Coordinated removal of repressive epigenetic modifications during induced reversal of cell identity

Abstract

Epigenetic modifications operate in concert to maintain cell identity, yet how these interconnected networks suppress alternative cell fates remains unknown. Here, we uncover a link between the removal of repressive histone H3K9 methylation and DNA methylation during the reprogramming of somatic cells to pluripotency. The H3K9me2 demethylase, Kdm3b, transcriptionally controls DNA hydroxymethylase Tet1 expression. Unexpectedly, in the absence of Kdm3b, loci that must be DNA demethylated are trapped in an intermediate hydroxymethylated (5hmC) state and do not resolve to unmethylated cytosine. Ectopic 5hmC trapping precludes the chromatin association of master pluripotency factor, POU5F1, and pluripotent gene activation. Increased Tet1 expression is important for the later intermediates of the reprogramming process. Taken together, coordinated removal of distinct chromatin modifications appears to be an important mechanism for altering cell identity.

Introduction

The functional specification of cell types in a multicellular organism is determined by both the levels of expression of transcription factors and the receptivity of the chromatin state of the target genes to the binding of such factors (Gaspar‐Maia et al, 2011; Holmberg & Perlmann, 2012; Iwafuchi‐Doi & Zaret, 2016). Once cell identity is established, chromatin modifications, including those on histones and DNA, can ensure that it is stably maintained through cell divisions. Cell fate can be altered by the overexpression of proteins from other cell types (Takahashi & Yamanaka, 2006; Jopling et al, 2011; Vierbuchen & Wernig, 2011). One of the most drastic examples of such cell fate conversions is the reprogramming of unipotent somatic cells to induced pluripotent stem cells (iPSCs) that share the ability to self‐renew indefinitely and differentiate into all other cell types with embryonic stem cells (ESCs) (Takahashi & Yamanaka, 2006; Pasque et al, 2011; Buganim et al, 2013). This process is accompanied by a global change in the epigenetic landscape and provides a model for investigating how chromatin modifications are unraveled to change cell identity (Polo et al, 2012; Sridharan et al, 2013; Hussein et al, 2014; Tonge et al, 2014; Chronis et al, 2017).

Temporal changes involving specific histone modifications have been identified by the isolation of reprogramming intermediates (Polo et al, 2012; Hussein et al, 2014; Chronis et al, 2017). Bivalent domains that contain both the activating histone modification, histone H3 lysine 4 trimethylation (H3K4me3), and the repressive H3K27me3 are more prevalent in pluripotent cells and are gained in a biphasic manner similar to gene expression changes (Polo et al, 2012; Hussein et al, 2014; Chronis et al, 2017). Somatic enhancers are decommissioned with a corresponding loss of enhancer‐associated histone modifications (Chronis et al, 2017), while some pluripotency‐specific enhancers are marked with H3K4me2 early in the reprogramming process (Koche et al, 2011). Large blocks of H3K9me3 form barriers to pluripotency factor binding and are depleted at the very end of reprogramming, but the role of H3K9me2 has not been well defined (Chen et al, 2012; Soufi et al, 2012; Tran et al, 2015). Besides histone modifications, DNA methylation patterns are remodeled at later stages (Mikkelsen et al, 2008; Polo et al, 2012; Hussein et al, 2014; Lee et al, 2014). 5‐methyl cytosine (5mC) is associated with gene repression when present at regulatory regions (Li & Zhang, 2014) and can be removed by the ten‐eleven translocation (TET) family of enzymes in conjunction with the thymine‐DNA glycosylase (Tdg) (Wu & Zhang, 2011; Pastor et al, 2013; Tran et al, 2019a). The Tet enzymes, Tet1, Tet2, and Tet3, convert 5mC to 5‐hydroxymethylcytosine (5hmC), 5‐formylcytosine (5fC), and 5‐carboxylcytosine (5caC) by iterative oxidation. 5fC and 5caC can be excised by Tdg to generate unmodified cytosine (Wu & Zhang, 2011, 2017; Kohli & Zhang, 2013; Pastor et al, 2013). Pluripotent stem cells have higher levels of 5hmC as compared to most somatic cells (Tahiliani et al, 2009; Globisch et al, 2010). 5hmC is thought to act as an epigenetic modification in addition to its role as an intermediate of DNA demethylation (Wu & Zhang, 2011; Kohli & Zhang, 2013; Pastor et al, 2013).

Although the importance of some of these epigenetic changes to reprogramming has been functionally corroborated by modifying the levels of the corresponding enzymes (Chen et al, 2012; Onder et al, 2012; Soufi et al, 2012; Sridharan et al, 2013; Tran et al, 2015), whether there is coordination between these changes remains unknown. We have now discovered that H3K9me2 demethylase Kdm3b impacts DNA demethylation by both transcriptional and post‐transcriptional mechanisms. In the absence of Kdm3b, Tet1 is not activated during reprogramming. Surprisingly, in Kdm3b‐deleted cells, several loci that should undergo conversion from 5mC to unmodified cytosine during reprogramming are trapped in a 5hmC intermediate state, indicating that the removal of H3K9me2 and processing of 5hmC to more oxidative forms may be functionally connected. The retention of 5hmC in Kdm3b‐deleted cells is anti‐correlated with chromatin association of pluripotency transcription factor POU5F1 (also called OCT4), which is required for pluripotency gene activation. Tet1 but not Tet2 has a critical role in remodeling of 5hmC locations at the very end of reprogramming. Taken together, our results provide a model by which repressive modifications act in concert to exclude association of a key reprogramming transcription factor thereby maintaining cell identity.

Results

Kdm3b is essential for the acquisition of pluripotency

During reprogramming, stalled intermediates called partially reprogrammed cells (pre‐iPSCs) are obtained. These cells can be clonally propagated, have completed the early stages of reprogramming such as the mesenchymal‐to‐epithelial transition, and have a transcriptome that is intermediate between a somatic mouse embryonic fibroblast (MEF) and iPSC (Mikkelsen et al, 2008; Sridharan et al, 2009; Esteban et al, 2010). We have shown that pre‐iPSCs can be converted to iPSCs at a very high efficiency of ~80% using the epigenetic modifier ascorbic acid (AA), and inhibitors of glycogen synthase kinase (GSK) and mitogen‐activated protein kinase (MAPK) pathways (2i) (Shinkai & Tachibana, 2011; Tran et al, 2015). The AA‐dependent effects of this conversion compromised upon depletion of Kdm3b (Chen et al, 2012; Tran et al, 2015).

We first depleted Kdm3b using short‐interfering RNA (siRNA) in MEFs expressing the reprogramming factors Pou5f1 (Oct4), Sox2, Klf4, and c‐Myc (OSKM) (Appendix Fig S1A). In serum‐containing media, without the addition of either AA or 2i, there was a significant reduction in the number of reprogrammed colonies as measured by expression of NANOG, an important pluripotency protein (Fig 1A, Appendix Fig S1B). A similar decrease was observed in serum replacement (KSR) conditions (Fig 1B, Appendix Fig S1C) and when reprogramming was conducted using only using Pou5f1, Sox2, and Klf4 (Fig 1C, Appendix Fig S1D) upon Kdm3b depletion. Pre‐iPSCs are obtained only in serum‐containing conditions and require c‐Myc as one of the reprogramming factors (Mikkelsen et al, 2008; Chen et al, 2012; Lee et al, 2014). Since these latter two conditions (Fig 1B and C) do not generate pre‐iPSCs, we conclude that Kdm3b is required for the acquisition of pluripotency irrespective of reprogramming media conditions and from both somatic cells and reprogramming intermediates.

Figure 1. Kdm3b plays an important role in reprogramming to pluripotency.

