Changes in microRNA expression associated with metastasis and survival in patients with uveal melanoma

Uveal melanoma (UM) is a major intraocular cancer that is molecularly distinct from cutaneous melanoma. Approximately half of patients with UM eventually develop metastasis. The prognosis of metastatic UM is poor, with a median overall survival (OS) of less than a year. In this study, we sought to identify microRNAs (miRNAs) associated with metastasis and OS in UM. We analyzed the miRNA expression and clinical outcomes data from The Cancer Genome Atlas (TCGA) dataset for UM. Differential expression analyses were conducted for each miRNA with respect ever-development of metastasis. Multiple survival analyses were done, using the Cox proportional hazards model, to evaluate interactions between miRNA expression, metastasis, and OS. A total of 22 miRNAs (3 upregulated and 19 downregulated) were differentially expressed between patients with vs. without metastatic UM. These 22 miRNAs could be grouped into four clusters based on similarities in expression patterns. Of the 22 miRNAs differentially expressed with respect to metastasis, 21 were significantly associated with OS. The expression of multiple miRNAs was significantly associated with metastasis and overall survival in patients with UM. Further investigation of these miRNAs as biomarkers and/or therapeutic targets is warranted in the push to improve outcomes for patients with metastatic UM.


INTRODUCTION
Uveal melanoma (UM) is the most common primary intraocular cancer occurring in adults [1,2]. The mortality of UM patients is approximately 40-50%; the leading contributor to mortality is development of metastasis, which occurs in up to 50% of UM patients [3,4]. For patients with metastatic UM, the 1-year survival rate is 20%, the 5-year survival rate is less than 5%, and the median overall survival is only 6-12 months [1,[5][6][7][8].
There is no effective therapeutic intervention to treat metastatic UM [4].
Given the biologic uniqueness of UM, its propensity to metastasize, and the poor survival outcomes and lack of adequate treatment for metastatic UM, there is a great need to uncover the molecular mechanisms of metastasis in UM and to discover predictive and prognostic biomarkers so as to better optimize understanding and management of this challenging oncologic condition.
Several genetic mutations and gene expression alterations have been associated with the molecular mechanisms responsible for the progression of UM [14]. It has also been shown that epigenetic modifications including microRNAs (miRNAs) are associated with the pathology and progression of UM [14]. miRNAs are small, non-coding RNAs approximately 22 nucleotides in length. Each microRNA can play a crucial role in regulating the expression of multiple genes. miRNAs typically regulate Research Paper gene expression by altering mRNA stability or repressing translation of mRNA to protein. Several miRNAs play an important role (tumor suppressors or tumor promoters) in cancer development and progression including metastasis [15][16][17][18][19][20]. However, data are limited regarding the role of miRNAs in metastatic UM [21][22][23]. Therefore, the purpose of this study was to identify miRNAs associated with UM metastasis and overall patient survival.

Demographics, key clinical data, and differences in OS for patients with metastatic vs. nonmetastatic UM
Profiling data from 1,598 miRNAs was available for primary-site tumor samples from 80 patients with UM in TCGA. Of those patients, 50 (62.5%) had never developed metastasis, while 30 (37.5%) had ever developed metastasis. Key demographic and clinical data for UM patients with metastatic and non-metastatic UM were identified and compared across the two groups. Compared to those with non-metastatic UM, patients with metastatic UM more often had epithelioid histology, less often had spindle cell histology, and had a higher proportion of advanced (stage III-IV) disease at initial diagnosis (p-values all < 0.05). There was no significant difference between non-metastatic and metastatic UM with respect to age, gender, or tumor thickness (Table 1). Survival analysis for metastatic vs. non-metastatic UM showed a major difference (hazard ratio [HR] = 15.24; p-value = 2.42 × 10 −4 ) with respect to overall survival (OS) between those groups ( Figure 1A).

miRNAs associated with survival status in UM
Cox proportional hazard analysis found that 64 miRNAs were significantly (adj. p-value < 0.001 and HR > 4 or HR < 0.2) correlated with patient survival (Supplementary Table 2). The 15 miRNAs most significantly associated with overall survival (OS) of patients (HR > 10 or HR < 0.10) are listed in Table 3; the Kaplan-Meier OS curves of these miRNAs are shown in Figure 4. Of the 22 miRNAs significantly differentially expressed in patients with metastasis, 21 were significantly (p < 0.05) associated with OS.

