If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Correspondence to: Dohyun Han, PhD, Proteomics Core Facility, Biomedical Research Institute, Seoul National University Hospital, 71 Daehak-ro, Jongno-gu, Seoul 03080, Republic of Korea. Tel: 82-2-2072-1719 Fax: 82-2-2072-4406,
Proteomics Core Facility, Biomedical Research Institute, Seoul National University Hospital, Seoul 03082, Republic of KoreaTransdisciplinary Department of Medicine and Advanced Technology, Seoul National University Hospital, Seoul 03080, Republic of Korea
Correspondence to: Maria Lee, MD, PhD, Department of Obstetrics and Gynecology, Seoul National University College of Medicine, Seoul National University Hospital, 101 Daehak-ro, Jongno-gu, Seoul 03080, Republic of Korea. Tel: 82-2-2072-2842, Fax: 82-2-762-3599,
Department of Obstetrics and Gynecology, Seoul National University College of Medicine, Seoul 03080, Republic of KoreaDepartment of Obstetrics and Gynecology, Seoul National University Hospital, Seoul 03080, Republic of Korea
We aimed to investigate novel, blood-based prognostic biomarkers in HGSOC.
MS-based label-free quantification was conducted using fresh-frozen plasma samples.
Candidate biomarkers were validated with an independent set of plasma samples via ELISA.
Plasma GSN was identified as an independent poor prognostic biomarker for progression-free survival.
We successfully developed models and nomograms predicting the 18-month PFS rate for clinical use.
Ovarian cancer is one of the most lethal female cancers. For accurate prognosis prediction, this study aimed to investigate novel, blood-based prognostic biomarkers for high-grade serous ovarian carcinoma (HGSOC) using mass spectrometry-based proteomics methods. We conducted label-free liquid chromatography–tandem mass spectrometry using frozen plasma samples (n=20) obtained from patients with newly diagnosed HGSOC. Based on progression-free survival (PFS), the samples were divided into two groups: good (PFS ≥18 months) and poor prognosis groups (PFS <18 months). Proteomic profiles were compared between the two groups. Referring to proteomics data that we previously obtained using frozen cancer tissues from chemotherapy-naïve HGSOC patients, overlapping protein biomarkers were selected as candidate biomarkers. Biomarkers were validated using an independent set of HGSOC plasma samples (n=202) via enzyme-linked immunosorbent assay (ELISA). To construct models predicting the 18-month PFS rate, we performed stepwise selection based on the area under the receiver operating characteristic curve (AUC) with five-fold cross-validation. Analysis of differentially expressed proteins in plasma samples revealed that 35 and 61 proteins were upregulated in the good and poor prognosis groups, respectively. Through hierarchical clustering and bioinformatic analyses, GSN, VCAN, SND1, SIGLEC14, CD163, and PRMT1 were selected as candidate biomarkers and were subjected to ELISA. In multivariate analysis, plasma GSN was identified as an independent poor prognostic biomarker for PFS (adjusted hazard ratio, 1.556; 95% confidence interval, 1.073–2.256; P=0.020). By combining clinical factors and ELISA results, we constructed several models to predict the 18-month PFS rate. A model consisting of four predictors (FIGO stage, residual tumor after surgery, and plasma levels of GSN and VCAN) showed the best predictive performance (mean validated AUC, 0.779). The newly developed model was converted to a nomogram for clinical use. Our study results provided insights into protein biomarkers, which might offer clues for developing therapeutic targets.
). The absence of disease-specific early symptoms and effective screening methods leads to ovarian cancer being diagnosed at an advanced stage and having high recurrence and mortality rates despite treatment, consisting of extensive cytoreductive surgery followed by taxane- and platinum-based chemotherapy (
). Meanwhile, ovarian cancer is not a single disease, but a heterogeneous disease comprising various histologic subtypes with different carcinogenic routes and clinical features. Among the subtypes of ovarian cancer, high-grade serous ovarian carcinoma (HGSOC) is the most common and responds very well to chemotherapy; however, it frequently relapses, with acquisition of chemoresistance (
), the management of HGSOC rapidly evolved. Maintenance therapy with poly-(ADP-ribose) polymerase (PARP) inhibitors, such as olaparib and niraparib, was incorporated into the primary treatment of HGSOC based on landmark phase III randomized controlled trials (RCTs) (
). After a complete or partial response to first-line platinum-based chemotherapy, olaparib maintenance therapy can be offered to patients with BRCA1/2 mutated, advanced HGSOC to improve survival outcomes, while niraparib maintenance therapy confers survival benefits in advanced HGSOC, regardless of BRCA1/2 mutational status or homologous recombination deficiency (HRD). Accurate prediction of prognosis is necessary to facilitate molecular profiling-based HGSOC treatment.
