Epigenetic memory in induced pluripotent stem cells.

Somatic cell nuclear transfer and transcription-factor-based reprogramming revert adult cells to an embryonic state, and yield pluripotent stem cells that can generate all tissues. Through different mechanisms and kinetics, these two reprogramming methods reset genomic methylation, an epigenetic modification of DNA that influences gene expression, leading us to hypothesize that the resulting pluripotent stem cells might have different properties. Here we observe that low-passage induced pluripotent stem cells (iPSCs) derived by factor-based reprogramming of adult murine tissues harbour residual DNA methylation signatures characteristic of their somatic tissue of origin, which favours their differentiation along lineages related to the donor cell, while restricting alternative cell fates. Such an 'epigenetic memory' of the donor tissue could be reset by differentiation and serial reprogramming, or by treatment of iPSCs with chromatin-modifying drugs. In contrast, the differentiation and methylation of nuclear-transfer-derived pluripotent stem cells were more similar to classical embryonic stem cells than were iPSCs. Our data indicate that nuclear transfer is more effective at establishing the ground state of pluripotency than factor-based reprogramming, which can leave an epigenetic memory of the tissue of origin that may influence efforts at directed differentiation for applications in disease modelling or treatment.

Introduction

Direct reprogramming of somatic cells with the transcription factors Oct4, Sox2, Klf4, and c-Myc[1] yields induced pluripotent stem cells (iPSC) with striking similarity to embryonic stem cells from fertilized embryos (fESC). Like fESC, iPSC form teratomas, differentiated tumors with tissues from all three embryonic germ layers, and when injected into murine blastocysts contribute to all tissues, including the germ line. iPSC from mouse embryo fibroblasts generate “all-iPSC mice” following injection into tetraploid blastocysts[2], thereby satisfying the most stringent criterion of pluripotency[3]. Embryonic tissues are the most efficiently reprogrammed, producing iPSC that are nearly identical to fESC. In contrast, reprogramming from accessible adult tissues, most applicable for modeling diseases and generating therapeutic cells, is inefficient and limited by barriers related to the differentiation state and age of the donor’s cells[456]. Aged cells have higher levels of Ink4/Arf, which limits the efficiency and fidelity of reprogramming[5]. Moreover, terminally differentiated blood cells reprogram less efficiently than blood progenitors[6]. As with cloning by nuclear transfer in frogs and mice, the efficiency and yield of reprogrammed genomes declines with increasing age and differentiation status of the donor cell[7], and varies with the methylation state of the donor nucleus[8].

Different tissues show variable susceptibility to reprogramming. Keratinocytes reprogram more readily than fibroblasts[9], and iPSC from stomach or liver cells harbor fewer integrated proviruses than fibroblasts, suggesting they require lower levels of the reprogramming factors to achieve pluripotency[10]. When differentiated into neurospheres, iPSC from adult tail-tip fibroblasts retain more teratoma-forming cells than iPSC from embryonic fibroblasts, again indicating heterogeneity based on the tissue of origin[11]. Moreover, cells can exist in intermediate states of reprogramming that interconvert with continuous passage or treatment with chromatin-modifying agents[1213]. Although generic iPSC are highly similar to fESC, in practice iPSC generated from various tissues may harbor significant differences, both functional and molecular.

Transcription factor reprogramming differs markedly from nuclear transfer, particularly with regard to DNA demethylation, which commences immediately upon transfer of a somatic nucleus into ooplasm[14], but occurs over days to weeks during the derivation of iPSC[13]. Because demethylation is a slow and inefficient process in factor-based reprogramming, we postulated that residual methylation might leave iPSC with an “epigenetic memory,” and that methylation might be more effectively erased by nuclear transfer. Here we compare the differentiation potential and genomic methylation of pluripotent stem cells (iPSC, ntESC, and fESC), and find evidence that iPSC indeed retain a methylation signature of their tissue of origin.

