1. Introduction
Melanoma is a highly aggressive malignant tumor originating from neural crest melanocytes [Reference Brancaccio, Russo, Lallas, Moscarella, Agozzino and Argenziano1, Reference Longo and Pellacani2]. Melanoma cells can infiltrate the tissues, lymph, and blood vessels. Melanoma accounts for 3% of all malignant tumors of the skin [Reference Eddy and Chen3], with its incidence rate increasing rapidly worldwide [Reference Namikawa and Yamazaki4, Reference Yde, Sjoegren, Heje and Stolle5]. In 2017, the incidence rate of melanoma in China was 0.9/10 million by age [Reference Wu, Wang, Wang, Yin, Lin and Zhou6]. Melanoma is not sensitive to radiotherapy, chemotherapy, and biological immunotherapy, and its invasiveness and early metastasis make it the deadliest among skin malignant tumors, with only a 6% five-year survival rate of stage IV patients [Reference Weber, Mandala and Del Vecchio7]. Thus, it is necessary to screen for reliable and effective biomarkers for the diagnosis and prognosis of melanoma.
MicroRNAs (miRNAs) are gene regulatory factors with excellent functions, ranging in size from 19 to 25 nucleotides [Reference Lu and Rothenberg8]. miRNAs participate in the posttranscriptional control of gene expression by binding to the 3′-untranslated region (3′-UTR) of the target mRNA and then inhibits or degrades the target gene [Reference Cannell, Kong and Bushell9]. This molecular-level imbalance in biological processes is a potential mechanism for the development of many diseases. Numerous miRNAs have been reportedly involved in melanoma development. Lu et al. found that five miRNAs, including miR-25, miR-204, miR-211, miR-510, and miR-513c, could be used as prognostic biomarkers for patients with cutaneous melanoma cancer [Reference Lu, Chen, Qu, Wang, Chen and He10]. Zhu et al. revealed that miRNA-378a-3p, miRNA-23b-3p, and miRNA-200c-3p are associated with the occurrence of skin cutaneous melanoma [Reference Zhu, Deng and Zhang11]. Li et al. suggested that miR-300 and miR-629 are associated with melanoma [Reference Huang, Zou and Tang12]. However, these studies screened the miRNAs related to melanoma based on a single dataset with a small sample size, and few studies have identified the miRNAs involved in the progression of melanoma. Therefore, miRNA biomarkers for melanoma progression should be identified based on multiple datasets.
This study aimed to screen miRNA biomarkers for melanoma progression. Using the data downloaded from the Gene Expression Omnibus (GEO) and the Cancer Genome Atlas (TCGA) databases, miRNAs related to melanoma progression were obtained, the miRNA-gene network was constructed, and prognosis-related miRNAs were screened. Moreover, cluster analysis was conducted based on miRNAs related to melanoma progression, and the differentially expressed genes (DEGs), tumor-infiltrating immune cells, and tumor mutation burden (TMB) between subtypes were analyzed. This study provided a powerful basis for identifying novel therapeutic targets for melanoma patients. A flowchart of this study is shown in Figure 1.
2. Materials and Methods
2.1. Data Collection
Four datasets that included some patients with melanoma were included from the GEO database [Reference Barrett, Troup and Wilhite13], including GSE34460 (8 benign melanoma, 9 primary melanoma, and 4 metastatic melanoma, detection platform: GPL15019 Agilent-031181 Unrestricted_Human_miRNA_V16.0_Microarray 030840 (miRBase release 14.0 miRNA ID version)), GSE35579 (11 benign melanoma, 20 primary melanoma, and 21 metastatic melanoma, detection platform: GPL15183 CRUK/Melton lab-Human melanoma-71-v2-microRNA expression profiling), GSE18509 (8 benign melanoma, 8 metastatic melanoma, detection platform: GPL9081 Agilent-016436 Human miRNA Microarray 1.0 G4472A (miRNA ID version)), and GSE24996 (8 benign melanoma, 15 primary melanoma, and 8 metastatic melanoma, detection platform: GPL6955 Agilent-016436 Human miRNA Microarray 1.0 (feature number version)). Moreover, the miRNA mature strand expression RNAseq data (log2 (RPM+1)), gene expression RNAseq data (log2 (norm_count+1)), and the survival information of the corresponding samples (overall survival (OS), disease-specific survival (DSS), and progression-free interval (PFI)) of skin melanoma (SKCM) were obtained from TCGA of UCSC-Xena [Reference Goldman, Craft, Brooks, Zhu and Haussler14]. The annotated MAF file processed by the MuTect software of somatic mutation data (SNPs and small INDELs) of TCGA-SKCM was obtained from the Genomic Data Commons (GDC) database (https://portal.gdc.cancer.gov) for subsequent mutation correlation analysis.
