Circulating miR-200c as a diagnostic and prognostic biomarker for gastric cancer

Consecutive GC patients from the Medical Oncology Unit at the University Hospital in La Coruña (Galicia, Spain) were eligible for the study. The inclusion criteria included a confirmed pathological diagnosis of gastric or gastro-oesophageal junction adenocarcinoma and no prior systemic medical therapy for cancer. The exclusion criteria included any other previous malignancy, coagulation disorders and a platelet count less than 20.0 x 10⁹ L^-1.

The diagnostic work-up included a clinical examination, blood sampling, endoscopy (when clinically indicated) and computed tomography (CT) scanning of the chest, abdomen and pelvis. The patients were followed up clinically with imaging every 8 to 12 weeks for the first 2 years and every 6 months thereafter to monitor disease progression.

The controls were recruited from the patients’ family and relatives. We only excluded subjects with a previous history of malignant disease. Thus, controls with different chronic but stable diseases (e.g., peptic disease, hypertension, diabetes mellitus or heart disease) were eligible and consecutively recruited. The control cohort was selected to include a sex and age distribution that was comparable to the patient group.

This study was approved by the Ethics Committee of Clinical Investigation of Galicia (Spain) and conducted in compliance with the Helsinki Declaration. Written informed consents were obtained from all the patients and the controls prior to their inclusion in the study.

Tumours and regional lymph nodes collected during surgery were processed on a routine diagnostic basis. Histological type, depth of invasion and nodal involvement were analysed and the disease was staged and graded according to the TNM and Laurent classification [31]. Residual disease status at the time of blood sampling was classified as R0 when no residual disease was present after surgery, R1 when microscopic residual disease was found and R2 in the presence of macroscopic disease. The patients from whom the blood was obtained before the start of neo-adjuvant treatment were categorised as R2. When surgery was not performed, the pathological diagnosis was based on endoscopic or radiological-guided biopsies.

To isolate the miRNA fraction, the RiboPure-Blood Kit was used with the alternate protocol: isolation of small RNAs (Applied Biosystems, Foster City, CA, USA). The procedure was performed using 0.5 ml of whole blood per preparation. The absorbances at 260/280 and 260/230 were assessed using a NanoDrop™ 1000 spectrophotometer (Nanodrop Technologies, Wilmington, DE, USA). The purified RNA was further processed using qRT-PCR or stored at −80°C until use.

Reverse-transcription (RT) PCR was performed with 25 ng (up to 6.6 μl) of total RNA using the mirVana™ qRT-PCR miRNA Detection Kit (Ambion, AM1558) with 2 μl 5X RT Buffer, 1 μl 1X RT Primer (Ambion, miR-200a, A30094; miR-200b, AM30095*; miR-200c, AM30096*; miR-141, AM2052*) and 0.4 μl of ArrayScript Enzyme Mix for a total volume of 10 μl.

For the PCR reaction, 10 μl of RT reaction and PCR Master Mix were used. The PCR Master Mix consisted of 5 μl 5X PCR buffer containing SYBR Green I, 0.2 μl SuperTaq 5 U/μl, 0.5 μl PCR primers and 9.3 μl of nuclease-free water for a total volume of 15 μl. Real-time PCR was performed on the LightCycler® 480 Instrument (Roche, Mannheim, Germany).

To control input variability and sample normalisation, primer sets specific for the small RNA species U6 snRNA (Ambion, AM30303) and 5S rRNA (Ambion, AM30302) were used. These primer sets were used not only as internal controls but also to verify the integrity of the RNA and the reverse transcription reaction. Any specimen with inadequate U6 snRNA or 5S rRNA expression would be excluded from the study.

For miR-141, miR-200b and miR-200c, the PCR cycling conditions and analysis were as follows: denaturation at 95°C for 8 seconds; cycling, 40 cycles of 95°C for 5 seconds, 60°C for 5 seconds and 72°C for 2 seconds; melting curve analysis, 1 cycle at 95°C for 5 seconds, 55°C for 1 minute 5 seconds and 95°C continuous; and finally, cooling at 40°C for 10 seconds. The conditions were identical for miR-200a, U6 snRNA and 5S rRNA, except the denaturation step was 1 cycle at 95°C for 6 seconds.

