Classification of breast cancer precursors through exhaled breath

Breast Cancer Research and Treatment ISSN 0167-6806 Volume 126 Number 3 Breast Cancer Res Treat (2011) 126:791-796 DOI 10.1007/s10549-010-1317x

1 23

Your article is protected by copyright and all rights are held exclusively by Springer Science+Business Media, LLC.. This e-offprint is for personal use only and shall not be selfarchived in electronic repositories. If you wish to self-archive your work, please use the accepted authors version for posting to your own website or your institutions repository. You may further deposit the accepted authors version on a funders repository at a funders request, provided it is not made publicly available until 12 months after publication.

1 23

Author's personal copy

Breast Cancer Res Treat (2011) 126:791796 DOI 10.1007/s10549-010-1317-x

BRIEF REPORT

Classication of breast cancer precursors through exhaled breath

Gregory Shuster Zahava Gallimidi Asnat Heyman Reiss Ekaterina Dovgolevsky Salem Billan Roxolyana Abdah-Bortnyak Abraham Kuten Ahuva Engel Ala Shiban Ulrike Tisch Hossam Haick

Received: 18 November 2010 / Accepted: 16 December 2010 / Published online: 29 December 2010 Springer Science+Business Media, LLC. 2010

Abstract Certain benign breast diseases are considered to be precursors of invasive breast cancer. Currently available techniques for diagnosing benign breast conditions lack accuracy. The purpose of this study was to deliver a proofof-concept for a novel method that is based on breath testing to identify breast cancer precursors. Within this context, the authors explored the possibility of using exhaled alveolar breath to identify and distinguish between benign breast conditions, malignant lesions, and healthy states, using a small-scale, case-controlled, cross-sectional clinical trial. Breath samples were collected from 36 volunteers and were analyzed using a tailor-made nanoscale articial NOSE (NA-NOSE). The NA-NOSE signals were analyzed using two independent methods: (i) principal component analysis, ANOVA and Students t-test and (ii) support vector machine analysis to detect statistically
Electronic supplementary material The online version of this article (doi:10.1007/s10549-010-1317-x) contains supplementary material, which is available to authorized users.
G. Shuster E. Dovgolevsky A. Shiban U. Tisch H. Haick (&) The Department of Chemical Engineering and Russell Berrie Nanotechnology Institute, TechnionIsrael Institute of Technology, 32000 Haifa, Israel e-mail: hhossam@technion.ac.il Z. Gallimidi A. H. Reiss A. Engel Breast Imaging Unit, Radiology Department, The Rambam Health Care Campus, 31096 Haifa, Israel Z. Gallimidi A. Kuten A. Engel Bruce Rappaport Faculty of Medicine, TechnionIsrael Institute of Technology, 31096 Haifa, Israel S. Billan R. Abdah-Bortnyak A. Kuten Oncology Division, Rambam Health Care Campus, 31096 Haifa, Israel

signicant differences between the sub-populations. The NA-NOSE could distinguish between all studied test populations. Breath testing with a NA-NOSE holds future potential as a cost-effective, fast, and reliable diagnostic test for breast cancer risk factors and precursors, with possible future potential as screening method. Keywords Breast cancer Benign Classication Volatile biomarkers Breath Sensor NA-NOSE Abbreviations IDC Inltrating ductal carcinoma DCIS Ductal carcinoma in situ BC Breast cancer NA-NOSE Nanoscale articial NOSE GCMS Gas-chromatography/mass-spectrometry PCA Principle component analysis SVM Support vector machine

Introduction Breast diseases can be classied either with disorders of the integument or with disorders of the reproductive system. The majority of breast diseases are benign. They nevertheless cause much distress until the diagnosis is conrmed through biopsy, and they frequently recur, necessitating further medical examinations. Some benign tumors, such as premalignant hyperplasia, are considered to be potential precursors of invasive breast cancer (IBC), which is the most common malignancy in women in the Western world [13]. Premalignant hyperplastic breast lesions may progress to carcinoma in situ and from there to IBC, according

