GDF11 upregulation independently predicts shorter overall-survival of uveal melanoma

Growth differentiation factor 11 (GDF11), is a member of the transforming growth factor-beta (TGF-β) superfamily and bone morphogenetic protein (BMP) subfamily. In this study, we aimed to assess the expression profile of GDF11, its prognostic value in terms of OS, as well as the potential mechanisms leading to its dysregulation in uveal melanoma. A retrospective study was conducted using our primary data and genetic, clinicopathological and overall survival (OS) data from the Cancer Genome Atlas-Uveal Melanoma (TCGA-UVM). Results showed that GDF11 expression was significantly higher in tumor tissues compared with that in adjacent normal tissues. High GDF11 expression was associated with uveal melanoma in advanced stages (IV), epithelioid cell dominant subtype, as well as extrascleral extension. Univariate analysis showed that older age, epithelioid cell dominant, with extrascleral extension and increased GDF11 expression were associated with unfavorable OS. Multivariate analysis confirmed that GDF11 expression was an independent prognostic indicator of unfavorable OS (HR: 1.704, 95%CI: 1.143–2.540, p = 0.009), after adjustment of age, histological subtypes and extrascleral extension. Among the 80 cases of uveal melanoma, only 3 cases had low-level copy gain (+1) and 2 cases had heterozygous loss (-1). No somatic mutations, including SNPs and small INDELs were observed in GDF11 DNA. The methylation of these four CpG sites had weakly (cg22950598 and cg23689080), moderately (cg09890930), or strongly (cg05511733) negative correlation with GDF11 expression. In addition, the patients with high methylation of these four sites had significantly better OS compared to the group with low methylation. Based on these findings, we infer that methylation modulated GDF11 expression might be a valuable prognostic biomarker regarding OS in uveal melanoma.


Introduction
Uveal melanoma arises from the melanocytes residing within the uvea and is the most common primary intraocular cancer in adults [1]. The risk of metastasis and death varied significantly among patients with different stages of tumor. For stage III tumors, the metastasis and death rates at five years were 44% and 27%. In comparison, all stage IV tumors had metastasis and death by one year [2]. Although long-term survival is uncommon in patients with metastatic tumors, certain subsets of patients had long-term survival [3,4]. However, the characteristics associated with long-term survival have not been fully understood.
Growth differentiation factor 11 (GDF11), is a member of the transforming growth factorbeta (TGF-β) superfamily and bone morphogenetic protein (BMP) subfamily [5]. GDF11 transmits signals through type I and II serine/threonine kinase receptors, which is similar to other TGF-β superfamily members [6]. Therefore, GDF11 can activate both Smad and non-Smad signals via binding to its receptors and subsequently regulate the expression of the downstream genes [6]. In the past decades, a series of studies showed that GDF11 is involved in embryonic development such as spinal cord anterior/posterior patterning and the development of urogenital system [6][7][8]. In addition, it plays an important role in the pathologic development of some diseases, such as cardiovascular disease, diabetes mellitus, and cancer.
In cancer biology, the role of GDF11 is conflicting. In patients with colorectal cancer, GDF11 was upregulated in tumor tissues compared with adjacent normal tissues [9]. In addition, high GDF11 expression is associated with a higher risk of lymph node metastasis and poorer overall survival [9]. In oral squamous cell carcinoma (OSCC), GDF11 overexpression is associated with induced epithelial to mesenchymal transition (EMT), cell migration as well as metastasis [10]. In comparison, GDF11 acts as a tumor suppressor in triple-negative breast cancer (TNBC) via promoting an epithelial and anti-invasive phenotype [11] and also suppresses the proliferation, migration and invasion of pancreatic cancer cells [12].
In this study, using primary tissue data and secondary data from the Cancer Genome Atlas-Uveal Melanoma (TCGA-UVM), we assessed the expression profile of GDF11, its prognostic value in terms of OS, as well as the potential mechanisms leading to its dysregulation.

Patients and methods
The study was approved by the ethics committee of Weifang People's Hospital, China. Written informed consent was obtained from all patients before this study.

