File S1 - Aberrant DNA Methylation in ES Cells

Supporting Information. Figure S1. Bisulfite methylation analysis of mouse ES cells. a. Heat map of RRBS methylation data for normal mouse tissues, undifferentiated and differentiated ES cells (NPC) and implantation embryos showing the top 2,000 CpG islands found modified (>25%) in NPCs. It should be noted that 70% of the aberrantly-methylated CpG islands detailed in Fig. 1 overlap with the set shown here. b. Methylation distribution in undifferentiated ES cells for 9,500 constitutively unmethylated (gray) CpG islands and the 2,000 CpG islands shown in a that were found to be methylated in NPCs (blue). c. Methylation distribution (IMS, derived from Fig. 1) in undifferentiated ES cells for the 9,500 constitutively unmethylated CpG islands (gray) as opposed to the 1,000 CpG islands (green) methylated following in vitro differentiation to endoderm. Figure S2. DNA methylation in human ES cells. Heat map analysis of CpG islands in human ES cells and normal tissues as determined by RRBS (Road map). Group A includes CpG islands that are abnormally methylated (>60%) in at least one undifferentiated or differentiated ES cell type but constitutively unmethylated (Figure S3. Scatter plot of H3K27me3 density in NPCs vs. undifferentiated mouse ES cells for 9,500 CpG islands (see Fig. S1). The 2,000 aberrantly methylated islands (marked in red) show a dramatic reduction in H3K27me3 density when ES cells are converted to NPCs (a). H3K27me3 density in mouse brain is shown for comparison (b). Figure S4. Markers of de novo methylation. a. Table showing the percentage of abnormally-methylated CpG islands that are marked with polycomb (H3K27me3>2) and their average algorithm score (A2) both in human and mouse ES cells. IMS data are from Figs. 1 and 2 and methylation sequencing data are from Figs. S1 and S2. The intrinsic ability of any CpG island to protect against de novo methylation can be expressed as an algorithm that takes into consideration underlying sequence features. Islands that are constitutively unmethylated have a high score (average of 2.0), while methylated islands have a low score (−1.80). Differentiated mouse ES-cell methylation targets (Fig. 1) have, on average, an intermediate score (0.8) that is significantly different (P−63) than the constitutively unmethylated CpG islands. b. Methylation levels of all CpG islands as a function of local H3K4me3 density for human ES cells before and after differentiation to endoderm. The average levels in a collection of fetal tissues is shown for comparison. Figure S5. DNA methylation in placenta. DNA from normal mouse embryos and from placenta were subject to mDIP microarray analysis. Heat map of 1,000 CpG islands methylated in placenta out of 9,500 background CpG islands (see Fig. 1). An estimate for the average percent methylation is also shown. Figure S6. DNA methylation in mouse pancreatic α and β cells. DNA from mouse pancreatic cell lines (TC and Min6) and natural β cells purified form fresh pancreatic islets were subject to mDIP microarray analysis. Out of 9,500 background CpG islands (Fig. 1), IMS Heat map of 2,000 CpG islands deemed methylated (IMS>0.75) in either the α or β cell lines as compared to DNA from embryos (where none have a positive binary methylation score). Only 5 are actually methylated in ex-vivo β cells. An estimate for the average percent methylation is also shown. Ex-vivo cells were obtained by preparing pancreatic islets from 2 to 8 month old transgenic mice carrying a Pdx1-GFP construct by ductal perfusion with collagenase. Islets were then hand-picked and dissociated to single cells with trypsin and subjected to FAC sorting using Anti-insulin antibodies. Figure S7. Methylation levels in cancer. Heat map of CpG islands deemed methylated in human ES cells (Fig. 2) compared to methylation levels (P value) in a number of different tumor samples from the Cancer Genome Atlas as determined by Infinium Human Methylation 450 array assay and methylation levels (%) in normal tissues as determined by RRBS from the Roadmap Epigenomics Project (www.roadmapepignenomics.org). The excess methylation seen in each tumor type is significantly greater than that observed in normal tissues (minimal P value−80). Table S1. Genes involved in differentiation and development that are associated with aberrantly methylated CpG islands. (DOC)

补充材料。
图S1 小鼠胚胎干细胞的亚硫酸氢盐甲基化分析（Bisulfite methylation analysis）。
a. 针对正常小鼠组织、未分化及分化后胚胎干细胞（神经前体细胞，NPC, Neural Progenitor Cells）、着床期胚胎的限制性位点关联亚硫酸氢盐测序（RRBS, Restriction Site-Associated Bisulfite Sequencing）甲基化数据绘制的热图（heat map），展示了在NPC中发生甲基化修饰（修饰率>25%）的前2000个CpG岛（CpG island）。需注意，图1中详述的异常甲基化CpG岛中有70%与本图展示的集合存在重叠。
b. 未分化胚胎干细胞中，9500个组成型未甲基化CpG岛（以灰色标注）以及图a中展示的2000个在NPC中被检测到甲基化的CpG岛（以蓝色标注）的甲基化分布情况。
c. 未分化胚胎干细胞中，9500个组成型未甲基化CpG岛（灰色）的甲基化分布（IMS数据源自图1），与之相对的是1000个经体外诱导分化为内胚层后发生甲基化的CpG岛（绿色）。
图S2 人类胚胎干细胞中的DNA甲基化。通过RRBS检测得到的人类胚胎干细胞与正常组织中的CpG岛热图分析（表观基因组路线图计划，Roadmap Epigenomics Project）。A组包含至少在一种未分化或分化的胚胎干细胞类型中发生异常甲基化（修饰率>60%）但组成型未甲基化的CpG岛。
图S3 针对9500个CpG岛（参见图S1）的NPC与未分化小鼠胚胎干细胞中H3K27me3密度散点图。2000个异常甲基化CpG岛（以红色标注）在胚胎干细胞转化为NPC后，其H3K27me3密度出现显著降低（a子图）；以小鼠脑组织中的H3K27me3密度作为对照展示（b子图）。
图S4 从头甲基化标志物。
a. 表格展示了人类与小鼠胚胎干细胞中，被多梳蛋白复合物标记（H3K27me3>2）的异常甲基化CpG岛占比，以及它们的平均算法评分（A2）。IMS数据源自图1与图2，甲基化测序数据源自图S1与图S2。任意CpG岛抵御从头甲基化的内在能力可通过一种考虑了底层序列特征的算法进行量化：组成型未甲基化的CpG岛评分较高（平均分为2.0），而发生甲基化的CpG岛评分较低（平均分为-1.80）；经分化的小鼠胚胎干细胞甲基化靶标（图1）平均得分为中间值0.8，与组成型未甲基化CpG岛的评分存在显著差异（P<10^-63）。
b. 人类胚胎干细胞在体外分化为内胚层前后，所有CpG岛的甲基化水平与局部H3K4me3密度的相关性。以一批胎儿组织的平均甲基化水平作为对照展示。
图S5 胎盘中的DNA甲基化。对正常小鼠胚胎及胎盘的DNA进行甲基化DNA免疫沉淀微阵列（mDIP microarray, methylated DNA immunoprecipitation microarray）分析。基于9500个背景CpG岛（参见图1），展示了胎盘中发生甲基化的1000个CpG岛的热图，同时展示了平均甲基化百分比的估算值。
图S6 小鼠胰腺α与β细胞中的DNA甲基化。对小鼠胰腺细胞系（TC与Min6）以及从新鲜胰岛中纯化得到的天然β细胞的DNA进行mDIP微阵列分析。在9500个背景CpG岛（图1）中，展示了在α或β细胞系中被判定为甲基化（IMS>0.75）的2000个CpG岛的IMS热图，与胚胎DNA（其中无二元甲基化评分阳性的位点）进行对照；而在离体β细胞中实际仅有5个位点发生甲基化，同时展示了平均甲基化百分比的估算值。
离体细胞的获取方式如下：通过胶原酶导管灌注法，从2至8月龄携带Pdx1-GFP转基因载体的小鼠中分离胰腺胰岛；随后手工挑选胰岛，用胰酶解离为单个细胞，再通过抗胰岛素抗体进行荧光激活细胞分选（FAC sorting, Fluorescence-Activated Cell Sorting）。
图S7 癌症中的甲基化水平。以人类胚胎干细胞中被判定为甲基化的CpG岛（图2）为对象绘制热图，与来自癌症基因组图谱（Cancer Genome Atlas）的多种肿瘤样本的甲基化水平（P值）进行对比：肿瘤样本的甲基化水平通过Infinium Human Methylation 450芯片检测得到，正常组织的甲基化水平则通过表观基因组路线图计划（Roadmap Epigenomics Project，www.roadmapepigenomics.org）中的RRBS检测得到。每种肿瘤类型中观察到的过度甲基化程度均显著高于正常组织中的水平（最小P值<10^-80）。
表S1 与异常甲基化CpG岛相关的、参与分化与发育过程的基因。（DOC格式文件）