1. Quantification of NANOG‐positive iPSC colonies obtained after 14 days of OSKM induction in MEFs treated with either non‐target or Kdm3b siRNA in FBS media. Values represent fold change over non‐target control providing relative quantitation. Experimental values are provided in Appendix Fig S1. Error bars represent the standard deviation of three biological replicates. **P < 0.01 by paired t‐test.
2. Quantification of NANOG‐positive iPSC colonies obtained after 12 or 14 days of OSKM induction in MEFs treated with either non‐targeting or Kdm3b siRNA in KSR media. Values represent fold change over non‐target control providing relative quantitation. Experimental values are provided in Appendix Fig S1. Error bars represent the standard deviation of three biological replicates. **P < 0.01 by paired t‐test.
3. Quantification of NANOG‐positive colonies after retroviral expression of OSKM (left) or OSK (right) in MEF treated with non‐targeting or siRNA targeting Kdm3. Nanog counts were performed on day 16 for OSKM or day 21 for OSK reprogramming. Values represent fold change over non‐target control providing relative quantitation. Experimental values are provided in Appendix Fig S1. Error bars represent the standard deviation of three biological replicates. *P < 0.05 by paired t‐test.
4. Immunoblot of H3K9me1/2 histone demethylase in WT or Kdm3b CRISPR‐targeted pre‐iPSC clones. These pre‐iPSCs carry a Nanog‐GFP reporter. Parent line (WT) and two Kdm3b‐KO clones using gRNA targeting either exon 1 (KO#1) or exon 2 (KO#2) are shown. RNA polymerase II levels are the loading control.
5. Quantification of Nanog‐GFP‐positive cells after 10 days of AA+2i treatment of WT or Kdm3b‐deleted pre‐iPSCs. Error bars represent the standard deviation of three or five biological replicates. ***P < 0.001 by paired t‐test.
6. Immunoblot of H3K9 methylation levels upon addition of H3K9 methyltransferase inhibitor UNC0638 in WT or Kdm3b‐KO pre‐iPSCs.
7. Quantification of percent Nanog‐GFP‐positive cells after 10 days of AA+2i or AA+2i with 5 μM UNC0638 in WT or Kdm3b‐deleted pre‐iPSCs. Error bars represent a standard deviation of three biological replicates. *P < 0.05 by paired t‐test.

Since reprogramming populations are a heterogeneous mixture of cells, we determined the mechanism of action of Kdm3b during the conversion of clonal pre‐iPSC to iPSCs. Using CRISPR‐Cas9 technology, we generated two independent Kdm3b knockout pre‐iPSC lines (Kdm3b‐KO). For this purpose, we started with a pre‐iPSC line that had been derived from MEFs with a Nanog‐GFP reporter (Tran et al, 2015). In the Kdm3b‐KO, the protein levels of KDM3A and JMJD1C which can have the same specificity for demethylating H3K9me1/me2 (Kim et al, 2010; Black et al, 2012) remain unaltered as compared to wild‐type cells (Fig 1D). However, the conversion to iPSCs was abolished in the Kdm3b‐KO upon AA+2i treatment (Fig 1E), indicating an essential role for Kdm3b in the acquisition of the iPSC state. The very small (< 1%) conversion observed in the Kdm3b may be sustained by Kdm3a, whose depletion has a smaller effect on reprogramming (Chen et al, 2012; Tran et al, 2015). In contrast, both Kdm3a and Kdm3b must be deleted to compromise the maintenance of pluripotency in embryonic stem cells (ESCs; Kuroki et al, 2018). Therefore, Kdm3b has a non‐redundant function in the acquisition but not the maintenance of pluripotency.

To determine whether the function of Kdm3b in reprogramming is derived from its catalytic activity, we lowered levels of H3K9me2 by using a chemical inhibitor (UNC0638) to G9a and Glp, the H3K9me1/me2 histone methyltransferases (Shinkai & Tachibana, 2011). Although this inhibitor is likely to have pleiotropic effects, the resulting depletion of H3K9me1/me2 (Fig 1F), partially rescues the phenotype of the Kdm3b‐KO to generate iPSCs (Fig 1G, Appendix Fig S1E). Therefore, the ability of Kdm3b to modulate the levels of H3K9me1/me2 is relevant for its function in the acquisition of pluripotency. Since global levels of H3K9me2 are not dramatically affected by Kdm3b loss (Fig 1F), the catalytic activity is likely to be relevant at specific genomic sites.

Tet1 upregulation is compromised in the Kdm3b‐KO

We next employed RNA sequencing to identify genes that are regulated by Kdm3b during the 10‐day AA+2i mediated conversion to iPSCs. Previously, we had dissected the individual contribution of AA and 2i to iPSC generation and found that by day 2, several AA‐dependent genes had already been upregulated (Tran et al, 2015). Therefore, we profiled WT and Kdm3b‐KO lines treated with vehicle alone (DMSO) or AA+2i for 2 or 10 days. A comparison of this gene expression profile with RNA‐seq data of MEFs, MEFs induced for 48 h of expression of OSKM, pre‐iPSCs, and mESC (Chronis et al, 2017; Data ref: Chronis et al, 2017), revealed that the gene expression of Kdm3b‐KO cells clustered with pre‐iPSCs and away from MEFs and OSKM‐induced MEFs (Fig 2A). Thus, it is unlikely that deletion of Kdm3b reverts pre‐iPSCs back to the original MEF state. WT day 10 samples clustered with mESCs, while the Kdm3b‐KO day 10 samples most closely resembled WT day 2 upon AA+2i treatment (Fig 2A). Since fewer genes were misregulated in the Kdm3b‐KO on day 2 than day 10 (Appendix Fig S1F–I, Dataset EV1), the expression changes on day 10 are likely to have the greatest contribution to iPSC generation.

Figure 2. Tet1 upregulation is a mediator of Kdm3b function during reprogramming.

1. Pearson's correlation of transcripts per million (TPM) of WT and Kdm3b‐KO pre‐iPSC upon DMSO or AA+2i treatment for 2 or 10 days and mESC, reprogramming populations from Chronis et al (2017), (labeled as CC).
2. k‐means clustering of the union of DEG (FDR > 0.99) between day 2 and DMSO in WT and Kdm3b‐KO pre‐iPSCs. Values represent a log₂‐fold change of day 2 over DMSO. Key pluripotency genes are shown on the right.
3. k‐means clustering of the union of DEG (FDR > 0.99) between day 10 and DMSO in WT and Kdm3b‐KO pre‐iPSCs. Values represent a log₂‐fold change of day 10 over DMSO. Key pluripotency genes or epigenetic regulators are shown on the right.
4. Expression of Tet enzymes and Tdg during AA+2i treatment in WT or Kdm3b‐KO pre‐iPSCs. Error bars represent the standard deviation of two biological replicates from the RNA‐seq sample.
5. Quantification of percent Nanog‐GFP cells of WT, Kdm3b‐KO, and Kdm3b‐KO pre‐iPSC clones expressing Tet1 catalytic domain (Tet1‐CD), Nanog, or both. Error bars represent a standard deviation of three biological replicates. *P < 0.05 by paired t‐test.

In general, developmentally regulated gene activation was compromised in the Kdm3b‐KO on both day 2 and day 10 (Appendix Fig S1F–I). On day 2, this set included Esrrb (Fig 2B), and by day 10, several more pluripotency genes, such as Dppa5a and Zfp42, were affected (Fig 2C‐ Cluster D). Chromatin immunoprecipitation of H3K9me2 had a poor signal‐to‐noise ratio, and whether the misregulated genes differed in the enrichment for this modification could not be conclusively determined (data not shown).

In addition to the pluripotency genes, a specific group of chromatin‐modifying proteins was exclusively upregulated by day 10 (Cluster A) in the WT but not the Kdm3b‐KO cells (Fig 2C). Among these genes, we were intrigued by Tet1, since DNA methylation patterns are thought to be reset at later stages of reprogramming (Mikkelsen et al, 2008; Hussein et al, 2014; Lee et al, 2014). From our RNA‐seq data, only the levels of Tet1, but not of Tet2, were affected in the Kdm3b‐KO (Fig 2D).

