease). 15
Table 2 summarizes the performance of the exhaled breath analysis with nanoarray 16
M

technology for different groups. Good performance was obtained for discriminating 17
early LC from benign nodules, with an accuracy of 87% by two sensor sets (Figure 1), 18
D

where each set comprise different combination of sensors and/or sensing features. The 19
use of 2 different sets of sensors provides additional validation for the results. For 20
TE

this comparison we found a relatively low sensitivity (75%) and a high specificity 21
(93%) with positive and negative predicted values (PPV and NPV) of 87.7% and 22
EP

87.5% respectively. 23
A subgroup of the LC patients included 19 patients diagnosed with the EGFR 24
C

mutation, and 34 LC patients who did not harbor any EGFR mutation. The clinical 25
characteristics of this subgroup are shown in Table 3. As expected, smoking was 26
AC

significantly lower in the positive EGFR mutation group. Using five different sensors, 27
the nanoarray sensors discriminated between LC patients with EGFR mutation 28
positive tumours and those with EGFR mutation negative tumours. Patients harboring 29
EGFR mutation were discriminated from wild-type by an accuracy of 83%, a 30
sensitivity of 79%, a specificity of 85% and PPV and NPV of 75% and 88% 31
respectively (Table 2). 32
We were interested to see whether LC of patients with a history of heavy smoking 33
could be differentiated from LC of never or distant past light smokers by volatolomic 34
9
 ACCEPTED MANUSCRIPT
signature. We used the smoking criteria of the NLST study in order to classify our 1
patients3. Of the 89 LC patients 50 patients met the criteria of current smokers or 2
those who had quit smoking during the past 15 years, and with at least 30 PY. We 3
found 26 patients who had never smoked or quit smoking more than 15 years before 4
the study with less than 30 PY. We analyzed the differences between these two 5
extreme groups and excluded 13 patients from smoking analysis: active smokers with 6

PT
less than 30 PY, those who had quit smoking during the past 15 years with less than 7
30 PY and those who had quit smoking more than 15 years ago with at least 30 PY. 8

RI
Breath analysis could discriminate between the heavier smokers and the lighter 9
smokers with accuracy of 76.3%, and a sensitivity and specificity of 78% and 73% 10

SC
respectively (Table 2). 11
12
13
Discussion
U 14
AN
By using highly sensitive nanoarray sensors and DFA as a powerful statistical 15
tool, we established the concept that different diseases have different 'breath-prints'. 16
M

The primary aim of this study is to discriminate LC patients who harbor EGFR 17
mutation from wild-type by 'breath-print' analysis using nanoarray sensors. Our 18
D

system was able to identify a unique breath-print in lung cancer patients with the 19
EGFR mutation with an accuracy of 83%. We used a unique set of 40 sensors which 20
TE

led to high accuracy in detecting the EGFR mutation in a clinical setting. 21

We also compare LC with benign nodules with similar results to those from 22
EP

previous studies.13,14,19,20,26,27 In our previous pilot study of patients from Colorado, 23

we showed that benign pulmonary nodules could be well discriminated from LC by a 24
unique 'breath-print'.25 The nanoarray sensors could also discriminate between adeno-
C

25
and squamous-cell carcinomas and between early stage and advanced disease. The 26
AC

current study demonstrated that early LC patients could be discriminated from 27

patients with benign pulmonary nodules with a sensitivity, specificity and accuracy of 28
75%, 93% and 87% respectively. Both PPV and NPV were high (approximately 88%) 29
indicating the efficiency and suitability of the test. However, because of the small 30
number of early LC patients (n=16) these results should be interpreted carefully. 31
To the best of our knowledge, this is the first clinical study of VOCs that 32
discriminates a mutation of lung cancer in humans by 'breath-print'. Previously, we 33
were able to detect mutations in in-vitro settings.28 Further studies are needed to 34
10
 ACCEPTED MANUSCRIPT
support this observation. Since we had a small number of patients with squamous-cell 1
carcinomas and SCLC we could not analyze histological differences; however this 2
technology has shown such discrimination in-vitro.24 We suggest that in case of a 3
negative EGFR mutation by histological testing, and when tissue sampling does not 4
provide adequate material for sequencing, a positive 'breath-print' for the EGFR 5
mutation could be considered a true-positive EGFR result for adenocarcinoma. 6

