Verification of Endometrial Tissue Biomarkers Previously Discovered Using Mass Spectrometry-Based Proteomics by Means of Immunohistochemistry in a Tissue Microarray Format Valerie Dube´ ,† Jo1 rg Grigull,‡ Leroi V. DeSouza,§ Shaun Ghanny,§ Terence J. Colgan,*,†,| Alexander D. Romaschin,|,⊥ and K. W. Michael Siu§ Pathology and Laboratory Medicine, Mount Sinai Hospital, 600 University Avenue, Toronto, Ontario, Canada M5G 1X5, Department of Mathematics and Statistics, York University, 4700 Keele Street, Toronto, Ontario, Canada M2J 1P3, Departments of Chemistry and Biology, and Centre of Research in Mass Spectrometry, York University, Toronto, Ontario, Canada M2J 1P3, Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto, Ontario, Canada M5G 1X5, and Division of Clinical Biochemistry, St. Michael’s Hospital, 30 Bond Street, Toronto, Ontario, Canada M5B 1W8 Received February 16, 2007
Verification of candidate protein biomarkers is a necessary step in moving from the initial discovery to application. Here, we report results of a verification exercise involving six candidate endometrial cancer biomarkers previously discovered using mass-tagging and multidimensional liquid chromatography/ tandem mass spectrometry (DeSouza L., et al. J. Proteome Res. 2005, 4, 377-386) on a cohort of 148 patient samples by means of immunohistochemistry on a tissue microarray format. A panel of the three best-performing biomarkers, chaperonin 10, pyruvate kinase M2, and R-1-antitrypsin, achieved a sensitivity of 0.85, specificity of 0.93, predictive value of 0.90, and positive predictive value of 0.88 in discriminating malignant from benign endometrium. The ruggedness of this panel of biomarkers was verified in a 2/3-training-set-1/3-test-set cross-validation analysis by randomly splitting the cohort in 10 ways. The roles of chaperonin 10 and pyruvate kinase M2 in tumorigenesis confirm them as credible cancer biomarkers. Keywords: Verification • Biomarkers • Endometrial carcinoma • Tissue proteomics • Immunohistochemistry • Tissue Microarray • Chaperonin-10 • Pyruvate kinase M2
Introduction Mass spectrometry-based proteomic technologies, especially those based on isotope-dilution with mass-tagging reagents, including cICAT and iTRAQ, have revolutionized the discovery and identification of candidate biomarkers via differentialexpression analyses of samples from diseased and healthy individuals.1-7 Many of these discovery studies tend to involve only a limited number of samples, and the candidate biomarkers identified in these studies await confirmation based on a larger set of samples. An oft practiced strategy in analytical chemistry is to verify using a completely independent technology, which presumably will not have the same type of potential biases and/or errors that the original technology has. A powerful technique in protein identification and quantification * Corresponding author: Dr. Terence J. Colgan, Room 6-502-3, Department of Pathology and Laboratory Medicine, Mount Sinai Hospital, Toronto, Canada M5G 1X5. Tel, (416) 586-4522; fax, (416) 586-8481; e-mail,
[email protected]. † Pathology and Laboratory Medicine, Mount Sinai Hospital. ‡ Department of Mathematics and Statistics, York University. § Departments of Chemistry and Biology, and Centre of Research in Mass Spectrometry, York University. | Department of Laboratory Medicine and Pathobiology, University of Toronto. ⊥ Division of Clinical Biochemistry, St. Michael’s Hospital.
2648
Journal of Proteome Research 2007, 6, 2648-2655
Published on Web 06/07/2007
that is in common use is immunochemistry. For a tissue sample, the practice of immunohistochemistry provides not only the identification and quantification of the assayed protein, but also the localization, that is, in which cells is the protein.8 In an earlier study, we reported the discovery and identification of nine potential markers for endometrial cancer (EmCa) using a combination of cICAT and iTRAQ tags with multidimensional liquid chromatography (LC) and tandem mass spectrometry (MS/MS).9 The differential-expression analyses were performed on homogenates of EmCa, normal proliferative, and normal secretory endometrium tissues. These cICAT and iTRAQ experiments are suitable for biomarker discovery, but are time-consuming and laborious, suggesting that other analytic methods for biomarker validation are desirable. Here, we describe results of a 148-sample verification exercise on six candidate biomarkers based on immunohistochemical detection in a tissue microarray (TMA) format. High-density TMA is a miniaturized histopathologic technique permitting high-throughput protein analysis.10 Further, it provides information regarding cellular localization and permits near simultaneous analysis of multiple proteins in successive sections from an identical tumor. A TMA is con10.1021/pr070087o CCC: $37.00
2007 American Chemical Society
research articles
Verification of Endometrial Tissue Biomarkers
structed by removing typically 0.6-2 mm tissue cores from individual donor paraffin blocks, under the supervision of a pathologist, and placing them into a single recipient block (the TMA). Sections taken from the completed TMA are used for immunohistochemistry, permitting reliable analyses of numerous tissue samples under identical histologic conditions. Despite the small size of the tissue cores within the TMA, studies have shown that the tissue is well-represented, if g3 cores from each sample are used.11-15
