Patients
This study included 40 cervical cancer patients admitted to the Gynecologic Oncology Unit, Fondazione Policlinico Universitario A. Gemelli, IRCCS, Roma, between January 2000 and October 2013. Staging was performed according to FIGO classification. Patients with a diagnosis of stage IB2-III LACC disease were evaluated in the study (Additional file 1: Table S1). Pre-treatment tumor tissue biopsies were obtained during staging procedures, the joint assessment by surgeon and pathologist allowing an unequivocal identification of tumor area to be sampled. Based on the large dimension of tumor size (more than 4 cm diameter), we failed to detect any site in the cervix showing normal morphology. Pre-treatment tumor specimens were either formalin-fixed paraffin-embedded (FFPE) for histopathological diagnosis or immediately frozen in liquid nitrogen for subsequent protein and nucleic acid extraction. Patients received preoperative CRT; RT was administered to the pelvic region (39.6–50.4 Gy) according to specific protocols, and concomitant chemotherapy included cisplatin and 5-fluorouracil or capecitabine [7, 12]. Seven or 8 weeks after the end of concomitant CRT, all cases were submitted to radical hysterectomy and pelvic ± aortic lymphadenectomy. After surgery, patients were triaged to routine follow-up procedure according to the previously reported schedule [7, 12].
To assess the amount of residual neoplastic tissue, FFPE tissue sections prepared from surgically resected specimens were evaluated by an expert pathologist as previously reported [13]. Specifically, pathologic complete response was defined as the absence of any residual tumor after treatment at any site level (residual tumor = None, pR0). Microscopic response included cases with persistence of only microscopic tumor foci at any site level (≤ 3 mm maximum dimension, pR1), while macroscopic response included cases with persistence of residual tumor > 3 mm (maximum dimension, pR2) [13]. In order to maximize the identification of potential differences in the biomarker profile associated with CRT responsiveness, we decided to focus our analysis on patients with complete response versus macroscopic residual tumor. Our cohort included 20 Sensitive (S, i.e. pathological complete response) and 20 Resistant (R, i.e. macroscopic residual tumor) patients. Additional file 1: Figure S1 describes the study flowchart.
Protein and nucleic acid extraction
Protein, DNA and RNA were isolated from tissue using AllPrep DNA/RNA/Protein Mini kit (Qiagen, Hilden, Germany), according to manufacturer’s instructions. DNA, RNA and proteins were independently purified and stored for subsequent analysis.
2D-DIGE-based proteomic analysis
Total proteins extracted from tumor tissue biopsies were further purified using Clean-Up kit (GE Healthcare). Proteomic profiles of 20 S and 20 R patients were comparatively analyzed through two-dimensional Difference In-Gel Electrophoresis technology (2D-DIGE) (GE Healthcare). Briefly, proteins were covalently labeled with CyDyes DIGE Fluors (Cy5 and Cy3), while a pool of all experimental samples was labeled with Cy2 to provide a common internal standard. After 2D electrophoretic separation (as reported in Additional file 1: Additional Materials and Methods), protein maps were visualized by Typhoon 9410 Imager (GE Healthcare), which was set at the appropriate wavelengths for each dye. All gels were scanned at 100 μm resolution and the photomultiplier tube was set between 525 and 680 V. Images were then exported to DeCyder (v 7.2, GE Healthcare) batch processor for Differential-In gel Analysis and elaborated by Biological Variation Analysis module for statistical analysis [14]. Univariate analysis one-way ANOVA was performed applying a false discovery rate filter to reduce the number of false positives. Protein spots with statistically significant variation (P ≤ 0.05), differential volume over 1.3-fold and a minimal representativeness among protein maps of 65% were identified as differentially represented. Differentially represented spots were automatically isolated from gel by the Ettan Spot Picker System (GE Healthcare). Multivariate analysis, consisting of hierarchical clustering and Principal Component Analysis (PCA), was performed using the DeCyder Extended Data Analysis module, which allowed to highlight the differences between the two groups of samples, excluding individual variability.
