Study design and patient samples
Written informed consent, was obtained from all patients. It included the permission to use blood for research purposes. The study was approved by institutional review boards and the hospital Clinical Research Ethics Committee. All patients were sporadic cases on the basis of family history of NPC. All patients first underwent neoadjuvant chemotherapy as described previously [6]. The treatment consisted of neoadjuvant chemotherapy and concurrent radiotherapy. Chemotherapy was made of 2 cycles of 5- fluoroucilat 700mg/m2/a day, performed on day 1 and day 4 (intraveneous injection). In addition, on day 1 after fluoroucilat, nedaplatin was infused (100mg/m2 over 2 hours). Nedaplatin treatment was repeated every 3 weeks. As for concurrent intensity modulated radiotherapy, it was performed with Nedaplatin which was given 60 minutes before radiation (100mg/m2) on days 1, 22 and 43. Radiotherapy consisted of external-beam radiotherapy to the nasopharynx (70–80 Gy), the lymph node–positive area (60–70 Gy), and the lymph node–negative area (50–60 Gy). Tumours were staged according to the 1997 American Joint Committee on Cancer (AJCC) Staging system.
Inclusion Criteria: Patients were included when they were between 40 and 70; radiotherapy was indicated for them; they had no endocrinologic or metabolic disorders and no uncontrolled hypertension or infections. They needed to have a normal liver, heart and kidney function.
Exclusion Criteria:Patients could not be included when they were intolerant to radiotherapy;they had not finished a prescribed treatment;they were unable to accept treatment.
This study was divided into two phases:
Phase I (Marker discovery): In this phase, 6 patients with loco-regionally advanced NPC underwent intensity-modulated radiotherapy (IMRT) concurrently performed with induction chemotherapy based on nedaplatin (NDP) and 5-FU. Serum samples were collected within the week before radiotherapy initiation and three months after radiotherapy completion. Among these 6 patients, 3 experienced recurrence or distant metastasis in the 5 years following chemoradiation completion,who belonged to the experimental group (EG);while the other 3 NPC patients experienced complete clinical response5 years after chemoradiation,who belonged to the control group (CG). Differences in miRNA profiles between EG and CG groups collected before and after chemoradiotherapy were assessed in serum samples. Two miRNA expression patterns were established by comparing miRNA profiles in these two groups using miRCURY LNA™ miRNA Arrays. The most frequently down-regulated miRNAs in CG group compared with EG group in both time points were identified by bioinformatics and used for further analysis in phase II.
Phase II (Marker selection and validation): The down-regulated miRNAs identified above were considered as possible candidates. Two batches of serum samples were collected from an independent cohort of 48 NPC patients within 7 days before the start of radiotherapy and three months after radiotherapy completion.29 NPC patients with recurrence or distant metastasis and 19 NPC patients in clinical remission within 5 years after CCRT composed the cohort. Putative miRNA markers identified in phase I were verified in these independent sets of serum samples using real-time quantitative RT-PCR.
miRNA array analysis
Trizol LS reagent (Invitrogen, Paisley, UK) was used to extract RNA. miRNAs were generated from the total RNA groups mentioned above. miRCURY locked nucleic acid (LNA) microarray platform (Exiqon, Denmark) was used [7]. Total RNA was labeled Hy3™ or Hy5™ fluorophores, using miRCURY™ Array Power Labeling kit (Exiqon, Denmark). Then, the reaction was spinned and left at 4°C. The two samples from the Hy3™ and Hy5™ labeling reactions were combined in ice. The samples were hybridized in an hybridization station using miRCURY™ LNA miRNA Array (v.11.0) containing Tm–normalized probes for 847 human miRNAs.Microarrays with labeled samples were hybridized at 56°C overnight using a heat-shrunk hybridization bag and washed with miRCURY Array Wash buffer kit (Exiqon, Denmark).After hybridization, the chip slides were immediately washed, dried and scanned . Each miRNA spot was replicated four times on the same slide and two microarray chips were used for each group. Scanning was performed with the Axon GenePix 4000B microarray scanner. GenePix pro V6.0 was used to read the raw intensity of the image. Signal intensities for each spot were scanned to produce the best within-slide normalization and minimize the intensity-dependent differences between the dyes. Then, they were calculated by subtracting local background (based on the median intensity of the area surrounding each spot) from total intensities using locally weighted scatter plot smoothing (Lowess, Locally Weighted Scatter plot Smoothing) Normalization (MIDAS, TIGR Microarray Data Analysis System). After normalization, the average values of each miRNA spot were used for statistics. The ratio between the green and red signals was calculated. Fold change >2.00 and fold change <0.5 (adjusted p-value<0.50) were used to screen both up and down regulated miRNAs. Hierarchical clustering to differentiate expressed miRNAs was generated using standard correlation to measure their similarity.