2.2. Data Processing
For the GSE34460, GSE18509, and GSE24996 datasets, the raw data were preprocessed using limma [Reference Smyth15] in the R package. The data preprocessing processes included data reading, RMA background adjustment, quantile normalization, and normalization. The expression matrix of probes was obtained, and the probes were annotated to miRNA ID according to the platform annotation information; then, the miRNA ID was converted using miRBase website [Reference Ana, Birgaoanu and Griffiths-Jones16], and the obtained miRNAs were used for the subsequent analysis. For the GSE35579 dataset, the processed probe expression matrix was downloaded, which was log-transformed, and the probes were annotated to miRNA ID according to the platform annotation information. Then, the miRNA ID was converted using miRBase website, and the obtained miRNAs were used for the subsequent analysis. For the TCGA-SKCM dataset, samples with an OS. time > 0 were selected for subsequent cluster analysis.
2.3. Differential Expression Analysis
Based on the GSE34460 and GSE35579 datasets, the typical Bayesian test provided by the R package limma package was used to screen the differentially expressed miRNAs (DEmiRNAs) between benign vs. primary, metastatic vs. benign, and metastatic vs. primary, with a cutoff value of P < 0.05 and |logFC| > 0.5. Then, all overlapping DEmiRNAs screened between benign vs. primary, metastatic vs. benign, and metastatic vs. primary were obtained in each dataset, and the common DEmiRNAs were obtained from the two datasets. Based on the common DEmiRNAs, the expressed values of the common DEmiRNAs in each sample of the GSE34460 and GSE35579 datasets were extracted. For the same group of samples, the average of each miRNA was calculated as the average level of the miRNA in the sample group. After comparing the expression levels of DEmiRNAs in the benign, primary, and metastatic groups, the DEmiRNAs were divided into four types: expression continuous upregulation (up-up), expression continuous downregulation (down-down), upregulated first and then downregulated (up-down), and downregulated first and then upregulated (down-up). The miRNAs that overlapped in the same types were considered miRNAs related to disease progression for subsequent analysis.
2.4. Prediction of Target Genes of miRNA and Enrichment Analysis
Based on the six databases, miRWalk, Microt4, miRanda, miRDB, RNA22, and TargetScan, the miRNA-gene interaction pairs were screened using the miRWalk 2.0 tool [Reference Dweep and Gretz17]. The miRNA-gene interaction pairs overlapped in the six databases were identified. Moreover, the “Skin Cutaneous Melanoma” was used as the keyword to screen the disease-related genes in the GeneCards database [Reference Stelzer, Rosen and Plaschkes18]. The disease-related genes were intersected with target genes of miRNA that overlapped in the six databases, and the miRNA-gene interaction pairs were further obtained, followed by visualization using Cytoscape [Reference Shannon, Markiel and Ozier19]. Additionally, clusterProfiler [Reference Yu, Wang, Han and He20] in the R package was used to perform enrichment analysis of the genes. The P value was corrected using Benjamini and Hochberg (BH), and the adjust P < 0.05 was regarded as a statistically significant difference.
2.5. Survival Analysis
Based on the miRNAs related to disease progression, the expression values of the miRNAs in each sample were extracted from the miRNA expression data in TCGA-SKCM samples. The patients were then allocated to high and low-miRNA expression groups according to the median expression value, and the Kaplan–Meier curve method was used to evaluate the association between miRNAs and prognosis.
