Gene Regulatory Networks Mediating Canonical Wnt Signal-Directed Control of Pluripotency and Differentiation in Embryo Stem Cells


  • Xiaoxiao Zhang,

    1. Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge, Massachusetts, USA
    2. Department of Molecular and Cellular Biology, Harvard University, Cambridge, Massachusetts, USA
    Search for more papers by this author
  • Kevin A. Peterson,

    1. Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge, Massachusetts, USA
    Search for more papers by this author
  • X. Shirley Liu,

    1. Department of Biostatistics and Computational Biology, Dana-Farber Cancer Institute, Harvard School of Public Health, Boston, Massachusetts, USA
    Search for more papers by this author
  • Andrew P. McMahon,

    Corresponding author
    1. Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge, Massachusetts, USA
    2. Department of Molecular and Cellular Biology, Harvard University, Cambridge, Massachusetts, USA
    3. Harvard Stem Cell Institute, Harvard University, Cambridge, Massachusetts, USA
    • Department of Stem Cell Biology and Regenerative Medicine, Broad-CIRM Center, W.M. Keck School of Medicine, University of Southern California, California 90089, USA. E-mail: Telephone: +1-323-442-3056; Fax: +1-323-442-8024

    Search for more papers by this author
  • Shinsuke Ohba

    1. Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge, Massachusetts, USA
    2. Division of Clinical Biotechnology, Center for Disease Biology and Integrative Medicine, The University of Tokyo Graduate School of Medicine, Tokyo, Japan
    Search for more papers by this author

  • Author contributions: X.Z. and S.O.: conception and design, collection and/or assembly of data, data analysis and interpretation, and manuscript writing; K.A.P.: collection and/or assembly of data and data analysis and interpretation; X.S.L.: data analysis and interpretation; A.P.M.: conception and design, data analysis and interpretation, financial support, manuscript writing, and final approval of manuscript. X.Z. and S.O. contributed equally to this article.


Canonical Wnt signaling supports the pluripotency of embryonic stem cells (ESCs) but also promotes differentiation of early mammalian cell lineages. To explain these paradoxical observations, we explored the gene regulatory networks at play. Canonical Wnt signaling is intertwined with the pluripotency network comprising Nanog, Oct4, and Sox2 in mouse ESCs. In defined media supporting the derivation and propagation of ESCs, Tcf3 and β-catenin interact with Oct4; Tcf3 binds to Sox motif within Oct-Sox composite motifs that are also bound by Oct4-Sox2 complexes. Furthermore, canonical Wnt signaling upregulates the activity of the Pou5f1 distal enhancer via the Sox motif in ESCs. When viewed in the context of published studies on Tcf3 and β-catenin mutants, our findings suggest Tcf3 counters pluripotency by competition with Sox2 at these sites, and Tcf3 inhibition is blocked by β-catenin entry into this complex. Wnt pathway stimulation also triggers β-catenin association at regulatory elements with classic Lef/Tcf motifs associated with differentiation programs. The failure to activate these targets in the presence of a mitogen-activated protein kinase kinase (MEK)/extracellular signal-regulated kinase (ERK) inhibitor essential for ESC culture suggests MEK/ERK signaling and canonical Wnt signaling combine to promote ESC differentiation. Stem Cells 2013;31:2667–2679


A central question in all stem cell-based systems is how the balance of stem cell maintenance and commitment is regulated. Embryonic stem cells (ESCs) derived directly from the early mammalian embryo provide a particularly attractive model given their capacity for long-term propagation as stem cells under defined culture conditions and their potential to generate all cell types of the adult organism [1]. The pluripotency of ESCs is dependent on a set of core transcriptional regulators, including Pou5f1/Oct4, Sox2, and Nanog (NOS) [[2, 3]]. The coexpression of Oct4, Sox2, and Klf4, a member of a family of transcriptional regulators with redundant roles in ESC maintenance [4], is sufficient for a broad range of differentiated cell types to acquire a pluripotent state that closely resembles that of ESCs [[5, 6]]. Direct analysis of the targets of these transcriptional regulators has demonstrated that core pluripotency factors co-occupy cis-regulatory elements near ESC-specific genes, providing strong evidence for coregulatory inputs into the pluripotency gene regulatory network as well as mutual reinforcement of each factor's own expression [[2, 7, 8]].