Results

Initially we sought to compare the in vivo engraftment potential of hematopoietic stem cells derived from fESC, ntESC, and iPSC in a mouse model of thalassemia. However, even in vitro we observed strikingly different blood-forming potential; thus, we focused here instead on understanding this phenomenon. Our initial set of pluripotent stem cells were derived from the hybrid C57BL/6 x CBA (B6/CBAF1) strain carrying a deletion in the beta-globin locus[15], which is otherwise irrelevant to this study (Fig. 1a). We isolated fESC cells from naturally fertilized embryos and derived ntESC cells from nuclei of dermal fibroblasts[8]. We infected early bone marrow cells (Kit+, Lin−, CD45+) or dermal fibroblasts from aged mice with retroviral vectors carrying Oct4, Sox2, Klf4, and Myc, and selected blood-derived and fibroblast-derived iPSC colonies (B-iPSC, F-iPSC). Hematopoietic progenitors and fibroblasts yielded a comparable frequency of reprogrammed colonies (0.02%), which consistent with prior reports[5], was lower than the yield from fibroblasts of a juvenile mouse (0.1%). We characterized the fESC, ntESC, and iPSC lines for expression of Oct4 and Nanog by immunohistochemistry, and demonstrated multi-lineage differentiation potential in teratomas (Fig. 1b, c; Supplementary Fig. 1). By criteria typically applied to human samples and appropriate for a therapeutic model[3], all stem cell lines manifest pluripotency.

Figure 1

Pluripotent stem cells and their characterization

a, Experimental schema. fESC, ntESC, F-iPSC, and B-iPSC were derived from B6/CBA F1 mice by reprogramming and/or cell culture, characterized for pluripotency by criteria applied to human cells, followed by differentiation analysis for osteogenic or hematopoietic lineages. b, Expression of pluripotency markers NANOG and OCT4 by immunohistochemistry. 4,6-Diamidino-2-phenylindole (DAPI) staining for total cell content. Feeder fibroblasts serve as internal negative controls. c, Teratoma analysis: tumor histology from indicated cell lines shows highly cystic structures consisting of differentiated elements of all three embryonic germ layers. Scale bar, 200μm.

Differentiation of pluripotent stem cells

To test blood potential, we differentiated multiple pluripotent stem cell clones into embryoid bodies (EBs), dissociated cells, and assayed for hematopoietic colony forming cells[16]. All pluripotent cells generated comparable EBs but markedly different numbers of hematopoietic colonies. Consistently, blood-derived B-iPSC yielded more hematopoietic colonies than F-iPSC (Fig. 2a). Hematopoietic colony formation from ntESC and fESC were higher than the iPSC lines.

Figure 2

Differentiation of cell lines

a, Hematopoietic colony number per 100,000 EB cells differentiated from indicated cell lines. b, Alizarin Red staining of osteogenic cultures of B-iPSC (left) and F-iPSC (right). Top: 3 cm culture dish; Bottom: stained colonies; scale bar, 500 μm. c, Quantification of elemental calcium by inductively coupled plasma – atomic emission spectroscopy[49] in 5×105 cells after osteogenic differentiation of indicated cell lines. d, Q-PCR of osteogenic genes, Bglap, Sp7, and Runx2 in indicated cell lines after osteogenic differentiation. Gene expression was normalized to Actin. n=number of independent clones tested. Error bars=s.d.

We then tested differentiation into osteoblasts, a mesenchymal lineage that can be derived from fibroblasts[1718]. By alizarin red staining, a marker of osteogenic cells[19], F-iPSC produced more sharply defined osteogenic colonies (Fig. 2b), deposited more elemental calcium (Fig. 2c), and showed higher expression of three osteoblast-associated genes (Fig. 2d) than B-iPSC. By these criteria, F-iPSC show enhanced osteogenic potential, reflecting a propensity to differentiate towards a mesenchymal lineage. In contrast, ntESC cells behaved comparably to fESC in hematopoietic and osteogenic assays.