In this regard, our research team has focused on discovering prognostic protein biomarkers in HGSOC using mass spectrometry (MS) proteomics. This emerging technology allows high-throughput and individualized characterization and quantification of proteins in biospecimens (
). Previously, we identified six protein biomarkers associated with progression-free survival (PFS) through the label-free quantitative proteomic analysis of frozen primary HGSOC tissues and validated them using immunohistochemical (IHC) staining in an independent sample set (
). Thus, we aimed to investigate whether we could identify novel, prognostic protein biomarkers for HGSOC from blood samples using MS-based proteomics. Biomarker candidates were validated using an enzyme-linked immunosorbent assay (ELISA) in an independent dataset. We also developed models to predict 18-month PFS rates in HGSOC patients.
1 Ethics statement
This study was approved by the Institutional Review Board of Seoul National University Hospital (SNUH; No. H-2010-152-1167) and was conducted in accordance with the Declaration of Helsinki. At our institution, we routinely asked patients with newly diagnosed ovarian cancer who were scheduled to undergo primary treatment to donate their biospecimens (e.g., blood, urine, and cancer tissues) for research purposes with written informed consent since June 2012.
2 Sample collection
In this study, we used plasma samples from HGSOC patients that were obtained one day before primary debulking surgery (PDS) or initiation of neoadjuvant chemotherapy (NAC) and stored at the SNUH Hunan Biobank. The process for the collection of plasma from whole blood was as follows: Collect 6 mL of blood sample into the EDTA tube, and centrifuge for 10 minutes at 1,551 g at 4°C. After centrifugation, carefully collect the plasma layer with a transfer pipette without disturbing the buffy coat layer. Pipette the 700 μL of plasma into a 1.8 mL labeled cryovial, up to 4 vials. Place all aliquots upright in a labeled rack in a -196°C LN2 tank. All the plasma samples used in this study had never been thawed before.
) patients whose disease relapsed within 18 months after PDS, that is, PFS <18 months (poor prognosis group) or those whose disease did not relapse for at least 18 months after PDS, that is, ≥18 months of PFS (good prognosis group). Twenty patients from the two groups (10 in each group) were selected for further proteomic analyses. The order of sample preparation was randomized and independent of the patient list. The proteomic profiles of the two groups were compared.
In the second phase (validation), we retrieved pre-treatment frozen plasma of patients who met the following conditions: (
) completed primary treatment, consisting of either PDS or NAC-IDS, followed by postoperative taxane- and platinum-based adjuvant chemotherapy. We excluded patients if they had enrolled in clinical trials for primary treatment; did not provide written informed consent; or were lost to follow-up during primary treatment or within 18 months after initiation of primary treatment, without relapse or disease progression. A total of 202 consecutive patients with HGSOC were included in this phase, and the sample size was adequate for multivariate survival analysis and further dvelopment of predictive models. The order of sample preparation was also randomized and independent of the patient list. ELISA was conducted with technical triplicates on pooling samples for the standard curve and batch control.
In the medical record review, we collected patients’ clinicopathologic data. Disease progression was ascertained based on computed tomography (CT) scans by applying the Response Evaluation Criteria in Solid Tumors version 1.1 (
). PFS and overall survival (OS) were defined as the time intervals from the date of initial diagnosis to the date of disease progression and to the date of cancer-related death or last follow-up, respectively.
4. Proteomic and bioinformatic analyses
The overall workflow of proteomic and bioinformatic analyses are depicted in Figure 1A.
4.1 Sample preparation
Protein digestion was performed using 2 μL of each plasma sample as previously described, with some modifications (
). Briefly, 23 μL of protein digestion buffer, including reduction and alkylation reagents, was added to 2 μL plasma samples in 96 well-plates. The mixture was boiled for 25 min at 60°C to denature and alkylate the proteins. After cooling samples to room temperature, protein digestion was performed at 37°C overnight using a trypsin/LysC mixture (Promega, Madison, WI, USA) at a 100:1 protein-to-protease ratio. The second digestion was performed at 37°C for 2 h using trypsin (enzyme-to-substrate ratio [w/w], 1:1000). All resulting peptides were acidified with 10% trifluoroacetic acid (TFA). The acidified peptides were loaded onto custom-made styrene divinylbenzene reversed-phase sulfonate-StageTips according to previously described procedures (
). The StageTip was washed three times with 100 μL 0.2% TFA. Three fractionations were performed using elution buffers with a step gradient of increasing acetonitrile (40%, 60%, and 80%) in 1% ammonium hydroxide. 1, 2, and 3. All the eluted peptides were dried using a SpeedVac centrifuge (Thermo Fisher Scientific, Waltham, MA, USA).
