Histology-Driven Data Mining of Lipid Signatures from Multiple

Jan 24, 2013 - E. Ellen Jones , Shaalee Dworski , Daniel Canals , Josefina Casas ... Nathan Heath Patterson , Balqis Alabdulkarim , Anthoula Lazaris ...
0 downloads 0 Views 2MB Size
Download PDF
Article pubs.acs.org/ac

Histology-Driven Data Mining of Lipid Signatures from Multiple Imaging Mass Spectrometry Analyses: Application to Human Colorectal Cancer Liver Metastasis Biopsies Aurélien Thomas,† Nathan Heath Patterson,† Martin M. Marcinkiewicz,‡ Anthoula Lazaris,§ Peter Metrakos,§ and Pierre Chaurand*,† †

Department of Chemistry, University of Montreal, Montreal, Quebec, Canada Cytochem Inc., 6465 Durocher Avenue, Montreal, Quebec, Canada § Department of Surgery, Division of Hepatobiliary and Transplant Surgery, McGill University Health Center, Quebec, Canada ‡

S Supporting Information *

ABSTRACT: Imaging mass spectrometry (IMS) represents an innovative tool in the cancer research pipeline, which is increasingly being used in clinical and pharmaceutical applications. The unique properties of the technique, especially the amount of data generated, make the handling of data from multiple IMS acquisitions challenging. This work presents a histology-driven IMS approach aiming to identify discriminant lipid signatures from the simultaneous mining of IMS data sets from multiple samples. The feasibility of the developed workflow is evaluated on a set of three human colorectal cancer liver metastasis (CRCLM) tissue sections. Lipid IMS on tissue sections was performed using MALDI-TOF/TOF MS in both negative and positive ionization modes after 1,5-diaminonaphthalene matrix deposition by sublimation. The combination of both positive and negative acquisition results was performed during data mining to simplify the process and interrogate a larger lipidome into a single analysis. To reduce the complexity of the IMS data sets, a sub data set was generated by randomly selecting a fixed number of spectra from a histologically defined region of interest, resulting in a 10-fold data reduction. Principal component analysis confirmed that the molecular selectivity of the regions of interest is maintained after data reduction. Partial least-squares and heat map analyses demonstrated a selective signature of the CRCLM, revealing lipids that are significantly up- and down-regulated in the tumor region. This comprehensive approach is thus of interest for defining disease signatures directly from IMS data sets by the use of combinatory data mining, opening novel routes of investigation for addressing the demands of the clinical setting.

L

molecular images and the histology of the tissue sections.14−18 Numerous studies have already demonstrated the power of IMS for lipidomic investigations of cancer biopsies, revealing important alterations of lipid expression.19−22 These lipid-based works, along with those performed on proteins, reflect the potential of IMS as a technique to identify new candidate biomarkers of disease.23−26 To date, two main strategies for the preliminary identification of biomarkers using IMS have been reported.27 The first approach, which is perhaps the most common, consists of acquiring molecular profiles in the region of interest (ROI) on the basis of a pathologist’s histological annotations.28−31 These MS profiles are then analyzed to identify signals differentially expressed in the regions of interest (for example, the tumor versus the normal region). IMS is then performed as a validation of the different discriminant ions and to determine where these are localized. Profiling approaches

ipids are the major constituents of biomembranes and act as regulators for various biological processes such as maintaining homeostasis, metabolism, cell adhesion and migration, signal transduction, and apoptosis.1,2 Given their important cellular roles, it is not surprising that modifications of lipid metabolism play a key role in the pathogenesis of many diseases, including atherosclerosis, Alzheimer’s disease, and cancer.3−6 Integration of lipidomic analysis strategies in the cancer research pipeline has created new opportunities to better understand the molecular mechanisms involved in disease onset and development.7,8 Such strategies have already provided novel biological insights, including the role of certain lipases in glycerolipid metabolism, a key mechanism in the generation of free fatty acids during cell proliferation.9 Incorporated within these innovative advances is imaging mass spectrometry (IMS), which is increasingly being applied to lipid analysis.10−13 Unlike conventional liquid chromatography−MS or shotgun approaches, IMS allows for the mapping of the expression of hundreds of lipids possible in an unbiased and specific fashion while maintaining a high correlation between the resulting © 2013 American Chemical Society

