SEARCH

SEARCH BY CITATION

Keywords:

  • pancreatic cancer;
  • formalin-fixed paraffin-embedded (FFPE);
  • laser microdissection (LMD);
  • liquid chromatography-tandem mass spectrometry (LC-MS/MS);
  • scheduled selected reaction monitoring (sSRM)

Abstract

  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Pancreatic cancer is among the most lethal malignancies worldwide. We aimed to identify novel prognostic markers by applying mass spectrometry (MS)-based proteomic analysis to formalin-fixed paraffin-embedded (FFPE) tissues. Resectable, node positive pancreatic ductal adenocarcinoma (PDAC) with poor (n = 4) and better (n = 4) outcomes, based on survival duration, with essentially the same clinicopathological backgrounds, and noncancerous pancreatic ducts (n = 5) were analyzed. Cancerous and noncancerous cells collected from FFPE tissue sections by laser microdissection (LMD) were processed for liquid chromatography (LC)-tandem MS (MS/MS). Candidate proteins were identified by semiquantitative comparison and then analyzed quantitatively using selected reaction monitoring (SRM)-based MS. To confirm the associations between candidate proteins and outcomes, we immunohistochemically analyzed a cohort of 87 cases. In result, totally 1,229 proteins were identified and 170 were selected as candidate proteins for SRM-based targeted proteomics. Fourteen proteins overexpressed in cancerous as compared to noncancerous tissue showed different expressions in the poor and better outcome groups. Among these proteins, we found that three novel proteins ECH1, OLFM4 and STML2 were overexpressed in poor group than in better group, and that one known protein GTR1 was expressed reciprocally. Kaplan–Meier analysis showed high expressions of all four proteins to correlate with significantly worse overall survival (p < 0.05). In conclusion, we identified four proteins as candidates of prognostic marker of PDAC. The combination of shotgun proteomics verified by SRM and validated by immunohistochemistry resulted in the prognostic marker discovery that will contribute the understanding of PDAC biology and therapeutic development.

Pancreatic cancer has the worst prognosis of any major malignancy, with a 5-year survival rate of less than 5% after diagnosis.1 The majority of exocrine pancreatic cancer is pancreatic ductal adenocarcinoma (PDAC).2 Radical surgical resection gives chance for cure, but only 10–20% of patients are candidates for curative resection, and even if curative resection is performed, high recurrence rates is expected.2 In an effort to improve survival rates, several studies have analyzed determinants of long-term survival in postoperative PDAC patients.3–7 The biggest problem of previous analyses of prognostic markers in PDAC is that the clinicopathological background is quite heterogeneous even in high volume centers including our departments. Multivariate analysis is usually used to evaluate clinical significance, but no previous studies compared biomarkers in well-balanced clinically matched cohorts. The biomarker in Union for International Cancer Control (UICC) Stage I disease (confined in the pancreas) may not represent biomarkers in UICC Stage IV (distantly metastasized). Patients with PDAC of UICC Stage IIA (extends to the surrounding organs, node negative) and IIB (extends to the surrounding organs, node positive) are most frequently encountered as candidates for curative resection.2 The clinical outcome of Stage IIA is different from that of Stage IIB because of node metastasis.2 Since UICC Stage IIB still contains wide heterogeneity. Japan Pancreas Society (JPS) Staging was used further stratification; JPS Stage III disease (extends to surrounding tissue but not to the major vessels) and JPS Stage IVa (invasion to the major vessels).2 From histological point of view, the outcome of Grade 2,3 (moderately and poorly differentiated, respectively) PDAC is different from that of Grade 1 (well differentiated) PDAC.2 Furthermore, preoperative or postoperative therapies affect the prognostic outcomes.4, 5 Thus, to find meaningful biomarkers, comparison of two groups of PDACs of same histological grade within one clinical stage treated similarly, but with different outcomes is vital.

