Naïve human pluripotent cells feature a methylation landscape devoid of blastocyst or germline memory

Human embryonic stem cells (hESCs) are in vitro pluripotent cell types with the capacity for unlimited self-renewal and differentiation, making them critical models for understanding mechanisms required for human embryo development and differentiation. Although hESCs are derived from pre-implantation human blastocysts, they are morphologically and transcriptionally similar to murine epiblast stem cells (EpiSCs), which are derived from post-implantation mouse embryos. As such, hESCs and EpiSCs are said to exhibit a “primed pluripotent state” while mouse ESCs derived from the preimplantation blastocyst exhibit a “naïve pluripotent state” corresponding to an earlier stage of development (Nichols and Smith, 2009).

A number of culture conditions have recently been developed that promote maintenance and self-renewal of naïve human pluripotent stem cells (Chan et al., 2013; Gafni et al., 2013; Takashima et al., 2014; Theunissen et al., 2014; Ware et al., 2014). Each protocol generates cell types with slightly different molecular characteristics, which may reflect metastable states in the spectrum of naïve to primed pluripotency. Recent meta-analysis of sequencing data indicates that two of these protocols generate cells with a close transcriptional resemblance to the human pre-implantation epiblast (Huang et al., 2014). In the first protocol, hESCs are transfected with KLF2 and NANOG and cultured in media with titrated two inhibitors plus leukemia inhibitory factor and Gö6983 (t2iL+Gö) (Takashima et al., 2014). In the second protocol, primed cells can be reverted by transferring to a media containing a cocktail of five inhibitors plus LIF, Activin and/or Fibroblast Growth Factor 2 (5iLA/F)(Theunissen et al., 2014). Using t2iL+Gö reversion of the H9 primed hESC line, it was shown that DNA methylation is globally reduced to the average level measured in human pre-implantation epiblasts (Takashima et al., 2014), with additional locus-specific erosion in the 5’ region of the LINE1 human specific (LIHS) retrotransposons (Gkountela et al., 2015). The DNA methylation profile of cells cultured in 5iLAF has never been evaluated.

Before studying the methylation pattern of 5iLAF cultured cells, we first wanted to confirm and characterize the naïve phenotype. We performed n=4 independent reversions of the hESC line UCLA1 (Diaz Perez et al., 2012) using 5iLAF (Theunissen et al., 2014). Upon reversion we observed a mixture of small, round colonies similar to naïve mESCs as well as flat, cobblestone-like colonies (Figure 1A,B). We evaluated one reversion using two classic human pluripotency surface markers called SSEA4 and TRA-1-81. Unlike primed UCLA1 hESCs, which are double positive for SSEA4 and TRA-1-81, the 5iLAF reverted hESCs have a large fraction of double negative cells (Figure 1A,B). Immunoflourescence staining showed that the SSEA4 and TRA-1-81 negative cells were still positive for OCT4 and NANOG (Figures S1A-F).

Next, we sorted the 5iLAF-cultured cells into SSEA4 positive and negative populations using fluorescence activated cell sorting (FACS) and re-plated the sorted cells onto MEFs in 5iLAF media (Figure 1C). We discovered that SSEA4 positive cells yielded mostly flat colonies, whereas SSEA4 negative cells yielded mostly round colonies. One passage after sorting, the SSEA4 negative population remained SSEA4 negative, indicating that this is a relatively stable state (Figure S1G). We then reverted two additional lines called UCLA4 and UCLA5 (Diaz Perez et al., 2012) and found that small, round colony morphology was always enriched in the SSEA4 negative fraction whereas the SSEA4 positive cells yielded mostly flat, cobblestone colonies (Figure S1H-K).