To determine whether Tet1 played a functional role in the Kdm3b‐KO, we expressed the Tet1 catalytic domain (CD) either alone or in combination with Nanog. A previous publication has shown Nanog targets the Tet proteins to relevant locations during reprogramming (Costa et al, 2013) (Appendix Fig S1J and K). Despite increased 5hmC levels in the presence of Tet1‐CD alone (Appendix Fig S1J), the conversion of the Kdm3b‐KO to iPSC was not improved (Fig 2E). Similarly, Nanog alone did not improve conversion to iPSCs (Fig 2E). However, when both the Tet1‐CD and Nanog were overexpressed together, there was a significant albeit incomplete rescue of the phenotype to the iPSC state (Fig 2E). Therefore, the compromise of Tet1 expression in the Kdm3b‐KO affects pluripotency acquisition.

Transient and sustained 5hmC‐enriched loci are associated with distinct regulatory elements during the transition to pluripotency

A triple knockout of the TET proteins compromises reprogramming by preventing the mesenchymal‐to‐epithelial transition (Hu et al, 2014). Having established that TET1 expression is important for later stages of reprogramming beyond that of pre‐iPSCs, which have already undergone MET, we wanted to determine the 5hmC landscape during the process. Therefore, we performed 5hmC enrichment, followed by next‐generation sequencing, on days 0, 2, and 10 of the transition of pre‐iPSCs to iPSCs and compared it to the pattern in mouse ESCs.

We found regions that undergo loss, gain, or transiently accumulate 5hmC. The loss (Tcea3) or gain of 5hmC (Epcam) could occur immediately on day 2 or could be delayed (Dppa4, Ano2) (Fig 3A), while transient accumulation peaked on day 2 (Gdf3) (Fig 3A). A net loss of 5hmC on day 10 was the most prevalent pattern followed by a transient accumulation and a net gain (Fig 3B). We performed k‐means clustering centered on the 5hmC peaks obtained on day 0, day 2, day 10, and mESCs, to determine how the patterns changed as the cells converted to iPSCs (Fig 3C). The regions with high 5hmC on day 0 (Clusters B, C, and D) tended to lose 5hmC (Fig 3C) until reaching similar levels observed in mESCs by day 10 (Appendix Fig S2A). Regions enriched on day 2 (Clusters E, F, and H) have a transient behavior where they are low on both day 0 and day 10 but increased 5hmC levels on day 2 (Fig 3C, Appendix Fig S2A). Lastly, sites that have high 5hmC at day 10 (Clusters A and G) show a gradual increase from day 0 to day 10 and reach levels comparable to mESCs (Fig 3C, Appendix Fig S2A). We also mapped the locations of 5hmC enrichment after 12 and 72 h of AA+2i exposure in ESCs (Appendix Fig S2A). From principal component analysis, at a genome‐wide level, the WT cells on day 10 of AA+2i treatment resembled ESCs that were exposed to AA+2i for 12 h more than ESCs grown in serum (Appendix Fig S2B). Some locations in Clusters C, D, and E also change in 5hmC enrichment in ESCs exposed to AA+2i (Appendix Fig S2A), suggesting that they may be particularly sensitive to culture conditions (Blaschke et al, 2013).

Figure 3. Dynamic 5hmC patterns during iPSC conversion.

1. Representative 5hmC profiles during AA+2i conversion of WT pre‐iPSCs. Illustration of the 5hmC pattern is shown in the bottom panel.
2. Venn diagram of called peaks at days 0, 2, and 10. Inset represents 5hmC pattern and the number of peaks displaying a pattern of enrichment.
3. Heatmap of k‐means clustered 5hmC patterns with histone modifications in pre‐iPSCs and mESCs from Chronis et al (2017).
4. Cis‐element annotation of 5hmC patterns in clusters obtained by k‐means in (C).

Combining Cis‐regulatory element annotation system (CEAS; Shin et al, 2009) with the distribution of histone modifications in pre‐iPSCs and ESCs (Chronis et al, 2017; Data ref: Chronis et al, 2017), we determined whether the patterns of 5hmC localization corresponded to particular genomic regions (Fig 3C and D). Clusters A and G which have a net gain of 5hmC enrichment on day 10 become enriched for H3K36me3 in ESCs, indicating association with regions of active transcriptional elongation (Fig 3C, Appendix Fig S2C). Cluster B loci with a net loss of 5hmC on day 10 had poised enhancer‐associated marks H3K4me1 and H3K4me2 in pre‐iPSCs that gained H3K27ac in ESCs, suggesting that the erasure of 5hmC leads to activation of enhancers (Fig 3C, Appendix Fig S2C). Since Cluster H loci that have a transient gain of 5hmC are devoid of most activating modifications in pre‐iPSCs (Fig 3C, Appendix Fig S2C) and demonstrate the greatest gain in promoter‐enriched H3K4me3 among all of the clusters (Fig 3C, Appendix Fig S2C) they are likely to represent promoters that become activated in ESCs. Taken together, at regulatory regions such as promoters and enhancers, the erasure of 5hmC is associated with an increase in activating histone modifications suggesting that it has a net negative effect on transcription. We hypothesized that these regulator regions with a transient or loss patterns of 5hmC enrichment (Clusters B, C, D, E, F, and H) could represent regions that begin as 5mC in somatic cells that are converted to 5hmC by the Tet enzymes and processed to unmodified cytosine through passive dilution or active excision via the Tdg enzyme (Wu & Zhang, 2011, 2017; Kohli & Zhang, 2013; Pastor et al, 2013).

Transient gain of 5hmC occurs in regions that lose 5mC during reprogramming

To determine whether these regions were indeed undergoing active DNA demethylation, we used bisulfite sequencing data from MEFs and iPSCs (Rais et al, 2013; Data Ref: Rais et al, 2013). Bisulfite sequencing does not distinguish between 5mC and 5hmC; therefore, we looked for regions that had a bisulfite signal in MEFs which was completely lost in iPSCs (Rais et al, 2013) (Methods). We found that the transient pattern of 5hmC in Cluster H was most enriched for loci with a complete loss of a bisulfite signal in these differentially methylated regions and included several important pluripotency loci such as Pou5f1, Gdf3, Dppa4, and Tcf7 (Fig 4A, Appendix Fig S2D). Therefore, the transient accumulation of 5hmC may directly lead to demethylation and further activation of pluripotency‐related loci. We further corroborated these results with an independent published dataset of Tet‐assisted bisulfite sequencing (TAB‐Seq) performed on day 0 and day 5 of OSKM reprogramming from MEFs (Hu et al, 2014; Data ref: Hu et al, 2014). TAB‐Seq can distinguish between 5mC and 5hmC. There was a significant overlap of 5hmC‐enriched regions from our study, with regions that gained 5hmC during MEF reprogramming on day 5 (Appendix Fig S2D and E) and included several pluripotency loci (Fig 4A).

Figure 4. Active DNA demethylation is important for pluripotency acquisition.

1. Representative profiles of regions that undergo demethylation through a 5hmC intermediate. Top—5hmC profile during AA+2i treatment. Middle—Percent methylation via bisulfite sequencing in MEFs and iPSCs, data from Rais et al (2013). Bottom—Percent 5hmC via Tet‐assisted bisulfite sequencing of MEFs expressing OSKM for 0 or 5 days, data from Hu et al (2014).
2. Left—Quantification of percent Nanog‐GFP‐positive colonies. Right—Total Nanog‐GFP‐positive colonies obtained after ten days of AA+2i treatment in Tdg‐depleted pre‐iPSC (Tdg) or control (Empty). Error bars represent a standard deviation of three biological replicates. *P < 0.05‐ paired t‐test.