2.6. Cluster Analysis of miRNAs
Based on the miRNAs related to disease progression and the expression value of miRNAs in each sample in TCGA-SKCM, bidirectional hierarchical clustering was conducted based on the centered Pearson correlation algorithm using pheatmap [Reference Johnson21] to identify different subtypes. The Kaplan–Meier curve method was used to evaluate the association between subtypes and prognosis.
2.7. Identification of DEGs between Subtypes
The gene expression matrix corresponding to each subtype sample was extracted, and the DEGs were identified between the two subtypes using limma in the R package. The P value was corrected using BH, and the adjust P < 0.05 and |logFC| > 0.5 were set as the cutoff values. Moreover, the DAVID tool [Reference Huang, Sherman and Lempicki22] was used to perform enrichment analysis on the common DEGs with a threshold value of P < 0.05 and count ≥ 2.
2.8. Tumor-Infiltrating Immune Cells
Based on the gene expression level of samples in each subtype, the CIBERSORT algorithm [Reference Charoentong, Finotello and Angelova23] was utilized to evaluate the proportion of immune cells in subtypes, and the differences in the proportion of immune cells in subtypes were analyzed using the Kruskal–Wallis test with a threshold value of P < 0.05.
2.9. TMB Analysis
The Oncoplot function in Maftools [Reference Mayakonda, Lin, Assenov, Plass and Koeffler24] was used to visualize the top 20 genes with the highest mutation frequency in each subtype based on the genetic mutation information in each sample. The TMB value was calculated, and the differences in TMB values in subtypes were analyzed using the t-test with a threshold value of P < 0.05.
2.10. Validation Analysis
To verify that the miRNAs were related to melanoma, the GSE18509 and GSE24996 datasets were used as the validation datasets. The expression values of miRNAs in each sample in the GSE18509 and GSE24996 datasets were extracted, and the differences in expression levels between benign and metastatic groups were analyzed using the t-test.
3. Results
3.1. Identification of DEmiRNAs
Based on the GSE34460 and GSE35579 datasets, DEmiRNAs between benign vs. primary, metastatic vs. benign, and metastatic vs. primary were identified. As given in Table 1, 80, 107, and 24 DEmiRNAs were obtained between benign vs. primary, metastatic vs. benign, and metastatic vs. primary in the GSE34460 dataset, respectively, and 138, 127, and 78 DEmiRNAs were screened between benign vs. primary, metastatic vs. benign, and metastatic vs. primary in the GSE35579 dataset, respectively. Thus, a total of 132 and 209 overlapping DEmiRNAs were obtained in the GSE34460 and GSE35579 datasets, respectively. Then, totally 61 common DEmiRNAs were screened (Figure 2(a)). A total of 27 DEmiRNAs that overlapped in the same types were considered as miRNAs related to disease progression, including 18 and 9 miRNAs in down-down- and up-up types, respectively (Figure 2(b)). Heatmaps of the 27 DEmiRNAs in the GSE34460 and GSE35579 datasets are shown in Figures 2(c) and 2(d), respectively.
3.2. Prediction of Target Genes of miRNA and Enrichment Analysis
A total of 2330 miRNA-gene interaction pairs containing 26 miRNAs and 1490 genes were screened using miRWalk 2.0. Then, the disease-related genes screened in the GeneCards database were intersected with target genes of miRNA, and a total of 604 miRNA-gene interaction pairs were further obtained, including 26 miRNAs and 377 genes. The miRNA-gene network is shown in Figure 3(a). hsa-miR-106b-5p, hsa-miR-27b-3p, and hsa-miR-141-3p had higher degree and were regulated by numerous genes (Table 2). The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis showed that the target genes of 14 miRNAs were involved in the MAPK and PI3K-Akt signaling pathways, and the top5 enriched pathways are shown in Figure 3(b).