Received: November 26, 2012 Accepted: January 24, 2013 Published: January 24, 2013 2860

dx.doi.org/10.1021/ac3034294 | Anal. Chem. 2013, 85, 2860−2866

Analytical Chemistry

Article

have several advantages: faster sample preparation and data acquisition, less complex and intuitive data analysis, and decreased requirement for powerful computational environments to treat the acquired data. However, profiling is not performed in the same manner as IMS (i.e., matrix deposition, number of acquired shots), which may compromise the crossvalidation of markers with ion images. Furthermore, in the profiling approach, the MS spectra are only obtained from the specific region of matrix deposition, making the retrospective analysis of a signal across all regions present in the tissue impossible.27 The second approach is based on a histologyindependent analysis consisting of the direct classification of entire imaging data sets to determine the histological molecular content from a completely unsupervised variance analysis, meaning no pathologist annotation is initially required.23,32,33 Although histology validation is necessary in the end, the main limitation of this approach is the demanding computation needed to perform such classification from entire tissue section data sets.34 This limitation is further emphasized when the number of samples is increased or the sample image is large.27 The goal of this feasibility study was to develop a histologydriven data mining approach for the simultaneous lipid IMS analysis of several human colorectal cancer liver metastasis (CRCLM) tissue sections. A preliminary set of three patients was analyzed in both negative and positive ionization modes by MALDI-TOF/TOF MS after 1,5-diaminonaphthalene (1,5DAN) matrix sublimation deposition. Both polarity data sets were combined to interrogate a larger lipidome into a single analysis. Data reduction was carried out in each region of interest by randomly selecting a fixed number of pixels within the region that does not compromise its selectivity. The described strategy was demonstrated to be useful for identifying specific molecular signatures from multiple IMS data sets.

■

from the isopentane, placed in a sterile prelabeled plastic pouch, and stored at −80 °C until it is sectioned. Frozen CRCLM slices of tissue were sectioned at a thickness of 10 μm using a Hacker/Bright cryostat (Hacker Instruments & Industries Inc., Winnsboro, CA) and thaw-mounted on indium−tin oxide-coated glass slides (Delta Technologies, Stillwater, MN). Sections were dried in a desiccator prior to matrix deposition. After IMS data acquisition and matrix removal by immersion in ethanol, the sections were stained with hematoxylin and eosin (H&E) using standard protocols to assess the histological accuracy of ion images. To evaluate tumor homogeneity among the three patients, serial CRCLM tissues were also investigated by immunostaining the Ki67 cell cycle marker of tumor cells (rabbit antibody, Thermo Scientific, West Palm Beach, FL) and Il-1β macrophage marker (mouse monoclonal antibody, Abcam, Cambridge, MA) using a Vectastain Elite ABC kit (Vector, Burlingame, CA). Serial sections were also stained with hematoxylin/eosin. All the patient CRCLM specimens (i.e., P1, P2, and P3) revealed histological normal liver tissue, residual tumor, and necrosis and inflammation sites surrounded by liver parenchyma metastases consisting of cancer cells in tubelike and acinar structures and intratumor stroma. Tumors were characterized by a columnar appearance of the cells in the glands, which in apparently older structures evolved into acini filled with cancer cell mass. The specimens displayed a presence of more or less abundant central necrosis. Inflammation cells such as macrophages and neutrophils were seen within central necrosis sites, intraluminarly in some malignant crypts, and in vascularized fibrotic areas (Figure S1, Supporting Information). Each cancer species displayed certain individual characteristics that differed in each in tumor extent, abundance of necrosis, volume of fibrotic area, pattern of the metastasis invasion front, and abundance of inflammatory cells. Compelling characteristics are shown in Table S1 (Supporting Information), which indicates that the three specimens studied present a relatively homogeneous group of macroscopic tumors. Matrix deposition was carried out using a sublimation apparatus (Chemglass Life Science, Vineland, NJ) as previously described.35 Briefly, the sublimation protocol was optimized for a fixed vacuum of 5 × 10−2 Torr, monitoring temperature, time of application, and amount of matrix deposited, measured with a high-precision balance. MALDI Mass Spectrometry. IMS of tissue sections was performed on a MALDI-TOF/TOF Ultraflextreme mass spectrometer equipped with a SmartBeamII Nd:YAG/355 nm laser operating at 1 kHz with the laser focus set at 50 μm diameter using the “small” setting (Bruker Daltonics, Billerica, MA). Laser fluence was optimized to obtain the best S/N ratio, keeping maximum resolution and optimal reproducibility. For IMS data acquisition, 100 shots were summed per array position. Consecutive positive and negative IMS measurements from the same tissue section were acquired with a spatial resolution of 200 μm after a pixel shift of 100 μm in both x and y dimensions. Furthermore, no modifications were made in the laser parameters (i.e., laser attenuation offset and focus setting) when switching polarities. Imaging data acquisitions were performed in reflectron geometry under optimized delayed extraction conditions with source accelerating voltages of +25 and −20 kV in a mass range of 640−1640 Da in positive and negative polarities, respectively. A mass resolution of ∼M/ΔM = 20 000 was typically achieved in the mass window of phospholipids (i.e., 600−1000). External calibration was carried

