Direct cancer tissue proteomics: a method to identify ... - Nature

Oncogene (2007) 26, 65–76

& 2007 Nature Publishing Group All rights reserved 0950-9232/07 $30.00 www.nature.com/onc

ORIGINAL ARTICLE

Direct cancer tissue proteomics: a method to identify candidate cancer biomarkers from formalin-fixed paraffin-embedded archival tissues S-I Hwang1,4, J Thumar1,4, DH Lundgren1, K Rezaul1, V Mayya1, L Wu1, J Eng2, ME Wright3 and DK Han1 1

Department of Cell Biology, Center for Vascular Biology, University of Connecticut School of Medicine, Farmington, CT, USA; Fred Hutchinson Cancer Research Center, Seattle, WA, USA and 3University of California Davis Genome Center, Davis, CA, USA

2

Successful treatment of multiple cancer types requires early detection and identification of reliable biomarkers present in specific cancer tissues. To test the feasibility of identifying proteins from archival cancer tissues, we have developed a methodology, termed direct tissue proteomics (DTP), which can be used to identify proteins directly from formalin-fixed paraffin-embedded prostate cancer tissue samples. Using minute prostate biopsy sections, we demonstrate the identification of 428 prostate-expressed proteins using the shotgun method. Because the DTP method is not quantitative, we employed the absolute quantification method and demonstrate picogram level quantification of prostate-specific antigen. In depth bioinformatics analysis of these expressed proteins affords the categorization of metabolic pathways that may be important for distinct stages of prostate carcinogenesis. Furthermore, we validate Wnt-3 as an upregulated protein in cancerous prostate cells by immunohistochemistry. We propose that this general strategy provides a roadmap for successful identification of critical molecular targets of multiple cancer types. Oncogene (2007) 26, 65–76. doi:10.1038/sj.onc.1209755; published online 26 June 2006 Keywords: direct tissue proteomics (DTP); LC-MS/MS; prostate cancer; Wnt; formalin-fixed paraffin-embedded tissues; prostate-specific antigen (PSA)

Introduction One of the most important challenges in the fight against cancer is the ability to detect cancer cells early in the disease (Parnes et al., 2005; Posadas et al., 2005; Smith et al., 2005; Wardwell and Massion, 2005). To achieve this overall goal, new and innovative technologies that will allow detection of early stages of cancer cells Correspondence: Dr DK Han, Department of Cell Biology, Center for Vascular Biology, University of Connecticut School of Medicine, 263 Farmington Avenue, Farmington, CT 06030, USA. E-mail: [email protected] 4 These authors contributed equally to this work. Received 9 December 2005; revised 7 April 2006; accepted 24 April 2006; published online 26 June 2006

sensitively and accurately are needed (Parnes et al., 2005; Posadas et al., 2005; Smith et al., 2005; Wardwell and Massion, 2005). This goal is central to reducing most cancer-associated deaths as the available cancer drugs and treatment procedures can lengthen the lifespan of most cancer patients if the disease is detected early (Posadas et al., 2005; Wardwell and Massion, 2005). A major shortcoming associated with the early detection and treatment of cancer is the lack of sensitive and robust technology to detect the signatures of cancer cells from minute quantities of available tissues or serum. The lack of technological platform has significantly slowed the identification of reliable biomarkers to accurately diagnose most types of cancers (Zangar et al., 2004; Posadas et al., 2005; Sweat, 2005; Tarro et al., 2005; Wardwell and Massion, 2005; Wright et al., 2005). Defining the molecular mechanisms that give rise to the cancer phenotype is also believed to represent a critical step in developing an effective therapeutic regimen for cancer patients (Tanneberger, 1977; Sawyers, 2002; Raben and Helfrich, 2004; Zangar et al., 2004; Meyerson and Carbone, 2005). Thus, effective treatment will require specific genotyping of expressed genes or proteins in the cancerous tissue. Once the expression profile associated with the underlying pathogenesis of the cancer is determined, one can presumably select a treatment regimen that is best suited for a specific type of cancer (Tanneberger, 1977; Sawyers, 2002; Raben and Helfrich, 2004; Zangar et al., 2004; Meyerson and Carbone, 2005). A number of large-scale expression studies characterizing both the levels of mRNA and proteins expressed in normal and cancerous tissue have been undertaken with the goal of identifying unique expression signatures associated with the cancer phenotype (Sawyers, 2002; Zangar et al., 2004; Meyerson and Carbone, 2005; Parnes et al., 2005). Although a wealth of information exists on the expression signature of mRNAs expressed by a specific type of cancer, very little data are available about the protein expression signatures that exist in normal and cancerous prostate tissue (Gygi et al., 1999; Nelson et al., 2000; Griffin et al., 2003; Wright et al., 2005). Although proteomic technology has been proposed as a general tool for the discovery of cancer biomarkers

Direct tissue proteomics analysis of prostate cancer-expressed proteins S-I Hwang et al

66

that will lead to new diagnostic and therapeutic targets of cancer, the promise has not been fully realized to date (Wilson, 2004; Wright et al., 2005). For example, the proteomic methods referred to as surface-enhanced laser desorption ionization mass spectrometry (SELDI-MS) and matrix-assisted laser desorption ionization time-offlight mass spectrometry (MALDI–TOF-MS) have been used to identify protein expression patterns directly from cancer tissues or serum (Petricoin et al., 2002). Although these methods can provide potential protein peak patterns that are statistically correlated with cancer phenotype, the exact chemical identity of protein biomarkers is never routinely identified. Additional protein enrichment and identification steps are necessary to unambiguously identify potential biomarkers or therapeutic targets. Once these biomarkers are conclusively identified, a more sensitive immunological or enzyme-based assay can be developed to accurately determine and measure the molecule of interest in normal and disease samples (Patterson, 2004; Zangar et al., 2004; Wright et al., 2005). Thus, deciphering the chemical identity of protein biomarkers is necessary if more robust and sensitive tests are to be developed to detect, monitor and possibly target these new protein biomarkers in cancer (Patterson, 2004; Zangar et al., 2004; Wright et al., 2005). In this report, we have employed a shotgun proteomics approach, which provides the chemical identity of proteins in cells, tissues and fluids. We have coined this strategy as direct tissue proteomics (DTP), as direct protein identification by tandem mass spectrometry (MS/MS) from tissues, we presumed, would allow detection of diagnostic biomarkers and therapeutic targets (Wolters et al., 2001; Cantin and Yates, 2004). Using this strategy, we have tested whether potential diagnostic biomarkers and/or therapeutic targets can be directly identified from the prostate cancer tissues. This method utilizes the advantages of shotgun proteomics and combines with a new protein extraction procedure that disrupts the crosslinked proteins from the previously formaldehyde-fixed paraffin-embedded tissue samples. This analysis led to the identification of over 400 proteins from prostate cancer biopsies, in which prostate-specific antigen (PSA) was quantified at picogram (pg) levels. We also detected expression of Wnt-3 as a possible regulator of prostate cancer cell proliferation. We discuss the broad utility of applying the DTP method to preventive medicine and the early detection of prostate cancer.

