The prognostic role of miR-181a-5p in advanced stages epithelial ovarian cancer patients enrolled in MITO16A/MaNGO-OV2 clinical trial

2.1 Cohort selection

An anonymized cohort of 398 stage III/IV EOC patients for whom formalin-fixed paraffin embedded (FFPE) primary tumor biopsies and complete clinical record and follow-up were available, was recruited within the MITO16A/MaNGO-OV2 clinical trial between October 2012 and November 2014. Details on the study design, procedures, clinical and demographic characteristics of patients enrolled and clinical evidences obtained have been previously published [17, 18, 19]. Briefly, the main inclusion criteria included: a diagnosis of EOC stage IIIB–IV, performance status 0–2 according to ECOG (Eastern Cooperative Oncology Group), no history of other tumor malignancy within the last five years, and no contraindications to bevacizumab administration. For all patients naive to chemotherapy, FFPE samples derived from primary tumor site and collected during the diagnostic laparoscopy—in the cases of patients who underwent neoadjuvant chemotherapy (NACT)—or during the primary debulking surgery were available. The primary endpoint of the trial was to identify clinical and biological factors that, at diagnosis, affect the different prognosis, in terms of progression-free (PFS) or overall survival (OS) of stage III/IV EOC patients, receiving frontline Pt-based chemotherapy followed by maintenance with bevacizumab. Details are as previously published [17]. The trial was conducted in compliance with the ethical principles outlined in the Declaration of Helsinki and the International Council for Harmonization guidelines on Good Clinical Practice. The protocol was presented and approved by Ethics Committee of the Istituto Nazionale Tumori IRCCS, Fondazione G. Pascale, Naples, Italy (Ref. No. 187/16E, 10 October 2016) and accepted by the institutional review board or ethics committee of each center. All patients, or their legal representatives, provided written informed consent before enrolment.

2.2 miRNAs isolation and miR-181a-5p expression analysis

Total RNA was extracted from two 1 mm cores of FFPE tissues using Qiagen miRNeasy FFPE Kit as previously described by Califano and colleagues [20] (miRNeasy FFPE kit, Qiagen, Venlo, Netherlands). RNA was measured by QIAxpert spectrophotometer (QIAexpert system, Qiagen, Venlo, Netherlands) and the presence of small RNAs was checked with a 2100 Bioanalyser (Agilent Technologies, Santa Clara, CA, USA). Aliquots were stored at −80 °C until use. Mature miR-181a-5p expression levels were analyzed by quantitative Reverse Transcription Polymerase Chain Reaction (qRT-PCR) by Sybr Green chemistry using the commercial miScript® Universal Primer, together with the miScript® Primer Assay (Qiagen, Venlo, Netherlands) on an Applied Biosystems 7900 HT instrument (Applied Biosystems, Waltham, MA, USA). Single replicate experiments were set using 384-well reaction plates in an automatic liquid handling station (epMotion 5075 LH; Eppendorf, Milan, Italy) and raw data generated with SDS Relative Quantification software (version 2.3, Ambion-ABI, Waltham, MA, USA). To correct for technical variability and obtain a more accurate measure of the miR-181a-5p expression, the raw CT (cycle threshold) values of miRNA-181a-5p were normalized using the average CT expression of three different housekeeping miRNAs, according to the ΔCT (delta CT) method as previously published [21, 22]. This method improves the reliability of the results, especially when comparing expression levels across different samples, as in this study with samples recruited across many different clinical centers. Briefly, three miRNAs, the expression of which displayed the minimal variation across samples, were selected as housekeeping miRNAs for data normalization (i.e., miR-191-5p, miR-16-5p and miR-24-3p). Then, for each housekeeping miRNA the range of CT expression was normalized by subtracting the minimal CT value (CTmin) from the raw CT values of all samples (CTnormalized = CT − CTmin), and results converted according to the following formula: 2^−CT normalized. For each sample analyzed, a normalization factor (NF) was calculated as the geometric mean of the 2^−CT normalized values of the three housekeeping miRNAs. To normalize the expression value of miR-181a-5p, the raw CT value of each sample, were first converted in linear scale according to the following formula: 2^−CT and then divided by the NF calculated on the same sample. To avoid biases due to low RNA quality from FFPE samples, only those samples with CT raw data below 35 were considered for downstream analysis.