123

Author's personal copy

792 Breast Cancer Res Treat (2011) 126:791796

to the multistep model for breast cancer (BC) progression [3]. Breast diseases are diagnosed through imaging techniques, such as mammography, ultrasound (US), and magnetic resonance imaging (MRI), in combination with biopsies. However, the accuracy of these methods for diagnosing benign breast conditions, which peak in women under the age of 50, is not optimal [1]. For mammography, the image quality depends on the breasts density and, therefore, it is primarily suitable for (post-) menopausal women, but not for young women at risk of developing heredity BC or for women with dense breast tissue. The accuracy of mammography is not optimal for diagnosing benign breast conditions, because they occur mainly in younger women. Also, between 11% and 20% of the lesions diagnosed as DCIS via mammography are observed to harbor invasive cancer on complete excision. US cannot identify breast tumors unambiguously, but is often employed as complementary diagnostic method after an inconclusive mammography or when the breast is dense. MRI has high sensitivity to breast diseases, but has significantly lower specicity than mammography, often leading to false positive IBC diagnosis that subject healthy women to unnecessary biopsies and other invasive follow-up tests. A complementary approach that overcomes many constraints of the conventional techniques relies on patterns of volatile biomarkers in the exhaled breath [4, 5]. This can be understood in terms of cell biology: BC is promoted by the progressive accumulation of genetic and epigenetic changes in the breast cells [6], leading to cellular oxidative stress [7, 8], and, thus, to the emission of BC-specic volatile biomarkers into the blood [912]. BC-promoting DNA methylation of certain genes has been observed even in benign lesions, indicating that epigenetic changes (very probably accompanied by measurable emission of volatile biomarkers) occur very early in the tumor genesis [3, 6]. Part of the BC volatile biomarkers in the blood are transmitted to the alveolar exhaled breath through exchange via the lung. It has been shown by gas chromatography combined with mass spectrometry (GCMS) that a combination of specic volatile biomarkers in breath samples can be used to identify BC [13, 14]. Amann and co-workers [15] have improved the accuracy of identication for monomethylated alkanes contained in exhaled breath that have been associated with BC and have investigated the biochemical background of cancer specic volatile biomarkers in exhaled breath [16, 17]. They proposed that the products of P-450 cytochrome catalyzed metabolization reactions and degradation products of membrane polyunsaturated fatty acids might be possible sources of monomethylated alkanes [15]. However, GCMS is time consuming, requires expensive equipment, experienced operators, and suitable pre-concentration methods.

Haick and co-workers [4] have recently demonstrated a breath test based on a nanoscale articial NOSE (TMNANOSE) that distinguishes patients having breast, lung, colon, and prostate cancer and healthy controls, irrespective of age, gender, lifestyle, smoking habits, and other confounding factors. The NA-NOSE is based on an array of highly cross-reactive gas sensors (here: chemiresistors based on different monolayer-capped metal nanoparticles) that can identify and separate different odors, even if the odorants are present at very low concentrations and their differences are very supple. Each sensor shows an individual response to all (or to a certain subset) of the volatile biomarkers that make up the cancer-odor. The odor is identied by analyzing the sensor signals with a statistical pattern recognition method. The authors have recently shown that a NA-NOSE can distinguish between malignant breast lesions and healthy states [4]. Here, the authors show that a tailor-made NA-NOSE can identify different benign breast conditions that are potential BC precursors. These benign conditions can be distinguished both from negative mammography and from malignant lesions. The identication of early BC precursors may improve treatment options and increase future BC survival rates. A cross-sectional case-controlled clinical trial is presented as proof-of-concept.

