Novel biomarkers distinguishing pancreatic head Cancer from distal cholangiocarcinoma based on proteomic analysis

We retrospectively searched patients with PHC and DCC, who underwent surgical resection between April 1998 and December 2016 in our hospital. None of the patients received radiation or neoadjuvant chemotherapy prior to surgery. ICC, perihilar cholangiocarcinoma, gallbladder cancer, and carcinoma of the ampulla of Vater were excluded. We adopted well, moderate, and poorly differentiated adenocarcinomas; rare histologic variants were not included. In total, 99 PHCs and 96 DCCs were examined. The pathological diagnosis and clinicopathological stage were determined according to the Union for International Cancer Control (UICC) 7th edition [22]. We selected 10 PHCs and 8 DCCs for proteomics. The remaining 89 PHCs and 88 DCCs samples served in the validation phase. We divided these 177 samples into three groups: small cohort group, large cohort group, and verification cohort group. The small cohort group (cohort 1) was composed of 12 PHCs and 12 DCCs for immunohistochemical screening. First, we performed immunohistochemical staining on the small cohort group and selected candidate proteins to minimize time and material consumption. The large cohort group (cohort 2), which was comprised of 72 PHCs and 74 DCCs, was used for the evaluation of diagnostic performance by immunohistochemistry. Based on immunohistochemical results of the large cohort group, we selected candidate proteins and generated biomarker panels consisting of three candidate proteins to improve diagnostic accuracy. All types of biomarker panels were evaluated in terms of sensitivity, specificity, positive likelihood ratio, and negative likelihood ratio. The remaining 5 PHC and 2 DCC cases were defined as clinically difficult cases to diagnose (cohort 3). These cases had completely opposite diagnoses between pre- and post-operation; 5 patients diagnosed as DCC preoperatively were consequently revealed as PHC by pathological findings, and vice versa. The pathologic diagnoses were carefully re-reviewed by a pathologist with an expertise in pancreaticobiliary malignancies to ensure accuracy. Finally, we applied the biomarker panels to the clinically difficult cases to verify the effectiveness of the biomarker panels.

Target lesions were initially detected on serial sections stained with hematoxylin and eosin. For laser micro-dissection, 10-μm sections were cut onto DIRECTOR slides (Expression Pathology, Gaithersburg, MD, USA). All sections were deparaffinized three times with xylene 5 min, rehydrated with graded ethanol solutions and distilled water, stained with hematoxylin, and then air-dried. Leica LMD6000 (Leica Microsystems GmbH, Germany) was used, and about 30,000 cancer cells (8 mm²) were collected into the cap of a 0.2 mL polymerase chain reaction tube. Peptide extraction was performed with a Liquid Tissue MS Protein Prep kit (Expression Pathology) following manufacturer’s instructions.

All peptide-mixture samples were analyzed in LC-MS/MS using a Finnigan LXQ linear ion-trap mass spectrometer (Thermo Fisher, San Jose, CA, USA) [18, 23]. A capillary reverse phase LC-MS/MS system (ZAPLOUS System: AMR, Tokyo, Japan) composed of a Paradigm MS4 (Michrom BioResources, Auburn, CA, USA), an HTC PAL autosampler (CTC Analytics, Zwingen, Switzerland), and a Finnigan LXQ linear ion-trap mass spectrometer was equipped with an ADVANCE nanospray ionization source (Michrom BioResources).

Protein identification was performed using MASCOT software (version 2.3.02, Matrix Science, UK). The MS/MS spectral raw data were searched against Homo sapiens entries in the Swiss-Prot database (Release 2011.03, 20,234 entries). Peptide and fragment mass tolerants were 2.0 Da and 1.0 Da, respectively. Trypsin specificity was applied with a maximum of 2 missed cleavages. Methionine oxidation and N-formylation including formyl (K), formyl (R), and formyl (N-terminus) were considered as variable modification. A P-value less than 0.05 was considered significant. Reported results were obtained from triplicate LC-MS runs for each sample.

The spectral counting method was utilized for comparison of protein expressions across all tissue samples. The spectral count value was determined by the number of peptide spectra with high confidence (Mascot ion score, P <  0.05). Fold changes in the expressed proteins on a base-2 logarithmic scale were calculated using the ratio from spectral counting (RSC) as defined by Old et al. [24]. RSC > 1 or < − 1 was consistent with fold changes > 2 or <  0.5, respectively. Spectral index (SpI) was also used to assess the differences in relative abundances of identified proteins between the 2 groups. SpI values ranged from − 1 to + 1 with values near 0 indicating similar relative peptide abundance between the 2 groups. Proteins satisfying the following conditions were candidates for the study: RSC > 1 or < − 1, SpI > 0.6 or < − 0.6, and P <  0.01 by the non-parametric G-test.

FFPE samples from 177 patients were used for a validation analysis. Sections (4-μm thick) cut from the FFPE blocks were deparaffinized with xylene and rehydrated with graded alcohol followed by distilled water. For antigen retrieval, tissue slides were heated in an autoclave at 120 degrees Celsius for 5 min in a citrate acid buffer (10 mM citric acid, pH 6.0). Sections were covered with the diluted primary antibody and incubated at 4 degrees overnight. Endogenous peroxidase activity was blocked by methanol containing 0.3% hydrogen peroxidase. The labeled antigens were identified by the horseradish peroxidase EnVision System (DAKO, Glostrup, Denmark) after incubating for 60 min at room temperature. Color development was visualized with 3,3′-diaminobenzidine tetrahydrochloride (Dojindo Lab, Kumamoto, Japan). The sections were lightly counterstained with hematoxylin. Technical details for immunohistochemistry are given in Additional file 1: Table S1.