DNA methylation of pluripotent stem cells

We hypothesized that the different pluripotent cells might harbor different patterns of genomic DNA methylation; thus, we performed Comprehensive High-throughput Array-based Relative Methylation (CHARM) analysis, which interrogates ~4.6 million CpG sites, including almost all CpG islands and nearby sequences termed shores[2021], but does not assess non-CpG methylation. We determined the number of differentially methylated regions (DMRs) between pair-wise comparisons, using a threshold area cutoff of 2, corresponding to a 5% false discovery rate (FDR[22]; Supplementary Table 1a). By this analysis, ntESC were most similar to fESC (only 229 DMRs), whereas F-iPSC differed most extensively (5304 DMRs). Relative to fESC, hypermethylated DMRs predominated for F-iPSC (3349=63%) and B-iPSC (516=74%). Highlighting their functional differences, 5202 DMRs were identified between B-iPSC and F-iPSC. We confirmed the results of CHARM analysis by bisulfite pyrosequencing of multiple loci (Supplementary Fig. 2).

Unsupervised hierarchical clustering of DMRs between B-iPSC and F-iPSC easily distinguished iPSC from ntESC and fESC, which cluster together (Fig. 3a). B-iPSC cluster nearer to ntESC and fESC than do F-iPSC, which represent a strikingly separate cluster. These data indicate that the methylation patterns of ntESC are more like fESC than are either iPSC.

Figure 3

Analysis of methylation in stem cell lines

a, Cluster dendrogram using probes from DMRs that distinguish B-iPSC and F-iPSC. Cell clones are described in Supplemental Table 5. b, Enrichment of DMRs for hematopoiesis and fibroblast-related transcription factors in B-iPSC and F-iPSC, relative to chance (blue histogram; 100,000 random permutations). Left panel: 20 of 74 hematopoiesis-related transcription factors overlap DMRs hypermethylated in F-IPSC (p=0.0034); Right panel: 115 of 764 fibroblast-specific genes overlap DMRs hypermethylated in B-iPSC (p=10−5).

Several lines of evidence support a mechanistic link between differential methylation and hematopoietic propensity of iPSC lines. First, literature survey of genes for the top 24 DMRs that distinguish B-iPSC and F-iPSC links 11 to hematopoiesis and 3 to osteogenesis (Supplementary Table 2). Of the 11 hematopoietic loci, 10 are hypermethylated in F-iPSC relative to B-iPSC. Second, of 74 hematopoietic transcription factors[23], 20 are in or near DMRs that are hypermethylated in F-iPSC versus B-iPSC, twice that predicted by chance (p=0.0034; Fig. 3b left panel, Supplementary Fig. 3a, and Supplementary Table 3). Similarly, of 764 fibroblast-specific genes, 115 are hypermethylated in B-iPSC, twice that predicted by chance (p=10−5; Fig 3b right panel). Given the correlation between methylation and transcriptional silencing[24], our data suggest that iPSC harbor epigenetic marks antagonistic to cell lineages distinct from the donor cell type.

We asked whether DMRs that distinguish B-iPSC from fESC might allow us to identify their hematopoietic lineage of origin. In a separate CHARM experiment [25], we examined genome-wide methylation in highly purified multipotent and lineage-specific hematopoietic progenitors. Comparing DMRs in B-iPSC to those that define hematopoietic progenitors, we observed that B-iPSC cluster alongside Common Myeloid Progenitors (CMP) and distant from Common Lymphoid Progenitors (CLP; Supplementary Fig. 4a and Supplementary Table 4), which is notable given that B-iPSC were derived from Kit+, lineage-negative myeloid marrow precursors. Next, we asked whether the tissue of origin (bone marrow vs fibroblast) could be identified by the methylation state of tissue specific DMRs in F-iPSC, B-iPSC, and Bl-iPSC (a B lymphocyte-derived iPSC line described below). Using DMRs that distinguish fibroblast and bone marrow, and examining methylation in iPSCs and somatic cells from two different genetic backgrounds (B6CBA and B6129), we found that F-iPSC cluster alongside fibroblasts, and distant from bone marrow (Supplementary Fig. 4b). Similarly, the hematopoietic-derived B-iPSC and Bl-iPSC grouped with somatic cells from bone marrow. Thus, residual methylation indicates the tissue of origin of iPSC, and for blood-derivatives even their precise lineage, further supporting the phenomenon of epigenetic memory in iPSC.