Several secreted factors are pivotal to maintaining ESC properties; their addition to culture medium replaces the requirement for serum and feeder cell support in the maintenance and propagation of ESCs [[9-11]]. In particular, leukemia inhibitor factor (LIF) acts through Stat3 to maintain the pluripotency of mouse ESCs (mESCs), whereas bone morphogenetic protein (BMP) activity-directed activation of inhibitor of DNA binding (Id) regulatory factors replaces serum requirements [[9, 12]].

Recent studies have identified two small molecule pathway modulators, PD0325901 (PD03) and CHIR99021 (CHIR), which substitute for LIF and BMP in defined ESC medium to enable the isolation and propagation of mouse ESCs [13], and for the first time ESCs from the rat [14]. PD03 is an inhibitor of mitogen-activated protein kinase kinase (MEK) [15]; MEK action lies downstream of several receptor tyrosine kinase-mediated signaling pathways [16] including the fibroblast growth factor (FGF) pathway. FGF signaling is critical in establishing and maintaining trophectodermal (TE) precursors, the first differentiated cell lineage to be established by the totipotent mammalian embryo [17]. CHIR inhibits glycogen synthase kinase-3 (GSK3); as GSK3β-directed phosphorylation and degradation of β-catenin suppresses canonical Wnt signaling, CHIR is a potent agonist of the Wnt signaling pathway [[15, 18]]. In canonical Wnt signaling, the accumulation of cytoplasmic β-catenin enables its nuclear entry and complexing with members of the Lef/Tcf family of transcriptional regulators [21]. In the absence of β-catenin, Lef/Tcf factors bind DNA directly at a consensus Lef/Tcf site and recruit transducin-like enhancer of split proteins to silence target gene activity. In contrast, their dimerization with β-catenin generates transcriptional activating complexes that bind to cis-regulatory modules activating target genes [22].

Analysis of ESC culture and embryonic development provides conflicting views of the role of β-catenin-dependent, canonical Wnt signaling on ESC cultures. Addition of recombinant Wnt3a, a Wnt ligand activating canonical Wnt signaling, together with LIF is reported to support ESC pluripotency in the absence of other factors [[23-25]]. Furthermore, CHIR-mediated stimulation of canonical Wnt signaling in the presence of PD03 blocks an intrinsic tendency of mouse ESCs to differentiate, enabling continued replication of ESCs in a pluripotent state [13]. BIO, another GSK-3 inhibitor, has been reported to maintain ESCs via upregulation of LIF [26] and enhance somatic cell reprogramming via cell-fusion through the accumulation of β-catenin [27]. Wnt signaling also promotes reprogramming to induced pluripotent cells (iPSCs), substituting for c-Myc in the efficient propagation of iPSCs derived from mouse embryonic fibroblasts infected with Sox2, Oct4, and Klf4 [28]. The downregulation of “stemness marker genes” in ESCs lacking functional β-catenin supports a role of canonical Wnt signaling in maintenance of pluripotency [29], although a study of an independently generated β-catenin-deficient mES cell line reached a different conclusion [30].

At the DNA level, genome-wide interaction studies of canonical Wnt signaling effectors have largely focused around transcription factor 7-like 1 (Tcf7l1, commonly known as Tcf3), a transcriptional component that is thought to predominantly repress Wnt-target genes. Tcf3 binding shows a strong intersection at sites cobound by major pluripotency regulators [[7, 31]]. Recent reports indicate that a loss of Tcf3 can substitute for CHIR in 2i, which is consistent with an inhibitory action of this member on the pluripotency program [32]. CHIR-mediated stimulation of β-catenin activity is proposed to both abrogate Tcf3 repression on the pluripotency network through a transactivation independent mechanism and to promote pluripotency through an interaction with Oct4 [[32-34]]. Together these data provide evidence for a canonical Wnt/β-catenin pathway action in promoting the pluripotent state of stem cells.