4.2 LC-MS/MS analysis
All liquid chromatography with tandem MS (LC-MS/MS) analyses were conducted using an Ultimate 3000 UHPLC system (Dionex, Sunnyvale, CA, USA) coupled with a Q-Exactive HF-X mass spectrometer (Thermo Fisher Scientific), as previously described, with some modifications (
). Peptides were separated on a two-column system equipped with a trap column (Thermo Fisher Scientific, Acclaim PepMap, C18 5 μm, 100 Å, 300 μm I.D. × 5 mm) and an analytical column (Thermo Fisher Scientific, EASY-Spray column, C18 1.9 μm, 100 Å, 75 μm I.D. × 50 cm) using 90-min gradients from 7% to 30% ACN at a flow rate of 300 nl/min. Column temperature was maintained at 60°C using a column heater. MaxQuant.Live version 1.2 was used for BoxCar acquisition (
). The MS1 resolution was set to 120,000 at m/z 200 for BoxCar, and the acquisition cycle comprised two BoxCar scans at 12 boxes (scaled width, 1 Th overlap) with a maximum ion injection time of 20.8 per box, with the individual AGC target set to 250,000. MS/MS spectra were acquired at a higher-energy collisional dissociation-normalized collision energy of 30, with a resolution of 17,500 at m/z 200. The maximum ion injection durations for the full and MS/MS scans were 20 ms and 100 ms, respectively.
4.3 Data processing
All raw MS files were processed using MaxQuant (version 188.8.131.52) (
). Primary searches were performed using 6 ppm precursor ion tolerance for total protein-level analysis. MS/MS ion tolerance was set at 20 ppm. Cysteine carbamidomethylation was used as a fixed modification. Protein N-acetylation and methionine oxidation are considered variable modifications. Enzyme specificity was set to full tryptic digestion. Peptides with a minimum length of six amino acids and up to two missed cleavages were considered. The required false discovery rate (FDR) was set to 1% at peptide, protein, and modification levels. To maximize the number of quantification events across samples, we enabled the “Match between Runs’ options on the MaxQuant platform. The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (
). First, we eliminated proteins identified as “reverse” and “only identified by site.” After filtering values of at least 70% in each group, missing values were imputed by random numbers drawn from a normal distribution with a width of 0.3 and a down-shift of 1.8. Finally, data were normalized using a width-adjustment function that subtracts the medians and scales all values in a sample to yield equal interquartile ranges (IQRs) (
). For pairwise proteome comparisons, we performed a two-sided t-test with a significance level (P value) of <0.05, and a fold-change of >1.5. Support vector machine analysis was performed using the R/Bioconductor package “GNC” (
). EnrichR uses the Fisher exact test to calculate P values. Statistical significance was set at P value <0.05, and GO analysis was used to identify significant GOBP terms.
5 Enzyme-linked immunosorbent assay
ELISA kits for human proteins were used to quantify the plasma levels of endogenous proteins, according to the manufacturer’s instructions. ELISA kits for gelsolin (GSN; abx253831), versican (VCAN; abx153474), staphylococcal nuclease, tudor domain containing 1 (SND1; abx383338), sialic acid-binding Ig-like lectin 14 (SIGLEC14; abx545882), and protein arginine methyltransferase 1 (PRMT1; abx258982) were purchased from Abbexa (Cambridge, UK), whereas a kit for CD163 (DC1630) was purchased from R&D Systems (Minneapolis, MN, USA).
After determining the optimal dilution factor for each protein, the concentrations of GSN, VCAN, SND1, CD163, SIGLEC14, and PRMT1 were measured and quantified in the pre-treatment and frozen plasma samples (n=202). Optical density at 450 nm was measured using a SPARK multimode microplate reader (Tecan Systems, Inc., San Jose, CA, USA).