3.3. Survival Analysis
To evaluate the relationship between the 27 miRNAs and prognosis, survival analysis was conducted using the Kaplan–Meier curve method, and the results showed that four miRNAs were associated with OS (hsa-let-7c-5p, hsa-miR-130b-3p, hsa-miR-142-3p, and hsa-miR-509-3p) (all P < 0.05, Figure 4(a)), and three miRNAs were correlated with DSS (hsa-let-7c-5p, hsa-miR-130b-3p, and hsa-miR-142-3p) (all P < 0.05, Figure 4(b)); however, there were no miRNAs related to PFI.
3.4. Cluster Analysis of miRNAs
Bidirectional hierarchical clustering was conducted based on the expression value of the 27 miRNAs in each sample in TCGA-SKCM. As shown in Figure 5(a), the bidirectional hierarchical clustering of the 27 miRNAs was classified into three subtypes (clusters 1, 2, and 3), and most miRNAs were highly expressed in cluster 2; however, miRNAs showed low expression in clusters 1 and 3. The principal component analysis (PCA) of three subtypes is shown in Figure 5(b). Moreover, survival analysis showed that clusters 1 and 2 were associated with poor OS, and cluster 3 correlated with better OS (Figure 5(c)). Moreover, the Kaplan–Meier curve was drawn between the two clusters, and the results showed that clusters 3 and 1 and clusters 3 and 2 had significant differences (all P < 0.05, Figures 5(d) and 5(e)). There was no significant difference between clusters 1 and 2 (P < 0.05, Figure 5(f)).
3.5. Identification of DEGs between Subtypes
A total of 1846, 679, and 2257 DEGs were identified between clusters 1 and 2, clusters 1 and 3, and clusters 2 and 3, respectively, and 47 common DEGs were identified (Figure 6(a)). PCA of the 47 common DEGs in clusters 1, 2, and 3 is shown in Figure 6(b). Moreover, enrichment analysis was conducted on 47 common DEGs. These were enriched in 13 gene ontology (GO)-biology process (BP) (Figure 6(c)) and three KEGG pathways (osteoclast differentiation, amoebiasis, and rheumatoid arthritis) (Figure 6(d)).
3.6. Tumor-Infiltrating Immune Cells
The CIBERSORT algorithm was used to evaluate the proportion of subtypes of immune cells, and the differences in these proportions were analyzed using the Kruskal–Wallis test. The results showed significant differences in four subtypes of immune cells, including activated dendritic cells (P < 0.05), naïve CD4 T cells (P < 0.05), M1 macrophages (P=0.02), and plasma cells (P=0.02). A box diagram of the proportion of 22 subtypes of immune cells is shown in Figure 7.
3.7. TMB Analysis
The waterfall plot of the top 20 genes with the highest mutation frequency in each subtype is shown in Figure 8(a), which revealed that the gene mutation frequency in cluster 2 was low, the gene mutation frequency in cluster 1 was high, and TTN, MUC16, BRAF, and DNAH5 in the three subtypes had high mutation frequencies. In addition, the differences in TMB values in subtypes were analyzed, and the TMB value in cluster 2 was significantly lower than that in the other two subtypes (P < 0.05) (Figure 8(b)).
3.8. Validation Analysis
The GSE18509 and GSE24996 datasets were used as validation datasets to verify that the 27 miRNAs were related to melanoma. As none of the probes in the GSE18509 dataset matched hsa-miR-424-3p and hsa-miR-23b-5p, 25 miRNAs were verified. The heatmap showed that 25 miRNAs were well distinguished between samples from the benign and metastatic groups (Figure 8(c)), suggesting that 25 miRNAs were related to melanoma. The differences in expression levels between benign and metastatic groups were evaluated using the t-test, and the box diagram showed that the expression levels of most miRNAs were significantly different, and the regulation trend was consistent with the results of the training dataset (Figure 8(d)). Moreover, the expression of 27 DEmiRNAs was analyzed using the GSE24996 dataset. Because the two probes in the GSE24996 dataset were not matched, 25 miRNAs were verified. The results showed that the expression of 22 DEmiRNAs was significantly different among the benign, primary, and metastatic groups (Figures 8(e) and 8(f)), suggesting that the above results were reliable.