To determine whether Tdg‐mediated 5hmC excision is critical in this process, we depleted Tdg in pre‐iPSCs (Fig 4B, Appendix Fig S3A) and subjected the cells to AA+2i‐mediated conversion. We found a significant reduction in the number of iPSC colonies (Fig 4B) indicating that active DNA demethylation is important for transition to iPSCs.

Overall, these data indicate that active DNA demethylation through Tdg contributes to the conversion to iPSCs.

5hmC enrichment is compromised in Kdm3b‐deficient cells affecting activation of pluripotency genes

Since Tet1 is not upregulated in Kdm3b‐KO pre‐iPSCs (Fig 2D) and Tet1‐CD overexpression can contribute to generating iPSCs from the Kdm3b‐KO (Fig 2E), we next wanted to investigate the interplay between the erasure of H3K9 methylation and 5hmC acquisition. Kdm3b‐KO cells presumably retain both H3K9 methylation and 5mC/5hmC at shared pluripotency loci that must be activated. Given that the H3K9me2 inhibition (Fig 1F) and Tet1‐CD overexpression (Fig 2E) can individually contribute to the phenotype of Kdm3b‐KO conversion to iPSCs, we wondered if the erasure of the two modifications was linked. To test this hypothesis, we exposed the Kdm3b‐KO pre‐iPSCs that overexpressed both Tet1‐CD and Nanog to the G9a/Glp inhibitor. Interestingly, exposure to UNC0638 increased the number of colonies obtained in conjunction with the increase in Tet1‐CD and Nanog (Fig 5A). The inhibition of H3K9 methyltransferases increased Tet1 levels in Kdm3b‐KO about threefold which is similar to the increase from RNA‐seq reads between day 0 and day 10 of reprogramming (Appendix Fig S3B, Fig 2D). These observations support a link between H3K9 and DNA demethylation.

Figure 5. Kdm3b‐KO cells retain 5hmC at pluripotency‐associated locations where it should be resolved in reprogramming.

1. Quantification of percent Nanog‐GFP‐positive cells of WT or Kdm3b‐KO pre‐IPSC clones expressing Nanog, Tet1‐CD, or Nanog+Tet1‐CD after 15 days of AA+2i or AA+2i with 4 or 5 μM of UNC0638. Error bars represent standard deviation of three biological replicates. **P < 0.01 by paired t‐test.
2. Left—Heatmap of K‐mean clustered 5hmC patterns in WT pre‐iPSC with 5hmC profile of Kdm3b‐KO pre‐iPSC treated with AA+2i for 0, 2, or 10 days. Middle—Metaplot of 5hmC signal on day 10 between WT (orange) and Kdm3b‐KO (black) for each cluster. Y‐axis scale is in normalized coverage data (1 × 10⁶/Total Count). Right—Illustration of 5hmC pattern in WT and Kdm3b‐KO during AA+2i treatment.
3. Representative 5hmC profiles during AA+2i conversion of WT and Kdm3b‐KO pre‐iPSC.
4. Scheme for identifying genes with both misregulated 5hmC and expression. Heatmap of twofold misregulated 5hmC and twofold differential expression between WT and Kdm3b‐KO after 10 days of AA+2i treatment (607 genes). Peaks are sorted by 5hmC signal in WT. RNA‐seq values represent a log₂‐fold change of day 10 over DMSO for either WT or Kdm3b‐KO pre‐iPSCs.
5. Box plot of expression for genes that are twofold misregulated 5hmC and expression between WT and Kdm3b‐KO on day 10 (607 genes). Middle line designates median value, while box ranges between 25^th and 75^th percentile and whiskers indicate minimum and maximum values. ****P < 0.0001‐ paired t‐test
6. Gene Ontology for genes that are twofold misregulated 5hmC and expression between WT and Kdm3b‐KO on day 10.

Given the different patterns of 5hmC that are established in WT cells, we hypothesized that the distribution of 5hmC may be affected by Kdm3b deletion. Therefore, we determined how 5hmC localization is affected in Kdm3b‐deleted intermediates. We found that 5hmC patterns were disrupted in all patterns that were observed in the wild‐type cells, but to varying degrees (Fig 5B and C, Appendix Fig S3C). Considering that the increase in Tet1 expression is compromised in the Kdm3b‐KO cells, it was not surprising that regions that gain 5hmC by day 10 were less enriched (Clusters A and G) (Fig 5B). Strikingly, several clusters that transiently accumulate 5hmC in WT cells (Cluster H) or lose 5hmC by day 10 (Cluster B) remain enriched for 5hmC in the Kdm3b‐KO (Fig 5B). This suggests that lack of Tet1 upregulation in the Kdm3b‐KO cells or the retention of H3K9me2 prevents the further oxidation of 5hmC. In fact, using principal component analysis, we found that the 5hmC enrichment in Kdm3b‐KO cells represents an intermediate point in the trajectory to becoming a pluripotent cell (Appendix Fig S2B).

We also found that 5hmC enrichment was significantly enhanced at a large majority of locations in Kdm3b‐KO ESCs as compared to WT mESCs (Appendix Fig S3D and E). Therefore, the functional link between H3K9me2 demethylation and resolution of 5hmC to more oxidative forms may be retained in different cell types.

We next asked whether the lack of 5hmC resolution in the Kdm3b‐KO cells contributed to altered gene expression during reprogramming. Therefore, we examined the relationship between those regions that had a greater than twofold difference in both 5hmC enrichment and gene expression between the WT and Kdm3b‐KO cells on day 10 (Fig 5D, Dataset EV2). The vast majority of the misregulated 5hmC locations were associated with genes that had to be upregulated during the conversion to iPSCs (Fig 5E, Dataset EV2). Interestingly, genes that remained repressed in the Kdm3b‐KO were evenly distributed among regions that gained or lost 5hmC in the Kdm3b‐KO (Fig 5C, Dataset EV2). There was a significant lack of upregulation of genes in the Kdm3b‐KO as compared to WT cells (Fig 5E). Gene Ontology analysis of the misregulated set of genes indicated compromised functional categories of embryonic and organ development that are associated with pluripotency and embryogenesis (Fig 5F). Therefore, the retention of 5hmC is associated with key pluripotency genes remaining inactivated, thus affecting reprogramming to iPSCs in the Kdm3b‐KO.

5hmC retention occludes POU5F1 binding

To determine how the retention of 5hmC could affect gene expression, we next asked whether there was an occlusion of binding of transcription factors. We mapped the binding of the reprogramming factors and pluripotency‐related transcription factors in pre‐iPSCs and ESCs onto the 5hmC clusters (Chronis et al, 2017; Data ref: Chronis et al, 2017; Xiong et al, 2016; Data ref: Xiong et al, 2016; Appendix Fig S4A). Among these factors, Cluster A was enriched for binding of Sall1 (Appendix Fig S4A and C). Notably, Sall1 was also found at several other clusters corroborating its role in targeting Tet1 and Tet2 in mESCs (Xiong et al, 2016). Strikingly, loci in Cluster B were already bound by Klf4 in pre‐iPSCs but became enriched for the binding of Pou5f1 only in ESCs. In the regions that had a net loss of 5hmC enrichment (Cluster C and some regions of Cluster H), there was a stronger enrichment of Pou5f1 in ESCs as compared to pre‐iPSCs (P‐value < 10⁻²⁰ by Wilcoxon signed rank test; Appendix Fig S4A and B).