Recently, efforts have been made to identify novel biomarker proteins by gene microarray and proteomics methods. Apart from a low gene to gene-product ratio, several studies have indicated that mRNA expression levels do not necessarily correlate with protein expression or disease progression, whereas profiles of proteins and their various isoforms can more accurately identify disease states, such as cancer.8 Furthermore, transcriptomics cannot predict the activation of key signaling molecules in important protein networks.9 These developments and issues have brought proteomics to the forefront of the search for biomarkers.10

Mass spectrometry (MS)-based proteomics is an indispensable tool for molecular and cellular biology,11 as well as for discovering biomarkers.12 Shotgun proteomics is a method of identifying proteins in complex mixtures using a combination of liquid chromatography (LC) and MS to provide global proteome profiles.13 Meanwhile, targeted proteomics based on selected reaction monitoring (SRM) is an appropriate method for accurately identifying and quantitating proteins of interest.14

The limitations of previous studies,15 using serum/plasma and raw tissue samples made us focus on formalin-fixed paraffin-embedded (FFPE) tissue. Serum/plasma contain very low amount of target specific proteins while very large amount of housekeeping proteins. Prospective collection of unfixed tissue samples is possible only in a large enough tumor and take years to correlate with clinical outcomes. On the other hand, FFPE tissue archived worldwide, derived from a routine process, provides clinicopathologically well-defined and clinically related tissue archive. Association of clinical data and experimental information will represent an extremely valuable, as yet untapped reservoir of protein biomarkers.16–18 Recent developments in extraction methodologies of covalently linked proteins have finally made analysis of FFPE tissue by MS possible, providing access to this vast and clinically important sample set for biomarker discovery.16–18

Another aspect of PDAC is desmoplastic stroma containing inflammatory cells that may disturb the identification of cancer-specific proteins.19 Using laser microdissection (LMD), isolation of cancer cells and accurate molecular analysis of previously unanswered questions was possible.

The present study, for the first time, revealed novel prognostic markers of resectable, node positive (UICC Stage IIB/JPS Stage III) PDAC using proteomics of cancer cells isolated with LMD of archived FFPE tissues by global shotgun proteomics using LC–tandem MS (MS/MS) and quantitative targeted proteomics using SRM-based MS. Furthermore, we verified three novel and one known proteins immunohistochemically as the candidate proteins for clinical use.

Patients and Methods

  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Patient characteristics and FFPE tissue samples

The study subjects were selected from 156 histologically diagnosed PDAC cases undergoing pancreatectomy during January 1998 through December 2007 period at Tohoku University Hospital. The selection algorism of case samples used for shotgun and targeted proteomics and the workflow of this study are presented in Figure 1. First, we selected 144 cases excluded the perioperative mortality, suitable for monitoring. Second, we chose 103 cases with microscopically complete resection (R0) and no para-aortic lymph node metastasis. We focused on 87 cases with no neoadjuvant chemotherapy. Furthermore, to identify novel prognostic markers, we standardized already known prognostic factors: (i) pathological stage,3, 4, 7 (ii) histological differentiation,3, 4 (iii) postoperative carbohydrate antigen 19-9 (CA19-9) level6 and (iv) adjuvant chemotherapy.4, 5 We excluded three cases because the observation period was too short, leaving eight cases divided into two groups, poor outcome (n = 4) and better outcome (n = 4). There was a significant difference in postoperative median survival time calculated by the Kaplan–Meier method and compared using the log-rank test (poor, 21.2 months; better, 66.4 months; p = 0.0067). The poor and better outcome groups did not differ in age, gender, tumor location, tumor size, lymph node metastasis or distant metastasis at the time of resection (Table 1). In addition, noncancerous pancreatic ductal tissues (n = 5) were obtained from bile duct and ampulla of vater carcinoma patients who underwent pancreaticoduodenectomy during the same period at Tohoku University Hospital, and whose cancer cells did not extend into the pancreas (Fig. 1). In total, samples from 13 cases were utilized as the discovery set for proteomic analysis. A cohort of 87 cases served as the validation set for immunohistochemical analysis.

thumbnail image

Figure 1. Patient selection for proteomic analysis and strategies for shotgun proteomics, targeted proteomics and immunohistochemical confirmation. Histological R0 specimen shows microscopically negative margins. Union for International Cancer Control (UICC) stage IIB: Cancer has spread to nearby lymph nodes and may also have spread to adjacent tissues and organs. Japan Pancreas Society (JPS) stage 3: Cancer has not extended into the portal vein, extra-pancreatic nerve plexus, or other organs.

Download figure to PowerPoint

Table 1. Patients characteristics of the discovery set
inline image

We employed FFPE tissues for this proteomic study with approval from the Tohoku University Ethics Committee (The number of the ethical approval: 2006-119), and informed consent was obtained from individual patients.