PT
The third major (and second novel) finding in this study is the difference in 7
'breath-prints' between LC of non-smokers or distant past light smokers and that of 8
heavy smokers (active or recent quitters). In previous studies16,33,34 we have found

RI
9
26
that the sensor array output is not affected by current smoking. Tan et al have made 10

SC
a similar finding. To the best of our knowledge this is the first publication of breath- 11
print analysis that distinguishes between lung cancer in heavy smokers and in light-to- 12
minimal smokers, suggesting a possible metabolic difference between lung cancers in 13
these two groups.
U 14
AN
There are several limitations to this study. Most important is the small study 15
population and in particular the small number of early lung cancer and EGFR 16
M

mutation. Second, the classification to benign nodules in most cases was determined 17
by two years of radiographically stability instead of by tissue diagnosis. Another 18
D

limitation is that the analysis regarding the smoking history is subject, as expected, to 19
recall bias. We expect that further larger studies with larger populations would 20
TE

provide more accurate results. 21

These encouraging findings support the field of VOCs as a marker for lung 22
EP

cancer using 'breath-print' analysis. There is potential in the future to identify 'breath- 23
prints' by sensors through a portable, inexpensive device that is easy to use and that 24
C

provides rapid results. This may become the method of choice in the field of VOC 25
analysis. 26
AC

27
28
Acknowledgments 29
The authors thank the European FP7 LCAOS program no. 258868 and the Israel 30

Cancer Association for their support in this study. 31

32
33

11
 ACCEPTED MANUSCRIPT
References 1

1. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2015. CA Cancer J Clin 2
2015;65(1):5–29. 3
2. Howlader N, AM N, Krapcho M, et al. SEER Cancer Statistics Review, 1975- 4
2010,. http://seer.cancer.gov/csr/1975_2010/ 2013; 5
3. Reduced Lung-Cancer Mortality with Low-Dose Computed Tomographic 6
Screening. N Engl J Med 2011;365(5):395–409. 7

PT
4. Wiener RS, Gould MK, Woloshin S, Schwartz LM, Clark JA. What do you 8
mean, a spot?: A qualitative analysis of patients’ reactions to discussions with 9
their physicians about pulmonary nodules. Chest. 2013;143(3):672–7. 10

RI
5. Shlomi D, Ben-Avi R, Balmor GR, Onn A, Peled N. Screening for lung cancer: 11
Time for large-scale screening by chest computed tomography. Eur Respir J 12
2014;44(1):217–38. 13

SC
6. Pauling L, Robinson AB, Teranishi R, Cary P. Quantitative analysis of urine 14
vapor and breath by gas-liquid partition chromatography. Proc Natl Acad Sci U 15
S A 1971;68(10):2374–6. 16

U
7. Konvalina G, Haick H. Sensors for breath testing: From nanomaterials to 17
comprehensive disease detection. Acc Chem Res 2014;47(1):66–76. 18
AN
8. Buszewski B, Kesy M, Ligor T, Amann A. Human exhaled air analytics: 19
Biomarkers of diseases. Biomed. Chromatogr. 2007;21(6):553–66. 20
9. Hakim M, Broza YY, Barash O, et al. Volatile organic compounds of lung 21
M

cancer and possible biochemical pathways. Chem Rev [Internet] 2012 [cited 22
2015 Aug 3];112(11):5949–66. Available from: 23
http://www.ncbi.nlm.nih.gov/pubmed/22991938 24
D

