Article Text

PDF

Survival prediction of gastric cancer by a seven-microRNA signature
  1. Xiaohua Li,
  2. Ying Zhang,
  3. Yafei Zhang,
  4. Jie Ding,
  5. Kaichun Wu,
  6. Daiming Fan
  1. State Key Laboratory of Cancer Biology & Xijing Hospital of Digestive Diseases, The Fourth Military Medical University, Xi'an, China
  1. Correspondence to Prof Daiming Fan, State Key Laboratory of Cancer Biology & Xijing Hospital of Digestive Diseases, The Fourth Military Medical University, 17 Changle Western Road, Xi'an 710032,China; fmmulxh{at}yahoo.com.cn

Abstract

Aims Several microarray studies have reported microRNA (miRNA) expression signatures that classify cancer patients into different prognostic groups. No study has evaluated the association between miRNA expression patterns and gastric cancer prognosis. In this study, we developed a seven-miRNA signature that is closely associated with survival of patients with gastric cancer.

Patients and methods MiRNA expression profile was analysed by real-time RT-PCR in 100 gastric cancer patients, which were randomly assigned to either the training set or the testing set. Cox proportional hazard regression and risk-score analysis were used to identify a stage-independent set of seven-miRNA signature in the training set that could classify patients with significantly different prognosis. This miRNA signature was further validated by the testing set and an independent cohort 60 patients.

Results We have identified a seven-miRNA signature (miR-10b, miR-21, miR-223, miR-338, let-7a, miR-30a-5p, miR-126) for overall survival (p=0.0009) and relapse-free survival (p=0.0005) of gastric cancer patients. Multivariate analysis shown that the risk signature was an independent predictor of overall survival (HR=3.046; 95% CI, 1.246 to 7.445, p=0.015) and relapse-free survival (HR=3.337; 95% CI, 1.298 to 8.580, p=0.012). Furthermore, the predictive value of this seven-miRNA signature was validated in the testing set of 50 patients and an independent set of 60 patients.

Conclusion Our seven-miRNA signature is closely associated with relapse-free and overall survival among patients with gastric cancer. The prognostic signature could be applicable to future decisions concerning treatment.

  • Gastric cancer
  • microRNA
  • prognosis prediction
  • cancer preventation
  • gastrointestinal neoplasia

Statistics from Altmetric.com

Gastric cancer is a heterogeneous disease. Even in patients with similar clinical and pathological features, the outcome varies: some are cured, whereas in others, the cancer recurs. Staging systems for gastric cancer that are based on clinical and pathological findings may have reached their limit of usefulness for predicting outcomes, but molecular methods add value.

MicroRNAs (miRNAs) are a wide class of small, noncoding RNAs that negatively regulate protein expression at the post-transcriptional level. MiRNAs can post-transcriptionally regulate the expression of hundreds of their target genes, thereby controlling a wide range of biological functions such as cellular proliferation, differentiation, and apoptosis.1 The expression of miRNAs was shown to be temporally and spatially regulated, whereas the disruption of their physiological expression patterns was associated with several examples of human tumourigenesis, suggesting that they may play a role as a novel class of oncogenes or tumour suppressor genes.2 Since one miRNA can regulate hundreds of downstream genes, the information gained from miRNA profiling may be complementary to that from the expression profiling of protein-coding genes.3 Recently, miRNA expression profiles have been shown to be potential tools for cancer diagnosis and prognosis. Several miRNAs were reported to be associated with the clinical outcome of chronic lymphocytic leukaemia,2 lung adenocarcinoma,3 4 breast cancer5 and pancreas cancers.6 7 However, whether a miRNA signature can predict clinical outcome of gastric cancer, including major histological or stage subgroups of gastric cancer, is still not known.

Here, based on the miRNA expression profile and risk score analysis, we identified seven-miRNA formula significantly correlated with clinical outcome in a set of patients with gastric cancer. The seven-miRNA signature was an independent predictor of relapse-free and overall survival. We also validated the signature with data from an independent cohort of 60 patients with gastric cancer.