2.3 Smad2 phosphorylation

Immunostaining was performed on 3 µm FFPE tissue sections mounted on poly-L-lysine-coated slides and dried at 37 °C overnight. As previously reported by Zannoni and colleagues [23], slides were deparaffinized in xylene and rehydrated conventionally, and the endogenous peroxidase was blocked with 3% H₂O₂ for 5 min. Antigen retrieval procedure was performed by microwave oven heating in 10 mM citric acid, pH 6.0 (2 times for 5 min). To reduce non-specific binding, the sections were incubated with 20% normal goat serum for 20 min, at room temperature. Cells expressing P-Smad2 were identified after overnight incubation at 4 °C by using the polyclonal rabbit anti-Smad2, phospho-specific (Ser465/467) antibody (diluted 1:1000; Millipore) in 20% normal goat serum. All slides were then processed with the labeled polymer EnVision-rabbit System-HRP (DAKO, Carpinteria, USA), 30 min, at room temperature. Diaminobenzidine was used as a chromogen (DAB substrate System, DAKO, Carpinteria, USA). Sections were counterstained with haematoxylin, dehydrated, cleared in xylene and mounted with Eukitt. Negative control was obtained by omission of the primary antibody. Positive control for P-Smad2 was represented by sections taken from the breast. The analysis of all tissue sections was done, without any prior knowledge of clinical parameters, by two authors by means of light microscopy and at least 5 fields were analyzed for each tissue section. The percentage of immunostained tumor cells was assessed at low magnification (5× objective lens) by evaluating the entire tumor area, and it was expressed as the percentage of positive cells, considering any level of staining intensity [16]. The percentage of cells exhibiting positive nuclear staining for P-Smad2 was scored as follows: 1 (1–10%), 2 (11–50%), 3 (51–75%) and 4 (>75%). Staining intensity was scored as 0 (absent), 1 (weak), 2 (moderate) and 3 (intense). Results were confirmed by repeating the staining experiment on serial sections from the same block. All specimens were reviewed and confirmed by hematoxylin-eosin staining by 2 independent experienced pathologists involved in the present study who had no prior knowledge of the clinicopathologic parameters and clinical outcomes of the patients. The % of tumor content has been also assessed on the best representative block, considering it optimal when ≥25%.

2.4 Statistical analysis

For descriptive statistics, we used median values and interquartile range (IQR) for continuous variables and absolute numbers and relative frequency for qualitative variables. For each biomarker a histogram was used to describe the distribution of values. To evaluate the correlation between biomarkers, a Spearman correlation test was used. To investigate the associations between each biomarker and the clinical prognostic factors the Wilcoxon rank sum test for dichotomous variables and the Kruskal-Wallis for categorical variables were used. The prognostic effect of each biomarker was tested on PFS and then on OS. PFS was defined as the time elapsing from the inclusion into the study to the first occurrence of either death for any cause or disease progression. OS was defined as the time elapsing from the inclusion into the study and death for any cause. For both PFS and OS Kaplan-Meier curves were drawn and compared with a two-sided log-rank test. The prognostic role of biomarkers was tested using a univariate and then with a multivariate Cox proportional model on both PFS and OS. In a first univariate analysis, each biomarker was tested as a continuous variable after testing the linearity assumption using fractional polynomials. In a second univariate analysis the biomarker was tested as a categorical variable searching for the best cut-off. The best cut-off was selected among the biomarker values choosing the value that minimized the p-value of Hazard Ratio (HR) for the categorical variable defined by the cut-off value. The best cut-off search was calculated on PFS and then applied to the OS. To adjust for overfitting HR estimates of best cut-off categories, a shrinkage procedure with 95% bootstrap-percentile method was performed [24].

Then a multivariable analysis was performed using as covariates: age (as category <65 vs. ≥65), ECOG performance status (PS) (0 vs. 1–2), residual disease (none; ≤1 cm; >1 cm; not operated), FIGO stage (III vs. IV), tumor histology (high-grade serous vs. other), BRCA status (mutated; wild type; missing) and each biomarker (considered as continuous variable and as categorical variable defined by best cut-off value). A multivariable Cox proportional analysis was performed for each biomarker and for both continuous variables and based on best cut-off categories. Exploratory subgroup analyses for PFS and OS were performed and reported in a forest plot including p-value for the first-order interactions between covariates and each biomarker, tested with the likelihood-ratio test of two nested models, with and without interaction. Data were analyzed using R software version 3.6.0 (R Foundation for Statistical Computing, Wien, Austria).

In the MITO16A/MaNGO-OV2 clinical trial protocol was reported for sample size calculation that the study had 80% power to identify a potential prognostic marker, expressed at least in 20% of the population, able to select a favorable subgroup with a HR of 0.60. With these assumptions and considering an alpha level of 0.05, 188 events of PFS were needed. In this study 253 cases were analyzed with both the techniques and 196 events for PFS were observed.

