A multi-omics study on cutaneous and uveal melanoma
皮肤和脉络膜黑色素瘤的多组学研究

Multiple Genomic Alteration Analysis
多组学基因改变分析

In order to ensure the data quality, mutation annotation format (MAF) files were downloaded from the TCGA data portal via R. First, the package “TCGAbiolinks” was used to analyze the MAF files of CM (104 files) and UM (80 files) separately. Second, we pooled the MAF files together to identify the top 20 most frequently mutated genes of the combined group which were presented with additional information such as “disease”, “gender” and “race”. All variant coordinates were transferred to genome reference hg38.
为了确保数据质量，我们通过 R 从 TCGA 数据门户下载了突变注释格式（MAF）文件。首先，使用“TCGAbiolinks”包分别分析了 CM（104 个文件）和 UM（80 个文件）的 MAF 文件。其次，我们将 MAF 文件合并，以识别联合组中突变频率最高的前 20 个基因，并提供了“疾病”、“性别”和“种族”等额外信息。所有变异坐标均转换为基因组参考 hg38。

Differentially Expressed Genes
差异表达基因

First, we downloaded the combined mRNA data which contained 103 CM samples and 80 UM samples. The combined data was then analyzed with an Array Array Intensity correlation which defined a square symmetric matrix of Pearson correlation among all 183 samples. Second, the combined mRNA data was processed with within-lane normalization and between-lane normalization. Ultimately, with a threshold of 0.25, differentially expressed genes (DEGs) were identified through edgeR function of the package TCGAbiolinks.
首先，我们下载了包含 103 例皮肤黑色素瘤（CM）样本和 80 例眼内黑色素瘤（UM）样本的联合 mRNA 数据。随后，使用阵列阵列强度相关性对联合数据进行了分析，定义了一个包含所有 183 个样本的皮尔森相关性对称矩阵。其次，联合 mRNA 数据进行了同通道内归一化和不同通道间归一化处理。最终，通过“TCGAbiolinks”包中的 edgeR 函数，使用阈值 0.25 识别出差异表达基因（DEGs）。

Gene Ontology Enrichment Analysis
基因本体论富集分析

With the identified DEGs, we performed the gene ontology (GO) enrichment analysis in three perspectives, biological process, cellular component and molecular function. Besides, 10 most significantly enriched pathways were also identified. The results were plotted by the -log(P-value corrected FDR).
通过已识别的差异表达基因（DEGs），我们从生物过程、细胞成分和分子功能三个角度进行了基因本体（GO）富集分析。此外，还确定了 10 个最显著富集的通路。结果通过-log（调整后的 FDR P 值）进行了绘图。

DNA Methylation  DNA 甲基化

DNA methylation analysis of both CM and UM were conducted on the platform “Illumina Human Methylation 450”. We searched for differentially methylated CpG sites, which were regarded as possible functional regions involving in gene transcriptional regulation. In order to find these regions, the beta-values of methylation were used for the comparison. First, the calculation of the mean methylation of each tumor was performed. Second, we calculated the P-value by using a Benjamini-Hochberg adjusted Wilcoxon test. Finally, a heatmap of the methylation of CM and UM was generated with R.
在“ Illumina Human Methylation 450”平台上对 CM 和 UM 的 DNA 甲基化进行了分析。我们搜索了差异甲基化的 CpG 位点，这些位点被认为是可能参与基因转录调控的功能区域。为了找到这些区域，我们使用甲基化β值进行比较。首先，计算每个肿瘤的平均甲基化水平。其次，我们使用 Benjamini-Hochberg 调整后的 Wilcoxon 检验计算 P 值。最后，使用 R 生成 CM 和 UM 的甲基化热图。

Integration of DNA Methylation and Gene Expression Data
DNA 甲基化和基因表达数据的整合

We conducted the integration of DNA methylation and gene expression data with the results from step 2 and step 4 by using the “TCGAvisualize_starburst” function in the “TCGAbiolinks” package.
我们使用“TCGAbiolinks”包中的“TCGAvisualize_starburst”函数，将步骤 2 和步骤 4 的结果进行了 DNA 甲基化和基因表达数据的整合。