Real-time quantitative PCR
Real-time PCR was done using GeneAmp Fast PCR Master Mix (Applied Biosystems) and ABI 7900HT real-time PCR machine. Table 1 summarizes the Oligonucleotides used in this study. To quantify the miRNAs expression in serum, U6 or miRNA-16 was respectively adopted as internal control. Appropriate internal normalization control was required to normalize sample-to-sample variations and relative quantification was applied in this qPCR. As no consensus on the use of internal normalization control in serum was defined for miRNA qPCR, we used expressions of miRNA-16 as the internal normalization control for miRNA quantification in serum. Although miRNA-16 turns less abundant than other miRNAs in the serum, miRNA-16 was selected as the normalization control as it proved higher stability and less variability. The threshold cycle (Ct) is defined as the fractional cycle number at which the fluorescence passes the fixed threshold. Each sample was run in triplicates for analysis.The relative amount of each miRNA was calculated using the equation 2‑ΔCt, in which ΔCT= (CTmiRNA-CTU6) or ΔCT= (CTmiRNA-CTmiRNA-16).
Analyzing miRNA targets and correlative article
The targeted mRNAs that have the potential binding sites for these miRNA to express in EG versus CG were searched in public databases endowed with prediction algorithms, such as TargetScan (http://targetscan.org), PicTar (http://pictar.mdc-berlin.de) and miRBase Target (http:// www.mirbase.org). The target genes we have chosen were significantly associated with different pathways. We also performed a PubMed search with various down-regulated miRNAs in our results from miRNA array analysis. Relevant publications dealing with miRNAs and their possible molecular mechanisms were obtained. Further, relevant articles were found by screening the references of these papers. In case of non-availability of the whole article, the abstract was taken into consideration despite the limited data provided. This procedure was conducted twice until the beginning of May 2019 to avoid any missed contribution.
Development of risk score
To analyse the data of miRNAs selection, we performed a multivariate Cox regression analysis using a backward stepwise approach to test if the signature was an independent prognostic factor of PFS and OS. The expression level of miRNA-26b, miRNA-29a, miRNA-125b miRNA-29b and miRNA-143 in NPC patients serum before and after treatment were used as candidate variates.PFS was used as dependent variable, p<0.15 was selected as variable standard, and p>0.2 was removed out of variable standard. We developed a formula to calculate every patient’s PFS and OS risk scores from the expression values of the five miRNAs, weighted by regression coefficient.
For PFS, Risk score = (0·792 × expression value of pre-therapy miRNA-29a) –(0·779× expression value of pre-therapy miRNA-125b) + (0·242 ×expression value of post-therapy miRNA-26b) . For OS: Risk score = (0·431× expression value of pre-therapy miRNA-29a) –(0·538× expression value of pre-therapy miRNA-125b) .
Statistical analysis
Data about miRNA expression was analyzed using BRB-Array Tools version 3.5.0 software (Richard Simon & BRB-ArrayTools Development Team), TIGR Multi-experiment viewer version 4.0 software (The Institute for Genomic Research, and Array Assist software, Stratagene), R software version 2.15.2 (The R Foundation for Statistical Computing c/o Institute for Statistics and Mathematics) and GraphPad software 5.0 (GraphPad Software Inc., San Diego, CA, USA). Hierarchical clustering (Manhattan distance and average linkage) and principal component analysis (PCA) were used to classify samples. Significance analysis of microarrays (SAM) and Student’s t-test based on multivariate permutation (with random variance model) were performed to identify miRNAs with significant differential expression between groups. In the validation phases, the expression levels of individual miRNAs in the different groups were compared using paired or unpaired t-test or ANOVA analysis for continuous variables. The Spearman rank order correlation test was used to examine correlation relationships between the levels of the miRNA markers. Progression-free survival (PFS) was calculated from the date of the start of a therapy to the documentation of PD according to the RECIST criteria. Overall survival (OS) was
defined as the interval between the date when therapy started and the date of the last follow-up or death from any cause. The survival rate was calculated using the Kaplan–Meier method. Univariable and multivariable Cox proportional hazards models including age, gender, tumour stage (according to the AJCC TNM classification), pathological type, expression level of ki-67 and miRNA were used to identify factors significantly associated with prognosis. Results were reported including hazard ratios (HRs) and 95% confidence intervals (CIs). The data were regarded as significantly different at P < 0.05.All statistical calculations were performed by the SPSS software(version 18.0,SPSS Inc., Chicago, IL, USA).