In order to determine whether the heterogeneity in SSEA4 expression was also observed when deriving hESC lines completely under naïve 5iLAF conditions, we derived n=2 new hESC lines, which we have called UCLA19n and UCLA20n after thawing n=7 day 5 vitrified human blastocysts. Colonies were uniformly round and flow cytometry revealed that UCLA19n was 85% SSEA4 negative (results not shown), whereas UCLA20n was almost completely SSEA4 negative (Figure 1D). In contrast, TRA-1-81 was expressed on a significant portion of SSEA4 negative cells in UCLA20n as well as reverted UCLA4 and UCLA5 hESC lines (Figure 1D, Figure S1I, Figure S1K). Therefore, loss of TRA-1-81 is not a consistent marker of naïve morphology, whereas absence of SSEA4 is a highly correlated feature of naïve round colony morphology. In summary, reversion of primed hESCs in 5iLAF generates a heterogeneous mixture of colonies, with SSEA4 negative hESCs correlating with small round colony morphology similar to naïve hESCs derived from the human pre-implantation blastocyst.

On the basis of morphology, we speculated that 5iLAF SSEA4 negative hESCs are the naïve population and thus transcriptionally resemble the cells of the human preimplantation epiblast. To address this, we performed RNA-sequencing of 5iLAF-cultured SSEA4 positive or SSEA4 negative fractions of UCLA1, and compared them to SSEA4 positive primed UCLA1 hESCs at equivalent passages (Table S1). We also performed RNA-Seq of UCLA20n at passage 20 after derivation. We did not analyze UCLA19n as it was found to be 70% polyploid by passage 14. Consistent with the expression patterns of genes associated with naïve pluripotency in mice, the 5iLAF SSEA4 negative cells and UCLA20n had elevated levels of NANOG as well as a dramatic up-regulation of KRUPPLE LIKE FACTOR (KLF) family transcription factors and reduced expression primed state master regulators such as ZINC FINGER OF THE CEREBELLUM (ZIC) family transcription factors and OTX2 (Buecker et al., 2014; Tang et al., 2011; Yang et al., 2014) (Figure 1E, Table S1). To further confirm the similarity of 5iLAF SSEA4 negative and UCLA20n hESCs to the human pre-implantation blastocyst we used the previously published single cell expression data from late pre-implantation epiblast and primed hESCs (Yan et al., 2013). We defined a set of “pre-implantation epiblast specific” and “primed-specific” genes, which showed >4 fold difference in expression between these two cell types (Table S1). Using these genes as a reference we found that the 5iLAF SSEA4 negative hESCs and UCLA20n had global up-regulation of naïve epiblast-specific genes, and down-regulation of primed-specific genes (Figure 1F).

In contrast, the SSEA4 positive hESCs sorted from 5iLAF cultures had an intermediate expression pattern between primed and naïve, suggesting that SSEA4 positive cells that stably self-renew in 5iLAF are partially reverted to the naïve state (Figure 1E,F). Comparing to published datasets, we found that the SSEA4 negative population in UCLA1 and the new hESC line UCLA20n is analogous to the original 5iLAF hESC lines created by reverting WIBR2 (Theunissen et al., 2014) and to t2iL+Gö-cultured hESCs created by reverting H9 (Takashima et al., 2014). In contrast, other published naïve methods showed a less pronounced shift toward the naïve state and failure to repress primed markers (Chan et al., 2013; Gafni et al., 2013; Ware et al., 2014). Interestingly, lines generated by these methods are also reported to be SSEA4 positive. Given these results, we focused our methylation analysis on the 5iLAF and t2iL+Gö conditions.

To determine the methylation pattern of hESCs in 5iLAF we performed whole genome bisulfite sequencing (WGBS) on 2-4 independent sorts of SSEA4 negative or SSEA4 positive reverted UCLA1 cells, SSEA4 negative UCLA20n cells and primed UCLA1 cells that had been in culture a similar length of time to the reverted lines. We discovered that similar to the levels observed in t2iL+Gö (Gkountela et al., 2015; Takashima et al., 2014) 5iLAF cultured SSEA4 negative hESCs and UCLA20n had an average CG methylation level that resembled the human blastocyst (Figure 2A) (Okae et al., 2014).