Materials and Methods Tissue Microarray. Three TMA paraffin blocks were constructed using a tissue microarray workstation MTA-1 (Beecher Instruments, Sun Prairie, WI). Following research ethics board approval, 148 anonymized endometrial tissues (cases) dating between 1998 and 2005 were retrieved from the Mount Sinai Hospital pathology archives for construction of an endometrial TMA. Sixty-three cases consisted of pathological endometrium comprised of 10 endometrial hyperplasia samples (two simple hyperplasia and eight complex hyperplasia) and 53 EmCa (39 endometrioid or Type 1 adenocarcinoma, 13 serous papillary/ clear cell or Type 2 carcinoma, and one carcinosarcoma). Eighty-five cases consisted of benign, physiologic endometrium (25 proliferative, 25 secretory, 25 atrophic, and 10 menstrual). Three tissue cores of 0.6 mm in diameter were selected from original paraffin blocks of each of the 148 cases, resulting in a total number of 444 cores for assessment. All cores were placed into the TMA using a horizontal zigzag method with pathologic cases interspersed among benign endometrial cases. TMA interpretation was then performed in the vertical axis so as to maximize independent scoring of each case’s three cores. Biomarkers. Six proteins were selected from the panel of putative biomarkers based on their biological importance and availability of antibodies.9,16-18 These six biomarkers and their antibody sources were chaperonin-10 (Cpn-10, rabbit polyclonal, Stressgen, Victoria, BC), calgranulin A (CalA or S100A8, monoclonal, RDI Research Diagnostics, Concord, MA), polymeric immunoglobulin receptor precursor (PIGR, monoclonal, U.S. Biological, Swampscott, MA), pyruvate kinase M2 isoform (PK-M2, monocloncal, ScheβoBiotech, Griessen, Germany), alpha1-antitrypsin (AAT, rabbit polyclonal, Dako Cytomation, Carpinteria, CA), and tissue inhibitor of metalloproteinases type 1 (TIMP-1, rabbit polyclonal, Lab Vision, Fremont, CA). The first five were potential biomarkers observed to be differentially expressed in our earlier work with multidimensional liquid chromatography (LC) and tandem mass spectrometry (MS/MS) using a combination of cICAT and iTRAQ tags.9 TIMP-1 was selected because of metalloproteinases role in angiogenesis and degradation of the extracellular matrix in tumor invasion.19 Antibody sources were identified for each of these six potential biomarkers. Procedures. Each antibody was first optimized with respect to dilution and the use of microwave heating in Tris-HCl buffer to expose the antigen (“antigen retrieval”). Immunohistochemical staining was then performed with the following four antibodies and dilutions: anti-calgranulin A (1/150), anti-PIGR (1/50), anti-PK-M2 (1/50), anti-TIMP-1 (1/50), and with the following two antibodies without antigen retrieval: anti-Cpn10 (1/2000), anti-AAT (1/3000). Following the application of a protein blocker for 10 min, deparaffinized tissue sections were first incubated with the primary antibodies for 1 h at room temperature followed by secondary antibody conjugated with biotin. Finally, streptavidin-biotin immunoperoxidase was
added and developed using using 3,3′-diaminobenzidine as the chromogen. The sections were then counterstained with 0.1% hematoxylin, dehydrated, and mounted. Immunohistochemical staining for each antibody was scored in four tissue compartments: epithelium, stroma and vessels, leukocytes, and intraluminal secretions. EmCa is a cancer of the endometrial epithelial cells. For the epithelial and stromal compartments, the intensity and quantity of staining was scored using the Allred method.20-22 First, a quantitative score was performed by estimating the proportion of immunohistochemically positive-staining cells (0, none; 1, < 1/100; 2, 1/100 to 1/10; 3, 1/10 to 1/3; 4, 1/3 to 2/3; and 5, > 2/3). Second, an intensity score was performed by evaluating the average staining intensity of the positive cells (0, none; 1, weak; 2, intermediate; and 3, strong). A total score (ranging from 0 to 8) was afterward obtained by adding the quantitative score and the intensity score for each of the 444 cores. Scoring was performed separately for the individual compartments: an average epithelial total (AET) score and average stromal total (AST) score were obtained for each case from the average of the mean total scores obtained from the three cores. The leukocytic and secretory compartments were scored as negative or positive. Biostatistical Analysis. Comparisons of biomarkers tissue expression between the diagnostic classes was performed through statistical analysis of immunohistochemical scoring data. Initially, a standard analysis was first performed with the χ2 test and Fischer exact test (with a cutoff p-value of 0.001) to evaluate global score distributions. The AET and AST scores for each of the six markers for the epithelial and stromal compartments for all pathological and benign cases were visualized using a K-means clustering in the EisenLab software, through the specification of a 4 × 4 node architecture (Spearman correlation, average linkage). Subsequently, diagnostic classifiers were trained on the epithelial immunohistochemical expression data. The data were explored initially using the BRB ArrayTools, a software tool developed at the National Instituties of Health and suitable for analysis and class prediction in highthroughput gene-expression data.23 Logistic regression predictors were then chosen as a statistical framework that is bestsuited for the analysis of the present discrete and semiquantitative immunohistochemical data over a range of diagnostic cutoff values. The specific logistic regression model was as follows: Let p denote the predicted probability that a case i whose observed marker values are given by the vector y(i) ) (y(i, marker 1), y(i, marker 2), y(i, marker 3)) is malignant. The logistic regression discriminator has the form p (case i is malignant| y(i)) ) exp(R +
∑β y(i,j))/[1 + exp(R + ∑β y(i,j))] j
j
where the index ‘i’ denotes the individual sample and ‘j’ is a summation index that runs over the markers. Logistic regression analyses of the AET scores from the pathological and benign cases, and then from the malignant and benign cases were performed, using the R statistics software and a cutoff p-value of 0.05 for significance of the logistic regression parameters. The sensitivity and specificity for each biomarker identified as statistically significant by logistic regression were then reported within a range of cut-points on the class posterior p(.|y) and quantified as receiver operating characteristic (ROC) curves.23,24 A resulting positive predictive value (PPV) was obtained, for a predicted probability using a standard cut-point of 0.5. The Journal of Proteome Research • Vol. 6, No. 7, 2007 2649
research articles
Dube´ et al.
