Potential role of miR-9 and miR-223 in recurrent ovarian cancer

Ovarian cancer is the leading cause of death from gynaecological malignancy in the western world [1]. Ovarian serous adenocarcinomas (OSC) are the commonest histotype and account for almost 50% of malignant neoplasms [2]. The majority of cases present in advanced stages and are treated with surgery and systemic chemotherapy. Current treatment is frequently followed by recurrence, which is often resistant to chemotherapy, as demonstrated by 15% long-term survivors [3]. Although focusing on known genes has already yielded new information, previously unknown noncoding RNAs, such as microRNAs (miRNAs), may also lend insight into the biology of ovarian cancer.

This new and surprisingly abundant class of RNA regulatory genes has been found to confer a novel layer of genetic regulation in cells. Active, mature miRNAs function as endogenous, highly conserved, small RNA's, 22 nucleotides long that silence gene expression by binding to target mRNAs. Their 5' end binds to its target complementary sequence in the 3'-untranslated region (3'UTR) of mRNA and given the degree of complementarity, miRNA binding appears to result in translational repression, or in some cases, cleavage of cognate mRNAs, causing partial or full silencing of the respective protein-coding genes [4]. An accumulating body of evidence reveals critical functions for miRNAs in various biological processes as diverse as proliferation, apoptosis, and cell differentiation [5] and given their diversity and abundance, miRNAs appear to functionally interact with various components of many cellular networks. Almost 30% of the human genome is estimated to be regulated by miRNAs [6]. Therefore, they should be considered one of the largest classes of gene regulators.

Functional miRNAs are excised from long endogenous transcripts by the sequential action of a pair of endonucleases (Drosha and Dicer) that reside in different compartments of the cell. In the nucleus, the primary microRNA (pri-miRNA) transcript is first cleaved by Drosha, liberating an approximately 60-to-80-nucleotide-long hairpin-shaped precursor miRNA (pre-miRNA). This pre-miRNA is then exported from the nucleus to the cytoplasm, where it undergoes a further processing by the Dicer enzyme and the resulting duplex is then loaded onto the RNA-induced silencing complex (RISC) in order to become more effective. Perfect base pairing between the RISC-bound miRNA and the target mRNA results in cleavage and degradation of the latter, whereas imperfect complementarity generally leads to translational repression of the target [7]. RISC recruits a multiprotein complex containing the anti-association factor eIF6. Depletion of eiF6 in human cells abrogates miRNA-mediated regulation of target protein and mRNA levels [8].

Not surprisingly, a variety of studies have linked aberrant microRNA expression to carcinogenesis where they act as both oncogenes and tumour suppressor genes [9]. Unique miRNA expression profiles have been able to classify various cancers. In one study, for example, the expression pattern of 217 microRNAs identified cancer type more accurately than messenger RNA [10].

A recent study reported aberrant miRNA expression in ovarian cancers compared to normal ovary [11]. In addition, direct evidence that miRNA is of critical importance in chemoresistance of human ovarian cancer has just been published [12]. In an attempt to understand the biology of recurrent ovarian cancer, we examined the expression of 180 miRNAs in primary and recurrent serous papillary adenocarcinomas, which our group has previously interrogated in a transcriptome study in recurrence [13, 14]. The aim of this study was to investigate the expression levels of miRNAs in recurrent ovarian cancer and to examine the expression levels of key components of the miRNA processing machinery. A flow chart of our experimental design is shown in Figure 1.

All samples were obtained at initial cytoreductive procedure from patients treated for ovarian cancer at St James's Hospital, Dublin, Ireland. The study had approval of the hospital ethics committee and informed consent was obtained from each patient by the research team prior to surgery. All tumors were staged according to the International Federation of Gynaecology and Obstetrics standards (FIGO).

This study demonstrates discrete miRNA signatures between primary and recurrent serous papillary adenocarcinomas on a histologically homogenous set of ovarian tissues from different patients. To our knowledge, this is the first ever study to examine the expression levels of miRNAs in recurrent cancer.

Real-time PCR can be employed to quantify miRNA expression profiles and study the potential function of miRNAs as well as miRNA precursors in cancer pathogenesis [21]. We utilized a novel stem-looped TaqMan RT-PCR method to quantify the expression profiles of 180 mature miRNAs in a collection of primary and recurrent serous papillary adenocarcinomas. This method offers several distinct advantages over conventional miRNA detection methods including a higher sensitivity, enhancing specificity for mature miRNAs and a fast and simple methodology [22].