Varying degrees of methylation have differing effects on transcription factor binding in vitro, with some proteins having a greater affinity for 5mC, 5caC, and 5fC as compared to 5hmC (Spruijt et al, 2013; Yin et al, 2017). Pou5f1 is capable of binding to 5mC‐containing probes and regions, but whether it can bind to 5hmC is unknown (Yin et al, 2017). Therefore, we clustered POU5F1‐bound regions in mESCs into two groups, one with high POU5F1 binding in both pre‐iPSC and mESC (Cluster 1) and sites that have increased binding in mESC as compared to pre‐iPSC (Cluster 2). For these clusters, we examined the 5hmC patterns during reprogramming (Fig 6A). Strikingly, in Cluster 2 that had diminished binding of POU5F1 in pre‐iPSCs as compared to mESCs, there was pre‐existing 5hmC enrichment (Fig 6A–C). In contrast, in Cluster 1 where the binding strength of POU5F1 was similar in pre‐iPSCs and ESCs, there was a depletion of 5hmC enrichment (Fig 6A–C). From this genome‐wide analysis, we focused on the pattern of binding of Pou5f1 in ESCs in regions that had the greatest loss in 5hmC enrichment between day 0 and day 10 (Fig 6D) or between day 2 and day 10 (Appendix Fig S4D). There was an inverse correlation between Pou5f1‐binding in ESCs and 5hmC enrichment on day 0 and day 2. Pou5f1 binding was enhanced in ESCs at locations where 5hmC was removed from pre‐iPSCs. This suggested that the retention of 5hmC in the Kdm3b‐KO could inhibit the recruitment or binding of Pou5f1.

Figure 6. Ectopic 5hmC retention is incompatible with POU5F1 chromatin association.

1. Heatmap of POU5F1‐bound peaks in both mESC and pre‐IPSC (Cluster 1) and mESC alone (Cluster 2) with 5hmC levels during days 0, 2, and 10 of AA+2i treatment.
2. Metaplots of POU5F1 binding for Cluster 1 (Left) and Cluster 2 (Right) regions in pre‐iPSCs and mESCs.
3. Metaplots of 5hmC Enrichment for Cluster 1 (Left) and Cluster 2 (Right) regions for days 0, 2, and 10 of AA+2i treatment.
4. Left—Scheme for identifying sites that lose 5hmC and their association with POU5F1 binding. Right—Heatmap of 5hmC profile during AA+2i treatment of WT pre‐iPSC and POU5F1 binding in pre‐iPSCs or mESCs at the loci identified in the left panel.
5. Representative 5hmC profiles during AA+2i conversion of WT and Kdm3b‐KO pre‐iPSC with POU5F1 binding in pre‐iPSCs and mESCs. Loci were assigned to the nearest gene.
6. ChIP‐PCR for POU5F1 during AA+2i reprogramming of WT and Kdm3b‐KO pre‐iPSCs. Values are fold change of enrichment between WT and Kdm3b‐KO (WT/Kdm3b‐KO). Error bars represent a standard deviation of two technical replicates. Three independent reprogramming experiments are shown.

To test this hypothesis, we performed chromatin immunoprecipitation for POU5F1 at selected loci that demonstrated an inverse relationship between POU5F1 binding and 5hmC enrichment in WT and Kdm3b‐KO cells (Fig 6E). Although the amount of enrichment of POU5F1 at each locus varied in the experiments due to differences in reprogramming efficiency, we noticed a consistent trend: By day 10, POU5F1 binding in the wild‐type cells was much greater than that in Kdm3b‐KO cells (Fig 6F) suggesting that a threshold of 5hmC must be overcome to enable POU5F1 association with chromatin (Fig 6F). Notably, the levels of POU5F1 protein are comparable in the wild‐type and Kdm3b‐KO cells at each timepoint of the conversion (Appendix Fig S4E) but do not compensate for the difference in enrichment of POU5F1 at specific loci.

Tet1 has a role late in reprogramming

Reprogramming of MEFs to pluripotency occurs in several phases with an early mesenchymal‐to‐epithelial transition marked by the upregulation of E‐cadherin. MEFs that are derived from a triple knockout of the Tet genes cannot complete the MET (Hu et al, 2014). Importantly WT pre‐iPSCs already express E‐cadherin (Cdh1), a marker for the epithelial transition (Mikkelsen et al, 2008; Sridharan et al, 2009; Esteban et al, 2010). Our results suggest that 5hmC regulation is important for the later stage of pluripotency gene activation. Several earlier reports have suggested that Tet2 is important for reprogramming, but the role of Tet1 has remained unclear (Doege et al, 2012; Chen et al, 2013; Gao et al, 2013; Schwarz et al, 2018). To establish whether the Tet enzymes function at different phases of reprogramming, we first determined their expression patterns. Since reprogramming is a low‐efficiency process where expression patterns can be obscured by bulk RNA sequencing, we analyzed our single‐cell transcription data during a reprogramming time course (Fig 7A; Tran et al, 2019b). We found that Tet2 is upregulated earlier in reprogramming similar to the activation of Cdh1 by day 6 (Fig 7B). By contrast, Tet1 is upregulated later at the end of reprogramming similar to the activation of pluripotency gene Nanog (Fig 7B).

Figure 7. Tet1 activation is important at later stages of pluripotency acquisition.

1. tSNE of Tet1, Tet2, Nanog, and Cdh1 expression in single‐cell RNA sequencing of MEF undergoing OSKM reprogramming from Tran et al (2019b). MEF, reprogramming (REPROG), and mESCs cells are highlighted in red dotted lines. Specific reprogramming timepoints are highlighted within the reprogramming cluster in arrows. Color corresponds to log₁₀ of transcript counts for each individual cell.
2. Bar graphs indicating number of cells expressing Tet1, Tet2, Cdh1, or Nanog from the reprogramming population (days 3, 6, 9, and 12). Number of cells per reprogramming population is highlighted within legend.
3. Scheme for deriving Tet1 and Tet2 double‐knockout pre‐iPSC lines. Tet1KO/KO Tet2fl/fl or Tet1WT/WT Tet2fl/fl MEFs were treated with retroviruses containing OSKM. Tet1KO/KO Tet2fl/fl or Tet1WT/WT Tet2fl/fl pre‐iPSC clones were isolated and treated with either adenoviral‐expressed Cre recombinase or empty vector to generate a population of pre‐iPSCs with Tet1KO/KO Tet2fl/fl, Tet1KO/KO Tet2KO/KO, Tet1WT/WT Tet2fl/fl, and Tet1WT/WT, Tet2KO/KO genotypes. Scale bar, 50 μm.
4. Quantification of percent Nanog‐positive cells by intracellular flow cytometry in each genotype as indicated of Tet1^WT/WT Tet2^fl/fl, Tet1^WT/WT Tet2^KO/KO, Tet1^KO/KO Tet2^fl/fl, and Tet1^KO/KO Tet2^KO/KO after 10 days of AA+2i treatment. Error bars represent a standard deviation of 3–6 different pre‐iPSC biological replicates. p‐values were calculated by unpaired t‐test. **P < 0.01.
5. Top—Model for action of Kdm3b and 5hmC during pre‐iPSC to iPSC conversion. Kdm3b‐mediated demethylation of H3K9me2 and 5hmC processing allows recruitment of POU5F1 and pluripotency gene upregulation. In Kdm3b‐deleted pre‐iPSCs, 5hmC is maintained and prevents POU5F1 binding at specific genes. Bottom—Distinct temporal expression of Tet1 and Tet2 during reprogramming.

To determine whether Tet1 and Tet2 can function redundantly after CDH1 upregulation, we derived Tet1 and Tet2 double KO (DKO) pre‐iPSC cell lines (Fig 7C). For this purpose, we infected MEFs with intact Tet1 and floxed Tet2 alleles or Tet1 deleted and Tet2 floxed with OSKM retroviruses (Fig 7C) and established independent pre‐iPSC cell lines that expressed CDH1 but not NANOG (Fig 7C, Appendix Fig S5A). Upon deletion of Tet2, reprogramming could still be completed indicating redundancy with Tet1 (Fig 7D, Appendix Fig S5B). When Tet1 was knocked out in the context of intact Tet2, CDH1‐expressing clones could be obtained (Appendix Fig S5C). In these clones, the conversion to iPSCs was compromised to about half the efficiency found in WT cells (Fig 7D). Importantly when both Tet1 and Tet2 were deleted (Appendix Fig S5D), there was no conversion to iPSCs (Fig 7D). Therefore, Tet enzyme function is critical for the late stages of reprogramming to iPSCs.