Conversely, canonical Wnt signaling has also been shown to induce specification of TE and mesendoderm lineages [[35-37]], and mouse embryos lacking β-catenin, Wnt3, or two Wnt coreceptors, Lrp5 and Lrp6, arrest prior to gastrulation linking canonical Wnt signaling to axial specification and mesendodermal induction [[38-43]]. The conflicting reports present a mechanistic paradox: how does β-catenin-dependent canonical Wnt signaling promotes both the stem cell state and the early commitment of pluripotent cells to specific cell lineages of the gastrulating embryo?

To address this question, we engineered a mESC line to produce a Biotin-FLAG epitope tagged form of β-catenin from the β-catenin (Ctnnb1) locus and performed genome-wide chromatin immunoprecipitation (ChIP) in conjunction with high-throughput sequencing (ChIP-seq) to directly address β-catenin target sites on canonical Wnt signaling activation in mouse ESCs. When these data are viewed in conjunction with extensive expression profiling of ESCs under pluripotency and differentiation promoting conditions, together with DNA binding studies of key pluripotency determinants and their complex formation with β-catenin, a mechanistic model emerges that can reconcile the opposing actions of canonical Wnt signaling discussed above.


For full details see supporting information Methods.

Generation of ESC Lines

Ctnnb1-BioFLneo and Ctnnb1-BioFLneo;BirA knock-in ESC lines generation are described in supporting information Methods.

ESC Culture

ESCs were cultured in serum+LIF complete media (CM) and 2i according to standard procedures (supporting information Methods).

Immunofluorescence, Immunoblot, and Coimmunoprecipitation Assays

Immunofluorescence and immunoblot were performed according to standard techniques described in supporting information Methods. Nuclear extracts (NE) were analyzed according to the manufacturer's instructions for Nuclear Complex Co-IP Kit (Active Motif, Carlsbad, CA, Coimmunoprecipitation experiments were performed according to the manufacturer's instructions for Protein A/G HP SpinTrap Buffer Kit (GE Healthcare, Little Chalfont UK,

ChIP, ChIP-Seq, and ChIP-qPCR

ChIP was performed according to Vokes et al. (2007) (supporting information Methods). The construction of ChIP-seq libraries was performed with ChIP-seq DNA Sample Prep Kit (Illumina, San Diego CA, according to manufacturer's instruction and sequenced on Genome Analyzer II (Illumina) machine. qPCR was performed with Biorad iQ SYBR Green Supermix (Bio-Rad, Hercules, CA, Fold enrichment was calculated by normalizing ChIP sample against input, and target region against control region as follows.

ΔCt = Ct (ChIP) − Ct (input); ΔΔCt = ΔCt (target region) − ΔCt (control region);

math formula.

Expression Profiling

Microarrays were conducted in triplicate using Affymetrix GeneChip mouse Gene 1.0 ST array (Affymetrix, Santa Clara, CA, according to the manufacturer's instructions, and data are normalized by robust multi-array average (RMA) in R (supporting information Methods).

Electrophoresis Mobility Shift Assay

NE were isolated from 293T cells overexpressing Pou5f1, Sox2, or Tcf7l1 as described [44]. Diogoxigenin-labeled probes were constructed using the DIG Gel Shift Kit (Roche, Basel, Switzerland, Binding reaction and subsequent electrophoresis mobility shift assay (EMSA) were performed according to the manufacture's instruction.

Bioinformatic Data Analysis

See supporting information Methods for complete description of data analysis.

Accession Numbers

Sequencing and microarray data have been deposited to GEO with accession number GSE43597.