Materials and methods Test population and breath collection The test population comprised 36 female volunteers, aged 2673. These included seven women with negative mammography, 16 with benign breast conditions (micro-calcications and benign tumors), and 13 women with malignant lesions. All volunteers were recruited from the Department of Breast Imaging, Rambam Healthcare Campus (Haifa, Israel), after conventional diagnosis (as reference standard) by mammography, followed, if necessary, by biopsy, and prior to any treatment. Ethical approval was obtained from the Rambam Healthcare Campus and Technions committee for supervision of human experiments, Haifa, Israel. The clinical trial was registered at ClinicalTrials.gov (registration no. NCT01234987). The volunteers gave their written informed consent before the breath collection, and all experiments were performed according to the guidelines of the Rambam Healthcare Campus and Technions committee for supervision of human experiments. The clinical details of all tested subjects are provided in Table S1. Alveolar breath was collected, using an ofine method that effectively separates the endogenous from the exogenous breath volatile biomarkers, and excludes the nasal entrainment. Exhaled breath is a mixture of alveolar air and

123

Author's personal copy

Breast Cancer Res Treat (2011) 126:791796 793

respiratory dead space air. The alveolar breath can be separated from the dead space air either by collecting the breath in a CO2 controlled manner[17] or by collecting the dead space air into a separate bag prior to the collection of the alveolar breath [4, 12]. In this study, the authors used the latter method (see supporting online information and Refs. 4, 12, 18). Each woman paid 12 visits within a period of 2 months or less, with a mammography routine checkup during the rst and, if indicated, a biopsy during the second visit. Breath samples were collected during these two visits. Breath analysis using the NA-NOSE The breath samples were tested with a NA-NOSE developed by Haick and coworkers. The NA-NOSE was composed of an array of seven cross-reactive gas sensors that were based on six types of spherical gold (Au) nanoparticles (NPs) (core size of 36 nm) with different organic molecules (tert-dodecanethiol, 2-ethylhexanethiol, 2-mercaptobenzyl alcohol, 2-mercaptobenzoazole, calixarene, and octadecylamine) and one type of cubic platinum (Pt) NPs with organic encapsulating molecule benzylmercaptan. The NPs were synthesized by Haick and co-workers, as described in detail in Refs [19, 20]. Chemical similarities between the NP ligands and the anticipated cancer biomarkers are important criteria for the choice of the NP sensors for a particular application (for details see Ref. [12]). Each sensor showed a characteristic response to all (or to a certain subset) of the volatile biomarkers found in the exhaled breath samples. The sensing principle of the NA-NOSE was described in Refs. [4, 12]. Note, however, that a different set of sensors was used in this study. The sensors were mounted into a custom polytetrauoroethylene circuit board inside a stainless steel test chamber with a volume of approximately 500 cm3. Each sensor of the array underwent a reversible change in electrical resistance when exposed to a vapor or analyte. The responses were unique because of chemical diversity of the sensor materials. An Agilent Multifunction switch 34980 was used to select the active sensor and measure the responses of all sensor array elements. The entire system was computer controlled. In a typical experiment, signals of the sensor array elements were collected for 5 min in vacuum, followed by 5 min of breath that lled the chamber housed the array, and then followed by another 5 min of vacuum environment. The cycles are typically repeated two times to test reproducibility. Statistical analysis The NA-NOSE signals were analyzed using two different approaches. In the rst approach, six sensors (tert-