LMD and protein extraction

Resected pancreatic tissues were fixed in 4% paraformaldehyde and routinely processed for paraffin sectioning. For shotgun proteomics, 10-μm sections were attached to DIRECTOR™ slides (Expression Pathology, MD), de-paraffinized three times with xylene for 5 min, rehydrated with graded ethanol solutions and distilled water, and then stained with hematoxylin. Stained, uncovered slides were air-dried and ∼30,000 cancerous and noncancerous pancreatic ductal cells (8 mm2) were collected into the cap of a 0.2-ml PCR tube using Leica LMD6000 (Leica Microsystems GmbH, Wetzler, Germany). Peptides were extracted using a Liquid Tissue™ MS Protein Prep kit (Expression Pathology) according to the manufacturer's instructions.16 Briefly, the cellular material, suspended in the liquid tissue buffer, was incubated at 95°C for 90 min, then cooled on ice for 3 min. Trypsin was added (∼15–18 U) followed by incubation at 37°C overnight. Dithiothreitol was added to a final concentration of 10 mM, and the samples were heated for 5 min at 95°C. The liquid tissue digestate was stored at −20°C until analysis.

Shotgun proteomics by LC-MS/MS

LC-MS/MS.

Exploratory LC−MS/MS analysis of a digested sample was performed using reversed-phase (RP) LC interfaced with an LTQ-Orbitrap hybrid MS (Thermo Fisher Scientific, CA) using a nanoelectrospray device (AMR, Tokyo, Japan), as previously reported.20 The RP-LC system (Paradigm MS4B, Michrom BioResources, CA) consisted of a peptide Cap-Trap cartridge (2.0 × 0.5 mm2 inside diameter) and an analytical column (L-column Micro, 150 × 0.2 mm2 L-C18, 3 μm, 12 nm, Chemical Evaluation Research Institute, Tokyo, Japan) fitted with an emitter tip (FortisTip, OmniSeparo-TJ, Hyogo, Japan). The Liquid Tissue™ solvents were evaporated, and peptides were redissolved with MS-grade water containing 0.1% trifluoroacetic acid and 2% acetonitrile. Peptide-mixture samples were loaded onto the trap cartridge and washed with mobile phase A (98% H2O with 2% acetonitrile and 0.1% formic acid) for concentration and desalting. Subsequently, the samples were eluted over 70 min from the analytical column via the trap cartridge using a linear gradient of 5−40% mobile phase B (10% H2O with 90% acetonitrile and 0.1% formic acid) at a flow-rate of 1 μl/min. General MS conditions were as follows: electrospray voltage, 3.0 kV, no sheath and auxiliary gas flow; ion transfer tube temperature, 200°C; collision energy (CE), 35%; ion selection threshold, 1,000 counts for MS/MS. MS/MS was performed on the top three ions in each MS scan using the dynamic exclusion principle, i.e., temporary (180 s) placement of a mass on an exclusion list after its MS/MS spectrum has been acquired.

Data analysis and protein identification.

All MS/MS spectral data were searched against Homo sapiens entries in the Swiss-Prot database (Release 57.13, 20,349 entries) using Mascot software (version_2.1.1, Matrix Science, London, UK). The peptide mass tolerance was 20 ppm, fragment mass tolerance 0.8 Da, and trypsin specificity was applied with a maximum of two missed cleavages. For variable peptide modifications, methionine oxidation and N-formylation, including formyl (K), formyl (R) and formyl (N-terminus), were taken into account. A p-value less than 0.05 was considered to be significant in protein identification. Results were obtained from triplicate LC-MS/MS runs for each sample.

Semiquantitative comparison using spectral counting.

To compare protein expression across all tissue samples from the results of shotgun analysis, we used the label-free spectral counting method.17 The number of peptide spectra with high confidence (Mascot ion score, p < 0.05) was used as the spectral count value. Fold changes in expressed proteins on a base-2 logarithmic scale were calculated using the protein ratio from spectral counting (Rsc).21 Rsc > 1 or < −1 corresponds to fold changes >2 or <0.5. Relative abundances of identified proteins were also obtained by using the normalized spectral abundance factor (NSAF).22 Candidate proteins of the two groups were chosen based on Rsc > 1 or < −1, and also statistical significance as indicated by p < 0.05 according to the nonparametric G-test.23

