Mining the Ovarian Cancer Ascites Proteome for Potential Ovarian Cancer Biomarkers

MATERIALS AND METHODS

Patients and Specimens—

Ascites fluid was obtained, with informed consent and Institutional Review Board approval, from women with advanced stage ovarian cancer undergoing paracentesis. These patients had stage IV serous ovarian carcinoma, and they had been treated previously with surgery plus carboplatin/paclitaxel chemotherapy.

Sample Collection and Preparation—

Ascites fluids were aliquoted in 1-ml portions and centrifuged at 16,000 × g for 30 min at 4 °C three times to separate the fluid from lipids and cellular components.

Gel Filtration—

Gel filtration was performed using a 0.75 × 60-cm TSK-Gel G3000SW column (Tosoh Bioscience) attached to an Agilent 1100 HPLC system. The column was equilibrated with either (i) phosphate/sulfate buffer (10 mm NaH₂PO₄, 10 mm Na₂SO₄, pH 6.8) or (ii) 100 mm ammonium bicarbonate buffer, pH 7.8. 500 μl of ascites were loaded onto the system at a flow rate of 0.5 ml/min for 1 h. Forty successive injections were performed, collecting eluted fractions at 1-min intervals, starting at 20 min (column void volume). A total of 39 fractions, containing 20 ml each, was collected for each buffer type. Gel filtration experiments were performed in duplicate for each buffer. Each fraction was then analyzed for kallikrein 6 (KLK6) and total protein followed by lyophilization to dryness.

Centrifugal Ultrafiltration—

15 ml of ascites were added to a prerinsed 50- or 100-kDa nominal molecular mass limit cutoff Amicon Ultra-15 centrifugal filter device (Millipore). After 5 min of centrifugation at 4000 × g in a swinging bucket rotor, unfiltered ascites was topped to 15 ml with water. This process was repeated until 15 ml of filtered ascites were obtained. The filtered ascites was then lyophilized to dryness and underwent trypsin digestion (see below). The 50-kDa-filtered ascites was analyzed directly by LC-MS/MS, whereas the 100-kDa-filtered ascites underwent strong cation exchange liquid chromatography (see below) prior to LC-MS/MS analysis with a reverse-phase C₁₈ column.

KLK6 ELISA—

The concentration of KLK6 in each eluted gel filtration fraction was measured by a sandwich-type immunoassay (20). In brief, a KLK6-specific monoclonal antibody (clone 27-4; developed in house) was first immobilized in a 96-well white polystyrene plate by incubating 250 ng/100 μl/well in a coating buffer (50 mmol/liter Tris, 0.05% sodium azide, pH 7.8) overnight. After washing three times with washing buffer (5 mmol/liter Tris, 150 mmol/liter NaCl, 0.05% Tween 20, pH 7.8), 50 μl of each pooled gel filtration fraction diluted 1:3 in 6% BSA or 50 μl of KLK6 standards were pipetted into each well in addition to 50 μl of assay buffer (50 mmol/liter Tris, 6% BSA, 0.01% goat IgG, 0.005% mouse IgG, 0.1% bovine IgG, 0.5 mol/liter KCl, 0.05% sodium azide, pH 7.8) and incubated for 1 h with shaking at room temperature. The plates were washed six times with the washing buffer after which biotinylated detection antibody solution (100 μl; 15 ng of anti-KLK6 (E24) monoclonal antibody in assay buffer) was added to each well and incubated for 1 h at room temperature with shaking. The plates were then washed six times with the washing buffer. Subsequently alkaline phosphatase-conjugated streptavidin solution (5 ng/well; Jackson ImmunoResearch Laboratories, Westgrove, PA) in 6% BSA buffer (in 50 mmol/liter Tris, 0.05% sodium azide, pH 7.8) was added to each well and incubated for 15 min with shaking at room temperature. The plates were washed six times with the washing buffer, and substrate buffer (100 μl; 0.1 mol/liter Tris buffer, pH 9.1) containing 1 mmol/liter substrate diflunisal phosphate, 0.1 mol/liter NaCl, and 1 mmol/liter MgCl₂ was added to each well and incubated for 10 min with shaking at room temperature. After adding 50 μl of developing solution containing Tb³⁺-EDTA complex, the fluorescence of each well was measured with a Wallac Envision 2103 multilabel reader. More details are given elsewhere (20).

