Dynamic Nanoparticle-Based Flexible Sensors - ACS Publications

Letter pubs.acs.org/NanoLett

Dynamic Nanoparticle-Based Flexible Sensors: Diagnosis of Ovarian Carcinoma from Exhaled Breath Nicole Kahn,† Ofer Lavie,‡ Moran Paz,‡ Yakir Segev,‡ and Hossam Haick*,† †

Department of Chemical Engineering and Russell Berrie Nanotechnology Institute, Technion−Israel Institute of Technology, Haifa 3200003, Israel ‡ Gynecological Oncology and Surgery Unit, Carmel Medical Center, Haifa 3436212, Israel

Downloaded by CENTRAL MICHIGAN UNIV on September 11, 2015 | http://pubs.acs.org Publication Date (Web): September 11, 2015 | doi: 10.1021/acs.nanolett.5b03052

S Supporting Information *

ABSTRACT: Flexible sensors based on molecularly modified gold nanoparticles (GNPs) were integrated into a dynamic cross-reactive diagnostic sensing array. Each bending state of the GNP-based flexible sensor gives unique nanoparticle spatial organization, altering the interaction between GNP ligands and volatile organic compounds (VOCs), which increases the amount of data obtainable from each sensor. Individual dynamic flexible sensor could selectively detect parts per billion (ppb) level VOCs that are linked with ovarian cancers in exhaled breath and discriminate them from environmental VOCs that exist in exhaled breath samples, but do not relate to ovarian cancer per se. Strain-related response successfully discriminated between exhaled breath collected from control subjects and those with ovarian cancer, with data from a single sensor being sufficient to obtain 82% accuracy, irrespective of important confounding factors, such as tobacco consumption and comorbidities. The approach raises the hope of achieving an extremely simple, inexpensive, portable, and noninvasive diagnostic procedure for cancer and other diseases. KEYWORDS: Nanoparticle, sensor, strain, volatile organic compound, breath, cancer We now report on a breath diagnostic array based on flexible GNP sensors that can be bent to multiple strain states under a single volume of gas. When GNP films are prepared on flexible substrates, the spatial organizational changes that result from bending or stretching allow them to be used as strain sensors.23−28 Sensor response varies with the type of VOC and concentration,29−31 and this allows the design of multiparametric sensing arrays composed of molecularly capped nanoparticles that can respond uniquely to specific VOC mixtures. Because of the dependence of film/VOC interaction on interparticle spacing,31−33 each bending state can be related to a single (virtual) sensor. This allows the measurement of both dynamic and static bending features that could allow extraction of multiple bending features on exposure to a single sample. The capabilities of this sensor approach in examining the breath VOCs for the diagnosis of ovarian cancer was carried out on real breath samples collected from 43 volunteers with ovarian cancer or control subjects (Figure 1). We particularly chose ovarian cancer because the currently available screening tests (pelvic examination, blood test for radioimmunoassay of CA 125, and transvaginal ultrasound) have limited specificity and sensitivity and are available only for high risk populations. Therefore, a complementary screening method for ovarian cancer, such as noninvasive breath analysis, could potentially be used in conjunction

A

promising approach for noninvasive and safe screening, diagnosis, and follow-up of disease conditions relies on the detection of volatile organic compounds (VOCs) in exhaled breath.1−5 Chemical sensor matrices based on molecularly modified nanoparticles (GNPs) may well become incorporated in a clinical and laboratory diagnostic tool because they are small, easy-to-use, and less expensive than other techniques, such as spectrometry.4,6,7 In general, an ideal GNP sensor for breath VOC analysis should be sensitive at very low concentrations of VOCs in the presence of water vapor because clinical samples are very humid.5,8 Furthermore, it should respond rapidly and reversibly differently to a range of VOCs, yet consistently respond to repeated exposures of the same compound. The sensor should rapidly return to its baseline state after evacuation of the VOC. Arrays of multiple chemiresistors on solid substrates have been used for diagnostic purposes, using pattern recognition methods and clinical data to build characteristic response profiles for the VOCs related to specific diseases in exhaled breath.9,10 Their application has been examined for various diseases, including lung,10−13 breast,14 colon,11 prostate,11 ovarian,15 gastric,16 and hepatic cancers,17 just as they can for noncancerous diseases (Parkinson and Alzheimer,18 chronic renal failure,19 acute kidney disease,20 pulmonary hypertension,21 and tuberculosis22). Nevertheless, solid-state chemiresistor arrays are limited by a small number of extractable sensing features for pattern recognition and therefore necessitate the inclusion of large number of specific sensors. © XXXX American Chemical Society