3.1 Cohort description

In the MITO16/MaNGO-OV2 population the median age at diagnosis was 59.2 years (49.9–66.5 years), and all patients received the same schedule of treatment. As frontline chemotherapy patients received carboplatin (area under the curve, AUC = 5) and paclitaxel (175 mg/m²) plus bevacizumab (15 mg/kg) on day 1 for six 3-weekly cycles, followed by bevacizumab monotherapy (15 mg/kg), up to a maximum of 22 cycles. All patients enrolled were diagnosed at advanced FIGO stages of disease, i.e., stage III (78.1%) or IV (21.9%). As reported in Table 1, the majority of patients were histologically classified as high-grade serous (HGS) subtype (333 out of 398, 83.7%) while the remaining 16.3% were ascribable to other EOC histotypes. Since the trial was designed before germline and somatic BRCA status was considered for the routine clinical practice, it was not an inclusion criterion for the trial enrollment, and thus analysis of pathogenic BRCA mutations was completed in 249 cases (62.5%), 77 of which were mutated (19.3%) (Table 1). At the time of the database lock, the median follow-up was 32.3 months (IQR = 24.1–40.4) with a median PFS of 20.8 months (95% CI = 19.1–22.0), and median OS of 41.1 months (95% CI = 39.1–43.5) [17]. The clinical and pathological features of the MITO16/MaNGO-OV2 population is representative of the majority of patients with advanced EOC and thus suitable for the translational aim of the study. As detailed in the Consort diagram reported in Supplementary Fig. 1, 40 out of 398 patients enrolled were not eligible for the downstream analysis due to the unavailability or the inadequacy of the tumor sample. Due to the low tumor content and/or the insufficient amount of nucleic acid obtained, other 68 patients were excluded from molecular analysis. Additional selection criteria for downstream analyses were applied for qRT-PCR and immunohistochemistry (IHC) experiments. Only samples with an unprocessed threshold cycle value (raw CT) lower than 35 and samples with minimal degradation or necrosis with a well-preserved morphology were included. Tissue handling and processing procedures were standardized in a single histopathological laboratory to reduce the bias of preanalytical variables. Overall, we successfully selected 324 cases for IHC analysis of P-Smad2 phosphorylation and 271 cases for miR-181a-5p expression profiles, regardless of the mutational status availability in the BRCA1 and BRCA2 genes. Of these, 253 cases were analyzed with both the techniques.

3.2 Analysis of miR-181 expression profile

To validate the clinical relevance of miR-181a-5p, a punctual analysis of its expression levels was performed using qRT-PCR in tumor biopsies of patients enrolled in the MITO16A/MaNGO-OV2 clinical trial (Table 1). Supplementary Table 1 and Supplementary Fig. 2 panel A show the percentage of normalized CT distribution of miR-181a-5p across the entire cohort of 271 cases (median = 9.7, IQR = 5.8–15.9). The analysis of the association of miR-181a-5p with clinical characteristics revealed a significant correlation with FIGO stage and tumor histology, with lower values in patients with FIGO stage IV (median = 7.2, IQR = 4.7–11.6, p = 0.033) and in patients with HGS-EOC (median = 9.5, IQR = 5.5–15.6, p = 0.0126). Among the 202 patients with available BRCA mutational status, no significant difference in miR-181a-5p expression levels was observed between BRCA wild-type (median = 9.9, IQR = 5.9–15.9) and BRCA mutated tumors (median = 8.1, IQR = 4.6–14.2, p = 0.1162).

3.3 Smad2 phosphorylation analysis

To validate the clinical relevance of Smad2 phosphorylation, we performed IHC analysis on the 324 cases selected. A representative IHC analysis showing high and low P-Smad2 expression levels is reported in Fig. 1. In cases with high P-Smad2 expression, the staining intensity was strong, as evidenced by dark brown nuclear staining, while cases with low P-Smad2 expression exhibited weak to moderate staining intensity. Nuclear expression of P-Smad2 was heterogeneous across the cohort of 324 cases, ranging from 0 to 90%, as depicted in Supplementary Fig. 2 panel B (median = 20.0, IQR = 10.0–60.0). However, since the intensity of staining was consistent among tumor samples and did not provide additional discriminatory power, it was not considered in the analysis. Instead, the analysis focused on the percentage of nuclear expression, which showed significant heterogeneity and clinical relevance. Differences in basal expression profiles of P-Smad2 showed a significant association with ECOG PS and residual disease, with higher values in patients with an ECOG PS of 1–2 (median = 30.0, IQR = 15.0–70.0, p = 0.0043) and in patients who were not eligible for surgery (median = 70.0, IQR = 55.0–80.0, p < 0.001). No trend or correlation was observed between P-Smad2 expression levels and BRCA status. The detailed analysis of the associations between miR-181a-5p, P-Smad2 and clinical characteristics is provided in Supplementary Table 2.