Figure 1. Graphic representation “heat map” of average epithelial total (AET) score and average stromal total (AST) score in 146 cases using all 6 biomarkers.
ROC curves reveal the number of malignant/pathological samples (M) and benign samples (B), predicted to be truly malignant/pathological (true positives/TP), and benign samples predicted to be truly benign (true negatives/TN), using the immunohistochemical data as training data on the logistic classifier.25 Sensitivity is defined as the quotient TP/M, that is, the fraction of true pathological/malignant cases predicted as being pathological/malignant. Likewise, specificity is defined as the quotient TN/B, that is, the fraction of true benign cases predicted as being benign. False positives (FP) are benign samples predicted to be malignant/pathological. The predictive value (PV) describes the proportion of correctly classified cases. The positive predictive value (PPV) is given by TP/(TP + FP). The predictive performances of the biomarkers were evaluated by randomly splitting the AET score data 10 times into a 2/3-training set (S_train) and a 1/3-testing set (S_test) and performing separate logistic regression analyses. The numbers of pathological/malignant samples and benign samples within the training and testing sets were identical across the 10 splits. In each of the 10 splits, a logistic regression model was specified, using only S_train. The model (which had different parameters across the 10 splits) was then used to predict classification as pathological/malignant or benign for each case in S_test. The performances were evaluated using ROC curves.
Results Fewer than 2% of the cores were absent from the TMA slides, reflecting overall technical quality and minimal sample losses. Data were available in 84/85 benign cases and in 62/63 pathological/malignant cases. All immunohistochemical staining mainly decorated the epithelial cytoplasm. Weak nuclear staining was noted for PIGR, CalA, and AAT. Surface membrane staining was noted for AAT, PK-M2, and TIMP-1. In addition, 2650
Journal of Proteome Research • Vol. 6, No. 7, 2007
PIGR showed intense positive staining in secretions, and CalA intense and constant positive staining of inflammatory cells.18 An initial survey of leukocytic and secretory results indicated that these two compartments were uninformative. We then focused exclusively upon the assessment and analysis of epithelial and stromal staining. Biostatistical analysis of epithelial immunohistochemical staining was analyzed at two diagnostic levels: pathologic (EmCa and hyperplastic endometrium) versus benign epithelium, and malignant (EmCa only) versus benign endometrium. Pathological versus Benign. The χ2 test and the Fischer exact test performed on the AET scores revealed a significant overexpression of AAT, Cpn-10, and PK-M2 in pathological cases compared to benign cases (p-values: 9.65 × 10-8, 1.39 × 10-5, and 3.44 × 10-7, respectively). PIGR was significantly underexpressed in pathological cases (p ) 3.44 × 10-7). CalA and TIMP-1 were not statistically discriminative. A graphical representation of the epithelial (AET) and stromal (AST) scores for all six biomarkers in all 146 cases is shown in Figure 1 as a heat map, following the K-means clustering analysis (E indicates epithelial, S indicates stromal values). The distribution of the cases suggests an overexpression of Cpn-10 and PK-M2 in the epithelium of pathological cases as compared to benign cases, the PK-M2 pattern being less apparent than the Cpn-10 and AAT patterns. This assessment was confirmed by box-plots that identified informative (Figure 2a), and uninformative markers (Figure 2b). PIGR is uninformative as a biomarker despite results of the χ2 test, which examines global distribution as opposed to individual expression in a given sample as a diagnostic test. Logistic regression of the AET score confirmed a statistically significant overexpression of Cpn-10 and PK-M2 in pathological cases compared to benign cases (p-values: 2.50 × 10-4 and
Verification of Endometrial Tissue Biomarkers
research articles
Figure 2. (a) Informative biomarkers (AAT, Cpn-10, PK-M2) useful in separating pathological from benign cases (n ) 146); (b) uninformative biomarkers (Calgranulin A, PIGR, and TIMP-1) not useful in separating pathological from benign cases (n ) 146).
2.76 × 10-4, respectively). Overexpression of AAT in pathological endometrium could not be confirmed by logistic regression (pvalue: 0.113), even though earlier test results seemed to demonstrate an overexpression in pathological cases (Figure 2a). PIGR, CalA, and TIMP-1 expressions were not statistically different between the pathological and benign cases (pvalues: 0.0661, 0.0590, and 0.858, respectively). Refined logistic regression was then performed using Cpn-10 and PK-M2, which confirmed overexpression of these two markers in pathological
endometrium (p-values: 6 × 10-8 and 1 × 10-7, respectively). These results and their coefficients are quantified in ROC curves (Figure 3); the resultant predictive performances of the logistic regression classifiers for the two biomarkers individually and in combination are shown in Table 1a. The results show that combining these two biomarkers improves predictive performances over those of each biomarker individually and results in a sensitivity of 0.77, a specificity of 0.87, a PV of 0.83, and a PPV of 0.81 for pathological cases as compared to benign cases. Journal of Proteome Research • Vol. 6, No. 7, 2007 2651
research articles
Dube´ et al.