Protein identification
Sampled spots were reduced, alkylated with iodoacetamide and digested with trypsin (Promega, USA) as previously reported [15]. Peptide digests were analyzed with an Easy-nanoLC (Proxeon, Odense, Denmark) connected through a nanospray source (Proxeon) to a LTQ XL mass spectrometer (Thermo Fischer Scientific, Waltham, USA) [16]. Experimental details are reported in Additional file 1: Additional Material and Methods. Raw mass spectrometric data were searched against a UniProtKB database of human sequences (2015/05; 167,637 protein entries) using MASCOT v2.3.02 software package (Matrix Science, UK). The following parameters were used for identification: a mass tolerance value of 1.8 Da for precursor ion and 0.8 Da for fragment ions, trypsin as proteolytic enzyme, a missed-cleavages maximum value of 2, Cys carbamidomethylation, pyroglutamate formation at Gln and Met oxidation as fixed and variable modifications, respectively. Protein candidates with more than 2 assigned peptide sequences, with MS/MS ion score > 30 and a peptide expectation value ≤0.05, were further evaluated by comparison with their calculated mass and pI values, using the experimental values obtained from 2D-DIGE. Definitive peptide assignment was always associated with manual spectra visualization and verification.
Assessment of mRNA expression profiles using Fluidigm 48.48 dynamic arrays
Quantity and quality of recovered RNA from cervical cancer samples were measured using Nanodrop (Thermo Fisher Scientific) and Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA), respectively. A total of 32 samples (i.e. 16 S and 16 R) were considered suitable for mRNA quantification. Analysis was carried out on total RNA using 48.48 dynamic array (Fluidigm Corporation, San Fransisco, CA, USA) and a Biomark platform, following the manufacturer’s protocol, as previously described [17]. Experimental details are reported in Additional file 1: Additional Material and Methods. The list of genes evaluated in the study, along with primers used, is shown in Additional file 1: Table S2.
Digital PCR
In order to validate the differential expression of the genes of interest (i.e. ANXA2, NDRG1 and STAT1) between S and R samples, their absolute quantification was determined using the QuantStudio3D digital PCR system (Thermo Fisher Scientific). cDNA was synthesized from 0.5 μg of total RNA by the use of iScript cDNA synthesis kit (Bio-Rad, CA, USA) and a total amount of 0.625 ng cDNA was then amplified in a 16 μl PCR volume, containing 0.8 μl of TaqMan assay (Thermo Fisher Scientific, Additional file 1: Table S3) for the genes of interest, and 8 μl of 2x QuantStudio 3D Master mix (Thermo Fisher Scientific). Data were analyzed with the QuantStudio 3D software (Thermo Fisher Scientific). Poisson distribution was used to estimate the average number of copies per reaction microliters. For each gene, the mean copy number was calculated in S and R populations.
RT-qPCR
Finally, results were verified also with conventional RT-qPCR, more suitable for future clinical application, on 15 S and 13 R, as previously described [18], using CFX Connect Real Time PCR Detection System (Bio-Rad), according to manufacturer’s instructions. Primers used to perform qPCR were the same of the nanofluidic PCR. All samples were amplified in triplicate and normalized to the housekeeping gene, B2M. The mean of threshold cycles (Ct, take-off point of reactions) for each specimen was used to obtain the fold change of gene expression level according to the -ΔΔCt method [19].
Bioinformatics
Identified proteins were uploaded into the STRING database v. 11.0 (
https://string-db.org/
), and protein-protein interaction networks were generated based on 10 interactors in both the first and second shell, highlighting the molecular actions when the interaction score was > 0.400.
Cell culture and transfections
CaSki and C-4I (European Collection of Authenticated Cell Cultures, ECACC) were purchased from Sigma-Aldrich (Darmstadt, Germany) and maintained in the specific medium supplemented with 10% v/v FBS, 1% w/v kanamycin, 1% w/v glutamine and 1% w/v MEM (Sigma-Aldrich) in a humidified incubator containing 5% CO2, at 37 °C. Cells were routinely tested for absence of mycoplasma with MycoAlert kit (LONZA, 169 Rockland, ME, USA). Predesigned SMARTpool siRNAs targeting ANXA2, NDRG1 and STAT1 and non-targeting control siRNA (siC) were purchased from Dharmacon (Lafayette, CO, USA). TransFectin lipid reagent (Bio-Rad) was used for transfection experiments as suggested by the supplier.