Differentially Expressed miRNAs
差异表达的 miRNA

We downloaded the combined miRNA data which contained 103 CM samples and 80 UM samples. By using the function “TCGAanalyze_DEA”, we identified the differentially expressed miRNAs between CM and UM patients. The “fdr.cut” and “logFC.cut” were set to be 0.01 and 1, respectively.
我们下载了包含 103 例皮肤黑色素瘤（CM）样本和 80 例眼内黑色素瘤（UM）样本的联合 miRNA 数据。使用“TCGAanalyze_DEA”功能，我们识别了 CM 和 UM 患者之间的差异表达 miRNA。“fdr.cut”和“logFC.cut”分别设置为 0.01 和 1。

ELMER Analysis on Identifying the Differentially Expressed Master Transcription Factors
ELMER 分析以识别差异表达的主转录因子

With the ELMER package, a Multi Assay Experiment (MAE) object including a DNA methylation matrix and a gene expression matrix was generated. For instance, in the hypomethylated direction, the samples of CM and UM were ranked by the beta values of the DNA methylation and 20% of the samples with the lowest methylation levels were processed with an unpaired one-tailed t-test to find out the comparatively hypomethylated probes in CM. Afterwards, the methylation of the probes and the expression of their 10 nearest upstream and downstream genes were tested for an inverse correlation. The differentially methylated probes were identified via the Mann-Whitney U test and the non-parametric test. Then, we integrated the significant probe-gene pairs to perform the motif enrichment analysis which finally helped to identify the candidate differentially expressed master regulatory transcription factors (TFs). The same method was also applied in the analysis of the hypermethylated direction.
使用 ELMER 包，生成了一个包含 DNA 甲基化矩阵和基因表达矩阵的多组学实验（MAE）对象。例如，在低甲基化方向上，根据 DNA 甲基化的 beta 值对 CM 和 UM 的样本进行排序，并对甲基化水平最低的 20% 样本进行配对单尾 t 检验，以找出 CM 中相对低甲基化的探针。之后，检测这些探针的甲基化水平与其 10 个最近的上游和下游基因的表达水平之间的反相关关系。通过 Mann-Whitney U 检验和非参数检验识别出差异甲基化探针，然后将显著的探针-基因对整合进行 motif 富集分析，最终识别出候选的差异表达主调控转录因子（TFs）。同样的方法也应用于高甲基化方向的分析。

Multiple Genomic Alteration Analysis
多重基因组改变分析

In Figure 1A and 2A, we noticed that single nucleotide polymorphism (SNP) was the most common variant type in both tumors when compared with insertion and deletion. Furthermore, in both tumors, “C>T” was the most frequent one in the six classes of single nucleotide variant (SNV) types. With regard to the variant classification, a same ranking order was observed in both groups, with missense mutation ranking first, followed by nonsense mutation, frameshift deletion and frameshift insertion. Besides, CM samples had a median variants' number of 223.5, with the maximum reaching more than 5000, whereas the median variants' number of UM was only 11 and the maximum was less than 400. This is concordant with a previous study which reported that CM had more mutational burden than UM^[9].
在 Figure 1A 和 2A 中，我们注意到与插入和删除相比，单核苷酸多态性（SNP）是两种肿瘤中最常见的变异类型。此外，在六类单核苷酸变异（SNV）类型中，“C>T”是最常见的。关于变异分类，两组中观察到相同的排名顺序，错义突变排名第一，其次是无义突变、移码缺失和移码插入。此外，CM 样本的变异中位数为 223.5，最大值超过 5000，而 UM 样本的变异中位数仅为 11，最大值不到 400。这与之前的一项研究一致，该研究报道 CM 的突变负担比 UM 更多 ^[9] 。

