Diagnostic plasma miRNA-profiles for ovarian cancer in patients with pelvic mass.

Ovarian cancer (OC) is the major cause of cancer death among women globally [ 1 ]. With approximately 324 new OC cases and 124 borderline tumours every year, Denmark has one of the highest incidence rates of OC in the world, with an estimate of 14 cases per 100,000 in the recent years [ 2 , 3 ]. Due to its asymptomatic features, OC is typically diagnosed at late stage (FIGO III-IV), in 65% of all cases, impacting on a poor 5-year overall survival of less than 50% [ 2 ]. Surgical debulking and precise surgical staging are essential for good OC management, where an optimal preoperative differentiation between OC and benign pelvic masses is critical to improve survival [ 4 , 5 ]. However, those procedures are reserved to specialised tertiary gynaecologic oncology centres. Moreover, due to the high unspecificity of the current diagnostic markers, only about 20–25% of patients with benign pelvic mass will be diagnosed with OC [ 5 , 6 ], underscoring the lack of reliable tests to secure the diagnosis. Therefore, the relative high prevalence, late diagnosis and consequent poor prognostics in OC are still a critical clinical challenge with unmet needs. For the past two decades micro-RNAs (miRNAs) have emerged as promising diagnostic and prognostic biomarkers. MiRNAs are small non-coding RNAs, ranging between 19-25nt which can regulate gene expression by binding to their complementary target messenger RNA (mRNA) [ 7 ]. Depending on the binding region of the miRNA to their target mRNA, it can lead mainly to suppression or degradation of its target [ 8 , 9 ]. Despite their ubiquitous nature miRNAs are tissue-specific, and their expression has also been associated with different stages of OC and clinical outcome [ 7 , 10 , 11 ]. Additionally, circulating miRNAs are highly stable in the body fluid, such as plasma, serum, urine, saliva, cerebrospinal fluid, amniotic and follicular fluids, presenting as a versatile non-invasive indicator for a wide range of cancers [ 12 – 14 ]. Thus, they may serve as potential biomarkers offering rapid and easy tests, and eliminating the necessity of performing invasive procedures, such as biopsies [ 15 – 18 ]. Early studies on genome-wide comparison of miRNA expression between OC and normal ovary tissue have found a set of miRNAs to be differentially represented [ 10 ]. More recently, a limited number of studies have suggested the existence of differential expression of circulating miRNA between healthy women and patients with pelvic masses. However, such studies were performed on very small sample sizes and did not focus on differentiating benign from malignant pelvic mass [ 7 , 13 , 19 ]. In this study, using an initial discovery cohort of 190 plasma samples from patients with pelvic masses, classified as benign control (n = 95) and malignant (n = 95), we investigated a panel of 46 miRNAs previously reported to be associated with ovarian and other cancers. We further validated our findings on an independent cohort with benign pelvic masses and late stage OC (FIGO III-IV).