MATERIALS AND METHODS

Chemicals and Reagents. Liquid chromatography grade solvents and 1,5-DAN were purchased from Sigma-Aldrich (St. Louis, MO). Patient Tissue Sampling, Sectioning, and Matrix Deposition. Tissue samples were procured from patients undergoing curative liver resection for CRCLM. Patients are all female and are between the ages of 56 and 65 and had undergone chemotherapy prior to liver resection. Full informed consent was obtained from all patients. This specimen collection is part of The MUHC Liver Diseases BioBank (Research Ethics Board (REB) approval of protocol SDR 11066) (MUHC = McGill University Health Center). Sample processing for MALDI IMS is crucial for ensuring the minimal degradation and architectural integrity. The tissues were frozen according to our established procedures and serially cut into whole-tumor sections 5−15 mm thick. Briefly, once the tumor, including surrounding liver tissue, is excised, it is immediately placed on plastic wrap on ice. Within 5 min the pathologist prepares a frozen section from one end of the tumor for diagnosis. Once the carcinoma and surgical margins are confirmed, the remaining tissue is released for research. After an elapsed time of 10 min on ice, to allow the tissue to cool, it is immersed in isopentane kept at −40 °C. Depending on the size of the sample, it is monitored to ensure complete freezing but not overfreezing, resulting in “freezer” burn, rendering the tissue more susceptible to cracking during downstream processing. Within 30 min, the tissue is removed 2861

dx.doi.org/10.1021/ac3034294 | Anal. Chem. 2013, 85, 2860−2866

Analytical Chemistry

Article

Figure 1. Generalized workflow of histology-driven data treatment for the simultaneous mining of multiple IMS data sets from both positive and negative polarity acquisitions: step 1, IMS of several tissue sections; step 2, ROI determination and data exportation; step 3, data reduction of both positive and negative polarity data sets; step 4, data matrix reconstruction; step 5, data mining on merged matrix and molecular identification.

