ARTICLE pubs.acs.org/jpr
Imaging Mass Spectrometry Provides Fingerprints for Distinguishing Hepatocellular Carcinoma from Cirrhosis Julie Le Faouder,†,‡ Samira Laouirem,‡ Manuel Chapelle,§ Miguel Albuquerque,‡ Jacques Belghiti,|| Franc) oise Degos,^ Valerie Paradis,‡,# Jean-Michel Camadro,§,z and Pierre Bedossa*,‡,# †
Institut Federatif de Recherche Claude Bernard, Universite Paris-Diderot, Paris, France - INSERM U773, Universite Paris-Diderot, Paris, France § Mass Spectrometry Facility, Jacques Monod Institute, UMR7592 Universite Paris-Diderot - CNRS, Paris, France Department of Liver Surgery, Beaujon Hospital, Assistance Publique-H^opitaux de Paris and Universite Paris-Diderot, France ^ Department of Hepatology, Beaujon Hospital, Assistance Publique-H^opitaux de Paris and Universite Paris-Diderot, France # Department of Pathology, Beaujon Hospital, Assistance Publique-H^opitaux de Paris and Universite Paris-Diderot, France z Molecular and Cellular Pathology Program, Jacques Monod Institute, UMR7592 Universite Paris-Diderot - CNRS, Paris, France
)
‡
bS Supporting Information ABSTRACT: MALDI imaging mass spectrometry (MALDI IMS) is a powerful tool for comprehending the spectrum of peptides/proteins expressed in tissue sections. The aim of the present study was to investigate, using MALDI IMS, the proteome of hepatocellular carcinomas (HCC) and to compare it with peritumoral cirrhosis so as to characterize new biomarkers of HCC. Frozen liver tissues corresponding to HCC and background cirrhosis (n = 30) were selected and subjected to MALDI IMS. We found a set of proteins/peptides with a differential intensity level that most accurately delineated cancer from adjacent cirrhotic tissue. Using a support vector machine algorithm, we generated a classification model in the train set that enabled segmenting images from the independent validation set and that in most cases matched histologic analysis. The most discriminating peak (m/z 8565) more intense in HCC was characterized as the monomeric ubiquitin. An immunohistochemical study in a large series of HCC/ cirrhosis sampled on tissue microarray supported that ubiquitin was overexpressed in HCC. We demonstrated also that this increase was not related to an upregulation of ubiquitin gene transcription in HCC, thus suggesting a post-transcriptional mechanism. This approach might provide a new tool for diagnosis of difficult HCC cases and an opportunity for identifying candidate biomarkers. KEYWORDS: MALDI imaging, biomarker, liver cancer, cirrhosis, classification, ubiquitin
’ INTRODUCTION Hepatocellular carcinoma (HCC) represents approximately 6% of all new cancers diagnosed worldwide. It is one of the most deadly malignancies and the third most frequent cause of death due to cancer among men.1 Over 80% of HCC develop from cirrhosis, mainly related to chronic hepatitis B and C virus infections.2 Since metabolic syndrome is also a cause of liver cancer, its high prevalence in the general population might contribute to an increase in the prevalence of HCC.3 HCC may thus become a major threat in coming years, and attempts at developing new diagnostic tools are justified. Proteomics could represent an interesting approach, via identification of new proteins/peptides specifically expressed in fluids or cancer cells; indeed, several studies have already been successfully performed on HCC.4 7 Some of them used proteins extracted from biological samples including tissue and comparative 2-D PAGE or 2-D DIGE analysis followed by mass spectrometry for molecular identification. Unfortunately, this approach r 2011 American Chemical Society
does not take into account tissue heterogeneity unless careful microdissection of tumor cells is performed. Furthermore, all these procedures require protein extraction protocols that might induce artificial protein modifications, thus compromising the physiological relevance of results. Direct analysis of tissue sections using MALDI IMS (matrixassisted laser desorption ionization imaging mass spectrometry) is a new approach that avoids this technical drawback. It uses protein mass to obtain biomolecular profiles that can be associated with specific histological features; thus, it does not require protein extraction or labeling.8 12 In brief, a thin tissue section is placed on a target slide and mass spectra are systematically recorded point by point throughout the tissue. The intensity of each m/z value (hereafter referred to as “peak”) over the array of pixels can be expressed as a 2D ion intensity map and generate images depicting localizations and relative intensities of different Received: April 21, 2011 Published: June 15, 2011 3755
dx.doi.org/10.1021/pr200372p | J. Proteome Res. 2011, 10, 3755–3765
Journal of Proteome Research
ARTICLE
For each case, one representative block of tissue sample with a mean area of 1 cm2 was selected for MALDI IMS. The block was cryosectioned and 10 μm thick sections were carefully placed on conductive indium tin-oxide-coated glass slides (Bruker Daltonics, Bremen, Germany), dried under a vacuum, briefly washed in 70 and 100% ethanol, dried again and directly covered with the matrix. The MALDI matrix was applied using the ImagePrep station (Bruker Daltonics) with standard protocol. The matrix was sinapinic acid at 10 mg/mL in water/acetonitrile 40:60 (v/v) with 0.2% trifluoroacetic acid. MALDI IMS Experiments
Figure 1. Overview of MALDI IMS workflow.