Furthermore, the CM and UM samples were combined together into a pooled analysis (Figure 3). We identified the 20 most frequently mutated genes in the combined group, with TTN (40%) ranking first, followed by MUC16 (37%), BRAF (29%), GNAQ (24%), PCLO (21%), DNAH5 (21%), GNA11 (20%), DNAH7 (18%), ADGRV1 (18%), LRP1B (17%), FAT4 (16%), XIRP2 (16%), CSMD1 (16%), DNAH9 (16%), HYDIN (16%), ANK3 (15%), MGAM (15%), THSD7B (15%), FLG (14%), ZFHX4 (14%). In the included 184 (104 CM and 80 UM) samples, 172 (93.48%) had at least one mutation in the abovementioned 20 genes.
此外，我们将 CM 和 UM 样本合并进行联合分析（ Figure 3 ）。我们确定了联合组中突变频率最高的 20 个基因，其中 TTN（40%）排名第一，其次是 MUC16（37%）、BRAF（29%）、GNAQ（24%）、PCLO（21%）、DNAH5（21%）、GNA11（20%）、DNAH7（18%）、ADGRV1（18%）、LRP1B（17%）、FAT4（16%）、XIRP2（16%）、CSMD1（16%）、DNAH9（16%）、HYDIN（16%）、ANK3（15%）、MGAM（15%）、THSD7B（15%）、FLG（14%）、ZFHX4（14%）。在包含的 184（104 个 CM 和 80 个 UM）样本中，有 172（93.48%）个样本在上述 20 个基因中至少有一个突变。

Interestingly, mutation on titin, or TTN, occurred in 71 (68%) of 104 CM samples, whereas only 2 (2.5%) of 80 UM samples were identified to have TTN mutations. In our study, 68 (65 CM and 3 UM) samples were identified to have MUC16 mutations. We also noticed an obvious absence of BRAF mutation in the UM group. In contrast to these top three mutated genes, 40 of the 44 GNAQ-mutated samples originated from UM, with the left four deriving from CM. Thirty-nine samples with mutated PCLO were detected in our study. Among them, 38 were CM samples and only one UM sample. Within the 38 mutated CM samples, the number of male samples (n=26) was more than twice as large as the number of female samples (n=12). Three cytoskeletal dynein genes DNAH5, DNAH7 and DNAH9 were also presented in the top 20 mutated genes. We noticed that only one UM sample was detected with DNAH7 mutation while the others were all CM samples. Similarly, few UM samples were found in the XIRP2-, ADGRV1-, LRP1B-, CSMD1-, HYDIN-, MGAM-, FLG-, ZFHX4-, ANK3- and FAT4-mutated groups.
有趣的是，在 104 例 CM 样本中有 71 例（68%）检测到了 TTN 基因的突变，而在 80 例 UM 样本中只有 2 例（2.5%）检测到了 TTN 基因的突变。在我们的研究中，共有 68 例样本检测到了 MUC16 基因的突变，其中 65 例为 CM 样本，3 例为 UM 样本。我们还注意到 UM 组中 BRAF 基因突变明显缺失。与这三种突变率最高的基因不同，44 例 GNAQ 突变样本中有 40 例来自 UM，其余 4 例来自 CM。我们研究中检测到了 39 例 PCLO 基因突变的样本，其中 38 例为 CM 样本，只有 1 例为 UM 样本。在 38 例 PCLO 突变的 CM 样本中，男性样本（n=26）的数量是女性样本（n=12）的两倍多。在前 20 种突变基因中，我们还发现了三个细胞骨架动力蛋白基因 DNAH5、DNAH7 和 DNAH9。我们注意到，只有 1 例 UM 样本检测到了 DNAH7 突变，其余均为 CM 样本。类似地，在 XIRP2、ADGRV1、LRP1B、CSMD1、HYDIN、MGAM、FLG、ZFHX4、ANK3 和 FAT4 突变的样本组中，UM 样本的数量也很少。

Transcriptomic Analysis  转录组分析

Totally 80 UM and 103 CM mRNA samples were included in the transcriptomic analysis. Finally, 4610 differentially expressed genes were identified. The top 20 differentially expressed genes (Table 1) were selected and ordered by the Delta value (the difference of gene expression between the two tumors multiplied logFC), i.e. KRT14, KRT6A, KRT16, KRT6B, KRT5, KRT17, KRT1, KRT6C, FN1, APOD, SERPINE2, KRT10, COL1A1, S100A9, SFN, TNC, S100A7, EEF2, S100A8, SPRR1B. Interestingly, 9 of the 20 identified genes belonged to the Keratin family which were abundant in the keratinocytes and were regarded as epithelial markers. And three members of S100 protein family (S100A7, S100A8 and S100A9) were also shown in Table 1.
总共包括了 80 个 UCM 和 103 个 CM 的 mRNA 样本进行转录组分析。最终共识别出 4610 个差异表达基因。选择并按Δ值（两肿瘤之间基因表达差异的对数倍数变化）排序的前 20 个差异表达基因，即 KRT14、KRT6A、KRT16、KRT6B、KRT5、KRT17、KRT1、KRT6C、FN1、APOD、SERPINE2、KRT10、COL1A1、S100A9、SFN、TNC、S100A7、EEF2、S100A8、SPRR1B。有趣的是，在这 20 个识别出的基因中有 9 个属于角蛋白家族，这些基因在角质细胞中丰富，被认为是上皮标志物。同时，S100 蛋白家族的三个成员（S100A7、S100A8 和 S100A9）也在其中。