Peak processing was performed on the reduced data set using ClinProTools v2.2 software (Bruker Daltonics), which is capable of importing the reduced data set file lists in .xml format. For each polarity data set, the mass spectra were internally recalibrated on common peaks and normalized by the total ion count. The peak-picking was performed over a spectrum averaged from all single spectra. These peaks were then found in each individual spectrum, and the corresponding intensities were plotted for computational purposes.32 The resulting positive and negative matrices were then merged together into a single matrix with a spreadsheet application. Supervised and unsupervised data mining were carried out on the merged matrix to identify data clusters for comparing lipid signatures of normal liver regions to those of tumor regions representative of the different patients. Principal component analysis (PCA), partial least-squares (PLS) analysis, and heat map data mining were performed using both Tanagra (http:// eric.univ-lyon2.fr/∼ricco/tanagra/) and Orange (http:// orange.biolab.si/) software, v1.4.45 and v2.6a1, respectively. Discriminant lipids were characterized by comparing the accurate mass measurements and the MS/MS fragmentation pattern obtained in LIFT-TOF/TOF mode with the LIPID MAPS prediction tool (http://www.lipidmaps.org/tools/index. html). MS/MS data were processed using FlexAnalysis v3.3 software (Bruker Daltonics).

out in quadratic mode using a lipid and peptide mixture to obtain five points of calibration over the investigated mass range. A mass accuracy better than 40 ppm was obtained across all tissue section images. Data Analysis. IMS data were reconstructed and visualized using FlexImaging v2.1 software (Bruker Daltonics). For each tissue section, in the positive ionization mode, ROIs were selected with the goal of being as representative as possible of the pathologist’s annotations in both the tumor and the “normal” regions. MS profiles were generated from average spectra using MALDIQuant in the R environment.36 A list of file paths leading to the pixel spectra enclosed in these regions was then exported as an .xml file (around 1000 pixels each). Following this, the ClinProtSpectra Import XML generator application was used to reduce the data by randomly selecting a user-determined number of pixels from the data set. A 10-fold reduction was performed, resulting in data sets of 100 pixels per ROI. The associated, reduced .xml format file list for the negative polarity was made by changing the file directory names in the one obtained for the positive mode using a text editor. This step ensures that both the positive and negative reduced data sets represent the neighboring desorption location within the tissues, and thus, the same data reduction is performed on both ionization polarities, meaning the spectra of the reduced data are offset only by the pixel offset defined during the dual positive/negative acquisition and are not entirely new random selections. 2862

dx.doi.org/10.1021/ac3034294 | Anal. Chem. 2013, 85, 2860−2866

Analytical Chemistry

Article

Figure 2. Principal component analysis score plot of the IMS data from the normal (A) and the tumor (B) regions from a single tissue section acquired in positive ionization mode. The complete and the reduced data sets are displayed as blue and red, respectively.

■

RESULTS AND DISCUSSION Histology-Driven Workflow for Multiplex IMS Data Mining. Regarding the complexity of IMS data sets, the most commonly used approach is to classify samples and identify discriminant ions from profiling analysis that can subsequently be confirmed by an IMS acquisition.28 The direct handling of several IMS data sets for the potential biomarker discovery needs to be generalized by the development of dedicated strategies.24,37 Within a histology-driven approach, the first step is to select the ROIs from the ion images on the basis of a pathologist’s annotations (Figure 1, steps 1 and 2). However, the peak processing capacity is usually limited to an amount of pixels that may not include the entire image in a 32 bit or even 64 bit computational environment. This threshold oftentimes limits the size of ROI selection, particularly when the study is performed on large tissue sections and/or a high number of patients. As shown in Figure 1, we developed our strategy to ensure that this ROI selection is as representative as possible of the pathologist’s annotations (i.e., without any size limitation). However, to stay within our computational capacity, a data reduction was performed by randomly selecting a fixed number of pixels within the ROI (Figure 1, step 3). For this study, we fixed this number to 100 pixels to make an approximately 10fold reduction of the initial size of the ROIs. It is important to note that this reduction rate can be easily adjusted to individual users’ computational capacities. The PCA of the complete and reduced data sets shows the reduction did not compromise the integrity of information coming from each ROI (Figure 2). The use of a 1,5-DAN matrix deposited by sublimation has been demonstrated to be of high efficiency for the IMS of lipids, providing rich information in both ionization modes.35 By using a fixed offset in the x and y dimensions of the grid array, images were acquired serially in both positive and negative polarities. The goal of data mining is to find, among multiple candidates, the most relevant markers for classifying the state of a disease. Combination of both polarity acquisitions is interesting for concurrently interrogating a larger lipidome from a single tissue section. However, the random selection performed during data reduction stipulates only a certain number of pixels will be selected in the positive reduced data set. The adjacent pixels in the negative grid array of those randomly selected in the positive grid array must also be selected for spatial continuity in the data merging, meaning selecting pixels that correspond to a nearly identical laser desorption area, offset diagonally only by half the pixel resolution (i.e., 100 μm in a 200 μm resolution image). Since the IMS measurement regions for positive and negative polarities were defined from the same optical image, the