10. Amann A, Mochalski P, Ruzsanyi V, Broza YY, Haick H. Assessment of the 25

TE

exhalation kinetics of volatile cancer biomarkers based on their 26

physicochemical properties. J Breath Res [Internet] 2014;8(1):16003. Available 27
from: http://www.ncbi.nlm.nih.gov/pubmed/24566039 28
11. Haick H, Broza YY, Mochalski P, Ruzsanyi V, Amann A. Assessment, origin, 29
EP

and implementation of breath volatile cancer markers. Chem Soc Rev [Internet] 30
2014;43(5):1423–49. Available from: 31
http://www.ncbi.nlm.nih.gov/pubmed/24305596 32
C

12. Phillips M, Herrera J, Krishnan S, Zain M, Greenberg J, Cataneo RN. Variation 33

in volatile organic compounds in the breath of normal humans. J Chromatogr B 34
AC

Biomed Sci Appl 1999;729(1–2):75–88. 35

13. Mazzone PJ, Hammel J, Dweik R et al. Diagnosis of lung cancer by the 36
analysis of exhaled breath with a colorimetric sensor array. Thorax 37
2007;62:565–568. 38
14. Schumer EM, Trivedi JR, Berkel V Van. High sensitivity for lung cancer 39
detection using analysis of exhaled carbonyl compounds. J Thorac Cardiovasc 40
Surg [Internet] 2016;150(6):1517–24. Available from: 41
http://dx.doi.org/10.1016/j.jtcvs.2015.08.092 42
15. Broza YY, Haick H. Nanomaterial-based sensors for detection of disease by 43
volatile organic compounds. Nanomedicine (Lond) [Internet] 2013;8(5):785– 44

12
 ACCEPTED MANUSCRIPT
806. Available from: http://www.ncbi.nlm.nih.gov/pubmed/23656265 1
16. Peng G, Hakim M, Broza YY et al. Detection of lung, breast, colorectal, and 2
prostate cancers from exhaled breath using a single array of nanosensors. Br J 3
Cancer 2010;103:542–551. 4
17. Peng G, Tisch U, Adams O et al. Diagnosing lung cancer in exhaled breath 5
using gold nanoparticles. Nat Nanotechnol 2009;4:669–673. 6
18. Peng G, Trock E HH. Detecting simulated patterns of lung cancer biomarkers 7
by random network of single-walled carbon nanotubes coated with 8

PT
nonpolymeric organic materials. Nano Lett 2008;8:3631–3635. 9
19. Dragonieri S, Annema JT, Schot R et al. An electronic nose in the 10
discrimination of patients with non-small cell lung cancer and COPD. Lung 11

RI
Cancer 2009;64:166–170. 12
20. Machado RF, Laskowski D, Deffenderfer O et al. Detection of lung cancer by 13
sensor array analyses of exhaled breath. Am J Respir Crit Care Med 14

SC
2005;171:1286–1291. 15
21. Tisch U HH. Arrays of chemisensitive monolayer-capped metallic 16
nanoparticles for diagnostic breath testing. Rev Chem Eng 2011;26:171–179. 17
22.
U
Tisch U HH. Nanomaterials for cross-reactive sensor arrays. MRS Bull 18
AN
2010;35:797–803. 19
23. Hakim M, Billan S, Tisch U et al. Diagnosis of head-and-neck cancer from 20
exhaled breath. Br J Cancer 2011;104:1649–1655. 21
M

24. Barash O, Peled N, Tisch U, Bunn P a., Hirsch FR, Haick H. Classification of 22
lung cancer histology by gold nanoparticle sensors. Nanomedicine 23
Nanotechnology, Biol Med 2012;8(5):580–9. 24
D