Fig. 1.Smad2 phosphorylation. Representative immunohistochemistry pictures showing high (panel a) and low (panel b) P-Smad2 expression in the examined series (magnification 20× and 40×).

3.4 Survival analysis

To assess the potential prognostic role of these two biomarkers, analysis was initially performed by testing the association between the expression levels of miR-181a-5p and P-Smad2 with PFS and OS (Table 2 and Supplementary Fig. 3). In a univariate analysis model, using miR-181a-5p as continuous variable, a significant non-linear association was found with both PFS (HR = 0.97; 95% CI = 0.95–0.99, p = 0.006) and OS (HR = 0.96; 95% CI = 0.93–1.00, p = 0.049). Including age, ECOG PS and residual disease as covariates, multivariate analysis confirmed the expression levels of miR-181a-5p as associated to PFS only (HR = 0.97; 95% CI = 0.94–0.99, p = 0.007), while no significant association with OS was observed. Focusing on P-Smad2 as continuous variable, both univariate (HR = 1.13; 95% CI = 1.05–1.21, p < 0.001) and multivariate analysis (HR = 1.09; 95% CI = 1.01–1.18, p = 0.026) showed a significant non-linear association with OS (Table 2). To categorize patients in two groups on the basis of their miR-181a-5p and P-Smad2 levels (“high” or “low”) and to test the prognostic relevance of these classification, we passed to consider these biomarkers as dichotomic variants using the cut-off levels previously published (miR-181a-5p, 55th percentile = 10.5). Considering the previous miR-181a-5p threshold its prognostic effect did not reach the level of significance at univariate analysis for PFS (HR = 1.14; 95% CI = 0.86–1.52, p = 0.362) while a trend was observed for OS (HR = 1.51; 95% CI = 1.00–2.28, p = 0.052).

As these cut-off values were calculated and tested in a smaller cohort of patients, who underwent different surgical approaches and chemotherapy treatment, we decided to exploit this larger cohort to optimize and refine our previous threshold. For this reason, we next performed an analysis of the best cut-off aimed to identify for each biomarker the best cut-off that minimizes the p-value of HR (see “Material and Methods”, “Statistical analysis” section). Using these cut-off values (5.0 for P-Smad2 and 2.4 for miR-181a-5p), we tested the abilities of the different biomarkers to predict patients’ prognosis using Kaplan-Meier curves both for PFS and OS (Fig. 2). In univariate analysis, before applying shrinkage procedure, a significant association was found for both biomarkers and endpoints (Table 3). After adjustment for overfitting, miR-181a-5p maintained a significant association with PFS (HR = 5.60, 95% CI = 1.40–22.43, p = 0.015), while P-Smad2 maintained a significant association with OS (HR = 2.23, 95% CI = 1.12–4.41, p = 0.022). The heterogeneity of the prognostic effects of P-Smad2 was also studied in a subgroup analysis considering BRCA status, histology, FIGO stage and residual disease with the aim to generate hypotheses. No significant interactions were found on both PFS and OS (Fig. 3). Due to the low number of patients in the low category of miR-181a-5p, the subgroup analysis was not done.

Fig. 2.Kaplan-Meier curves. Kaplan-Meier curves for PFS (panel a) and OS (panel b) according on best cut-off value of P-Smad2 (left panel: cutoff value of 5.0) and miR-181a-5p (right panel: cutoff value of 2.4).

Fig. 3.Exploratory subgroup analysis of P-Smad2 for PFS and OS. PFS: Progression-Free Survival; FIGO: International Federation of Gynecology and Obstetrics; OS: Overall Survival; BRCA: BReast CAncer genes.

In the multivariate analysis after applying shrinkage procedure only a significant association was found for miR-181a-5p with PFS (HR = 5.25, 95% CI = 1.32–20.80, p = 0.018). Focusing on high-grade serous histotype the prognostic effect of miR-181a-5p in multivariate analysis results in HR = 5.40, (95% CI = 1.69–17.21, p = 0.004) and HR = 2.84, (95% CI = 0.68–11.83, p = 0.152) for PFS and OS respectively, thus indicating the highest predictive value that miR-181a-5p has in this specific histotype. A schematic summary of the workflow and the final results obtained is presented in Fig. 4.

Fig. 4.Schematic representation of the workflow and final results. The figure illustrates the key steps of the methodology and the main outcomes obtained. Abbreviations: Carbo: carboplatin; Beva: Bevacizumab; Pt-S: Platinum-sensitive; Pt-R: Platinum-resistant; FFPE: formalin-fixed paraffin embedded; IHC: Immunohistochemistry; P-Smad2: phosphorylated Smad2.