Results Development of DTP technology The overall goal of this study is to determine if proteins can be conclusively identified from small quantities of biopsy tissue samples from clinically relevant prostate cancers. This study addressed three main questions: (1) how many proteins can be conclusively identified in small quantities of prostate cancer biopsy tissues using Oncogene

the shotgun proteomic method?, (2) can a current prostate cancer protein biomarker, such as PSA, be robustly identified using this approach? and (3) can additional proteins involved in prostate tumorigenesis also be detected using this method? More importantly, we asked whether a general methodology could be developed that would allow for robust protein identification directly from archival tissue samples that were previously formaldehyde-fixed and paraffin-embedded. We believe that the DTP methodology provides an opportunity for investigators to further interrogate the expression levels of disease-associated proteins in large volumes of archived tissues that represent various stages of pathological conditions. This ability to conclusively identify proteins from archival tissues will clearly be an important technological advancement to identify new diagnostic and prognostic proteins biomarkers. As a proof-of-principle demonstration, commercially available prostate tissue arrays were used for the DTP method development. The tissue array contained five normal and 25 cancer biopsy sections in duplicates, providing a reasonable number to examine the feasibility of the method. Furthermore, all of the biopsy sections were 4 mm thick and 2 mm in diameter in dimension, and thus, allowed us to compare unique proteins that can be conclusively identified by proteomics methodology. The overall outline of our study is shown in Figure 1a. In order to assess the usefulness of the proteomic study, we first carefully examined and categorized the prostate cancer according to the Gleason scoring criteria (Gleason, 1988, 1992). Characteristic cellular morphology associated with each of the stages of carcinogenesis was examined in the entire region of the tissue and two most representative Gleason scores were assigned (Supplementary Figure 1). Once the Gleason scoring was completed, as shown in Figure 1b and Supplementary Figure 1, we subdivided all of the tissues into four broad categories: (1) normal prostate tissue consisting of biopsy samples from five patients, (2) low-grade cancer consisting of eight biopsy tissues with Gleason scores of 2–5, (3) medium-grade cancer consisting of 26 biopsy tissues from Gleason scores of 6–7 and (4) high-grade cancers comprising 16 biopsy tissues that showed the Gleason scores of 8–10. Using these established criteria, from a tissue array containing five controls and 25 cancerous prostate tissues, we tested the feasibility of proteomic analysis. In contrast to the MALDI and SELDI peak profiling proteomic approaches, our goal was to identify proteins conclusively from these biopsy tissues. To initiate our analyses, all steps of the DTP method were optimized: (1) reversing the paraformaldehyde crosslinks from the tissues biopsies, (2) trypsin digestion of tissues, (3) separating the tryptic peptides using reverse-phase chromatography and (4) protein identification using the m-LC–MS/MS procedures previously optimized for complex tryptic mixtures. The schematic outline of the DTP procedure is shown in Figure 1a and c. After a series of optimization steps, we developed a procedure that yielded successful identification of a large number of peptides from the prostate tissue array slide.


67

Figure 1 Overall strategy for the DTP method. (a) A schematic flow diagram depicting the steps for proteomic identification, quantification and validation of prostate tissue arrays. (b) A representative H&E image from each of the four categories of Gleason scoring is shown. (c) Schematic diagram of the prostate tissue array slide used for DTP procedure. The array slide contains 60 tissue biopsies and each of the biopsy sections are 4 mm thick and 2 mm in diameter. The tissue samples are arranged in six columns (c, from left to right) and each column contain 10 sections (1–10, from top to bottom) and contain two representative needle biopsy sections from each patient.

Details describing optimization of the DTP method are contained in the Materials and methods. Two tissue arrays processed in replicate led to the generation of thousands of MS/MS spectra. Database searching of all of these spectra using the SEQUEST followed by stringent data filtering coupled with a decoy database search to establish low false-positive identifications resulted in the successful identification of 12 631 peptides. Subsequent data analysis grouping the peptides into unique proteins using a suite of bioinformatics software tools resulted in a list of 428 unique proteins with high confidence identification. Bioinformatics characterization of identified proteins from prostate tissue array A major strength of the DTP method is that proteins specifically expressed in distinct Gleason categories can be identified using the appropriate bioinformatic software. Thus, we implemented the bioinformatic tools INTERSECT and PROTEOME-3D (Lundgren et al., 2003). INTERSECT allowed us to sort proteins common and unique between the four Gleason categories (Figure 2a and Supplementary Table 1). Based on our filtering criteria, we observed 18 proteins in the

control group, eight proteins in the low-grade, 77 proteins in the medium-grade and 60 proteins in the high-grade cancers (Figure 2a). In addition, the INTERSECT tool allowed the categorization of proteins that are detectable only in two or three specific categories, such as 46 proteins common only to the medium- and high-grade prostate cancers. Although we cannot conclusively categorize the Gleason stage-specific expressed proteins owing to the fact that we have not identified proteins comprehensively from these tissues, bioinformatics categorization of these proteins allowed us to narrow the list for validation experiments. Next, we utilized the PROTEOME-3D software tool, allowing the user to analyse large-scale experimental data sets in the context of publicly available knowledgebase (Lundgren et al., 2003). Specifically, this software tool automatically downloads user-defined information of the identified proteins for in-depth functional and pathway analysis. Thus, we loaded the identified proteins onto this software tool, and examined the proteins that are involved in distinct metabolic pathways. As shown in Figure 2b, we were able to assign the identified proteins into 24 broad functional categories according to the Gene Ontology (GO) classification. As dysregulation of energy metabolism has long Oncogene