25. Peled N, Hakim M, Bunn P a., et al. Non-invasive Breath Analysis of 25

Pulmonary Nodules. J Thorac Oncol 2012;7(10):1528–33. 26
TE

26. Tan J-L, Yong Z-X, Liam C-K. Using a chemiresistor-based alkane sensor to 27
distinguish exhaled breaths of lung cancer patients from subjects with no lung 28
cancer. J Thorac Dis [Internet] 2016;8(10):2772–83. Available from: 29
EP

http://jtd.amegroups.com/article/view/9989/8875 30
27. Peralbo-Molina A, Calderón-Santiago M, Priego-Capote F, Jurado-Gámez B, 31
Luque de Castro MD. Identification of metabolomics panels for potential lung 32
C

cancer screening by analysis of exhaled breath condensate. J Breath Res 33

[Internet] 2016;10(2):26002. Available from: http://stacks.iop.org/1752- 34
AC

7163/10/i=2/a=026002?key=crossref.80f887b036b844f96b371281d6ba0d2a 35
28. Peled N, Barash O, Tisch U, Ionescu R, Broza YY. Volatile fingerprints of 36
cancer specific genetic mutations. Nanomedicine Nanotechnology, Biol Med 37
[Internet] 2013;9(6):758–66. Available from: 38
http://dx.doi.org/10.1016/j.nano.2013.01.008 39
29. Davies MPA, Barash O, Jeries R, et al. Unique volatolomic signatures of TP53 40
and KRAS in lung cells. Br J Cancer [Internet] 2014;111(6):1213–21. 41
Available from: http://dx.doi.org/10.1038/bjc.2014.411 42
30. Handa H, Usuba A, Maddula S. Exhaled Breath Analysis for Lung Cancer 43
Detection Using Ion Mobility Spectrometry. 2014;1–13. 44

13
 ACCEPTED MANUSCRIPT
31. Dovgolevsky E, Haick H. Direct observation of the transition point between 1
Quasi-spherical and cubic nanoparticles in a two-step seed-mediated growth 2
method. Small 2008;4(11):2059–66. 3
32. Dovgolevsky E, Tisch U, Haick H. Chemically sensitive resistors based on 4
monolayer-capped cubic nanoparticles: towards configurable nanoporous 5
sensors. Small [Internet] 2009 [cited 2015 Aug 3];5(10):1158–61. Available 6
from: http://www.ncbi.nlm.nih.gov/pubmed/19274647 7
33. Amal H, Leja M, Funka K, et al. Breath testing as potential colorectal cancer 8

PT
screening tool. Int J cancer [Internet] 2016 [cited 2016 Dec 2];138(1):229–36. 9
Available from: http://www.ncbi.nlm.nih.gov/pubmed/26212114 10
34. Amal H, Leja M, Funka K, et al. Detection of precancerous gastric lesions and 11

RI
gastric cancer through exhaled breath. Gut [Internet] 2016 [cited 2016 Dec 12
2];65(3):400–7. Available from: 13
http://www.ncbi.nlm.nih.gov/pubmed/25869737 14

SC
15
16

U
17
AN
Figure Legends 18
19
Figure 1: Discriminant factor analysis (DFA) plots calculated from the responses of 20
M

the nanoarray sensors for benign pulmonary nodules and early LC. Each point 21
represents one patient. The positions of the mean values are marked with □, the boxes 22
D

correspond to the first and third quartile, and the error bars correspond to the standard 23
deviations. CV1: first canonical variable (see text). 24
TE

25

Figure 2: Discriminant factor analysis (DFA) plot calculated from the responses of 26
EP

the nanoarray sensors for LC patients with and without the EGFR mutation. Each 27
point represents one patient. The positions of the mean values are marked with □, the 28
boxes correspond to the first and third quartile, and the error bars correspond to the 29
C

standard deviations. CV1: first canonical variable (see text). 30

AC

31

14
 ACCEPTED MANUSCRIPT
Table 1: Clinical Characteristics of the Tested Patients

Benign Nodule Early Lung Advanced Total p-value

Cancer Lung Cancer Patients

No. (%) 30 16 73 119

(25) (13) (62)
Age (years) 56.93±10.68 68.18±10.3 64.25±12 63±12 0.0025
Avg ± SD

PT
Gender (%)
Male 20 (67) 6 (38) 43 (59) 69 (58) 0.156
Female 10 (33) 10 (62) 30 (41) 50 (42)