We verified that the amplification of each PCR product was specific using a melting curve analysis. The amplification efficiency was determined for both target and reference genes. Each assay was performed at least in triplicate. The quantification cycle (Cq) was performed using LightCycler 480 Quantification software (Roche, Mannheim, Germany). For further data analysis, only those miRNAs with a Cq value equal to or below 35, representing detection of one single-molecule template [32], were considered. Positive and negative controls were included in each experiment.

The Relative Expression Software Tool (REST) was used to analyse the relative miRNA expression in each sample and to determine the fold difference for every miRNA [33]. The expression levels of the target miRNAs were standardised using an index containing 5S rRNA and U6 snRNA.

miRNA analyses were performed with no knowledge of the clinical or follow-up data.

To analyse the expression of the miR-200 family in gastric cancer, the OE19 and MKN-45 human gastric cell lines were used. The cell lines were maintained in Dulbecco’s modified Eagle’s medium (DMEM) with high glucose and MegaCell^TM RPMI-1640 medium (both provided by Sigma–Aldrich Química, Madrid, Spain) supplemented with 10% inactivated foetal calf serum, 1% penicillin, 1% streptomycin and 1% amphotericin at 37°C in 5% CO₂. The cells were recovered with 1% trypsin–1% EDTA cell-dissociating reagent.

The isolation of total RNA (including miRNA) from the cell cultures was performed using the mirVana^TM miRNA isolation kit (Ambion, Inc. AM1560). The procedure was performed using 10⁷ cultured cells at 70% confluence.

The miR-200 expression profiles in paired samples (n = 160) of non-tumour gastric mucosa and GC were obtained using bioinformatic analysis of the data originally published by Ueda T, et al. [10]. The microarray expression was downloaded through the public repository ArrayExpress (experiment number E-TABM-341. http://​www.​ebi.​ac.​uk/​arrayexpress/​). Only the normalised expression values were used for subsequent analysis. The differential expression levels were calculated using a moderate t-test implemented in the Bioconductor limma package (R statistical software). The comparisons were performed using t-test and pairwise t-tests. The resulting p values were adjusted for multiple testing using Benjamini-Hochberg’s adjustment [34, 35].

The primary aims were to estimate the diagnostic accuracy and usefulness of miRNA as measured by qRT-PCR in the blood of GC patients as a clinical biomarker and to determine its potential prognostic value. The study was performed following the proposed guidelines of the Early Detection Research Network [36]. The design and results are presented in accordance with the REMARK [37] and MIQE guidelines [38].

The receiver operating characteristic (ROC) curve was constructed by plotting sensitivity (Y-axis) vs 1-specificity (X-axis) and the areas under the curve (AUC) were calculated. The diagnostic performance including sensitivity, specificity, positive and negative predictive values and accuracy of miR-200c quantification was also estimated [36]. The potential correlation among blood miRNA levels and the clinical and pathological features of the study subjects were analysed. The normality of the distribution of miRNA expression was analysed using the Kolmogorov-Smirnov test. Thus, parametric or non-parametric statistics were used, as appropriate. The relationships between miR-200c levels and the quantitative clinical variables were analysed using the Pearson correlation.

Progression-free survival (PFS) was measured as the time between the baseline blood sampling for miRNA analysis and the documentation of first tumour progression, based on clinical and radiological findings, or death (events). Overall survival (OS) was measured from the time at which the baseline blood sample was obtained to the date of death from any cause or date of last follow-up. The patients who were alive and progression-free at the time of analysis were censored by using the time between the blood assessment and their most recent follow-up evaluations. The distributions of time-to-event end points, namely PFS and OS, were estimated using the Kaplan-Meier method and compared using the Breslow-Wilcoxon test.

Multivariate survival analyses (PFS and OS) were performed using Cox regression models. We estimated hazard ratios (HRs), 95% CI and p values. All statistical tests were two-sided and p values less than 0.05 were considered significant. SPSS Statistics 19.0 for Windows (IBM Corporation, Armonk, NY, USA, 2011) and Graph Pad Prism 5 (GraphPad Software, La Jolla, CA, USA, 2007) were used for data analyses.

