Proteomic characterization of ovarian cancers identifying annexin‐A4, phosphoserine aminotransferase, cellular retinoic acid‐binding protein 2, and serpin B5 as histology‐specific biomarkers

Abstract

Numerous studies have suggested that the different histological subtypes of ovarian carcinoma (i.e. clear cell, endometrioid, mucinous, and serous) have distinct clinical histories and characteristics; however, most studies that have aimed to determine biomarker have not performed comprehensive analyses based on subtype specificity. In the present study, we performed two‐dimensional gel electrophoresis‐based differential proteomic analysis of the different histological subtypes of ovarian carcinoma using tissue specimens from 39 patients. Seventy‐seven protein spots (55 unique proteins) were found to be up‐ or downregulated in a subtype‐specific manner. The most significant difference was observed for: (i) annexin‐A4 (ANXA4) and phosphoserine aminotransferase (PSAT1), which are expressed strongly in clear cell carcinoma; (ii) cellular retinoic acid‐binding protein 2 (CRABP2), which is expressed specifically in serous carcinoma; and (iii) serpin B5 (SPB5), which is upregulated in mucinous carcinoma. Validation of these candidates by western blotting using a 34 additional test sample set resulted in an expression pattern that was consistent with the screening and revealed that differential expression was independent of cancer stage or tumor grade within each subtype. Thus, the present study reinforces the notion that ovarian cancer subtypes can be clearly delineated on a molecular basis into four histopathological groups, and we propose that ANXA4, PSAT1, CRABP2, and SPB5 are candidate subtype‐specific biomarkers that can help define the basis of tumor histology at a molecular level. (Cancer Sci 2012; 103: 747–755)

Epithelial ovarian carcinomas (EOC) are grouped histopathologically into serous, endometrioid, mucinous, and clear cell carcinomas. These subtypes have been suggested to differ in their oncogenic mechanisms,1 precursor lesions,2, 3 risk factors,4 and molecular signatures as determined by mRNA expression profiling.5 Clinically, EOC subtypes respond differently to chemotherapy. For example, clear cell carcinoma is relatively resistant to chemotherapy, with approximately 30% of cases responding to combination therapy with carboplatin and paclitaxel (TC regimen),6 resulting in a relatively poor prognosis and high recurrence rate.7 Clear cell carcinoma is also unique in that its incidence is markedly higher in Japan than in Western countries and is rising.8 Although current clinical practices regard EOC as a single disease entity and treatment regimens are not subtype specific, the evidence from the literature suggests that EOC is a multifaceted disease with subtypes that have distinct clinical features. However, the recent NCI State of Science meeting proposed that separate clinical trials for mucinous and clear cell EOC subtypes be undertaken.9 Moreover, a recent study argued that biomarkers that take subtype specificity into account could convey more information by demonstrating that the association between prognosis and Wilms’ tumor‐1 expression was unique to the serous subtype.10

The aim of the present study was to discover better candidate biomarkers for EOC that were associated with the histopathological classification by identifying proteins that were correlated with specific subtypes. Using clinical tissue specimens we undertook the expression profiling of clear cell, endometrioid, mucinous and serous carcinoma using two‐dimensional gel electrophoresis (2‐DE)‐based differential proteomic analysis. Although extensive profiling studies have been performed using mRNA microarray,11, 12, 13, 14, 15 only a handful of studies have undertaken proteomic analysis of EOC16, 17, 18, 19 and many of the studies have experimental shortcomings, such as insufficient sample size, analysis of only a single subtype, or analyses based on cell lines.

To overcome these issues, we used 39 fresh frozen ovarian cancer specimens for differential analysis and 34 additional specimens for validation of a number of promising targets. This is the first report to describe the comprehensive proteomic expression profile of EOC subtypes.

Materials and Methods

Ovarian cancer specimens

Ovarian cancer samples were collected from patients who had provided informed consent by Keio University Hospital, with the approval of the Ethics Committee of Keio University (Approval No. 15‐96‐6). The characteristics of the samples used are summarized in Table 1, with details provided in Table S1. Resected tumor samples were snap frozen in liquid nitrogen and stored at −150°C until protein extraction. Cancer staging in the present study adhered to the International Federation of Gynecology and Obstetrics (FIGO) criteria.20

Table 1.

Summary of the ovarian cancer specimens used in the present study

	Epithelial ovarian carcinoma subtype
	Clear cell	Endometrioid	Mucinous	Serous
n	13	10	6	10
FIGO stage
Stage I	10	4	6	0
Stage II	0	2	0	5
Stage III	3	4	0	5
Tumor grade
Grade 1	N/A	5	6	0
Grade 2	N/A	2	0	6
Grade 3	N/A	3	0	4
Mean age at diagnosis (years)	55	48	54	54

FIGO, the International Federation of Gynecology and Obstetrics.