RI
Smoking: (PY) 27.83±29.86 29.89±34.9 47.57±39.87 40.14±38.05 0.044
Avg± SD
Active (%) 14 (47) 2 (13) 14 (19) 30 (25)
Former 10 (33) 9 (56) 47 (64) 66 (55)

SC
Never 6 (20) 5 (31) 12 (17) 23 (20)
BMI (kg/m2) avg 27.5±6.39 26.4±3.84 25.18±4.47 25.82±4.7 0.08
± SD

U
Medical History:
COPD 2 0 10 12
AN
Emphysema 2 0 6 8
IHD 1 2 9 12
HTN 10 7 26 43
DM 3 4 18 25
M

Nodule Size (mm) 0.0001

Range 3-37 8-60 NA-advance
D

Avg ± SD 13.47±12.88 31.6±15.02 disease

Stages:
TE

Stage I 9
Stage II 7
Stage III 20
Stage IV 53
EP

PY: pack-years; BMI: body mass index; COPD: chronic obstructive pulmonary disease; IHD:
ischemic heart disease; HTN: hypertension; DM: diabetes mellitus.
C
AC
 ACCEPTED MANUSCRIPT

Table 2: Breath print analysis to detect lung cancer, EGFR mutation and level of smoking

PT
RI
SC
Groups No. TP TN FP FN Sensitivity Specificity Accuracy PPV NPV
(%) (%) (%) (%) (%)

U
Early Lung Cancer vs. Benign Nodules 16 vs. 30 12 28 2 4 75 93.3 86.96 87.71 87.50
(2 sensors sets)

AN
EGFR mutation vs. wild type 19 vs. 34 15 29 5 4 78.95 85.29 83.02 75 87.88

Lung Cancer in Heavy smokers vs. Never or Light 50 vs. 26 39 19 7 11 78 73.08 76.32 84.78 63.33

M
distant past smokers

D
TE
EGFR: epidermal growth factor receptor; TP: true positive; TN: true negative; FP: false positive; FN: false negative; PPV: positive predictive value; NPV:
negative predictive value.
C EP
AC
 ACCEPTED MANUSCRIPT
Table 3: Clinical Characteristics of Patients with EGFR Test

EGFR EGFR Total p-value

Positive Negative Patients

No. 19 34 53
(%) (36) (64)

Age (years)

PT
Avg ± SD 65.73±13 65.77±11.3 65.76±11.8 0.495

Gender

RI
Male (%) 10 (53) 22 (65) 32 (60)
Female 9 (44) 12 (35) 21 (40) 0.388

SC
Smoking:
(PY)
Avg± SD 24.97±30.97 53.51±44.12 43.63±42.04 0.004
Active (%) 2 (10) 7 (20) 9 (17)

U
Former 11 (58) 23 (68) 34 (64)
Never 6 (32) 4 (12) 10 (19)
AN
BMI (kg/m2)
avg ± SD 26.78±3.81 23.94±4.53 24.93±4.47 0.014
M

Medical
History:
COPD 3 3 6 0.493
D

Emphysema 0 3 3
IHD 3 3 6
HTN 8 13 21
TE

DM 3 9 12

Stage:
Stage I 2 2 4
EP

Stage II 1 4 5 0.446
Stage III 1 6 7
Stage IV 15 22 37
C
AC

EGFR: epidermal growth factor receptor; PY: pack-years; BMI: body mass index; COPD:
chronic obstructive pulmonary disease; IHD: ischemic heart disease; HTN: hypertension;
DM: diabetes mellitus.
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC