A 3‐miRNA signature predicts survival of patients with hypopharyngeal squamous cell carcinoma after post‐operative radiotherapy

Abstract Since the prognosis of hypopharyngeal squamous cell carcinoma (HSCC) remains poor, identification of miRNA as a potential prognostic biomarker for HSCC may help improve personalized therapy. In the 2 cohorts with a total of 511 patients with HSCC (discovery: N = 372 and validation: N = 139) after post‐operative radiotherapy, we used miRNA microarray and qRT‐PCR to screen out the significant miRNAs which might predict survival. Associations of miRNAs and the signature score of these miRNAs with survival were performed by Kaplan‐Meier survival analysis and multivariate Cox hazard model. Among 9 candidate, miRNAs, miR‐200a‐3p, miR‐30b‐5p, miR‐3161, miR‐3605‐5p, miR‐378b and miR‐4451 were up‐regulated, while miR‐200c‐3p, miR‐429 and miR‐4701 were down‐regulated after validation. Moreover, the patients with high expression of miR‐200a‐3p, miR‐30b‐5p and miR‐4451 had significantly worse overall survival (OS) and disease‐specific survival (DSS) than did those with low expression (log‐rank P < .05). Patients with a high‐risk score had significant worse OS and DSS than those with low‐risk score. Finally, after adjusting for other important prognostic confounders, patients with high expression of miR‐200a‐3p, miR‐30b‐5p and miR‐4451 had significantly high risk of overall death and death owing to HSCC and patients with a high‐risk score has approximately 2‐fold increased risk in overall death and death owing to HSCC compared with those with a low‐risk score. These findings indicated that the 3‐miRNA‐based signature may be a novel independent prognostic biomarker for patients given surgery and post‐operative radiotherapy, supporting that these miRNAs may jointly predict survival of HSCC.


| INTRODUC TI ON
Hypopharyngeal squamous cell carcinoma (HSCC) is an aggressive malignancy among head and neck squamous cell cancers. It accounts for approximately 4% of the total cases of head and neck cancers. 1 Although the incidence is approximately 0.7-0.95 per million personyears with a declining trend over the past 4 decades, 1-3 an estimated 3000 new cases are diagnosed with HSCC annually in the United States. 4 The combined treatments of surgery, radiotherapy and chemotherapy have been applied, while the overall prognosis of patients with HSCC remains poor because of the high rate of local recurrence and nodal metastasis. The average 5-year overall survival has improved only from 37.5% to 41.3% between before 1990 and after 1990 according to the SEER data. 5 Many proteins have been selected for prognostic biomarkers of HSCC, such as the truncating mutation of TP53, CD105 expression, CD98, and Oct4 and osteopontin in tumours, 6 and however, the proteins or genome DNA has the tendency of degradation, which limits their clinical application. Thus, a molecule with more stable characteristics is more suitable to be a candidate biomarker. miR-NAs are small non-coding RNAs with potent regulatory functions in cell growth, proliferation and apoptosis. 7 It has been demonstrated that expression levels of miRNAs which function as tumour suppressors and oncogenes are dysregulated in different cancers. 8 A growing body of evidence indicates that miRNA profiles are associated with particular cancers and could serve as potential biomarkers for tumour diagnosis and prognosis. 9 Unlike the proteins, miRNAs are quite stable in clinical archived samples such as serum samples or formalin-fixed paraffin-embedded (FFPE) tissue specimens. In addition, miRNAs could be detected and their degradation of miRNA expression was very minimal in various types of cancer tissue samples after a long-term storage. 10 Thus, human miRNAs from FFPE tissue specimens are highly stable, and FFPE tissue-based miRNAs hold great promise as ideal candidate biomarkers for diagnosis and prognosis of HSCC in clinical practice.
Our previous study showed that the expression level of miR-4451 in patients with HSCC was significantly associated with prognosis. 11 To improve the efficiency and accuracy of prediction of prognosis in HSCC, we developed a risk model based on these three-miRNA signature for improved prognostic prediction of HSCC. The 3-miRNA signature allows us to assess the combined effects of a panel of miRNA expression profiles that act in the same pathway. Evaluating such combined effects may amplify the associations of individual miRNA expression with the prognosis of HSCC.