68

Figure 2 Bioinformatics characterization of identified proteins from the prostate tissue arrays. (a) Venn diagram showing the number of identified proteins from each of the four Gleason categories. Proteins common to some categories are not shown in the figure: control and high-grade prostate cancers, 10; medium and low-grade, 6. (b) Functional characterization of identified proteins into 24 GO categories. (c) Classification of identified proteins in energy-generating metabolic pathways. The classification table is shown in the right panel. Identified metabolic enzymes are highlighted in blue and marked with the classification number.

been implicated in the pathogenesis of prostate cancer, we next grouped the identified proteins into energygenerating metabolic pathways. From these analyses, key enzymes from five major pathways were found to be associated with cancer cells (Figure 2c). For example, among the six pathways that were analysed, preferential identification of enzymes that are critical for gluconeogenesis were detected in the low- and medium-stage prostate cancers (Figure 2c). In addition, metabolic enzymes associated with the tricarboxylic acid cycle were detected in multiple stages of the prostate cancers (Figure 2c). In contrast, a large majority of the glycolysis enzymes were found in normal tissue as well as in the cancerous prostate tissues. However, phosphoglucose isomerase, phosphoglycerate kinase and phosphogylceromutase enzymes were detected in the prostate cancers. Detection of known markers of prostate cancer cells including the PSA and the measure of its relative abundance We next examined whether the most widely used prostate antigen, PSA, can be detected by the DTP method. We carefully examined the presence of tryptic peptides from PSA in the 1.5 million spectra generated from the fragmentation of mass spectrometry peaks. Oncogene

Towards this goal, we made use of the INTERACT software tool, which allows the sorting of one protein from a large number of MS/MS spectra generated from over 120 LC–MS/MS analyses. Sorting for the KLK3_HUMAN (Kallikrein 3/p-30 antigen/PSA precursor) resulted in the identification of 214 tryptic peptides from the PSA with up to 61% peptide coverage of the protein (Figure 3a and b). MS/MS analyses and the resulting SEQUEST scores revealed that approximately half of the identified peptides were above the cutoff scores (Figure 3d; please see Materials and methods). Consistent with the known expression of PSA in normal prostate and cancerous prostate tissue, we were able to identify tryptic PSA peptides in all four states of the prostate tissues. In order to get a sense of relative abundance of PSA in these samples, we analysed the number of independent identifications in each of the stages of the prostate tissue. As shown in Figure 3c, normalizing the number of independent identifications of PSA as a measure of abundance revealed that the highest level of PSA was detected in the medium-stage prostate cancers (Materials and methods). These results establish that the DTP strategy can conclusively identify the most well-established diagnostic marker from minute quantities of previously formalin-fixed paraffinembedded tissues.


69

Figure 3 Identification of PSA. (a) MS/MS spectrum of a PSA peptide, its amino-acid sequence, theoretically predicted fragment ion series and experimentally determined fragment ion series are shown. Identified b and y ions are indicated with red and blue colors, respectively. (b) Peptide sequence coverage of PSA protein. Each of the independently identified tryptic peptides are highlighted with red. Identified peptides are also alternatively underlined to denote consecutively identified peptides. The peptide LSEPAELTDAVK is highlighted in blue. The MS/MS fragment ions from this peptide is shown in (a) and two of the fragment ions were used for the multiple-ion monitoring experiment as shown in Figure 4. (c) Estimation of the relative abundance of PSA peptides in normal and cancerous prostate cancer biopsies. Independently identified PSA peptide numbers were divided by the number of a-actin peptides to have a measure of PSA protein abundance as described previously. (d) Variation in the SEQUEST cross-correlation scores seen during the identification of PSA protein. Up to top five scores for each of the peptides are shown.

We next tested the limit of sensitivity of our methodology using PSA protein as a known biomarker. To achieve this goal, we utilized the strategy outlined by Steven Gygi and colleagues termed absolute quantification (AQUA) (Gerber et al., 2003). This methodology utilizes a standard peptide with known quantity to compare against a biological sample to establish the absolute quantity of protein in the mixture. To quantify PSA in the prostate cancer tissue, we synthesized a PSA standard peptide (LSEPAELTDAVK) using the heavyisotope-containing amino acids (Figure 4a), spiked-in 100 femptomole (fmol) of standard peptide in the tryptic digested tissue sections and quantified the amount of PSA in normal controls and in the cancerous prostate tissues. As shown in Figure 4b–d, multiple-reaction monitoring (MRM) analyses of PSA standard and endogenous peptide fragment ions revealed that endogenous PSA peptide from the prostate cancer tissue samples co-eluted with the standard peptide. The range of PSA quantified directly from the tissues was 0.5–140 pg. These results demonstrate that a known biomarker

protein such as PSA can be sensitively quantified in needle-biopsy samples of cancer tissue. However, it is known that serum PSA levels do not always predict the presence of prostate cancers (Ford et al., 1985; Coombs et al., 1998; Gupta et al., 2004). It is anticipated that proteomic techniques can provide the identification of additional protein biomarkers that may be used in conjunction with PSA to increase the specificity and sensitivity in prostate cancer detection. Thus, we examined the list of identified proteins to search for additional proteins that have been implicated in prostate carcinogenesis using a number of bioinformatics tools. Our first analysis utilized the PROTEOME-3D tool (Lundgren et al., 2003). Specifically, we examined the presence of proteins from three broad categories: androgen responsive and androgen receptor regulators, known oncoproteins and angiogenic and stromal-associated proteins (Supplementary Table 3). In order to have a sense of abundance of these proteins in the analysed tissue samples, we extracted the number of times the peptides from these proteins were Oncogene


70

Figure 4 Absolute quantification of PSA from prostate biopsy tissue sections. (a) Schematic diagram demonstrating method development phase of the experiment. A PSA peptide was synthesized with deuterium-labeled valine (8 Da heavier than normal valine). (b) Quantification phase of the experiment for measuring the absolute quantities of PSA in the prostate biopsy sections. Two fragment ions from the standard peptide and the endogenous PSA peptide were used for the MRM experiment. (c) Extracted ion chromatograms of standard PSA peptides (black) and endogenous peptides (highlighted) from normal control prostate (blue) and three cancer grades (green, orange and red) are shown. Quantified values in fmols were calculated from the AUC (area under the curve) values using the formula shown in (b). (d) Quantification values of PSA from a total of five normal samples and 15 cancerous samples are shown. The values in fmols are converted to pg amounts. Prostate biopsy sample numbers that were quantified and the respective quantification values are indicated in the lower legends.