dodecanethiol, 2-ethylhexanethiol, 2-mercaptobenzyl alcohol, 2-mercaptobenzoazole, calixarene, and cubic Pt NPs benzylmercaptan) that showed statistically signicant differences between (pre) malignant breast lesions and healthy states were used. From these six sensors, the authors read out seven features and analyzed them with standard principal component analysis (PCA) [21]. The seven features used in this study were DR at the start of signal, DR/R at the start of signal, DR in the middle of signal, DR/R in the middle of signal, DR at the end of signal, DR/R at the end of signal, and the rate of signal drop. R is hereby the electrical resistance and DR is its steady state change upon exposure to the breath sample. The PCA of the features allowed a visual two-dimensional presentation of the BCrelated variability in the multidimensional data. A detailed description of the signal analysis will be published elsewhere as a technical paper. PCA determines the linear combinations of the sensing signals so that the maximum variance between all data points can be obtained in mutually orthogonal dimensions. Thus, PCA effectively reduces the multidimensional experimental data to its main components and improves the human perception of the data. However, PCA does not classify the data. Objective cluster identication was achieved by studying the statistical distribution of the rst principal component (containing more than 65% of the variance of the data) with one-way ANOVA. Separation between the test groups was analyzed using the Students t-test. In an independent, complimentary approach, support vector machines (SVM) analysis was used to classify the experimental data, using a total of four features from the NA-NOSE output, based around the response values of the start and end of the response, along with the rate of change in those areas: response start peak value, response start peak change rate, response end summation, and response end change rate summation. The SVM classication was based on two sensors (Au NPs capped with octadecylamine and the cubic Pt NPs capped with benzylmercaptan), and two features per sensor. These were chosen by the SVM algorithm to give best separation between the analyzed groups. Cross validation was utilized to evaluate the specicity and sensitivity [22, 23]. SVM analysis is a supervised learning method that nds the best separating line between two data sets, through computerized analysis of the sensing signal and automatic choice of the most suitable set of sensing features. It can be used for data classication and pattern recognition [22]. The advantage of SVM over ANOVA is that it does not require normal distribution of the data points around the average value. Therefore, SVM is more suitable when dealing with smaller data sets. The sub-populations were compared by building a multi-class classier based on a linear nu-SVC

123

Author's personal copy

794 Breast Cancer Res Treat (2011) 126:791796

SVM classier [23]. Cross validation was utilized to evaluate the specicity and sensitivity by randomly dividing each sub-population into two sets, which were then used as training and test set. All possible combinations of division into two sets were tested and the results were averaged. It was noted that the results were stable against changing the number of folds in the cross validation.

Results and discussion Figure 1a shows the rst principle component (PC1), which contains more than 65% of the variability within the data. Each point in the graph represents one test person. The authors observed a separation between breast cancer patients (malignant lesions) and healthy controls (including women with negative mammography and benign breast conditions) on the PC1 axis, with positive values for the former and negative values for the latter. These results are in agreement with the previous study [4]. Note, however, that the authors had to consider the rst two principal components, PC1 and PC2, for analysis of the sensing signals of the earlier NA-NOSE. The PCA results show that the variability within the data is related to the presence of pre-malignant and malignant breast lesions. Two patients were misclassied in each group (see Fig. 1a). Note that one of the misclassied cancer patients was diagnosed with tubular carcinoma, which is a welldifferentiated variant of IDC with an exceptionally favorable prognosis. Possibly, the tubular carcinoma cells have a weak volatile biomarker signature, because they are relatively similar to the healthy ductal cells. The mammography of the two healthy subjects that were misclassied as BC patients showed micro-calcications (see Table S1), which are associated with extra cell activity in the breast tissue. The extra cell growth is usually benign, but sometimes tight clusters of micro-calcications can indicate very early BC. The observed micro-calcications might possibly stem from early stage, high grade (i.e., fast developing) BC that releases relatively large quantities of BC marker volatile biomarkers while the tumor is still too small for observation. The statistical distribution of the PC1 values was studied using ANOVA and Students t-test. Note that the Students t-test is based on normal distribution of the data points and equal variances within the two groups that are compared. The authors therefore did not consider the four misclassied test persons described above in this part of the analysis. The PC1 values of the healthy and BC test groups were distributed around -3.43 and 6.22, respectively (see Fig. 1a, Table 1a). The error bars in Fig. 1a represent the standard deviations (containing 68% of the PC1 values under the assumption of normal distribution), and the 95%