Genome-Wide Profiling of the Canonical Wnt Regulatory Network in mESCs

To take advantage of in vivo biotinylation and FLAG-tag technologies in analyzing canonical Wnt signaling in mESCs, we generated a Ctnnb1-Biotin-3xFLAG knock-in ESC line (Ctnnb1-BioFLneo ESC) using gene-targeting strategies [[45, 46]]. The modified allele places a carboxyl-terminal epitope tag on β-catenin comprising three tandem copies of a FLAG (3× FLAG) epitope [47] and a short peptide that serves as a substrate for in vivo biotinylation in cells expressing the Escherichia coli Biotin ligase, BirA [[46, 48, 49]] (supporting information Fig. S1A). Correct targeting of the modified Ctnnb1 knocked-in allele was confirmed by long-range PCR (supporting information Fig. S1B). Ctnnb1-BioFLneo ESCs were then engineered to stably express BirA to ensure biotinylation of β-catenin-BioFL proteins (Ctnnb1-BioFLneo; BirA ESC).

We confirmed the integrity, specificity, and activity of the allele through the following observations. First, production and localization of β-catenin-BioFL protein were comparable to that of the wild-type protein (supporting information Fig. S1C, S1D). Second, biotinylated β-catenin-BioFL proteins were detected using streptavidin-conjugated reagents in Ctnnb1-BioFLneo;BirA ESCs, but not in Ctnnb1-BioFLneo ESCs (supporting information Fig. S1D). Finally, biotinylated β-catenin-BioFL appeared to function normally; mice homozygous for Ctnnb1-BioFL alleles carrying BirA ligase are viable with no apparent abnormalities (S.O. and A.P.M., manuscript in preparation). As a control cell line for subsequent analyses, we also generated a BirA-expressing ESC line (BirA ESC; supporting information Fig. S1E).

To better understand roles of canonical Wnt signaling in ESC biology, we set out to identify genomic targets of β-catenin, applying ChIP-seq to Ctnnb1-BioFLneo;BirA ESCs. Ctnnb1-BioFLneo;BirA ESCs were cultured on feeder cells in standard condition with serum+LIF CM, and treated with CHIR (CM+CHIR) for 16 hours, then subjected to a series of ChIP-seq procedures. β-Catenin-DNA complexes were pulled down using anti-FLAG antibody (FLAG-ChIP) or streptavidin (Biotin-ChIP) in parallel. DNA obtained from each ChIP procedure was independently sequenced. We also repeated FLAG-ChIP on Ctnnb1-BioFLneo;BirA ESCs without CHIR treatment.

We obtained 15,947 and 16,069 binding regions for Biotin-ChIP and FLAG-ChIP replicates, respectively, from ChIP-seq of CHIR-treated Ctnnb1-BioFLneo;BirA ESCs (supporting information Fig. S2A--S2C; see supporting information Methods for peak calling program and criteria). In contrast, only a small number of regions were bound by β-catenin in CM without CHIR (data not shown), suggesting only background levels of endogenous canonical Wnt signaling in CM-supplemented feeder-supported cultures. The intersection of the two CHIR-dependent datasets identified 9,885 regions (62.0% in Biotin-ChIP and 61.5% in FLAG-ChIP) in common (Pearson correlation = 0.89, supporting information Fig. S2C). Shared peaks have a higher intensity and peak ranking than ChIP regions unique to a single dataset suggesting that the intersection represents the most robust set of bone-fide interaction sites (supporting information Fig. S2D). This independently validated intersection formed the foundation for subsequent analysis (Fig. 1A) and representative peaks associated with pluripotency sustaining transcriptional components were validated by qPCR (supporting information Fig. S2E). All binding sites were annotated relative to ref-seq gene predictions (Fig. 1B and supporting information Table S1). When compared across the genome, β-catenin-associated regions show enrichment within 10 kb of the transcriptional start site (TSS), and a relative depletion in intronic and exonic regions (Fig. 1B). Approximately 16% of all annotated genes are associated with β-catenin binding in CHIR-treated mESCs using a “gene" definition as the region 10 kb upstream of the TSS plus the gene body.

