This article is mentioned in:

Introduction

Ovarian cancer is the fifth leading cause of cancer deaths in women and has the highest overall mortality rate and poor 5-year survival. More than 90% of ovarian cancer cases are epithelial ovarian cancer (EOC), which represents a series of etiologically and molecularly distinct disease. Although early diagnosis and therapy are considered to be the most effective methods to improve the outcome of patients with any cancer, the majority of EOC patients often do not manifest clinical symptoms and receive medical intervention when their tumor cells have disseminated to the peritoneal cavity. Currently, the standard therapy for EOC is surgical resection followed by postoperative chemotherapy with carboplatin and paclitaxel (PTX). Despite initial responsiveness to cisplatin-based chemotherapy, surgical and chemotherapy is far from satisfactory and most patients eventually develop drug-resistant tumors and succumb to the recurrent disease. This is why the majority of EOC patients with advanced disease relapse within 5 years, and little progress has been made in improving overall survival rates. Previous studies have proposed numerous factors to influence drug-resistance, such as ATP-dependent efflux pumps, extracellular microenvironment, DNA repair mechanism, modification of the drug target, drug-induced cytotoxicity, disruptions in apoptotic signaling pathways and changes in the expression of protein associated with tumor resistance (1,2). Cisplatin has been used to treat various cancers primarily by causing DNA damage and has been accepted worldwide as a first-line anticancer drug for EOC chemotherapy. In this regard, to identify those patients who have potential recurrence and to overcome chemoresistance and therefore improving patient outcome are the serious challenges in the management of EOC patients. Nevertheless, none of the identified biomarkers for drug resistance have been proven acceptable for routine clinical use. Hence, identification of clinical reliable biomarkers has come to the forefront of investigation. Moreover, if a non-invasive but sensitive blood assay that can monitor responses to chemotherapy was available, it would be invaluable for guiding chemotherapy and greatly improving the overall survival rate of EOC patients.

Mass spectrometry (MS) is an important high-throughput, industrially stable, information-rich technique for profiling small molecular compounds and is widely used to assess potential diagnostic and prognostic biomarkers. We applied isobaric tags for relative and absolute quantitation (iTRAQ)-based quantitative proteomic approach (Fig. 1A) and HPLC-micrOTOF-Q II high-resolution mass spectrometer-based metabolic analysis to compare and identify proteins and metabolites with differential profile in normal control (NC) group, benign ovarian cyst (BOC), platinum-sensitive (PTS) and platinum-resistant (PTR) cohort of serum samples. The emergence of resistance to platinum-based therapy is the main clinical endpoint of this experiment. iTRAQ proteomic analysis that combines 2D-LC and MALDI-TOF-MS/MS is an established technique in which total proteins are enzymatically digested into a large array of small peptide fragments and then directly analyzed by liquid chromatography-mass spectrometry (LC-MS). A total of 64 proteins with different expression levels were identified. In the list of differentially identified proteins via this method, most of these proteins were in accordance with the previously published literatures and associated with cancers. In addition, to further explore the new biomarkers predicting the responses to cisplatin, four of these proteins (FN1, SERPINA1, ORM1 and GPX3) were confirmed in a large patient cohort using western blotting and commercial enzyme-linked immunosorbent assay (ELISA), respectively. Moreover, HPLC-micrOTOF-Q II MS coupled with multivariate analysis was utilized, good separations were obtained for PTR, PTS vs. health controls. Finally, six substances with low molecular weight were identified based on the database using nuclear magnetic resonance (NMR). Receiver operating characteristic (ROC) curve analysis was then used to elucidate the potential power of FN1, SERPINA1, ORM1 and six small molecular metabolites for discriminating between the PTS and PTR group. The findings of this study are expected to reveal new proteins and metabolites related to platinum resistance and to provide candidate biomarkers to predict clinical response to chemotherapy. However, we do need an effective serum marker to predict the patients who have no response to cisplatin chemotherapy and will progress or recur during or after chemotherapy that cannot be easily judged from ultrasound or CT scan. This prediction is fundamental since patients that are resistant might benefit from a different combinational chemotherapy.

Figure 1 (A) Overall workflow for screening and validation experiments. (B) Comparison of human serum after depletion of the targeted high abundant proteins. Samples were separated by 10% SDS-PAGE and visualized by Coomassie Brilliant Blue staining: lane 1, human crude serum; lane 2, low-abundant proteins. (C) Equal amounts of the depleted sera from NC, PTS, and PTR individuals were loaded in each lane, according to the concentrations determined by the BCA kit. NC, normal control; PTS, platinum-sensitive; PTR, platinum-resistant.

Materials and methods

Human serum samples

After Institutional Review Board approval (Ethics Committees of the Affiliated Tumor Hospital of Guangxi Medical University, Nanning, China), we obtained specimens (Table I) between September 1998 and March 2013, including EOC and BOC specimens. EOC cases were assigned to the PTR and PTS group. The FIGO classification was used for clinical staging, and the Gynecologic Oncology Group criteria were used for histological grading. NC blood samples were voluntarily donated by healthy individuals. Patients were eligible to participate in this trial if they had a pathologically confirmed diagnosis of BOC or EOC. After giving informed consent, serum samples were collected by clean venipuncture, centrifuged for 10 min at 3,000 × g at 4°C and stored at −80°C until further analysis.

Table I Clinical characteristics of 40 cases of samples used in screening, 129 cases in validating and 132 samples in comparing metabolomic profiles between NC, BOC, PTS and PTR groups.

Table I

Clinical characteristics of 40 cases of samples used in screening, 129 cases in validating and 132 samples in comparing metabolomic profiles between NC, BOC, PTS and PTR groups.