Fig. 1 Graphical representation of the PC1 values from the six-sensor NA-NOSE for a healthy controls and breast cancer individuals, b for the healthy sub-populations with negative mammography and with benign breast conditions, and the breast cancer subgroups with DCIS (open symbols) and IDC (full symbols). Each point represents one patient. The positions of the PC1 mean values are marked with ?, the boxes correspond to their 95% condence limits, and the error bars corresponds to the standard deviation of PC1. The four misclassied individuals that are marked in the graphs were not considered for the statistical analysis

condence intervals (CIs) of the PC1 mean values are represented as boxes. The CIs are relatively large, as a result of the small test population. Nevertheless, they do not overlap, but are, on the contrary, well separated (P \ 0.0001) (cf. Table 1b). The presented analysis of the distribution of the PC1 values shows the breast-lesionrelated variability is statistically signicant. Furthermore, the NA-NOSE was able to identify benign breast conditions such as micro-calcications and benign lesions. Figure 1b shows the differences between the PC1 values corresponding to negative mammography, benign breast conditions, and malignant lesions. ANOVA and Students t test yielded statistically signicant differences (i) between benign breast conditions and negative

123

Author's personal copy

Breast Cancer Res Treat (2011) 126:791796 Table 1 One-way ANOVA analysis of the PC1 values for the correctly classied subjects: Mean value of PC1, standard deviation (SD), as well as upper and lower 95% condence limit (CL) for (a) Sub-population (a) (b) BC Healthy Malignant Lesion Benign Breast Conditions Clear Mammography
a,b

795 healthy controls and cancer patients, and (b) for the healthy subpopulations with clear mammography and with benign breast conditions, and the breast cancer patients with malignant lesion Mean PC1 6.22 -3.43 6.22 -2.56 -5.15 SD 3.51 2.35 3.51 2.01 2.13 Lower 95% CL 3.86 -4.50 4.59 -4.01 -7.20 Upper 95 CL 8.59 -2.37 7.85 -1.12 -3.11

No. of subjects 11a 21b 11a 14b 7

Note that four clearly misclassied patients (two with malignant lesions and two with begin breast conditions) were excluded from the ANOVA analysis (cf. Fig 1)

mammography within the healthy test group (P = 0.04) and (ii) between benign breast conditions and malignant lesions (P \ 0.0001) (cf. Tables 1b and 2b). The relatively large 95% CIs of the two healthy sub-populations overlap only marginally, as can be seen in Fig. 1b. The authors expect that a larger clinical trial resulting in smaller CIs would enhance the separation. It should be noted that a false positive separation between the tested sub-populations, resulting from the statistical PCA analysis, can safely be ruled out, because a separation can be observed, though less pronounced, in the signal of each of the constituent sensor. Using the 6-sensor NA-NOSE in conjunction with PCA analysis improves the signal-to-noise ratio of the sensing signal, and, hence, increases separation between the sub-populations. The results also indicate that it might be possible to distinguish between sub-populations corresponding to IBC and DCIS patients within the breast cancer test group, so that the studied NA-NOSE could identify and separate between women with negative mammography, benign breast diseases, DCIS, and IDC. However, this study included only two DCIS patients, and a larger clinical trial would be necessary to conrm this promising preliminary result. Using PCA, the authors could separate negative mammography, benign breast conditions, and malignant lesions using PC1 alone, which contained only 65% of the datas variance. In a complimentary approach, the authors used