Reprogrammed state of iPSC and ntESC

We postulated that the differing methylation signatures of B-iPSC, F-iPSC, and ntESC reflect disparate reprogramming, and confirmed this by two independent computational analyses. First, we overlapped DMRs that distinguish B-iPSC, F-iPSC, and ntESC from fESC with genes specifically expressed in undifferentiated murine fESC[26]. By this analysis, ntESC showed the fewest DMRs at loci corresponding to the most highly expressed fESC-specific genes, and B-iPSC showed fewer DMRs at these loci than F-iPSC (Supplementary Fig. 5a). Second, we overlapped DMRs with the DNA binding locations for seven transcription factors that compose a core protein network of pluripotency[27], and found the fewest DMRs at core transcription factor binding sites in ntESC, and less overlap in B-iPSC than in F-iPSC (Supplementary Table 5). These analyses indicate that F-iPSC harbor more residual methylation than B-iPSC at loci directly linked to the gene expression and pluripotency networks of fESC, whereas ntESC show the least differential methylation and appear closest to fESC at these critical loci.

Further analysis of Oct4 and Nanog indicates that although both are detected by immunohistochemistry in B-iPSC and F-iPSC (Fig. 1b), Oct4 mRNA is fully expressed from a demethylated promoter in both types of iPSC, whereas Nanog mRNA is sub-optimally expressed from a promoter that retains considerable methylation in F-iPSC (Supplementary Fig. 6). When assessed by blastocyst chimerism, B-iPSC contribute to all tissues, including the germ line, whereas F-iPSC contribute only poorly (Supplementary Fig. 7a), although they can be found in SSEA1+ germ cells of the gonadal ridge (Supplementary Fig. 7b). Thus, while both B-iPSC and F-iPSC generate robust multi-lineage teratomas, satisfying criteria for pluripotency typically applied to human cells[3], broader functional assessments available in the mouse system confirm their differential degree of reprogramming. In this comparison of iPSC derived from accessible tissues of aged adult mice, bone marrow yields stem cells with superior features of pluripotency, but neither iPSC is equivalent to ntESC or fESC.

Stringently-defined pluripotent stem cells

To determine if blood-forming potential differs among cell lines that satisfy more stringent criteria for pluripotency, we analyzed lines derived from a uniform genetic background (B6/129F1) that all express a Nanog-eGFP reporter gene[28], and for which pluripotency was demonstrated by blastocyst chimerism and transmission through the germ line (Fig. 4a, upper schema; Supplementary Table 6). These studies involve “secondary” iPSC lines derived from neural progenitor cells (NP-iPSC[29]) and B-lymphocytes (Bl-iPSC[30]) of mice chimerized with iPSC carrying proviruses that express doxycycline-inducible reprogramming factors from identical proviral integration sites. NP-iPSC and Bl-iPSC were compared to ntESC generated from neural progenitor cells (NP-ntESC[8]), blood progenitor cells (B-ntESC[31]), and fibroblasts (F-ntESC[31, 32]), as well as fESC.

Figure 4

Stringently-defined pluripotent stem cells and their characterization

a, Experimental schema. Four horizontal lines indicate integrated proviruses carrying dox-inducible reprogramming factors in some experiments. Characteristics of individual clones in all subsequent panels can be found in Supplemental Table 5. b, Hematopoietic colony number per 100,000 EB cells differentiated from indicated cell lines. n=number of independent clones tested. Error bars=s.d., added for clones repeated three or more times. c, Cluster dendrogram using probes from DMRs that distinguish Bl-iPSC and NP-iPSC.

All cell lines were differentiated into embryoid bodies and assayed for hematopoietic colony forming activity[16]. Across multiple clones, we observed higher blood forming potential of iPSC derived from B lymphocytes (Bl-iPSC) than from neural progenitors (NP-iPSC; Fig. 4b). In contrast, we observed that ntESC, regardless of tissue origin (fibroblasts, neural progenitors, or T-cells), and fESC displayed an equivalently robust blood forming potential (Fig. 4b). In this independent set of iPSC lines, qualified as pluripotent by stringent criteria, we again observed consistent differences in blood formation, with blood derivatives showing more robust hematopoiesis in vitro than neural derivatives.