Clinical characteristics	NC	BOC	PTS	PTR	P-value
iTRAQ (n=40)
No. of patients	10	10	10	10	-
iTRAQ-labeled sample	113 (run 1)	114 (run 1)	115 (run 1)	116 (run 1)	-
	117 (run 2)	118 (run 2)	119 (run 2)	121 (run 2)	-
Age (years) (mean ± SD)	39.80±5.53	44.1±18.11	47.21±12.64	44.85±16.17	-
Histological type
Serous	-	6/10	4/10	6/10	-
Mucinous	-	2/10	2/10	1/10	-
Other	-	2/10	4/10	3/10	-
FIGO stage
I–II	-	-	3/10	3/10	-
III–IV	-	-	7/10	7/10	-
Tumor grade
Well-differentiated	-	-	1/10	2/10	-
Moderately-differentiated	-	-	1/10	1/10	-
Poorly-differentiated	-	-	8/10	7/10	-
ELISA (n=129)
No. of patients	33	-	52	44	-
Age (years) (mean ± SD)	39.71±10.05	-	46.33±10.65	47.08±10.85	-
Histological type
Serous	-	-	26/52	17/44	-
Mucinous	-	-	7/52	6/44	-
Other	-	-	19/52	21/44	-
TNM stage
I–II	-	-	19/52	2/44	-
III–IV	-	-	33/52	42/44	-
Tumor grade
Well-differentiated	-	-	14/52	7/44	-
Moderately-differentiated	-	-	9/52	10/44	-
Poorly-differentiated	-	-	29/52	27/44	-
FN1 (mean ± SD)	69.14±13.29	-	62.41±12.78	71.08±13.19	0.004
ORM1 (mean ± SD)	157.43±18.26	-	173.64±22.69	221.12±34.60	0.000
SERPINA1 (mean ± SD)	756.19±244.39	-	685.69±204.59	816.26±245.53	0.021
Metabolomics (n=132)
No. of patients	41	9	45	37	-
Age (years) (mean ± SD)	39.61±9.25	42.56±15.44	46.56±10.03	47.41±12.46	-
Histological type
Serous	-	-	23/45	11/37	-
Mucinous	-	-	6/45	4/37	-
Other	-	-	14/45	19/37	-
NA	-	-	2/45	3/37	-
FIGO stage
I–II	-	-	13/45	3/37	-
III–IV	-	-	31/45	32/37	-
NA	-	-	1/45	2/37	-
Tumor grade
Well-differentiated	-	-	6/45	3/37	-
Moderately-differentiated	-	-	5/45	4/37	-
Poorly-differentiated	-	-	27/45	20/37	-
NA	-	-	7/45	10/37	-
Primary therapy outcome
Success	-	-	22/45	13/37	-
CR+PR	-	-	21/45	23/37	-
SD+PD	-	-	2/45	1/37	-
NA

[i] NC, normal control; BOC, benign ovarian cyst; PTS, platinum-sensitive; PTR, platinum-resistant; iTRAQ, isobaric tags for relative and absolute quantitation; ELISA, enzyme-linked immunosorbent assay; NA, not available.

Chemicals and reagents

The iTRAQ™ Reagent kit and mass calibration standards were purchased from Applied Biosystems (Bedford, MA, USA). Sequencing grade trypsin was obtained from Promega (Madison, WI, USA). Amicon Ultra-15 Certifugal Filter Units (3 kDa) were purchased from EMD Millipore (Billerica, MA, USA). All the solvents and chemicals used in this experiment were of LC-MS or analytical grade. HPLC grade water and acetonitrile (ACN) were purchased from Merck KGaA (Darmstadt, Germany). BCA assay kit was purchased from Pierce Biotechnology, Inc. (Rockford, IL, USA). Methyl methanethiosulfonate (MMTS) and methanol were obtained from Thermo Fisher Scientific (Rockford, IL, USA). Triethylammonium bicarbonate (TEAB), trifluoroacetic acid (TFA), formic acid and α-cyano-4-hydroxycinnamic acid (CHCA) were all obtained from Sigma (St. Louis, MO, USA).

Proteomic analysis

Depletion of high abundant proteins

Pooled serum samples were depleted of the 14 most highly abundant proteins using antibody-based depletion with Human 14 Multiple Affinity Removal System (MARS Hu-14; Agilent Technologies, Inc., Palo Alto, CA, USA), according to the manufacturer’s instructions. Crude serum samples were thawed on ice. Equal amounts of blood (20 μl) from 10 individuals in each group were pooled. Thereafter, MARS Hu-14 column was used to deplete specific 14 high-abundant proteins ~94% of total protein mass from human serum (Fig. 1B). The total protein concentrations of the depleted sera were determined using the BCA Protein Assay Reagent Kit (Pierce Biotechnology, Inc.) (Fig. 1C).

iTRAQ labeling

Prior to iTRAQ analysis, aliquots of 100 μg protein from each of the four sample pools were reduced using dissolution buffer (0.5 M TEAB) to a volume of 20 μl. To each of the four pools, 1 μl denaturant (2% SDS) and 2 μl reducing reagent [50 mM tris(2-carboxyethyl)phosphine] were added. Each pool was incubated at 60°C for 1 h. Cysteines were stopped by adding 1 μl cystine-blocking reagent (200 mM MMTS in isopropanol), and samples were incubated for additional 10 min at room temperature. The samples were digested with Sequencing Grade Modified Trypsin (Promega) at a protein-to-trypsin ratio of 30:1, at 37°C overnight. After that, peptides from each of the four depleted serum pools were labeled with 8-plex iTRAQ reagents (AB SCIEX, Foster City, CA, USA) according to the manufacturer’s instructions. The labels were applied in the following order: NC pool (113 Da), BOC pool (114 Da), PTS pool (115 Da); PTR pool (116 Da), so as to run the same sample in duplicate in each run. The four labeled samples were then evaporated to a volume of roughly 30 μl using a SpeedVac Concentrator and combined as a mixture followed by cleaning up by a strong cation exchange (SCX) column.

2D-LC-ESI-MS/MS

The combined peptide sample was subjected to SCX chromatography employing a PolySulfoethyl A column (2.1×200 nm; PolyLC, Inc., Columbia, MD, USA), on a high-pressure LC-pump (1200 series; Agilent Technologies, Inc.). The mixed sample was diluted in 10 mM KH2PO4 (pH 3.0), 25% v/v ACN (mobile phase A). Peptides were eluted with a linear gradient of 0–500 mM KCl (mobile phase B: 25% v/v ACN, 10 mM KH2PO4, 500 mM KCl, pH 3.0) for 115 min at a flow rate of 0.2 ml/min. Fractions were collected at 2-min intervals, and 16 fractions were collected from 24.5 to 98.5 min. Each SCX fraction was desalted using C18 Spin Columns (The Nest Group, Inc., Southborough, MA, USA) following the manufacturer’s instructions and then vacuum centrifuged to dryness. The peptide fractions were separated on a nano-reverse-phase LC system (Tempo™ LC MALDI Spotting System; Applied Biosystems), using a Magic C18AQ column (150 mm × 200 μm, 3 μm, 200 Å; Michrom Bioresources, Inc., Auburn, CA, USA), at a flow rate of 2 μl/min. A binary gradient with buffer A (98% H2O, 2% ACN, and 0.1% TFA) and buffer B (2% H2O, 98% ACN, and 0.1% TFA) was employed as the mobile phase. The peptide solutions were first loaded for 20 min using buffer A only on the pre-column, and the separation occurred over a period of 110 min. The elution from the column was mixed in 1:1 ratio with 5 mg/ml CHCA with a flow rate of 2 μl/min, and spotted onto the MALDI plates in a 44×28 spot array format. MS and MS/MS analysis was performed on a TOF-TOF 5800 MALDI platform (Applied Biosystems). MS spectra were recorded in the positive-ion reflector mode covering 700/800–4000 mass-to-charge ratio (m/z) acquiring 1,500 laser shots per spectrum (30 subspectra of 50 shots). After screening of all LC-MALDI sample positions the fragmentation of automatically selected precursors was performed at a collision energy of 2 kV with collision-induced dissociation gas (air). Up to 20 of the most intense ion signals per spot position, characterized by an S/N >45, were selected as precursors for MS/MS acquisition.