Received: August 2, 2015 Revised: September 8, 2015

A

DOI: 10.1021/acs.nanolett.5b03052 Nano Lett. XXXX, XXX, XXX−XXX

Letter


Nano Letters

Figure 1. Scheme illustrating experimental plan. (a) A sensing chamber allows the exposure of multiple sensors to a controlled VOC environment, while a crosshead on a linear motor is remotely manipulated to an exerted controlled strain. (b) The sensors are strained in multiple bending steps; at each increment of increasing strain, the nanoparticle film morphology is affected, leading to unique responses to the volatile organic compounds present. (c) Sensor resistance is measured during sensor bending and analyte exposure, allowing the collection of multiple bending-related features. Sensor behavior was stable and repeatable from one bending cycle to the other. The maximum drift in the baseline resistance was ∼7%, while ΔR/Rb changed only slightly (2%) after 10,000 bending cycles.26 (d) Sensors were prepared by the layer-by-layer deposition method involving repeated cycles of bi-ligand nanoparticles and dithiol linker molecules, allowing precise control over film morphology and thickness, using (e) a total of five different thiol ligands to modify molecularly the nanoparticles.

face down over a gap in the chamber floor, and edges were gripped to allow electrode bending while preventing substrate stretching. The VOC and bending response of these sensors was analyzed using the setup illustrated in Figure 1. GNP thin layers were optimized using this system, as described below. CBMT sensors of varying film thicknesses (characterized by UV absorption) were prepared, and film morphology was examined using high resolution SEM (SI, section 5). Unbent VOC response increased with film thickness, which had a strong effect on strain response in a VOC-dependent manner. These results are discussed in detail in the SI, section 6. The general assumption is that any VOC that can be adsorbed in the GNPs film would increase the bending response relative to that under atmosphere, thus increasing the interparticular space and allowing room for additional adsorption that increases the film swelling. However, adjacent ligand−ligand interactions and ligand−VOC interactions define the response mechanisms, described in greater detail in the SI, section 2. To explore these interactions, sensors were prepared with the ligands indicated in Figure 1e at a single film thickness (0.3 OD) and exposed to the seven VOCs mentioned above. The results are shown in Figure 2 and more fully discussed in SI, section 7. The overall bending response (ΔR/Rb) to a strain of 0.15% was measured and normalized to the bending response under nitrogen.

with the currently available tests, permitting screening of larger populations and detecting earlier stages of the disease. Kapton-based silver electrodes were overlaid with thin films of thiol-capped GNPs (listed in Figure 1), using hexadecanedithiol as a cross-linker in the layer-by-layer method.34 Aromatic ligands were chosen to increase film conductance with delocalized electrons, with a variety of electron withdrawing and donating substituents to increase the range of possible interactions with different VOCs. These sensors were exposed to a range of concentrations of VOCs that were linked in earlier work by mass spectrometry to ovarian cancer conditions in exhaled breath (styrene, nonanal, 2-ethylhexanol, 3-heptanone, decanal, and hexadecane), as well as to confounding environmental factors (ethanol35 and propionitrile36) that might exist in the breath, but were unrelated to the disease per se. For detailed information about the preparation of the GNPs, fabrication of the GNP-based flexible sensors, and on the surface characterization of the devices, see the Supporting Information (SI), sections 1.1 and 1.2. Signal analysis and bending-related feature extraction are detailed in SI, sections 3 and 4, respectively. An array of 10 sensors was exposed to different VOCs associated with ovarian cancer using a closed chamber connected to a controlled gas source while measuring the electrical resistance of the sensor. Dynamic bending was measured by controlling a computerized mobile cross-head. Sensors were laid B