We used a twofold change as a cut-off for miRNA differential expression in our training set. One might argue, much smaller changes could also be significant, reflecting the fact that miRNAs target many genes of even similar function. In this regard, disruption of cooperating genes can be a likely event with subtle miRNA expression [23]. However, unsupervised hierarchical clustering failed to segregate samples in primary and recurrent cohorts (3vs3). Instead, a single primary "chemoresistant" tumour clustered together with the recurrent ones suggesting inherent genetic differences between chemosensitive and chemoresistant groups. This finding enforces the potential advantage of utilizing miRNA expression patterns to predict relapse in ovarian cancer.

Because the expression of some miRNAs is highly tissue specific [24], it was actually expected that not all of the miRNAs would be detected in ovarian cancer samples. Furthermore, because of the tissue specificity, different sets of miRNAs are likely to be deregulated in cancers of different cellular origin, although it has been reported that the miRNA signatures of different cancer types could share some individual miRNAs [25]. Of the miRNAs we detected here as dysregulated, many are similarly dysregulated in other cancers: down-regulation of miR-125b in breast cancer [26], the let-7 miRNA family in lung cancer [27], miR-21 which activates PTEN in hepatocellular cancer [19], miR-155 in B cell lymphomas [28], miR-221 and miR-222 that clearly differentiate papillary thyroid carcinomas (PTC) from normal thyroid tissues (See Additional file 1). miR-221 and miR-222 block endothelial cell migration, proliferation and angiogenesis in vitro by targeting the stem cell factor receptor c-Kit [29]. miR-21 expression was shared by the signatures of six solid cancers (breast, colon, lung, pancreas, stomach, prostate) in a large scale miRome analysis [25] suggesting that miRNAs are involved in common molecular pathways. However, the majority of the differentially expressed miRNAs detected in this study have not been commonly reported to be deregulated in other cancers, implying that primary and recurrent ovarian cancers may have miRNA signatures specific for this cancer type.

Real time PCR was further employed to detect expression levels of a subset of miRNAs in an expanded cohort of fresh frozen serous papillary ovarian adenocarcinomas. Selected miRNAs comprised miR-223 (up) and miR-9 (down), which had demonstrated the greatest changes in expression levels in the initial cohort. Similar patterns of altered expression were observed for the top dysregulated genes (Figure 4).

We then set out to determine the expression patterns of selected miRNAs across a large panel of FFPE pathological tissues representing primary and recurrent serous ovarian adenocarcinomas, routinely used in hospital pathology laboratories. Unlike mRNAs, miRNAs remain largely intact in these routinely collected clinical samples [30]. "Recurrent" samples are rather limited, thus valuable, in common clinical practice therefore, LCM was performed in a subset of recurrent samples, as it can harvest tumour cells, in order to give histologically pure enriched cell populations [31]. Few earlier studies have addressed the expression characteristics of miRNA genes from FFPE tissues [30]. Our results suggest human archival tissue is a reliable source of study material for analyzing miRNA expression levels.

This miRNA "recurrent metastatic signature", comprised miR-223 and miR-9 may deserve some special consideration. miR-223 was the most upregulated in recurrent cancers when compared to primary. Recent work by Chen et al. identified among others miR-223 as a murine hematopoietic-specific miRNA and showed functionality of these miRNA through in vitro and in vivo overexpression studies [32]. miR-223 was highly expressed in cell lines of myeloid origin suggesting important regulatory roles in human hematopoiesis and oncogenesis [33]. More recently, it has been reported that miR-223 is a key member of a regulatory circuit that controls granulocytic differentiation and clinical response of acute promyelocytic leukemia (APL) blasts to all-trans retinoic acid (ATRA) [34]. ATRAs appear to be new promising drugs as they have been shown to arrest growth of ovarian carcinoma cells [35]. The potential role of miR-223 in immune and anti-inflammatory response has been recently described [36].

miR-9 was the most downregulated in recurrent cancers and appears to play a functional role in the determination of neural fates in embryonic stem (ES) cell differentiation [37]. miR-9 was also transcriptionally downregulated in a methylation-dependent way, in early breast cancer development [38].