Taken together, these data provide evidence for a model for gene activation, where the retention of 5hmC diminishes the recruitment or binding of POU5F1, which prevents upregulation of pluripotency‐related gene expression. Activation of Tet1 in a Kdm3b‐dependent manner erases DNA methylation at H3K9‐methylated loci (Fig 7E).

Discussion

In this study, we have uncovered a transcriptional and post‐transcriptional link that functionally connects DNA demethylation by H3K9me1/2 demethylation. Such crosstalk is significant because chromatin modifications often work in concert with each other to establish cell fate. In the context of differentiation from the pluripotent state, the H3K9me1/me2 methyltransferase, G9a, recruits DNMT3a and DNMT3b to pluripotency loci to ensure chromatin compaction and gene silencing (Epsztejn‐Litman et al, 2008). From our results, the coordinated removal of these repressive marks is required for the reversal of cell identity.

H3K9 methylation is generally associated with transcriptional repression and silencing of repeat elements of the genome and is enriched in heterochromatin. H3K9me2 and H3K9me3 are also found at overlapping locations and can be catalyzed by enzymes that have redundant functions (Black et al, 2012; Becker et al, 2016). The specific role of H3K9me2 has been difficult to parse due to the megabase scale regions that the modification is enriched in that are thought to overlap with the lamina‐associated domains (Wen et al, 2009; Lienert et al, 2011; Poleshko et al, 2017). Single deletion of Kdm3a or Kdm3b has little effect in mESCs; however, dual deletion of Kdm3a and Kdm3b leads to cell death, indicating redundant functions in the maintenance of pluripotency (Kuroki et al, 2018). These observations have a larger implication for chromatin biology in general where enzymes with the same specificity may associate with distinct protein complexes to mediate cell‐type‐specific effects.

Additionally, deletion of Kdm3b in stalled intermediates prevented Tet1 expression and resulted in misregulated 5hmC enrichment during iPSC conversion. The role of the Tet proteins has been explored before in reprogramming from MEFs, in CEBP‐alpha‐mediated rapid reprogramming of B cells, and also in the conversion of stalled intermediates where a significant effect on the mesenchymal‐to‐epithelial transition was observed (Doege et al, 2012; Chen et al, 2013; Di Stefano et al, 2014; Hu et al, 2014; Sardina et al, 2018). Here, we find an important role of Tet1 in later stages of reprogramming. In fact, Tet1 can replace Pou5f1 in the reprogramming cocktail (Gao et al, 2013). In this context, it is interesting that we find that the continued retention of 5hmC is incompatible with Pou5f1 binding at certain locations. It is possible that, in these locations, Pou5f1 can act as a pioneer factor to bring in other pluripotency components, but in the absence of Pou5f1, Tet1 itself is able to allow the binding of other ESC‐specific transcription factors.

Chromatin modifications often work cooperatively to regulate gene expression. The erasure of H3K9me2 and the correct targeting of Tet1 activity function cooperatively to convert the Kdm3b‐KO into iPSCs. In the Kdm3b‐KO, the Tet proteins can initiate the oxidation of 5mC, but perhaps the continued presence of H3K9me2 in the same nucleosome affects catalytic activity (Fig 7E). Kdm3b may directly recruit Tet1 increasing the concentration locally at H3K9me2 sites. In this regard, a direct interaction between the H3K9me2/me3 demethylase Kdm4c and Tet1 has been reported in mESCs (Sim et al, 2017). It is also possible that there is an indirect connection as observed in early development. In the female pronucleus of the zygote, H3K9me2‐containing locations are bound and protected from Tet3‐mediated 5mC oxidation by the protein Dppa3 (Nakamura et al, 2012). Although Dppa3 is only expressed at the very end of reprogramming, it is possible that other H3K9me2 reader proteins may repel Tet1 recruitment or activity. These mechanisms will be interesting to investigate in future studies.

Materials and Methods

MEF cell isolation

Mouse embryonic fibroblasts were isolated from E13.5 embryos as described in Tran et al (Tran et al, 2015) from reprogrammable mice homozygous for the Pou5f1‐2A‐Klf4‐2A‐IRES‐Sox2‐2A‐c‐Myc (OKSM) transgene at the Col1a locus and for the reverse tetracycline transactivator (rtTA) allele at the Rosa26 locus (Sridharan et al, 2013). Mice were maintained according to protocols approved by the UW‐Madison IACUC.

Cell culture and reprogramming

Mouse embryonic fibroblasts were thawed and maintained in ESC media (knockout DMEM, 15% FBS, L‐glutamine, Pen/Strep, NEAA, 2‐mercaptoethanol, and leukemia inhibitory factor) for 2 days before plating. On day −1, MEFs were plated and induced on day 0 with 2 μg/ml doxycycline for OKSM expression (Table 1). For KSR reprogramming, media was switched on day 3 to ESC media replaced with 15% Knockout serum replacement (Thermo). ESCs (V6.5 line) were maintained in ESC media on a feeder layer of irradiated MEFs. For pre‐iPSCs (12‐1 line) (Sridharan et al, 2013) carrying a Nanog‐GFP reporter (Tran et al, 2015) reprogramming, cells were plated at 200,000 cells per well (6 well) on day 0 and treated with ascorbic acid (Sigma, A8960) (50 μg/ml), 2i: 1 μM PD‐0325901 (Stemgent, 04‐0006‐10) and 3 μM CHIR‐99021 (Stemgent, 04‐0004‐10), and 4–5 μM UNC0638 (Fisher Scientific, 501150439) when noted. Media was replaced every 2 days. Cells were split to 200,000 cells on day 4 and evaluated for conversion efficiency on day 10 or 15 by counting Nanog‐GFP‐positive colonies and/or by flow cytometry.

Table 1.

Primers used in this study

Gene name	5’	3’
Nanog (RT–PCR)	CATCCCGAGAACTATTCTTGCT	GAGGCAGGTCTTCAGAGGAA
Tet‐Cd (RT–PCR)	GACCAAGTGTGGCTGCTGT	CCTGAGAGCTCTTCCCTTCC
Cdh1 (RT–PCR)	GCCACCAGATGATGATACCC	GGAGCCACATCATTTCGAGT
Kdm3b (RT–PCR)	TGCTAATGGGATGCATCTGT	CGGCTGAAAATCTCTTGTAGC
Tdg (RT–PCR)	TTCAACGGCGTCTCTGAAG	CCCAGGGTAGTGATGTCCTT
Tet2 (Genotype)	AAGAATTGCTACAGGCCTGC	TTCTTTAGCCCTTGCTGAGC
Tet2 (Genotype)		TAGAGGGAGGGGGCATAAGT
Esrrb (ChIP Enrich)	AAGCCTTTGGATTTGGAGGT	AATCACCACAGGCTTGTTCC
D1Pas1 (ChIP Enrich)	ATTTCAGGAGTGGCTGTTGG	ATTGTCAAGTCAGCCGAACC
Tsc22d1 (ChIP Enrich)	AAATTGTCGGTCCAAACTGC	CACGCCTCCTAATCCTTTCA

Generating Kdm3b‐KO pre‐iPSCs

Kdm3b‐KO pre‐iPSC line was generated by cloning gRNA targeting exon 1 or exon 2 (See “Crispr gRNA used to target Kdm3b” Table 2) of Kdm3b into the gRNA cloning vector (Addgene, 41824). Pre‐iPSC was transfected with 1 μg gRNA cloning vector and 1 μg hCas9 (Addgene, 41815) using ViaFect Transfection Reagent (Promega, E4981) in a 6‐well plate. Cells were plated at low density to isolate single clones and screened via Western blot for Kdm3b.

Table 2.