Ionizing radiation and cisplatin treatments
All irradiations of cells were performed with an IBL 437C γ-irradiator (Schering, Gir-Sur-Yvette Cedex, France) provided with a 137Cs source and a dose rate of 2.05 Gy/min. Non-transfected cells or cells transfected with control siRNA or targeting siRNA were irradiated in small Petri dishes. Cisplatin (Sigma-Aldrich) was dissolved and stored as a stock solution (10 mM) at − 20 °C.
Clonogenic assay
For the clonogenic assay, cells were irradiated in the dose range 0–6 Gy and/or treated with different cisplatin concentrations. Cells (2000–6000/dish for C-4I and 250–750/dish for CaSki cell lines, according to the radiation dose) were plated in Petri dishes 24 h before IR or cisplatin treatment. Ten to 14 days after IR, surviving colonies with more than 50 cells were counted after fixation with ice cold methanol and staining with 0.5% w/v crystal violet. Normalization to untreated control in each condition allowed to calculate the plating efficiency (PE), defined as the number of colonies counted/number of cells plated × 100 [20]. The surviving percentage was expressed as [n° of colonies in treated sample/(n° of plated cells × PE /100)] × 100. The dose that inhibited 50% of the colony-forming ability compared with the untreated control (IC50), was calculated to compare radiosensitivity between cell lines. A dose-response curve was also generated for cisplatin treatment and the corresponding IC50 value was determined for each cell line. For the combined treatments, the cisplatin dose that inhibited 30% of the colony-forming ability (IC30) was added to the plates immediately after γ-irradiation.
Cell cycle analysis by flow cytometry
Fluorescence flow cytometry was used to analyse alterations in cell cycle profiles 24 and 48 h after 2 Gy IR in non-silenced and silenced cells. At the end of each incubation period, adherent cells were trypsinized, harvested and washed several times with cold PBS. Cells were then counted, gently fixed in 70% v/v cold ethanol, and incubated at − 20 °C for no longer than 7 days. Prior to DNA staining, fixed cells were spun down and treated with RNase (100 μg/ml) for 10 min to ensure that only DNA was stained. Then, 1 × 106/ml cells were stained with propidium iodide (0.5 mg/ml) and stored at 4 °C overnight. The day after, stained cells were subjected to flow cytometry analysis, which was performed using a 6-parameters (2 scatter and 4 fluorescence signals) flow cytometer EPICS-XL (Beckman Coulter, Brea, USA). A minimum of 30,000 cells of interest were acquired for each sample, at a low flow rate (< 200 events/sec). Analysis of cell cycle perturbation was performed with the ModFit LT software (Verity Software House, Bury St. Edmunds, UK). Pulse shape processing was used to exclude cell doublets from the analysis.
Intracellular ROS measurements
Intracellular ROS accumulation was measured with DCFH-DA assay. Non-silenced and silenced cells (24 h after silencing) were plated in 96 wells black plates (1 × 104 cells/well). Intracellular ROS generation was determined at the specified interval following IR using 2′,7′- dichlorofluorescein diacetate (DCFH-DA) (Sigma-Aldrich). Fluorescence was determined at 495 nm excitation and 530 nm emission using the Enspire multimode plate reader (PerkinElmer, Walthman, MA, USA). NAC (N-acetyl-L-cysteine, 100 μM) was added 30 min prior to the irradiation in both CaSki and C-4I cells, as ROS inhibitor, to confirm assay specificity.