Database searches and criteria

Peptide matching, protein identification, and relative protein quantification for the iTRAQ experiment were performed with ProteinPilot v4.0 software (Applied Biosystems) in which the paragon search algorithm was applied. MS/MS spectra were searched against the UniProt/Swiss-Prot database for species of Homo sapiens. The database was searched using the following parameters: trypsin was used as the digestion agent, MMTS as a fixed modification of cysteine, thorough as search effort, and biological modification as the ID focus. Identifications of proteins were only accepted with a ‘local false discovery rate (FDR)’ estimation of ≤5% and an unused ProtScore ≥1.3 (>95% CI). In addition, proteins were considered for further statistical analysis when meet the following standards: one or more unique peptides with 95% confidence had to be identified; proteins were considered up- or downregulated when their fold changes were >1.3 or <0.77. The results obtained from ProteinPilot were exported to Microsoft Excel for manual interpretation. The protein lists from the two iTRAQ experiments (run 1 and 2; Table II) were merged with ratios calculated to the reference pool.

Table II Differentially expressed proteins identified by iTRAQ between PTR group compared to PTS group.

Table II

Differentially expressed proteins identified by iTRAQ between PTR group compared to PTS group.

Unused protein score	Sequence coverage (%)	Accession no.	Name	Gene symbol	Species	iTRAQ ratio		Expression pattern
						Run 1	Run 2
13.01	47.80	P02763	α-1-acid glycoprotein 1	ORM1a	Human	3.32	2.34	Up
4.56	41.80	P19652	α-1-acid glycoprotein 2	ORM2a	Human	-	2.33	Up
8.52	46.50	D1MGQ2	α-2 globin chain	HBA2	Human	1.58	1.56	Up
23.74	72.70	P02647	Apolipoprotein A-I	APOA1a	Human	2.00	1.72	Up
3.52	68.00	P02652	Apolipoprotein A-II	APOA2	Human	1.33	-	Up
2.99	23.60	Q13790	Apolipoprotein F	APOF	Human	1.36	-	Up
2.00	30.00	P08519	Apolipoprotein(a)	LPA	Human	-	1.57	Up
19.91	35.80	P22792	Carboxypeptidase N subunit 2	CPN2	Human	-	1.45	Up
8.27	30.10	P06276	Cholinesterase	BCHEa	Human	-	1.31	Up
1.60	25.20	P31146	Coronin-1A	CORO1A	Human	1.81	-	Up
1.86	35.30	Q5VVP7	C-reactive protein, pentraxin-related	CRP	Human	3.82	2.46	Up
17.63	55.50	E9KL23	Epididymis secretory sperm binding protein Li 44a	SERPINA1a	Human	6.02	3.80	Up
4.23	44.00	P02675	Fibrinogen β chain	FGBa	Human	2.42	-	Up
111.13	54.60	P02751	Fibronectin	FN1	Human	2.52	-	Up
16.43	55.20	O75636	Ficolin-3	FCN3	Human	-	1.37	Up
63.09	79.10	P00738	Haptoglobin	HPa	Human	2.71	2.59	Up
18.65	84.40	D9YZU5	Hemoglobin, β	HBBa	Human	1.76	1.46	Up
1.46	14.30	P01880	Ig δ chain C region	IGHD	Human	-	2.54	Up
2.03	39.30	P01860	Ig γ-3 chain C region	IGHG3	Human	1.36	1.47	Up
3.89	51.90	P0CG05	Ig λ-2 chain C regions	IGLC2	Human	-	1.72	Up
13.88	41.20	P01871	Ig μ chain C region	IGHM	Human	-	1.37	Up
224.13	70.60	P02751–8	Isoform 8 of fibronectin	FN1a	Human	-	2.35	Up
1.35	43.70	Q9BWP8–9	Isoform 9 of collectin-11	COLEC11	Human	1.34	-	Up
25.71	70.90	P02750	Leucine-rich α-2-glycoprotein	LRG1	Human	-	1.55	Up
2.43	19.00	Q6Q3G8	Lysosomal-associated membrane protein 2, isoform CRA_b	LAMP2a	Human	1.81	-	Up
2.23	6.70	Q6UXB8	Peptidase inhibitor 16	PI16	Human	-	1.30	Up
2.56	47.40	P06702	Protein S100A9	S100A9a	Human	1.36	-	Up
94.91	82.10	P02787	Serotransferrin	TFa	Human	1.63	1.58	Up
75.06	73.60	P02768	Serum albumin	ALBa	Human	-	1.50	Up
3.27	70.00	E9PR14	Serum amyloid A protein	SAA2	Human	4.57	2.63	Up
2.04	17.80	E7ES66	Uncharacterized protein	GP1BA	Human	1.50	-	Up
5.51	37.80	B4E1D3	Uncharacterized protein	FGB	Human	-	2.46	Up
2.01	20.40	C9JC84	Uncharacterized protein	FGG	Human	-	2.60	Up
8.99	21.90	P04275	von Willebrand factor	VWFa	Human	-	1.33	Up
499.38	71.20	P04114	Apolipoprotein B-100	APOBa	Human	-	0.74	Down
3.82	61.50	B2R526	Apolipoprotein C-I	APOC1	Human	0.63	-	Down
6.00	65.40	P02655	Apolipoprotein C-II	APOC2	Human	0.62	0.68	Down
7.18	61.50	B0YIW2	Apolipoprotein C-III variant 1	APOC3a	Human	0.60	0.58	Down
23.95	77.30	P02649	Apolipoprotein E	APOEa	Human	-	0.60	Down
8.96	43.20	Q96KN2	β-Ala-His dipeptidase	CNDP1	Human	-	0.71	Down
2.01	30.60	P49913	Cathelicidin antimicrobial peptide	CAMP	Human	0.71	-	Down
4.00	21.70	P00488	Coagulation factor XIII A chain	F13A1a	Human	0.75	0.67	Down
2.60	21.90	Q12860	Contactin-1	CNTN1a	Human	0.72	-	Down
2.10	22.60	P22352	Glutathione peroxidase 3	GPX3a	Human	-	0.76	Down
5.52	43.20	Q1KLZ0	HCG15971, isoform CRA_a	PS1TP5BP1a	Human	0.62	0.39	Down
50.43	70.10	P05546	Heparin cofactor II	SERPIND1	Human	-	0.76	Down
3.78	31.50	P26927	Hepatocyte growth factor-like protein	MST1	Human	0.75	-	Down
110.79	52.90	P19823	Inter-α-trypsin inhibitor heavy chain H2	ITIH2a	Human	-	0.77	Down
1.66	14.60	Q9NPH3	Interleukin-1 receptor accessory protein	IL1RAP	Human	0.52	-	Down
2.00	36.20	O14791–2	Isoform 2 of apolipoprotein L1	APOL1	Human	-	0.76	Down
4.00	19.20	P16070–5	Isoform 5 of CD44 antigen	CD44a	Human	0.74	-	Down
204.25	80.80	B7ZKJ8	ITIH4 protein	ITIH4	Human	-	0.59	Down
44.35	54.80	P01042	Kininogen-1	KNG1a	Human	0.77	-	Down
8.29	43.20	Q5SQS3	Mannan-binding lectin	MBL2a	Human	0.69	0.59	Down
1.33	11.80	P04180	Phosphatidylcholine-sterol acyltransferase	LCAT	Human	0.64	-	Down
2.00	13.60	Q53Y44	Profilin	PFN1	Human	-	0.68	Down
99.83	76.50	P00734	Prothrombin	F2a	Human	-	0.65	Down
21.20	60.10	P02743	Serum amyloid P component	APCS	Human	-	0.58	Down
9.43	27.90	P27169	Serum paraoxonase/arylesterase 1	PON1a	Human	0.66	-	Down
2.00	17.70	O00391	Sulfhydryl oxidase 1	QSOX1a	Human	-	0.37	Down
2.00	23.10	D3DUS9	Triosephosphate isomerase	TPI1a	Human	-	0.70	Down
3.05	51.60	P67936	Tropomyosin α-4 chain	TPM4	Human	-	0.68	Down
6.31	26.60	A6NHF2	Uncharacterized protein	BTD	Human	0.41	-	Down
2.00	21.50	E7EX29	Uncharacterized protein	YWHAZa	Human	0.75	-	Down