6 Model construction
We constructed regression-based models to predict 18-month PFS rates using clinical variables and the ELISA results for protein biomarkers in patients with HGSOC (n=202). The 18-month PFS rate was defined by binarizing the PFS for 18 months. Each of the six identified protein biomarkers was binarized based on the optimal cutoff obtained from maximally selected log-rank statistics (maxstat) (
). To select important predictors for the 18-month PFS rate, stepwise selection was performed based on the area under the receiver operating characteristic curve (AUC). During stepwise selection, predictors contributing to AUC improvement were selected in a stepwise fashion (
). From variable selection to model evaluation, five-fold cross-validation was used, considering the two-class proportions of the 18-month PFS rate. The AUC, sensitivity, and specificity were used as evaluation measures. The optimal cutoff for calculating sensitivity and specificity was determined as a value corresponding to the maximum value of balanced accuracy, defined as the average of the sensitivity and specificity. Based on the logistic regression model including the selected predictors, we developed a nomogram for clinical use.
R statistical software (version 4.0.3; R Foundation for Statistical Computing, Vienna, Austria) was used to construct predictive models and plot nomograms.
7 Statistical analysis
Clinicopathologic characteristics were compared between the good and poor prognosis groups by using Student’s t and Mann–Whitney U tests for continuous variables and Pearson’s chi-squared and Fisher’s exact tests for categorical variables. The Pearson’s correlation coefﬁcient test was used to measure the relationship between continuous variables. For survival analysis, we used the Kaplan–Meier method with the log-rank test. In the multivariate analysis, a Cox proportional hazards model was constructed, and adjusted hazard ratios (aHRs) and 95% confidence intervals (CIs) were calculated.
Statistical analyses were performed using SPSS Statistics (version 25.0; IBM Corp., Armonk, NY, USA) and GraphPad Prism 5 (GraphPad Inc., La Jolla, CA, USA). All statistical tests were two-sided, and a P value <0.05 was considered statistically significant.
1 Characteristics of patients in the discovery phase
The clinicopathologic characteristics of 20 patients with HGSOC for whom proteomic analysis was performed are presented in Supplementary Table 1. The mean patient age was 54.9 years, which was similar between the good and poor prognosis groups (P=0.609). Between the two groups, there was no differences in parity, menopausal status, initial serum CA-125 levels, International Federation of Gynecology and Obstetrics (FIGO) stage, residual tumor after PDS, and total number of cycles of post-operative adjuvant chemotherapy (Supplementary Table 2). In relation to germline BRCA mutational status, seven and two patients had BRCA1 and BRCA2 mutations, respectively, while the other 11 patients harbored wild-type BRCA1/2. None of the patients received first-line PARP inhibitor maintenance therapy. The median length of observations was 34.0 months, during which 15 patients experienced disease recurrence. Patients in the good prognosis group had a significantly better PFS than those in the poor prognosis group (median, 48.4 vs. 12.4 months; P<0.001).
2. Results of proteomic and bioinformatic analyses
2.1 Global proteomic analysis of plasma samples
To identify prognostic biomarkers for HGSOC, we performed MS-based label-free quantification using frozen plasma samples from chemotherapy-naïve patients (n=20). To increase the proteome depth, we applied BoxCar acquisition using a small amount (2 μL) of plasma sample, without depletion of highly abundant proteins. In total, 1912 proteins were identified at the protein FDR 1% level. An average of 1082 protein groups were quantified per sample (Figure 1B). Signal intensities for the quantified proteins overall spanned approximately seven orders of magnitude (Figure 1C), and included several previously reported ovarian cancer marker candidates, such as HE4, MSLN, VCAM-1, CEA, CRP, PROZ, LCAT, and M-CSF. Details of the identified and quantified proteins are presented in Supplementary Table 3.
To identify the differences within and between groups, the protein profiles were plotted as multi-scatter plots. Pearson correlation coefficient (PCC) values for proteome pairs were calculated (Supplementary Figure 2). The intra-group correlation displayed average PCCs of 0.84 and 0.83 in the good and poor response groups, respectively. The average intergroup PCC value, between the good and poor response group, was 0.82.
2.2 Label-free quantification
Next, we assessed significant quantitative differences between samples from patients with good and poor prognosis, based on pairwise comparisons. First, we compared the good and poor prognosis groups via principal component analysis of a filtered list with approximately 1028 proteins (with 70% valid iBAQ values in at least one group). Although tumor proteomes were correlated regardless of prognosis (Supplementary Figure 2), the two good and poor response groups were separated independently (Figure 2A).
Pairwise comparisons via t-test and filtering (P<0.05; fold-change, >1.5) revealed significant alterations in 96 proteins, of which 35 proteins had higher expression in the good prognosis group than the poor prognosis group. The other 61 proteins had higher expression in the poor prognosis group than the good prognosis group (Figure 2B and Supplementary Table 4). Using the stringent filtering criterion of FDR <0.05, PLXND1, SIGLEC14, SND1, and PRMT1 were found to be upregulated in patients with a poor prognosis. GO enrichment analysis based on biological processes revealed that proteins upregulated in the good prognosis group were significantly enriched for terms such as “actin filament organization,” “regulation of lipase activity,” “cellular response to chemical stress,” “glucose 6-phosphate metabolic process,” and “regulation of cell death” (Figure 2C and Supplementary Table 5). In contrast, proteins upregulated in the poor prognosis group were significantly enriched in “neutrophil degranulation,” “neutrophil-mediated immunity,” “aspartate metabolic process,” and “negative regulation of lipoprotein particle clearance” GO-BPs (Figure 2D and Supplementary Table 5).
2.3 Selection of candidate prognostic biomarkers
Potential plasma biomarker candidates for sequential validation experiments were first selected among differentially expression proteins (DEPs) that met one or more of the following criteria: (
). Consequently we first selected 18 potential biomarkers (GSN, VCAN, SND1, SIGLEC14, CD163, PRMT1, PLXND1, F12, HPR, HSPA5, ACY1, CD248, C5, GRHPR, MCAM, PPP1R7, STAB1, and UGGT1). Among the 14 proteins that overlapped with our previous tissue data, six proteins (GSN, VCAN, CD163, F12, HPR, and HSPA5) were selected according to concordant expression patterns between tissue and plasma. We further selected prognostic biomarker candidates on the basis of the following parameters: (
) proteins could be detected in our validation cohort using the selected ELISA kits. Finally, GSN, VCAN, SND1, SIGLEC14, CD163, and PRMT1 were selected as candidate biomarkers for the validation stage (Supplementary Figure 3).
2.4 Validation of protein biomarkers through ELISA
Protein biomarkers underwent prognostic validation by using independent plasma samples obtained from patients with HGSOC (n=202). Clinicopathologic characteristics of the patients are presented in Table 1. Of all patients, 88.6% had advanced-stage (FIGO stage III–IV) disease and 92.1% underwent PDS, rather than NAC followed by IDS. Optimal debulking (with no gross residual tumor) was achieved in 71.8% of cases. Germline and/or somatic BRCA1/2 testing was conducted in 158 patients (78.2%), and 36.1% (57/158) had mutations in BRCA1 or BRCA2. Three patients received first-line PARP inhibitor maintenance therapy (olaparib). The median length of observation was 43.8 months, during which 134 patients (66.3%) experienced relapse and 30 (14.9%) died of the disease. The median PFS was 24.6 months, and the 18-month PFS rate was 62.9% (127/202) (Supplementary Figure 4).
Table 1Clinicopathologic characteristics of the patients in validation phase
In addition to PSR, the patients who completed taxane- and platinum-based chemotherapy and did not experience disease recurrence during at least six months of follow-up period were considered platinum-sensitive.
Abbreviations: BEV, bevacizumab; CA-125, cancer antigen 125; FIGO, International Federation of Gynecology and Obstetrics; NAC, neoadjuvant chemotherapy; IDS, interval debulking surgery; PDS, primary debulking surgery; PRR, platinum-resistant recurrence; PSR, platinum-sensitive recurrence; SD, standard deviation.
a PSR was defined as relapse ≥6 months after completion of taxane- and platinum-based chemotherapy, whereas PRR as relapse <6 months.
b In addition to PSR, the patients who completed taxane- and platinum-based chemotherapy and did not experience disease recurrence during at least six months of follow-up period were considered platinum-sensitive.
c Germline and/or somatic BRCA1/2 mutational status.
Table 1 also compares clinicopathologic characteristics between the good and poor prognosis groups. Patients in the poor prognosis group (n=75) were significantly older (P=0.029) and had more advanced disease (P=0.001), compared to those in the good prognosis group (n=127). While the two groups had similar proportion of PDS (P=0.267), optimal debulking was less frequently achieved in the poor prognosis group (52.0% vs. 83.5%; P<0.001). Among the patients who received germline and/or somatic BRCA1/2 testing, BRCA1/2 mutations were less frequently observed in the poor prognosis group (21.4% vs. 44.1%; P=0.004). Comparing the survival outcomes, the poor prognosis group showed worse PFS (median, 12.5 vs. 54.1 months; P<0.001) and OS (5-year OS rate, 57.1% vs. 94.0%; P<0.001), compared to the good prognosis group (Supplementary Figure 4).