independently identified in normal prostate tissues and three Gleason categories of low, medium and high (Supplementary Table 3). We found a number of proteins previously implicated in prostate cancer progression, and the peptide counts of these proteins suggest possible overexpression of these proteins during prostate carcinogenesis. One of the proteins that we have chosen to further validate is the Wnt-3 protein, which was identified with multiple peptides although most of these peptides were identified with lower SEQUEST cross-correlation scores (Supplementary Figure 2a–d). We chose Wnt-3 as the Oncogene

Wnt family of proteins is known to cause oncogenic transformation in a number of cell systems including the prostate cells (Nantermet et al., 2004; Verras et al., 2004; Zhu et al., 2004; Cronauer et al., 2005). Recently, Wnt3-related Wnt-3a protein was shown to support the androgen-independent growth of LNCaP prostate cancer cells (Nantermet et al., 2004; Verras et al., 2004; Zhu et al., 2004; Cronauer et al., 2005). Furthermore, the Wnt proteins are secreted protein ligands for cell surface receptors of the frizzled and lipoprotein receptor-related protein family (Nantermet et al., 2004; Verras et al., 2004; Zhu et al., 2004;


71

Cronauer et al., 2005). Thus, the possibility that these proteins might be detectable in serum as a diagnostic marker along with PSA becomes a strong possibility. Thus, Wnt-3 protein was further validated using immunohistochemistry on the prostate cancer tissue arrays. We examined the expression of Wnt-3 protein in the normal prostate tissues, and in the low-, medium- and high-grade cancers. As shown in Figure 5a, Wnt-3 protein expression was low in normal glands of prostate with strong immuno-reactivity restricted to a few clusters of basal epithelial cells of the prostate glands. No detectable staining was observed when the primary antibody against Wnt-3 protein was omitted (Supplementary Figure 3a). In contrast, significant upregulation of Wnt-3 protein was detectable in the luminal epithelial cells of the prostatic intraepithelial neoplasia (Figure 5b). In more advanced prostate cancers, strong immunoreactivity was seen in the neoplastic and invasive cells (Figure 5c and d). In our immunohistochemistry experiments, we found observable differences in background staining (Figure 5 and Supplemental Figure 3). In these experiments, we performed the staining of all of the spots on the same slide, thereby, effectively removing the variability owing to the antigen retrieval or the differences in staining efficiency. Therefore, the consistent and observable differences in background stroma staining can be attributed to differences in fixation conditions and/or to the presence of secreted Wnt-3 in the prostate tissue. Scoring the intensities from control and cancerous prostate sections (low, medium

and high grades) revealed approximately two-fold stronger immunoreactivity in the medium- and highgrade prostate cancers (Table 1). Quantification of background-to-signal ratios from 200 representative regions using the Image J software further confirmed that approximately two-fold increase in staining intensities was seen in the cancerous glands of the prostate. These results confirm Wnt-3 protein expression in the prostate cancer cells and support the idea that DTP approaches can provide additional protein targets that may participate in prostate carcinogenesis. Additionally, these results demonstrate that the DTP method can be used to identify proteins from the source cancer tissues that may possess diagnostic, prognostic and therapeutic value against cancer. Discussion Our overall goals for this study are as follows: (1) to prove that protein identification directly from archival cancer tissues (previously formalin-fixed and paraffinembedded) is feasible; (2) to demonstrate that a commonly used prostate cancer biomarker (PSA) can be detected and quantified by this method; and (3) to demonstrate that this method allows identification of biologically interesting proteins such as Wnt-3. Indeed, the primary focus of this study is not about the discovery of novel biomarkers; every potential biomarker requires a thorough validation step using a large number of cancer tissue samples. Interestingly, we found

Figure 5 Validation of Wnt-3 expression in the prostate cancer samples. (a) Basal epithelial cells from normal prostate glands specifically react with anti-Wnt-3 antibody as shown by brown peroxidase reaction product. (b) Luminal cells of the prostate intraepithelial neoplasia showing positive staining for Wnt-3. (c and d) Strong Wnt-3 staining in neoplastic glands with Gleason pattern 3 (c) and pattern 5 (d). Magnification 40. Oncogene


Oncogene

1 2

Abbreviation: IHC, immunohistochemistry. aFive patients from each grade of cancer and normal glands were taken for this IHC analysis. Eight serial sections from each individual were examined by Wnt-3 immunohistochemistry. bThe number of glands cannot be accurately determined owing to the fusion of adjacent glands and the presence of invasive epithelial cells between glands. c Semiquantitative grading of IHC staining: grade 0 ¼ no staining, grade 1 ¼ weak staining, grade 2 ¼ moderate staining, grade 3 ¼ strong staining.

2 2 2 1 1 2 2 2.2 2.2 3 1 1 1 1 1 0.8

0 0 0 0 0 6 5 7 7 10 3 3 2 3

2

10 12 14 24 30 4 5 5 6 10 1 1 2

3

0.8

b b

3

0.8

5 4

3 10 2 20 5 10 4 12 3 14 2 24 2 4

Patientsa No. of glands examined from eight serial sections No. of glands stained with Wnt3 IHC staining No. of glands negative for Wnt3 IHC staining Average grade of stainingc

1 6

3 5

4 4

5 4

1 20

2 13

3 12

4 10

5 10

1 30

Medium-grade prostate cancer (Gleason score 6–7) Low-grade prostate cancer (Gleason score 2–5) Normal prostate glands Semiquantitative analysis of IHC of Wnt3 at various stages of prostate cancer

Table 1 Intensity scoring of Wnt-3 immunoreactivity in normal and cancerous prostate biopsy sections

1 20b

High-grade prostate cancer (Gleason score 8–10)