Total Protein Assay—

Total protein of each ascites fraction was quantified using a Coomassie (Bradford) protein assay reagent (Pierce). 5 μl of each pooled gel filtration fraction and 5 μl of water were loaded in duplicate in a microtiter plate along with the reagent, and protein concentrations were estimated by reference to absorbance obtained for a series of bovine albumin standard protein dilutions.

Trypsin Digestion—

Each lyophilized sample was denatured using 8 m urea and reduced with DTT (final concentration, 13 mm; Sigma) at 50 °C followed by alkylation with 500 mm iodoacetamide (Sigma) with shaking at room temperature in the dark. The samples were then desalted using a NAP5 column (GE Healthcare). Samples were lyophilized and resuspended in trypsin buffer (1:50 ratio of trypsin (Promega, sequencing grade modified porcine trypsin):protein concentration; 120 μl of 50 mm ammonium bicarbonate, 100 μl of methanol, 150 μl of water) overnight in a 37 °C waterbath and then lyophilized to dryness.

Strong Cation Exchange Liquid Chromatography—

Trypsin-digested lyophilized sample was reconstituted in 120 μl of mobile phase A (0.26 m formic acid in 10% acetonitrile). The samples were directly loaded onto a PolySULFOETHYL A™ column (The Nest Group, Inc.) containing a hydrophilic, anionic polymer (poly-2-sulfoethyl aspartamide) with a pore size of 200 Å and a diameter of 5 μm. A 1-h fractionation run was performed using HPLC with an Agilent 1100 system at a flow rate of 200 μl/min. A linear gradient of mobile phase B (0.26 m formic acid in 10% acetonitrile and 1 m ammonium formate) was added as the elution buffer. The eluate was monitored at a wavelength of 280 nm. Forty fractions, 200 μl each, were collected every minute after the start of the elution gradient. These 40 fractions were pooled into eight combined fractions (each pool consisting of five fractions) and concentrated to ∼200 μl using a SpeedVac system preceding mass spectrometry analysis. Prior to each run, a protein cation exchange standard (Bio-Rad) was applied to evaluate column performance.

Mass Spectrometry—

The samples from each pooled fraction of each individual separation experiment were desalted using a ZipTip C₁₈ pipette tip (Millipore) and eluted in 4 μl of Buffer B (90% acetonitrile, 0.1% formic acid, 10% water, 0.02% trifluoroacetic acid). 80 μl of Buffer A (95% water, 0.1% formic acid, 5% acetonitrile, 0.02% trifluoroacetic acid) were added to each sample, and 40 μl were loaded on an Agilent 1100 HPLC system by the autosampler and injected onto a 2-cm C₁₈ trap column (inner diameter, 200 μm). Peptides were eluted from the trap column onto a resolving 5-cm analytical C₁₈ column (inner diameter, 75 μm) with an 8-μm tip (New Objective). This liquid chromatography setup was coupled on line to a two-dimensional linear ion trap (LTQ, Thermo Inc.) mass spectrometer using a nano-ESI source in data-dependent mode. Each fraction underwent a 120-min gradient, and eluted peptides were subjected to MS/MS. Data files were created using the Mascot Daemon (version 2.2) and extract_msn. The parameters for data file creation were: minimum mass, 300 Da; maximum mass, 4000 Da; automatic precursor charge selection; minimum peaks, 10 per MS/MS scan for acquisition; and minimum scans per group, 1.

Data Analysis—

The resulting mass spectra from each fraction were analyzed using Mascot (Matrix Science, London, UK; version 2.2) and X!Tandem (Global Proteome Machine Manager, version 2006.06.01) search engines on the non-redundant International Protein Index (IPI) human database (version 3.27) that included the forward and reversed sequences for calculating false positive error of each protein. Up to one missed cleavage was allowed, and searches were performed with fixed carbamidomethylation of cysteines and variable oxidation of methionine residues. A fragment tolerance of 0.4 Da and a parent tolerance of 3.0 Da were used for both search engines with trypsin as the digestion enzyme. The resulting files were all loaded into Scaffold (version 2.0; Proteome Software Inc., Portland, OR), which validated each MS/MS-based peptide and protein identification. Peptide identifications were accepted if they could be established at greater than 95% probability as specified by the PeptideProphet algorithm (21). Protein identifications were accepted if they could be established at greater than 95% probability and contained at least one identified peptide. Protein probabilities were assigned by the ProteinProphet algorithm (22). The DAT and XML files for each fraction were inputted into Scaffold to cross-validate Mascot and X!Tandem data files. All biological samples were searched with multidimensional protein identification technology option clicked. Sample reports were exported from Scaffold, and each protein identification was assigned a cellular localization based on information available from Swiss-Prot, Gene Ontology, Euk-mPLoc (23), and other publicly available databases.