Six protein biomarkers, GSN, VCAN, SND1, SIGLEC14, CD163, and PRMT1, were subjected to further prognostic validation using ELISA (Supplementary Table 6). The ELISA results are summarized in Table 2 and Supplementary Figure 5. Table 2 also compares ELISA results between the good and poor prognosis groups. Plasma GSN levels were significantly higher in the poor prognosis group than those in the good prognosis group (median, 23.150 vs. 19.300 ng/mL; P=0.001). However, plasma levels of VCAN, SND1, SIGLEC14, CD163, and PRMT1 were similar between the two groups.
Table 2ELISA results of the six plasma protein biomarkers
No correlation was observed between serum CA-125 levels and the plasma levels of each protein biomarker (Supplementary Table 7). Plasma GSN levels were significantly correlated with plasma VCAN (Pearson’s correlation coefficient r=0.224; P=0.001), SND1 (r=0.177; P=0.012), and CD163 levels (r=0.351; P<0.001), but the correlations were weak. A weak positive correlation was also observed between plasma VCAN and SND1 levels (r=0.167; P=0.017). Plasma VCAN levels were significantly moderately correlated with plasma SIGLEC14 levels (r=0.501; P<0.001), and weakly correlated with plasma CD163 levels (r=0.341; P<0.001). Using the cut-off values determined by maxstat (
), the validation set was divided into high (≥ cut-off value) and low (< cut-off value) plasma level groups for each protein.
We then compared the clinicopathologic characteristics of the patients with high and low plasma levels of the six protein biomarkers (Supplementary Table 8). Patients with high GSN levels (n=62) were significantly older (P=0.001), had higher initial serum CA-125 levels (P=0.043), more advanced disease (P=0.012), less commonly achieved optimal debulking (P=0.011), and more commonly showed platinum resistance (P=0.040) than did those with low GSN levels (n=140). For VCAN, high plasma levels were associated with old age at the initial diagnosis (P<0.001). For SND1, high plasma levels were associated with advanced disease (P=0.032) and suboptimal debulking (P=0.027). However, for SIGLEC14, CD163, and PRMT1, no significant differences in patient age, FIGO stage, or residual tumor after surgery were observed between the high and low expression groups.
In assessing the platinum sensitivity of patients with respect to the plasma levels of each protein biomarker, we observed a significant difference only for GSN. Patients with high GSN levels were less sensitive to platinum-based chemotherapy than those with low GSN levels (77.4% vs. 88.6%; P=0.040).
In survival analysis, the high GSN group showed significantly worse PFS than did the low GSN group (median, 15.6 vs. 29.4 months; P=0.001). In contrast, the high VCAN group showed significantly better PFS than did the low VCAN group (median, not reached vs. 23.2 months; P=0.042). PFS was also better in the high than in the low SND1 group, but the difference was not statistically significant (median, 40.2 vs. 22.6 months; P=0.066). No differences in PFS were observed between groups with high and with low plasma levels of SIGLEC14, CD163, and PRMT1 (Figure 3).
In the multivariate analysis adjusted for patient age, FIGO stage, and residual tumor after surgery, a high plasma GSN level was identified as an independent poor prognostic biomarker for PFS (aHR, 1.556; 95% CI, 1.073–2.256; P=0.020). However, subsequent multivariate analyses revealed no influence of VCAN (aHR, 0.617; 95% CI 0.370–1.030; P=0.065) and SND1 (aHR, 0.789; 95% CI 0.454–1.372; P=0.401) plasma levels on PFS (Table 3).
Table 3Factors associated with progression-free survival.
Residual tumor after surgery
Abbreviations: aHR, adjusted hazard ratio; CI, confidence interval; CA-125, cancer antigen 125; FIGO, International Federation of Gynecology and Obstetrics; HR, hazard ratio.
3 Development of models predicting 18-month PFS rate
Next, we constructed regression-based models predicting the 18-month PFS rate using clinical variables and plasma levels of five plasma protein biomarkers in HGSOC patients (n=202). Herein, SND1 was excluded as the high SND1 group showed better PFS than did the low SND1 group in the validation phase, which was contrary to the results in the development phase. Through stepwise selection methods, four predictors were selected: FIGO stage, residual tumor after surgery, GSN, and VCAN. Various models were developed using these predictors. Each predictive model underwent five-fold cross-validation to compute the AUC. Among them, the model using cutoff plasma values for GSN (24.350 ng/mL) and VCAN (5.832 ng/mL) showed the best predictive performance, with an AUC of 0.779 (Figure 5 and Supplementary Table 9). This model also showed better predictive performance than did the model using continuous values for plasma GSN and VCAN levels, and those replacing the two protein biomarkers, GSN and VCAN, with serum CA-125 levels (Supplementary Table 9).