Resetting differentiation propensity

Finally, we asked whether we could rescue the poor blood-forming potential of NP-iPSC by differentiation into hematopoietic lineages[1633], followed by a tertiary round of reprogramming back to pluripotency by doxycycline induction of the endogenous reprogramming factors (Fig. 4a, lower schema). As a control, we differentiated NP-iPSCs into neural stem cells[8], followed by tertiary reprogramming to pluripotency. Resulting iPSC clones were selected for expression of the Nanog-eGFP reporter and shown to express Oct4 and Nanog by immunohistochemistry (Supplementary Fig. 8) and to chimerize murine blastocysts (Supplementary Fig. 9). The tertiary blood-derived B-NP-iPSC showed higher hematopoietic colony-forming potential than the tertiary NSC-NP-iPSC (Fig. 4b), and generated larger hematopoietic colonies with more cells per colony (Supplementary Fig. 10a, b). These data indicate that the poor blood-forming potential of secondary NP-iPSC can be enhanced by differentiation into hematopoietic progeny, followed by tertiary reprogramming. In contrast, tertiary reprogramming via neural intermediates yields iPSC that retain poor hematopoietic potential.

The reduced blood potential of NP-iPSC might be explained by residual epigenetic marks that restrict blood fates or a lack of epigenetic marks that enable blood formation. We sought to determine whether treatment of NP-iPSC with pharmacologic modulators of gene expression and DNA methylation might reactivate latent hematopoietic potential. We treated NP-iPSC in vitro with Trichostatin A (TSA), a potent inhibitor of histone deacetylase[34], and 5-azacytidine (AZA), a methylation-resistant cytosine analogue[35]. After 18 days of drug treatment, the resulting cells displayed higher blood forming activity (NP-iPSC-TSA-AZA; Fig. 4b). For unclear reasons, tertiary reprogramming through blood intermediates or drug treatment of NP-iPSC produced altered ratios of colony sub-types, perhaps suggesting different efficiencies of lineage reprogramming.

Methylation in secondary and tertiary iPSC

We used CHARM to examine the methylome of the germ-line competent pluripotent stem cells, the tertiary reprogrammed B-NP-iPSC and NSC-NP-iPSC, and the drug-treated NP-iPSC (Fig. 4a). In pair-wise comparisons (Supplementary Table 1b), the NP-iPSC showed only a small number of DMRs relative to fESC (553), fewer than the numbers of DMRs distinguishing ntESC from fESC (679), indicating that selection using the Nanog-GFP reporter and derivation from young donor tissue yields more equivalently reprogrammed cells. Despite equivalent Nanog-GFP expression, B lymphocyte-derived Bl-iPSC harbored more DMRs (1485) relative to fESC than did the NP-iPSC. Cluster dendrogram analysis, employing the most variable DMRs that distinguish Bl-iPSC and NP-iPSC, showed NP-iPSC to be more similar to fESC than are Bl-iPSC, which represent a distinct cluster (Fig. 4c). These data suggest that neural progenitors are more completely reprogrammed to an ESC-like state than blood donor cells. Cluster dendrogram analysis failed to distinguish among NP-iPSC, ntESC, and fESC, but assessment of the overlap of DMRs with loci for highly expressed ESC-specific genes[26] and core pluripotency transcription factor binding sites[27] indicated differences among these three pluripotent cell types, and reveal that ntESC have the fewest DMRs affecting these critical loci (Supplementary Fig. 5b).