Figure 4. A 2/3 vs 1/3 cross-validation of the logistic regression predictor of two biomarkers (Cpn-10, PK-M2) in TMA. Figure 3. Refined logistic regression predictor using two biomarkers (Cpn-10 and PK-M2): pathological endometrium (n ) 62) vs benign (n ) 84) ROCs. Table 1. Coefficients of Predictive Performance for Single Biomarkers biomarker
sensitivity
specificity
PV
PPV
(a) Pathological versus Benign Endometrium 0.69 0.80 0.75 0.72 0.65 0.79 0.73 0.69 0.77 0.87 0.83 0.81 (b) Malignant versus Benign Endometrium Cpn-10 0.73 0.83 0.79 0.73 PK-M2 0.65 0.79 0.74 0.65 AAT 0.71 0.79 0.76 0.67 Panel 0.85 0.93 0.90 0.88
Cpn-10 PK-M2 Panel
AUC
0.84 0.79 0.91 0.84 0.75 0.81 0.92
The area-under-the-curve (AUC) value for the ROC curve of the Cpn-10 and PK-M2 panel is 0.91. The reproducibility of the predictive performance of the logistic regression classifier is evaluated by cross-validation analysis6 by splitting the data set randomly into a 2/3-training set and 1/3-testing set: the 62 pathological and 84 benign cases were randomly divided into 10 training and 10 testing sets of 41 pathological/56 benign cases and 21 pathological/28 benign cases in each split, respectively. For each training set, a logistic regression model was specified using the expression values of the two markers Cpn10 and PK-M2. Figure 4 shows the ROC curves for the training (dashed lines) and testing (full lines) sets. The AUC values for these ROC curves are shown in Table 2a. Malignant versus Benign. The χ2 test and the Fischer exact test performed on the AET scores again revealed a significant overexpression of AAT, Cpn-10, and PK-M2 in malignant cases (p-values: 2.08 × 10-8, 6.10 × 10-6, and 1.06 × 10-6, respectively). As before, PIGR was significantly underexpressed in malignant cases (p-value: 1.06 × 10-6), which was not borne out by individualistic examination via logistic regression as opposed to global expression evaluation by means of the χ2 test. CalA and TIMP-1 were not statistically discriminative. Logistic regression of the AET scores revealed overexpression of AAT, Cpn-10, and PK-M2 in malignant cases compared to 2652
Journal of Proteome Research • Vol. 6, No. 7, 2007
Table 2. Quantitative Summaries of Cross-Validation ROCs data split
train
test
(a) AUC Values for 62 Pathological versus 84 Benign Cases Using Two Biomarkers (Cpn-10 and PK-M2) 1 0.94 0.86 2 0.92 0.87 3 0.93 0.87 4 0.92 0.87 5 0.93 0.86 6 0.94 0.83 7 0.94 0.81 8 0.91 0.91 9 0.88 0.95 10 0.91 0.91 (b) AUC Values for 52 Malignant versus 84 Benign Cases Using All Six Biomarkers 1 0.97 0.86 2 0.96 0.89 3 0.95 0.89 4 0.95 0.90 5 0.93 0.95 6 0.96 0.88 7 0.95 0.92 8 0.95 0.92 9 0.94 0.94 10 0.94 0.94
benign cases (p-values: 3.59 × 10-2, 1.08 × 10-3, and 2.40 × 10-4, respectively). CalA was marginally underexpressed in malignant cases (p-value: 0.0327). The differences in the expressions of PIGR and TIMP-1 were not statistically significant, using a p-value cutoff of 0.05. Refined logistic regression was then performed using only three significant biomarkers, AAT, Cpn-10, and PK-M2. These results and their coefficients were again quantified in ROC curves (Figure 5); the resultant predictive performances of the logistic regression classifier for the three biomarkers individually and in combination are shown in Table 1b. The panel of three biomarkers, AAT, Cpn10, and PK-M2, improves predictive performances over each biomarker taken individually and results in a sensitivity of 0.85, a specificity of 0.93, a PV of 0.90, a PPV of 0.88, and an AUC of 0.92 for malignant versus benign cases. Validation and reproducibility predictive performances of all six TMA biomarkers for malignant compared to benign cases
Verification of Endometrial Tissue Biomarkers
research articles Type 2 carcinomas. A prior 40-sample iTRAQ-labeling study of 12 Type 1 and 8 Type 2 EmCa samples did suggest different proteomic profiles.26 Any attempt to discriminate between the two types of EmCa would require a much larger cohort of malignant samples, especially Type 2 carcinomas.
Figure 5. Refined logistic regression predictor using three biomarkers (Cpn-10, PK-M2, and AAT): EmCa (n ) 52) vs benign (n ) 83) ROCs.
Figure 6. A 2/3 vs 1/3 cross-validation of the logistic regression predictive performances of six biomarkers in TMA: EmCa (n ) 52) vs Benign (n ) 84).
was assessed by splitting the AET score data into a 2/3-training set (S_train) and 1/3-test test (S_test). Figure 6 displays the ROC curves for 10 cross-validation data splits (2/3 training versus 1/3 testing). The AUC values for the separate ROC curves in each data split are given in Table 2b.