The present study was approved by the Keio University Institutional Review Board.

Reagents

DeStreak Reagent, IPG buffer, Immobiline DryStrip gel, HRP‐conjugated anti‐rabbit IgG, the ECL detection kit and chemiluminescence films were purchased from GE Healthcare (Buckinghamshire, UK). The HRP‐conjugated anti‐β‐actin antibody (Sp2/0‐Ag14) was obtained from Abcam (Cambridge, UK). Sequence grade modified trypsin was from Promega (Madison, WI, USA). ZipTip μC18 was from Millipore (Madison, WI, USA). The MALDI matrix, α‐cyano‐4‐hydroxycinnamic acid, was purchased from Shimadzu‐GLC (Tokyo, Japan).

Two‐dimensional gel electrophoresis

Frozen tissue blocks were first pulverized in liquid nitrogen, transferred to screw‐cap microtubes, and homogenized in buffer composed of 50 mM HEPES–NaOH, pH 7.5, 100 mM NaCl, 2% CHAPS, and 1% Triton X‐100. Lysis was performed by using a ball mill homogenizer operated at 4000/min for 1 min with zirconium beads (1 mm diameter; Tomy Seiko, Tokyo, Japan). After centrifugation at 20 000g for 15 min, the supernatant was collected for 2‐DE analysis. For 2‐DE, tissue extracts containing 300 μg total protein were desalted by 15% trichloroacetic acid precipitation followed by cold ethanol/ether washing and then redissolved in buffer containing 6 M urea, 2 M thiourea, 2% CHAPS, 1% Triton X‐100, 0.5% DeStreak Reagent and 0.5% IPG buffer. Isoelectric focusing was performed using a 13‐cm Immobiline DryStrip, pH 3–10, non‐linear gel, whereas a 10–18% linear gradient polyacrylamide gel was used for second‐dimension SDS‐PAGE. Protein spots were visualized by 0.1% Coomassie brilliant blue G‐250 stain and gel images were acquired using a GS‐800 Imaging Densitometer (Bio‐Rad Laboratories, Hercules, CA, USA).

Analysis of 2‐DE data

The 2‐DE gel images were analyzed using Progenesis PG200 software ver. 2006.2160.3 (Nonlinear Dynamics, Newcastle, UK) to compute spot detection and density calculations. Detected spots were matched across all gels while manually flagging poor‐quality spots. The background level was computed according to the modal density in the 50‐pixel margin surrounding each spot and subtracted accordingly. Spot volume was normalized by dividing each spot volume by the sum of all matched spots (excluding large spots derived from blood, such as albumin), followed by further division by the β‐actin spot on the same gel, thereby collectively correcting for the variation in total loading amount and blood proteins. As a consistency criterion, all spots observed at frequency of <50% in all histological subtypes were excluded from further analyses. Absent spots were given an arbitrary expression value of either the smallest value among the matched spot set or one‐third of the mean, whichever gave the smaller value. In all, 323 spots that fulfilled the consistency criterion were identified by mass spectrometry (Table S2). Subtype‐specific expression was tested by the Mann–Whitney U‐test. For each spot, P‐values were calculated with respect to each subtype testing the null hypotheses median_subtype = median_all. The Mann–Whitney U‐test, Spearman's correlation (rho) test, and principal component analysis were performed using SPSS ver. 15.0J (IBM, Tokyo, Japan). Computed principal components were subsequently transformed by varimax rotation and the first two principal components were used for scatter plot display. Gene ontology (GO) annotation for the listed proteins was collected from QuickGO (http://www.ebi.ac.uk/QuickGO/, accessed 20 May 2011) using the GO slim algorithm.21

Protein identification

Protein spots were excised to gel pieces that were 1.5 mm in diameter and transferred to a 96‐well microtiter plate. In‐gel trypsin digestion was performed as described by Shevchenko et al.22 using 50 ng sequencing grade modified trypsin per gel piece. Extracted peptides were desalted by ZipTip μC18 and eluted directly onto a matrix‐assisted laser desorption ionization (MALDI) target plate. The MALDI matrix, α‐cyano‐4‐hydroxycinnamic acid, was prepared at a concentration of 3 mg/mL in aqueous 50% acetonitrile and 10 mM ammonium dihydrogen phosphate, and 0.5 μL was applied to each well. After drying, peptide mass fingerprints were acquired using the AXIMA‐CFRplus MALDI time‐of‐flight mass spectrometry (MALDI‐TOF MS) instrument (Shimadzu/Kratos, Kyoto, Japan) for m/z 700–3500. Proteins were identified by MASCOT Server ver. 2.3.02 (Matrix Sciences, London, UK), searching against 20 239 human sequences of UniProt (http://www.uniprot.org/, accessed 1 Jun 2011, database release 2011_05). The search parameters were as follows: enzyme, trypsin; allow up to one missed cleavage; carbamidomethyl (cysteine) fixed modification; oxidation (methionine) and acetylation (protein N‐terminus) variable modifications; peptide tolerance 0.15 Da. Protein spots were regarded as identified only when confirmed by at least two independent identification results acquired from the same position of different 2‐DE gels of ovarian cancer samples. For acid ceramidase only, protein identification was assisted by an MS/MS ion search by AXIMA‐QIT (Shimadzu/Kratos) using a fresh preparation.