Relative to fESC, hypermethylated DMRs predominated in NP-iPSC and Bl-iPSC (417 (75%) and 1423 (96%), respectively; Supplementary Table 1b), confirming that even when pluripotency is documented by stringent criteria, iPSC retain residual methylation. By analysis of overlapping DMRs, Bl-iPSC cluster with progenitors of the lymphoid lineage (CLP) rather than the myeloid lineage (Supplementary Fig. 4a; Supplementary Table 4[25]). To illustrate this point, the Gcnt2 gene, which encodes the enzyme responsible for the blood group I antigen, and Gata2, a regulator of hematopoiesis and erythropoiesis, are both hypermethylated and transcriptionally silent in the lymphoid lineage[25]. Bl-iPSC showed hypermethylation at these loci relative to fESC, whereas the myeloid-derived B-iPSC did not (Supplementary Fig. 11). Thus, a methylation signature correctly identifies the blood lineage of origin of B-lymphocyte derived iPSC.

Furthermore, we found that neural-related genes tended to be differentially methylated between Bl-iPSC and NP-iPSC (Supplementary Fig. 3b). Treatment of NP-iPSC with TSA and AZA enhances blood-forming potential and increases hypomethylated DMRs (626; Supplementary Table 1c). We found significant overlap between these DMRs and genes enriched in mouse hematopoietic stem cells (MSigDG signature STEMCELL_HEMATOPOIETIC_UP; Supplementary Fig. 3c), suggesting that drug treatment erases inhibitory methylation signatures at hematopoietic loci.

We compared DMRs in iPSCs with high hematopotietic potential (B-NP-iPSC and NP-iPSC-TSA-AZA) to those with low hematopoietic potential (NP-iPSC and NSC-NP-iPSC), and found that B-NP-iPSC and NP-iPSC-TSA-AZA harbored higher gene-body methylation of Wnt3 (Supplementary Fig. 12), a gene which along with its homologue Wnt3a plays a major role in blood development from fESC[36]. The blood-deficient NP-iPSC and NSC-NP-iPSC lines lacked gene body methylation. While promoter methylation is repressive, gene body methylation is seen in active genes[37]. When iPSC were differentiated into embryoid bodies, the blood-prone NP-iPSC-TSA-AZA showed higher levels of Wnt3/3a expression than the blood-deficient NP-iPSC (Supplementary Fig. 13a). Interestingly, supplementation of the culture media with Wnt3a during embryoid body differentiation restored blood-forming potential in the blood-deficient NP-iPSC and NSC-NP-iPSC lines, but had little effect on the already robust hematopoietic potential of B-NP-iPSC (Supplementary Fig. 13b). Albeit preliminary, these data correlate differential gene body methylation and expression of the Wnt3 locus with enhanced blood-forming potential in iPSC lines.

Discussion

Here we demonstrate that iPSC retain an epigenetic memory of their tissue of origin. Our data reveal several important principles that relate to the technical limitations inherent in the process of reprogramming, and which in practice influence the differentiation propensity of specific isolates of iPSC.

First, tissue source influences the efficiency and fidelity of reprogramming[91011]. From aged mice, blood cells were reprogrammed more closely to fESC than dermal fibroblasts, which yielded only incompletely reprogrammed cells. Neural progenitor-derived iPSC were most similar to fESC, consistent with evidence that such cells can be reprogrammed with fewer transcription factors[38]. Whereas neural progenitors are not readily accessible, iPSC can be generated by direct reprogramming of human blood[39].

Second, analysis of DNA methylation reveals substantial differences between iPSC and embryo-derived ESC (ntESC and fESC). iPSC derived from non-hematopoietic cells (neural progenitors and fibroblasts) retain residual methylation at loci required for hematopoietic fate, which manifests as reduced blood-forming potential in vitro. Residual methylation signatures link iPSC to their tissue of origin, and even discriminate between the myeloid and lymphoid origins of blood-derived iPSC. Prior studies reporting residual hypermethylation in iPSC[3722] did not establish a link between DMRs at specific loci, tissue of origin, and altered differentiation potential. While residual methylation is mostly repressive, we have shown for Wnt3 that residual gene body methylation in blood-derived iPSC is associated with enhanced blood potential. Interestingly, the poor blood potential of neural progenitor-derived iPSC, which lack this epigenetic mark and express lower levels of endogenous Wnt3, can be enhanced by supplementing differentiating cultures with exogenous Wnt3a cytokine, indicating that manipulating culture conditions can overcome epigenetic barriers.