Specimens
Tumor samples and their adjacent morphologically normal tissues (> 0.3 cm from the tumor) were obtained from 19 uveal melanoma patients who received primary enucleation and without prior radiation or chemotherapy, at the Department of Ophthalmology, Weifang People's Hospital, China. The tissues were snap-frozen in nitrogen and stored at −80˚C before RNA extraction. All patients were given diagnoses according to current ophthalmologic criteria. Tumor diameter varied from 8 mm to 15 mm (mean±SD: 10.8±1.7 mm). Histopathological analysis showed that 6 tumors were epitheloid, 10 were spindle cell, and 3 were mixed histology.

Infinium HumanMethylation450 BeadChip assay
Genomic DNA (gDNA) methylation status of the primary tissue samples was examined using the Infinium HumanMethylation450 BeadChip kit (450K) (Illumina, San Diego, CA, USA) according to manufacturer's instructions. Briefly, gDNA was extracted from the tumor and adjacent normal tissues and was treated using the EZ DNA Methylation kit (Zymo Research, Orange, CA, USA) for bisulfite modification. The samples after purification were subjected to hybridization on Infinium HumanMethylation450 (450K) BeadChips, following the protocol recommended by the manufacturer. The signal intensities and the methylation level of each CpG site were determined using the GenomeStudio software. Beta values were calculated following the formula: β = M/(U+M+100), in which M refers to the fluorescence level of the methylation probe and U means the methylation level of the unmethylated probe.

Data mining in the cancer Genome Atlas-Uveal melanomas (UVM)
The level 3 data in TCGA-UVM were downloaded by using the UCSC Xena browser (https:// xenabrowser.net). In this dataset, 80 patients with primary uveal melanomas, who had no history of neoadjuvant treatment were included. The basic information of the patients can be obtained from: https://portal.gdc.cancer.gov/projects/TCGA-UVM.
Their clinicopathological and survival data including age at diagnosis, gender, histological type, pathological stage, pathological nodal status, pathological metastasis status, tumor diameter (mm), tumor thickness (mm), extrascleral extension and OS status and OS in days were downloaded for survival analysis. Among the 80 patients included, 23 cases were deceased by the end of the project, among which 19 cases were due to metastatic uveal melanoma.

Statistical analysis
GDF11 expression in different groups was compared using one-way ANOVA followed by Tukey post-hoc test or using Welch's t-test. Kaplan-Meier curves of overall survival (OS) were generated by GraphPad Prism v6.0 (GraphPad Software Inc., La Jolla, CA, USA). Two types of grouping were performed to separate the patients: 1, based on median GDF11 expression irrespective of methylation status, 2, strictly based on GDF11 methylation status irrespective of its expression status. Receiver operating characteristic (ROC) curve of GDF11 expression for death detection was plotted and area under the curve (AUC) was calculated. According to the AUC values, the accuracy of a prognostic test can be roughly classified into five categories: 0.90-1 = excellent, 0.80-0.90 = good; 0.70-0.80 = fair; 0.60-0.70 = poor and 0.50-0.60 = fail [13]. The association between GDF11 expression and the clinicopathological parameters was examined by using χ 2 test by two-sided Fisher's exact test. The difference between the survival curves was assessed using the Log-rank test. Univariate and multivariate Cox regression models were applied to evaluate the independent prognostic value of GDF11, as a continuous variable. Linear regression analysis was conducted to evaluate the correlation between GDF11 expression and the methylation value of each CpG sites. p < 0.05 was considered statistically significant.

GDF11 expression profile in uveal melanoma
By performing qPCR assay, we examined GDF11 expression in 19 cases of uveal melanoma tumor tissues and adjacent normal tissues. Results showed that GDF11 was generally upregulated in the tumor tissues compared with adjacent normal tissues (Fig 1). To further explore the association between GDF11 expression and the phenotypes of uveal melanoma, we GDF11 and overall-survival of uveal melanoma retrieved RNA-seq and clinicopathological data in TCGA-UVM. By grouping the tumor cases according to pathological stages, we found that stage IV tumors had the highest GDF11 expression (Fig 2A and 2B). In comparison, the difference between stage II and stage III tumors was not significant (Fig 2A and 2B). By comparing GDF11 expression between epithelioid cell dominant and spindle cell dominant subtypes, we found that the more malignant epithelioid cell dominant subtype had significantly higher GDF11 expression (Fig 2C and 2D). In addition, we also found that the samples with extrascleral extension had significantly upregulated GDF11 expression compared to the negative cases (Fig 2E and 2F). Notably, GDF11 expression was substantially higher in the deceased cases compared with that in the living cases (Fig 2G).