GO Enrichment Analysis  GO 富集分析

Basing on the results of DEGs, we performed a GO enrichment analysis to classify the DEGs. Three GO types (biological processes, cellular components, molecular functions) and pathways were investigated. The results were ordered by -log₁₀(FDR).
基于 DEGs 的结果，我们进行了 GO 富集分析以分类 DEGs。研究了三种 GO 类型（生物过程、细胞成分、分子功能）和途径。结果按-log ₁₀ （FDR）排序。

In the 10 identified biological processes, we observed three processes with epidermal characteristics, i.e. the epidermal cell differentiation, epidermis development and keratinocyte differentiation. The DEGs also influence the cell and biological adhesion process. Additionally, two sensory organ-associated process, i.e. sensory perception and cognition, were also identified in the top 10 most significant biological processes.
在识别出的 10 个生物过程中，我们观察到 3 个具有表皮特征的过程，即表皮细胞分化、表皮发育和角质形成细胞分化。差异表达基因（DEGs）还影响细胞和生物黏附过程。此外，在前 10 个最显著的生物过程中，还识别出了两个与感觉器官相关的过程，即感觉感知和认知。

From the perspective of the cellular component, it's noteworthy that eight of the 10 components were associated with cellular membrane, including plasma membrane, extracellular region part, extracellular region, intrinsic to plasma membrane, integral to plasma membrane, proteinaceous extracellular matrix, extracellular matrix and extracellular space.
从细胞组分的角度来看，值得注意的是，在 10 个组分中有 8 个与细胞膜相关，包括细胞膜、细胞外区域部分、细胞外区域、膜内在部分、膜整合部分、蛋白质性细胞外基质、细胞外基质和细胞外空间。

With regard to the molecular function, the DEGs were enriched majorly in the binding functions. Despite the eight identified binding functions, the structural molecule activity and chemokine activity were also presented in the top 10 functions. Finally, the DEGs were integrated to identify the most significantly enriched pathways. We noticed that four of the 10 identified pathways were associated with the immune system, including the granulocyte adhesion and diapedesis, agranulocyte adhesion and diapedesis, T helper cell differentiation and the role of nuclear factor of activated T cells (NFAT) in regulation of the immune response. Other presented pathways like hepatic fibrosis/hepatic stellate cell activation and colorectal cancer metastasis signaling were also identified as DEGs-enriched pathways.
在分子功能方面，差异表达基因（DEGs）主要富集在结合功能中。尽管识别出了八种结合功能，但在前 10 种功能中，结构分子活性和趋化因子活性也有所体现。最后，将 DEGs 整合以识别最显著富集的信号通路。我们注意到，在识别出的 10 个通路中，有 4 个与免疫系统相关，包括中性粒细胞黏附和渗移、无粒细胞黏附和渗移、T 辅助细胞分化以及活化 T 细胞核因子（NFAT）在免疫反应调节中的作用。其他如肝纤维化/肝星状细胞激活和结直肠癌转移信号通路也作为 DEGs 富集的通路被识别出来。

DNA Methylation  DNA 甲基化

The methylation data of both tumors (104 CM and 80 UM samples) were acquired through the platform of Infinium Human Methylation 450. First, a comprehensive study was performed to check the mean DNA methylation level of the two tumors. A significant difference of DNA methylation level between the two tumors was confirmed by the t test with P<0.001. Second, a heatmap (Figure 4) was generated to visualize the DNA methylation level across all CM and UM samples. While notable methylation differences could be observed between CM and UM, no obvious disparities were detected in genders and races.
两种肿瘤（104 例皮肤黑色素瘤和 80 例眼内黑色素瘤样本）的甲基化数据通过 Infinium Human Methylation 450 平台获取。首先，进行了全面的研究以检查两种肿瘤的平均 DNA 甲基化水平。t 检验显示两肿瘤之间的 DNA 甲基化水平差异显著，P<0.001。其次，生成了热图（ Figure 4 ）以可视化所有 CM 和 UM 样本的 DNA 甲基化水平。虽然在 CM 和 UM 之间可以观察到明显的甲基化差异，但在性别和种族方面未发现明显差异。