Targeted proteomics by SRM-based MS

The candidate proteins identified by shotgun proteomics and spectral counting were verified by SRM-based MS analysis. The capillary RP LC-MS/MS system comprised a Paradigm MS4 (Michrom BioResources) connected to a 4000 QTRAP hybrid system (AB Sciex, CA) operating in positive ion mode. For all SRM studies, quadrupoles were operated at unit/unit resolution, and CE was determined using the equation: CE = 0.044 × m/z + 6 for doubly charged precursor ions. The scheduled SRM (sSRM) mode was utilized in this study, with the sSRM detection window set at 180 s. The internal standard used was the specific SRM transition of the peptide AGFAGDDAPR (m/z 488.7), which is the doubly charged actin beta (ACTB) peptide, to the singly charged fragment (m/z 630.3).18 This internal standard is referred to as the in-sample internal standard (ISIS) since ACTB is a housekeeping protein.24 Peak areas of each transition were normalized using the equation: Normalized peak area = peak area × (500,000/peak area of 488.7/630.3). All sequences were checked using the BLAST program [National Center for Biotechnology Information] and compared with the Swiss-Prot human database. The averaged values of the poor and better outcome groups, and of the cancerous and noncancerous tissues, based on triplicate runs, were compared. Expression differences of at least 2- or 0.5-fold, respectively, were defined as significant.

Immunohistochemistry

For immunohistochemical analysis, 4-μm FFPE tissue sections were de-paraffinized with xylene and rehydrated with ethanol solutions and distilled water. Antigen retrieval was performed by heating the sections in 10 mmol/L citrate buffer (pH 6.0) at 121°C for 5 min with an autoclave. Rabbit polyclonal enoyl CoA hydratase 1 (ECH1) antibody (HPA002907, Sigma, MO, 1:250, http://www.sigmaaldrich.com/catalog/product/sigma/hpa002907?lang=en&region=US), mouse monoclonal glucose transporter type 1 (GTR1) antibody (ab40084, Abcam, MA, 1:200, http://www.abcam.com/Glucose-Transporter-GLUT1-antibody-SPM498-ab40084.html), rabbit polyclonal olfactomedin-4 (OLFM4) antibody (ab96280, Abcam, 1:100, http://www.abcam.com/OLFM4-antibody-ab96280.html) and mouse monoclonal stomatin-like protein 2 (STML2) antibody (60052-1-Ig, ProteinTech Group, IL, 1:100, http://www.ptglab.com/Products/STOML2-Antibody-60052-1-Ig.htm) were used as the primary antibodies. After blocking endogenous peroxidase with methanol containing 0.3% hydrogen peroxidase, the labeled antigens were detected by the horseradish peroxidase EnVision+ System (DAKO, Glostrup, Denmark) and visualized by 3,3′-diaminobenzidine tetrahydrochloride as a chromogen. The sections were lightly counterstained with hematoxylin. Appropriate positive controls were used, in part with reference to the Human Protein Atlas (http://www.proteinatlas.org/). Negative control experiments were conducted by replacing the primary antibody with phosphate-buffered saline.

After completely reviewing all slides of immunostained sections for each carcinoma, three of the authors (T.T., T.O. and T.S.) independently and blindly classified the cases into two groups. GTR1 immunostaining was quantified by grading the proportion of cancerous and noncancerous pancreatic ductal cells showing strong and distinctive membranous immunoreactivity. If GTR1 staining was absent or less than 5% of cells were immunoreactive, the specimen was classified as negative, while specimens in which more than 5% of cells stained for GTR1 were classified as positive.25 There have been no papers regarding ECH1, OLFM4 and STML2 expression in PDAC. Thus, if the percentages of cancerous and noncancerous pancreatic ductal cells staining positive for one of these proteins exceeded 10%, we categorized that specimen as positive for the protein being examined, while the negative group for the relevant protein comprised those with fewer than 10% positive cells.

Statistical analysis

The χ2-test was used to compare categorical variables. In univariate analysis, survival rates were calculated by the Kaplan–Meier method and compared using the log-rank test. p < 0.05 was considered statistically significant. JMP software version 9.0 (SAS Institute, NC) was used for all analyses.

