As patterns of gene expression correlate with disease phenotype and patient outcome, mRNA expression profiling has been used to classify disease risks as well as prediction of outcome[1–4]. In addition, survival analysis with gene expression profiles is beneficial for identifying new prognostic targets of diverse diseases.

Many miRNAs have been also found to be correlated with clinical outcome in specific cancer types [5–10]. miRNAs are a class of small and endogenous RNA molecules that regulate their target mRNAs through translational repression or mRNA degradation [11]. In tumors, many miRNAs can be aberrantly expressed, leading to potentially abnormal regulation of their target mRNAs. Although over 1,000 human miRNAs may be encoded in human genome, the potential therapeutic markers provided by miRNAs for a diverse spectrum of diseases are still unexplored.

Recently, several investigations have put emphasis on the integrated analysis of miRNAs and mRNAs in clinical outcomes [12–17]. In general, there are different approaches for the joint analysis of miRNA and mRNA data. For example, several miRNAs associated with survival rate can be extracted by survival analysis and then their relationship of inverse correlation can be identified based on analyzing miRNA and mRNA expression profiles. Most approaches do not test the clinical outcome by considering miRNA expression and mRNAs expression simultaneously. At minimum, this analysis requires the size of the cohort to be large enough for statistically significant measurement outcomes and the paired samples with both mRNA and miRNA expressions should be given. The Cancer Genome Atlas (TCGA) [18] provides different types of genomic datasets and we systematically integrated multi-omics data for cancer clinical outcome prediction [19].

Here, we present the survival analysis considering the regulatory relationships of miRNAs and their target genes. We tested clinical outcomes with the patient survival information, miRNA expression profiles and gene expression profiles for ovarian cancer and glioblastoma multiforme (GBM) through the integrated analysis of miRNA-mRNA interaction pairs. We found the miRNA-mRNA pairs with an inversely correlated expression profile that have significant survival differences between two patient groups. The results presented here suggest that this analysis can help to discover expression signatures of miRNAs and their target mRNAs that can be employed for therapeutics in human cancers.

Therefore, we extracted miRNA-mRNA interaction pairs that showed significant difference between the HL group and LH group in survival rate using more stringent criteria. In Figure 2(A), we plotted the distribution of p-values corresponding to survival tests for 10,574 validated pairs in the ovarian cancer dataset. Each test was conducted from two groups containing different patients based on the expression values of miRNA and mRNA (Figure 2(B)). We obtained the threshold (corresponding to p-value < 3.66e^-3) by calculating the false discovery rate (0.01) from a null model with a total of 776,016 pairs including most non-target relations. A total of 163 significant pairs with validated interactions were selected and they contained 22 miRNAs and 156 mRNAs. Expression patterns of all pairs are shown in Figure 2(C). The list of all significant interacting pairs for ovarian cancer is described in Additional file 1 Supplemental Table 1 and the list for GBM is also shown in Additional file 1 Supplemental Table 2. The survival difference between HL group and LH group suggests that the selected miRNA and its target mRNA might affect the progression of cancer in a coordinated fashion.

Among top 20 ranked miRNA-mRNA interaction pairs, the roles of four targets including PLAGL1, MTUS1, MEF and IFNGR1 have been reported in ovarian cancer (Table 2). miR-148a was down-regulated in ovarian cancer cell lines and might be involved in the carcinogenesis of ovarian cancer [21, 22]. Previous research reported that overexpression of miR-148b in ovarian cancer tissues was not associated with any of the pathological features of patients with ovarian cancer. It suggested that miR-148b might be involved in the early stage of ovarian carcinogenesis [23]. Although other miRNAs including miR-196a, miR-374b, miR-124, and miR-98 have not been reported for associations with ovarian cancer, they are related with the oncogenic phenotype or their expression of other cancers [24–27].

Several miRNAs are dominated in significant miRNA-mRNA pairs. miR-98 and miR-148b-3p have more than 446 and 187 target genes, respectively. miR-98 or miR-148b-3p itself shows a significant survival difference between high and low expression (pvalue < 0.0077 and pvalue < 0.0014), while miR-124-3p with 979 targets genes shows a borderline significance (pvalue < 0.0051).

LOT1(PLAGL1/ZAC1) is known to possess anti-proliferative effects and is frequently silenced in ovarian cancer and breast cancer [28]. Previous studies suggest that a shortage of the PLAGL1 protein might impair its role in regulating the cell cycle and interfere with apoptosis. Consequently, cells may grow and divide too quickly in an uncontrolled manner. Mitochondrial tumor suppressor 1 (MTUS1) is a newly identified candidate tumor suppressor gene [29]. Previous studies have shown that MTUS1 expression levels are down-regulated in cancers of the colon, ovary, pancreas, head and neck, and breast cancer [30]. MEF (myeloid ELF1-like factor, also known as ELF4) is expressed in a significant proportion of ovarian carcinomas [31]. The oncogenic activity of MEF was shown by the ability of MEF to transform NIH3T3 cells, and induce the formation of tumors in nude mice. The expression level of IFNGR1 in a typical ovarian cancer population varies, with 22% of them displaying a complete loss of the IFNγ receptor [32]. Low levels of receptor expression seem to have a negative effect on survival and are unrelated to other pathologic variables. Therefore, low expression of IFNGR1 could be regarded as an independent prognostic marker in ovarian cancer. Although we have found only several previously reported functional roles related to ovarian cancer, they hint at the possibility that other target genes might be associated with ovarian cancer development and progression.

Previous quantitative real-time PCR revealed overexpression of CDKN1A in primary GBM[38]. This result suggests that CDKN1A expression is regarded as a putative marker to distinguish primary GBM from secondary GBM. The gene for methylthioadenosine phosphorylase (MTAP) is located closely to the gene CDKN2A. MTAP-deficiency in many tumors that have been most resistant to treatment occurs commonly. Especially 70% of glioblastoma lack MTAP. Amplification of KIT in 17 (4.4%) glioblastomas was reveled from screening of 390 glioblastomas[40]. A borderline positive association (p=0.0579) between KIT amplification and TP53 mutation was also observed. ATM expression is correlated with radioresistance in primary GBM cells in culture. Genes encoding components of the DNA-damage response (DDR) pathway are frequently altered in human GBM patients and the ATM/Chk2/p53 cascade suppresses GBM formation[41].

Figure 3 displays the network of miRNA-target mRNA interactions that are associated with the difference of clinical outcome of OV cancer patients. It shows that many target genes in the network, indicated by highlighted nodes, are associated with neoplasm or cancer-related conditions. Some are interaction pairs where HL groups show good outcomes while others are when LH groups have good outcome. This network implies that the increased expression of miR-148b is significantly associated with survival rate for ovarian cancer, which agrees with the previous observation [23]. In contrast, the low level of expression of miR-124 is shown to have strong association with survival rate. The complete network of interactions can be found in Additional file 1 Supplementary Figure 1.

We have introduced the integrated analysis of survival test with datasets of miRNA and mRNA expression profiles, as well as clinical information in ovarian cancer and GBM. We have seen that the combined expression patterns between miRNAs and mRNAs can distinguish between risk groups related to co-regulation of both. In addition, we have presented supporting evidence for functional roles of miRNA and their targets in specific cancer from the literature. Our approach can be utilized to detect clinical and therapeutic miRNAs and their targets related to outcome of several cancers.