a The peptides identified with 95% confidence.

{ label (or @symbol) needed for fn[@id='tfn3-ijo-49-04-1651'] } iTRAQ, isobaric tags for relative and absolute quantitation; PTR, platinum-resistant; PTS, platinum- sensitive.

Western blotting

To confirm the identity of the proteins discovered by iTRAQ, western blotting was performed (Fig. 3A–E). Briefly, equal volumes of non-depleted serum from NC, PTS and PTR individuals (n=9 per group) were electrophoretically separated by 10% SDS-PAGE and transferred to a 0.45-mm polyvinylidene fluoride membranes (EMD Millipore) using a Bio-Rad wet transfer apparatus. Anti-human ORM1 (2 μg/ml), FN1 (1:1,000 dilution), SERPINA1 (1:1,000) and GPX3 (1:500) antibodies were from R&D Systems, Inc. (cat. no. MAB3694), Sigma (cat. no. F3648), OriGene Technologies, Inc., (cat. no. TA500376) and Abcam (cat. no. ab27325). Secondary antibodies were DyLight 680 anti-mouse (cat. no. 072-06-18-06, 1:5,000 dilution; KPL, Inc.) and IRDye 680RD donkey anti-rabbit (cat. no. 926-32223, 1:5,000 dilution; LI-COR Biosciences). Membranes were blocked with 5% skim milk in phosphate-buffered saline (PBS) with 0.1% Tween-20 for 2 h at room temperature. The concentration of primary and secondary antibodies was consistent as recommended in the instructions. Then membranes were incubated with primary antibodies overnight at 4°C, followed by fluorescent secondary antibodies (1:5,000) for 1 h at ambient temperature. After washing three times in PBST, proteins were detected with Odyssey infrared imaging system (LI-COR Biosciences, Lincoln, NE, USA) following the manufacturer’s instructions.

Figure 3 (A–D) Western blotting of SERPINA1, ORM1, GPX3 and FN1 protein expressions in human serum. Western blotting was carried out using antibodies against SERPINA1, ORM1, GPX3 and FN1. Consistent with the proteomic results, the expressions of SERPINA1, ORM1 and FN1 were significantly upregulated (p<0.05), whereas, GPX3 protein expression was significantly downregulated (p<0.05) in PTR group, compared to the PTS group. (E) The overall protein profile of the NC, PTS and PTR groups. The gel was stained with Coomassie Brilliant Blue to ensure equal loadings during immunoblotting analysis. The expression levels of the four differential proteins were normalized based on the Coomassie Brilliant Blue staining. *P<0.05. ROC curve analysis using the ELISA profile of the (F) SERPINA1, (G) ORM1 and (H) FN1 for the 129 sera analyzed. SPSS software was used to construct ROC curves and to calculate the AUC. The solid lines in the ROC curves represent the plot of 1-specificity vs. sensitivity. Histogram was used to elucidate the profile of ELISA. Mean ± SD was used to express protein level. *P<0.05, **p<0.01. (I) Clinical diagnostic performance of the three proteins in PTR and PTS groups. PTR, platinum-resistant; PTS, platinum-sensitive; NC, normal control; ROC, receiver operating characteristic; ELISA, enzyme-linked immunosorbent assay; AUC, area under the curve.

ELISA

Based on the iTRAQ and western blotting findings above we selected four targets, SERPINA1, ORM1, GPX3 and FN1, the protein markers potentially associated with PTR, for the validation using ELISA method. ORM1, FN1, SERPINA1 ELISA kits were obtained from and utilized according to the manufacture’s instructions. ORM1 and FN1 serum samples were diluted 100-fold, and SERPINA1 serum samples were diluted 50-fold. All samples and standards were tested in triplicate. Absorbance was determined using Power Scan 4 multiplex microplate reader (DS Pharma Biomedical Co. Ltd., Osaka, Japan) and analysis of results was conducted by SPSS 16.0 software.

Statistical analysis

Statistical analyses were performed with SPSS 16.0 software (SPSS, Inc., Chicago, IL, USA). A comparative analysis of multiple groups was analyzed by one-way ANOVA or Kruskal-Wallis test and multiple comparisons were performed with the least significant difference test. Results are presented as means ± SD. ROC curves were used to determine the diagnostic value of the markers. P<0.05 was considered statistically significant.

Metabolical analysis

Serum sample preparation

Prior to serum preparation, samples were thawed on ice for 1 h. A total of 200 μl of serum (stored at −80°C) was resuspended with 800 μl cold CAN (stored at −20°C) mixed thoroughly, and precipitated on ice for 2 h. The samples were centrifuged at 4°C and 12,000 × g for 15 min. The supernatant was then transferred to a new tube. Subsequently, serums were lyophilized for 24 h using a freeze dryer (Beijing Songyuan Huaxing Biotechnology Co., Ltd., Beijing, China). Methanol (200 μl) was added to lyophilized samples, vortexed, sonicated for 5 min, and centrifuged (12,000 × g, 4°C, 5 min). The supernatant (150 μl) was collected for further analysis.