DEGs and DNA Methylation
DEGs 和 DNA 甲基化

DNA methylation locating at promoters has been confirmed to have a silencing effect on the respective genes^[10]. To investigate the methylation differences between CM and UM, a starburst plot was produced based on the differentially methylated CpG sites and nearby genes (Figure 5). DNA methylation [log₁₀(FDR-corrected P-value)] and DEGs' expression [log₁₀(FDR-corrected P-value)] were plotted on the x-axis and y-axis, respectively. Both of the horizontal and vertical black dashed lines presented an FDR-adjusted P value of 10⁻⁵. We noticed that most DEGs were downregulated in UM. Within the downregulated DEGs, hypo- and hypermethylation presented no obvious difference.
位于启动子区域的 DNA 甲基化已被证实对相应基因具有沉默效应 ^[10] 。为了研究 CM 和 UM 之间的甲基化差异，基于差异甲基化的 CpG 位点和附近基因，生成了星爆图 Figure 5 。DNA 甲基化[对数 FDR 校正 P 值]在 x 轴上绘制，DEGs 的表达[对数 FDR 校正 P 值]在 y 轴上绘制。横纵坐标上的黑色虚线分别表示 FDR 调整后的 P 值为 10 ⁻⁵ 。我们注意到，大多数 DEGs 在 UM 中下调。在下调的 DEGs 中，低甲基化和高甲基化之间没有明显差异。

miRNA

Totally 485 differentially expressed miRNAs were identified. The top 20 miRNAs were presented and ordered by the absolute value of logFC in Table 2. Despite miR-124-1, the other 19 miRNAs were all downregulated in UM. Recent investigations indicated that miR-205, miR-206, miR-203a, miR-124, miR-105-1, miR-200c and miR-516b was significantly downregulated in several types of malignancies^[11]–^[16].
总共识别出 485 个差异表达的 miRNA。按 logFC 的绝对值排序，列出了前 20 个 miRNA。尽管 miR-124-1 下调，其他 19 个 miRNA 在 UM 中均下调。最近的研究表明，miR-205、miR-206、miR-203a、miR-124、miR-105-1、miR-200c 和 miR-516b 在多种恶性肿瘤中显著下调[ ^[11] – ^[16] ]。

Differentially Expressed Master Transcription Factors
差异表达主转录因子

In the hypomethylated (comparative hypomethylated in CM than in UM) direction, 39 enrichment motifs were identified while 66 were identified in the hypermethylated direction. Basing on the identified enriched motifs, we found 114 and 116 potential differentially expressed master TFs in hypomethylated and hypermethylated direction, respectively. Many of the identified master regulatory TFs were associated with the epithelial mesenchymal transition (EMT). Particularly, the acknowledged core EMT-TFs, i.e. SNAI1, SNAI2, TWIST1 and SIX1 were found to be relatively more hypomethylated in the CM than in the UM^[17].
在低甲基化方向（CM 相比 UM 低甲基化），共识别出 39 个富集 motif，而在高甲基化方向则识别出 66 个富集 motif。基于识别出的富集 motif，我们分别发现了 114 个和 116 个潜在差异表达的主调控 TFs，分别对应低甲基化和高甲基化方向。许多识别出的主调控 TFs 与上皮-间质转化（EMT）相关。特别是公认的 EMT 核心 TFs，即 SNAI1、SNAI2、TWIST1 和 SIX1，在 CM 中的低甲基化程度明显高于 UM。