SVM analysis to nd the best separating line between three data sets, by applying a multi-class classier to the entire experimental data [22]. The specicity and sensitivity were determined through cross validation as described in Statistical analysis. The numbers of correct and incorrect patient classications are listed in Table 3. Benign breast conditions classied as such are true positive (TP), benign breast conditions classied either as clear mammography or as malignant lesions are false negative (FN). Malignant lesions and clear mammography classied as either are true negative (TN) and malignant lesions and clear mammography classied as benign breast conditions are false positive (FP). From this, the authors can extract a sensitivity of 94% and a specicity of 80% for detecting benign breast conditions through breath testing with the NA-NOSE alone. To validate the results, label shufing was repeatedly applied to the samples, and the SVM algorithm was ran in same manner as in the true-labels case. The mean classication accuracy in the shufed populations was much lower than in the true-label population, and was close to random (showing no classication ability). The shufing test validates that the presented classication is stable and not related to noise. A comprehensive chemical analysis of the collected breath samples, using GCMS, to understand the biochemistry pathways of the volatile biomarkers associated with the different breast diseases, is currently underway and will be presented elsewhere.

Table 2 Students t-test for detecting statistically signicant differences (a) between healthy controls and breast cancer patients, and (b) between each pair of the four sub-populations: healthy controls with Sub-population 1 (a) (b) Healthy Negative mammography Benign breast condition Negative mammography Sub-population 2 BC Benign breast conditions Malignant lesion Malignant lesion

clear mammography, healthy controls with benign breast conditions, and breast cancer patients Difference of the PC1 mean values 9.64 2.59 8.78 11.37 Lower 95% CL difference 12.15 0.08 6.60 8.75 Upper 95% CL difference 7.13 5.09 10.96 13.98 P value \0.0001 0.04 \0.0001 \0.0001

123

Author's personal copy

796 Table 3 Number of correct and incorrect patient classications using supportive vector machine (SVM) and cross validation Negative mammography 5 0 0 Benign breast conditions 2 15 2 Malignant lesions 0 1 11 Classied as Negative mammography Benign breast conditions Malignant lesions Breast Cancer Res Treat (2011) 126:791796 5. Songa G, Qina T, Liua H, Xub GB, Pana YY, Xionga FX, Gua KS, Suna GP, Chena ZD (2010) Quantitative breath analysis of volatile organic compounds of lung cancer patients. Lung Cancer 67:227231 6. Conzen SD, Grushko TA, Olopade OI (2008) Cancer of the breast. In: Vincent T, DeVita J, Lawrence TS, Rosenberg SA (eds) DeVita, Hellman, and Rosenbergs cancer: principles & practice of oncology, vol 2, 8th edn. Wolters Kluwer, Philadelphia, pp 15951653 7. Vousden KH, Ryan KM (2009) p53 and metabolism. Nat Rev Cancer 9(10):691700. doi:nrc2715[pii]10.1038/nrc2715 8. Okunieff P, Fenton B, Chen Y (2005) Past, present, and future of oxygen in cancer research. Adv Exp Med Biol 566:213222 9. Kneepkens CMF, Lepage G, Roy CC (1994) The potential of the hydrocarbon breath test as a measure of lipid peroxidation. Free Rad Biol Med 17(2):127160 10. Gordon SM, Szidon JP, Krotoszynski BK, Gibbons RD, Oneill HJ (1985) Volatile organic-compounds in exhaled air from patients with lung-cancer. Clin Chem 31(8):12781282 11. Preti G, Labows JN, Kostelc JG, Aldinger S, Daniele R (1988) Analysis of lung air from patients with bronchogenic carcinoma and controls using gas chromatographymass spectrometry. J Chromatogr 432:111 12. Peng G, Tisch U, Adams U, Hakim M, Shehada N, Broza YY, Billan S, Abdah-Bortnyak R, Kuten A, Haick H (2009) Diagnosing lung cancer in exhaled breath using gold nanoparticles. Nat Nanotechnol 4(10):669673 13. Phillips M, Cataneo RN, Ditkoff BA, Fisher P, Greenberg J, Gunawardena R, Kwon CS, Tietje O, Wong C (2006) Prediction of breast cancer using volatile biomarkers in the breath. Breast Cancer Res Treat 99(1):1921 14. Phillips M, Cataneo RN, Saunders C, Hope P, Schmitt P, Wai J (2010) Volatile biomarkers in the breath of women with breast cancer. J Breath Res 4:026003 15. Krkosova Z, Kubinec R, Sojak L, Amann A (2008) Temperatureprogrammed gas chromatography linear retention indices of all C4C30 monomethylalkanes on methylsilicone OV-1 stationary phase. J Chromatogr A 1179(1):5968 16. Filipiak W, Sponring A, Filipiak A, Ager C, Schubert J, Miekisch W, Amann A, Troppmair J (2010) TDGCMS analysis of volatile metabolites of human lung cancer and normal cells in vitro. Cancer Epidemiol Biomark Prev 19(1):182195 17. Bajtarevic A, Ager C, Pienz M, Klieber M, Schwarz K, Ligor M, Ligor T, Filipiak W, Denz H, Fiegl M, Hilbe W, Weiss W, Lukas P, Jamnig H, Hackl M, Haidenberger A, Buszewski B, Miekisch W, Schubert J, Amann A (2009) Noninvasive detection of lung cancer by analysis of exhaled breath. BMC Cancer 9:348 18. Peng G, Trock E, Haick H (2008) Detecting simulated patterns of lung cancer biomarkers by random network of single-walled carbon nanotubes coated with nonpolymeric organic materials. Nano Lett 8(11):36313635 19. Dovgolevsky E, Haick H (2008) Direct observation of the transition point between quasi-spherical and cubic nanoparticles in two-step seed-mediated growth method. Small 4(11):20592066 20. Dovgolevsky E, Tisch U, Haick H (2009) Chemically sensitive resistors based on monolayer-capped cubic nanoparticles: towards congurable nanoporous sensors. Small 5(10): 11581161 21. Roeck F, Barsan N, Weimar U (2008) Electronic nose: current status and future trends. Chem Rev 108(2):705725 22. Cortes C, Vapnik V (1995) Support-vector networks. Mach Learn 30(3):273297 23. Hall M, Frank E, Holmes G, Pfahringer B, Reutemann P, Witten IH (2009) The WEKA data mining software: an update. SIGKDD Explor Newsl 11(1):1018