Letter


Nano Letters

biomarkers. NTMBT and tBBT sensors had low responses to water (Figure 2), indicating a potentially good performance as elements of a breath-focused diagnostic array. CBMT sensors also had relatively small responses to water vapor and provided distinct responses to the six ovarian cancer breath biomarkers examined with little overlap between them. These results indicate that film thickness can control sensor sensitivity, while the type of ligand can be chosen to optimize the response to specific VOCs. To verify the ability of these sensors to differentiate between VOCs at the low concentrations relevant to clinical diagnostic uses, the sensors were exposed to styrene, nonanal, ethanol, and propionitrile at ppb levels. NOTE: styrene and nonanal serve as breath biomarkers for the differentiation between ovarian cancer and control samples; ethanol serves as a confounding factor for alcoholism;35 and propionitrile serves as a confounding factor for smoking.36 The sensors could differentiate between the two nonpolar VOCs and the confounding factors with an accuracy of 91% using leave-one-out cross validation. In addition to the multiple dynamic and static parameters collected from each bending step, a single larger strain step was added to the end of each exposure to allow for a single parameter that could be more

Figure 2. Normalized overall bending response (ΔR/Rb) for sensors of identical film thickness (OD = 0.3) under VOC exposure at a concentration of P/P0 = 0.04. The strain exerted was 0.15%.

The extent of response to water is particularly important; water significantly contaminates exhaled breath, and a strong response to water can mask the presence of relevant breath

Figure 3. Overall bending response (ΔR/Rb) to exertion of a strain of 0.27% from an unbent state of the nine sensors used as follows: (A) CBMT OD 0.3 sintered; (B) CBMT OD 0.4 sintered; (C) ETP; (D) CBMT OD 0.2; (E) CBMT OD 0.3; (F) CBMT OD 0.4; (G) NTMBT; (H) NT; and (I) tBBT. C


Letter


Nano Letters easily compared between sensors and VOCs. The results of this comparison are summarized in Figure 3. Many conclusions can be drawn from the graphs in Figure 3. Specific sensors are generally capable of differentiating both VOC type and concentration. The most significant error bars were visualized with the sintered sensors, which were notably noisier than the nonsintered sensors. However, a concentrationdependent strain response to nonanal was visualized for both sensors (Figure 3A,B). The ETP sensor was not strongly concentration-dependent (Figure 3C), but did allow for excellent differentiation between the confounding factors and ovarian cancer biomarkers. It is possible that this sensor can differentiate between polar and nonpolar VOCs. CBMT 0.3 OD provided concentration dependence for nonanal and styrene, but no clear concentration-dependence for the confounding factors (Figure 3E). All nonsintered CBMT sensors gave relatively strong responses to nonanal (Figures 3D−F). These results are particularly interesting for two reasons. First, they indicate the potential of individual sensors to respond to specific VOCs, suggesting that an array of sensors could be sensitive to a wide range of biomarkers. Second, they show that the VOC-dependent strain response is sensitive to concentrations down to 50 ppb for specific sensors, indicating that they are appropriate for diagnostic applications based on biomarker concentration in exhaled breath. Clinical samples were used to assess the potential of the system as a diagnostic tool. Breath samples were collected from volunteers from 43 subjects, of which 26 were controls and 17 had previously been diagnosed with epithelial ovarian cancer before initiating any therapeutic treatment (details given in Table 1 below). The ability of each individual sensor in the array to

Table 2. Results from Single Sensor Analysis of the Clinical Data, Providing the Separation Ability of Each Individual Sensor and the Number of Features Used for Each Analysis 0.3OD CBMT (sintered) 0.4OD CBMT (sintered) 0.2OD CBMT 0.3OD CBMT 0.4OD CBMT NTMBT ETP NT tBBT

number of patients

age

smoker (Y/N)