4. Discussion
miRNAs are aberrantly expressed in numerous cancers [Reference Lee and Dutta25], including melanoma. This study aimed to screen miRNA biomarkers of melanoma using data downloaded from public databases. hsa-miR-106b-5p, hsa-miR-27b-3p, and hsa-miR-141-3p had a higher degree and were regulated by numerous genes in the miRNA-gene network, and the target genes of miRNAs were involved in the MAPK and PI3K-Akt signaling pathways. Moreover, four miRNAs were associated with prognosis: hsa-let-7c-5p, hsa-miR-130b-3p, hsa-miR-142-3p, and hsa-miR-509-3p. Moreover, the bidirectional hierarchical clustering of the 27 miRNAs was classified into three subtypes, and TMB and four subtypes of immune cells, including activated dendritic cells, naïve CD4 T cells, M1 macrophages, and plasma cells, showed significant differences among the three subtypes.
Sixty-one common DEmiRNAs were screened between benign vs. primary, metastatic vs. benign, and metastatic vs. primary in the GSE34460 and GSE35579 datasets. Among these, 27 DEmiRNAs were related to disease progression. Moreover, the miRNA-gene network was constructed, and hsa-miR-106b-5p, hsa-miR-27b-3p, and hsa-miR-141-3p had a higher degree and were regulated by numerous genes. Chen et al. found that hsa-miR-106-5p promotes the cell cycle progression of malignant melanoma by targeting PTEN [Reference Chen, Chen, Chen, Ma, Shi and Zhou26]. Bonnu et al. revealed that hsa-miR-106b-5p regulates prostate cancer by interfering with the endoplasmic reticulum stress repair pathway and reducing the expression of tumor suppressor genes involved in many biological processes [Reference Bonnu, Ramadhani and Saputro27]. Liu et al. suggested that hsa-miR-106b-5p promotes colorectal cancer cell migration and invasion [Reference Liu, An, Cao, Li, Jia and Lei28]. Gencia et al. found that hsa-miR-27b-3p showed altered expression in all three melanoma types [Reference Gencia, Baderca and Avram29]. Miao et al. indicated that hsa-miR-27b-3p inhibits glioma development by targeting YAP1 [Reference Miao, Li, Gu, Yi, Su and Cheng30]. Verrando et al. found that transnonachlor reduces the level of hsa-miR-141-3p in human melanocytes in vitro and promotes the characteristics of melanoma cells [Reference Verrando, Capovilla and Rahmani31]. In addition, the KEGG pathway enrichment analysis showed that the target genes of 14 miRNAs were involved in the MAPK and PI3K-Akt signaling pathways. Chen et al. illustrated that FARP1 promotes cell proliferation by regulating the MAPK signaling pathway in cutaneous melanoma [Reference Chen and Wang32]. Wu et al. found that lncRNA OR3A4 facilitates the invasion and migration of melanoma cells through the PI3K/Akt signaling pathway [Reference Wu, Zhou, Yu, Wu and Xie33]. Thus, miRNAs may be involved in the progression of melanoma via the MAPK and PI3K-Akt signaling pathways.
Three miRNAs, hsa-let-7c-5p, hsa-miR-130b-3p, and hsa-miR-142-3p, were related to both OS and DFF, and hsa-miR-509-3p was associated with OS. Liu et al. found that hsa-let-7c-5p was associated with melanoma metastasis [Reference Liu, Tang, Chen, Jiang and Fang34]. Fu et al. suggested that hsa-let-7c-5p restrains breast cancer cell proliferation and facilitates apoptosis by targeting ERCC6 [Reference Fu, Mao, Wang, Ding and Li35]. Murria et al. identified miRNAs associated with melanoma survival, and the results suggested poorer survival in melanomas with hsa-miR-130b-3p overexpression [Reference Murria Estal, de Unamuno Bustos and Pérez Simó36]. Tembe et al. indicated that hsa-miR-142-3p is related with the prognosis of patients with metastatic melanoma [Reference Tembe, Schramm and Stark37]. Li et al. found that GPC6 is an early biomarker for metastatic progression of melanoma, which can be regulated by hsa-miR-509-3p [Reference Lisowska, Pindel and Pietruczuk38]. Patil et al. revealed that hsa-miR-509-3p inhibits the migration, invasion, and proliferation of osteosarcoma cells [Reference Patil, Palat and Pan39]. Our results were in line with those reported in previous studies, suggesting that hsa-let-7c-5p, hsa-miR-130b-3p, hsa-miR-142-3p, and hsa-miR-509-3p are associated with the prognosis of patients with melanoma.