Materials and methods

Patients and tissue specimens

For the analysis of whether the miRNAs expression signature could predict clinical outcomes of gastric cancer, 100 primary gastric cancer samples were randomly separated into training (n=50) and testing (n=50) datasets. There were no significant differences in clinicopathological features between the two sets (table 1). Formalin fixed paraffin-embedded specimens of gastric cancer and information about the patients were collected from Gastrointestinal Surgery in Xijing hospital, Xi'an, China. Primary gastric cancer in these patients was diagnosed and treated at Xijing hospital from 1990 to 1997. Total RNA, with efficient recovery of small RNAs, was isolated from 20 μm sections from formalin-fixed, paraffin-embedded tissue blocks, using the RecoverAll Total Nucleic Acid Isolation Kit (Ambion, Woodward, Austin, Texas, USA). We validated the miRNA signature using an independent cohort of 60 consecutive patients (table 1) who underwent surgical resection of gastric cancer at Tangdu hospital between 1995 and 2001. All patients did not perform any therapy before recruitment to this research. In addition, all patients received FAM (5-fluorouracil (5-FU), mitomycin (MMC), and adriamycin (ADR)). plan chemotherapy after surgery during our research. The protocols used in the study were approved by the Hospital's Protection of Human Subjects Committee.

Table 1

Clinical characteristic of the training, testing and validation cohorts

MicroRNA profiling and Taqman real-time PCR analysis

Stem-loop reverse transcription for mature miRNA was done as described previously.8 All reagents were obtained from Applied Biosystems (Foster City, California, USA). The TaqMan MicroRNA Human Array V.1.0 Kit (Early Access) was used to detect the miRNA expression profiling. Briefly, 5 ng of total RNA were reverse transcribed to cDNA with stem-loop primers and the Taqman MicroRNA Reverse Transcription kit. Quantitative real-time PCR was done using an Applied Biosystems 7500 Real-time PCR System and a Taqman 2× Universal PCR Master Mix. All PCR primers were from the Taqman MicroRNA Assays. The small nuclear U6 RNA was used as an internal control for normalisation and quantification the MicroRNAs expression as previously reported.9

Statistical analysis

The statistical analysis was carried out as previously described.3 Briefly, the expression level of each miRNAs was assessed. Hazard ratios (HRs) from univariate Cox regression analysis were used to determine which miRNAs were associated with death from any cause or recurrence of cancer. Protective miRNAs were defined as those associated with a HR for death of less than one; risk miRNAs were defined as those associated with a HR for death of more than one. Overall survival time was calculated from the date of diagnosis until death or the last follow-up contact. To construct a model for the prediction of survival, univariate Cox proportional-hazards analysis was performed, with overall survival as the dependent variable.2

The special miRNAs related to survival were selected from the training set with the permutation test.3 Briefly, we used permutation p values for miRNA selection with 0.05 as the p value cut off. The hypothesis for permutation testing for Cox proportional hazard regression is that the survival time is independent of the expression level of that miRNA. The details are described as following: (1) For each miRNA, the p value is estimated by univariate Cox proportional hazard regression. This p value is called original p value. (2) The survival time and status (death or alive) are randomly assigned to each patient. The univariate Cox proportional hazard regression is performed to re-estimate the p value for each miRNA. (3) Step 2 is repeated 10 000 times to obtain the p value distribution for each miRNA. The proportion of p values smaller than the original p value is calculated. This gives the permutation p value. To avoid the effect of extreme values and set the number of patients in the two groups (high vs low risk signature) equal in the training dataset, the 50th percentile (median) was chosen as the cut off value. In the testing dataset, both the regression coefficients of risk score and the cut off value derived from the training dataset were applied directly. In order to further validate the performance of this risk formula, an independent cohort from a different hospital were used to validate the detected biomarkers. The Kaplan-Meier method was used to estimate overall survival and relapse-free survival. Differences in survival between high-risk and the low-risk patients were analysed using the two-sided log-rank test.