Using regression-based models, nomograms were then developed for clinical use. Finally, we fitted a user-friendly interface onto the developed nomograms and posted them on a website (http://asiansgo.org/software/nomogram_ovarian).
Our proteomic analysis study identified plasma protein biomarkers that might be associated with the prognosis of HGSOC. In validation with ELISA, high plasma levels of GSN were associated with worse PFS, while VCAN, SND1, SIGLEC14, CD163, and PRMT1 did not affect the survival outcomes of patients with HGSOC. We also developed models and nomograms to predict the 18-month PFS rate for clinical purposes.
GSN, a calcium-dependent multi-functional actin-binding protein, has cytoplasmic and plasma isoforms, which are encoded by the same gene (
). Plasma GSN is a well-known poor prognostic biomarker for PFS and OS in patients with ovarian cancer. In addition, the expression and secretion of GSN were higher in chemoresistant ovarian cancer cells than in chemosensitive ovarian cancer cells (
). Consistently, the current study showed that high plasma GSN levels were associated with poor prognostic factors, such as advanced-stage disease and residual tumor after surgery, loss of platinum sensitivity, and reduced PFS. Recently, Asare–Werehene et al. demonstrated that plasma GSN confers chemoresistance in ovarian cancer by inhibiting the antitumor functions of macrophages through apoptosis and modulating the tumor microenvironment (
). In advanced-stage serous ovarian cancer, Ghosh et al. reported that high VCAN expression in the tumor stroma was associated with increased angiogenesis and significantly worse PFS and OS than low VCAN expression (
). However, such an association seems to differ depending on the specimen type. In contrast to this study, we measured plasma VCAN levels instead of tissue expression and observed that VCAN did not affect PFS in patients with HGSOC.
SND1, a component of the RNA-induced silencing complex, is an oncogene involved in tumorigenesis, tumor progression, and metastasis in multiple malignancies, including breast cancer (
). In contrast, our study showed that high or low plasma SND1 levels did not affect PFS in HGSOC patients. Such inconsistent results between our study and previous studies might originate from differences in specimen types, histological subtypes, and sample sizes. To the best of our knowledge, no previous study has investigated the relationship between plasma SND1 levels and survival outcomes in ovarian cancer. Therefore, further prospective studies are warranted to investigate the relationship between plasma SND1 levels and survival outcomes.
CD163, a multifunctional receptor containing a scavenger receptor cysteine-rich domain, is specifically expressed in monocytes and macrophages and can be cleaved from the cell membrane of monocytes and macrophages (
). In contrast, no reduction in PFS due to high plasma CD163 levels was observed in our study. While a previous study examined serum samples of patients with all histological subtypes and grades of epithelial ovarian cancer, the current study examined plasma samples of patients with HGSOC. Such differences may underlie the inconsistent results.
PRMT1 mediates epigenetic modifications. Aberrant expression of PRMT1 has been reported to be involved in tumorigenesis (
). They found that high PRMT1 expression was associated with platinum resistance and reduced OS. In contrast, we could not identify any association between plasma PRMT1 levels and response to platinum-based chemotherapy or PFS.
SIGLEC family proteins play diverse immune and non-immune regulatory roles in the tumor microenvironment and participate in tumor progression. Facilitating tumor immune escape is one of the mechanisms by which tumors progress (
). Compared with other SGILEC family proteins, the prognostic role of SIGLEC14 in ovarian cancer is not fully understood. We observed no association between plasma SIGLEC14 levels and PFS in HGSOC patients.
In the current study, we developed two regression-based models and nomograms to predict the 18-month PFS rate in patients with newly diagnosed HGSOC. In both models, only two (GSN and VCAN) of the six protein biomarkers were selected and incorporated. Although independent multivariate analyses indicated GSN as the solitary independent prognostic factor for PFS, the addition of VCAN to GSN seems to confer further improvement in performance in predicting the 18-month PFS rate. In the validation phase, we could not conduct external validation due to the scarcity of resources and time to conduct a prospective multicenter study that could collect plasma samples from the enrolled subjects. Instead, we implemented five-fold cross-validation, a well-established statistical method, to prevent overfitting and increase the robustness and prediction accuracy of the developed model. A further increase in the predictive performance is expected if the multi-omics data of HGSOC patients are integrated into the developed models.