Western blotting

Target proteins annexin A4 (ANXA4), cellular retinoic acid‐binding protein 2 (CRABP2), serpin B5 (SPB5), and phosphoserine aminotransferase (PSAT1) were quantified by western blotting using the same tissue lysates as used for biomarker screening plus an independent sample set. Total protein (5 μg) was separated by tricine‐buffered SDS‐PAGE and blotted on a PVDF membrane. Polyclonal antibodies were produced by immunization of rabbits with human full‐length recombinant ANXA4, CRABP2, SPB5, and PSAT1, and used at a dilution of 1 : 2000 using 5% skim milk as a blocking agent. The HRP‐conjugated secondary antibody was used at a dilution of 1 : 5000 in 2% BSA. Reactivity was visualized on chemiluminescence films using an ECL detection kit (GE Healthcare). Directly after detection, the membranes were washed twice in 100 mM glycine–HCl pH 2.5 buffer, blocked in 5% skim milk and reprobed with HRP‐conjugated anti‐β‐actin as a loading control.

Immunohistochemistry

Formalin‐fixed, paraffin‐embedded tissues from EOC patients were sectioned at 4 μm. All slides were deparaffinized in xylene and rehydrated by stepwise immersion (for 2 min each) in 100%, 100%, 90%, 80%, and finally 70% ethanol. Residual peroxidase activity was quenched in 0.3% hydrogen peroxide in methanol for 30 min. Immunohistochemical staining was performed using the VECTASTAIN Elite ABC kit and ImmPACT DAB solution (Vector Laboratories, Burlingame, CA, USA) according to the manufacturer's instructions. The primary antibodies were anti‐ANXA4 antibody (clone D‐2; 1 : 200 dilution; Santa Cruz), anti‐SPB5 antibody (clone EAW24; 1 : 20 dilution; Thermo Fisher Scientific, Waltham, MA, USA), anti‐CRABP2 antibody (clone TA52; 1 : 2000 dilution; Chemicon, Temecula, CA, USA), and anti‐PSAT1 polyclonal antibody (product code 10501‐1‐AP; 1 : 100 dilution; Proteintech, Chicago, IL, USA). For antigen retrieval for SPB5, CRABP2 and PSAT1 staining, slides were boiled in 10 mM citrate buffer (pH 5.0) at 110°C for 1 min in an autoclave and anti‐SPB5 antibody was diluted in Can Get Signal Immunostain Solution B (Toyobo, Osaka, Japan). Stained sections were lightly counterstained with hematoxylin and then examined under a light microscope (DM6000B; Leica Microsystems, Wetzlar, Germany).

Results

Proteomic profiling by 2‐DE

Surgically dissected tissue blocks of 39 ovarian cancer patients (Table 1) were prepared and the total protein content in each was differentially analyzed by 2‐DE to compare clinically important histological subtypes: clear cell, endometrioid, mucinous, and serous adenocarcinoma. The screening process is summarized in Figure 1. Representative 2‐DE images of each subtype are shown in Figure 2(a), detecting a total of 800–900 protein spots. As described previously,23 each spot volume was normalized against the β‐actin spot. After spot matching and the application of consistency criteria, 323 spots were selected for further analysis and were identified by MALDI‐TOF MS, giving 217 unique protein identities. The expression levels of the identified spots in each subtype were compared against all other subtypes combined by the Mann–Whitney U‐test to elucidate subtype‐specific features. Table 2 lists those proteins that were significantly up‐ or downregulated (P < 0.0125, considering Bonferroni's correction). In total, 77 spots (55 proteins) were identified as potential subtype‐specific biomarker candidates. According to the cellular_component of the GO annotation, the screened proteins were mostly cytoplasmic and demonstrated that the influence of blood on our tissue proteome analysis was minimal. As indicated in parentheses following the protein names in Table 2, redundancy in protein identification was frequently observed owing to multiple conformations of the same protein separating into individual spots. Differential expression observed in only one of the multiple spots was possibly indicative of a subtype‐specific change in protein modification, such as phosphorylation, acetylation, glycosylation, or site‐specific protease cleavage. For example, Spots 1 and 4 of purine nucleoside phosphorylase were oppositely regulated in clear cell and serous carcinoma. For ANXA4 and PSAT1, all spots observed were concomitantly upregulated in clear cell carcinoma, reflecting the total change in protein expression.