Two authors (T. T. and T. S.) evaluated all slides of immunostained sections and scored for percentage of immunoreactive cells (labeling index). Immunoreactivity was regarded as positive if more than 10% tumor cells showed staining. Nuclear immunoreactivities for annexin A10 (ANXA10) and parathymosin (PTMS) were assessed, while cytoplasmic staining was considered positive for the other 13 proteins.

All statistical analyses were performed using JMP software version 13.0 (SAS Institute, Cary, NC, USA). Fisher’s exact test was applied to assess the statistical significance of differences in the validation phase. To evaluate diagnostic performance, we used 2 by 2 contingency tables with binomial 95% confidence intervals.

In the shotgun proteomic analysis, we identified 1361 proteins in PHC and 1274 in DCC. In total, 1820 proteins were identified. The identified proteins were semi-quantitatively compared using the spectral counting method. Candidate proteins were chosen based on Rsc > 1 or < − 1, SpI > 0.6 or < − 0.6, and also statistical significance (p <  0.01 by nonparametric G test). We selected 18 proteins with different expression levels in the PHC and DCC groups. Shotgun proteomics revealed 5 proteins to be overexpressed in PHC as compared with in DCC, and 13 proteins to be overexpressed in DCC. Three ribosomal proteins were excluded from further analysis. Therefore, we set these 15 proteins as candidates for the validation phase (Table 2).

The 15 candidates were first validated in the small cohort group including 12 PHC and 12 DCC to access cell type specific staining. Table 3 shows the positive rate for each protein in both tumor groups. Among the 15 proteins, we found the immunohistochemical results of the small cohort group were compatible with those of the shotgun proteomic analysis. We imposed the conditions to select for further evaluation; the positive rate was higher than 50% in either group and fold change between the two groups was more than 1.5-fold. ANXA10, cytokeratin-17 (KRT17), transmembrane protein 109 (TMEM109), sodium/potassium-transporting ATPase subunit beta-1 (ATP1B1), and PTMS demonstrated favorable results and were, therefore, transferred to the next validation stage using the large cohort group of 72 PHC and 74 DCC. The representative staining patterns of these 5 proteins are shown in Additional file 2: Figure S1.

The result of the immunohistochemical analysis in the large cohort group is described in Table 3. The diagnostic performance of candidate proteins when discriminating between PHC and DCC are listed in Table 4. All 5 proteins yielded p values below 0.05 in the Fisher’s exact test. KRT17 stained 76.4% of PHC and 28.4% of DCC. KRT17 provided the best diagnostic performance for the diagnosis of PHC (76.4% sensitivity, 71.6% specificity, 2.69 positive likelihood ratio, 0.33 negative likelihood ratio). Similar to KRT17, ANXA10 staining was frequently positive in PHC (81.9%) and moderately in DCC (48.7%). KRT17, ANXA10, and TMEM109 were assumed to be expressed in PHC by proteomic analysis; whereas, PTMS and ATP1B1 demonstrated an overexpression in DCC. Immunohistochemical staining of PTMS revealed more positively in DCC (60.8%) compared to PHC (27.8%). Positive staining for ATP1B1 was observed in 60.8% of DCC and 41.7% of PHC samples. PTMS exhibited better performance (60.8% sensitivity, 72.2% specificity, 2.19 positive likelihood ratio, 0.54 negative likelihood ratio) than ATP1B1.

In addition to single diagnostic markers, we generated biomarker panels consisting of our candidate proteins to improve diagnostic accuracy. As the result of the staining patterns in the large cohort group, PHC exhibited a KRT17+/ANXA10+/PTMS- staining pattern, which was seen in 34 cases (47.2%). Six of 74 DCC cases (8.1%) showed this combinatorial pattern. When used for the diagnosis of PHC, the sensitivity, specificity, and positive/negative likelihood ratio were 47.2, 91.9% and 5.82/0.574, respectively (Table 4). In contrast, DCC had a tendency to show a KRT17−/ANXA10−/PTMS+ staining profile in 17 cases (23.0%). Of 72 PHC cases, this immunoreactive pattern was demonstrated in only 2 cases (2.8%). The sensitivity, specificity, and positive/negative likelihood ratio of the KRT17−/ANXA10−/PTMS+ immunophenotype were 23.0, 97.2%, and 8.27/0.792, respectively, for the diagnosis of DCC (Table 4). These panels yielded low sensitivity, but high specificity compared with single diagnostic markers.

We selected 5 PHC and 2 DCC clinically difficult cases, which had opposite diagnoses between preoperative investigations and pathological findings. We applied 5 candidate proteins to these cases to achieve correct diagnoses (Fig. 2). KRT17+/ANXA10+/PTMS- panel diagnosed correctly 4 of 5 PHC cases, which were considered as DCC in preoperative imaging; whereas, KRT17−/ANXA10−/PTMS+ panel diagnosed 2 of 2 DCC cases correctly.