peaks above the histological image. Since this technology analyzes intact tissue, thereby avoiding homogenization and purification steps, the spatial distribution of molecules within the tissue is preserved and the risk of nonrelevant preanalytical protein alteration is reduced. Although imaging proteomics bears significant technical limitations, it is a rapidly expanding technique due, in particular, to refining of effective protocols for sample preparation, analysis of biomolecular data sets and integration of results into histological images.8,13 21 In the present study, we found specific peaks, the differential intensity of which accurately delineated cancer from adjacent cirrhotic tissue. We generated a classification model discriminating between HCC and the peritumoral area in tissue sections of an independent validation group that matched histologic analysis. This tool might be very useful for liver pathologists, since distinguishing well-differentiated HCC from cirrhotic macronodules is becoming increasingly difficult in clinical practice.22
’ EXPERIMENTAL SECTION Tissue Samples
We retrospectively retrieved surgical resections or transplantations for HCC in cirrhotic livers at our center. The etiology of chronic liver disease was assessed according to standard clinical, biological, serological and virological tests. We selected 20 cases of resected HCC for which frozen tissue blocks of both tumoral and nontumoral tissue were available. These cases made up the train set. For the validation study, we prospectively included 10 new cases of HCC in cirrhosis. For each of these, a tissue fragment that included both tumoral and nontumoral tissue was carefully sampled at the HCC margin and stored at 80 C until use for the validation study. Serial hematoxylin and eosin (H&E)-stained sections of frozen tissue were consistently used to judge the adequacy of the frozen material. For each case, standard pathological analysis of HCC was performed. Study protocols were in conformity with ethical guidelines of the 1975 Declaration of Helsinki and approved by the local institutional review board and ethical committee. All subjects gave informed consent for participating in the study. Sample Preparation for MALDI IMS
A workflow of the complete process is shown in Figure 1.
The MALDI measurement was performed on an Autoflex III MALDI-TOF/TOF mass spectrometer with a Smartbeam laser using FlexControl 3.0 and FlexImaging 2.1 software packages (Bruker Daltonics). Ions were detected in positive linear mode at a mass range of m/z 2000 20 000 with a sampling rate of 0.1 GS/s. The lateral resolution (distance between raster points) was set to 200 μm and a total of 500 laser shots were accumulated per pixel at constant laser power. A ready-made protein standard (Bruker Daltonics) was employed for calibration of spectra, which was done externally on the same target before each measurement. Peak Selection for Identification
For each of the 20 cases in the train set, mass spectra were extracted from regions of interest (ROIs) corresponding to the entire areas of tumor and cirrhosis previously defined in FlexImaging 2.1. Then the “peak statistic calculation” was performed in ClinProTools 2.2, for each sample, by loading the two classes (tumor and cirrhosis). Thus, comparative analysis of the whole set of spectra of the tumoral vs nontumoral tissue samples was performed to generate a list of differentially expressed peaks. A peak was considered differentially expressed for a p-value
