Cytobrush and urine samples were obtained from cervical cancer patients and healthy volunteers at the Chulabhorn Hospital from July 2014 to April 2015. All participants received detailed information regarding the study objectives and gave written informed consent. The present study was approved by the Ethical Review Board of the Chulabhorn Hospital (Bangkok, Thailand; no. 31/2554). All participants were subsequently performed HPV genotyping analysis and normal routine cervical cancer screening at Chulabhorn Hospital. The samples for HPV genotyping were obtained using a cytobrush by gynecologists or well-trained general practitioners during pelvic examinations at Chulabhorn Hospital. The first morning urine samples were collected and fractionated within 24 h. Each specimen was collected in 10 ml conical tubes and the protease inhibitor cocktails (Sigma-Aldrich; Merck KGaA, Darmstadt, Germany) were added. The detail of the 13 urine samples from HPV-negative females who were diagnosed as non-cervical cancer was shown in Table IA. While, the other 24 urine samples were collected from females with cervical cancer at different stages (S1B1, SIB2, SIIA, SIIA1, SIIB, SIIIB, SIVA and SIVB) as shown in Table IB.

The Bethesda 2001 report system was used as the standard protocol for classification of all cervical cytology slides (18). The different grades, such as squamous cell carcinoma and adenocarcinoma, were determined using the cervical cytology slides by qualified pathologists via a normal routine at Chulabhorn Hospital.

The Linear array HPV testing (Roche Diagnostics Indianapolis, IN, USA) was employed to identify HPV genotypes. This kit was for identification of 37 HPV genotypes including 12 high-risk, 8 probable high-risk and 17 low-risk types. Briefly, 450 bp fragments of the L1 region of HPV were first amplified by polymerase chain reaction, followed by hybridization using a reverse line blot system for the simultaneous detection of up to 37 HPV genotypes (i.e., genotypes 6, 11, 16, 18, 26, 31, 33, 35, 39, 40, 42, 45, 51, 52, 53, 54, 55, 56, 58, 59, 61, 62, 64, 66, 67, 68, 69, 70, 71, 72, 73, 81, 82, 83, 84, IS39, and CP6108) (19).

Urine samples were stored at 4°C and then centrifuged at 1,120 × g for 5 min at 4°C to remove cell debris. The supernatants were precipitated with 50% v/v acetone for 16 h followed by centrifugation at 12,000 × g for 10 min at 4°C. The pellets were kept at −80°C until further use.

The pellets of urine samples from 5 normal (N9-N13) and from 6 cervical cancer patients at different stages (C19-C24) were resuspended in 50 mM ammonium bicarbonate (NH₄HCO₃) separately and the protein concentration was determined by spectrophotometry using the Bradford method (20). The samples were pooled by obtaining 5 µg from each of the 5 normal urine samples and 5 µg from each of the 6 cancer urine samples. A total of 5 µg of the pooled samples were separately reduced with 100 mM DTT (10 mM final concentration) for 5 min at 95°C and alkylated with 1/10 volume of 200 mM iodoacetamide prior to incubation for 30 min in the dark at room temperature. Digestion was performed overnight at 37°C by adding 1:50 (w/w) of sequencing grade trypsin (Promega Corporation, Madison, WI, USA) to the protein in each sample. The digestion reaction was stopped by adding formic acid to reach a 1% final concentration and the samples were dried completely by SpeedVac.

To prepare samples for label-free LC-MS/MS quantification, the digested samples were dissolved in 0.1% formic acid in H₂O and separated on a nanoACQUITY system (Waters Corporation, Milford, MA, USA). All pooled urine samples were run in triplicate. The samples were injected into a nanoACQUITY UPLC column (1.7 µm BEH, 75 µm × 200 mm C18) at a flow rate of 300 nl/min. The column temperature was maintained at 40°C. The LC gradient (1–50% B in 70 min, 50–90% B in 5 min, followed by 15 min on 90% B) was performed using 0.1% formic acid in H₂O as solvent A and 0.1% formic acid in ACN as solvent B. The eluting peptides were analyzed directly via MS/MS on an amaZon speed ion trap mass spectrometer (Bruker Corporation, Billerica, MA, USA) equipped with a captive-electrospray ion source. The positive mode was used with a spray voltage of 1,300 V and the capillary temperature was set at 150°C. Mass spectra were acquired from 400–1,400 m/z using parameters optimized at 922 m/z with a target of 500,000 set for ion charge control and a maximum acquisition time of 100 msec. The scan range was 50–3,000 m/z. MS/MS data were processed by Bruker Compass 1.4 software (Bruker Corporation).