Third, the differentiation propensity and methylation profile of iPSC can be reset. When blood-deficient neural progenitor-derived iPSC (NP-iPSC) are differentiated into blood and then reprogrammed to pluripotency, their blood-forming potential is markedly increased. Alternatively, treatment of NP-iPSC with chromatin-modifying compounds increases blood-forming potential and is associated with reduced methylation at hematopoietic loci. For some applications, epigenetic memory of the donor cell may be advantageous, as directed differentiation to specific tissue fates remains a challenge.

Fourth, nuclear transfer-derived ESC are more faithfully reprogrammed than most iPSC generated from adult somatic tissues. Like the immediate and rapid demethylation of the sperm pronucleus following fertilization, somatic nuclei are rapidly demethylated by nuclear transfer into ooplasm, prompting speculation that the egg harbors an active demethylase[14]. In contrast, demethylation is a late phenomenon in factor-based reprogramming, and likely occurs passively[13]. Studying how ooplasm erases methylation might identify biochemical functions that would enhance factor-based reprogramming. Failure to demethylate pluripotency genes is associated with intermediate or partial states of reprogramming[122940], and knock-down of the maintenance methyltranferase DNMT1 or treatment with the demethylating agent 5-AZA can convert intermediate states to full pluripotency. Demethylation appears passage dependent[13], and reprogramming efficiency correlates with the rate of cell division and the passage number[41]. In our experiments, we compared pluripotent stem cells of comparable low passage number (Supplemental Table 5), but continued serial passage may homogenize the differentiation potential of pluripotent cell types.

The mRNA expression program of iPSC and fESC are strikingly similar[42]. Minor differences in mRNA and microRNA expression have been reported[40], but removal of transgenes reduces the differences[43]. The Dlk1-Dio3 locus, whose expression correlates with capacity to generate “all-iPSC” mice[44], is not differentially methylated and expressed in at least some iPSC lines that manifest epigenetic memory (our unpublished observations). Thus even the most stringently-defined iPSC might retain epigenetic memory. Importantly, differences between iPSC and fESC may not manifest until differentiation, when the specific loci that retain residual epigenetic marks are expressed, influencing cell fates. Methylation is but one molecular feature of “epigenetic memory” in iPSC. Faulty restoration of bivalent domains, which mark developmental loci with both active and repressive histone modifications [45], and loss of pioneer factors, which in fESC and iPSC occupy enhancers of genes expressed only in differentiated cells[46], represent two other potential mechanisms.

Although ideal, generic iPSC may be functionally and molecularly indistinguishable from fESC, we have shown in practice that even rigorously selected iPSC can retain epigenetic marks characteristic of the donor cell that influence differentiation propensity. Epigenetic differences are unlikely to be essential features of iPSC, but rather reflect stochastic variations associated with the technical challenges of achieving complete reprogramming. Given that we lack reporter genes for selecting human iPSC, and cannot qualify their pluripotency by assaying embryo chimerism, the behavior of human cells will likely be influenced by epigenetic memory. Human ESC can also manifest variable differentiation potential[47]. These data highlight our limited understanding of the epigenetic heterogeneity of pluripotent stem cells, and the need for improved methods to ensure reprogramming of somatic cells to a fully naïve, “ground state” of pluripotency[48].

METHODS SUMMARY

ES cells and iPSCs were cultured in ESC media containing 15% FBS, and 1,000 U/ml of LIF. For the reprogramming of somatic cells, retrovirus expressing Oct4, Sox2, Klf4, and Myc were introduced. For the somatic cells containing inducible reprogramming factors, the media was supplemented with 2 ng/ml of doxycycline. For DNA and RNA isolation, fESC or iPSCs were trypsinized and re-plated onto new tissue culture dishes for 45 minutes to remove feeder cells, and nucleic acids were extracted from the non-adherent cell suspension. Genomic DNA methylation analysis and pyrosequencing were performed by previously published methods[2021].

Full Methods and any associated references are available in the online version of the paper at www.nature.com/nature.