Target genes and pathways regulated by miRNAs associated with UM metastasis
A comprehensive search of experimentally validated target genes regulated by miRNAs found to be associated with UM metastasis was performed using Ingenuity Pathway Analysis software (QIAGEN, Redwood City, CA, USA). Further bioinformatics analyses were performed to discover pathways and biological processes associated with these target genes. The top canonical pathways associated with these genes include p53 signaling, regulation of epithelial-mesenchymal transition pathway, cell cycle G1/S checkpoint regulation, ILK signaling, and PTEN signaling ( Table 4). The top biological functions include apoptosis, necrosis, growth of tumor, cell proliferation, invasion, movement, migration, and cell cycle progression (Table 4). We also performed network analysis to discover the interactions between the target genes. The top-scoring network is shown in Figure 5. The network analysis revealed that the hub genes of the network are MYC, VIM, AR, ERBB2, HIF1A, FOS, KRAS, VEGF, PKA, ELAVL1, and GSK3B. These hub genes interact with many (> 15) nodes on the interaction network and are likely important for gene expression dynamics (mechanism). The top upstream regulators of the target genes are TP53, EGF, TGFB1, PTEN, MYC,   (Table 4). These upstream regulators are the predicted transcriptional regulators in the pathway.

DISCUSSION
The development of metastases plays an important role in UM patient prognosis. Molecular biomarkers associated with UM metastasis may help in accurately identifying high-risk patients and in discovering potential therapeutic targets for metastatic UM treatment. MicroRNAs are small single-stranded endogenous noncoding RNAs, which are involved in the posttranscriptional regulation of expression of their targeted mRNAs. It has been established that aberrant expression of miRNAs leads to progression and metastasis of several cancers. In the past decade, several studies have examined the role of microRNAs in pathogenesis and progression of UM by utilizing plasma [24], serum [25], cell lines [26][27][28][29], and clinical tissue specimens [21][22][23]30]. Uveal melanoma is a very rare cancer, making it difficult to obtain a large number of samples from a single institution. TCGA is a landmark cancer genomics dataset which has molecularly characterized cancer and matched normal samples and is an especially important resource for rare cancers such as UM. TCGA also provides data on various clinical and demographic parameters associated with UM and analyzes this larger sample set with careful experimental design and proper control groups. The use of this larger sample size increases the statistical power of the analysis. Also, to minimize the false positives, we have used a very stringent cutoff to select differentially expressed miRNAs. This approach enabled us to identify several novel miRNAs potentially related to UM metastasis which may have clinical, biological, or mechanistic relevance to UM and may expand our understanding of UM tumor progression.
In this study, we found 22 miRNAs highly dysregulated (> 2-fold change and p < 0.01) in UM patients with (vs. without) ever-development of metastasis. 21 (95%) of those miRNAs associated with metastasis were also significantly associated with poor OS. The 22 miRNAs associated with metastasis could be divided into four distinct clusters based on highly correlated expression patterns within each cluster.
Cluser-4 included the three most significantly upregulated miRNAs, including miR-592, miR-708-5p, and miR-199a-5p. Previous studies have also shown that miR-199a regulates melanoma metastasis related genes and may provide new therapeutic targets [19,21]. In a recent study, higher expression of miRNA 199a was observed in UM with liver metastasis [22]. Using a genome-wide microarray based approach, another study found that expression of miRNA-199a was one of the most significant discriminators of low metastasis and high metastasis risk of UM patients [21].
The additional bioinformatic analyses we performed identified the target genes and pathways regulated by the miRNAs found to be associated with UM. Several key transcription regulators (TP53, MYC, SP1, TP63, E2F1), growth factors (EGF, TGFB1, FGF2, HGF) and other key regulators including PTEN, ERBB2, ESR1, KRAS and PI3K-complex were found as key targets using this analysis. Previous studies have also reported constitutive activation of these oncogenic pathways in primary UM [34,35]. Biological functions related to metastasis including cell cycle progression, cell proliferation, invasion, movement and migration were significantly enriched in the target genes.
With the recent advancement of molecular technologies, miRNAs have newfound potential to serve as viable therapeutic tools. Molecular approaches such as AMOs (anti-miR oligonucleotides), LNA anti-miRs, antagomirs, miRNA sponges, and S-miRs (small molecule inhibitors to target specific miRNAs) are available to inhibit the miRNAs overexpressed in cancer [36][37][38][39][40]. On the other hand, molecular approaches to restore the decreased expression of miRNAs downregulated in cancer are also available and include miRNA mimics (double-stranded synthetic RNAs that mimic endogenous miRNAs) and miRNA expression vectors. Several studies have used miRNA replacement therapy in experimental models [41][42][43]. Additionally, in a recent study, aptamer-miRNA conjugates were used as a novel tool for targeted delivery of miRNAs [44]. Several miRNA-based therapies are already in clinical trials, for example, miR-16 mimics are under phase 1 clinical trials for patients with recurrent thoracic cancer [45].
A major limitation of this study was a lack of experimental validation of the findings with either a separate dataset from patient samples or through in vivo or in vitro experiments. Further, as with other investigations primarily based on data from TCGA, our analyses used retrospectively obtained data and TCGA UM patient population may not be fully generalizable to some UM patient populations with demographic or clinical features under-represented in TCGA.
In conclusion, this study identified, in primarysite tumor samples, altered miRNA expression patterns associated with ever-development of metastasis in patients with uveal melanoma. We found several known tumor suppressor miRNAs to be downregulated in UM patients with metastasis. These results support the increasingly accepted concept that miRNAs play a major role in metastasis. Our finding of 95% overlap between (a) miRNAs associated with UM metastasis and (b) miRNAs associated with poor survival in patients with UM warrants further investigation those overlapping miRNAs. Future evaluation of the 21 overlapping miRNA as prognostic biomarkers and/or therapeutic targets may be a step toward improved outcomes for those with metastatic UM, a patient population that suffers from high mortality and a lack of effective treatment options.

