LETTERS Epigenetic reversion of post-implantation epiblast to pluripotent embryonic stem cells Siqin Bao1*, Fuchou Tang1*, Xihe Li2, Katsuhiko Hayashi1{, Astrid Gillich1, Kaiqin Lao3 & M. Azim Surani1 The pluripotent state, which is first established in the primitive ectoderm cells of blastocysts, is lost progressively and irreversibly during subsequent development1. For example, development of post-implantation epiblast cells from primitive ectoderm involves significant transcriptional and epigenetic changes, including DNA methylation and X chromosome inactivation2, which create a robust epigenetic barrier and prevent their reversion to a primitive-ectoderm-like state. Epiblast cells are refractory to leuk- aemia inhibitory factor (LIF)���STAT3 signalling, but they respond to activin/basic fibroblast growth factor to form self-renewing epi- blast stem cells (EpiSCs), which exhibit essential properties of epi- blast cells3,4 and that differ from embryonic stem (ES) cells derived from primitive ectoderm5. Here we show reprogramming of advancedepiblastcellsfromembryonicday5.5���7.5mouseembryos with uniform expression of N-cadherin and inactive X chro- mosome to ES-cell-like cells (rESCs) in response to LIF���STAT3 signalling. Cultured epiblast cells overcome the epigenetic barrier progressively as they proceed with the erasure of key properties of epiblast cells, resulting in DNA demethylation, X reactivation and expression of E-cadherin. The accompanying changes in the tran- scriptome result in a loss of phenotypic and epigenetic memory of epiblast cells. Using this approach, we report reversion of estab- lished EpiSCs to rESCs. Moreover, unlike epiblast and EpiSCs, rESCs contribute to somatic tissues and germ cells in chimaeras. Further studies may reveal how signalling-induced epigenetic reprogramming may promote reacquisition of pluripotency. Previous studies showed that epiblast cells, unlike primitive ecto- derm cells, are refractory to LIF���STAT3 signalling instead they respond to activin/basic fibroblast growth factor (bFGF) to generate EpiSCs3,4, which are more like epiblast with an inactive X chro- mosome, and differ from ES cells. Here we re-examined advanced post-implantation epiblast cells to see if they could revert to ES-cell- like cells in response to LIF���STAT3 signalling. First, epiblast tissue carrying an Oct4���DPE���green fluorescent protein (GFP) reporter6 was isolated from mouse embryos on embr- yonic days (E) E5.5���E7.5. This reporter, with only the distal enhancer for Oct4 (also known as Pou5f1), shows preferential expression in the primitive ectoderm, primordial germ cells (PGCs) and ES cells, but not in the epiblast or EpiSCs6. Notably, the distal enhancer constitutes an ���enhanceosome��� representing the densest binding locus for the key pluripotency-specific transcripts in ES cells7, which makes it likelythat its activation will only occur if all pluripotency factors are expressed optimally some of these must be lacking or suboptimal in the epiblast and EpiSCs. Next, the epiblast tissue was dissected to remove the most prox- imal region (the site of existing PGCs and PGC precursors2) and the outer visceral endoderm (Fig. 1a). Notably, unlike previous studies where the epiblast tissue was left intact3,4, we trypsin-digested indi- vidual epiblast into a single-cell suspension to break up existing cell��� cell interactions and promote establishment of a new signalling- induced transcriptional network in vitro. The resulting cells were cultured in LIF���fetal calf serum (FCS) on mouse embryonic fibro- blast (MEF) feeder cells, a standard condition used for the derivation of ES cells from primitive ectoderm, and of induced pluripotent stem (iPS) cells from somatic cells5,8,9. After 4���7 days, most cultures revealed large colonies (Fig. 1a and Supplementary Table 1) with many alkaline-phosphatase-positive cells (Fig. 1a), but no detectable Oct4���DPE���GFP expression, indicating that the distal enhancer was yet inactive (Fig. 1a). We could propagate the cultured epiblast (cEpi) colonies in LIF���FCS after collagenase treatment without detectable changes for at least 20 passages. After culture of cEpi cells for 14���35 days, we started to detect clusters of GFP-positive cells within cEpi colonies (Fig. 1b, c), indi- cating activation of the distal enhanceosome of the Oct4���DPE���GFP reporter6. Subsequent culture of GFP-positive cells was carried out after disruption of cEpi colonies by treatment with