Figure 1.

Genome-wide mapping of β-catenin binding regions in mouse embryonic stem cells cultured in complete media. (A): Venn diagram showing the overlap between β-catenin Biotin ChIP-seq and FLAG ChIP-seq peaks. (B): Genome-wide distribution of β-catenin binding regions relative to mouse genes compared with random control region genomic distribution. Binding regions were annotated as exon, introns, 5′ untranslated region (5′ UTR), 3′ UTR, within 0–1 kb upstream of TSS (TSSup1k), within 1–10 kb upstream of TSS (TSSup10k), within 0–1 kb downstream of TES (TESdown1k), within 1–10 kb downstream of TES (TESdown10k), or >10 kb away from the nearest genes (intergenic). (C, D): Top enriched motifs recovered from de novo motif analysis of β-catenin binding regions. Left panels show motif logos. HMG box motif is highlighted in light blue, and POU family motif in light red. Right panels show histogram of motifs ± 300 bp around peak summit of β-catenin (orange) or matched control peak (blue). (E, F): Gene ontology terms enriched for β-catenin peaks containing Lef/Tcf motif (E) or Oct-Sox motif (F) using GREAT. The −log10 of the raw binomial p-value is reported. Abbreviation: ChIP, chromatin immunoprecipitation.

Motif analysis was performed on 400 base-pair regions centered on the peak summit of β-catenin association to identify statistically enriched DNA motifs within the dataset (supporting information Fig. S2F). As expected, the DNA target site for Lef/Tcf factors, the DNA-binding partner for β-catenin, was highly enriched in the dataset: 35.0% of all β-catenin peaks predicted a Lef/Tcf site, versus 11.1% in matched control regions (two-proportion z test, p-value <1e-350). Strikingly, an Oct-Sox composite motif was also highly enriched (26.1% of all β-catenin peaks, p-value <1e-324) and like Lef/Tcf predictions, centered on the predicted peak of β-catenin binding (Fig. 1C, 1D). We also identified motifs that matched binding sites for Klf4, Zic, Esrrb, E2a, and AP-2 (supporting information Fig. S2F). The distribution of these suggests enrichment in the region but a less direct association with β-catenin binding (data not shown). Importantly, the enrichment of Lef/Tcf sites provides strong support for the quality of the dataset, while the enrichment of Oct-Sox motifs near β-catenin peak summits suggests an interplay between β-catenin and the pluripotency circuit on canonical Wnt signaling stimulation. Interestingly, gene ontology analysis using Genomic Regions Enrichment of Annotations Tool [50] revealed that while embryogenesis-related and Wnt receptor signaling pathway-related genes were both enriched in Lef/Tcf motif-containing and Oct-Sox motif-containing β-catenin peaks, the former category was also enriched in mesoderm development-related term, and the latter stem cell- and neural-related terms (Fig. 1E, 1F).

Analysis of β-Catenin, Tcf3, Sox2, Oct4, and Nanog Interactions at Target Genes Points to Distinct Enhancer Modules Mediating the Actions of Canonical Wnt Signaling

The recovery of Oct-Sox motif within β-catenin binding regions prompted us to compare the β-catenin data with previously published ChIP-seq data for 19 transcription factors (TFs) associated with maintenance of pluripotency, induction of iPS cells, and Wnt action (supporting information Fig. S3A): the core pluripotency factors Nanog/Oct4/Sox2 (NOS, refer supporting information Fig. S3B, S3C for the comparison of two independent datasets) [[7, 8]]; Smad1/Stat3, effectors of key mESCs signaling pathways [8]; Tcfcp2l1/Tbx3/Klf4/C-myc/N-myc/Zfx, reprogramming factors important for self-renewal [[8, 51]]; Ring1b/Ezh2/Suz12, components of polycomb repressive complexes [[8, 52]]; Esrrb/Nr5a2, nuclear receptors linked to the ESC state [[8, 53]]; and Tcf3, the most abundant of the Lef/Tcf family of canonical Wnt transcriptional effectors in mESCs [7].