The Role of Epithelial Mesenchymal Transition
上皮-间质转化的作用

The top 20 most frequently mutated genes supported the acknowledged key role of BRAF and GNAQ/11 in the etiopathogenesis of CM and UM, respectively^[18]–^[19]. While GNAQ/11 mutations were found almost uniquely in the UM, the majority of the other 18 mutated genes (including BRAF) were presented by the CM samples. Within the 18 mutated genes, MUC16, PCLO, DNAH5, DNAH7, DNAH9, ADGRV1, ANK3, FAT4 and XIRP2 encoded cytoskeleton- and extracellular matrix-related proteins (CECMPs) which were reported to provide the most common class of cancer mutants and estimated to play a role in the EMT^[20]–^[23].
最常见的 20 个突变基因支持了 BRAF 和 GNAQ/11 在 CM 和 UM 发病机制中的公认关键作用—— ^[18] – ^[19] 。虽然 GNAQ/11 突变几乎仅限于 UM，但其他 18 个突变基因（包括 BRAF）主要出现在 CM 样本中。在这 18 个突变基因中，MUC16、PCLO、DNAH5、DNAH7、DNAH9、ADGRV1、ANK3、FAT4 和 XIRP2 编码细胞骨架和细胞外基质相关蛋白（CECMPs），这些蛋白被认为是癌症中最常见的突变类型，并被认为在上皮-间质转化（EMT）中发挥作用—— ^[20] – ^[23] 。

On the transcriptomic level, it's noteworthy that eight of top 20-DEGs are keratin family members which were regarded as epithelial markers. And several previous studies implied that keratins may play a role in the progression of CM^[24]. Besides, the DEGs-listed FN1, SERPINE 2, TNC, EEF2 and COL1A1 were also regarded as EMT-associated genes which could contribute to the proliferation, differentiation and migration of tumors^[25]–^[29].
在转录组水平上，值得注意的是，最常见的 20 个差异表达基因中有 8 个是角蛋白家族成员，被认为是上皮标志物。此前的一些研究提示角蛋白可能在 CM 的进展中发挥作用—— ^[24] 。此外，差异表达基因列表中的 FN1、SERPINE 2、TNC、EEF2 和 COL1A1 也被认为是与 EMT 相关的基因，能够促进肿瘤的增殖、分化和迁移—— ^[25] – ^[29] 。

Additionally, 13 of the top 20 differentially expressed miRNAs were previously reported to participate in the EMT. And they were all less expressed in CM samples than in UM samples. Ten of the 13 miRNAs, i.e. miR-205, 520c, 520b, 203a, 520f, 122, 124, 526b and 200c, were reported to negatively regulate the EMT through a variety of downstream genes like CDH1, ZEB1, ZEB2, ADAM9, TGFBR2, SNAI2, ERK, MYC, CCND1, JUN and SNAI1 etc^[30]–^[37]. Thus, we estimate that the relative low expression of these miRNAs in CM would promote the EMT. Nevertheless, the other three miRNAs, miR-520g, 105 and 1269a were reported to positively regulate the EMT through cooperating with SMAD7, TNF, TGFB and SOX4^[38]–^[40]. Therefore, more in-depth studies are needed to help elucidate the role of miRNAs in CM and UM.
此外，20 个差异表达的 miRNA 中有 13 个之前被报道参与了上皮-间质转化（EMT）。这些 miRNA 在 CM 样本中的表达量均低于 UM 样本。在这 13 个 miRNA 中，有 10 个被报道通过多种下游基因（如 CDH1、ZEB1、ZEB2、ADAM9、TGFBR2、SNAI2、ERK、MYC、CCND1、JUN 和 SNAI1 等）负向调控 EMT ^[30] – ^[37] 。因此，我们估计这些 miRNA 在 CM 中的相对低表达会促进 EMT。然而，另外 3 个 miRNA（miR-520g、105 和 1269a）被报道通过与 SMAD7、TNF、TGFB 和 SOX4 等合作正向调控 EMT ^[38] – ^[40] 。因此，需要更深入的研究来阐明 miRNA 在 CM 和 UM 中的作用。

In addition, the role of the EMT in differentiating CM and UM was also confirmed by the analysis of differentially expressed master regulatory TFs. In the 114 identified hypomethylated master regulatory TFs (TFs hypomethylated in CM compared to UM), the first three TFs were LEF1, E2F2 and TRIM62. They were all suggested to positively regulate the EMT^[41]–^[43]. After a brief MeSH (Medical Subject Headings) search on “Google Scholar” (https://scholar.google.com/) until Nov. 01, 2018, at least 65 TFs were previously reported to participate in the EMT, with 57 positively-regulating TFs (including three acknowledged key EMT-TFs, SNAI1, TWIST1 and SIX1) and 8 negatively-regulating TFs.
此外，通过分析差异表达的主调控转录因子（TFs），EMT 在区分 CM 和 UM 中的作用也得到了确认。在 114 个被识别为低甲基化的主调控 TFs 中（与 UM 相比，CM 中的 TFs 低甲基化），前三位分别是 LEF1、E2F2 和 TRIM62。它们都被认为能够正向调节 EMT。截至 2018 年 11 月 1 日在“Google Scholar”上进行简要的 MeSH（医学主题词）搜索，至少有 65 个 TFs 被报道参与 EMT，其中 57 个是正向调节 TFs（包括三个公认的 EMT 关键 TFs，SNAI1、TWIST1 和 SIX1），8 个是负向调节 TFs。