72

that a significantly large number of proteins can be identified from minute quantities of previously formalinfixed paraffin-embedded tissues. One significant finding is that we were able to identify and quantify PSA directly from the tissue array samples using the AQUA strategy. Another significant finding from our study is the identification and validation of Wnt-3 protein in human prostate cancer cells. These results suggest that even though the starting material is very small, the DTP strategy may provide novel protein biomarkers or therapeutic targets, in addition to the high-abundant proteins that are present in the prostate and prostate cancers. From the methodological standpoint, this study represents a proof-of-principle demonstration for protein identification directly from archival prostate cancer tissue. It is conceivable that large-scale genomic and proteomic characterization of cancer tissues will provide novel insights into the early detection and effective treatment of this disease. Concerted effort to tackle the cancer genome and identify key genes and proteins that are responsible for the pathogenesis has now been proposed (Wright et al., 2003; Patterson, 2004; Wilson, 2004; Cox et al., 2005; Stephens et al., 2005). Whether or not this daunting mission can be successfully accomplished is not certain. However, if one assumes that there are only limited genetic mechanisms for each tissue- or cell-specific cancer, it may be possible to therapeutically target the underlying pathogenic mechanisms (Sawyers, 2002; Wright et al., 2003, 2005; DePinho and Polyak, 2004; Stephens et al., 2005). It is well known that predominant pathways are involved in the pathogenic mechanisms of cancer in different tissues. For example, cancers of the hormone-sensitive tissues such as breast and prostate tissues are sensitive to therapeutic mechanisms associated with the hormone receptors. In addition, in a number of lymphoid cancers, chromosomal break and translocation of known cancercausing genes next to the highly expressed regions specific to the immune cells are responsible for the pathogenic mechanisms of these cancers (Sawyers, 2002; Wright et al., 2003; DePinho and Polyak, 2004; Stephens et al., 2005; Wright et al., 2005). Thus, it may be possible to fully characterize predominant pathways that control the cancer development of cells from each tissue type. Comprehensive genomic and proteomic characterization of cancer cells and tissues and quantification of biomarkers and predominant pathway components may provide a foundation for the overall goal of detecting and treating cancer effectively. We believe that DTP method will allow the identification of molecular targets associated with pathologically well-defined regions of cancer and other human diseases. There are now efforts to systematically search for common mutations in 12 500 tumor samples from 50 major cancer types. Furthermore, efforts to sequence the coding regions of approximately 2000 known genes that have been implicated in cancers have also been recently proposed. These efforts, when completed, will likely yield important information about the common muta-


73

tions that are present in major cancer types and may provide insights into the underlying mechanisms for each of the cancer types (Sharpless and DePinho, 2005; Tonon et al., 2005). Complementing this effort will be the characterization of the proteome complement of these cancer types, as the proteome from each of the cancer types will provide additional mechanistic insights into the consequences of genetic mutations. For example, mutations in oncogenes or tumor suppressor genes may cause dominant-positive or -negative effects (George, 2002; Sawyers, 2002; Raben and Helfrich, 2004; Cox et al., 2005; Meyerson and Carbone, 2005; Sharpless and DePinho, 2005). These effects may be exhibited by compensatory up- or downregulation of key proteins. Thus, proteomic identification of these proteins may help explain key pathways that are activated in each cancer type and DTP method will be useful in this regard. Furthermore, some proteins that are activated as a consequence of gene activation/inactivation may be used as surrogate biomarkers for early cancer detection (Wilson, 2004; Wright et al., 2005). We have systematically compared the identified protein list from this study with a number of published cDNA microarray data sets (Velasco et al., 2004) and found very little overlap between the regulated mRNAs and identified proteins. These results suggest that more comprehensive characterization of proteomes and mRNAs from normal and cancerous prostates is required for meaningful comparative efforts. Compounding this effort is also the fact that cancer tissues are heterogeneous in nature and their cellular to stroma ratios can be very different in distinct areas. Thus, comparing two normal or tissue samples may not be meaningful unless careful efforts are made to normalize the cells and matrix components. In addition, most proteomic shotgun proteomic experiments suffer from under-sampling of the expressed proteins, as the complexity of the expressed proteins, especially from tissue samples, is huge. Moreover, the duty cycle of the

latest mass spectrometers are still relatively slow, and thus, cannot identify most of the expressed proteins that are detectable by the mass spectrometer. Thus, comparative or subtractive proteomics techniques are quite limited and at the present form, could not provide comprehensive list of expressed proteins. However, to achieve these goals, a number of technical challenges have to be overcome in the field of proteomics. One of the challenges is the ability to analyse proteins from relatively pure cell populations. As most cancer cells are mixed with normal cells, it is important to develop methodologies to purify or select for cancer cells for proteomic or genomic analysis. In addition, it is also important to develop methods so that one can identify proteins comprehensively, including the low-abundant proteins and membrane proteins, as key enzymes and regulators, although in low abundance, can have a major impact on the growth of the cell. Thus, additional methodological improvements are needed to select pure cancer cell populations from mixed tissue samples and to identify proteins comprehensively, including the low-abundant proteins and membrane proteins. A related issue also is the ability to extract most of the proteins from previously formalin-fixed paraffin-embedded tissue samples. During the preparation of our manuscript, David Krizman’s group published a paper describing the proteomic analysis of formalin-fixed prostate cancer tissue (Hood et al., 2005). In contrast to our paper, owing to the use of commercial kit, no detailed information regarding the exact buffer composition was given in recent paper (Hood et al., 2005). We have continued to test different buffer conditions for the DTP method to improve and have a sense of extraction efficiency by comparing protein yield before and after the formaldehyde fixation and found that our current method using the acetonitrile buffer (30% acetonitrile, 100 mM ammonium bicarbonate) can extract from 13 to 42% of the total extractable proteins (Table 2). We compared five different buffer conditions

Table 2 Analysis of protein recovery after paraformaldehyde fixation Before paraformaldehyde fixation Buffer A Exp. Exp. Exp. Exp.