miRNAs have been recently implicated in the development of ovarian cancer [11]. Specifically, in the serous histotype, 37 of 41 differentially expressed miRNAs were down-regulated relative to normal ovary. Notably, miR-9 downregulation was commonly shared between the serous and endometrioid histotypes compared to normal ovaries. The absence of other common miRNAs between the two studies, including miR-223 that was not identified in that study could be due to different experimental design (we carried out miRNA profiling in primary vs recurrent and not malignant vs normal) but might also reflect different experimental approaches or tumour heterogeneity. It also suggests that different miRNAs are involved in different stages of ovarian carcinogenesis. Iorio et al also identified miR-214 as "serous specific", which has been recently linked with chemoresistance in ovarian cancer by inducing cisplatin resistance through targeting the PTEN/Akt pathway [12]. miR-214 was also not included in our 60 miRNA-signature of dysregulated genes, possibly due to different selection criteria of control cells [39] or tumour heterogeneity.

Our team has previously examined gene expression levels in the training sample and identified over and under represented genes with full functional annotation [14]. In the current study, miRGen webserver was used to determine lists of putative mRNA targets of differentially expressed miRNAs. Although these lists may contain an unpredictable number of false positives, miRGen appears superior to other approaches, providing higher specificity for individual miRNAs [40]. It was interesting to observe that some of the differentially regulated miRNAs in this study correlate with our transcriptome findings [13] (See also Table 1) as rather few studies have attempted to link mRNA with miRNA data in different cancers [41, 42]. We provide strong evidence that miR-9 is a putative tumour suppressor gene given that predicted targets for miR-9 include members of the Claudin and FGF family that were upregulated in the recurrent tumours.

Common significant biological pathways involving predicted gene targets for miR-9 and miR-223 were identified using PANTHER. It appears that these pathways contain traits that distinguish primary and recurrent ovarian cancers. They have all previously been well characterised in ovarian cancer and chemoresistance. In this respect, if downregulated miRNAs are expected to activate overexpression of mRNA targets, it is fair to suggest that miR-9 downregulation can possibly lead to drug resistance.

The target list for miR-223 and miR-9 was interestingly enriched for genes involved in FGF cell signaling (FGFR2 and FGF respectively), which is in concordance with our transcriptome results [13]. This is also consistent with the very latest discovery that FGF signaling is crucial for collective invasion of carcinoma cells following the tracks generated by fibroblasts [43].

The mechanism of miRNA deregulation in recurrent ovarian cancer has not been thoroughly exploited. Fold changes <2 were observed for Dicer and Drosha at a mRNA level in our initial cohort and extended set of tumours with same histology. Inclusion of non serous tumours in the previously homogenous set conferred no statistical significance. To determine whether the observed gene expression pattern was altered at the protein level, primary and recurrent ovarian tumors (n = 71) were analysed immunohistochemically on TMA constructs containing multiple sections, using specific antibodies for Dicer and EIF6. Both proteins are localized in the cytoplasm and have roles in the processing of mature miRNAs.

Our results show that Dicer, Drosha and eIF6 may not have a substantive role in recurrent ovarian cancer and may only be involved in early tumorigenesis as shown by Chiosea et al [44]. We are unsure whether this represents a delay in processing or a complete block on the miRNA machinery, but it is possible that altered miRNA machinery is still functional and other components also participate, such as argonautes and p-bodies [45]. Our data do not exclude other important mechanisms, such as DNA copy number alterations [46] or disrupted epigenetic regulation [7], but rather suggest a multistep model for the control of miRNA dysregulation. The exact nature of the regulatory mechanism awaits further investigation.

In summary, we propose a miRNA model for the molecular pathogenesis of recurrent ovarian cancer (Figure 9). Some of the differentially deregulated miRNAs identified correlate with our previous transcriptome findings. Based on integrated transcriptome and miRNA analysis, miR-9 and miR-223 can be of potential importance as biomarkers in recurrent ovarian cancer, given their abundance, stability and easy detection. Future use of this miRNA signature as a blood-based diagnostic and prognostic tool can ultimately be translated into designing drugs that might offer a promising treatment to selected ovarian cancer patients.