Results

  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Protein identification by shotgun proteomics and semiquantitative comparison

In shotgun proteomic analysis, 845 proteins in the better outcome group, 924 in the poor outcome group, and 730 in the noncancerous pancreatic ductal tissue, i.e., 1,229 proteins in total, were identified (Fig. 2a). The identified proteins were semiquantitatively compared using spectral counting analysis (Figs. 2b and 2c). The graph is nearly symmetrical around its center, and housekeeping proteins such as ACTB and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were highly expressed in cancerous and noncancerous cells with minimum variation in this semiquantitative spectral counting analysis. Thus, ACTB peptide was utilized as ISIS for SRM-based targeted proteomics.

thumbnail image

Figure 2. (a) Venn map of proteins identified from four cases with poor outcomes and four cases with better outcomes, i.e., longer survival, and five noncancerous pancreatic ducts. (b, c) The protein ratio from spectral counting (Rsc) and the normalized spectral abundance factor (NSAF) values calculated for the proteins identified (x-axis). A comparison of protein expressions: (b) Poor outcome group versus better outcome group. Proteins significantly overexpressed in the poor outcome group are near the right side of the x-axis. (c) Cancerous versus noncancerous cells. Proteins significantly overexpressed in cancerous cells are near the right side of the x-axis. Actin beta (ACTB) is located near the center of the x-axis in (b) and (c).

Download figure to PowerPoint

Candidate proteins were chosen based on Rsc > 1 or < −1, and also statistical significance (p < 0.05 by nonparametric G-test). We selected 106 proteins with different expression levels in the poor and better outcome groups, including 53 proteins overexpressed in the poor and 53 overexpressed in the better outcome group. In addition, we chose 64 proteins overexpressed in cancerous as compared to noncancerous tissue, though there was no significant difference between the poor and better outcome groups by semiquantitative comparison. Thus, in total, we focused on 170 candidate proteins.

Quantitative verification by SRM-based targeted proteomics

To confirm expressions of the candidate proteins, we quantitatively analyzed the discovery set with shotgun proteomic data using SRM-based targeted proteomics. Targeted peptides for SRM were selected with reference to data obtained from the shotgun LC-MS/MS analysis. Unfortunately, in 47 of the 170 candidate proteins, we could not identify specific peptides suitable for quantitative analysis of the targeted proteins. For the 123 candidate proteins for which specific peptides were identified, preliminary analysis of the project control (mixtures of equal aliquots of all patient samples) was conducted to select readily detectable SRM transitions and to confirm the retention time of each peptide. In 12 of the 123 candidates, SRM transitions were not detectable by preliminary analysis of the project control (mixtures of equal aliquots of all patient samples). We added ACTB (ISIS) to the remaining 111 proteins, such that an SRM assay consisting of 112 proteins (183 peptides and 697 transitions) with sufficient sensitivity was finally developed. The 69 proteins differed in expression levels between the poor and better outcome groups and 42 were overexpressed in cancerous as compared to noncancerous tissue in the shotgun analysis, along with ACTB.

The SRM quantitative analysis verified that 35 proteins were overexpressed in cancerous as compared to noncancerous tissue. Among them, fourteen proteins showed different expressions in the poor and better outcome groups (Table 2). Seven proteins were differentially expressed by poor and better outcome groups, but the expression level as a whole was not different from noncancerous ductal tissue. (Of these seven, four proteins were less expressed in better group, while three were less expressed in poor group.)

Table 2. List of the candidate proteins by SRM analysis
inline image

Validation by immunohistochemical analysis

We focused on the 14 proteins that showed different expression levels in the poor and better outcome groups as well as being overexpressed in cancerous tissue (Table 2, Fig. 3a). First, to confirm expressions of the 14 candidate proteins in tissue sections, we immunohistochemically analyzed the discovery set. Some of these proteins were not immunoreactive, while others were equally positive in both cancerous and noncancerous cells for the available antibodies. Among the 14 candidate proteins, we found that the immunohistochemical results of ECH1, GTR1, OLFM4 and STML2 were most compatible with those of SRM analysis (Fig. 3b). ECH1, OLFM4 and STML2 were overexpressed in the poor outcome group, while GTR1 was overexpressed in the better outcome group. Second, we immunohistochemically analyzed the 87 validation set (Fig. 1, 56 men, 31 women; median age, 65 years) suitable for monitoring, with microscopic complete resection (R0), and no neoadjuvant chemotherapy, to confirm the associations between these four proteins and the outcomes after resection. Median survival of patients in this cohort was 19.1 months and 22.7% were alive at 5 years. The correlation between the immunoreactivity and the clinicopathological parameters of the 87 cases are shown in Table 3. A significant correlation was detected between high GTR1 expression and more aggressive histological differentiation (p = 0.0420), and higher postoperative CA19-9 level (p = 0.0376). There was no significant correlation between the expression of the other three proteins and examined clinicopathological parameters. Kaplan–Meier analysis revealed patients with cancer cells expressing ECH1, OLFM4 and STML2 to have significantly worse overall survival (OS) than those without these proteins (Fig. 3c, ECH1, p = 0.0204; OLFM4, p = 0.0435 and STML2, p = 0.0408), as expected from SRM analysis. Contrary to the expectation from SRM results, but in accordance with the more aggressive differentiation and higher postoperative CA19-9, a high level of GTR1 expression also correlated with poor OS (Fig. 3c, p = 0.0056).