Crispr gRNA used to target Kdm3b

Exon 1	CTCACTACCGCCGCCGCCGT
Exon 2	ACTTGTTTGGGCTCCCCGGA

Generating Tet1‐CD‐ and Nanog‐overexpressing pre‐iPSC lines

Tet1‐CD‐overexpressing pre‐iPSC lines were generated by using plasmid pcDNA3‐Flag‐Tet1‐CD (Addgene plasmid #70129). After selection with G418, positive clones were identified by FLAG Western blot (Sigma‐Aldrich, F3165) and ability to increase 5hmC via dot blot (See Dot Blot section in methods). Nanog‐overexpressing Kdm3b‐KO and Kdm3b‐KO + TetCD pre‐iPSC lines were generated using pPB‐CAG‐Nanog‐pA‐pgk‐hph (Addgene plasmid #74915). After selection with hygromycin, positive clones were identified by Nanog Western blot or intracellular flow cytometry (Cosmo Bio Co., RCAB0002P‐F).

Tet1, Tet2 KO, and Tdg KD pre‐iPSC generation

Tet1^KO/KO and Tet2^fl/fl MEFs (a gift from Prof. Kathrin Plath, UCLA) were infected with pMX retrovirus encoding reprogramming factors Pou5f1, Sox2, Klf4, and c‐Myc. Colony‐like structures were picked on day 14 and passaged on irradiated MEFs. Colonies were immunostained for both CDH1 and NANOG, and clones with high CDH1 that lacked NANOG expression were designated as pre‐iPSCs. Deletion of Tet2 allele on Tet1^KO/KO pre‐iPSCs was accomplished by transduction of either AdEmpty or AdCre lentivirus from Vector Development Lab for 1 h in serum‐free media. Tet2 deletion was confirmed by genotype PCR following conditions from Jackson Labs (#017573). Generated Tet1^KO/KO Tet2^fl/fl, Tet1^WT/WT Tet2^fl/f, Tet1^WT/WTTet2^KO/KO, and Tet1^WT/WT Tet2^KO/KO pre‐iPSC lines were tested for reprogramming upon AA+2i treatment. Tdg KD pre‐iPSC is generated by transducing WT pre‐iPSCs with either shRNA targeting Tdg or luciferase. Knockdown was validated by RT–PCR.

Intracellular flow cytometry

Samples were harvested and fixed using BD Cytofix/Cytoperm solution (BD, BD554714) for 30 min. Cells were washed and treated with BD Plus Permeabilization Buffer (BD, BDB561651) for 15 minutes on ice. Cells were washed and stained using anti‐NANOG antibodies conjugated to Alexa Fluor 647 (eBioscience, 560279, 1:2,000) for 1 h at room temperature. Samples were washed and run on BD Accuri Flow Cytometer, and data were analyzed using FlowJo software.

Dot blot

5hmC dot blots were performed essentially as in Tran et al (2015). Briefly, cells were lysed overnight, and DNA was precipitated and quantified by Qubit. Samples were cross‐linked onto a nitrocellulose membrane by backing at 100°C for 10 min. Membrane was blocked for 1 h in 5% milk, blotted with 5hmC diluted 1:10,000 (Active Motif, 39770) overnight at 4°C, and detected with chemiluminescence.

Western blot

Equivalent number of cells was lysed using a syringe in 1X Laemmli sample buffer. Samples were run on polyacrylamide gels and transferred onto nitrocellulose membranes overnight at 4°C. Membranes were blocked in 5% milk for 30 min at room temperature and blotted for primary antibodies overnight at 4°C. Antibodies used were H3K9me1 1:5,000 (Abcam Ab9045), H3K9me2 1:1,000 (Abcam Ab1220), Kdm3b 1:1,000 (Cell Signaling 5377), Kdm3a 1:1,000 (Proteintech, 12835‐1‐AP), 1:1,000 Jmjd1c (MBL, D356‐3), 1:1,000 POU5F1 (Santa Cruz, sc‐5279 or sc‐8628), 1:1,000 alpha‐Tubulin (Cell Signaling, 3873), and 1:1,000 NSD3 (Thermo Fisher, PA5‐28972).

Immunofluorescence

Immunofluorescence was performed essentially as in Sridharan et al (2009). Briefly, coverslips were fixed in 4% paraformaldehyde and permeabilized in 5% Triton X‐100. Coverslips were washed and blocked in 5% normal donkey serum followed by overnight primary antibody staining at 4°C. Coverslips were washed in PBS with 0.1% Tween and stained with secondary antibodies for 30 min at room temperature, followed by washing and mounting on slides with Aqua‐Poly/Mount. Antibodies used were H3K9me2 1:100 (Abcam, Ab1220), H3K9me3 1:100 (Active Motif, 39162), 1:100 E‐cadherin (eBioscience, 14‐3249‐82), and 1:100 Nanog (Cosmo Bio, RCAB0002P‐F).

RNA sequencing

RNA sequencing was performed as in Tran et al (2015). Briefly, RNA was extracted using a Qiagen RNeasy Mini Kit. A single‐end cDNA library was constructed using 2 μg of RNA via TruSeq RNA Sample Preparation Kit (Illumina, RS‐122‐2002) according to the manufacturer's instruction. Samples were multiplexed and sequenced on a HiSeq 2500 at the University of Wisconsin—Biotechnology Center. Experiments were performed in biological duplicate.

Single‐cell RNA sequencing

Single‐cell data were obtained from Tran et al (2019b). (GSE108222) using MEF reprogramming populations (days 3, 6, 9, and 12) and ESCs (See “Datasets used in the study” Table 3). Data were processed as in Tran et al (2019b). Briefly, FASTQ files were aligned to mm10 genome using Spliced Transcripts Alignment to a Reference (STAR) and single cells were identified using Illumina BaseSpace SureCell RNA Single‐Cell Application. Count tables were generated from SureCell RNA Single‐Cell Application and were processed through Monocle 2 (Trapnell et al, 2014) on R version 3.4.3 (http://cole-trapnell-lab.github.io/monocle-release/docs/) to generate t‐SNE using 2 PC components, which represents at least 50% of the variance within the data. Identification of MEF, reprogramming, and ESC populations was performed by plotting the location of each population within the t‐SNE. To identify percentage co‐expression across MEF, reprogramming, and ESC cells, we used the CellTypeHierarchy function in Monocle 2. Cells were considered positive if they had at least one transcript count for targeted gene.

Table 3.

Datasets used in the study

H3K36me3 (mESCs)	GSM2417108
H3K36me3 (pre‐iPSCs)	GSM2417107
H3K4me1 (mESCs)	GSM2417088
H3K4me1 (pre‐iPSCs)	GSM2417087
H3K4me2 (mESCs)	GSM2417084
H3K4me2 (pre‐iPSCs)	GSM2417083
H3K4me3 (mESCs)	GSM2417080
H3K4me3 (pre‐IPSCs)	GSM2417079
H3K27ac (mESCs)	GSM2417096
H3K27ac (pre‐iPSCs)	GSM2417095
Bisulfite (MEFs)	GSM1564675
Bisulfite (iPSCs)	GSM1564678
TARRBS (MEF)	GSM1275170
TARRBS (D5 OKSM)	GSM1275172
Pou5f1 (mESCs)	GSM2417142
Pou5f1 (pre‐iPSCs)	GSM2417134
Sox2 (mESCs)	GSM2417143
Sox2 (pre‐iPSCs)	GSM2417135
Klf4 (mESCs)	GSM2417144
Klf4 (pre‐IPSCs)	GSM2417136
c‐Myc (mESCs)	GSM2417145
c‐Myc (pre‐iPSCs)	GSM2417137
Nanog (mESCs)	GSM2417187
Esrrb (mESCs)	GSM2417188
Sall1 (mESCs)	GSM1386970
Sall4 (mESCs)	GSM1386979
Tet1 (mESCs)	GSM1386972