Furthermore, we used multivariate Cox regression analysis, stage stratification and histology stratification to investigate whether the micro-RNA signature is an independent predictor of overall survival and disease free survival in gastric cancer patients, especially whether it is independent of stage. The detail two statistical approaches were performed as previously described.3 All statistical analyses were conducted using SPSS V.11.05 software. Two-tailed tests and p values <0.05 for significance were used.

Results

The reproducibility of the real-time PCR

In order to set up a protocol for the preparation of total RNA including miRNAs from formalin-fixed, paraffin-embedded tissue blocks, we first examined the real-time PCR reproducibility of this protocol. A quantitive RT-PCR-based detection of specific miRNAs, which are known to be highly abundant in gastric cancer tissues (miR-17-5p, miR-2110, miR-2511) were used in our protocol. U6 RNA was also used for normalisation. The quantitive RT-PCR-based detection of four selected RNAs was precise with a mean intra-assay coefficient of variation of 19.72% (donor 1) and 8.41% (donor 2, see online table 1), respectively. The inter-assay coefficient of variation of three RNA-extraction and RNA-detection repeats averaged 11.95% (donor 1) and 32.92% (donor 2), indicating that the extraction and quantification of miRNAs from formalin-fixed, paraffin-embedded tissue blocks was robust and reproducible (see online table 1).

The seven-miRNA signature and survival

According to the permutation test from the training set, we found seven miRNAs are significantly associated with the patient survival by Cox proportional hazard regression in the training data set. HR from the univariate Cox regression analysis showed that the levels of expression of seven miRNAs correlated with death from any cause: three (let-7a, miR-126, miR-30a-5p) were protective miRNAs (associated with a HR of less than one) and four (miR-10b, miR-21, miR-223, miR-338) were risk miRNAs (associated with a HR of more than one). For example, the negative weighting value assigned to let-7a indicates that higher expression correlates with longer survival. The positive value for miR-21 indicates that higher expression correlates with shorter survival. The seven miRNAs regression coefficients were as follows: miR-10b, 0.32; miR-21, 0.25; miR-223, 0.14; miR-338, 0.19; let-7a, −0.22; miR-30a-5p, −0.15; and miR-126, −0.18. Then a patient's risk score was derived by a summation of each miRNA expression level times its corresponding coefficient as follows: risk score=(0.32×expression level of miR-10b)+(0.25×expression level of miR-21)+(0.14×expression level of miR-223)+(0.19×expression level of miR-338)−(0.22×expression level of let-7a)−(0.15×expression level of miR-30a-5p)−(0.18×expression level of miR-126).

The risk score was used to classify patients into high or low risk signature, in which a high risk score indicated a poorer survival for patients.2 According to this risk-score formula, the risk score of all patients in training set were calculated. Then, we ranked patients on the basis of their risk scores and divided them into a high-risk group or low-risk group using the median risk score as the cut off point. Table 1 lists the characteristics of the patients in the first analysis. Among the 50 patients in the training cohort, tumours with high-risk scores expressed risk miRNAs, whereas tumours with low-risk scores expressed protective miRNAs (figure 1A). Patients with a high-risk score signature had a lower median overall survival than those with a low-risk score signature (23 months vs not reached) (figure 1B). Tumours with a high-risk score signature were associated with a lower median relapse-free survival than tumours with a low-risk score signature (14 months vs not reached) (figure 1B).

Figure 1

The seven-miRNA signature and survival of 100 patients with gastric cancer. A shows the miRNA-expression profiles of the tumour specimens (according to the colour scale shown); each column represents an individual patient. The magnitude of the corresponding risk scores is represented by the slope of the red triangle. B, C show Kaplan–Meier estimates of survival according to the seven-miRNA signature in training (B) and testing sets (C).