Figure 1.

Overall workflow of the screening performed in the present study. 2‐DE, two‐dimensional gel electrophoresis; PCA, principal component analysis.

Figure 2.

Seventy‐seven differentially expressed spots were identified by two‐dimensional (2D) gel electrophoresis analysis of epithelial ovarian carcinoma subtypes. (a) Representative 2D gel images of clear cell, endometrioid, serous, and mucinous tumor lysates, separated over a pH 3–10NL gel in the first dimension and a 10–18% gradient gel in the second dimension. Arrows indicate the position of annexin‐A4 spots. (b) Scatter plot of the first two components computed using the expression values of 77 subtype‐specific spots with P < 0.0125, showing separation of clear cell carcinoma from the other subtypes. Individual tumors are annotated with histological subtype as indicated.

Table 2.

Differentially expressed proteins that characterize histological subtypes of epithelial ovarian carcinoma, screened by two‐dimensional gel electrophoresis and identified by matrix‐assisted laser desorption ionization time‐of‐flight mass spectrometry (proteins identified from multiple individual spots are indicated in parentheses and are numbered in order of abundance)

UniProt ID	Gene symbol	Protein name	P‐valuea	Fold change	Cellular componentb
Clear cell carcinoma
Upregulated
P00491	PNP	Purine nucleoside phosphorylase (#1 of 4 spots)	1.3E‐06	2.38	Cytoplasm
P60174	TPI1	Triosephosphate isomerase (#3 of 4)	7.6E‐06	2.36	Cytoplasm, nucleus
P07942	LAMB1	Laminin subunit β‐1	7.7E‐06	4.71	Plasma membrane, extracellular space
P09525	ANXA4	Annexin A4 (#1 of 2)	1.4E‐05	2.52	Cytoplasm
Q9Y617	PSAT1	Phosphoserine aminotransferase (#1 of 2)	2.4E‐05	2.49	n/a
P09525	ANXA4	Annexin A4 (#2 of 2)	4.0E‐05	3.97	Cytoplasm
P07195	LDHB	l‐Lactate dehydrogenase B chain	4.0E‐05	1.62	Cytoplasm, mitochondrion
P23381	WARS	Tryptophanyl‐tRNA synthetase, cytoplasmic	5.6E‐05	4.06	Cytoplasm
O00764	PDXK	Pyridoxal kinase	1.2E‐04	1.91	Cytoplasm
P11047	LAMC1	Laminin subunit γ‐1	1.4E‐04	3.31	Extracellular space
Q9Y617	PSAT1	Phosphoserine aminotransferase (#2 of 2)	1.4E‐04	4.06	n/a
Q16762	TST	Thiosulfate sulfurtransferase (#2 of 2)	1.4E‐04	2.54	Mitochondrion, plasma membrane
P30040	ERP29	Endoplasmic reticulum resident protein 29	1.9E‐04	1.86	Cytoplasm
Q8WW59	SPRYD4	SPRY domain‐containing protein 4	1.9E‐04	1.74	Mitochondrion, nucleus
P04179	SOD2	Superoxide dismutase (Mn), mitochondrial	2.2E‐04	2.39	Cytoplasm, mitochondrion
P22352	GPX3	Glutathione peroxidase 3	3.4E‐04	2.25	Extracellular space
P32119	PRDX2	Peroxiredoxin‐2 (#2 of 3)	8.6E‐04	2.35	Cytoplasm, mitochondrion
Q99536	VAT1	Synaptic vesicle membrane protein VAT‐1 homolog	9.8E‐04	1.76	Cytoplasm
Q13510	ASAH1	Acid ceramidase	0.0014	2.48	Cytoplasm
P02794	FRIH	Ferritin heavy chain	0.0016	1.57	Cytoplasm, mitochondrion
P40261	NNMT	Nicotinamide N‐methyltransferase	0.0018	2.44	Cytoplasm
Q5EBM0	CMPK2	UMP‐CMP kinase 2, mitochondrial	0.0031	2.14	Mitochondrion
O15305	PMM2	Phosphomannomutase 2	0.0035	1.42	Cytoplasm
P00491	PNP	Purine nucleoside phosphorylase (#4 of 4)	0.0035	2.42	Cytoplasm
P30041	PRDX6	Peroxiredoxin‐6	0.0039	1.78	Cytoplasm, mitochondrion
Q9HCC0	MCCC2	Methylcrotonoyl‐CoA carboxylase β chain, mitochondrial	0.0053	1.76	Mitochondrion
P02743	SAMP	Serum amyloid P‐component	0.0071	1.98	Extracellular space
P21964	COMT	Catechol O‐methyltransferase (#1 of 2)	0.0079	1.45	Cytoplasm, mitochondrion, plasma membrane
P30084	ECHS1	Enoyl‐CoA hydratase, mitochondrial	0.0079	1.44	Mitochondrion
P00918	CA2	Carbonic anhydrase 2	0.0087	1.77	Cytoplasm, extracellular space
P04083	ANXA1	Annexin A1 (#2 of 3)	0.010	1.52	Cytoplasm, nucleus, plasma membrane, extracellular space