Hyperlipidemia (Y:N)

control ovarian cancer

26 17

56 ± 13 56 ± 11

4:22 3:14

8:18 6:11

specificity

accuracy

64.7% 58.8% 83.4% 76.5% 76.5% 58.8% 76.5% 58.8% 70.6%

69.2% 61.5% 80.8% 80.8% 69.2% 61.5% 73.1% 80.8% 80.8%

67.4% 60.5% 81.4% 79.1% 72.1% 60.5% 74.4% 72.1% 76.7%

multiple sensors. The values of 83.4% sensitivity, 80.8% specificity, and 81.8% accuracy compare favorably with previously published results,15 a study that had relied on an array of six different sensors to assess the differentiation between volunteers with positive ovarian cancer and tumor-free control subjects (which obtained 79% sensitivity, 100% specificity, and 89% accuracy). 0.3OD CBMT and tBBT also provided reasonable separation. This data is significant and attests to the impressive potential of flexible sensors in multiparametric sensing arrays. Leave-one-out analysis was used with DFA to separate the diagnostic groups for all the sensors in the array. Sensitivity of 82.3%, a specificity of 84.6%, and an accuracy of 83.7% were achieved. A student’s t-test was used to assess the statistical significance of each feature’s separation, and p-values of 0.002, 0.004, and 0.02 were obtained for each feature, respectively. Kfold analysis (k = 4) was used to verify the results. Sensitivity of 81.3%, a specificity of 82.9%, and an accuracy of 83.6% were achieved (a graph of the separation and graphs of the features used in the separation are show in the Figure 4). Confounding factors, such as eating or drinking before sample collection, hyperlipimia, smoking, and the order in which samples were tested, were insignificant. The features used to analyze the effect of these confounding factors were those found using leave-one-out analysis (results have been plotted in Figure 4). These factors did not cause artificial separation in the diagnostic study. In conclusion, this study provides a proof-of-concept for an integrated system that allows collection of resistance data in realtime under bending and exposure. In addition to diagnostic applications, this system could also be used to improve our understanding of the mechanisms of interdependent strain and VOC response. The utilization of these bending features for sensing applications is illustrated by a successful diagnostic trial. This use of flexible substrates could potentially become the next generation of current GNP-based solid state sensors, due to the increased number of features that can be measured and extracted from each exposure, increasing sensing ability and allowing the use of a smaller array of sensors, while still achieving diagnostic success. The clinical results obtained using the flexible sensors are promising and comprise an excellent proof of concept of a relatively simple system that would allow the extraction of a significant amount of useful data. Larger-scale clinical trials are required to clarify the capabilities and limitations of this diagnostic approach. While ovarian cancer has been targeted in this study, the ability of the sensor array methodology to identify other diseases by analyzing collected breath can be targeted.

Table 1. Details of Participants in the Clinical Study group

sensitivity

analyze the clinical data was assessed. The rationale behind this approach is that numerous bending-related features from single sensor can be extracted from each bending step, and each sensor can be bent to multiple increments of strain under a single volume of VOCs mixture. It may also be possible that specific ligand chemistries or film morphologies are more proficient at sensing specific VOCs or mixtures, in which case extraction of bending features from flexible sensors could be a method of functionalizing sensing arrays for specific needs. Discriminant Factor Analysis (DFA) was used with the leaveone-out method to find the sensitivity, specificity, and accuracy of each sensor using bending-related features from only one sensor. Briefly, DFA is a supervised statistical analysis method, aimed at finding the best possible separation between two previously known groups. The condition in the analysis is maximal variance between the two groups while maintaining minimal variance between members of the same group. The output of the DFA is a set of canonical variables (CVs) that are the dimensions that meet the prior requirement, with CV1 being the dimension with the highest differentiation power. Various functions can be used in DFA, but in this study we used a linear function for classification (more details in SI, section 3). Table 2 shows that 0.2OD CBMT alone could provide separation that compared favorably with that obtained using features from a wide array of D


Letter


Nano Letters

Figure 4. (A) Separation of the OC-positive and control groups (OC = ovarian cancer). (B) The two features that passed student t tests. (C) Confounding factors that could potentially affect diagnostic separation that proved to be insignificant.