I II III IV

Exp. V

24.472.5% 24.5% 51.5% N/Aa Buffer A 47.4%

Buffer D

After paraformaldehyde fixation Buffer A

Buffer D

100% 2275.8% 100% 21.58% 100% 42.3% N/Aa 13.5% b Prostate tissue array sample

83720% 100% 90.9% 100%

Buffer D 100%

Buffer Ec 177.2%

Abbreviation: BCA, bicinchoninic acid. Quantitative protein recovery from five independent experiments is shown. Jurkat T-lymphoma cell line was used for quantification of protein recovery for experiments I and II. Freshly isolated mouse kidney was used for experiment III. Human coronary artery previously fixed with paraformaldehyde and embedded in paraffin was used for experiment IV. As freshly isolated human coronary artery before paraformaldehyde fixation is not available for comparison, Buffer D-extracted protein quantity was assumed to be 100% and compared with buffer A. The extracted proteins were quantified by the use of BCA protein assay (PIERCE Biotechnology, Rockford, IL, USA) and densitometric quantification of Coomassie-stained gel bands from SDS–PAGE gels. Experiment I was performed in triplicate to assess the variability. aN/A, not analysed as this is previously formalin-fixed paraffin-embedded archival coronary artery sample. bExp. V: Proteins were extracted with three different buffers from prostate tissue array samples and run on SDS–PAGE gel. The gel was stained with silver nitrate and then quantified by densitometry and Image J software. cBuffer E: 0.1% RapiGest (Waters, Milford, MA, USA) in 50 mM NH4HCO3 at 941C for 30 min. Oncogene


74

and found an optimal buffer and extraction conditions for Buffer D: 2% sodium dodecyl sulfate (SDS)containing radioimmunoprecipitation assay (RIPA) buffer (150 mM NaCl, 10 mM Tris-HCl pH 7.2, 2% SDS, 1% Triton X-100, 1% deoxycholate, 5 mM ethylenediamenetetraacetic acid (EDTA) at 941C for 30 min and at 601C for 3 h). As shown in Table 2, comparison of freshly isolated cell line or mouse tissue with or without the paraformaldehyde fixation, followed by extraction with either Buffer A (acetonitrile buffer) or Buffer D revealed that 80–100% of the proteins can be extracted by Buffer D. We have used acetonitrile buffer in this study as this buffer is compatible with trypsin digestion and direct LC–MS/MS analysis. This point is critical especially for minute quantities of clinically relevant sample analysis such as thin sections of needle biopsy tissue. However, if larger quantities of tissue samples are available, we recommend the use of 2% SDS-containing RIPA buffer for the DTP method for increased protein yield. Furthermore, we tested the extraction efficiency with another detergent, RapiGest (Waters, Milford, MA, USA), which is compatible with mass spectrometry (Table 2). The extraction efficiency of 0.1% RapiGest in 50 mM NH4HCO3 was found to be B77% higher than Buffer D by densitometric quantification of silver-stained gel. Once the comprehensive information regarding the gene mutation and proteome library is achieved from cancer cells, it will be useful to reconstruct the pathways that are likely amplified or inactivated, and correlate the mechanistic data with the clinical outcome and responsiveness to therapy. It is also conceivable that the identified protein targets that participate in the underlying pathogenic mechanisms will be targeted for patient- and cancer-specific therapeutic strategies to achieve the most beneficial outcome. We believe DTP method will help establish proteomic libraries of major cancer types. This information will be useful for better understanding of the disease pathogenic mechanisms, as well as for successful prevention and treatment of cancer.

Materials and methods Prostate cancer tissue characterization Tissue array sections stained with hematoxylin and eosin (H&E) were carefully examined under the light microscope for the presence of cancerous cells according to the Gleason criteria (Gleason, 1988, 1992). The tissue array contained prostate tissues from five normal individuals and 25 cancer patients. The H&E-stained tissue sections from each of the patients consist of two separate regions of the biopsy tissue, each 4 mm thick and 2 mm in diameter. A combined total of 60 sections were carefully examined under the light microscope, given the representative scores and subdivided into four broad categories: (1) normal prostate tissue, (2) low-grade cancers comprising Gleason scores from 2 to 5, (3) medium-grade cancers comprising Gleason scores from 6 to 7 and (4) highgrade cancers comprising Gleason scores from 8 to 10. This scoring scheme, first reported by Gleason, and since then changed to a new scoring scheme, was chosen, as this outdated Oncogene

categorization was most suitable for our sample size. Characteristic gland types from each of the sections were examined and two predominant scores were given and combined to generate the final Gleason scores. Two of the most representative regions from each of the sections and corresponding Gleason scores are shown (Supplementary Figure 1). Optimization of protein extraction from paraformaldehyde-fixed paraffin-embedded prostate cancer tissue arrays Our first goal was to isolate and identify proteins from previously formaldehyde-fixed paraffin-embedded tissues. Paraffin-embedded tissue arrays were de-waxed by heating for 1 h at 651C followed by the removal of paraffin by three 5-min extractions in 100% toluene at room temperature. Residual toluene was removed from the slides by immersion in 100% ethanol and air-dried in the fume hood for 5 min. We have used the tissue size as a normalization factor as virtually all of these sections were corebiopsy samples with identical dimensions (2 mm rings with 4 mm thickness). Estimation of cell numbers on each spot by manual counting of defined microscopic fields revealed B40 000–60 000 cells/sample. Estimation of protein quantities by pooling two batches of five spots each in Buffer A, followed by protein quantification using the bicinchoninic acid kit revealed that each spot contained 0.63–0.74 mg of extractable proteins. As the sample quantities are limited, fearing sample loss from sub-fractionation using the strong cation exchange method, we have chosen to analyse them directly by liquid chromatography–tandem mass spectrometry (LC–MS/MS) procedure. We typically find that repeated sequencing of the same sample resulted in 70–80% overlap in protein identification between two analyses, whereas much wider overlap is seen when different samples are compared (50–85%). We next tested and optimized a method for successful protein identification directly from the prostate cancer tissue arrays. We chose a buffer system that was optimal to break the formaldehyde crosslinks, re-hydrate the proteins from the tissue arrays, digest the proteins directly by sequencing-grade modified trypsin and directly analyse digested tissues by the m-capillary-LC–MS/MS. Conditions to optimally extract paraffin-embedded proteins and identify them using the shotgun proteomics approach were determined empirically. In short, the following steps were performed: (1) tissue sections were extracted from the glass slides using a razor blade and transferred to a 0.5 ml GenAmp microfuge tube (Applied Biosystems, Foster City, CA, USA); (2) 60 ml Buffer A (100 mM ammonium bicarbonate and 30% acetonitrile) or Buffer D (150 mM NaCl, 10 mM Tris-HCl, pH 7.2, 2% SDS, 1% Triton X-100, 1% deoxycholate, 5 mM EDTA) was added to each tube and heated to 941C for 30 min followed by 601C for 3 h to rehydrate proteins and hydrolyse the formaldehyde crosslinks; (3) samples were then incubated with 1 mg of sequencing-grade modified trypsin for 18 h at 371C (1:20 dilution of Buffer D is required for efficient digestion) and (4) samples were then lyophilized and re-suspended in 12 ml of Buffer B (5% acetonitrile, 0.5% acetic acid, 0.005% heptafluorobuteric acid) and analysed on a Finnigan LTQ-linear ion-trap mass spectrometer coupled to the nano-electrospray source. Data-dependent mass spectrometry analysis The digested tissue samples were loaded onto a self-pack C18 microcapillary using the Famos autosampler coupled in-line to a 1100 Hewlett Packard binary pump connected to the mass spectrometer. Peptide mixtures were eluted into the mass spectrometer over 106 min using an acetonitrile gradient. The details of the