P24666	ACP1	Low molecular weight phosphotyrosine protein phosphatase (#2 of 3)	0.010	2.25	Cytoplasm, nucleus, plasma membrane
Q01105	SET	Protein SET	0.011	2.06	Cytoplasm, nucleus
P37837	TALDO	Transaldolase	0.011	1.80	Cytoplasm
Downregulated
Q14019	COTL	Coactosin‐like protein	4.0E‐05	0.55	Cytoplasm
P06702	S100A9	Protein S100‐A9 (#1 of 4)	9.1E‐05	0.60	Cytoplasm, nucleus, plasma membrane, extracellular space
Q9P1F3	C6ORF115	Costars family protein C6orf115	3.0E‐04	0.37	n/a
Q01469	FABP5	Fatty acid‐binding protein, epidermal (#2 of 2)	5.9E‐04	0.35	Cytoplasm
P29373	CRABP2	Cellular retinoic acid‐binding protein 2	0.0028	0.24	Cytoplasm, nucleus
Q00169	PITPNA	Phosphatidylinositol transfer protein α isoform	0.0031	0.64	Cytoplasm
P07108	DBI	Acyl‐CoA‐binding protein	0.0048	0.60	n/a
P12429	ANXA3	Annexin A3	0.0053	0.59	Cytoplasm, plasma membrane
Q13228	SELENBP1	Selenium‐binding protein 1	0.0059	0.62	Cytoplasm, nucleus
O95865	DDAH2	N G,N G‐Dimethylarginine dimethylaminohydrolase 2	0.0071	0.63	Cytoplasm
P26447	S100A4	Protein S100‐A4	0.0071	0.36	Cytoplasm, nucleus
P52907	CAPZA1	F‐Actin capping protein subunit α‐1	0.0079	0.72	Cytoplasm
P36952	SPB5	Serpin B5	0.0105	0.25	Cytoplasm, extracellular space
Endometrioid carcinoma
Upregulated
Q7L266	ASRGL1	l‐Asparaginase	0.0046	3.59	Cytoplasm, nucleus
Q9P1F3	C6ORF115	Costars family protein C6orf115	0.0089	1.96	n/a
Downregulated
P09525	ANXA4	Annexin A4 (#2 of 2)	0.0007	0.21	Cytoplasm
O00299	CLIC1	Chloride intracellular channel protein 1 (#1 of 3)	0.0022	0.65	Cytoplasm, nucleus, plasma membrane
Q92597	NDRG1	Protein NDRG1	0.0032	0.52	Cytoplasm, nucleus, plasma membrane
P40925	MDH1	Malate dehydrogenase, cytoplasmic	0.0052	0.58	Cytoplasm, mitochondrion
P50583	NUDT1	Bis(5′‐nucleosyl)‐tetraphosphatase (asymmetrical)	0.0110	0.41	Mitochondrion
Mucinous carcinoma
Upregulated
P36952	SPB5	Serpin B5	1.0E‐04	4.76	Cytoplasm, extracellular space
P18085	ARF4	ADP‐ribosylation factor 4 (#2 of 2)	0.0016	2.93	Cytoplasm, plasma membrane
P37802	TAGLN2	Transgelin‐2 (#1 of 2)	0.0045	3.62	Nucleus, plasma membrane
P12429	ANXA3	Annexin A3	0.0052	1.60	Cytoplasm, plasma membrane
Downregulated
P32119	PRDX2	Peroxiredoxin‐2 (#2 of 3)	0.0045	0.39	Cytoplasm, mitochondrion
Q13938	CAPS	Calcyphosin	0.0052	0.27	Cytoplasm
P02794	FRIH	Ferritin heavy chain	0.0061	0.53	Cytoplasm, mitochondrion
O00764	PDXK	Pyridoxal kinase	0.011	0.52	Cytoplasm
P24666	ACP1	Low molecular weight phosphotyrosine protein phosphatase (#1 of 2)	0.011	0.43	Cytoplasm, nucleus, plasma membrane
Serous carcinoma
Upregulated
P29373	CRABP2	Cellular retinoic acid‐binding protein 2	9.4E‐05	3.82	Cytoplasm, nucleus
O43598	C6ORF108	Deoxyribonucleoside 5′‐monophosphate N‐glycosidase	0.0010	2.21	Cytoplasm, nucleus
P24666	ACP1	Low molecular weight phosphotyrosine protein phosphatase (#1 of 2)	0.0010	1.68	Cytoplasm, nucleus, plasma membrane
P61769	B2M	β‐2‐Microglobulin	0.0029	1.68	Cytoplasm, plasma membrane, extracellular space
P00491	PNP	Purine nucleoside phosphorylase (#4 of 4)	0.0032	1.58	Cytoplasm
P42771	CDKN2A	Cyclin‐dependent kinase inhibitor 2A, isoforms 1/2/3 (#3 of 3)	0.0037	3.35	Cytoplasm, nucleus
O95865	DDAH2	N G,N G‐Dimethylarginine dimethylaminohydrolase 2	0.0046	1.59	Cytoplasm
P42771	CDKN2A	Cyclin‐dependent kinase inhibitor 2A, isoforms 1/2/3 (#1 of 3)	0.0058	1.58	Cytoplasm, nucleus
P52907	CAPZA1	F‐Actin capping protein subunit α‐1	0.0080	2.36	Cytoplasm
P26447	S100A4	Protein S100‐A4	0.0080	1.39	Cytoplasm, nucleus
P60174	TPI1	Triosephosphate isomerase (#4 of 4)	0.0110	1.33	Cytoplasm, nucleus
Downregulated
P00491	PNP	Purine nucleoside phosphorylase (#1 of 4)	0.0022	0.51	Cytoplasm
Q13510	ASAH1	Acid ceramidase	0.0065	0.42	Cytoplasm
Q9HCC0	MCCC2	Methylcrotonoyl‐CoA carboxylase β chain, mitochondrial	0.0080	0.53	Mitochondrion