| Patients and samples
The patients included in this study were diagnosed with HSCC pathologically and underwent surgery with post-operative radiotherapy

| qRT-PCR analysis
After high-throughput screening by microarray, the candidate miR-NAs were verified by qRT-PCR analysis. As previously described, 12 the 100 ng of total RNA of each sample was polyadenylated and reversely transcripted to cDNA. Then, a quantified PCR was carried out in the Qiagen miScript system (Qiagen) on the ABI 7900HT platform (ABI). The U48 was selected as endogenous control. The Cq value was generated by SDS software 2.4 (ABI). Through data processing by Thermo Fisher Cloud (https ://apps.therm ofish er.com/apps/dashb oard/), the Cq values were adjusted by interplate calibrator (IC), and the relative expression of candidate miRNA was calculated as 2^-ΔCq, where ΔCq = Cq (candidate miRNA) − Cq (endogenous control).

| miRNA signature score and survival analysis
The expression of each miRNA was categorized into two groups

| Patients' characteristics
According to inclusion and exclusion criteria, a total of 436 HSCC patients' medical records were reviewed in discovery cohort.
Among them, 408 patients received surgery and post-operative radiotherapy. Eleven patients were lost for follow-up, and 8 patients without correct diagnosis were excluded. In the remaining 389 patients who had FFPE tissues available, 8 cases had massive necrosis in samples and 9 cases had low concentration of RNA output for qRT-PCR. These patients were also excluded. Thus, a total of 372 patients were included in discovery cohort. Similarly, we used the same inclusion and exclusion criteria as in discovery cohort, and a total of 139 out of 157 patients with HSCC were included in the validation cohort.
The clinical and epidemiological variables were not significantly different between the two cohorts (

| Identification of candidate miRNAs by microarray
By miRNA microarray, differentially expressed miRNAs were selected under two criteria. A total of 14 miRNAs were found to match the 1st criteria: FC ≥ 1.5 and P-value ≤ .05 between the two groups.
Because miRNAs with higher miRBase ID might be artefacts, we excluded miR-5088-5p, miR-6808-5p and miR-6813-3p. In our preliminary experiment, because we found the PCR reactions for miR-3195 and miR-4688 generated unsatisfactory dissociation curve and segregated strips in electrophoresis, these two miRNAs were also excluded.
Finally, totally 9 miRNAs were selected as candidate miR-NAs as shown in Figure  and a high grade of T (T4) ( Table 2).

| Survival analysis on candidate miRNAs
As shown in  were associated with better OS and DSS, while the low expression of miR-30b-5p was only associated with better DSS, rather than OS.

| Association of candidate miRNAs with risk of OS
As shown in

| Effect of a signature score of miRNAs on risk of OS
The HRs of miR-200a-3p, miR-30b-5p or miR-4451 were all >1, so all the 3 miRNAs were risky to OS and DSS. As we described in the method, high expressions of miRNAs were defined as a value of 1, whereas low expression as 0. By summating the values of miRNA in this signature, a patient might get a score of 0-3. The higher score meant a patient expressed more risky miRNAs.
According to the risk signature score, we divided the patients into a low-risk group (score = 0 or 1) and a high-risk group (score = 2 or 3). In the discovery cohort, the average OS and DSS of patients with the low-risk/high-risk group were 67.8 months (95% shown in Figure 3, the patients with high-risk score had significantly worse OS and DSS than those with low-risk score (log-rank:  (Table 4). Similarly, as in the discovery cohort, after validation in the validation cohort, we found that the OS and DSS were shorter in high-risk group than in low-risk group, while the high score of miRNAs also increased risk of overall death and death due to the disease compared with the patients with a low-risk score.

| Diagnostic value of miRNA signature score for prognosis prediction
As shown in Figure 4 and Table 5