Tumor-infiltrating immune cells play a vital role in promoting or inhibiting tumor growth [Reference Wang, Liu, Ly, Xu, Qu and Zhang40]. The bidirectional hierarchical clustering of the 27 miRNAs was classified into three subtypes, and four subtypes of immune cells, including activated dendritic cells, naïve CD4 T cells, M1 macrophages, and plasma cells, showed significant differences among the three subtypes. Among the immune cells involved in cancer immunity, properly activated plasmacytoid dendritic cells play a vital role in building a bridge between inherent and adaptive immune responses and directly eliminating cancer cells [Reference Monti, Consoli, Vescovi, Bugatti and Vermi41]. Dendritic cells with potent antigen-presenting abilities have long been considered a critical factor in antitumor immunity [Reference Veglia and Gabrilovich42]. Su et al. found that the abundance of naïve CD4+ T cells is closely related to Tregs, suggesting that the prognosis of breast cancer patients is poor [Reference Su, Liao and Liu43]. Tumor-associated macrophages could be used as treatment targets in oncology [Reference Mantovani, Marchesi, Malesci, Laghi and Allavena44, Reference Xia, Rao, Yao, Wang, Ning and Chen45]. Plasma cells are the main effectors of adaptive immunity and are responsible for producing antibodies to protect the body against pathogens. Numerous studies have reported that plasma cells are related to the pathogenesis of cancer, such as triple-negative breast cancer [Reference Bar, Theate and Haussy46], lung cancer [Reference Fujimoto, Yoshizawa and Sumiyoshi47], and colorectal cancer [Reference Berntsson, Nodin, Eberhard, Micke and Jirström48]. Moreover, TMB could be considered an immunotherapy biomarker for cancer [Reference Chan, Yarchoan and Jaffee49–Reference Barroso-Sousa, Jain and Cohen51]. In this study, the TMB value in cluster 2 was significantly lower than that in the other two subtypes, suggesting that these miRNAs might play important roles in the pathogenesis of melanoma by involving tumor-infiltrating immune cells and TMB.
However, this study has certain limitations. First, the data analyzed in this study were downloaded from public databases, and external validation is required. Moreover, the sample size should be expanded in future studies. Relevant experiments must be performed to verify the miRNAs identified in our bioinformatics analyses.
5. Conclusions
In summary, this study identified numerous novel miRNA biomarkers, including hsa-miR-106b-5p, hsa-miR-27b-3p, and hsa-miR-141-3p, which play important roles in melanoma and might provide new insights into the pathogenesis of melanoma.
Data Availability
The data supporting the findings of this study are available from the GEO database (GSE34460, GSE35579, GSE18509, and GSE24996 datasets).
Ethical Approval
The animal experiments in this study were approved by the Ethics Committee of the First Affiliated Hospital of Harbin Medical University.
Disclosure
Xuerui Wu and Yu Wang are the co-first authors.
Conflicts of Interest
The authors declare that there are no conflicts of interest.
Authors’ Contributions
XW, LC, and SZ conceptualized and designed the study. XW, CC, YX, and YW acquired data. LC, XW, and SZ analyzed and interpreted data, XW, CC, YX, and YW performed statistical analysis. LC obtained fund and revised the manuscript for important intellectual content. XW and SZ drafted the manuscript. All authors have read and approved the final manuscript. Xuerui Wu and Yu Wang contributed equally to this work.
Acknowledgments
This work was supported by the Research Innovation Fund of the First Affiliated Hospital of Harbin Medical University (2019M07).