After obtaining the risk score formula and choosing the 50th percentile cut off point from the training set, we then applied this risk score formula and cut off point to the testing set. Results in the testing set were similar to those in the training cohort. Among the 50 patients, tumours with high risk scores expressed risk miRNAs, whereas tumours with low risk scores expressed protective miRNAs (figure 1A). Patients with a high-risk score signature had a lower median overall survival than those with a low-risk score signature (22 months vs not reached) (figure 1C). Tumours with a high-risk score signature were associated with a lower median relapse-free survival than tumours with a low-risk score signature (16 months vs not reached) (figure 1C).

According to Cox multivariate regression analysis, the high-risk seven-miRNA signature, tumour stage III and IV were significantly associated with death from any cause among the training set (table 2), and the high-risk seven-miRNA signature, tumour stage III and IV were significantly associated with recurrence of cancer as well (the high-risk signature vs the low-risk signature, HR=3.337, p=0.012; stage III and IV vs stage I or II disease, HR=3.811, p=0.005). The similar results were also found in the testing set (table 2).

Table 2

HR for death from any cause among patients with gastric cancer, according to multivariate cox regression analysis

Validation of the seven-miRNA signature

In order to confirm the above findings, we used another independent cohort to examine the seven-miRNA signature. The clinical characteristics of the 60 patients in the validation cohort are also listed in table 1. Patients were classified as high-risk or low-risk groups based on their miRNA risk score signature. Patients with a high-risk score signature had a lower median overall survival than those with a low-risk score signature (20 months vs not reached) (figure 2A). Tumours with a high-risk score signature were associated with a lower median relapse-free survival than tumours with a low-risk score signature (15 months vs not reached) (figure 2B).

Figure 2

Kaplan-Meier estimates of overall survival and relapse-free survival of the validation cohort patients according to the seven-miRNA signature.

Seven-miRNA signature is independent from stage or histology

With the Cox multivariate regression analysis and stepwise variable selection, we want to test whether the seven-miRNA signature is an independent prognostic factor associated with patient survival in this independent cohort. The miRNA signature, age, sex, stage and histology were used as covariates. According to Cox multivariate regression analysis, the high-risk seven-miRNA signature (HR=2.625, p=0.037) and tumour stage (HR=3.35, p=0.0002) were significantly associated with death from any cause (table 2) and the prognostic ability of the miRNA signature is independent from stage or histology (table 2).

Seven-miRNA signature can predict patient survival within cancer stages, histological subgroups and Lauren classification

We further used the combination of the testing and independent cohort to examine whether the seven-miRNA signature can distinguish high-risk versus low-risk groups of patients within each stage stratum (gastric cancer stage I, II, III or IV) and to predict their survival.

We found that among patients with four stages of gastric cancer, the high-risk survival curve all lies below the low-risk curve (figure 3). For stage I, II and IV disease, the miRNA signature is significantly associated with the overall survival and relapse-free survival of gastric cancer patients (p=0.0231, p=0.0368; p=0.0028, p=0.0124; p=0.0146, p=0.0086 respectively). Among patients with stage III disease, the overall survival and relapse-free survival did not differ significantly between those with high-risk and those with low-risk miRNA signatures (p=0.3227, p=0.1323), probably owing to the small number of patients. To reach the overall conclusion that the survival prediction of miRNA signature is independent from stage, we conducted the overall χ2 that combines the four log-rank tests together. The p value is less than 0.005 for the overall survival and for the relapse-free survival (see online table 2).

Figure 3

Kaplan-Meier estimates of overall survival and relapse-free survival according to the seven-miRNA signature differential stages of gastric cancer patients in the combination of the testing and independent sets. (A) Stage I (n=31). (B) Stage II (n=27). (C) Stage III (n=22). (D) Stage IV (n=30).

Next, we stratified the gastric cancer patients by the histological subtype of differentiated or undifferentiated and Lauren classification of intestinal or diffuse. The miRNA signature can predict patient survival within each gastric cancer histology subtype and Lauren classification (figure 4). The overall χ2 test that combines the two log-rank tests together gives significance p values (p<0.005) for overall survival and relapse-free survival, respectively (see online table 2).