Metabolic signature via LC-MS

LC-MS analysis was performed using Agilent HPLC (1290 series) fitted with a Zorbax Rx-C8 column (5 μm, 150×2.1 mm; Agilent Technologies, Inc.) and coupled to a Bruker Daltonics’ micrOTOF-Q II high-resolution mass spectrometer. The flow rate was 0.25 ml/min, injection volume 5 μl and column temperature 30°C. The mobile phase was consisted of solvent A (0.1% formic acid, 99.9% water) and solvent B (0.1% formic acid, 99.9% ACN). HPLC conditions were 15% solvent B changing linearly to 40% solvent B over 5 min, to 80% solvent B over 10 min, 80% solvent B over 5 min, to 90% solvent B over 5 min, and then 90% solvent B over 15 min. Finally, mobile phase constituents reverted to starting conditions for 5 min re-equilibration. Total analysis time was 45 min. Mass spectral analysis was operated on a micrOTOF-Q II high-resolution mass spectrometer (Bruker Daltonics) linked to an Agilent HPLC (1290 series) by HyStar software (Bruker Daltonics). Electrospray ionization (ESI) (positive ion mode) was used to identify the molecular ion mass [M+H]. Source parameters are: ESI capillary voltage, 4,500 V; nebulizing gas pressure, 1 bar; drying gas flow, 6 l/min; and drying gas temperature, 220°C. Data were acquired in a mass range of 50–1,500 m/z.

Data analysis

Following LC-MS, raw MS data were converted into a matrix that is compatible with multivariate statistical analysis and interpretation by using an in-house set of tools, such as the Compass software package (Bruker Daltonics). Signals obtained from each sample in the chromatogram were segmented into a series of regions characterized by retention time and m/z using the Compass software, furthermore, the theoretical m/z values were compared with the experimental values from MS signals. Based on the exact m/z, elemental formulas were generated using the DataAnalysis software (Bruker Daltonics). C, H, N, O, P and S were the elements of the formulas. The lists of generated formulas were searched against the METLIN database (https://metlin.scripps.edu/) to identify compounds. Principal component analysis (PCA) was then operated utilizing Profile Analysis software (Compass software package; Bruker Daltonics).

Results

Proteomic analysis

Serum proteomic data analysis

To enhance the detection of the lower abundance proteins, most of the 14 abundant proteins were removed in equal volumes from each sample. Technical replicate samples were used to increase the reliability of the iTRAQ technique for relative quantitation. The relative expression levels, statistical parameters and the peptide information of identified proteins for each pool were obtained from two (replicate) peptide spectra data as described above. Subsequently, all the identified proteins were filtered with manually selected filter exclusion parameters. Thus, in the first iTRAQ data set (run 1), identification of 197 proteins was made. Similarly, 184 proteins were discovered in the second iTRAQ data set (run 2). The proteins identified from the two iTRAQ data sets were subsequently combined, and a total of 248 unique proteins were identified and quantified. Proteins were considered up- or downregulated when their ratios were >1.3 or <0.77 (Fig. 2C). Therefore 64 proteins were screened out as candidate biomarkers in one or two separate experiments as differentially expressed proteins between PTS and PTR sets: 33 of which were increased (PTR/PTS >1.3) and 31 were decreased (PTR/PTS <0.77). Candidate biomarkers selected by these criteria are summarized in Table II. For better understanding of the data structure in our experiment, a clustering algorithm for grouping proteins was required (Fig. 2B), and k-means clustering approach was then performed to group the data based on the degree of similarity between the PTS and PTR group. The different trends of identified proteins during PTS and PTR can be grouped into six subsets with a similar expression pattern (Fig. 2E). In the first subset, most of the 20 proteins were upregulated in a stepwise way in NC, BOC, PTS and PTR groups. Most are extracellular proteins involved in cell adhesion, cell communication and immune system process. The trend of these proteins differentially expressed in PTS and PTR groups were of interest as these could provide leads for potentially useful biomarkers of platinum status. In the second set, all the 10 proteins were specifically increased in PTS group and are therefore interesting candidates for EOC diagnosis or prognostic studies.

Figure 2 (A) Analysis of the 64 differential proteins between PTS and PTR group showing cellular component, molecular function and biological process. (B) Hierarchical clustering analysis representation of the altered serum protein profiles identified in PTS and PTR patients using a shotgun quantitative proteomic approach. Red, overexpressed genes; green, underexpressed genes. (C) Logarithmic plot of unused score of proteins in PTS vs. PTR pools. Red pentagrams, green diamonds and blue triangles denote R/S ratio >1.3-folds, <0.77-folds, and between them, respectively. (D) Biological interaction network analysis of proteins that showed differential abundance between PTS and PTR patients. (E–J) k-means clustering of differential protein profiles. The cluster gene number is labeled in the upper-left corner of each figure. Each line in the charts represents one differentially expressed protein. The shape of the line reveals the trend of differential expressed proteins in four groups. PTS, platinum-sensitive; PTR, platinum-resistant.

Gene ontology analysis

Data analysis of 64 unique proteins identified by two iTRAQ experiments was performed using the Blast2GO database (http://www.blast2go.com/b2ghome) and Web Gene Ontology Annotation Plot (WEGO: http://wego.genomics.org.cn/cgi-bin/wego/index.pl) to class each protein into its respective cellular components, molecular function and biological process (Fig. 2A). For the 64 differential proteins, the subcellular distributions were enriched mainly in extracellular region (90.6 and 78.1%) (the two numbers represent the upregulated and downregulated proteins proportion of the total, respectively), which imply that most of these proteins are secretary proteins. According to GO molecular function analysis, the top three common functional annotations were binding (87.5 and 93.8%), catalytic (28.1 and 56.3%), and enzyme regulator (25 and 31.3%). Most of the differential proteins were involved in biological regulation (90.6 and 84.4%), response to stimulus (81.3 and 87.5%), and pigmentation (84.4 and 71.9%). To clearly show the expression trend of differential proteins during cancer progress, k-means clustering method was used to classify the 64 protein. The results are shown in Fig. 2E. For further text mining, PANTHER Classification System (http://www.pantherdb.org/) was used to carry out the GO analysis. During tumor progression, 20 proteins were upregulated gradually in cluster A and most of these members participated in immune system process, cell adhesion, and cell communication, which would make sense in connection with drug resistance.