Discussion Results of this immunohistochemical TMA study have independently validated three previously discovered and identified biomarkers, Cpn-10, PK-M2, and AAT, for EmCa on a different and very much larger set of 148 samples. The major advantage of TMA is the ability to handle and compare many samples nearly simultaneously. Performing iTRAQ analysis on that many samples is currently intractable. Differential-expression analysis using mass spectrometry to first discover candidate biomarkers, followed by immunohistochemical validation on a TMA format, appears to be a valuable combination of strategies that complement each other. This study successfully identified a panel of three biomarkers capable of discriminating malignant (Table 1b) and pathologic (Table 1a) from benign endometrium; sample size did not permit a statistically valid examination of whether these, or other, biomarkers, could successfully separate Type 1 from
Of the three validated EmCa biomarkers, Cpn-10 and PKM2 offer the best tumorigenetic plausibility for their discriminating role. AAT is a major protein in serum and is secreted in large quantities by the liver, which lessens its potential utility as a biomarker for early, blood-based detection. Cpn-10 (or HSP-10) is a 10-kDa, 101 amino-acid protein. Mitochondrial Cpn-10 is a chaperone involved in protein folding, which in the presence of ATP mediates protein assembly.27 Our earlier studies involving analysis of the intact protein have shown that Cpn-10 carries two post-translational modifications at the N terminus.28 Cpn-10 was first discovered as “early pregnancy factor” in 1987, when it was found to be released from the fetal placental unit within 6 h of conception and persisted for at least the first 6 months of pregnancy, and was considered to be a suppressor of the maternal immune response.29 More recently, the extracellular form of Cpn-10 has been shown to be a product of neoplastic cell proliferation and acts as a growth factor for these cells.30 Additional studies have demonstrated that Cpn-10 acts as an obligatory autocrine growth factor for cultured tumor and transformed cells.31 Extracellular Cpn-10 is secreted by endometrial and ovarian cancers,32 and has been identified as a potential biomarker in a number of malignancies.33 Strategies to block Cpn-10 have been hampered by the inability to produce moderate or high affinity antibodies, and as a result, assessment of the full therapeutic potential of blocking the action of extracellular Cpn-10 in vivo remains to be explored.34 PK-M2 is expressed exclusively in metastatic cells and only in healthy lung tissue; it is structurally different from its other isoforms (L, R, and M1).35-38 The M1 and M2 isoforms are splice variants and vary in sequence in 22 positions in the fructose1,6-diphosphate (FDP) binding domain.39 PK-M2 plays a key role in the survival of cancers in hypoxic environments and in the provision of metabolites for rapid cell division.35,40 In tumor cells, PK-M2 oscillates from an inactive dimer to an active tetramer based on allosteric activation by FDP.41,42 By contrast, the M1 isoform is unresponsive to this activation.43 The M1 isoform is present in tissues requiring rapid energy production, such as muscle and brain. PK-M2, the onco-fetal form, is characteristic of cells with high rates of nucleic synthesis, and is progressively replaced in tissues after birth.44 Subsequently, PK-M2 is found in all proliferating cells, such as embryonic cells, adult stem cells, and particularly tumor cells. During tumor formation, a shift in the isozyme composition of pyruvate kinase always takes place in such a manner that the tissue-specific isozyme disappears and is replaced by the M2 isoform. Because of the high rate of nucleic acid synthesis, the metabolome of tumor cells is classified as nucleogenic in order to distinguish it from the metabolome of differentiated tissues.45 The only normal differentiated human tissue in which PK-M2 is found is lung. Paradoxically, in lung and tumor sections, immunological staining of PK-M2 with monoclonal antibodies specific to the M2 isoform always reveals strong staining of the tumor cells, but absence of staining in normal lung tissue.46 The reason for this apparent paradox is due to differences in the quaternary structure of the protein in normal lung as opposed to tumor tissue. The tumor form of PK-M2 is predominantly dimeric, whereas in lung and normal proliferating Journal of Proteome Research • Vol. 6, No. 7, 2007 2653
research articles cells PK-M2 is predominantly found in the tetrameric form.47,48 The dimeric form of PK-M2 has low affinity for its substrate phosphoenol pyruvate (PEP), while the tetrameric form has high affinity. At physiological PEP concentrations, the tetrameric form is highly active, whereas the dimeric form is nearly inactive. The tetrameric form is associated with a glycolytic enzyme complex consisting of other glycolytic and nonglycolytic enzymes. A cyclic dissociation and association of PK-M2 with the glycolytic enzyme complex allows tumor cells to simultaneously adapt to hypoxic environments, requiring direct substrate-level glycolitic ATP production via PK-M2, and sequential accumulation of key metabolites for rapid cell proliferation.45 Sufficient primary sequence difference exists between the M2, L and R isoforms in the FDP-binding region to allow specific inhibition of this domain by drugs. Blocking PK-M2 allosteric activation has the potential to drive tumor cells into energy starvation. Structural variations in the FDPbinding pocket offer an excellent opportunity to discriminate between pyruvate kinase isoforms, and hence offer tumorspecific targets.43 Allosteric binding pockets distant from catalytic sites have been shown to be effective enzyme regulators and drugable sites. Elevated levels of plasma PK-M2 in ovarian cancer patients have recently been reported.49 Some of the TMA findings in this study are in apparent contrast with our earlier MS-based findings; resolution of these paradoxes serves to highlight the value of TMA verification in tissue proteomics. In this TMA study, AAT is a positive marker, yet an earlier MS-based proteomic analysis of tissue homogenates identified AAT as a negative marker.9 This MS finding, however, was based upon an integrated signal from all tissue compartments, which include blood vessels, secretion, stroma, and not merely epithelium. It is of note that AAT is significantly overexpressed in the stroma of benign tissues (Figure 1), and there is typically more stroma than glands in the benign endometrium. Similarly, whereas PIGR was previously found to be a positive marker in MS analysis of tissue homogenates, it has not been found to be an informative marker in this TMA study. It is known, however, that PIGR is a secreted protein, and intense PIGR staining was observed in the lumen of the glands of some EmCa samples. Secretion is a compartment that would have contributed to the MS findings, but would have not been reflected in an epithelial immunohistochemical score in this study. This study has achieved its primary purpose of validating biomarkers discovered using a combination of cICAT and iTRAQ tags with multidimensional LC and MS/MS, using an IHC/TMA format combined with bioinformatics analysis. This step represents only the first in a multistep process toward clinical application. The additional steps toward clinical application would require exhaustive examination of biomarker expression in endometrial tissue which can demonstrate an enormous plasticity of its physiologic, preneoplastic, and malignant states. For example, a detailed study of the performance of biomarkers in distinguishing endometrioid carcinoma from its precursor lesion, atypical endometrial hyperplasia may be required. Such studies can be very challenging since mixed lesions, for example, carcinomas of both Types 1 and 2 or endometriod carcinomas with adjacent preneoplastic hyperplasia, are not uncommon. Finally, the most promising biomarkers would then be selected for use in a diagnostic platform (e.g., histology, exfoliative cytology, aspiration cytology, or fluid/ serum analysis), and their performance compared to existing screening, diagnostic, and monitoring methods. 2654
Journal of Proteome Research • Vol. 6, No. 7, 2007
Dube´ et al.