5hmC enrichment and library preparation

5hmC enrichment was performed as in Li et al (2012). Briefly, DNA was sonicated to 300–600 bp fragments using Covaris S220. 50 μg of sonicated DNA was labeled with UDP‐6‐N3‐Glc (Active Motif Cat, 55020) at a final concentration of 100 μM using T4‐β‐glucosyltransferase (NEB, M0357S) for 1 h at 37°C. Labeled DNA was purified using QIAquick Nucleotide Removal Kit (Qiagen, 28304) and eluted in 30 μl. A biotinylation reaction was performed using DBCO‐S‐S‐PEG3‐Biotin (Click Chemistry Tools, A112P3‐10) conjugate at a final concentration of 150 mM at 37°C for 2 h. DNA was purified using QIAquick Nucleotide Removal Kit and eluted with 100 μl of water. DNA was quantified, and all samples were normalized before 5hmC capture. 5hmC was enriched by addition of 50 μl of Dynabeads MyOne Streptavidin C1 beads (Invitrogen Cat, 65001) after equilibration with washing buffer (10 mM Tris–HCl pH7.5, 1 mM EDTA, 2 M NaCl, 0.01% Tween‐20). Labeled DNA was incubated with beads for 15 min and washed three times with washing buffer. Beads were eluted using 50 mM of DTT for 2 h at room temperature. Buffer exchange was done using Micro Bio‐Spin 6 Columns (Bio‐Rad, 7326221) and purified using a Qiagen MinElute PCR Purification Kit (Qiagen, 28004). Library preparation was accomplished by NuGEN Ultralow Library System (NuGEN, 0344‐32) via the manufacture's protocol. Samples were multiplexed and sequenced on the HiSeq 2500 at the University of Wisconsin—Biotechnology Center in biological triplicate, and consensus peaks were used for further analysis.

Chromatin immunoprecipitation

Chromatin was prepared by cross‐linking cells with 1% formaldehyde followed by nuclei isolation and sonicated using a Covaris S220. Equal portions of protein A and protein G magnetic beads (Thermo Fisher Scientific, 1002D and 1004D, respectively) were washed and incubated with 5 μg of Pou5f1 antibody (Santa Cruz, sc‐5279 lot#D1119 or sc‐8628). Pre‐bound beads were incubated with 25 μg of chromatin overnight at 4°C rotating. Following IP, beads were washed twice with low salt buffer (50 mM HEPES, 0.1% SDS, 1% Triton X‐100, 0.1% deoxycholate, 1 mM EDTA, and 140 mM NaCl), high salt buffer (50 mM HEPES, 0.1% SDS, 1% Triton X‐100, 0.1% deoxycholate, 1 mM EDTA, and 500 mM NaCl), LiCl buffer (20 mM Tris, 0.5% NP‐40, 0.5% deoxycholate, 1 mM EDTA, and 250 mM LiCl), and TE. Elution of DNA was performed by incubating beads with elution buffer (50 mM Tris, 1 mM EDTA, and 0.1% SDS) followed by reverse cross‐linking at 65°C overnight with 1% SDS. Samples were treated with RNase A for 30 min followed by 2 h of proteinase K digestion. Magnetic beads were removed, and DNA was purified by phenol:chloroform extraction followed by ethanol precipitation.

RNA sequencing analysis

Sequenced reads were trimmed, filtered by quality, and adapter sequence clipped using the FASTX‐Toolkit. Accepted sequences were aligned to mouse reference genome constructed to include only annotated genes (i.e., NM_ RefSeqs) by Bowtie 2 alignment. Gene expression of aligned reads was quantified using RSEM (Li & Dewey, 2011). Differential expression (DE) was calculated using EBSeq program which uses a Bayes approach to calculate the distribution of expression (Leng et al, 2013). Genes were considered differentially expressed with a posterior probability > 0.99. k‐means clustering was performed using Gene Cluster 3.0 (de Hoon et al, 2004) using eight clusters and visualized by Java TreeView. Identification (Saldanha, 2004) of enriched functional annotation for individual clusters was determined by DAVID Bioinformatics Resources (Huang et al, 2009). A Venn diagram for overlapped genes was performed using web software Venny (http://bioinfogp.cnb.csic.es/tools/venny/).

5hmC enrichment processing

FASTQ files were aligned to the mm9 genome using Bowtie aligner with default parameters. BAM and index files were generated using the SAMtools toolkit (Li et al, 2009) with the “view” and “index” commands, respectively. Peak calling was performed using the macs2 (Shin et al, 2009) “callpeak” command between target and input samples with a P‐value cutoff of 1e‐4. Intersect peaks between replicates were considered high confidence peaks and used for downstream analysis. Intersect peaks between Kdm3b‐KO clone 1 and Kdm3b‐KO clone 2 were used for downstream analysis. To generate Venn graphs between day 0, 2, and 10 peaks, called peaks were overlapped via R package ChIPpeakAnno using (Zhu et al, 2010) the “findOverlapsOfPeak” command. For clustering 5hmC peaks, day 0, 2, 10 and mESC 5hmC peaks were normalized to an average signal across peaks and were k‐means clustered into eight clusters using the EaSeq (Lerdrup et al, 2016) “Cluster” tool. Peak annotation was done using the EaSeq “Annotate” tool with the center of the peak set at nearest start or end of a gene. Metaplots were generated by calculating the average ChIP density of an individual cluster using the “Average” tool and were merged using the “Overlay” tool in EaSeq. To determine distribution of 5hmC peak to genomic region, the Cis‐regulatory element annotation system (CEAS; Shin et al, 2009) was with the mm9 gene annotation table file and default settings. Metaplot for 5hmC signal across a gene set was done by ngs.plot using default settings (Shen et al, 2014).

ChIP‐Seq processing

FASTQ files from published data were obtained from Gene Expression Omnibus (GEO) and associated GSM number can be found in “Datasets used in study” (Table 3). OSKM, Nanog, and Esrrb in pre‐iPSCs and mESCs were from Chronis et al (Chronis et al 2017; Data ref: Chronis et al, 2017) (GSE90895), and Sall1, Sall4, and Tet1 binding data in mESCs were from Xiong et al (Xiong et al, 2016; Data ref: Xiong et al, 2016) (GSE57700). The generation of BAM, index, and peak calling files of published data was performed identical to 5hmC enrichment processing (see 5hmC processing). Overlap between transcription factor binding and 5hmC peaks was performed using ChIPpeakAnno using the “findOverlapsOfPeaks” tool with a minimum of 50 bp. The post‐processed reduced representation bisulfite sequence (RRBS) in MEFs and iPSCs was from Rais et al (Rais et al, 2013; Data ref: Rais et al, 2013) obtained from GEO (GSE49767). The post‐processed Tet‐assisted reduced representation bisulfite sequence (TARRBS) in MEF and MEFs overexpression OSKM for 5 days was from Hu et al (Hu et al, 2014; Data ref: Hu et al, 2014) obtained from GEO (GSE52741). To identify differentially methylated regions (DMR) between MEFs and iPSCs, we took cytosine that had at least 3 methylation counts in MEFs and 0 methylation counts in iPSCs from post‐processed data.

Statistical analysis

Significance was calculated using the two‐tailed paired t‐test for Figs 1, 2, 4, 5, and 7 using GraphPad Prism Software and Wilcoxon signed rank test for Appendix Figs 3E, 4B and C using R command “wilcox.test” with parameters “paired=TRUE”.

Author contributions

KAT conceived ideas, conducted experiments, performed bioinformatic analysis, prepared figures, and contributed to writing the paper. CMD performed experiments, contributed to preparing figures, and wrote the paper. RS conceived the project, procured funding, interpreted data, and wrote the paper.

Conflict of interest

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Dataset EV1

Dataset EV2

Review Process File

The EMBO Journal (2019) 38: e101681

Data availability

The accession number for the gene expression data and 5hmC enrichment reported in this paper is GEO: GSE119077 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE119077).