75 methods for the mass spectrometer data acquisition, termed ‘top 6 method’, are described elsewhere (Mayya et al., 2005). Using 60 biopsy tissue spots and performing two independent analyses from two array slides, thousands of MS/MS attempts were performed. The generated data in .RAW format were submitted for database searching using the SEQUEST algorithm (Eng et al., 1994). Identified proteins were filtered, sorted and analysed by a suite of bioinformatic software tools as described below (Han et al., 2001; Rezaul et al., 2005). Data analysis and interpretation Each of the MS/MS spectra generated from fragmentation of the peptides was searched against the locally installed human protein database with over 120 000 entries. Each of the top 10 matches was then listed in the output files. SEQUEST search parameters include the option to identify some post-translational modifications on peptides such as the methionine oxidation ( þ 16), phosphorylation on serine, threonine and tyrosine residues ( þ 80) and the presence of both tryptic ends allowing one missed cleavage. The resulting output data files were grouped into four categories based on the Gleason scoring criteria as described above. All the output files in each of the four categories were further filtered using the INTERACT software tool (Han et al., 2001; Rezaul et al., 2005). In brief, a set of commonly accepted stringent SEQUEST scoring criteria was used to filter the output: Xcorr of 1.9 for þ 1 peptides, 2.8 for þ 2 peptides and 3.7 for þ 3 peptides, and DCn of 0.1 or above. In addition to the SEQUEST scores, protein identification was confirmed by manual inspection of the spectra. Furthermore, statistical significance of peptide and protein identification was assessed using the PeptideProphet and ProteinProphet Software tools as described previously (Han et al., 2001; Keller et al., 2002; Nesvizhskii et al., 2003; Rezaul et al., 2005). In addition, we used the decoy-database searching approach to estimate the level of false-positive peptide assignments (Peng et al., 2003). We used reasonably stringent criteria to eliminate most of the false-positive identifications and we are reporting the list of proteins with approximately 1.1% false-positive rates for multiple peptide containing proteins. Furthermore, we manually validated the single hit proteins. Using these stringent criteria, we identified 428 unique proteins derived form the list of 12 631 identified peptides. Among the proteins that were identified, 361 proteins were identified with multiple peptides and 67 proteins with single peptides. The identified proteins were analysed by the use of PROTEOME-3D software (Lundgren et al., 2003). This tool allows the categorization of proteins into defined metabolic pathways, oncogenes/tumor suppressor genes, function groups, subcellular location and categorization according to the GO terminology. Finally, a stage-specific prostate expression library was generated by the use of INTERSECT software tool (DH Lundgren et al., unpublished data). This tool also allows the selection of candidate prognostic and diagnostic proteins that are expressed in different stages of the cancer tissues for additional validation experiments. Immunohistochemical validation of Wnt-3 expression in prostate cancer cells Briefly, prostate tissue arrays were de-paraffinized with toluene, re-hydrated with graded alcohol washes and subjected to antigen retrieval in a steamer for 30 min in sodium citrate buffer (pH 6.0). The tissue slides were then blocked for the endogenous peroxidase activity using hydrogen peroxide. The slides were then blocked for nonspecific binding using normal goat serum (1:20 dilution in phosphate-buffered saline (PBS), pH 7.4), followed by blocking for endogenous

biotin by using the Victor kit (Burlingame, CA, USA). Slides were then incubated for 1 h with the following primary antibodies diluted in PBS: goat anti-human Wnt-3 antibody (Santa Cruz Biotechnology, Santa Cruz, CA, USA, C-15 antibody catalog No. SC-5210 diluted 1:50), anti-human b-catenin antibody (Santa Cruz Biotechnology, Santa Cruz, CA, USA, C-18 antibody catalog No. SC-1496 diluted 1:50). Cell proliferation status and stromal cell composition of the prostate tissue sections were also examined by immunostaining with anti-proliferating cell nuclear antigen antibody and antismooth muscle a actin antibody (Sigma, St Louis, MO, USA). Each of the immunostaining experiments was carried out together with negative control staining by omission of primary antibody. Slides were then washed, incubated with biotinylated anti-mouse or anti-goat secondary antibody for 30 min at room temperature and incubated with the preformed avidin– biotin–horseradish peroxidase complex. The slides were then developed with the horseradish peroxidase substrate, diaminobenzidine, counter stained with hematoxylin, dehydrated and mounted. Absolute quantification of PSA in normal and cancerous prostate tissues In order to quantify the absolute amounts of PSA in the prostate cancer tissues, we made use of a technique termed AQUA (24). In brief, a tryptic peptide between positions 126 and 137 of PSA (LSEPAELTDAVK) was chosen. The synthetic peptide contained an octa-deuterated valine (Isotec, St Louis, MO, USA) at position 136 and is 8 Da heavier than the endogenous peptide. Accurate quantity of the standard peptide in a suspended solution was measured by quantitative amino-acid analysis. The peptide synthesis and amino-acid analysis were carried out by Molecular Biology Resource Facility at the University of Oklahoma Health Science Center. Optimization steps include acetonitrile gradient for peptide elution and selecting the MRM ions for reproducible quantification (Figure 4). Finally, 100 fmol of standard PSA peptide was combined with the tissue digest, and the mixture was loaded onto a C18 reverse-phase micro-column. The mass spectrometer is operated in the MRM mode to fragment and detect the PSA standard peptide (641.3 m/z) and the endogenous PSA peptide (637.3 m/z). Two of the fragment ions from the standard peptide and the endogenous peptides were used for the MRM experiment. Absolute quantity of PSA is obtained by computing the quantity of endogenous PSA peptide with the standard peptide (Figure 4). Five normal prostate tissue samples and 15 cancerous tissue samples were quantified using this approach.