Through pair-wise cobinding analyses, we were able to classify binding patterns of these regulatory factors into several clusters; notably Tcf3, Nanog, Sox2, Smad1, and Oct4 interactions most closely resembled those observed through β-catenin ChIP-seq (Fig. 2A). Given that β-catenin regulates gene expression through Tcf transcription factors, of which Tcf3 is most abundantly expressed in the mESCs, we first did two-way intersection of β-catenin and Tcf3 binding peaks taking only the β-catenin::Tcf3 regions to increase the credibility of binding events. A further intersection with NOS peaks, produced Group-A (β-catenin::Tcf3) and Group-B (β-catenin::Tcf3::NOS) (Fig. 2B): comparison of these two categories provides an insight into whether canonical Wnt signaling action differs in the presence of NOS. We explored motif enrichment, chromatin state, and functional properties of predicted target genes adjacent to these binding regions. A clear consensus Lef/Tcf motif was the most over-represented motif in the Group-A (p-value <1e-336, two-proportion z test) (Fig. 2C), while the most enriched motif in Group-B closely resembled the published Oct-Sox motif (p-value <1e-561, two-proportion z test) (Fig. 2D). Stem cell- and ectoderm-related terms were enriched in Group-B targets, while axis specification and mesoderm terms were over-represented in Group-A targets (Fig. 2E, 2F and supporting information Table S2). In terms of chromatin state, both groups displayed a strong H3K4me2 signature, an indicator of poised or active enhancer regions [55], but the signature was more prominent among Group-B regions, suggestive of a more active state in ESCs (Fig. 2G). Consistent with this view, Group-B displayed a stronger H3K4me1 and H3K27ac active enhancer signature than Group-A (Fig. 2G) [[56-58]]. We also intersected Group-A and Group-B-associated genes with Group-1 and Group-2 targets as defined by Yi et al. [33]. Group-1 consists of genes where the response to Wnt3a addition to wild-type ESCs is similar to that observed on Tcf3 ablation (Tcf3 KO): that is, genes predicted to be regulated by Wnt3a antagonism of Tcf3 repression of target activity. Group-2 consists of genes that show a similar response to Wnt3a independent of Tcf3 activity: that is, Wnt3-dependent and Tcf3-independent targets. We observed a strong enrichment (p-value <1e-14.4, two-proportion z test) of Group-A and Group-B-regulated genes with the genes showing a transcriptional response to Wnt3a in a Tcf3-dependent or Tcf3-independent manner lending additional evidence to support the conclusion that β-catenin ChIP identifies likely regulatory regions mediating transcriptional regulation by Wnt3a signaling and Tcf3 binding. However, we did not observe a clear pair-wise segregation between Group-A/B genes and Group-1/2 genes.

Figure 2.

Characterization of β-catenin and embryonic stem cell (ESC) pluripotency factors binding. (A): Heat map depicting the correlation of β-catenin and ESC factors bindings. Red: positive correlation; blue: negative correlation. (B): Venn diagram of β-catenin, Tcf3, and intersection of Nanog/Oct4/Sox2 peak regions. Two groups of peaks are highlighted: Group-A: β-catenin::Tcf3, and Group-B: β-catenin::Tcf3::NOS. (C, D): Enriched motifs in Group-A and Group-B. Red: motif occurrence in β-catenin peaks; gray: motif occurrence in matched control regions with the same coverage. p-Value was calculated according to two-proportion z test. (E, F): Functional annotation of Group-A and Group-B regions using GREAT. The −log10 of the raw binomial p-value is shown. (G): Aggregation plots of H3K4me1, H3K4me2, and H3K27ac signals ±3 kb around the peak summit for binding regions in Group-A (red) and Group-B (blue) as well as corresponding matched control regions with SE bars (black and gray). The analysis is done using HOMER [54]. Bin size 100 bp. Abbreviation: ChIP, chromatin immunoprecipitation.