Figure 4

Kaplan-Meier estimates of overall survival and relapse-free survival according to the seven-miRNA signature differential histological types or Lauren classification of gastric cancer patients in the combination of the testing and independent sets. (A) Differentiated (n=53). (B) Undifferentiated (n=57). (C) Intestinal (n=50). (D) Diffuse (n=60).

Discussion

In this study, we identified an RT-PCR–based seven-miRNA signature using risk scores based on miRNA profile of 100 paraffin-embedded specimens from patients with gastric cancer. The presence of a high-risk seven-miRNA signature in the gastric cancer tumours was associated with an increased risk of recurrence and decreased overall survival, even after stratifying patients by stage, histology (differentiated or undifferentiated carcinoma) subgroups or Lauren classification. These results suggest that miRNAs may play an important role in the molecular pathogenesis, clinical cancer progression and prognosis of gastric cancer.

Our selection of miRNAs in the training set was not only validated in the testing set, but also in an independent cohort results were also validated in an independent cohort of 60 patients who were treated at the Xi'an TangDu Hospital. Thus, we believe that the data we obtained using the seven-miRNA signature are reliable. The identification of seven-miRNA signature that is closely associated with the outcomes in patients with gastric cancer has important clinical implications. Current therapy for patients with early stage gastric cancer usually consists of surgical resection without adjuvant treatment. Radiotherapy and/or chemotherapy are only effective in some patients with gastric cancer. We propose that patients who have tumours with a high-risk miRNA signature could benefit from this type of adjuvant therapy, whereas those with a low-risk miRNA signature could be spared what may be unnecessary treatment. Prospective, large-scale, multicentre studies are necessary to test this idea.

The identification of seven miRNAs that can predict the clinical outcome in patients with gastric cancer may reveal targets for the development of therapy for gastric cancer. Protective miRNA let-7a inhibits tumour cell proliferation and anchorage-independent growth through repression of the HMGA2 and RAS oncogenes.12 Recently studies also demonstrated that let-7a expression is associated with prolonged survival in non-small cell lung cancer patients.13 Another protective miRNA is miR-126, which inhibits neoplastic cell growth, invasion and metastasis via downregulating phosphatidylinositol 3-kinase, IRS-1, HOXA9.14–16 Regarding the potential role of the four risky miRNAs in this miRNA signature, miR-10b was reported to act as an oncogene in breast cancer that promotes proliferation and metastasis by downregulating the tumour suppressor HOXD10.17 It has been reported that miR-21 also acts as an oncogene that promotes proliferation, invasion and metastasis, and inhibits apoptosis through downregulating tumour suppressor PDCD4 and tropomyosin 1.18 19 Overexpressed miR-21 is associated with shorten survival in breast cancer, lung cancer and gastric cancer patients.20–22 Also, miR-223 was reported to act as an oncogene in gastric cancer.10 However, a recent report suggested that miR-223 directly targets tumour suppressor Stathmin1 to inhibit hepatocacinoma development and progression.23 It is well known that one miRNA can regulate many targets. Therefore, it may be possible that the same miRNA may play opposite roles in cancer progression, both as a tumour suppressor in certain cancers and as an oncogene in others. To our best knowledge, the molecular mechanism of miR-338, miR-126 and miR-30a-5p in cancer biology has not been reported yet.

In conclusion, the seven-miRNA signature we identified is closely associated with the clinical outcome in patients with surgically resected gastric cancer. This signature could be useful in stratifying patients according to risk in trials of adjuvant treatment of the disease.

References

View Abstract

Footnotes

  • Competing interests None.

  • Ethics approval This study was conducted with the approval of The protocols used in the study were approved by the Hospital's Protection of Human Subjects Committee.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles

  • Digest
    Emad El-Omar Severine Vermeire Alexander Gerbes