RESULTS

Complex biological fluids such as serum and ascites fluid contain thousands of proteins with a concentration range spanning at least 9 orders of magnitude (19). The major challenge preventing in-depth analysis of these proteomes by mass spectrometry is the presence of abundant proteins such as albumin and immunoglobulins, which make up 65–97% of serum proteins. These abundant proteins limit the ionization efficiency during mass spectrometric analysis, preventing the identification of low abundance proteins. Although various techniques have been used for albumin and immunoglobulin depletion (24, 25), we chose to perform size exclusion chromatography and centrifugal ultrafiltration to fractionate ascites fluid on the basis of molecular mass. Because the top 20 most abundant serum proteins have molecular masses greater than 30 kDa, we arbitrarily chose 30 kDa as the approximate molecular mass cutoff for the identification of the ascites fluid subproteome.

Identification of Proteins by Mass Spectrometry-Gel Filtration—

20 ml of ascites fluid from one patient with disseminated ovarian cancer were used, and size exclusion chromatography was performed in duplicate using two different mobile phase buffer solutions: phosphate/sulfate and ammonium bicarbonate. After performing KLK6 ELISA and total protein assay on the eluate, fractions containing KLK6 (∼30 kDa) and lower molecular mass proteins were selected for further fractionation or trypsin digestion and mass spectrometry (Figs. 1 and 2). Although with this method the majority of albumin and immunoglobulins were removed, some early fractions still contained a significant amount of total protein (Fig. 1). Hence 10 fractions starting from the KLK6 peak were collected and refractionated with gel filtration to remove additional amounts of high abundance proteins. Four hundred and four proteins were identified with the ammonium bicarbonate buffer, and 231 proteins were identified using the phosphate/sulfate buffer system (duplicate analysis with both systems). There was a 46% overlap between the two buffer systems; when data were combined, a total of 434 unique proteins were identified (supplemental Fig. 1). Only 30 additional proteins were identified with the phosphate/sulfate buffer system.

Identification of Proteins by Mass Spectrometry-Ultrafiltration—

15 ml of ascites from two different patients with late stage ovarian cancer underwent ultrafiltration using Millipore centrifugal ultrafiltration devices with a nominal molecular mass cutoff of 50 and 100 kDa. These cutoffs were chosen based on the guidelines provided by the manufacturer regarding yields of proteins in the eluates. To obtain a good yield of filtrated proteins with molecular mass of ≤30 kDa, it was suggested to use ultrafiltration devices with a nominal molecular mass cutoff 2–4 times higher than the desired protein mass. Eighty-eight and 121 proteins were identified from the 50- and 100-kDa filtrates, respectively. There was a 45% overlap between the 50- and 100-kDa filtrates; when data were combined, a total of 144 unique proteins were identified (supplemental Fig. 2). When all data were combined (gel filtration and ultrafiltration) the number of unique proteins identified was 445 (Fig. 3).

A complete list of proteins identified along with their number of unique peptides in each experiment is presented in supplemental Table 1. Redundancies and false positive proteins were removed from the list. Supplemental Table 2 contains detailed information on all proteins identified for each experiment, including number of unique peptides identified per protein, peptide sequences, precursor ion mass, and charge states. A total of 445 unique proteins were identified from all six individual experiments; 215 more proteins were identified in the soluble ascites fraction than the previously published proteome of ascites fluid by Gortzak-Uzan et al. (17). Our false positive rate was 2.4%.