Among the 116 TFs, which were identified as relatively more hypermethylated in CM when compared with UM, we found at least 42 EMT-associated TFs after a brief MeSH search on “Google Scholar”. While 19 of them were reported to negatively regulate the EMT, 23 were found to promote the EMT. Theoretically, the hypermethylation of the 19 negatively-regulating TFs would contribute to the EMT which would support our hypothesis that CM undergoes a more obvious EMT than UM. On the contrary, the hypermethylation of the 23 positively-regulating TFs were supposed to suppress the EMT. Thus, more relevant studies are needed to present a more precise landscape of the TFs' regulating network in these two tumors.
在 116 个被识别为在 CM 中相对高甲基化的 TF 中，我们通过在“Google Scholar”上进行简要的 MeSH 搜索，发现了至少 42 个与 EMT 相关的 TF。其中，有 19 个 TF 被报道负向调节 EMT，而 23 个 TF 被发现促进 EMT。理论上，19 个负向调节 TF 的高甲基化会促进 EMT，这支持了我们的假设，即 CM 比 UM 经历更明显的 EMT。相反，23 个正向调节 TF 的高甲基化可能会抑制 EMT。因此，需要更多的相关研究来呈现这两个肿瘤中 TF 调节网络的更精确图景。

In summary, our multi-omics results indicate that many distinctions between CM and UM were associated with the EMT.
总之，我们的多组学结果表明，CM 和 UM 之间的许多差异与上皮-间质转化（EMT）有关。

Metastatic Routes  转移途径

While CM was estimated to spread through hematologic as well as lymphatic dissemination, UM was believed to only metastasize hematogenously^[44]. The lymphatic endothelial cells were reported to secrete chemokines such as chemokine (C-C motif) ligand 21 (CCL21) and stroma cell derived factor-1 (SDF-1) which could stimulate the chemotaxis via the respective receptors C-C chemokine receptor 7 (CCR7) and leucine aminopeptidase 3 (LAP3) on tumor cells^[45]–^[46]. Besides, integrin α4β1 expressed on lymphatic vessels was also confirmed to play an essential role in the cancer-associated lymphangiogenesis through binding the corresponding ligands such as vascular cell adhesion molecule-1 (VCAM-1), fibronectin 1 (FN1)^[47]–^[48]. Because the lymphatic dissemination was not estimated to be involved in the metastasis of UM, we conject that the expression of CCR7, LAP3, VCAM-1 and FN1 might be lower expressed in UM when compared with CM. Supporting this, the DEGs' list showed that CCR7 (logFC=-2.378070086), LAP3 (logFC=-1.11), VCAM-1 (logFC=-1.63) and FN1 (logFC=-4.63) were all significantly lower expressed in UM. In addition, we also checked the expression of VEGFC and VEGFD in the DEGs' list because of the established involvement of VEGFC/D induced lymphangiogenesis in tumor metastasis^[49]. While VEGFC and VEGFD were not presented in the DEGs, we observed that VEGFA was lower expressed in UM with a logFC of -1.69 (i.e. significantly higher expressed in CM). Basing on a study in mouse, Björndahl et al^[50] and colleagues confirmed that the VEGFA induced lymphangiogenesis was independent from the VEGFC/D pathway. Thus, further studies are encouraged to examine the potential role of VEGFA in the lymph angiogenesis of CM.
虽然 CM 被认为通过血液和淋巴系统扩散，而 UM 则仅通过血液转移 ^[44] 。淋巴管内皮细胞被报道分泌趋化因子如 C-C 基序趋化因子配体 21（CCL21）和基质细胞衍生因子-1（SDF-1），这些趋化因子可通过相应的受体 C-C 基序趋化因子受体 7（CCR7）和亮氨肽酶 3（LAP3）刺激肿瘤细胞的趋化性 ^[45] – ^[46] 。此外，淋巴管上表达的整合素α4β1 也被证实通过结合相应的配体如血管细胞黏附分子-1（VCAM-1）、纤维连接蛋白 1（FN1）在与癌症相关的淋巴管生成中发挥重要作用 ^[47] – ^[48] 。由于淋巴系统扩散未被认为参与 UM 的转移，我们推测 UM 中 CCR7、LAP3、VCAM-1 和 FN1 的表达可能低于 CM。支持这一观点的是，差异表达基因列表显示，CCR7（logFC=-2.378070086）、LAP3（logFC=-1.11）、VCAM-1（logFC=-1.63）和 FN1（logFC=-4.63）在 UM 中的表达均显著降低。 此外，我们还检查了 DEGs 列表中 VEGFC 和 VEGFD 的表达情况，因为已有研究表明 VEGFC/D 诱导的淋巴管生成在肿瘤转移中起作用 ^[49] 。虽然 VEGFC 和 VEGFD 未出现在 DEGs 中，但我们发现 UM 中 VEGFA 的表达量较低，logFC 为-1.69（即在 CM 中显著高表达）。基于 Björndahl 等人的研究 ^[50] ，他们证实 VEGFA 诱导的淋巴管生成与 VEGFC/D 途径无关。因此，鼓励进一步研究 VEGFA 在 CM 淋巴管生成中的潜在作用。