To investigate the differential expression levels of the miR-200 cluster, we used real-time PCR to analyse the expression levels of miR-200a, 200b, 200c and miR-141 in total RNA extracted from the GC cell lines OE-19 and MKN-45. We compared the miRNA expression profiles, calculated using REST as described, with those of normal human blood (a control group consisting of pooled RNA obtained from 19 healthy donor blood samples). The relative expression ratios of every target miRNA were significantly higher in the GC cell lines compared with the control blood. In OE-19 cells, the miRNAs were upregulated by a mean factor of 6.61x10⁵, 9.99 x10³, 4.47 x10⁵ and 2.54 x10⁵ for miR-200c (p < 0.001), 141 (p = 0.018), 200a (p < 0.001) and 200b (p < 0.001), respectively. In MKN-45 cells, the miRNAs were upregulated by a mean factor of 4.94 x10⁵, 5.79 x10³, 2.86 x10⁵ and 1.30 x10⁵ for miR-200c (p = 0.033), 141 (p < 0.001), 200a (p < 0.001) and 200b (p < 0.001), respectively. Thus, the highest fold-change observed in the GC cell lines relative to control blood was 5.78 x10⁵ for miR-200c. In addition to the miRNA expression data analysis obtained by REST, we compared the raw Cq data for every miRNA in the control blood and gastric cancer cell lines. In the blood, the mean Cq was lower for miR-141 (Cq = 28) compared with miR-200a (Cq = 35), 200b (Cq = 35) and 200c (Cq = 35). These differences were significant (ANOVA, p < 0.001; Bonferroni post hoc test, p < 0.001) suggesting an increased background miR-141 expression in non-tumour blood relative to the other miR-200 family members. In the GC cell lines, the mean Cqs were 15.3, 16.7, 17.7 and 16.1 for miR-141, 200a, 200b and 200c, respectively, without significant differences (ANOVA, p = 0.133; Figure 1).

To ascertain whether the miR-200 cluster signature differs between GC and non-tumour mucosa and between the two histological subtypes of GC, the expression profiles were retrieved from Ueda et al. [10]. We observed that the miR-200 family was not differentially expressed in the paired non-tumour mucosa and cancer samples. Furthermore, miR-141, 200a, 200b and 200c were not differentially expressed between the GC histological subtypes (diffuse and intestinal). Additional file 1: Figures S1, S2 and Additional File 1: Tables S1, S2.

From November 2006 to July 2010, 52 patients with histologically proven GC were consecutively recruited for this study. The clinical characteristics of the patients are shown in Table 1. The control cohort included 15 cases. The mean age was 65.3 years (standard error of the mean [SEM], 1.9; range, 49 to 74 years) in the control group and 65.9 years (SEM, 1.32; range, 43 to 85) in the patient group (t test, p = 0.82). The ratio of males to females was similar among the controls and the patients (Yates-corrected χ², p = 0.07).

The blood was obtained after R0 surgery in 20 patients (38.5%). In 32 patients, the blood samples were obtained in the presence of residual or metastatic disease, both of which were categorised as R2 at the time of blood sampling. In the patients receiving surgery (71.2%; 37/52), the number of lymph nodes analysed was 19 (range, 0–76; St. D, 16.2). Chemotherapy was administered to 44 patients (84.6%).

All patients were followed until death or study completion. The last date of follow-up for the survivors was September 5, 2011. Disease progression events occurred in 38 patients (73.1%). The median PFS was 6 months (95% CI, 1.4 to 10.6 months). There were 7 relapses among stage I–III patients and 31 progressions of metastatic disease. The median OS was 15 months (95% CI, 11.1 to 18.9 months) and 35 patients (67.3%) died of advanced disease. Most of the PFS events (29/38; 76.3%) and OS (18/35; 51.4%) events occurred in the first 9 months of follow-up. The mean (SEM) follow-up time for the patients still alive at the time of the analysis was 26.3 (3.7) months (median, 24 months; range, 6 to 53 months).

As described above, we found that miR-200c was not only upregulated in GC cell-lines compared with control blood, it was expressed at the highest levels of all miR-200 family members. Thus, miR-200c was selected for clinical validation.

Real-time quantitative assessment of miR-200c was performed using 67 blood samples (52 patients and 15 controls). The mean relative miR-200c expression (Figure 2) was 16.2 (SEM, 5.6; CI 95%, 4.1 to 28.3) in controls, 90.3 (SEM, 17.4; CI 95%, 53.9 to 126.6) in stage I-III patients and 114.6 (SEM, 16.3; CI 95%, 81.4 to 114.9) in stage IV GC patients (p = 0.018; Kruskal-Wallis test. Bonferroni post hoc test: stage I-III vs control, p = 0.018; stage IV vs control, p < 0.001). The confidence interval with the alpha level of significance at 99% estimated using the Monte Carlo test was 0.015 to 0.022.