Progenesis label-free LC-MS software version 3.1 (Nonlinear Dynamics, Ltd., Newcastle upon Tyne, UK) was used to process the raw data obtained from LC-MS/MS and to calculate the significant changes. The retention time of each sample was aligned and the reference sample was set up. For all of the replicates, their retention times were aligned to the established reference and the intensities of the peak were then normalized. Three criteria were used to filter all data prior to exporting the MS/MS output files to identify proteins using Mascot software (www.matrixscience.com), including: i) An analysis of variance (ANOVA) with Tukey's post hoc test score shown between experimental groups of P≤0.05; ii) featured mass peaks with +2, +3 and +4 as the charge states; and iii) only data with >1 isotopes or peptides. From the aforementioned Progenesis software, all of the exported MS/MS spectra were generated using a Mascot generic file. Mascot software version 2.4.0 was used for the identification of peptides. The SwissProt human protein database (Matrix Science, Ltd., London, UK; www.matrixscience.com) program was set up as follows: The MS/MS mass tolerance was set at 0.6 Da, peptide mass tolerance was set to 1.2 Da, carbamidomethylation was set as a fixed modification and ≤1 missed cleavages were allowed. A false discovery rate threshold of 1% was applied and identification of two or more unique peptides and two or more peptides were required for positive identification, respectively.

Venn Diagrams web tool (http://www.interactivenn.net) was used to compare the expression of proteins from cervical cancer patients between identified proteins by LC-MS/MS and the database of Human Proteome Atlas (www.proteinatlas.org/humanproteome/pathology/cervical+cancer). The classification and functional analysis of differential expressed proteins was performed using the PANTHER database (www.pantherdb.org/). The Gene Ontology (GO) standard was employed to categorize proteins according to ‘Molecular function’ and ‘Biological process’. The Search Tool for Retrieval of Interacting Genes/Proteins (STRING) database (string-db.org) version 9.0 was used to obtain protein-protein interactions. Protein mode from STRING was analyzed, and the interactions from upregulated and downregulated urinary expressed proteins were identified.

The expression levels of proteins were analyzed by western blot analysis. Leucine-rich α-2-glycoprotein (LRG1), multimerin-1 (MMRN1), S100 calcium-binding protein A8 (S100A8), serpin B3 (SERPINB3) and cluster of differentiation-44 antigen (CD44) were selected to be validated for expression. The pellets of 8 normal (N1-N8) and 18 different stages (C1-C18) of cancer urine samples were resuspended in 50 mM ammonium bicarbonate (NH₄HCO₃). A total of 20 µg of urine samples were diluted in sample buffer (50 mM Tris, pH 6.8, 10% glycerol, 2% SDS and 0.01% bromophenol blue) and resolved by SDS-PAGE. Following SDS-PAGE, the proteins were transferred onto 0.20 µm polyvinylidene difluoride membranes (Pall Life Sciences, Port Washington, NY, USA). The membranes were blocked with 5% (w/v) non-fat dry milk in TBST [50 mM Tris, pH 8.0, 150 mM NaCl, and 0.1% Tween-20 (v/v)] for 1 h. Membranes were washed 10 min for 3 times in TBST and incubated at 4°C overnight with the following primary antibodies: Mouse anti-LRG1 (1:2,000; ab57992; Abcam, Cambridge, UK), rabbit anti-MMRN1 (1:2,000; ab130585; Abcam), rabbit anti-S100A8 (1:1,000; ab92331; Abcam), rabbit anti-CD44 (1:2,000; ab51037; Abcam) and rabbit anti-SerpinB3 (1:5,000; ab126752; Abcam). Then the membranes were washed 3 times in TBST, and incubated with horseradish peroxidase-conjugated anti-mouse (P0260) or anti-rabbit (P0217) antibodies secondary antibody (Dako; Agilent Technologies, Inc., Santa Clara, CA, USA) at room temperature for 1 h. Following washing 3 times with TBST, the reaction was detected by chemiluminescence with the WesternBright ECL detection kit (Advansta Inc., Menlo Park, CA, USA). Darkroom development techniques were employed to expose and acquire images of the membranes.