In summary, the data suggest that two groups of β-catenin binding regions are involved in transcription of two distinct categories of target genes through two distinct mechanisms, one through a Lef/Tcf-mediated DNA interaction of β-catenin with poised enhancers around differentiation-related genes (Group-A) and one through cooperative interactions of NOS and β-catenin/Lef/Tcf with active enhancers around stem cell- and ectoderm-related genes (Group-B). A CisGenome browser screenshot of representative genes of Group-A (Cdx2, Fig. 3A; Axin2, supporting information Fig. S4) and Group-B (Nanog, Fig. 3B) shows the strong relative signal intensity of β-catenin over Nanog/Oct4/Sox2 for Group-A versus Group-B-associated genes.

Figure 3.

CisGenome browser screenshots showing combinatorial binding pattern of β-catenin and core pluripotency factors in complete media. β-Catenin binding to known Wnt-target genes related to differentiation (Cdx2, A), and pluripotency (Nanog, B). Endogenous association of β-catenin is also displayed in the absence of CHIR stimulation (CHIR−). Tcf3, Nanog, Oct4, Sox2, and histone modification profiles displayed here are from published datasets (see Results). A zoom in on highlighted regions with motif annotation is displayed to the right side.

Activation of Canonical Wnt Signaling Directs Early Mesoderm Differentiation

To connect DNA association profiles of these factors with canonical Wnt signaling-mediated gene expression, we intersected β-catenin ChIP peaks with neighboring genes, with a focus on those genes that displayed differential expression between mESCs cultured in CM+CHIR and CM+XAV939 (XAV) (supporting information Table S3). XAV is a tankyrase inhibitor antagonizing Wnt signaling [59]. Among the intersected gene set, canonical Wnt signaling target genes, early mesoderm-, and TE-related genes ranked top on the most upregulated genes. However, pluripotency-related genes showed no strong differential expression (supporting information Table S3).

The integrity of our microarray dataset was supported by two analyses, prediction of canonical Wnt signaling target genes, and correlation with published expression datasets. First, we applied an approach based upon an empirical finding that the potential of a gene being a direct target of a given TF decreases monotonically as a function of the distance of the binding site to that gene's TSS [60]. Second, using a rank product-based ranking method, we made a probabilistic prediction of β-catenin target gene list (Fig. 4A and supporting information Table S4) [62]. Using a false discovery rate of 10%, we obtained 376 and 362 putative direct target genes that were positively and negatively regulated by CHIR, respectively, in CM. This method accurately predicted known target genes of canonical Wnt signaling, such as Axin2, T, Sp5, Lef1, Cdx2, and Tcfcp2l1. Interestingly, the identification of Porcn, which encodes a key factor in the palmitoylation and secretion of Wnt ligands [63], suggests a hitherto unrecognized positive feedback loop in canonical Wnt signaling.

Figure 4.

Integration of β-catenin chromatin immunoprecipitation (ChIP)-seq and expression profiling in mESCs treated with an activator or inhibitor of canonical Wnt signaling. (A): Scatter plot of β-catenin direct target gene prediction based on distance weighted regulatory potential score from ChIP-seq and t value of differential CHIR/XAV expression in CM. Red dots: upregulated genes with false discovery rate (FDR) <0.10; blue dots: downregulated genes with FDR <0.10. The darker red/blue represents the higher likelihood for a gene being β-catenin direct target. The horizontal and vertical histograms reflect the distribution of the index for distance weighted regulatory potential and differential expression t value, respectively. Representative genes are labeled. (B): Correlation of top 1,000 genes of high MEC/ESC or NEC/ESC expression ratio (from microarray data in Shen et al. [61]) (x-axis) with their differential expression fold changes in CM+CHIR/CM+XAV (y-axis). Genes were ranked by their expression ratio in MEC versus ESC or NEC versus ESC from high to low. For the top 1,000 genes in the two ranks, their expression ratio in CM+CHIR versus CM+XAV was checked. Bins represent the top 20 genes, then the top 40 genes, and so forth, as determined by the MEC/ESC ratio or NEC/ESC ratio. The correlation of MEC/ESC or NEC/ESC ratio with CM+CHIR/CM+XAV ratio for each bin was calculated and plotted. Abbreviations: CM, complete media; ESC, embryonic stem cell; MEC, mesendoderm cell; NEC, neural ectoderm cell.