designations of the corresponding pixels were identical over both polarity acquisitions, implying that only the file directory name in the reduced pixel list has to be changed (Figure 1, step 3). Due to this, the reduced matrices obtained can be easily merged on the basis of the pixel coordinates to perform a single, combinatory data mining process (Figure 1, steps 4 and 5). Identification of the Colorectal Cancer Metastasis Signature. Figure S2 (Supporting Information) presents typical average positive full scan mass spectra of normal and CRCLM regions. As can be visually verified, the lipid profiles are differentially expressed in the tumor region compared to the normal region. This result was confirmed by the PLS analysis performed on the merged matrix from the three patient tissue sections (Figure 3).

Figure 3. Partial least-squares plot of the adjacent normal (blue) and colorectal cancer liver metastasis (red) regions for the simultaneous mining of IMS data sets from three patients.

The metastasis and the adjacent normal liver regions were highly classified with more than 82% total variance explained by the first projection axis, presenting positive and negative coordinates on PLS1 for the tumor and the normal regions, respectively. The merging of both positive and negative matrices resulted in a slight enhancement of the explained variance in the first projection, mainly as a result of the greater number of variables (Figure S3, Supporting Information). Early detection and prognosis of disease requires the use of a highly sensitive and specific marker. Facing the complexity of finding such a single “perfect” target, biomarker research is evolving toward the determination of multiple candidates to provide a molecular pattern, or signature, of a given disease phenotype.38−40 Accordingly, the top 20 discriminant ions were selected using a Mann−Whitney U test (p < 0.05). As demonstrated in Figure 4, the heat map analysis reveals a signature for the CRCLM by reflecting the modification of the discriminant lipid expression between the different ROIs of the 2863

dx.doi.org/10.1021/ac3034294 | Anal. Chem. 2013, 85, 2860−2866

Analytical Chemistry

Article

Figure 4. Heat map and bidirectional hierarchical clustering analysis showing the expression intensity of the top 20 discriminant ions for colorectal cancer liver metastasis and adjacent normal liver regions from three patients. Discriminant ions are determined by a Mann−Whitney U test. The letter in brackets identifies the polarity of the ionization mode, with P and N for positive and negative, respectively. A black and red gradient represents the down- and up-regulation of lipids, respectively. Each cell represents a sum for the associated m/z of 20 pixels among the 100 pixels randomly selected for each ROI during data reduction, corresponding to 5 cells per ROI. P1, P2, and P3 correspond to the data sets for patient 1, patient 2, and patient 3, respectively. Top and side clustering were performed by m/z and patient ROIs, respectively.

three patients. This modification of expression was emphasized by the hierarchical cluster dendograms. Indeed, this bidirectional classification highlighted two groups of lipids, which were up- and down-regulated in the CRCLM regions. As expected, these groups were composed of both positive and negative ions, which is in agreement with the discussion above of variance being related to matrix merging. IMS of the Discriminant Lipid Species. On the basis of the heat map results, we confirmed the observed marker patterns by retrospectively visualizing the associated ion images in each patient sample. Ion images presented in Figure 5 and Figure S4 (Supporting Information) were taken from the three patients by serially acquiring both positive and negative polarities from the same tissue section with a spatial resolution of 200 μm and an offset of 100 μm on both x and y dimensions with respect to both grid arrays. As expected, the ion images confirmed the lipid signature established by the heat map. For instance, the negative m/z 762 was highly down-regulated in the tumor, where the signal was practically nonexistent (Figure 5B). In contrast, the positive m/z 746 was overexpressed in the tumor region (Figure 5C). Due to this regioselectivity, a composite ion image (for example, the sum of m/z 762 and 778) was generated to efficiently exhibit the tumor and normal regions within the biopsy (Figure 5D). A strong correlation can