■

ASSOCIATED CONTENT

S Supporting Information *

■

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.nanolett.5b03052. Synthesis of nanoparticles; sensor preparation and characterization; exposure under high concentrations of VOCs; exposure under low concentrations of VOCs; collection of breath samples; sensing mechanism of GNPbased chemiresistive sensors; signal analysis; bendingrelated feature extraction; effect of film thickness on film morphology; effect of thickness of VOC and strain

response; effect of ligand type on bending and VOC response (PDF)

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS The research leading to these results has received funding from the Horizon2020 ICT Program (grant agreement no. 644031). E


Letter

Nano Letters

(29) Olichwer, N.; Leib, E. W.; Halfar, A. H.; Petrov, A.; Vossmeyer, T. ACS Appl. Mater. Interfaces 2012, 4, 6151−6161. (30) Wang, L.; Luo, J.; Yin, J.; Zhang, H.; Wu, J.; Shi, X.; Crew, E.; Xu, Z.-Q.; Rendeng, Q.; Lu, S.; Poliks, M.; Sammakia, B.; Zhong, C.-J. J. Mater. Chem. 2010, 20, 907−915. (31) Yin, J.; Hu, P.; Luo, J.; Wang, L.; Cohen, M. F.; Zhong, C.-J. ACS Nano 2011, 5, 6516−6526. (32) Zamborini, P.; Leopold, M. C.; Hicks, J. F.; Kulesza, P. J.; Malik, M. A.; Murray, R. W. J. Am. Chem. Soc. 2002, 124, 8958−8964. (33) Haick, H. J. Phys. D: Appl. Phys. 2007, 40, 7173−7186. (34) Wang, L.; Shi, X.; Kariuki, N. N.; Schadt, M.; Wang, G. R.; Rendeng, Q.; Choi, J.; Luo, J.; Lu, S. J. Am. Chem. Soc. 2007, 129, 2161− 2170. (35) Bendtsen, P.; Hultberg, J.; Carlsson, M.; Jones, A. W. Alcohol.: Clin. Exp. Res. 1999, 23, 1446−1451. (36) Rodgman, A.; Perfetti, T. A. The Chemical Components of Tobacco and Tobacco Smoke; CRC Press: Boca Raton, FL, 2013.

The authors thank Dr. Viki Kloper for her generous help with figure graphics and Dr. Haitham Amal (Technion) for support and assistance.