The limitations of immunohistochemistry would suggest that immediate introduction of the two validated biomarkers is likely premature, and that further work is required on the utility of these, and other, biomarkers in the histologic diagnosis of disease. First, modification of biomarker antigenicity may occur during tissue fixation, initial paraffin embedding, TMA construction,50 and storage of unstained TMA slides.51 Second, immunological staining intensity within a laboratory can vary from run to run, the archival age of the tissue,11 and the longevity, source, and purity of the antibody. Third, immunohistochemical results vary substantially among laboratories, due to difficult-to-reproduce conditions.8 Fourth, immunohistochemical scoring in this study was undertaken by a single reviewer (V.D.); whether this scoring can be replicated among various observers is currently unknown. The standardization of all these potential barriers remains a large, and possibly insurmountable, challenge for the utility of immunohistochemistry in the identification of disease, using only a selected few biomarkers.8 Nevertheless, it is possible that such biomarkers used in an immunohistochemical analysis could be an adjunctive technique to the current histopathologic diagnosis of EmCa. In conclusion, this IHC-TMA study has validated two credible cancer biomarkers, PK-M2 and Cpn10, following their initial identification in a MS-based proteomic experiment.
Acknowledgment. The authors gratefully acknowledge the support and cooperation of their surgical and clinical laboratory collaborators, including Drs. K. Joan Murphy (University Health Network), Wusun Paek (Mount Sinai Hospital), and Titus Owalabi, Ali Qizlbash, and Denis MacDonald (North York General Hospital), and the assistance of Muntajib Alhaq, Raquel Salamanca, and Maria Mendes (Mount Sinai Hospital). This work was supported primarily by the Canadian Cancer Society’s Research Grant No.016172 administered by the National Cancer Institute of Canada (A.D.R., T.J.C., K.W.M.S.). The Ontario Genomics Institute and Genome Canada provided some personnel support. Infrastructural support was provided by the Ontario Research and Development Challenge Fund, and Applied Biosystems/MDS SCIEX. V.D. acknowledges clinical fellowship support from Laval University and Mount Sinai Hospital. References (1) Li, C.; Hong, Y.; Tan, Y.-X.; Zhou, H.; Ai, J.-H.; Li, S.-J.; Zhang, L.; Xia, Q.-C. Wu, J.-R.; Wang, H.-Y.; Zeng, R. Accurate qualitative and quantitative proteomic analysis of clinical hepatocellular carcinoma using laser capture microdissection coupled with isotope-coded affinity tag and two-dimensional liquid chromatography mass spectrometry. Mol. Cell. Proteomics 2004, 3 399409. (2) Veenstra, T. D.; Conrads, T. P.; Hood, B. L.; Avellino, A.; Ellenbogen, R.; Morrison, R. Biomarkers: mining the biofluid proteome. Mol. Cell. Proteomics 2005, 4, 409-418. (3) Caldwell, R. L.; Caprioli, R. M. Tissue profiling by mass spectrometry. Mol. Cell. Proteomics 2005; 4, 394-401. (4) Maruvada, P.; Wang, W.; Wagner, P. D.; Srivastava, S.; Biomarkers in molecular medicine: cancer detection and diagnosis. Biomarkers Cancer Res. 2005, 38, 9-15. (5) Chaurand, P.; Schwartz, S. A.; Capriole, R. M. Assessing protein patterns in disease using imaging mass spectrometry. J. Prot. Res. 2004, 3, 245-252. (6) Ransohoff, D. F. Rules of evidence for cancer molecular-marker discovery and validation. Nat. Rev. Cancer 2004, 4, 309-314. (7) Wilkins, M.; Appel, R.; Van Eyk, J.; Chung, M.; Go¨rg, A.; Hecker, M.; Huber, L.; Langen, H.; Link, A.; Paik, Y.; Patterson, S.; Pennington, S.; Rabilloud, T.; Simpson, R.; Weiss, W.; Dunn, M. Guidelines for the next 10 years of proteomics. Proteomics 2006, 6, 4-8.