Dataset
The Cancer Genome Atlas (TCGA; RRID: SCR_003193) is one of the foremost data repositories providing molecular characterization of more than 20,000 primary cancers, including unprecedented amounts of miRNA sequence data (~11,000 libraries) across 33 cancer types [20]. We therefore chose TCGA as the dataset for this investigation. We utilized this high-quality data for our study to analyze the differential miRNA expression between patients with and without metastasis. For the sake of this analysis, "metastasis" refers to metastasis at either initial presentation/diagnosis or recurrence (i. e., everdevelopment of metastasis).
TCGA miRNA expression data was generated using the Illumina HiSeq/GA miRseq and was reported as counts normalized to reads per million mapped reads (RPM). The uveal melanoma miRNA dataset was downloaded from UCSC Xena browser [46]. The dataset includes miRNA expression data from each sampled tumor, as well as corresponding demographic and clinical information such as patient survival and presence of metastases. For statistical analyses, expression values were log2 transformed to achieve a normal distribution. All statistical analyses were performed using the R language and environment for statistical computing (R version 3.5.2; R Foundation for Statistical Computing; https://www.r-project.org; RRID: SCR_001905).

Differential expression analysis
miRNA expression observations were normalized and differential miRNA expression between metastatic and nonmetastatic tumors was analyzed for all evaluable miRNAs in TCGA UM dataset using the LIMMA package (RRID: SCR_010943) [47]. P-values were adjusted using the false discovery rate (FDR) method. Also, to minimize the false positives, a cut-off of fold change > 2 and adj. p-value < 0.01 was used to select the differentially expressed miRNAs.

Survival analysis
The survival difference between patients with vs. without metastatic UM in this TCGA dataset was calculated using the Cox proportional hazard model [48]. Independently, we performed survival analyses for each miRNA in the UM TCGA dataset. For each miRNA, subjects were separated into high-expression or lowexpression groups relative to the median expression value. Cox proportional hazard models were fitted for each miRNA. The p-values for HRs were computed and adjusted using the FDR method.

Bioinformatics analyses
Ingenuity Pathway Analysis (IPA) software was used to identify the target genes of miRNAs found to be associated with UM metastasis. Bioinformatics analyses of the target genes were performed using the IPA software for identification of enriched canonical pathways and biological functions. The prediction of upstream regulators was also done using the IPA software.

Author contributions
AV contributed to conceptualization. AV and TJL contributed to methodology, data curation, and formal analysis. AV, TJL, and JJW contributed review and interpretation of data analysis. AV, TJL, and JJW contributed to writing. AV, TJL, and JJW contributed to editing. AS contributed to bioinformatics analyses to identify gene targets and pathways. JJW contributed to final review and submission.