thumbnail image

Figure 3. (a) Scatter plot of the normalized peak area by SRM-based quantitative analysis. The levels of expressions of these 14 proteins differed between the poor and better outcome groups by at least 2- or 0.5-fold and were overexpressed in cancerous as compared to noncancerous tissue by at least 2-fold. (b) Representative immunohistochemical staining results for four candidate proteins in noncancerous pancreatic duct, and the poor and better outcome groups. Positive staining results for (A) ECH1, (B) GTR1, (C) OLFM4 and (D) STML2 are shown in brown. Scale bar = 100 μm, original magnification, ×200. (c) Survival analysis based on immunoreactivities of the four candidate proteins by the Kaplan–Meier method. (A) Positive staining results for ECH1, (B) GTR1, (C) OLFM4 and (D) STML2 were significantly associated with poor OS.

Download figure to PowerPoint

Table 3. Patients characteristics of the validation set
inline image

Discussion

  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

PDAC is an aggressive malignancy and carries an extremely poor prognosis due to delayed diagnosis, early metastasis and resistance to most cytotoxic agents.26 Thus, it is very important to find new diagnostic, prognostic and therapeutic biomarkers. Various proteomic studies of pancreatic tissues, juice, serum/plasma, and cell lines have recently attempted to identify proteins expressed differently in PDAC to unravel the abnormal signaling pathways underlying oncogenesis, and to detect new biomarkers.9, 19, 26 However, there have been only a few reports regarding MS-based proteomic analysis using FFPE tissues from PDAC cases. We selected poor and better outcome groups of resectable, node positive PDAC carefully adjusted for clinicopathological backgrounds and collected cells of interest from FFPE tissue by LMD. The present study design allowed us to use the best advantages of proteomic analysis with FFPE tissues. However, it is ideal to increase the number of discovery set samples. We and others performed the power calculation for this type of proteomics using FFPE (data not shown). As the number of individuals increase, the number of proteins detected by shotgun analysis increases, but it tends to reach plateau after exceeding 10 samples. As we show in Figure 1, four patients with better outcome and four patients with poor outcome were max of our available specimen. Since the operative procedure, clinical assessment of extent, pathological evaluation and postoperative therapy may different interinstitutionally, we preferred to match the patient background using our samples, rather than compromised with other uncontrollable factors.

The foundation for any biomarker discovery effort is identification of proteins showing different expressions in disease and control samples.27 Global shotgun proteomics has advantages of broad proteome coverage in unbiased fashion, thereby increasing chances to discover novel biomarkers, although largely qualitative rather than quantitative.28 In shotgun proteomics, two fundamental quantitative approaches have emerged, i.e., the label-free and the stable isotope labeling methods. We used the label-free spectral counting approach for selecting candidate proteins, because this method is relatively easy, inexpensive, and requires no chemical modifications. However, semiquantification by spectral counting is not fast enough to probe every ion detected, especially low abundant ions.29 Therefore, we confirmed expressions of the candidate proteins using SRM-based targeted proteomics.

Quantification of hundreds of candidates by affinity-based methods, the broadly used Western blot or ELISA approaches, is impractical, and are also limited in abilities to detect multiple proteins in the same sample.14 Recently, quantification assays based on SRM have been extensively investigated for protein verification purposes.