In a total of 380 patients, 190 were diagnosed with benign pelvic masses, and 190 with malignant condition, categorised according to their respective FIGO stage and histological classification. Samples with not enough plasma material or poor RNA quality were further excluded in order to avoid technical outliers (14 OC and 13 benign cases). CA-125 levels and body mass index (BMI) were also measured. Patient demographics are shown on Table 1 . *Including serous adenofibroma **Including dermoid cysts. All OC cases in this study were traced by the Danish Central Population Register (CPR) and date of death or emigration were registered until January 11th, 2011. In addition, information about treatment (surgery and chemotherapy) and cause of death was retrieved from the DGCD database. In total, 127 patients completed the follow-up period for this study–a total of 73 OC patients died from OC (median follow-up time: 18 months, range: 1–68) and 54 patients were still alive (median follow-up time: 56 months, range: 38–75). The profile of plasma miRNA was investigated with a panel of 46 miRNAs ( S2 Table ). The panel was assembled based on miRNA known to be associated with OC or other types of cancer. By univariate analysis we found 4 miRNAs to be significantly up-regulated–hsa-miR-21-5p (OR = 2.15; p = 0.0046), has-miR-200c-3p (OR = 1.88; p = 0.0001), hsa-miR-221-3p (OR = 2.43; p = 0.0006), and hsa-miR-484 (OR = 1.38; p = 0.0181)–and 2 miRNAs down-regulated–hsa-miR-195-5p (OR = 0.73; p = 0.0071) and hsa-miR-451a (OR = 0.68; p = 0.0046)–in patients with malignant pelvic mass compared to patients with a benign ovarian tumour ( Table 2 ). OR: odds ratio. AUC: area under the curve. All p-values < 0.05 were considered as statistically significant. We further investigated the ability of all candidate miRNAs to discriminate between benign and malignant cases by individually calculating the Receiver Operating Characteristic (ROC) curves, including for CA-125. Among the 6 candidates, hsa-miR-200c-3p showed the highest predictive efficiency (AUC = 0.78), with an OR = 1.88 (95% CI:1.42–2.49), followed by hsa-miR-221-3p (AUC = 0.65), hsa-miR-195-5p (AUC = 0.63), hsa-miR-21-5p (AUC = 0.63), hsa-miR-484 (AUC = 0.63), and hsa-miR-451a (AUC = 0.62) ( Fig 1 ). CA-125 measurements had the highest index, with AUC = 0.92. Receive Operating Characteristic (ROC) curves were performed for each miRNA, and CA-125. To confirm our findings, we implemented further statistical analysis on our discovery cohort. To ensure that such findings were not biased in the first place by an incomplete set of data in our discovery cohort, multiple imputation with 25 iterations was done. Multivariate analysis on the imputed dataset was performed including all 6 candidate miRNAs in order to explore association between them and the presence of malignant pelvic mass. We found 2 miRNAs–hsa-miR-200c-3p and hsa-miR-221-3p –to be strongly associated with OC (p< 0.0001 for the logistic model). The discrimination for the combination of these 2 miRNAs, showed AUC = 0.75 ± 0.04. The multivariate model from the discovery cohort was then applied to the validation cohort demonstrating an increased accuracy efficiency (AUC = 0.82), with a sensitivity of 80% and 64% specificity ( Fig 2 ) which corresponded to a PPV of 71% and NPV 73%. A ROC curve was performed for the combination of miR-200c-3p and miR-221-3p validated by an independent cohort. Considering that CA-125 is currently used in the clinic for OC management, it alone presents as a very poor predictive marker, with limited specificity. Here, we tested the combination of CA-125 with hsa-miR-200c-3p and hsa-miR-221-3p together for its predictive ability. The performance of CA-125 in the validation cohort showed an AUC = 0.94, whilst the combination of both miRNAs and CA-125, as well as age, had a modest improvement at a high level, AUC = 0.96. However, the contribution of the miRNA’s was statistically significant (p = 0.004) with excellent discrimination, presenting the substantial overall predictive performance, with sensitivity = 80% and specificity = 95%, and PPV = 94% and NPV = 80% ( Fig 3 ; S3 Table ). A ROC curve was performed for the combination of both miRNAs together with CA-125. Noteworthy, the number of serous adenocarcinoma subtype samples is remarkably higher in both of cohorts compared to the other subtypes. However, the expression levels of both hsa-miR-200c-3p and hsa-miR-221-3p is similar among all subtypes, except for the endometrioid adenocarcinoma subtype (p = 0.038), likely due to the small sample size (n = 4) and the large standard deviation in that subtype ( S2 Fig ). In addition, further statistical analysis considering only serous adenocarcinoma did not show any increase on the performance of those markers (AUC = 0.95).

We found that plasma miRNA profiles distinguished patients with late stage OC from benign ovarian conditions consistently in both the discovery and validation cohorts. Moreover, the combination of miRNAs we found in this study and routinely used serum CA-125 levels [ 46 ] may improve the diagnosis of the disease in regard to sensitivity. In the future, we seek to explore our findings by further investigating whether those markers are also capable of identifying early stage OC patients. In this way, this approach might show very valuable for OC patient treatment, overall survival and quality of life.