be observed between the selected ion images and the pathologist annotations (from the H&E staining, Figure 5A), which validates the location of the metastasis area at the molecular level. An interesting aspect of the lipid IMS is the possibility to directly characterize the regioselective species by on-tissue MS/ MS fragmentation.41 Although not necessary for classification or prognosis, the molecular identification remains essential for highlighting and exploring biological processes of interest or even finding a potential therapeutic target.42 For instance, different MS/MS spectra are presented in Figure S5 (Supporting Information). As shown, m/z 762 and 778 are identified as the phosphatidylethanolamines (PEs) 38:6 and PE-p 40:4, respectively.

■

CONCLUSIONS

This work presents a histology-driven approach for the multiplex handling of several IMS data sets. From the image acquisitions to the molecular characterization, this integrated IMS strategy resulted in the identification of a potential human CRCLM lipid signature. Although evaluated on a small sample size, we demonstrated that discriminant lipid ion images can be retrospectively extracted from multiple IMS data sets by data mining. The 2864

dx.doi.org/10.1021/ac3034294 | Anal. Chem. 2013, 85, 2860−2866

Analytical Chemistry

Article

acknowledge financial support from the Canadian Foundation for Innovation, the Natural Sciences and Engineering Research Council of Canada, and Le Fonds de Recherche du Québec Nature et Technologies. A.T. is a recipient of a fellowship supported by the Swiss National Science Foundation.

■

Figure 5. Discriminant ion images of colorectal cancer liver metastasis of patient P2 retrospectively selected on the basis of the lipid signature obtained from the heat map (Figure 4). Ion images are correlated to a serial section after H&E staining (A). Ion images B and C were obtained from the negative and positive acquisitions, respectively. Ion image D is a composite of two negative m/z acquisitions, downregulated (762) and up-regulated (778), in the tumor region.

use of a 1,5-DAN matrix to acquire serially both negative and positive polarity IMS data from a single tissue section contributes to an augmentation in the amount of useful information obtainable, consequently confirming the potential of lipid IMS for clinical investigations. Enriched by a decade of remarkable developments, the IMS technique is evolving to face the demands of the clinical setting. Among the ongoing efforts, the means to overcome IMS data complexity remains challenging, especially in large patient cohort studies. Such large clinical IMS studies could be used to establish robust molecular signatures as a basis for diagnosis and prognosis of disease or to track disease in response to therapy. The development of comprehensive data mining strategies, as presented in this work, is needed to translate the technology to effective large-scale clinical applications.

■

ASSOCIATED CONTENT

S Supporting Information *

Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org.