Quantitative analysis by SRM has some limitations. Although proteins give rise to tens or hundreds of tryptic peptides, only some of these peptides are suitable for quantitative analysis. The choice of peptides suitable for SRM was guided primarily by four important criteria: the peptides should (i) be tryptic and contain no missed cleavage sites, (ii) have a high score (Mascot), (iii) contain no cysteine residues (to avoid carbamidomethylation) and (iv) have a sequence specific for the selected protein.30 Unfortunately, in 47 of the 170 candidate proteins, we were not able to identify specific peptides suitable for SRM, and in 12 proteins, SRM transitions were not detectable in the preliminary analysis of the project control.

SRM quantitative analysis revealed 42 proteins in total to be potential diagnostic or prognostic markers (Table 2). We identified previously reported PDAC markers (CEA-related cell adhesion molecule 6, Protein S100-P) in an unbiased fashion, suggesting this workflow to be extremely useful for biomarker research in a high-throughput setting. In this study, we focused on 14 proteins showing different expression levels in the poor and better outcome groups and that were overexpressed in cancerous as compared to noncancerous tissues. These 14 proteins are potential prognostic markers and therapeutic targets because of its low expression in noncancerous tissue. For the other 28 proteins besides 14 proteins, further study is needed in the future.

From among these 14 proteins, we found ECH1, GTR1, OLFM4 and STML2 match with SRM findings by immunohistochemical validation, and then performed large scale validations for these four proteins. Kaplan–Meier analysis revealed patients with carcinoma cells expressing ECH1, OLFM4 and STML2 to have significantly worse OS than those without these proteins, as expected from SRM analysis (Fig. 3c). The expression of GTR1 resulted in lower survival rate in Kaplan–Meier analysis, contrary to the expectation from SRM result, but in accordance with the relations with invasive factors (Table 3, Fig. 3c). We did not perform Western blot analysis using these antibodies to reconfirm the specificity. Even in manufacturers' data sheet available in each URL, Western blot using the antibodies may result in multiple bands, that is, due to nonspecific binding, sample preparation, experimental condition or other reasons. Western blot also do not always guarantee the specificity of immunohistochemistry. These limitations of immunohistochemistry are, however, mostly supplemented by the specific peptide proof by MS analysis. Since we carefully chosen the antibodies that completely matched with the SRM results in discovery set, we think the immunohistochemical results in validation set are reliable.

GTR1 expression is reportedly associated with a worse prognosis in PDAC patients.25, 31 On the other hand, ECH1, OLFM4 and STML2 were newly discovered to be associated with the postoperative outcomes of PDAC patients.

ECH1, enoyl CoA hydratase 1, is an auxiliary enzyme of the unsaturated fatty acid β-oxidation pathway.32 Various studies have suggested ECH1 to be associated with tumor progression.33–36 Abnormal expression of ECH1 has been linked to hepatocellular carcinoma secondary to hepatitis C virus infection35 and the pathogenesis of gastric cancer.33 ECH1 was upregulated in the Hca-F cell line, from a mouse hepatocellular carcinoma with a high rate of lymphatic metastasis.34 In addition, downregulation of ECH1 can inhibit the cell proliferation of Hca-F, while maintaining more Hca-F cells in the S phase and fewer in the G1 phase, and reduce the adhesion and migration capacities of Hca-F calls.36

OLFM4, olfactomedin-4, a glycoprotein originally described as human granulocyte colony stimulating factor stimulated clone-1 (hGC-1), was identified as the novel antiapoptotic molecule GW112.37 OLFM4 binds to the apoptosis-promoting factor GRIM-19 to inhibit its function, and decreases the expressions of GADD153 and c-Abl which are known to be involved in apoptosis via various stress factors leading to DNA damage.37 Aberrant expression of OLFM4 has been observed in some cancerous tissues.38 In PDAC cells, OLFM4 mRNA is highly expressed during the S phase of the cell cycle, and the cell cycle is arrested at S phase by downregulation of OLFM4 mRNA expression.39 This observation indicates that OLFM4 promotes proliferation of PDAC cells by favoring transition from the S to the G2/M phase.