All patients in the current study were recruited from the Pelvic Mass study, initiated in September 2004 and finalized in August 2007. Patients were referred to surgery for suspected pelvic tumor, following PET/CT exams within two weeks prior to procedure. Furthermore, all patients were subjected to abdominal and vaginal ultrasound, and serum CA-125 level and menopausal state assessments. Exclusion criteria were pregnancy, previous cancer or borderline tumour, patient unable to understand any information about the study, and cancellation of planned surgery because of no suspicion of pelvic disease after further examinations. All tissue specimens obtained during the exploratory surgery were examined by a specialized pathologist. All patients were registered in the Danish Gynaecologic Cancer Database (DGCD), which is a national compulsory research and quality on-line database. All proceedings were carried out according to the precepts of the Declaration of Helsinki, including written informed consent from all participating patients. The study has been approved by the Danish National Committee for Research Ethics, Capital Region (approval codes KF01-227/03 and KF01-143/04). Blood samples were secured less than 14 days before surgery. They were collected as described previously [ 20 ], and all handling procedures were accomplished within 6 h. RNA purification and reverse transcription followed by quantitative PCR (RT-qPCR) were performed as previously described [ 19 ]. Briefly, total RNA was purified from 380 plasma samples (100 μl/sample) using Norgen Total RNA Purification Kit (Norgen Biotek Corp. Ontario, Canada), with addition of 10mM DTT. As a technical control for further RT-qPCR analysis, C . elegans synthetic cel-mir-54 and cel-mir-238 (TAG Copenhagen A/S) were mixed to the lysis buffer prior to adding to each sample. Purified RNA was stored at -20°C until use. Reverse transcription was performed on a panel of 46 miRNAs (TaqMan microRNA Reverse Transcription Kit), followed by preamplification (TaqMan PreAmp master mix), and RT-qPCR using TaqMan miRNA assays (Applied Biosystems, Foster city, CA, USA). For the qPCR, we used a Fluidigm BioMark system (Fluidigm Corp. USA) allowing for a duplicate assay in one operational run. Four 96.96 gene expression chips (Fluidigm Corp. USA) were used. The panel of 46 miRNAs is described on S1 Table . For data analysis the auto detectors settings were chosen for signal detection threshold. Average raw Cq values above 35 were excluded from the data set. Each remaining average Cq value was normalized by subtracting the average of Cq of cel-miR-54 and cel-miR-238 for that particular sample, yielding the -ΔCq (average Cq of cel-miR-54 and cel-miR-238 –average Cq of hsa-miR). The -ΔCq values were further normalized with the average -ΔCq of all detected miRNAs in all samples to further correct for variations in total input RNA. In the end, we used the average -ΔCq of 23 miRNAs to subtract in the discovery cohort and 29 miRNAs in the validation cohort. These double-normalized expression values were used for all statistical analyses. Univariate analysis was performed on all miRNAs in the discovery cohort, identifying any with a p-value < 0.05. The statistical analysis approach was logistic regression modelling for the probability of malignant pelvic mass versus benign. The miRNAs were then subjected to a multivariate analysis on this cohort in order to select for further analysis. To account for missing values, multiple imputation [ 21 ] was then employed, assuming missing at random (MAR). The multivariate analysis was performed with 25 imputations for those miRNAs that passed the univariate analysis, selecting for candidates with p<0.005. Model fitting was assessed using martingale residuals and the Hosmer-Lemeshow test [ 22 ]. The final analysis included the selected miRNAs and clinical covariates assessed in the validation cohort. Results are presented by odd ratios (OR) with 95% confidence intervals (CI) and the area under the receiver operating characteristic curve (AUC). In addition, the positive predictive values (PPV), as well as the negative predictive values (NPV), were calculated for cutoffs corresponding to 80% sensitivity. Statistical calculations were done using SAS (SAS Institute, Cary, N.C., USA, v9.4) and R (R Development Core Team, Vienna, Austria) [ 23 ]. An overview of data handling and statistical analysis is shown on S1 Fig .

(DOCX) Click here for additional data file. (DOCX) Click here for additional data file. (DOCX) Click here for additional data file. (PDF) Click here for additional data file. (PDF) Click here for additional data file.