| D ISCUSS I ON
In this study, we have identified a three-miRNA signature that was significantly associated with OS and DSS of patients with HSCC undergoing surgery and radiotherapy. The high expression of each of these miRNAs may predict an increased risk of death, while the combined 3-miRNA-based signature is novel and more robust prognostic F I G U R E 4 ROC curve of 3 significant miRNA expression/signature score for predicting prognosis of OS and DSS biomarkers for patients with HSCC after post-operative radiotherapy, with a higher risk score having worse survival.
Squamous cell carcinoma in the hypopharyngeal region has progressive behaviour and poor prognosis. Prediction of prognosis of patients with HSCC is critical to develop novel strategy for the personalized cancer therapy. Currently, the traditional TNM system remains useful for the prediction of survival in patients with HSCC.
To improve the efficiency and accuracy of the TNM system, identification of novel prognostic molecular biomarkers may help improve the prediction besides the proper combination of T, N and M stages for classifying the patients into different prognostic groups. [13][14][15] miRNA has many advantages to act as biomarkers for tumour diagnosis and prognosis since it is more stable than protein, mRNA and ge- acted as a tumour suppressor in SCCHN. 11 MiR-99a was found to be down-regulated in SCCHN, especially in SCC of oral cavity, [18][19][20] subsequently contributing to the survival. 21 In SCCHN, miR-405 was found to inhibit tumour proliferation by targeting CDK6. 22 Wang found miR-203 could inhibit tumour growth and metastasis through PDPN. 23 MiR-15a was up-regulated in HPV-positive HSCC and might induce tumour apoptosis via BCL2L2 and BCL2. 24,25 However, many miRNAs are oncogenic by targeting tumour suppressor genes. miR-21 was widely accepted as 'oncomiR', which was overexpressed in various types of cancers including lymphoma, oesophageal cancer, LSCC, HSCC and SCC of oral cavity. 26,27 Similarly, these oncogenic miRNAs also included miR-16, miR-155, miR-130b and miR-184. 28,29 In our current study, we found that miR-200a-3p, miR-30b-5p and miR-4451 were up-regulated in patients with HSCC and caused poor prognosis, indicating an oncogenic role of these miRNAs in prognosis of HSCC.
In previous studies, miR-200a-3p was reported to be highly expressed in several types of tumours 30,31 and led to worse prognosis, 32 indicating that miR-200a-3p could be an early biomarker and a potential novel target for cancer therapeutic interventions. In vivo study, miR-200a-3p was found to promote cancer cell proliferation via targeting CRMP1 and inactivating tumour suppressor gene RHOA, 31 while miR-200a-3p acted as a tumour suppressor gene in renal cell carcinoma or hepatocellular carcinoma. 33,34 Various studies reported the oncogenic role of miR-30b-5p, which was found to be up-regulated in bladder cancer and medulloblastoma. 35, 36 Shao found that miR-30b-5p was correlated with advanced OSCC via increasing the copy number of miR-3b-5p. 37 Gaziel-Sovran reported that miR-30b/30d regulated the GalNAc transferase, enhancing invasion and immunosuppression of melanoma cells during TA B L E 5 AUG of ROC analysis for prognosis of OS and DSS metastasis. 38 The high expression of miR-30b-5p was significantly associated with poor prognosis of patients with glioblastoma through mediating PRRT2. 39 However, unlike the classic oncomiRs, miR-30b-5p was also reported as a tumour suppressor in oesophageal cancer, non-small-cell lung cancer or hepatocellular carcinoma. [40][41][42] Unlike miR-200a-3p and miR-30b-5p, which were widely studied in different cancers, miR-4451 was identified more recently, for which few studies were focused on its function or mechanisms in human cancer development and prognosis. Our previous study showed miR-4451 was highly expressed in patients with HSCC, and high expression of miR-4451 was associated with a shorter survival of patient with HSCC, 12 while the molecular mechanisms behind the association of miR-4451 high expression with worse survival need more in-depth studies.
A miRNA signature-based method as a tool has been widely used to predict cancer risk and prognosis. As each of these miR-NAs appeared to have a minor or moderate effect on OS and DSS, the combination of these miRNAs into a signature may more efficiently predict cancer outcome. Such a miRNA signature-based classifier is of more powerful for the prediction of prognosis or early diagnosis. In our current study, we identified a novel classifier based on a 3-miRNA signature (including miR-200a-3p, miR-30b-5p and miR-4451), which can more accurately predict the prognosis of HSCC. The patients with high-risk score had worse prognosis. Thus, a miRNA signaturebased method may be used to evaluate the collective effects of these miRNAs on the risk of death overall in patients with HSCC.
Although significant association of these miRNAs with prognosis was found, future bioinformatics analyses or in in vivo and in vitro experiments are needed to develop further mechanisms underlying the signature.
In conclusion, our study identified a 3-miRNA signature as potential independent prognostic predictor for patients with HSCC.
Future further research and functional study are needed to explore the underlying mechanisms of these significant miRNAs in signature.

ACK N OWLED G EM ENTS
We thank technicians Wenming Jia, Shuai Chen, Qian Sun and Juan Zhao for laboratory procedures, Dr Xiao Wang and Dr Chunyan Zhai for pathologic grading, technician Hainan Ren and Shunxue Hu for FFPE sample preparations and Xiubin Sun for statistical analysis.

AUTH O R CO NTR I B UTI O N S
XX and ZL performed experiments and wrote the manuscript. FZ, DL and XP assisted with the study design, data analysis and editing of the manuscript. NG and GL helped with editing of the manuscript.