research articles
Verification of Endometrial Tissue Biomarkers (8) Cregger, M.; Berger, A.; Rimm, D. Immunohistochemistry and quantitative analysis of protein expression. Arch. Pathol. Lab. Med. 2006, 130, 1026-1030. (9) DeSouza, L.; Diehl, G.; Rodrigues, M.; Guo, J.; Romaschin, A.; Colgan, T.; Siu, K. W. M. Search for cancer markers from endometrial tissues using differentially labeled tags iTRAQ and cICAT with multidimensional liquid chromatography and tandem mass spectrometry. J. Prot. Res. 2005, 4, 377-386. (10) Kononen, J.; Bubendorf, L.; Kallioniemi, A.; Barlund, M.; Schraml, P.; Leighton, S.; Torhorst, J.; Mihatsch, M.; Sauter, G.; Kallioniemi, O. Tissue microarrays for high-throughput molecular profiling of tumor specimens. Nat. Med. 1998, 4, 844-847. (11) Eguiluz, C.; Viguera, E.; Millan, L.; Perez, J. Multitissue array review: a chonological description of tissue array techniques applications and procedures. Path. Res. Pract. 2006, 202, 561568. (12) Packeisen, J.; Buerger, H.; Krech, R.; Boecker, W. Tissue microarrays: a new approach for quality control in immunohistochemistry. J. Clin. Pathol. 2002, 55, 613-615. (13) Packeisen, J.; Korsching, E.; Herbst, H.; Boecker, W.; Buerger, H. Demystified ... tissue microarray technology. J. Clin. Pathol. 2003, 56, 198-204. (14) Simon, R.; Sauter, G. Tissue microarrays for miniaturized highthroughput molecular profiling of tumors. Exp. Hematol. 2002, 30, 1365-1372. (15) Torhorst, J.; Bucher, C.; Kononen, J.; Haas, P.; Zuber, M.; Ko¨chli, O.; Mross, F.; Dieterich, H.; Moch, H.; Mihatsch, M.; Kallioniemi, O.; Sauter, G. Tissue microarrays for rapid linking of molecular changes to clinical endpoints. Am. J. Pathol. 2001, 159, 22492256. (16) Guo, J.; Colgan, T.; DeSouza, L.; Rodriguez, M.; Romaschin, A.; Siu, K. W. M. Direct analysis of laser capture microdissected endometrial carcinoma and epithelium by MALDI mass spectrometry. Rapid Commun. Mass Spectrom. 2005, 19, 2762-2766. (17) Yang, E.; Guo, J.; Diehl, G.; DeSouza, L.; Rodriguez, M.; Romaschin, A.; Colgan, T.; Siu, K. W. M. Protein expression profiling of endometrial malignancies reveals a new tumor marker: chaperonin 10. J. Proteome Res. 2004, 3, 636-643. (18) Guo, J.; Yang, E.; DeSouza, L.; Diehl, G.; Rodriguez, M.; Romaschin, A.; Colgan, T.; Siu, K. W. M. A strategy for high-resolution protein identification in surface-enhanced laser desorption/ ionization mass spectrometry: calgranulin A and chaperonin 10 as protein markers for endometrial carcinoma. Proteomics 2005, 5, 1-14. (19) Ariztia, E. V.; Lee, C. J.; Gogoi, R.; Fishman, D. A. The tumor microenvironment: key to early detection. Crit. Rev. Clin. Lab. Sci. 2006, 43, 393-425. (20) Allred, D.; Clark, G.; Elledge, R.; Fuqua, S.; Brown, R.; Chamness, G.; Osborne, C.; McGuire, W. Association of p53 protein expression with tumor cell proliferation rate and clinical outcome in node-negative breast cancer. J. Natl. Cancer Inst. 1993, 85, 200206. (21) Allred, D.; Harvey, J.; Berardo, M.; Clark, G. Prognostic and predictive factors in breast cancer by immunohistochemical analysis. Mod. Pathol. 1998, 11, 155-168. (22) Harvey, J.; Clark, G.; Osborne, C.; Allred, D. Estrogen receptor status by immunohsitochemistry is superior to the ligand-binding assay for predicting response to adjuvant endocrine therapy in breast cancer. J. Clin. Oncol. 1999, 17, 1474-1481. (23) Hastie, T.; Tibshirani, R.; Friedman, J. The Elements of Statistical Learning; Springer: New York, 2001. (24) Pepe, M. The Statistical Evaluation of Medical Tests for Classification and Prediction; Oxford University Press: Oxford, 2003. (25) Simon, R.; Korn, E.; McShane, L.; Radmacher, M.; Wright, G.; Zhao, Y. Design and Analysis of DNA Microarray Investigations; Springer: New York, 2004. (26) DeSouza, L.V.; Grigull, J.; Ghanny, S.; Dube´, V.; Romaschin, A. D.; Colgan, T. J.; Siu, K. W. M. Endometrial carcinoma biomarker discovery and verification using differentially tagged clinical samples with multidimensional liquid chromatography and tandem mass spectrometry. Mol. Cell. Proteomics [Online early access]. Published Online: March 19, 2007. http://www.mcponline.org/cgi/content/abstract/M600378-MCP200v1. (27) Meyer, A.; Gillespie, J.; Walther, D.; Millet, I.; Doniach, S.; Frydman, J. Closing the folding chamber of the eukaryotic chaperonin requires the transition state of ATP hydrolysis. Cell 2003, 113, 369-381. (28) Raspopov, S.; El-Farawamy, A.; Thomson, B.; Siu, K. W. M. Infrared multiphoton dissociation in quadrupole time-of-flight mass spectrometry to-down characterization of proteins. Anal. Chem. 2006, 78, 4572-4577.