trypsin, which is detrimental for cEpi cells but promotes propagation of ES-cell-like cells. With further passaging, we established ES-cell-like cells with uniform GFP expression (Fig. 1b). We call these cells reprogrammed epiblast ES-cell-like cells (rESCs). The frequency of rESC derivation was relatively high at around 22��� 36%, which notably did not diminish with developmental age from E5.5 to E7.5 (Supplementary Table 1). Furthermore, the epiblast cells at the start were uniformly negative for Oct4���DPE���GFP expression and positive for N-cadherin and inactive X-chromosome (see below). If reversion had occurred from rare epiblast cells, we might have seen a reduction in the frequency of rESC derivation. Notably, no ES-cell-like cells have previously been reported from embryos as late as E7.5 rather, they have only been derived from primitive ecto- derm present in the inner cell mass in blastocysts5,7,9. Furthermore, pluripotent EPL10 and FAB-SC11 were derived from pre-implantation or implanting blastocysts and not from post-implantation embryos (Supplementary Table 2) their epigenetic state was not reported. To gain an insight into the reversion process, we examined changes in gene expression (Fig. 2a, b). Transcriptome analysis revealed that cEpi cells, like EpiSCs3,4, were closely related to the epiblast. Thus, cEpi cells showed strong expression of Eomes, Fgf5, Sox17, Gata6, Lefty1 and Cer1. However, there was little expression of stella (also called Dppa3), Pecam1, Rex1 (also called Zfp42) and Fbxo15, but their expression increased in rESCs with a concomitant loss of Fgf5, Eomes and Sox17 (Fig. 2a). Expression of key pluripotency genes Oct4, Sox2 and Nanog also increased in rESCs (Fig. 2a and Supplementary Fig. 1, 2b). Expression of Eomes and Fgf5 was slightly higher in early passage rESCs (passage 4 (P4)) compared to the levels in P24 cells 1Wellcome Trust/Cancer Research UK Gurdon Institute, University of Cambridge, Tennis Court Road, Cambridge, CB2 1QN, UK. 2College of Life Science, Inner Mongolia University/ Mengniu RB CO. Ltd., West No. 1 Daxue Road, Huhhot, Inner Mongolia 010021, China. 3Molecular Cell Biology, Applied Biosystems, 850 Lincoln Centre Drive, Foster City, California 94404, USA. {Present address: Department of Anatomy and Cell Biology, Graduate School of Medicine, Kyoto University, Yoshida-Konoe-Cho, Sako-Ku, Kyoto 606-8501, Japan. *These authors contributed equally to this work. Vol 461|29 October 2009|doi:10.1038/nature08534 1292 Macmillan Publishers Limited. All rights reserved ��2009

(Fig. 2a), suggesting a loss of residual ���memory��� of their epiblast origin (see below). Thus, comprehensive microarray analysis con- firmed that rESCs are similar to ES cells and differ from cEpi cells and EpiSCs (Fig. 2b and Supplementary Fig. 3). These changes in the transcriptome are consistent with the distal enhancer-driven activa- tion of the Oct4���DPE���GFP reporter in rESCs. We next asked if LIF���STAT3 signalling is critical for reprogram- ming, and found that STAT3 is phosphorylated in cEpi cells, suggest- ing that they respond to LIF signalling (Supplementary Fig. 4a). Notably, addition of the Janus protein tyrosine kinase inhibitor (JAK inhibitor 1 Calbiochem) that prevents phosphorylation of tyrosine 705 of STAT3 initially allowed some cEpi colonies to develop, but they gradually differentiated and failed to form rESCs (Sup- plementary Table 3). Thus, LIF���STAT3 seems to be crucial for the propagation and reprogramming of cEpi cells to rESCs. Furthermore, culture of rESCs with the JAK inhibitor caused a marked reversal towards the cEpi-like transcriptome (Supplementary Fig. 4b). A num- ber of STAT3 targets and their expression have been identified in ES cells7, including Fbxo15, Rex1 and Stat3 itself, as well as the epigenetic modifiers Lin28, Ezh2 and Mbd3, suggesting that STAT3 has the poten- tial to influence the transcriptional and epigenetic state of cEpi cells. Next, we examined epigenetic changes in the epiblast during rever- sion to rESCs. Notably, reactivation of the late-replicating inactive X chromosome12,13 during reversion to rESCs would indicate a major epigenetic change14. Consistently, we found that all the E6.5 epiblast cells (96 of 96 cells Supplementary Fig. 5a) had the characteristic accumulation of histone H3 lysine 27 trimethylation (H3K27me3) associated with the inactive X chromosome (ref. 15). After 12 days, nearly all cEpi cells (99 of 100 cells Fig. 3a) still had the H3K27me3 ���spot���, which declined to 62% after 25 days (37 of 60 cells), and to only 9% after 35 days of culture (7 of 80 cells). This suggests continu- ing epigenetic reprogramming towards rESCs, which uniformly lacked the H3K27me3 ���spot���. These observations on H3K27me3 sug- gest initiation of X reactivation, a hallmark of epigenetic reprogram- ming, although further studies are needed to confirm completion of the process. Similar changes are also seen after somatic nuclear trans- plantation into oocytes, in mouse iPS cells, and in ES-cell-somatic cell hybrids16���19. We also examined DNA methylation of the promoter regions of stella and Rex1 both of these genes (and others like Pecam1) are repressed in the epiblast but active in primitive ectoderm and ES cells20. Although stella and Rex1 were unmethylated in the epiblast, they became transiently methylated in cEpi cells before undergoing demethylation (Fig. 3b), which is consistent with the activation of the Stella���GFP reporter in rESCs (Supplementary Fig. 2a). Once estab- lished, the rESC epigenotype was stable, heritable and distinct from EpiSCs, which retain key properties of epiblast cells, including inac- tive X, and in which stella and Rex1 are methylated and repressed (Figs 2a and 3b). These findings are relevant to human ES cells, which resemble mouse EpiSCs but not mouse ES cells or rESCs. To observe thedynamic nature of reprogramming to rESCs, we also examined changes in the expression of E-cadherin and N-cadherin (Fig. 4a). Whereas expression of both E-cadherin and N-cadherin was detected in E6.5 epiblast uniformly, trypsin-digestion of these cells before culture led to the loss of these adhesion molecules. During subsequent culture, we detected heterogeneous N-cadherin expres- sion in cEpi cells, but in continuing culture there was a complete loss of N-cadherin, which was replaced by uniform expression of E-cadherin in rESCs, consistent with the evidence that LIF���STAT3 promotes upregulation of E-cadherin11. Next, we asked if rESCs might have originated from early PGCs because they can undergo dedifferentiation into pluripotent embryonic germ cells that are similar to ES cells. To reduce this likelihood, the epiblast tissue from E6.5���E7.5 mouse embryos was dissected away from the most proximal region, the site of PGC precursors and PGC, respec- tively.Inparticular,E7.5epiblastalsoshowsalossofcompetencetoform additional PGCs21. Second, PGCs require bFGF and stem cell factor for proliferation and dedifferentiation into pluripotent embryonic germ cells, and cannot survive in culture conditions used for cEpi cells. Furthermore, rESCs, unlike embryonic germ cells, retain methylation of imprinted genes22,23 (Supplementary Fig. 5c). Thus, cumulative evidence makes it unlikely that rESCs could originate from PGCs. In view of our observations, we asked if established EpiSCs cultured in activin/bFGF could also undergo reversion to rESCs in response to LIF���STAT3. We chose two EpiSC lines: one with the Oct4���DPE���GFP reporter (passage 20) and the other with an X-GFP reporter the latter was FACS sorted to establish lines with repressed reporter on the inactive X chromosome (passage 23) (Supplementary Fig. 5b). When these EpiSCs with stably repressed GFP reporters were cultured for 10���20 days (Supplementary Fig. 6a, b), we detected GFP-positive a b c E6.5 embryo Epiblast Single-cell suspension Day 6 cEpi colony Day 20 in culture Day 21 in culture Day 22 in culture rESC AP staining AP staining Oct4�����PE���GFP+ Oct4�����PE���GFP��� Oct4�����PE���GFP+ Epiblast Single cells cEpi rESC Self-renew 2���5 weeks Self-renew Figure 1 | Reprogramming epiblast cells from mouse E6.5 embryos to generate rESCs. a, Derivation of cEpi cells from E6.5 epiblast. Epiblast tissue had the proximal region (white line, far left panel) and visceral endoderm removed a single-cell suspension (black arrow) was cultured, which formed cEpi colonies. Note alkaline phosphatase (AP)-positive cells in cEpi colonies in the far right panel. b, Derivation of rESCs from cEpi colonies. Note the appearance of clusters of Oct4���DPE���GFP-positive cells in cEpi colonies (black arrowheads), and corresponding white arrowheads for GFP in the panel below. Scale bars, 100 mm (a and b). c, Schematic representation of reprogramming of epiblast through cEpi cells, and finally rESCs in response to LIF���STAT3 signalling. NATURE|Vol 461|29 October 2009 LETTERS 1293 Macmillan Publishers Limited. All rights reserved ��2009