Fluorescence microscopy
Cells were seeded in 6-well plates containing coverslip in complete growth medium and, after treatment, fixed in 4% paraformaldehyde for 20 min at 20 °C, and permeabilized in 0.5% v/v Triton X-100 in PBS for 10 min, prior to be blocked with 5% v/v goat serum and 0.1% v/v Triton X-100 in PBS for 1 h. For ANXA2, permeabilization and blocking were performed with 0.2% w/v saponin in presence of 5% v/v goat serum and 0.2% w/v BSA in PBS, for 1 h. Immunofluorescence staining was obtained using anti-annexin A2 (1:125, ab41803, Abcam, Cambridge, UK), anti-NDRG1 (1:50, HPA006881, Sigma-Aldrich), anti-PARP1 (1:800, 46D11, Cell Signaling Technology, Danvers, MA), anti-STAT1 (1:100, HPA000931, Sigma-Aldrich) and anti-α-tubulin (1:200; clone DM1A, Merck Millipore, MA, USA), following overnight incubation at 4 °C. After washing, cells were incubated with secondary antibody anti-mouse Alexa Fluor-488 conjugate or anti-rabbit Alexa Fluor-488 and Alexa Fluor-567 conjugate (1:200) (Thermo Fisher Scientific), in the dark for 30 min at 20 °C. Coverslip was mounted onto slides using an antifade mounting reagent containing DAPI. Slides were observed under a fluorescence microscope (Leica Biosystems, Newcastle, UK) using a 100X oil immersion objective.
Western blot
Western blot analysis of total cell lysates was performed as previously described [21]. In detail, total cellular proteins were obtained lysing the cells with RIPA buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% w/v Nonidet P-40, 0.5% w/v Na-deoxicholate, 0.1% w/v SDS, 1 mM EDTA) in the presence of proteases and phosphatase inhibitors. Equal amounts of protein were separated by SDS polyacrylamide gel electrophoresis, blotted to PVDF and transferred using the Trans-Blot Turbo Transfer System (Bio-Rad) with 25 V, 1.0 A, for 30 min. After blocking in 5% non-fat milk (Biorad), membranes were probed with the following primary antibodies: anti-annexin A2 (C10, Santa Cruz Biotechnology, CA, USA); anti-BAX (clone N-20, Santa Cruz); anti-Bcl2 (NCL-bcl-2-486, Leica); anti-cleaved caspase-3 (5A1E, Cell Signaling Technology); anti-cyclin B1 (Y106, Abcam); anti-NDRG1 (HPA006881, Sigma-Aldrich); anti-p21WAF1 (AB1, Merck Millipore); anti-p53 (DO-7, Dako, Glostrup, Denmark); anti-PARP1 (46D11, Cell Signaling Technology); anti-STAT1 (HPA000931 Sigma-Aldrich); anti- β-actin (1:5000, A5441, Sigma-Aldrich); anti-HSP70 (H5147, Sigma-Aldrich), at 4 °C, overnight. After incubation with secondary horseradish peroxidase-conjugated antibodies (Bio-Rad), specific proteins were visualized by the enhanced chemiluminescence system (Amersham Biosciences, Buckinghamshire, UK) using a VersaDoc imaging system (Bio-Rad).
Statistical analysis
Means and standard error of the mean (SEM) were calculated for all data points, from at least three different independent experiments. Clonogenic survival curves were analyzed with v6.0 GraphPad through a Linear-Quadratic (LQ) model, historically used to describe radiation-induced clonogenic cell death [22]. The mathematical formulation of the LQ-model is given by:
$$ SF(D)/{SF}_0=100\ast {e}^{-\alpha \ast D-\beta \ast {D}^2} $$
(Eq. 1)
where α is the coefficient for the linear dose term, β is the coefficient of quadratic component of the survival curve and SF0 is the plating efficiency (the surviving fraction at Dose = 0 Gy). Linear-quadratic curves were obtained by applying the least-squares fitting method. For all pair-wise comparisons, F test was performed to evaluate if the variances were significantly different. Subsequently, unpaired Student’s t-test (or unpaired t-test with Welch’s correction if unequal variances) was used to analyze and compare the means. Paired t-test was used to compare experimental results over different independent experiments, when applicable. A statistically significant difference was considered when P < 0.05.
The dataset collected through qPCR analysis of ANXA2, NDRG1 and STAT1 genes was also used as training dataset to implement and optimize a Random Forest (RF) method [23] to classify two groups of patients (i.e. 28 samples: 15 Sensitive and 13 Resistant) according to their response to therapy. The software environments in which RF was trained and tested are R (
https://www.r-project.org/
) and Orange (
http://orange.biolab.si/
), both open source software tools for statistical analysis of data.