n/a, no annotated terms; UMP‐CMP, uridine monophosphate‐cytidine monophosphate.

^a P‐values were calculated by the Mann–Whitney U‐test.

^bThe gene ontology (GO) terms searched for were: cytoplasm, mitochondrion, nucleus, plasma membrane and extracellular space.

Principal component analysis (PCA) was performed using the expression values of the 77 selected spots to visualize the expression pattern of 39 samples (Fig. 2b). The pattern clearly showed that clear cell carcinoma (rectangles in Fig. 2b) was most distinct from other subtypes, and that all samples of clear cell carcinoma were very homogeneous, with 11 samples forming a very confined cluster. The PCA plot also conveyed that clear cell and serous carcinomas were most distant, suggesting that the list of candidate biomarkers would perform best in distinguishing these two subtypes. Endometrioid samples exhibited remarkable overlap with other subtypes, conveying its expressional heterogeneity, concordant with the discovery of the smallest number of specifically up‐ or downregulated proteins for this subtype (two upregulated and five downregulated).

Validation by western blotting

The four proteins that exhibited the largest quantitative specificity to represent each subtype, namely ANXA4, PSAT1, CRABP2, and SPB5, were validated by western blotting (Fig. 3a). Because this validation method was not structure specific, candidates of subtype‐specific protein modification (i.e. only one of multiple conformations was altered) were not selected for analysis. Figure 3(b) shows the box‐plot of the expression levels of each protein for each of the four subtypes of EOC, quantitated by normalizing the bands against β‐actin detected from the same membrane. The results show significant upregulation of ANXA4 and PSAT1 in clear cell carcinoma (P = 8.3E‐07 and 4.9E‐08, respectively, Mann–Whitney U‐test), upregulation of CRABP2 in serous carcinoma (P = 3.5E‐05), and upregulation of SPB5 in mucinous carcinoma (P = 0.013). Moreover, a possible correlation between the expression of these proteins and cancer stage (FIGO I, II, or III/IV) or grade (G1, G2 or G3; excluding clear cell carcinoma) was tested by Spearman's ρ calculation (Fig. 3c). There was significant correlation between CRABP2 expression and both cancer stage (ρ = 0.339, P = 0.004) and tumor grade (ρ = 0.669, P = 1.1E‐7). However, this strongly reflected the characteristics of the serous subtype with which CRABP2 expression was associated. When the serous subtype was considered independently, there was no significant association between CRABP2 levels and tumor grade or stage. Instead, we observed a weak correlation between SPB5 expression and tumor grade in serous carcinoma (ρ = 0.499, P = 0.025).

Figure 3.