Increased GDF11 expression independently predicts unfavorable OS in uveal melanoma
By performing ROC analysis regarding OS, GDF11 expression had an AUC value of 0.753 ( Fig  3A), suggesting that high GDF11 expression might be a fair marker of unfavorable OS. Then, Kaplan-Meier curves of OS were generated by setting median GDF11 expression as the cutoff. Results that the patients with high GDF11 expression had significantly shorter OS, compared to the patients with low GDF11 expression (p = 0.001, Fig 3B). By comparing the clinicopathological parameters between the high and low GDF11 expression groups, we found that the high expression group was significantly older (mean ± SD, 64.7±12.45 vs. 58.60 ± 14.83, p = 0.05), had a higher proportion of epithelioid cell dominant subtype (25/40 vs. 9/40, p < 0.001), more patients in advanced stages, thicker tumors (> 10 mm vs. � 10 mm, 27/40 vs. 16/40, p = 0.024) and a higher death rate (18/40 vs. 5/40, p = 0.003) ( Table 1). By performing univariate analysis, we found that older age, epithelioid cell dominant, with extrascleral extension and increased GDF11 expression were associated with unfavorable OS (Table 2). In multivariate analysis, GDF11 expression was an independent prognostic indicator of unfavorable OS (HR: 1.704, 95%CI: 1.143-2.540, p = 0.009) ( Table 2).

Genetic and epigenetic related mechanisms underlying the dysregulation of GDF11 in uveal melanoma
Since we identified that GDF11 expression might be a valuable prognostic indicator in uveal melanoma, we then tried to assess the potential mechanisms of its dysregulation. By examining GDF11 DNA CNAs in uveal melanoma, we found that DNA amplification and deletion were not frequent. Among the 80 cases of uveal melanoma, only 3 cases had low-level copy gain (+1) and 2 cases had heterozygous loss (-1) (Fig 4A). However, although the low-level copy gain cases had significantly elevated GDF11 expression compared to the copy neutral (0) cases, the GDF11 heterozygous loss did not necessarily result in GDF11 downregulation (Fig 4B). In addition, no somatic mutations, including SNPs and small INDELs were observed in GDF11 DNA (Fig 4A), suggesting that GDF11 dysregulation was less likely to be influenced by genetic alteration.
Then, we further explored the association between GDF11 expression and its DNA methylation, an epigenetic mechanism influencing gene expression. The methylation status of 15  CpG sites in GDF11 DNA was measured in Illumina Infinium Human Methylation 450K Bead-Chip. In the heatmap, we found that the methylation of some CpG sites (cg22950598, cg09890930, cg05511733 and cg23689080) were negatively correlated with GDF11 expression (Fig 5A). By performing linear regression analysis, we confirmed that the methylation of these four CpG sites had weakly (cg22950598 and cg23689080), moderately (cg09890930), or strongly (cg05511733) negative correlation with GDF11 expression (Fig 5B). Although another CpG site cg15466281 also showed a moderately negative correlation with GDF11 expression (Pearson's r = -0.55), the average methylation of this site was low in uveal melanoma (mean ± SD: 0.04 ± 0.04) (Fig 5B). Therefore, the influence of this site on GDF11 expression was limited. Then, we assessed the association between the average methylation of cg22950598, cg09890930, cg05511733 and cg23689080 and OS of uveal melanoma. Kaplan-Meier showed that the group with high methylation had significantly better OS compared to the group with low methylation (Fig 5C). This finding further confirmed that methylation modulated GDF11 expression was a valuable prognostic biomarker in uveal melanoma. Since the status of cg05511733 was strongly and negatively correlated with GDF11 expression, we compared the methylation level of this CpG site between the 19 primary tumor and adjacent normal tissues. Results showed that the adjacent normal group had a significantly higher level of methylation than the tumor group (p<0.001, Fig 5D).