The receiver operating characteristic (ROC) curve and area under the curve (AUC) were used to determine the performance of the five individual potential biomarkers by employing GraphPad Prism software version 7 (GraphPad Software, Inc., La Jolla, CA, USA). The combination of the five potential biomarkers was also analyzed using logistic regression with SPSS software version 11 (IBM Corp., Armonk, NY, USA).

The body filtrates from urine such as water, electrolytes, salts, nitrogenous waste and proteins are derived from the plasma, kidneys and urogenital tract. The amount of urinary protein excretion normally is <150 mg/day and needs to be concentrated for use in experiments (21). Label-free quantitative proteomics is the best choice for analyzing samples with a low level of proteins. The pooled urinary samples from 5 normal individuals (N9-N13) and 6 cervical cancer patients (C19-C24) were analyzed by LC-MS/MS, as shown in Table IIA and B. Significant changes were calculated using Progenesis label-free LC-MS software version 3.1. A list of proteins with 2 or more matched peptides was compiled. A total 133 proteins with 1% FDR were identified. A total of 60 upregulated and 73 downregulated urinary proteins were expressed when comparing normal and cervical cancer samples. Serotransferrin (TF) was observed to be the most significantly increased protein (2.91-fold, P=0.000004) while Kininogen-1 (KNG1) was the most significantly decreased protein (1.73-fold, P=0.000003). The results also revealed that LRG1, low-density lipoprotein receptor-related protein 2 (LRP2), Myosin light chain kinase (MYLK), SPARC-like 1, Zinc finger CCHC domain-containing protein 8, MMRN1 and nectin-2 were notable proteins that had >2 fold expression while aminopeptidase N had downregulated protein expression (5.31 fold) and another 10 proteins had between 3.0–4.33 fold altered expression including protein S100A8, trefoil factor 2, apolipoprotein E (APOE), apolipoprotein D (APOD), DNA (cytosine-5)-methyltransferase 3B (DNMT3B), CD44, SERPINB3, macrophage colony-stimulating factor 1 (CSF1), epidymal secretory protein E1 and ribonuclease pancreatic. The large molecular weight proteins of >100 kDa were obtained using this method, including LRP2 (521.6 kDa), MYLK (210.7 kDa) and MMRN1 (138 kDa). These proteins of interest (LRG1, MMRN1, S100A8, SERPINB3 and CD44) were selected to confirm their expression by western blot analysis.

Venn diagram comparing protein expression from cervical cancer between identified proteins by LC-MS/MS and the database of Human Protein Atlas was shown in Fig. 1A. Twenty-one identified proteins were overlapped with The Human Protein Atlas Database, including ABCB5, AMY2A, ANGPTL2, C3, CD44, CD55, COL6A1, CRNN, CRYL1, DNAH3 ESAM, GALNS, HEXA, KRT17, MACF1, MXRA8, PGK2, PIP, PTGDS, SERPINB3 and WFDC2. Interestingly, 112 identified proteins have never been reported in the database of Human Protein Atlas for cervical cancer.

The up- and downregulated urinary proteins from Table IIA and B were analyzed by GO with the PANTHER database system to obtain ‘Molecular function’ and ‘Biological process’ data, as presented in pie charts in Fig. 1B-E. The molecular functions of 60 upregulated proteins from Fig. 1B were classified into 5 groups: Binding (50%), catalytic activity (36.8%), structural molecular activity (5.3%), transporter activity (5.3%) and receptor activity (2.6%), while the biological processes of these proteins (Fig. 1C) were classified into 11 groups with the highest percentage observed for cellular process (24.7%), followed by metabolic process (16.9%) and response to stimulus (10.1%). The molecular functions of the 73 downregulated proteins, as presented in Fig. 1D, were classified into 6 groups and were mainly involved in binding (37.5%) and catalytic activity (37.5%), while the biological processes of these proteins (Fig. 1E) were classified into 12 groups and were mainly involved in cellular process (25.8%) and response to stimulus (9.3%).