Cellular Localization of Identified Proteins—

Each unique protein identified was classified according to its cellular localization based on information available from Swiss-Prot, Gene Ontology, and other publicly available databases. Fig. 4 shows the cellular distribution of the 445 proteins with known localizations. When one protein is localized in more than one cellular compartment, all of the categories were accounted for non-exclusively. This resulted in a total percentage greater than 100%. Of the proteins, 14% were not classified as none of the sources were informative. The majority of the classified proteins were extracellular (40%) and membrane-bound (12%). One hundred and fifty-seven of the 445 identified ascites fluid proteins were also identified in the Plasma Proteome Database (supplemental Table 1). This does not mean that the remaining 288 proteins are exclusive to ascites fluid as the plasma proteome is incomplete and still requires more in-depth analysis. Our data suggest that many of the proteins identified are secreted by the tumor cells or the tumor microenvironment.

Identification of Candidate Biomarkers—

To identify biomarker candidates, we applied a set of filtering criteria to our list of unique proteins (Fig. 5). 1) We removed proteins that are not extracellular or membranous: from our list of 445 unique proteins, we eliminated 148 proteins, resulting in a shortened list of 289 extracellular and membranous proteins (supplemental Table 3). We chose to focus on extracellular and membranous proteins as these proteins have the highest potential of being found in the circulation and hence can be detected by non-invasive serum-based tests. 2) We removed all known high abundance serum proteins (concentration >5 μg/ml) such as albumin, immunoglobulins, and complement-related proteins: of the 289 extracellular and membranous proteins, 130 were classified as high abundance serum proteins and were removed, leaving a list of 159 proteins. 3) We removed proteins previously studied in the serum of ovarian cancer patients: the 159 remaining proteins were individually searched in PubMed. Twenty-five of these proteins were examined in the past as candidate ovarian cancer biomarkers (Table I). Five of those belong to the kallikrein family of biomarkers as described previously by our group (40–45). 4) We removed proteins found in only one fractionation protocol and with a single unique peptide: 43 proteins were removed with 91 proteins remaining for further selection. 5) We have previously completed analysis of the proteome and secretome (secreted and membrane-bound proteins) of four ovarian cancer cell lines (HTB75, serous; TOV112D, endometrioid; TOV21G, clear cell; and RMUG-S, mucinous) and identified a total of 1689 proteins.² One hundred and fifty-four proteins overlapped with our list of 445 ascites proteins, whereas 73 of these proteins overlapped with our list of 289 extracellular and membranous ascites fluid proteins (supplemental Table 4). We used the results of the cell line analysis to further confirm the proteins identified within our ascites samples. Of our 91 remaining proteins, 52 proteins were identified as extracellular or membranous proteins in the supernatant of at least one of the four ovarian cancer cell lines studied. These remaining 52 proteins, which passed all of our selection criteria, represent our panel of candidate ovarian cancer biomarkers (Table II).

DISCUSSION

One of the main obstacles in proteomics analysis of biological fluids is the presence of high abundance proteins (26), mainly albumin and immunoglobulins. In human serum the top 10 most abundant proteins comprise over 95% of all proteins present in this fluid (27), and the top 20 are greater than 30 kDa in molecular mass. These abundant proteins, especially human serum albumin, generate massive amounts of ions that often result in the inaccurate representation and identification of ions from the low abundance proteins because of the limited number of ions passed on for tandem mass spectrometry analysis. Hence these high abundance proteins must be depleted or removed to efficiently identify the proteins of low molecular mass and low abundance by mass spectrometry. Various methods have been used previously for the removal of albumin or immunoglobulins such as dye affinity resins or protein A/G beads (28, 29), yet these approaches are limited as albumin and other high abundance proteins often act as transport proteins by binding (and thereby concentrating) many low abundance proteins and peptides. Thus, removal of serum albumin and other abundant proteins may inadvertently remove many small proteins and peptides of interest (30).

Alternative approaches for biomarker discovery without the problems associated with high abundance molecules include analysis of tissue culture supernatants of cancer cell lines grown in serum-free media (31). Also recently Faça et al. (32) characterized the cell surface proteome and the proteins released into the extracellular milieu of three ovarian cancer cell lines and identified over 6000 proteins as candidate biomarkers and therapeutic targets.

By utilizing different separation methods in combination with mass spectrometry, we identified 445 proteins within the soluble fraction of ascites fluid. Recently the proteome of ascites fluid was reported by Gortzak-Uzan et al. (17). Although they identified over 2200 proteins, only 229 were found in the soluble fraction of ascites. Because the serum proteome contains thousands of proteins, the list of 229 proteins is unlikely to represent the full proteome of soluble ascites fluid. As serum protein concentrations range over 9 orders of magnitude, we aimed to identify a more extensive proteome of ascites fluid, focusing on low molecular mass (≤30 kDa).