■

REFERENCES

(1) Dowhan, W. Annu. Rev. Biochem. 1997, 66, 199−232. (2) Shevchenko, A.; Simons, K. Nat. Rev. Mol. Cell. Biol. 2010, 11, 593−598. (3) Hammad, L. A.; Wu, G.; Saleh, M. M.; Klouckova, I.; Dobrolecki, L. E.; Hickey, R. J.; Schnaper, L.; Novotny, M. V.; Mechref, Y. Rapid Commun. Mass Spectrom. 2009, 23, 863−876. (4) Maxfield, F. R.; Tabas, I. Nature 2005, 438, 612−621. (5) Thomas, A.; Deglon, J.; Lenglet, S.; Mach, F.; Mangin, P.; Wolfender, J. L.; Steffens, S.; Staub, C. Anal. Chem. 2010, 82, 6687− 6694. (6) Oresic, M.; Hanninen, V. A.; Vidal-Puig, A. Trends Biotechnol. 2008, 26, 647−652. (7) Wenk, M. R. Cell 2010, 143, 888−895. (8) Fernandis, A. Z.; Wenk, M. R. J. Chromatogr., B 2009, 877, 2830− 2835. (9) Singh, R.; Kaushik, S.; Wang, Y.; Xiang, Y.; Novak, I.; Komatsu, M.; Tanaka, K.; Cuervo, A. M.; Czaja, M. J. Nature 2009, 458, 1131− 1135. (10) Kim, Y.; Shanta, S. R.; Zhou, L. H.; Kim, K. P. Exp. Mol. Med. 2010, 42 (1), 1−11. (11) Chughtai, K.; Heeren, R. M. Chem. Rev. 2010, 110, 3237−3277. (12) Berry, K. A.; Hankin, J. A.; Barkley, R. M.; Spraggins, J. M.; Caprioli, R. M.; Murphy, R. C. Chem. Rev. 2011, 111, 6491−6512. (13) Touboul, D.; Roy, S.; Germain, D. P.; Chaminade, P.; Brunelle, A.; Laprévote, O. Int. J. Mass Spectrom. 2007, 260, 158−165. (14) Chaurand, P.; Sanders, M. E.; Jensen, R. A.; Caprioli, R. M. Am. J. Pathol. 2004, 165, 1057−1068. (15) Blanksby, S. J.; Mitchell, T. W. Annu. Rev. Anal. Chem. 2010, 3, 433−465. (16) Thomas, A.; Lenglet, S.; Chaurand, P.; Deglon, J.; Mangin, P.; Mach, F.; Steffens, S.; Wolfender, J. L.; Staub, C. Thromb. Haemostasis 2011, 106, 20−33. (17) Chaurand, P.; Schwartz, S. A.; Billheimer, D.; Xu, B. J.; Crecelius, A.; Caprioli, R. M. Anal. Chem. 2004, 76, 1145−1155. (18) Chaurand, P. J. Proteomics 2012, 75, 4883−4892. (19) Dill, A. L.; Ifa, D. R.; Manicke, N. E.; Costa, A. B.; Ramos-Vara, J. A.; Knapp, D. W.; Cooks, R. G. Anal. Chem. 2009, 81, 8758−8764. (20) Dill, A. L.; Eberlin, L. S.; Zheng, C.; Costa, A. B.; Ifa, D. R.; Cheng, L.; Masterson, T. A.; Koch, M. O.; Vitek, O.; Cooks, R. G. Anal. Bioanal. Chem. 2010, 398, 2969−2978. (21) Shimma, S.; Sugiura, Y.; Hayasaka, T.; Hoshikawa, Y.; Noda, T.; Setou, M. J. Chromatogr., B 2007, 855, 98−103. (22) Park, Y. S.; Yoo, C. W.; Lee, S. C.; Park, S. J.; Oh, J. H.; Yoo, B. C.; Paik, S. S.; Lee, K. G.; Jin, S. Y.; Kim, S. C.; Kim, K. P.; Kim, Y. H.; Choi, D.; Kim, H. K. Clin. Chim. Acta 2011, 412, 1978−1982. (23) Willems, S. M.; van Remoortere, A.; van Zeijl, R.; Deelder, A. M.; McDonnell, L. A.; Hogendoorn, P. C. J. Pathol. 2010, 222, 400− 409. (24) Schwamborn, K. J. Proteomics 2012, 75, 4990−4998. (25) Lemaire, R.; Menguellet, S. A.; Stauber, J.; Marchaudon, V.; Lucot, J. P.; Collinet, P.; Farine, M. O.; Vinatier, D.; Day, R.; Ducoroy, P.; Salzet, M.; Fournier, I. J. Proteome Res. 2007, 6, 4127−4134. (26) Cazares, L. H.; Troyer, D.; Mendrinos, S.; Lance, R. A.; Nyalwidhe, J. O.; Beydoun, H. A.; Clements, M. A.; Drake, R. R.; Semmes, O. J. Clin. Cancer Res. 2009, 15, 5541−5551. (27) Jones, E. A.; Deininger, S. O.; Hogendoorn, P. C.; Deelder, A. M.; McDonnell, L. A. J. Proteomics 2012, 75, 4962−4989. (28) Hardesty, W. M.; Kelley, M. C.; Mi, D.; Low, R. L.; Caprioli, R. M. J. Proteomics 2011, 74, 1002−1014. (29) Groseclose, M. R.; Massion, P. P.; Chaurand, P.; Caprioli, R. M. Proteomics 2008, 8, 3715−3524.