Expression patterns for annexin A4 (ANXA4), cellular retinoic acid‐binding protein 2 (CRABP2), serpin B5 (SPB5), and phosphoserine aminotransferase (PSAT1) were validated by western blotting using 74 tumor specimens. (a) Protein (5 μg) prepared from the same tissue lysate as that analyzed by two‐dimensional (2D) gel electrophoresis was resolved by 10–16% tricine–SDS‐PAGE, blotted onto a PVDF membrane, and detected by antibodies against each protein. Combined results of four independent blots are shown. One lane corresponds to one patient of the tumor type indicated at the top of the blots. (b) Box plots summarizing the expression of ANXA4, SPB5, PSAT1, and CRABP2 in each histological subtype. Expression was quantitated by normalizing the band volumes against the corresponding loading control band, detected by reprobing the same membrane. (c) Dot plots summarizing the expression of CRABP2 according to the International Federation of Gynecology and Obstetrics (FIGO) stage and tumor grade. The slope of the line of best fit is proportional to Spearman's correlation coefficient (ρ).

Immunohistochemistry

We performed immunohistochemical staining for ANXA4. PSAT1, CRABP2, and SPB5 in formalin‐fixed, paraffin‐embedded tissue specimens representing clear cell, endometrioid, mucinous, and serous carcinoma to demonstrate differential staining and localization of candidate proteins (Fig. 4). Overall, staining was confined to the cancer cells of each subtype and the degree of staining was concordant with the results of western blotting. Clear cell carcinoma showed strongly positive staining for ANXA4 and PSAT1, and both cytosolic and nuclear staining were observed. Mucinous carcinoma was positive for ANXA4 and SPB5 (mainly in the cytosol), weakly positive for CRABP2, and negative for PSAT1. Serous carcinoma was very strongly positive for CRABP2, weakly positive for PSAT1, and negative for ANXA4 and SPB5. Endometrioid carcinoma was weakly positive for ANXA4, CRABP2, and PSAT1 and negative for SPB5. Using the panel of biomarkers, it was demonstrated that different tumor subtypes show characteristic immunohistochemical staining patterns. No stromal staining was observed throughout the analysis, except for serous stromal cells that showed a marginal response to CRABP2 staining.

Figure 4.

Immunohistochemical analysis of primary epithelial ovarian carcinomas using formalin‐fixed, paraffin‐embedded tissue sections. (a) Representative H&E‐stained images of clear cell, endometrioid, mucinous, and serous carcinomas. (b) Annexin A4 (ANXA4) staining was strongly positive in clear cell and mucinous carcinomas. (c) Phosphoserine aminotransferase (PSAT1) staining was strongly positive in clear cell carcinoma, weakly positive in endometrioid and serous carcinomas, and negative in mucinous carcinoma. (d) Cellular retinoic acid‐binding protein 2 (CRABP2) staining was strongly positive in serous carcinoma, positive in endometrioid carcinoma, and marginally positive in mucinous carcinoma. (e) Serpin B5 (SPB5) staining was positive in mucinous carcinoma only. All sections were incubated for the same duration following 3′,3‐diaminobenzidine development. Scale lines, 100 μm.

Discussion

The present study is the first comprehensive proteomic characterization of ovarian cancer using clinical tumor specimens. The results identify molecular signatures that distinguish the histopathology of ovarian cancer. One advantage of 2‐DE proteomic analysis over gene expression profiling using a cDNA microarray is that quantitative differences of target proteins are readily reproducible by antibody‐based verification, and this was clearly illustrated in the present study by western blotting (Fig. 3) and differential staining by immunohistochemistry (Fig. 4). Thus, proteomic profiling defined key “effector” proteins that could contribute directly to the histopathological and clinical differences in the various subtypes of ovarian cancers. Moreover, considering the growing attention towards post‐transcriptional protein regulation in systems biology,24 our proteomic data should complement previous studies by gene expression profiling and uniquely identify previously overlooked features.

In the present study, we identified ANXA4 and PSAT1 as the most quantitative signature that best delineated the clear cell subtype from other subtypes. To our knowledge, upregulation of PSAT1 associated with ovarian cancer has not been reported previously. It is known that PSAT1 is part of the serine biosynthetic process25 and its aberrant expression has been reported in colon cancer;26 yet its role as an oncogene remains to be fully elucidated. A member of the calcium‐dependent phospholipid‐binding protein family, ANXA4 has been widely implicated to be associated with various cancers, including renal clear cell carcinoma,27 pancreatic adenocarcinoma,28 and ovarian clear cell carcinoma.18 Although previous reports argue that ANXA4 expression directly confers cisplatin resistance,29 which is a clinical characteristic of clear cell carcinoma, our western blotting data (Fig. 3) revealed very consistent basal expression of ANXA4 among all ovarian carcinoma samples tested. Moreover, on 2‐DE separation, ANXA4 appeared in two distinct spots (Fig. 2a), arising from post‐translational modification, and our immunohistological staining of ANXA4 suggested different localization according to EOC subtype. Such structural or functional changes to target molecules are promising biomarker candidates.