In mammals the methylation pattern of the blastocyst is shaped by events during gametogenesis and early embryogenesis. The male pronucleus is selectively demethylated in early embryonic development, with only a few regions such as paternally methylated imprinted loci protected from DNA demethylation (Okae et al., 2014; Smith et al., 2014; Smith et al., 2012). Thus in humans the methylation pattern of the blastocyst strongly resembles Metaphase II human oocytes (Figure 2B, C). In contrast, the methylation landscapes of SSEA4 negative 5iLAF-cultured hESCs, UCLA20n and t2iL+Gö-cultured cells are only weakly correlated with the human blastocyst and human oocyte (Figure 2B, C). Naïve cells, even if cultured by different methodologies or derived directly from the human blastocyst, converge toward a methylation pattern that is different from the pre-implantation human blastocyst (Figure 2B,C). A striking example of this is trend is observed at 332 CpG islands identified previously as “transient maternal imprints”: sites that are highly methylated in oocytes and the maternal chromosomes of blastocyst that lose methylation upon implantation (Smith et al., 2014). We discovered that reversion does not regenerate methylation at these sites, nor is methylation retained at these transient maternal imprints in the UCLA20n hESC line (Figure S2A).

An additional, striking deviation from oocyte and blastocyst methylation patterns in 5iLAF and t2iL+Go cultured cells occurred at stable imprints. These are regions where DNA methylation is established exclusively during germ-cell development. These methylated sites are protected from DNA demethylation during pre-implantation embryo development, remaining differentially methylated in somatic cells through the life of the organism and promoting a parent-of-origin specific expression pattern in the neighboring genes. We examined DNA methylation at 29 maternally methylated stable primary imprints and 2 paternally methylated stable primary imprints (Okae et al., 2014) (Figure 2D, Table S2). There is roughly 50% methylation in somatic tissue and slightly below 50% methylation in blastocysts as expected. In the primed UCLA1 hESCs used in our study the median methylation of these imprinted sites was close to 50%, though some imprints were hyper or hypomethylated, similar to what has been observed previously for other hESC lines (Rugg-Gunn et al., 2007). Strikingly, the 5iLAF SSEA4 negative hESCs and UCLA20n had near complete loss of methylation from all thirty-one primary stable imprints evaluated in our study, with loss over many imprints also found in t2iL+Gö (Figure 2D,E). Taking advantage of single nucleotide polymorphisms (SNPs) present in the UCLA1 hESC line, we observed a shift in allelic expression upon reversion in 5iLAF from monoallelic to biallelic expression of several imprinted genes including H19 and SNRPN. (Figure S2B, Table S2). In order to determine whether methylation could be restored at imprinted genes by reverting the naïve hESCs back to a primed state, we cultured 5iLAF SSEA4 negative and primed UCLA1 cells in primed epiblast like cell (EpiLC) media (Hayashi et al., 2011) for sixteen days. During this time, we discovered that 5iLAF SSEA4 negative cells showed a global shift toward expression of primed-specific genes and gained DNA methylation genome-wide (Figure S2C). However, increased methylation over imprinted regions was very modest, and biallelic expression was still observed (Figure S2D, Table S2). Thus, when lost, imprinting is not re-established upon culturing in primed conditions, a similar scenario to the rescue of global DNA methylation but not imprint methylation in Dnmt1 knockout ESCs by the re-expression of Dnmt1 (Holm et al., 2005). Furthermore, consistent with data observed in hESCs cultured in t2iL+Gö (Gkountela et al., 2015), young LINE elements also show dramatic promoter hypomethylation in 5iLAF (Figure S2E).