Throughout the study, patients’ BRCA1/2 mutational status and the use of first-line PARP inhibitor maintenance treatment were not considered, because of their low frequency in our study. In particular, only a few patients were eligible (n=3 in the second phase). Such a low frequency might have originated from the socio-medical environment in Korea. In October 2019 and August 2020, the Korean Ministry of Food and Drug Safety approved olaparib and niraparib as first-line maintenance therapies based on the SOLO1 (
), respectively. Furthermore, it was not until October 2021 that the National Health Insurance System started to cover both olaparib and niraparib in patients with BRCA1/2 mutated HGSOC. Despite the approval of PARP inhibitors, HGSOC patients find these difficult to use beyond insurance coverage because of their high cost.
In the era of precision cancer medicine, it is critical to predict prognosis or survival outcomes precisely. Our results indicated that adding plasma levels of GSN and VCAN to the clinical factors in predictive models improved the models’ performance. Applying the developed models, if a patient with HGSOC was predicted to be at high risk of disease progression within 18 months from the initiation of primary treatment, physicians might consider incorporating bevacizumab into conventional taxane- and platinum-based chemotherapy. In particular, based on the germline/somatic BRCA1/2 mutational status and homologous recombination deficiency, first-line PARP inhibitor maintenance therapy may be recommended more strongly (
). After completion of chemotherapy, a high-risk patient might undergo more intensive surveillance.
Our study had several limitations. First, as this study had a retrospective design, inevitable issues, such as selection bias, might exist. Second, the sample size might have been insufficient for discovering and validating plasma protein biomarkers. In particular, in the validation phase, we failed to observe a relationship of high plasma SND1 levels with poor prognosis, which was marked in the development phase. Third, external validation of the developed models is needed. Fourth, we only investigated statistical correlations, but did not evaluate the biological interactions between protein biomarkers. Lastly, we did not investigate longitudinal changes in each plasma protein biomarker over the course of the primary treatment. Such information might enable us to calculate the kinetics of each biomarker during a specific period and predict the primary treatment success more accurately.
In conclusion, we successfully generated proteomic profiles of plasma samples from HGSOC patients. A subsequent ELISA study assessed the prognostic value of the six protein biomarkers. Plasma GSN was identified as a poor prognostic biomarker for PFS in HGSOC, but plasma VCAN, SND1, SIGLEC14, CD163, and PRMT1 levels were not. Combined with clinical factors, we developed models and nomograms to predict the 18-month PFS rate for clinical purposes. Our study results provided insights into the protein biomarkers that might potentially develop HGSOC and offered clues for developing therapeutic targets. Further translational and prospective validation studies are needed.
DH, and ML conceived the project. SIK, SH, KD, DH, and ML provided methodology. DH and ML provided resources. DH and ML provided software. SIK, SH, KD, and DH conducted formal analysis. SIK, HHC, and ML performed funding acquisition. HSK, HHC, JWK, NHP, and YSS performed validations. SIK, SH, and DH wrote the manuscript, and all authors contributed to the final version.
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentralproteomexchange.org) via the PRIDE partner repository with the dataset identifier PXD034636.
CONFLICT OF INTEREST
The authors declare that they have no conflict of interest
The biospecimens and data used in this study were provided by the Biobank of Seoul National University Hospital, a member of Korea Biobank Network. This work was supported by the Korea Medical Device Development Fund grant funded by the Korean government: the Ministry of Trade, Industry and Energy (No. RS-2020-KD000028). This work was also supported by the National Research Foundation (NRF) of Korea (No. NRF-2020R1G1A1005711) and the Cooperative Research Program of Basic Medical Science and Clinical Science from the Seoul National University College of Medicine (No. 800-20210297).
To investigate novel, blood-based prognostic biomarkers, we conducted label-free liquid chromatography–tandem mass spectrometry using fresh-frozen plasma samples obtained from patients with newly diagnosed high-grade serous ovarian carcinoma (HGSOC). Six plasma proteins were selected as candidate biomarkers and underwent validation with an independent set of HGSOC plasma samples via enzyme-linked immunosorbent assay (ELISA). Plasma GSN was identified as an independent poor prognostic biomarker for progression-free survival. By combining clinical factors and ELISA results, we successfully developed models and nomograms to predict the 18-month PFS rate for clinical use.
Se Ik Kim: Methodology, Formal analysis, Funding acquisition, Writing - Original Draft, Writing - Review & Editing