Abbreviations DTP, direct tissue proteomic; fmol, femptomole; pg, picogram; LC–MS/MS, liquid chromatography–tandem mass spectrometry; TCA cycle, tricarboxylic acid cycle; MALDI–TOF-MS, matrix-assisted laser desorption ionization-time-of-flight mass spectrometry; SELDI–TOF, surface-enhanced laser desorption ionization mass spectrometry. Acknowledgements We thank Michael Fong for his expert assistance in graphics and tables, and members of Han Lab for helpful discussion. This work was supported by R01 HL 67569, P01 HL70694, RR019436, funds from Neag Comprehensive Cancer Center and UConn Cancer Golf Proceeds. Oncogene


76 References Cantin GT, Yates III JR. (2004). J Chromatogr A 1053: 7–14. Coombs GS, Bergstrom RC, Pellequer JL, Baker SI, Navre M, Smith MM et al. (1998). Chem Biol 5: 475–488. Cox C, Bignell G, Greenman C, Stabenau A, Warren W, Stephens P et al. (2005). Proc Natl Acad Sci USA 102: 4542–4547. Cronauer MV, Schulz WA, Ackermann R, Burchardt M. (2005). Int J Oncol 26: 1033–1040. DePinho RA, Polyak K. (2004). Nat Genet 36: 932–934. Eng JK, McCormack AL, Yates III JR. (1994). J Am Soc Mass Spectrom 5: 976–989. Ford TF, Butcher DN, Masters JR, Parkinson MC. (1985). Br J Urol 57: 50–55. George DJ. (2002). Urology 60: 115–121; discussion 122. Gerber SA, Rush J, Stemman O, Kirschner MW, Gygi SP. (2003). Proc Natl Acad Sci USA 100: 6940–6945. Gleason DF. (1988). NCI Monogr 7: 15–18. Gleason DF. (1992). Hum Pathol 23: 273–279. Griffin TJ, Lock CM, Li XJ, Patel A, Chervetsova I, Lee H et al. (2003). Anal Chem 75: 867–874. Gupta A, Roehrborn C, Aragaki C. (2004). J Urol 171: 2384–2385 (author reply 2385). Gygi SP, Rochon Y, Franza BR, Aebersold R. (1999). Mol Cell Biol 19: 1720–1730. Han DK, Eng J, Zhou H, Aebersold R. (2001). Nat Biotechnol 19: 946–951. Hood BL, Darfler MM, Guiel TG, Furusato B, Lucas DA, Ringeisen BR et al. (2005). Mol Cell Proteomics 4: 1741–1753. Keller A, Nesvizhskii AI, Kolker E, Aebersold R. (2002). Anal Chem 74: 5383–5392. Lundgren DH, Eng J, Wright ME, Han DK. (2003). Mol Cell Proteomics 2: 1164–1176. Mayya V, Rezaul K, Cong YS, Han D. (2005). Mol Cell Proteomics 4: 214–223. Meyerson M, Carbone D. (2005). J Clin Oncol 23: 3219– 3226. Nantermet PV, Xu J, Yu Y, Hodor P, Holder D, Adamski S et al. (2004). J Biol Chem 279: 1310–1322. Nelson PS, Han D, Rochon Y, Corthals GL, Lin B, Monson A et al. (2000). Electrophoresis 21: 1823–1831.

Nesvizhskii AI, Keller A, Kolker E, Aebersold R. (2003). Anal Chem 75: 4646–4658. Parnes HL, Thompson IM, Ford LG. (2005). J Clin Oncol 23: 368–377. Patterson SD. (2004). Arthritis Rheum 50: 3741–3744. Peng J, Elias JE, Thoreen CC, Licklider LJ, Gygi SP. (2003). J Proteome Res 2: 43–50. Petricoin EF, Zoon KC, Kohn EC, Barrett JC, Liotta LA. (2002). Nat Rev Drug Discov 1: 683–695. Posadas EM, Simpkins F, Liotta LA, MacDonald C, Kohn EC. (2005). Ann Oncol 16: 16–22. Raben D, Helfrich B. (2004). Clin Lung Cancer 6: 48–57. Rezaul K, Wu L, Mayya V, Hwang SI, Han D. (2005). Mol Cell Proteomics 4: 169–181. Sawyers CL. (2002). Curr Opin Genet Dev 12: 111–115. Sharpless NE, DePinho RA. (2005). Nature 436: 636–637. Smith RA, Cokkinides V, Eyre HJ. (2005). CA Cancer J Clin 55: 31–44; quiz 55–56. Stephens P, Edkins S, Davies H, Greenman C, Cox C, Hunter C et al. (2005). Nat Genet 37: 590–592. Sweat GT. (2005). Postgrad Med 117: 45–50. Tanneberger S. (1977). Arch Geschwulstforsch 47: 755–765. Tarro G, Perna A, Esposito C. (2005). J Cell Physiol 203: 1–5. Tonon G, Wong KK, Maulik G, Brennan C, Feng B, Zhang Y et al. (2005). Proc Natl Acad Sci USA 102: 9625–9630. Velasco AM, Gillis KA, Li Y, Brown EL, Sadler TM, Achilleos M et al. (2004). Endocrinology 145: 3913–3924. Verras M, Brown J, Li X, Nusse R, Sun Z. (2004). Cancer Res 64: 8860–8866. Wardwell NR, Massion PP. (2005). Semin Oncol 32: 259–268. Wilson JF. (2004). Ann Intern Med 140: 317–319. Wolters DA, Washburn MP, Yates III JR. (2001). Anal Chem 73: 5683–5690. Wright ME, Eng J, Sherman J, Hockenbery DM, Nelson PS, Galitski T et al. (2003). Genome Biol 5: R4. Wright ME, Han DK, Aebersold R. (2005). Mol Cell Proteomics 4: 545–554. Zangar RC, Varnum SM, Covington CY, Smith RD. (2004). Dis Markers 20: 135–148. Zhu H, Mazor M, Kawano Y, Walker MM, Leung HY, Armstrong K et al. (2004). Cancer Res 64: 7918–7926.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc)

Oncogene