An ROC curve was constructed (Figure 3). Comparing the relative miR-200c levels in controls and patients, the AUC was 0.715 (95% CI, 0.597–0.833; p = 0.012). According to the ROC curve, the relative blood level of miR-200c of 62.4 was defined to be the optimal cutoff value for differentiating GC patients and controls (Youden's index). With this cutoff value for miR-200c, the sensitivity, specificity, positive and negative predictive values and accuracy values of 65.4% (95% CI, 50.8 to 77.7), 100% (95% CI, 74.7 to 99.4), 100% (95% CI, 87.4 to 99.7), 45.5% (95% CI, 28.5 to 63.4) and 73.1% (95% CI, 60.7 to 82.9), respectively, were achieved. The relative expression values for miR-200c in blood above this cutoff point were found in 17.6% of stage I-II patients, in 20.6% of stage III patients and in 67.7% of stage IV GC patients (p < 0.001; exact test). These findings suggested that elevated blood miR-200c could be detected in the early stages of GC and therefore facilitate early disease detection.

The clinical and pathological characteristics and the miR-200c expression levels in the blood from cancer patients are shown in Table 2. The relative expression levels for miR-200c in the blood of GC patients were not associated with any of the parameters analysed. Furthermore, the percentage of patients with miR-200c levels above its mean value (mean, 104.8; SEM, 12) was not associated with clinical and pathological characteristics.

The correlations of potential prognostic factors and miR-200c levels in the blood in gastric cancer patients are shown in Table 3. There was only a significant correlation (Pearson’s r = 0.438, p = 0.016) between miR-200c levels and the number of lymph node metastases.

To generate survival curves, we converted continuous miR-200c expression levels measured using qRT-PCR to a dichotomous variable, using its mean levels of expression as a threshold (10). Using this approach, miR-200c was overexpressed in the blood of 53.8% (28/52) of patients. The mean values (with SEM) in the low and high expression groups for miR-200c were 23.9 (7.9) and 174.1 (8.5), respectively (Mann–Whitney test, p < 0.001). The percentage of patients with miR200c overexpression tended to increase with TNM stage: 33.3% (3/9) in stage I-II patients, 50% in stage III patients (6/12) and 61.3% (19/31) in stage IV patients (p = 0.076).

The Kaplan-Meier curves for patient OS and PFS categorised according to miR-200c expression levels in the blood are shown in Figure 4. The increased blood expression of miR-200c was significantly associated with a poor overall survival (Breslow-Wilcoxon text; p = 0.016). The median and mean OS for the group with high miR-200c expression levels were 9 months (95% CI, 1.7–16.3) and 17.4 months (95% CI, 11.2–23.6), respectively. In the group with low miR-200c blood expression levels, the median OS was 24 months (95% CI, 8.1–39.9) and the mean OS was 29.2 months (95% CI, 20.9–37.6).

The relative strength of blood expression levels of miR-200c as an independent prognostic factor was evaluated by performing a Cox multivariate analysis. The details of this analysis are listed in Table 4. With the inclusion of miR-200c expression levels in the model, the independent predictors of PFS were as follows: stage IV disease (HR = 5.52; p = 0.005), residual disease (HR = 4.29; p = 0.023); and miR-200c overexpression (HR = 2.27; p = 0.028). Similar results were achieved for the prediction of OS including stage IV disease (HR = 8.6; p < 0.001), weight loss higher than 10% (HR = 2.38; p = 0.024) and miR-200c overexpression (HR = 2.24; p = 0.028) in the model. Residual disease, Eastern Cooperative Oncology Group (ECOG) performance status and age were not independent prognostic factors for OS.

To further explore the relationship between miR-200c expression levels and outcomes, we estimated the hazard ratios associated with the miR-200c level as a continuous variable by performing Cox multivariate regression models. Concordant results were achieved for the prediction of PFS, considering miR-200c expression levels as a continuous variable (HR = 1.004; 95% CI; p = 0.045) in the multivariate Cox model including stage IV and residual disease. Likewise, the risk of death was higher with increasing miR-200c relative blood expression levels (HR = 1.007; 95% CI, 1.003 to 1.012; p = 0.003) independent of stage and weight loss.