Conclusion The authors have delivered a proof of concept for the ability of a tailor-made NA-NOSE to identify with high sensitivity and specicity women with benign breast conditions in a population that contains women with negative mammography and women with malignant lesions. The multi-dimensional sensing signal was statistically analyzed using two entirely independent, complementary approaches, PCA and SVM analysis. Both methods yielded similar results. This method could form the basis of a future cost-effective, fast, and reliable diagnostic test for possible breast cancer precursors such as benign tumors and microcalcications. Moreover, this method holds future potential as screening test that would be more comfortable than mammography and would not involve X-ray irradiation.
Acknowledgments The research was funded by the Marie Curie Excellence Grant of the European Commissions FP6 program (H.H.), the Alfred Mann Institute (H.H.), and the Friends for an Earlier Breast Cancer Test (H.H.). The authors acknowledge Ms. Rana Bassal and Ms. Nisreen Shehada (Technion-IIT) for their assistance in breath collection. H.H. is a Knight of the Order of Academic Palms and holds the Horev Chair for Leaders in Science and Technology. Conicts of interest related to the study. All co-authors declare no competing interests

References
1. Guray M, Sahin AA (2006) Benign breast diseases: classication, diagnosis, and management. Oncologist 11:435449 2. American Cancer Society (2009) Breast cancer facts and gures 20092010. American Cancer Society, Inc., Atlanta 3. Wiechmann L, Kuerer HM (2008) The molecular journey from ductal carcinoma in situ to invasive breast cancer. Cancer 112(10):21302142 4. Peng G, Hakim M, Broza YY, Billan S, Abdah-Bortnyak R, Kuten A, Tisch U, Haick H (2010) Detection of lung, breast, colorectal, and prostate cancers from exhaled breath using a single array of nanosensors. Br J Cancer 103:542551

123