Liver Tropism of the Metastatic UM
转移性 UM 的肝 tropism

Though CM has no metastatic propensity for the liver, metastatic UM has an obvious unexplained liver tropism. Previous studies of UM have shown that the increased expression of Met gene in the primary tumors would promote the liver metastasis^[51]. It was estimated that the liver propensity occurred because of the high HGF expression in the liver cell membrane surface. The activation of Met by binding hepatocyte growth factor (HGF) would lead to a series of downstream reactions which resulted in tumor cell proliferation and progression^[52]. Our DEGs' results confirmed the relatively higher expression of Met in UM than in CM (logFC=2.04). Besides, Krüger et al^[53] reported that the inhibition of matrix metalloproteinases (MMPs) would increase the expression level of HGF in the liver, thus promote the metastasis formation. In agreement with this, 13 MMP mRNAs in the DEGs were lower expressed in UM than in CM.
尽管皮肤黑色素瘤（CM）没有向肝脏转移的倾向，但黏膜黑色素瘤（UM）却有明显的、无法解释的肝脏倾向。先前对 UM 的研究表明，原发肿瘤中 Met 基因的高表达会促进肝脏转移 ^[51] 。据估计，这种肝脏倾向是因为肝脏细胞膜表面 HGF 表达较高。HGF 与 Met 结合后激活 Met，会引发一系列下游反应，导致肿瘤细胞的增殖和进展 ^[52] 。我们的差异表达基因（DEGs）结果证实，UM 中 Met 的表达量明显高于 CM（logFC=2.04）。此外，Krüger 等人 ^[53] 报道，抑制基质金属蛋白酶（MMPs）会增加肝脏中 HGF 的表达水平，从而促进转移灶的形成。与此一致的是，DEGs 中有 13 种 MMP mRNA 在 UM 中的表达量低于 CM。

Moreover, basing on a study on UM, Laurent et al^[54] have reported that the high expression of PTP4A3 could be seen as a predictor for liver tropism. Correspondingly, our study has proved that the PTP4A3 mRNA was significantly higher expressed in UM than in CM (logFC=1.91).
此外，基于对 UVM 的研究，Laurent 等人 ^[54] 报道高表达的 PTP4A3 可以被视为肝倾向的预测因子。相应地，我们的研究证明在 UVM 中 PTP4A3 mRNA 的表达显著高于 CM（logFC=1.91）。

In summary, the comparison between CM and UM from multiple perspectives presented a comprehensive illustration which provides us a better understanding of these two tumors and new insights for finding better treatment strategies.
综上所述，从多个角度比较 CM 和 UM，为我们全面展示了这两种肿瘤的特点，使我们对其有了更深入的理解，并为寻找更好的治疗策略提供了新的见解。