With regard to other proteins that were upregulated in clear cell carcinoma, 75% were metabolic enzymes involved in fundamental biological processes such as aerobic respiration (triosephosphate isomerase, L‐lactate dehydrogenase B chain, carbonic anhydrase 2, transaldolase), cofactor processing (nicotinamide N‐methyltransferase, catechol O‐methyltransferase, methylcrotonoyl‐CoA carboxylase beta chain, pyridoxal kinase), and nucleoside metabolism (purine nucleoside phosphorylase, uridine monophosphate‐cytidine monophosphate kinase 2). The wide spectrum of upregulated pathways suggests that the clear cell subtype of EOC is highly metabolically active, despite its slow tumor progression. Moreover, high levels of antioxidative enzymes, such as glutathione peroxidase 3, peroxiredoxin‐2, peroxiredoxin‐6, and superoxide dismutase, may be responsible for resistance to apoptosis induced by oxidative stress or chemotherapy.30, 31 These features are worth taking into consideration in the development of effective therapies against the clear cell subtype of EOC.

It is known that CRABP2 is strongly and specifically expressed in serous carcinoma; it contains a retinoic acid‐binding domain and sensitizes the cellular response to the antiproliferative signaling by retinoic acid.32 Whether serous carcinoma responds to retinoic acid signaling requires further investigation; if so, there is the chance that retinoic acid signaling would be a serous subtype‐specific therapeutic target. In the present study, CRABP2 was almost entirely absent in clear cell carcinoma and exhibited the most significant difference between the serous and clear cell subtypes in our analysis. The absence of CRABP2 could explain a unique differentiation of clear cell carcinoma, and it could also be associated with endometriosis because CRABP2 is known to be strongly suppressed in endometriosis tissues.33 Interestingly, the level of fatty acid binding protein 5, another retinoic acid signaling transducer that competes with CRABP2,32 was slightly downregulated in clear cell carcinoma, but was generally uniform in other subtypes, suggesting that CRABP2 was deregulated independently of other retinoic acid signaling pathways. Similar deregulation has been reported in astrocytic gliomas, in which CRABP2 expression is associated with a poor prognosis.34 Another intriguing protein upregulated in serous carcinoma is N ^G,N ^G‐dimethylarginine dimethylaminohydrolase 2, which plays an important role in the nitric oxide signaling pathway via regulation of asymmetric dimethylarginine, an endogenous inhibitor of nitric oxide synthase.35 Other proteins upregulated in the serous subtype of EOC include low molecular weight phosphotyrosine protein phosphatase, known to regulate EphA2‐mediated signaling,36 protein S100‐A4, which is involved in nuclear factor‐κB signaling, and cyclin‐dependent kinase inhibitor 2A, which is involved in cell cycle regulation. Overall, the molecular signature of serous carcinoma is characterized by upregulation of proteins involved in multiple, apparently unrelated signaling pathways. Such prevalence of deregulated signaling pathways implicates dysfunction of the common central regulator, for example, TP53, which is mutated in most cases of serous carcinoma.37

Screening for mucinous‐specific features was confounded by unmatched sample size. Nevertheless, SPB5 expressed aberrantly in mucinous carcinoma is known as a tumor suppressor gene in breast cancer and as an oncogene in various malignant cancers.38 Our findings here agree with the recent study that reported aberrant SPB5 expression in mucinous ovarian carcinoma, where borderline tumors showed higher tendency of nuclear localization.39 Moreover, our study revealed that endometrioid carcinoma exhibited a very heterogeneous molecular profile, with only seven proteins identified as endometrioid‐specific features, and presented a wide range of principal components. We therefore speculate that there is the possibility that morphological classification of endometrioid carcinoma may be subdivided on the basis of specific molecular signatures to better characterize the clinical features.

In conclusion, the present proteomic characterization study provides evidence supporting the proposal that EOC subtypes are clinically distinct entities. Of particular interest are the correlations with the malignant and chemoresistant features of clear cell carcinoma. Better biomarkers associated with the different histopathological classifications of EOC, coupled with prognostic or chemosensitivity data, could suggest new avenues for clinical therapy. A molecular profile that correlates to histopathological classification would lead to a more detailed biological basis for this disease and eventually allow for the individualization of patient care.

Disclosure Statement

The authors have no conflict of interest to disclose.

Supporting information

Table S1. List of all ovarian carcinoma specimens used in the present study.

Table S2. Full list of the spot set used for biomarker screening shown with protein identity.