Discussion
In this study, by using data from TCGA-UVM, we demonstrated that high GDF11 expression was associated with uveal melanoma in advanced stages (IV), epithelioid cell dominant subtype, as well as extrascleral extension. More importantly, we confirmed that GDF11 expression was an independent prognostic indicator of unfavorable OS (HR: 1.704, 95%CI: 1.143-2.540, p = 0.009), after adjustment of age, histological subtypes and extrascleral extension. These findings suggest that GDF11 might serve as a valuable prognostic biomarker in uveal melanoma.
GDF11 firstly binds to Activin receptor II (ActRIIA and ActRIIB), and then recruits Activin receptor I (ActRI) including activin receptor-like kinase 1 (ALK1), ALK4, ALK5 and ALK7 [14,15]. After that, their complex activates canonical Smad signaling via including Smad2/3 and Smad1/5/8 [6]. Besides, the complex can also activate non-Smad signals such as Rho-like GTPase, MAP kinases (including p38, ERK and JNK), and phosphatidylinositol-3-kinase/AKT [16]. Several recent studies found that GDF11 acts an important stimulator of angiogenesis. It stimulates angiogenesis via in focal cerebral ischemia/reperfusion rats via ALK5 [17], and also stimulates pulmonary artery endothelial cell (PAEC) proliferation, migration, tube formation, via activating ALK1/p-Smad1/5/8 and ALK5/p-Smad2/3 signals [18]. Angiogenesis plays a critical role in the progression and metastasis of uveal melanoma [19,20]. Inhibition of angiogenesis via the neddylation pathway inhibited hepatic metastasis in uveal melanoma, using NOD-SCID mouse xenograft model [21]. These mechanisms help to explain the association between GDF11 upregulation and the poor OS of uveal melanoma. Although some studies reported that GDF11 might be a tumor suppressor in some other cancers, such as TNBC [11], the contradictory results might be a result of the dual role of TGF-β in the different stages of cancer. In normal cells and early carcinomas, TGF-β signaling pathways mainly exerts tumor suppressive effect. However, the protective effects of TGF-β signaling are usually lost, which in turn switches to promote tumor progression, invasion and metastasis [22].
The mechanisms underlying GDF11 dysregulation were quite complex in different diseases and might be tissue specific. In BALB/c-3T3 cells, GDF11 expression is activated by the histone deacetylase (HDAC) inhibitor trichostatin A, while is repressed by HDAC3 [23]. In PAECs, the transcription factor zinc finger protein 740 directly binds to the GDF11 promoter and enhances its transcription [18]. In this study, we explored the potential genetic and epigenetic (typically methylation) alterations in GDF11 DNA in uveal melanoma. Results showed that DNA CNAs were not frequent in uveal melanoma. In addition, no somatic mutations, including SNPs and small INDELs were observed in GDF11 DNA. However, we found that that the methylation of these four CpG sites had weakly (cg22950598 and cg23689080), moderately (cg09890930), or strongly (cg05511733) negative correlation with GDF11 expression. Also, we demonstrated that the patients with high methylation of these four sites had significantly better OS compared to the group with low methylation. In addition, using data from primary samples, we confirmed that the adjacent normal group had a significantly higher level of cg05511733 methylation than the tumor group. These findings indicated that methylation is an important mechanism of GDF11 dysregulation in uveal melanoma.
This study also has some limitations. Firstly, we only explored the association between GDF11 expression in tumor tissues and the OS of uveal melanoma patients. In the future, it is quite necessary to explore whether it has prognostic value as a circulating protein found in serum. In addition, since only OS data was recorded in TCGA-UVM, we had not evaluated the prognostic value regarding disease-free survival (DFS). This also needs to be assessed in a large patient cohort in the following studies.

Conclusion
Methylation modulated GDF11 expression might be a valuable prognostic biomarker in terms of OS in uveal melanoma.