Using different mobile phase systems (ammonium bicarbonate and phosphate/sulfate) and size exclusion chromatography, we identified 404 and 231 proteins, respectively. Combining these proteins with the 88 and 121 proteins identified from the 50- and 100-kDa centrifugal ultrafiltration experiments, we identified a total of 445 unique proteins within ascites fluid; this is more than any other published proteome of soluble ascites and almost doubles the proteins reported earlier (17). Our results indicate that by combining different sample fractionation methods a greater coverage of the ascites fluid proteome can be obtained, thus allowing for better biomarker selection. The overlap between our own soluble ascites fluid proteome (445 proteins) and that of Gortzak-Uzan (17) (220 proteins) was 28% (supplemental Table 3).

Although ascites fluid is the buildup of peritoneal fluid accumulated from infiltrated circulating serum, its composition may be different due to the presence of the burdening ovarian tumor. By comparing our identified ovarian ascites fluid subproteome with the human plasma proteome database, only 34% of our proteins were common. Even taking into account that the plasma proteome is not complete and that our focus was on low molecular mass proteins, our data suggest that ascites fluid has a significantly different composition than plasma, and its proteins reflect the contribution of the tumor microenvironment.

Classification of the identified proteins by Swiss-Prot, Gene Ontology as well as other publicly available databases indicated that 52% of the proteins within ascites fluid are extracellular or membranous (Fig. 4) as would be expected for an extracellular biological fluid. With over half of the proteins defined as extracellular or membranous, it is highly plausible that many of these proteins are shed by the tumor cells, allowing for an efficient selection of candidate ovarian cancer biomarkers.

The proteins identified within ascites fluid reflect the pathobiological state of ovarian cancer. Because ascites accumulation is often linked to advanced ovarian cancer, it is likely that many of these identified proteins represent promising new biomarkers. On the other hand, not all proteins in ascites represent tumor-associated antigens. By applying an arbitrary set of selection criteria, we were able to minimize the list of candidate biomarker proteins to a more manageable number (∼50) for further selection and validation. From the list of extracellular and membranous proteins, we eliminated high abundance proteins, proteins previously studied as serological biomarkers for ovarian cancer, proteins identified with a single unique peptide from only one fractionation protocol, and proteins that were not identified in at least one supernatant of four different ovarian cancer cell lines. The identification of 25 known secreted or membrane-bound ovarian cancer biomarkers (Table I) supports our view that the outlined approach can identify novel biomarkers.

From our panel of 52 candidate biomarkers (Table II), 31 proteins were also identified within the ascites fluid proteome by Gortzak-Uzan et al. (17) (supplemental Table 5). However, these authors did not select any of these proteins for further investigation as they had applied a different set of criteria for biomarker selection. This underlines the fact that despite successful identification of proteins in fluids by mass spectrometry the criteria for narrowing down the list of candidates are also of paramount importance.

Although many of our filtering criteria were somewhat subjective, our discovery strategy appears to be efficient as 25 of our 289 identified extracellular or membranous proteins were previously studied as potential serum ovarian cancer biomarkers (Table I). It is very likely that our list of 52 candidates also includes novel ovarian cancer biomarkers. Although the most widely studied biomarker for ovarian cancer, CA125, was not identified in any of our experiments, this can be explained by the fact that CA125 is highly glycosylated with a molecular mass ranging from 190 to 2700 kDa (33) and was therefore excluded during sample preparation. We acknowledge that other glycosylated proteins of molecular mass of ≥30 kDa may have not been identified. Additionally many of our candidates (Table II) have molecular masses ≥30 kDa, implying that they are likely fragmented in ascites fluid. This has also been reported by others who observed truncated forms of transthyretin and cleavage fragment of inter-α-trypsin inhibitor heavy chain H4 (35).

The major challenge in biomarker discovery using proteomics is the validation phase. In the future, we intend to validate some, or all, of our 52 candidate ovarian cancer biomarkers by using ELISAs or other quantitative techniques and serum as the fluid of choice. Such analysis is currently problematic because of the lack of immunological reagents or assays that have the capability of measuring low levels of these antigens in serum.