Given the problem with maintenance of imprint methylation in naïve cells, we considered the possibility that the 5iLAF SSEA4 positive cells may represent a useful intermediate. However, we discovered that the SSEA4 positive cells showed intermediate levels of global and imprint methylation loss (Figure 2C, 2D), with biallelic expression of SNRPN and H19 (Table S2). We also compared the methylation loss to naïve human stem cell (NHSM) media (Gafni et al., 2013), which shows the smallest transcriptional shift towards naïve pluripotency (Figure 1E,F). In order to directly compare our methylome data to Gafni et al., 2013 we modified all whole-genome data sets to simulate the Reduced Representation Bisulfite Sequencing approach used by Gafni to measure DNA methylation. We discovered that imprint methylation was unperturbed in NHSM (Figure S2F), but very little global change in methylation was observed either (Figure 2F).

Consistent with an initial report of karyotypic instability in 5iLAF culture (Theunissen et al., 2014), we discovered that twenty-four passages after reversion the 5iLAF UCLA1 hESCs developed widespread karyotypic abnormalities, which was not observed in the first thirteen passages following reversion (Figure S2G). Similarly, UCLA20n had evidence of trisomies at chromosomes 3, 7, 12 and 20 by passage 14 and as discussed above, UCLA19n was 70% polyploid at passage 15 (Figure S2H). Therefore, karyotypic instability may also be a frequent consequence of naïve hESC culture.

To determine the cause of the failure to maintain DNA methylation, we analyzed changes in RNA and protein levels of DNA methylation and demethylation machinery. We found that the RNA and protein levels of the de novo DNA methyltransferase DNMT3B dropped sharply in the 5iLAF SSEA4 negative cells, while DNMT3A was unchanged and DNMT3L increased dramatically relative to primed hESCs. UHRF1 RNA levels were slightly elevated in naïve hESCs. However, at the protein level, we observe a 65% loss of UHRF1, and both DNMT1 RNA and protein levels were reduced by 50% in the naïve state. Furthermore, expression of the 5mC oxidases TET1 and TET2 increased substantially in the naïve state. (Figure 2G, 2H).

In the current study we have shown that naïve hESCs have a transcriptional program enriched in human pre-implantation-specific genes but with a global DNA methylation landscape that is distinct from the normal state of the human preimplantation blastocyst. The negative effect of the loss of “transient imprints”, and the failure to recapitulate the oocyte-like methylation pattern is unclear. However, the loss of stable primary imprints is potentially serious in human pluripotent stem cell reserach. Correct imprinting is necessary for organism survival, and a number of rare human medical disorders have been linked to aberrant imprinting (Butler, 2009). Of note, murine embryonic germ cell lines are transcriptionally similar to murine ESCs but have widespread loss of imprints and contribute poorly to chimeras (Leitch et al., 2013; Oliveros-Etter et al., 2015; Tada et al., 1998), demonstrating the importance of imprints in correct differentiation of pluripotent cells in vivo. Accompanying the loss of imprints, we observed extensive karyotypic abnormalities in cells after prolonged culture in 5iLAF. Loss of DNA methylation has been linked to karyotypic instability (Haaf, 1995).

We note that methylation at the imprinted loci is clearly depressed relative to surrounding regions. This may reflect the observation that many imprinted loci are promoters or regulatory elements that are active in the blastocyst (Rugg-Gunn et al., 2007). Thus if methylation is partially eroded at the imprint, the relevant transcription factors bind and cause further demethylation (as is generally the case at these genetic elements). In other words, methylation may be a very weak barrier to locus activation in 5iLAF. Similar dynamics may be at work at L1HS elements.

Although we observed a reduction in DNMT3B protein in the naive cells, we propose that this has only modest effects on creating the 5iLAF methylome given that DNMT3A^−/− DNMT3B^−/− DKO primed hESCs maintain primary imprints and show only modest DNA demethylation even after extended culture (Liao et al., 2015). We therefore propose that a combination of impaired maintenance methylation and increased TET activity could explain the majority of the 5iLAF hypomethylation phenotype. In a cell type with impaired maintenance and some continuous de novo methylation (imparted by DNMT3A and the remaining DNMT3B), DNA methylation levels will reach a steady state, but memory of previous methylation will be lost with DNA replication.