■

REFERENCES

(1) Haick, H.; Broza, Y. Y.; Mochalski, P.; Ruzsanyi, V.; Amann, A. Chem. Soc. Rev. 2014, 43, 1423−1449. (2) Amann, A.; Mochalski, P.; Ruzsanyi, V.; Broza, Y. Y.; Haick, H. J. Breath Res. 2014, 8, 016003. (3) Shehada, N.; Brönstrup, G.; Funka, K.; Christiansen, S.; Leja, M.; Haick, H. Nano Lett. 2015, 15 (2), 1288−1295. (4) Broza, Y. Y.; Haick, H. Nanomedicine (London, U. K.) 2013, 8, 785− 806. (5) Konvalina, G.; Haick, H. ACS Appl. Mater. Interfaces 2012, 4, 317− 325. (6) Tisch, U.; Haick, H. Rev. Chem. Eng. 2010, 26, 171−179. (7) Nakhleh, M.; Broza, Y. Y.; Haick, H. Nanomedicine (London, U. K.) 2014, 9, 1991−2002. (8) Konvalina, G.; Haick, H. Acc. Chem. Res. 2014, 47, 66−76. (9) Han, L.; Shi, X.; Wu, W.; Kirk, F. L.; Luo, J.; Wang, L.; Mott, D.; Cousineau, L.; Lim, S. I. I.; Lu, S.; Zhong, C.-J. Sens. Actuators, B 2005, 106, 431−441. (10) Peng, G.; Tisch, U.; Adams, O.; Hakim, M.; Shehada, N.; Broza, Y. Y.; Billan, S.; Abdah-Bortnyak, R.; Kuten, A.; Haick, H. Nat. Nanotechnol. 2009, 4, 669−673. (11) Peng, G.; Hakim, M.; Broza, Y. Y.; Billan, S.; Abdah-Bortnyak, R.; Kuten, A.; Tisch, U.; Haick, H. Br. J. Cancer 2010, 103, 542−551. (12) Peled, N.; Hakim, M.; Bunn, P. A., Jr; Miller, Y. E.; Kennedy, T. C.; Mattei, J.; Mitchell, J. D.; Hirsch, F. R.; Haick, H. J. Thorac. Oncol. 2012, 7, 1528−1533. (13) Broza, Y. Y.; Kremer, R.; Tisch, U.; Gevorkyan, A.; Shiban, A.; Best, L. A.; Haick, H. Nanomedicine 2013, 9, 15−21. (14) Shuster, G.; Gallimidi, Z.; Reiss, A. H.; Dovgolevsky, E.; Billan, S.; Abdah-Bortnyak, R.; Kuten, A.; Engel, A.; Shiban, A.; Tisch, U.; Haick, H. Breast Cancer Res. Treat. 2011, 126, 791−796. (15) Amal, H.; Shi, D. Y.; Ionescu, R.; Zhang, W.; Hua, Q. L.; Pan, Y. Y.; Tao, L.; Liu, H.; Haick, H. Int. J. Cancer 2015, 136, E614−E622. (16) Xu, Z. Q.; Broza, Y. Y.; Ionsecu, R.; Tisch, U.; Ding, L.; Liu, H.; Song, Q.; Pan, Y. Y.; Xiong, F. X.; Gu, K. S. Br. J. Cancer 2013, 108, 941− 950. (17) Haick, H.; Ding, L.; Liu, B.-B.; Tisch, U.; Xu, Z.-Q.; Shi, D.-Y.; Zhao, Y.; Chen, J.; Sun, R.-X.; Liu, H. Int. J. Nanomed. 2012, 7, 4135. (18) Tisch, U.; Schlesinger, I.; Ionescu, R.; Nassar, M.; Axelrod, N.; Robertman, D.; Tessler, Y.; Marmur, A.; Aharon-Peretz, J.; Haick, H. Nanomedicine (London, U. K.) 2013, 8, 43−56. (19) Marom, O.; Nakhoul, F.; Tisch, U.; Shiban, A.; Abassi, Z.; Haick, H. Nanomedicine (London, U. K.) 2012, 7, 639−650. (20) Nakhleh, M. K.; Amal, H.; Awad, H.; Abu-Saleh, N.; Jeries, R.; Haick, H.; Abassi, Z. Nanomedicine 2014, 10, 1767−1776. (21) Cohen-Kaminsky, S.; Nakhleh, M.; Perros, F.; Montani, D.; Girerd, B.; Garcia, G.; Simonneau, G.; Haick, H.; Humbert, M. Am. J. Respir. Crit. Care Med. 2013, 188, 756−759. (22) Nakhleh, M. K.; Jeries, R.; Binder, A.; Broza, Y. Y.; Pascoe, M.; Dheda, K.; Haick, H. Eur. Respir. J. 2014, 43, 1522−1525. (23) Segev-Bar, M.; Konvalina, G.; Haick, H. Adv. Mater. 2015, 27, 1779−1784. (24) Konvalina, G.; Leshansky, A.; Haick, H. Adv. Funct. Mater. 2015, 25, 2411−2419. (25) Segev-Bar, M.; Haick, H. ACS Nano 2013, 7, 8366−8378. (26) Segev-Bar, M.; Landman, A.; Nir-Shapira, M.; Shuster, G.; Haick, H. ACS Appl. Mater. Interfaces 2013, 5, 5531−5541. (27) Moreira, H.; Grisolia, J.; Sangeetha, N. M.; Decorde, N.; Farcau, C.; Viallet, B.; Chen, K.; Viau, G.; Ressier, L. Nanotechnology 2013, 24, 095701. (28) Herrmann, J.; Müller, K. H.; Reda, T.; Baxter, G. R.; Raguse, B.; de Groot, G. J. J. B.; Chai, R.; Roberts, M.; Wieczorek, L. Appl. Phys. Lett. 2007, 91, 183105. F


Dynamic Nanoparticle-Based Flexible Sensors - ACS Publications

Recommend Documents