Aberrant DNA methylation in ES cells

Abstract

Both mouse and human embryonic stem cells can be differentiated in vitro to produce a variety of somatic cell types. Using a new developmental tracing approach, we show that these cells are subject to massive aberrant CpG island de novo methylation that is exacerbated by differentiation in vitro. Bioinformatics analysis indicates that there are two distinct forms of abnormal de novo methylation, global as opposed to targeted, and in each case the resulting pattern is determined by molecular rules correlated with local pre-existing histone modification profiles. Since much of the abnormal methylation generated in vitro appears to be stably maintained, this modification may inhibit normal differentiation and could predispose to cancer if cells are used for replacement therapy. Excess CpG island methylation is also observed in normal placenta, suggesting that this process may be governed by an inherent program.

Figure 1. Excess methylation in differentiated mouse ES cells.

DNA from normal mouse tissues, undifferentiated and differentiated ES cells, post implantation embryos and teratomas were subject to mDIP microarray analysis. Heat map of 1,000(IMS>0.75) in at least one of the ES cell types out of 9,500 CpG islands constitutively unmethylated (IMS<0) in all tissue samples (see Materials and Methods). A number of different ES cell lines were used in this study. Embryoid bodies and retinoic acid treated cells were derived from TT2, while endoderm and mesoderm were derived from EB5. An estimate for the average percent of methylation in fetal tissues as compared to in vitro differentiated cells is also shown.

Figure 2. Excess methylation in human ES cells.

DNA from normal human fetal tissues, undifferentiated and in vitro differentiated ES cells were subject to mDIP microarray analysis. Heat map shows 950(IMS>0.75) in at least one of the differentiated ES cell types out of 13,000 CpG islands constitutively unmethylated (IMS<0) in fetal tissue samples. Retinoic acid treated cells and embryoid bodies were derived from CSES2. An estimate for the average percent methylation in fetal tissues as compared to in vitro differentiated cells is also shown.

Figure 3. De novo methylation is proportional to H3K27me3 concentration.

a. IMS (Fig. 1) of all 9,500 constitutively unmethylated CpG islands in undifferentiated, endoderm-differentiated ES cells and adult tissue DNA graphed as a function of H3K27me3 density . Each point represents the average IMS within a 5 unit span. b. Average methylation levels of all constitutively unmethylated CpG islands in endoderm (IMS), NPCs (%) and adult tissue (%) were graphed against H3K27me3 density partitioned into ten bins. The X-axis also shows average H3K27me3 levels in each percentile. c. Methylation levels in adult tissue, ES differentiated into endoderm and NPCs of all constitutively unmethylated CpG islands with above background concentrations (>2) of H3K27me3 graphed against their H3K4me3 density in ES.

Figure 4. Resetting de novo methylation in vivo.

Blastocysts injected with ES cells carrying a GFP expression vector were transplanted into pseudo-pregnant mice. Whole embryos were isolated at 16 dpc and sorted for GFP⁺ and GFP⁻ cells. DNA from these cells was then treated with bisulfite and deep-sequenced (Ion Torrent) at four different specific CpG island sequences. a. 7,000 individual molecules of island A containing nine individual CpGs with yellow indicating methylation. b. Graph showing percent methylation for islands A, B, C and D.