The STRING database was used for analysis of the protein-protein interactions and was able to predict the similar functions of proteins by accessing various free resources. The ‘molecular function’, ‘cellular component’ and ‘biological process’ from GO and KEGG pathways enrichment analyses were obtained. The present study analyzed the 60 and 73 identified proteins that were up- and downregulated, respectively (Table II), when the pooled urinary samples of normal individuals and patients with cervical cancer were compared. The results for the interaction map of the upregulated proteins was presented in Fig. 2A. The results for the associations between these proteins revealed that 40 of the 60 proteins had good interactions. The big cluster was involved in protein binding functions from molecular function (GO), metabolic process, blood coagulation, cell-cell adhesion, regulation of cell morphogenesis and cell motility from biological process (GO). A total of 19 proteins were involved in binding including, albumin (ALB), fibrinogen β-chain (FGB), fibrinogen γ-chain (FGG), fibronectin (FN1), haptoglobin, nesprin-1, zinc-α-2-glycoprotein, vitamin D-binding protein, TF, LRP2, protein RRP5 homolog and mitogen-activated protein kinase kinase kinase 7 (MAP3K7) while 12 proteins were involved in blood coagulation, such as fibrinolysis, and fibrin clot formation including, ALB, FGG, FGB, FN1, TF, MMRN1, apolipoprotein A, urokinase-type plasminogen activator, complement C3, ceruloplasmin and lysosomal-trafficking regulator. There were some proteins of interest that were involved in multiple functions: For example, MAP3K7 was not only involved in binding functions but was also involved in cell-cell adhesion and metabolic process; and microtubule-actin cross-linking factor 1 was involved in metabolic processes, the regulation of cell morphogenesis for differentiation and the regulation of anatomical structure morphogenesis. MMRN1 was also involved in the stress response and cell adhesion function.

Fig. 2B presents the STRING analysis of 73 downregulated proteins from the same comparison. The results revealed that 49 of the 73 proteins had good interactions. The big cluster was involved in different protein functions such as stress-response, cell adhesion, leukocyte migration, wounding response and extracellular matrix organization from biological process (GO). A total of 20 proteins were involved in the stress response with fold change range of 2 to 3.62 including DNMT3B (3.41-fold), APOD (3.62-fold), CSF1 (3.55-fold) and APOE (3.04-fold). Some proteins were involved in multiple functions: For example, CD44 and S100A8 were involved in all functions associated with biological process while MMRN1 were involved in only the stress response and cell adhesion function.

Notable proteins were selected to confirm the association with the urine of normal individuals and patients with different stages of cervical cancer including 8 normal (N1-N8), 2 samples of stage IB1 (C1-C2), 9 samples of stage II (C3-C11), 4 samples of stage IB1 (C12-C15) and 3 samples of stage IVB (C16-C18), as determined by western blot analysis. The expression of LRG1, MMRN1, S100A8, SERPINB3 and CD44 were confirmed as shown in Fig. 3. LRG1 and MMRN1 were upregulated in stages IIB, IIIB and IVB of cervical cancer while S100A8, SERPINB3 and CD44 had reduced expression in stage I and in some stage II cancer samples. Notably, CD44 was highly expressed in normal and stage IB1, and had low expression in stages II, III and IV.

The results from immunodetection by western blot analysis revealed that LRG1, MMRN1, CD44, S100A8 and SERPINB3 could be potential biomarkers for cervical cancer. All five proteins were tested individually and in combination for sensitivity and specificity using logistic regression on SPSS software and ROC curves, respectively. The ROC curves were plotted for these proteins to discriminate between cervical cancer (all stages) and control urine samples (Fig. 4). The results showed the AUC of LRG1, MMRN1, CD44, S100A8 and SERPINB3 were 0.993, 0.785, 0.938, 0.931 and 0.986, respectively. LRG1 and SERPINB3 had the highest AUC among all 5 proteins. The combination of the 5 biomarker panel was calculated and the value of the AUC was increased to 1 (100% sensitivity, 87.5% specificity), which was better than LRG1 or SERPINB3 individually.