Pathway and biological interaction network analyses

Enrichment analysis of associated diseases and drugs was performed for the differentially expressed proteins that met our thresholds (fold, rank) using the IPAD browser (http://bioinfo.hsc.unt.edu/IPAD/) tools. Results of associated diseases and drug analysis showed that 30 proteins (14 upregulated, 16 downregulated) among them were significantly associated with EOC (indicated as pentagrams in Table II) and GPX3 was associated with cisplatin. In addition, DAVID Bioinformatics Resources 6.7 (http://david.abcc.ncifcrf.gov/home.jsp) was used to investigate possible interactions between the 30 proteins associated with EOC, which revealed that the differential proteins were significantly enriched in complement and coagulation cascades and ECM-receptor interaction. To model the signaling network potentially affected in the context of platinum status, the 17 focus proteins with fold changes between PTS and PTR group >1.5 were then subjected to network analysis using STRING software (http://string-db.org/). The network analysis identified ALB, APOA1, SERPINA1, FN1, ORM1 and TF as the major molecules affected in PTR patients (Fig. 2D). Candidate proteins with the most extreme deviation from the NC, BOC, PTS groups, including SERPINA1 (3.8-fold increased), ORM1 (3.32-fold increased), FN1 (2.35-fold increased) and cisplatin-associated GPX3 (1.35-fold decreased), were commonly identified in two iTRAQ experiments, and chosen for further analyses.

Co-occurrence analysis with COREMINE

COREMINE was used to performe co-occurrence analysis based on literature. The 64 differentially expressed proteins and the following list of keywords were used to interrogate the tools: drug resistance, neoplasm; drug resistance; drug resistance, multiple. In order to restrict the number of proteins potentially associated with drug resistance or MDR, p<0.01 was considered statistically significant. The cumulative frequency top 50 protein lists out of connected proteins, which showed a p<0.01 and the 64 differentially expressed proteins were compared to look for the degree of overlap. Finally, proteomic and co-occurrence analysis shared the following 11 proteins: ALB, CRP, FN1, S100A8, TF, VWF, APOC2, APOE, CAT, CD44, F2.

Western blotting

A total of 27 serum samples, composing 9 from NC group, 9 from PTS group, and 9 from PTR group, were subjected to western blotting against SERPINA1, ORM1, GPX3 and FN1. These proteins were selected for western blotting primarily the following factors: big fold changes of differential expression, correlation with cancer/drug resistance from a literature-based text mining, the expression trend in four pools and the availability of commercial antibodies. Our results indicated that three of the four candidates have similar trends with the proteomic results (SERPINA1, ORM1, FN1) in the serum of PTR cases, compared to PTS cases, which implied the credibility of proteomic analysis (Fig. 3A–D). One-way ANONA was applied to calculate means ± SD from each group along with p-values.

Clinical relevance of SERPINA1, ORM1, GPX3 and FN1

In the initial experiment 10 NC, 10 PTS and 10 PTR samples were used to validate the expression levels of SERPINA1, ORM1, GPX3 and FN1. The results illustrated that statistical significant difference between PTS and PTR was seen for SERPINA1, ORM1 and FN1, but not GPX3 (data not shown). However, the expression of GPX3, which was observed to be downregulated (FC=1.35) in proteomic analysis, was not significantly different (p>0.05) by ELISA. Consequently, we carried a full validation study for SERPINA1, ORM1 and FN1, using the entire 129 samples collected (data are shown in Table I). Consistent with the iTRAQ results in the previous experiment, relative quantitation of SERPINA1, ORM1 and FN1 (Fig. 3F–I) between PTS 52 samples and PTR 44 samples were all found to be significantly upregulated (p<0.05). Likewise, to further evaluate the diagnostic significance of these three proteins, a ROC curve analysis was constructed for each protein by plotting sensitivity vs. specificity. The overall predictive accuracy of each protein was reflected by the area under the ROC curve (AUC), a commonly used indicator for estimating the diagnostic efficacy of a potential biomarker. FN1 and SERPINA1 with ROC areas of 0.679 and 0.666, respectively, suggest that their use as a biomarker may not be reliable. Unlike the FN1 and SERPINA1, the AUC for ORM1 was 0.91 and its sensitivity and specificity for predicting PTR was 71 and 97.4%, respectively, which could clearly separate the PTS patients from the PTR individuals. These results highlight a potential role for ORM1 in the response to platinum therapy.

Metabolic analysis

A total of 25,800 metabolic features was observed in our study. Data of identified compounds were subjected to t-test analysis to identify significant metabolic patters and variations. Compounds having p<0.01 and fold-change >2 were considered as statistically significant. PCA and multivariate statistics were then applied to identify key PTR-associated metabolic perturbations in PTR compared to PTS. Unsupervised PCA of the resultant data showed clear metabolic separation of PTR from PTS along the first principal component, and clear distinctions of EOC (PTR and PTS) from healthy individuals along the second principal component (Fig. 4A). The BOC sera were not obviously grouped because of their limited sample size. PCA loading plots (Fig. 4B) provided six metabolite features contributing to the separation of groups along PC1 and PC2. Six known compounds were identified using NMR based on database (Table III). The levels of the six potential biomarkers in blood from PTR, PTS and NC group were determined by LC-MS/MS. Compared to PTS, PTR exhibit a specific metabolic trait characterized by decreased levels of calycanthidine and increased levels of 1-monopalmitin, ricinoleic acid methl ester, polyoxyethylene (600)mono-ricinoleate/glycidyl stearate. Furthermore, the concentration of dodemorph was higher and of C16 sphinganine was lower in the EOC compared to NC (Table III). ROC curve was used to calculate sensitivity and specificity of the four biomarkers for PTR compared with PTS. The AUC for 1-monopalmitin, ricinoleic acid methyl ester, polyoxyethylene (600)mono-ricinoleate and calycanthidine was 0.892, 0.900, 0.883 and 0.109, respectively, and their sensitivity and specificity for predicting PTR were 83.8 and 75%; 81.1 and 86.4%; 83.8 and 75%; 90.9 and 73%, respectively, which could clearly separate the PTS patients from the PTR individuals. The combinational four biomarkers achieved an AUC value (AUC=0.925) while the statistical analysis provided 86.5% sensitivity and 81.8% specificity for the prediction of PTR (Fig. 5).

Figure 4 Multivariate statistical analysis. (A) PCA score plots. Blue rings, PTS; red triangles, PTR; blue plus, BOC; green multiplication, NC. (B) PCA loading plots. Selecting compounds far away from the center were assumed to have a greater contribution to the classification of PTS, PTR, BOC, and NC are numbered. PCA, principal component analysis; PTS, platinum-sensitive; PTR, platinum-resistant; BOC, benign ovarian cyst; NC, normal control.