AUTHOR INFORMATION

Corresponding Author

*Phone: (514) 343-2088. Fax: (514) 343-7586. E-mail: pierre. [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We acknowledge Jade Laveaux Charbonneau (University of Montreal) for her contribution in the IMS analysis and Ayat Salman (Clinical Research Associate, McGill University Health Center), who obtained the consent of the patients. We also 2865

dx.doi.org/10.1021/ac3034294 | Anal. Chem. 2013, 85, 2860−2866

Analytical Chemistry

Article

(30) Balluff, B.; Elsner, M.; Kowarsch, A.; Rauser, S.; Meding, S.; Schuhmacher, C.; Feith, M.; Herrmann, K.; Rocken, C.; Schmid, R. M.; Hofler, H.; Walch, A.; Ebert, M. P. J. Proteome Res. 2010, 9, 6317− 6322. (31) Cornett, D. S.; Mobley, J. A.; Dias, E. C.; Andersson, M.; Arteaga, C. L.; Sanders, M. E.; Caprioli, R. M. Mol. Cell. Proteomics 2006, 5, 1975−1983. (32) Deininger, S. O.; Ebert, M. P.; Futterer, A.; Gerhard, M.; Rocken, C. J. Proteome Res. 2008, 7, 5230−5236. (33) Amstalden van Hove, E. R.; Blackwell, T. R.; Klinkert, I.; Eijkel, G. B.; Heeren, R. M.; Glunde, K. Cancer Res. 2010, 70, 9012−9021. (34) McDonnell, L. A.; van Remoortere, A.; de Velde, N.; van Zeijl, R. J.; Deelder, A. M. J. Am. Soc. Mass Spectrom. 2010, 21, 1969−1978. (35) Thomas, A.; Charbonneau, J. L.; Fournaise, E.; Chaurand, P. Anal. Chem. 2012, 84, 2048−2054. (36) Gibb, S.; Strimmer, K. Bioinformatics 2012, 28, 2270−2271. (37) Schwamborn, K.; Caprioli, R. M. Nat. Rev. Cancer 2010, 10, 639−646. (38) Gustafsson, J. O.; Oehler, M. K.; Ruszkiewicz, A.; McColl, S. R.; Hoffmann, P. Int. J. Mol. Sci. 2011, 12, 773−794. (39) Moore, R. G.; MacLaughlan, S.; Bast, R. C., Jr. Gynecol. Oncol. 2010, 116, 240−245. (40) Minerva, L.; Ceulemans, A.; Baggerman, G.; Arckens, L. Proteomics Clin. Appl. 2012, 6, 581−595. (41) Chaurand, P.; Cornett, D. S.; Angel, P. M.; Caprioli, R. M. Mol. Cell. Proteomics 2011, 10, O110 004259. (42) Yanagisawa, K.; Shyr, Y.; Xu, B. J.; Massion, P. P.; Larsen, P. H.; White, B. C.; Roberts, J. R.; Edgerton, M.; Gonzalez, A.; Nadaf, S.; Moore, J. H.; Caprioli, R. M.; Carbone, D. P. Lancet 2003, 362, 433− 439.

2866

dx.doi.org/10.1021/ac3034294 | Anal. Chem. 2013, 85, 2860−2866