STML2, stomatin-like protein 2, was cloned and characterized in human erythrocytes.40 STML2 is a member of the stomatin superfamily, based on its stomatin consensus signature sequence. STML2 and stomatin have, however, very low overall homology, and STML2 lacks the typical amino-terminal transmembrane domain, present in other stomatins. STML2 was shown to be overexpressed in many human cancer tissues, including esophageal squamous cell carcinoma41 and lung cancer.42 Although the function of STML2 remains unknown, STML2 has been suggested to link stomatin or other integral membrane proteins to the peripheral cytoskeleton and to play a role in regulating ion channel conductance or the organization of sphingolipids and cholesterol-rich lipid rafts.40 Furthermore, antisense transfection of the STML2 gene was reported to decrease cell growth, tumorigenicity and cell adhesion.41

There were no significant correlations between the expression of ECH1, OLFM4 and STML2 and clinicopathological parameters of validation set. These results suggest that we could identify these new independent prognostic markers by this screening tactics.

GTR1 (also called GLUT1), glucose transporter member 1, is a member of the GLUT family of facilitative glucose transporters that mediate Na+-independent cellular uptake of glucose.43 Previous studies have shown enhanced glycolytic metabolism in malignant neoplasms.44 Increased glucose uptake is one of the major metabolic changes found in malignant tissues. GTR1 is reportedly overexpressed in a wide range of human cancers, and studies have shown that GTR1 expression correlates with aggressive behavior and poor prognosis.25, 31, 45, 46 In this study, GTR1 positive group contained more cases of moderately and poorly differentiated adenocarcinoma and higher level of postoperative CA19-9 than GTR1 negative group (Table 3) and showed significantly lower survival rate (Fig. 3c), contrary to the expectations from SRM results (Table 2, Figs. 3a and 3b). This may suggest the limitation and possibility of biomarker search using state-of-the-art proteomics technology GTR1 was definitely over expressed in cancerous tissue compared to the noncancerous ductal tissue (Table 3). The expression ratio poor/better group was 0.4. Suggesting GTR1 expression may correlate with some factors associated with better group. It is necessary to validate the biomarkers using the appropriate validation set and methods.

Both, poor and better group were treated by gemcitabine (GEM) postoperatively. Thus, these proteins may alter the sensitivity of cancer cells to GEM. The cohort of 87 validation set includes 52 treated with GEM, 23 treated with 5-fluorouracil (5FU) and 12 with no adjuvant chemotherapy. In addition, 29 patients with adjuvant chemotherapy treated with GEM based or other chemotherapies after recurrence. Thus, to determine the relationship between the expression of these four proteins and GEM sensitivity, further cellular and molecular biological investigation is required.

Although further study is needed to determine the precise roles of these proteins in the malignant behaviors of cells, their expressions may have prognostic significance as well as potentially being useful for selecting PDAC treatments. Herein, the expressions of candidate proteins were confirmed immunohistochemically, though serum levels of some of these proteins have already been reported to possibly contribute to predicting outcomes of other cancers. It would be interesting to examine whether serum levels of the candidate proteins are associated with poor outcomes in larger numbers of postoperative PDAC patients. In addition, endoscopic ultrasonography-fine needle aspiration (EUS-FNA) and brushing cytology have recently been recognized as very useful diagnostic tools in PDAC. Thus, immunocytochemical study of candidate proteins using tissue samples obtained by these methods may aid in diagnosing and selecting optimal treatments for PDAC patients preoperatively. As no candidate proteins other than ECH1, GTR1, OLFM4 and STML2 showed good immunoreactivity for the available antibodies, further study using other antibodies or nonimmunohistochemical methods is needed. Among these proteins, there may be novel and useful diagnostic or prognostic markers and therapeutic targets for PDAC.

In conclusion, we identified several proteins possibly associated with postoperative outcomes of PDAC using MS-based proteomics approaches with archived FFPE tissues. By careful matching of the specimen in discovery set, we could identify independent prognostic markers of resectable, node positive PDAC. Immunohistochemically, we demonstrated high expressions of ECH1, GTR1, OLFM4 and STML2 to be significantly associated with poor OS by Kaplan–Meier analysis. Measurement of the expressions of these proteins is potentially useful for obtaining diagnostic, prognostic and treatment information for PDAC patients. This strategy, making optimal use of FFPE tissues with their corresponding pathological and clinical records, offers new opportunities to identify biomarkers and therapeutic targets.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

We thank Emiko Shibuya and Keiko Inabe (Department of Surgery, Tohoku University, Sendai, Japan), Megumi Maeda and Seiko Yamada (Leica Microsystems K.K., Tokyo, Japan) and Taro Takemura (National Institute for Materials Science, Tsukuba, Japan) for technical assistance.

References

  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References