Figure 5 (A–E) The ROC curves of the candidate biomarkers. (F) Clinical diagnostic performance of the four metabolites in PTR and PTS groups. ROC, receiver operating characteristic; PTR, platinum-resistant; PTS, platinum-sensitive.

Table III Relevant analytical data for the metabolites identified in PTR, PTS, BOC and NC groups.

Table III

Relevant analytical data for the metabolites identified in PTR, PTS, BOC and NC groups.

Retention time (min)	Adduct	m/z	Error (mDa)	σ value	Molecular formula	Trend	CAS/PubChem	Possible metabolite
19.9	[M+H]	282.2807	−1.5	0.0064	C18H35NO	Up (EOC/NC)	1593-77-7	Dodemorph
9.4	[M+H]	274.2777	−3.6	0.0329	C16H35NO2	Down (EOC/NC)	4266342	C16 sphinganine
16.9	[M+H]	313.2775	−3.9	0.0334	C19H36O3	Up (PTR/PTS)	141-24-2	Ricinoleic acid methyl ester
16.8	[M+H]	331.2871	−2.8	0.006	C19H38O3	Up (PTR/PTS)	542-44-9	1-Monopalmitin
19.2	[M+H]	341.3071	−2.1	0.0067	C21H40O3	Up (PTR/PTS)	977137-78-2	Polyoxyethylene (600)mono-ricinoleate
19.2	[M+H]	341.3071	−2.1	0.0067	C21H40O3	Up (PTR/PTS)	7460-84-6	Glycidyl stearate
19.9	[M+H]	361.2419	−4.6	0.0235	C23H28N4	Down (PTR/PTS)	5516-85-8	Calycanthidine

[i] PTR, platinum-resistant; PTS, platinum-sensitive; BOC, benign ovarian cyst; NC, normal control; m/z, mass-to-charge ratio; EOC, epithelial ovarian cancer.

Discussion

Cisplatin is one of chemotherapeutical agents commonly used to treat EOC, which causes DNA damage via forming inter- and/or intrastrand DNA adduct lesions and eventually cytotoxicity. However, the benefits of chemotherapy can be attenuated because of the emergence of platinum resistance. To eradicate the mechanisms of platinum resistance in EOC is a difficult task. The recent development of proteomic approaches applied to investigate drug-resistance mechanisms has greatly helped in addressing these issues. Comparative proteomic approach is a powerful tool, which might help to guide future research and cross validation of various proteomic profiling with a high throughput. In our experiment, a panel of 64 different proteins that have altered expression in PTR patients were compared to the parental PTS group using a shotgun quantitative proteomics approach, and four of these proteins were confirmed with western blotting and ELISA. The results of serum FN1, SERPINA1, GPX3 and ORM1 from 2D-LC-MS/MS analysis were validated in a 139 cohort using a different methodology. Western blotting and ELISA confirmed that the serum level of FN1, SERPINA1 and ORM1 was upregulated in PTR group, which indicated that the FN1, SERPINA1 and ORM1 serum levels might be a tool for screening and diagnosis of PTR. However, it should be noted that although the change in direction (up- or down-regulated) of GPX3 detected by western blotting between PTS and PTR group was consistent with iTRAQ, the changes measured by ELISA assay in 43 patients was not statistically significant (p>0.05). The differences in fold change determined by iTRAQ, western blotting and ELISA can be attributed to methodological factors such as the use of isobaric tags and/or differences inherent in the technical method. ROC curve analysis was applied to find the cut-off value of serum FN1, SERPINA1 and ORM1 to discriminate between PTS and PTR group. The sensitivity and specificity were calculated. ROC curves show, ORM1 with 71% sensitivity and 97.4% specificity could give a higher accuracy (Fig. 3). However, FN1 and SERPINA1 were not reliable for clinical diagnosis because of low sensitivity and specificity. Our comprehensive study of proteomics led to the possibility that monitoring the level of serum ORM1 could be clinically useful for the screening and diagnosis of PTR patients.

α-1-antitrypsin (SERPINA1) is an inhibitor of serine proteases principally secreted by hepatocytes, but also by monocytes, neutrophils, macrophages and alveolar epithelial cells, and plays a critical role in modulating host immunity, inhibiting T lymphocyte-mediated antitumor function and thereby accelerated tumor proliferation, and metastasis (3–5). Moreover, there have been numerous studies documenting a link between SERPINA1 and various cancers, although for most the mechanism for the linkage is unclear. Giving the expression levels of SERPINA1 in rat bladder tumor tissues were 2.5-fold higher than those in normal bladder tissues using two-dimensional difference gel electrophoresis (2D-DIGE) (6). Similar conclusions were also obtained in pancreatic tumors, hepatocellular carcinoma, non-small cell lung cancer, gastric cancer in gastric juice, prostate cancer patients and malignancy in insulinomas. Much attention has been focused on the role of SERPINA1 as a tumor suppressor, but no report has shown directly the relation between SERPINA1 and chemotherapy drugs. In our study, the contribution of SERPINA1 to drug resistance was implicated in human serum samples. SERPINA1 shows one of the largest fold increases (3.8-fold increased) in protein expression level in PTR cohort compared with PTS cohort (p<0.05). Nevertheless, our data are in good agreement with prior studies elicited above, but disagree with the results of Normandin et al (7).

Fibronectin is a glycoprotein that is involved in cell adhesion, signal transduction and migration processes including embryogenesis, wound healing, blood coagulation, host defense, and metastasis, especially possibly suppression of apoptosis (8–10). There have been many reports on the relation between FN1 and human tumors. Similar thesis reported that fibronectin was involved in Ras, Erk, Akt and ECM pathways and mediate various signals such as cancer cell adhesion, growth migration and invasion (11,12). Akiyama et al showed that FN1 played a causal role in tumor neovascularization and metastasis (13). In addition, a recent study found that FN1 is one of the key genes in regulating SOX2 cell migration, invasion, colony formation and drug resistance in ovarian cancer cells (14–16). Qian et al also indicated that FN1 is targeted by let-7g to promote mammary carcinoma cell migration and invasion via p44/42 MAPK and MMPs (17). FN1 was also suggested as a marker for renal cell carcinoma aggressiveness (18,19). Moreover, FN1 was shown to be a direct target gene for miR-1 and miR-200. While miR-1 may play a role as a tumor suppressor gene in laryngeal carcinoma. Similarly, miR-200 is crucial for the maintenance of epithelial identity, behavior, and sensitivity to chemotherapy in ovarian cancer cell line (20,21), which confirmed our previous observation by miRNA microarrays with samples obtained from the same patients as this study (22). All these findings suggest that a functional relation is present between FN1 and platinum response, which supports our data in EOC.