(29) Noonan, F.; Halliday, W.; Morton, H.; Clunie, G. Early pregnancy factor is immunosuppressive. Nature 1979, 278, 649-651. (30) Quinn, K.; Morton, H. Effect of monoclonal antibodies to early pregnancy factor (EPF) on the in vivo growth of transplantable murine tumours. Cancer Immunol. Immunother. 1992, 34, 265271. (31) Quinn, K.; Athanasis-Plastis, S.; Wong, T-Y.; Rolfe, B.; Cavanagh, A.; Morton, H. Monoclonal antibodies to early pregnancy factor perturb tumour cell growth. Clin. Exp. Immunol. 1990, 80, 100108. (32) Akyol, S.; Gercel-Taylor, C.; Reynolds, L.; Taylor, D. HSP-10 in ovarian cancer: expression and suppression of T-cell signaling. Gynecol. Oncol. 2006, 101, 481-86. (33) Cappello, F.; Bellafiore, M.; David, S.; Anzalone, R.; Zummo, G. Ten kilodalton heat shock protein (HSP10) is overexpressed during carcinogenesis of large bowel and uterine exocervix. Cancer Lett. 2003, 196, 35-41. (34) Somodevilla-Torres, M.; Hillyard, N.; Morton, H.; Alewood, D.; Halliday, J.; Alewood, P.; Vesey, D.; Walsh, M.; Cavanagh, A. Preparation and characterization of polyclonal antibodies against human chaperonin 10. Cell Stress Chaperones 2000, 5, 14-20. (35) Eigenbrodt, E.; Reinacher, M.; Scheefers-Borchel, U.; Scheefers, H.; Friis, R. Double role of pyruvate kinase type M2 in the expansion of phosphometabolite pools found in tumor cells. Crit. Rev. Oncog. 1992, 3, 91-115. (36) Gumi’nska, M.; Ignacak, J.; K¸ edryna, T.; Stachurska, B. Tumorspecific pyruvate kinase isoenzyme M2 involved in biochemical strategy of energy generation in neoplastic cells. Acta Biochim. Pol. 1997, 44, 711-724. (37) Hacker, H.; Steinberg, P.; Bannasch, P. Pyruvate kinase isoenzyme shift from L-type to M2-type is a late event in hepatocarcinogenesis induced in rats by a choline-deficient/dl-ethioninesupplemented diet. Carcinogenesis 1998, 19, 99-107. (38) Eigenbrodt, E.; Gerbracht, U.; Mazurek, S.; et al. Carbohydrate Metabolism and Neoplasia: New Perspective for Diagnosis and Therapy; Academic Press: San Diego, CA, 1994, 311-385. (39) Noguchi, T.; Inoue, H.; Tanaka, T. The M1 and M2 type isozymes of rat pyruvate kinase are produced from the same gene by alternative RNA slicing. J. Biol. Chem. 1986, 261, 13807-13812. (40) Mazurek, S.; Boschek, C.; Eigenbrodt, E. The role of phosphometabolites in cell proliferation, energy metabolism and tumor therapy. Biomembrane 1997, 29, 315-330. (41) Mazurek, S.; Eigenbrodt, E. The tumor metabolome. Anticancer Res. 2003, 23, 1149-1154. (42) Ashizawa, K.; Willingham, M.; Liang, C.; Cheng, S. In vivo regulation of monomer-tetramer conversion of pyruvate kinase subtypeM2 by glucose is mediated by fructose 1,6 biophosphate. J. Biol. Chem. 1991, 266, 16842-16846. (43) Dombrauckas, J.; Santarsiero, B.; Mesecar, A. Structural basis for tumor pyruvate kinase M2 allosteric regulation and catalysis. Biochemistry 2005, 44, 9417-9429. (44) Yamada, K.; Noguchi, T. Alteration of isozyme gene expression during cell differentiation and oncogenesis. Nippon Rinsho 1995, 53, 1112-1118. (45) Mazurek, S.; Boschek, B.; Hugo, F.; Eigenbrodt, E. Pyruvate kinase M2 and its role in tumor growth and spreading. Sem. Cancer Biol. 2005, 15, 300-308. (46) Shneider, J.; Neu, K.; Grimm, H.; Velcovsky, H.; Weisse, G.; Eigenbrodt, E. Tumour M2 pyruvate kinase in lung cancer patients: immunohistochemical detection and disease monitoring. Anticancer Res. 2002, 22, 311-318. (47) Presek, P.; Reinacher, M.; Eigenbrodt, E. Pyruvate kinase type M2 is phosphorylated at tyrosine residues in cells transformed by Rous sarcoma virus. FEBS Lett. 1988, 242, 194-198. (48) Mazurek, S.; Zweschke, W.; Jansen-Du ¨ rr, P.; Eigenbrodt, E. Effects of the human papilloma virus HPV-16 E7 oncoprotein on glycolysis and glutaminolysis: role of pyruvate kinase type M2 and the glycolytic enzyme complex. Biochem. J. 2001, 356, 247256. (49) Ahmed, A.; Dew, T.; Papadopoulos, A.; Devaja, O.; Lawton, F.; Raju, K.; Sherwood, R. M2-PK as a novel marker in ovarian cancer, a prospective cohort study. Int. J. Gynecol. Ca 2006, 16 (S3), 624694. (50) Chiriboga, L.; Osman, I.; Mikhail, C.; Lau, C. Tissue microarrays, tread carefully (letter). Lab. Invest. 2004, 84, 1677. (51) Mirlacher, M.; Kasper, M.; Storz, M.; Knecht, Y.; Durmuller, U.; Simon, R.; Mihatsch, M.; Sauter, G. Influence of slide aging on results of translational research studies using immunohistochemistry. Mod. Pathol. 2004, 17, 1414-1420.
PR070087O Journal of Proteome Research • Vol. 6, No. 7, 2007 2655