As a member of guutathione peroxidases, GPX3 is located in 5q23 and has critical roles in the detoxification of hydrogen peroxide and other oxygen-free radicals. Previous studies have demonstrated that GPX3 had a broader downregulated pattern in a variety of cancers, such as ovarian, cervical, thyroid, head and neck, lung, colorectal, gastric, gallbladder, breast, and esophageal cancers than in healthy controls. These reports suggest that GPX3 contains a tumor-suppressor function. The mechanisms involved in mediating the GPX3 tumor-suppressor function are mainly due to promoter hypermethylation (23), the downregulation of c-Met expression (24,25), and the role of antioxidant enzymes which are involved in reactive oxygen species (ROS) metabolism. As a messenger molecule, ROS might increase cancer cell proliferation, genetic mutations, instability, and thereby invasion and angiogenesis (26). In addition, ROS also mediates the induction of tumor cell death via many chemotherapeutic agents such as platinum (27). Although the researchers failed to measure the serum concentration of GPX3, this statement is supported by our results. However, GPX3 is identified to be highly expressed in clear cell adenocarcinoma compared to control tissues at a DNA, mRNA and protein level on cell lines and clinical samples of ovarian clear cell adenocarcinoma (28,29). Although the molecular biological mechanism is not clarified, these results might indicate that GPX3 activity is tumor-specific. In our present study, GPX3 was shown to be downregulated in PTR group compared with PTS group, which confirmed the previous results, but the exact mechanism, in response to anticancer drugs remains to be further understood.

The α-1-acid glycoprotein primarily synthesized by the liver is an acute-phase reactant with immunomodulatory and immunosuppressive properties (30) and its serum levels are increased by inflammation, stress, and chronic disease such as cancer (31). Two main biological functions were involved in α-1-acid glycoprotein, binding and transporting of endogenous substances or drugs, and a strong immunomodulatory function. Previous investigations in patients with carcinoma of the breast, lung, ovary and endometrium have suggested that serum ORM1 concentrations were increased two times higher than that in healthy individuals, and ORM1 might act as blocking agent protecting tumor cells against immunological attack, thereby contributing to the ‘immune escape’ of the tumor (32,33). ORM1 can also interfere with cytokine function by inducing the secretion of soluble TNFα receptor and IL-1, -6 and -12 receptor antagonist (30,34). Although the mechanisms by which ORM1 mediates its functions are not fully understood, ORM1 has been shown to bind to the chemokine receptor CCR5 in macrophages, the asialoglycoprotein receptor in hepatocytes, the surface lectin-like receptor Siglec-5 in neutrophils and can also modulate TNFα-induced phosphorylation of p38 MAPK, MEK1/2, c-Jun N-terminal kinase which is required for angiogenesis in macrophages (35–39), but not VEGF-induced signaling. In addition, ORM1 has been shown to enhance endothelial cell migration and capillary tube formation in vitro (40). Moreover, several reports suggested that the serum levels of α-1-acid glycoprotein influenced the pharmacokinetics (PK)/pharmacodynamics (PD) of chemotherapy drugs such as docetaxel, PTX and imatinib (41–44). As these reports remarked, α-1-acid glycoprotein may function as a carrier of PTX from the serum into the liver via the α-1-acid glycoprotein receptors, and this might result in the enhancement of the PTX metabolism. Although ORM1 has been reported to be associated with cancers or metabolisms of chemotherapy drugs according to previous reports, no studies have underlined the importance of ORM1 in cisplatin resistance in PTR patients, and this is the first time that ORM1 was identified as an important biomarker of response to cisplatin-based chemotherapy. The mechanism of this phenomenon may be attributed to the PK/PD changes of cisplatin, however, further studies will be required to fully understand ORM1 functional roles in drug resistance.

In the present study, we undertook a non-destructive metabolomic technique (HPLC-micrOTOF-Q II MS/MS) to investigate the metabolic traits. All the six metabolites in our experiment were identified as fatty acid or derivatives. Profiling of metabolomics elucidated changes in the levels of fatty acid metabolism, which confirmed our previous observations by proteome approach and conclusions of many addressed articles on chemotherapeutic resistance and metabolism, and served as an insightful reference to the mechanism research of drug resistance. Fatty acid synthese (FASN) providing proliferating cancer cell lipids for membrane biogenesis was assumed to have metabolic characteristics of cancel cells (45). Expression level of FASN is significantly upregulated in kinds of neoplasm and correlates with poor prognosis, but in a health individual is very low even undetectable, suggesting that FASN serves as a metabolic oncogene (46). Fatty acids were used by proliferating tumor cells for membrane assembly, lipid modifications of proteins, and as an efficient source of energy, all are required to sustain neoplasm growth and survival (47). Furthermore, it is shown that FASN is overexpressed in drug-resistant breast neoplasm cell line (MCF7/AdVp3000), and that reducing the expression of FASN increased the drug sensitivity in MCF7 and MDA-MB-468 (breast cancer cell lines) (48). Analogously, FASN was reported to be associated with acquired trastuzumab/docetaxel/5-fluorouracil resistance in breast cancer or radiation and gemcitabine in pancreatic neoplasm. FASN also played an active role in chemotherapy resistance of HER-2/neu-induced breast neoplasm. FASN not only played a key role in acquired resistant phenotype but also in inherent resistant phenotype in hepatocellular carcinoma (49). Roodhart et al identified two platinum-induced polyunsaturated fatty acids which induce resistance to chemotherapeutic drugs. When the central enzymes associated with the production of polyunsaturated fatty acids were blocked, the mesenchymal stem cells induced resistance which was prevented (50). All the above further confirmed the metabolism abnormality of fatty acid is induced by PTR.

In conclusion, we identified a panel of new ovarian epithelial cancer serum protein biomarkers, which have an indicator value for platinum status and allow patients who have a high chance of being resistant to cisplatin-based chemotherapy to receive an alternative therapy. Although thousands of metabolites were identified, links were weak and annotated only a small proportion of the total analytes. In further studies, the role of these differentially proteins or compounds in cisplatin resistance needs to be validated on a large scale to evaluate the clinical benefit of using these candidate biomarkers for diagnosis or prognosis analyses. The contribution of the identified biomarkers in cisplatin resistance should also be explored to help understand and design chemosensitizing agents. In addition, our study demonstrated that metabolomics and proteomics could validate one another partially and their combination could better elucidate the mechanism of drug resistance and provide candidate molecular targets for personalizing therapeutic interventions and treatment efficacy monitoring.

Acknowledgements

This study was supported by the National Natural Science Foundation of China (grant no. 81572579), Guangxi Scientific Research and Technological Development Program Topics (no. 14124004) and the Specialized Research Fund for the Doctoral Program of Higher Education (nos. 20124503110003 and 2013.01–2016.10).

References