We calculated the correlation of the upregulated genes in our data with the published microarray data comparing gene expression between ESCs, mesendoderm cells (MECs), and neural ectoderm cells (NECs) [61]. The top 100 genes displaying a high MEC/ESC expression ratio showed some correlation (> 0.5) with genes exhibiting a high CHIR/XAV expression ratio. No correlation was observed with genes associated with neural ectoderm development (high NEC/ESC) (Fig. 4B), consistent with the known role of canonical Wnt signaling in inducing mesendoderm and suppressing neural ectoderm development [[43, 64, 65]].

Similarity of β-Catenin Chromatin Binding Between CM+CHIR and 2i

To examine the interactions of β-catenin and pluripotency network components under the 2i conditions, Ctnnb1-BioFLneo;BirA ESCs were cultured under 2i supplemented with LIF to enhance colonogenicity [13]. ChIP-qPCR for Sox2, Oct4, and Tcf3 binding were conducted at cis-elements near pluripotency genes in 2i, 2i+LIF, and CM conditions. The bindings of the three factors were comparable in all three conditions (Fig. 5A, p-value >.05 from two sample t test).

Figure 5.

Roles of small molecules CHIR and PD03 in 2i. (A): ChIP-qPCR for Oct4, Sox2, and Tcf3 interaction at defined regulatory regions surrounding pluripotency target genes in mouse embryonic stem cells (mESCs) cultured in 2i, 2i+LIF, and CM. Data represent the mean of biological replicates. (B): Experimental scheme for studying the role of CHIR and PD03 in CM and 2i-adapted mESCs. ChIP-qPCR and microarray analysis were performed in 2i-adapted mESCs cultured for 24 hours with DMSO (control), PD03, CHIR, 2i, and 2i+LIF. Cells at passage 20 under the 2i+LIF condition were subjected to each assay. Cells that had been maintained in CM on feeder cells were cultured for 24 hours in CM with CHIR or XAV prior to microarray analysis. (C): ChIP-qPCR for β-catenin at selected loci near pluripotency-related genes (upper), differentiation-related and Wnt-target genes (lower) on 2i-adapted mESCs. ChIP using anti-FLAG antibodies was performed according to the experimental scheme described in (B). Data show the mean and SEM for three biological replicates. (D): K-means clustering was used to classify genes with expression fold change >2 in at least one comparison group of 2i/PD03, 2i/CHIR, 2i/DMSO, 2iLIF/2i, 2iLIF/DMSO, and CM+CHIR/CM+XAV. A total of 388 genes were clustered into six clusters. (E): Individual component maps are shown for each pair-wise comparison. Top left: 2i/PD03; middle left: 2i/CHIR; bottom left: 2i/DMSO; top right: 2iLIF/2i; middle right: 2iLIF/DMSO; bottom right: CM+CHIR/CM+XAV. In general, red indicates upregulation and blue downregulation. The number by each color bar is the actual number of fold change. (F): Five base pair core Ets motif logo TCCTW from TRANSFAC motif M00339. (G): Enrichment of Ets core motif ± 500 bp around β-catenin peak summit (orange) compared with matched control regions (blue). Abbreviations: CM, complete media; DMSO, dimethyl sulfoxide; LIF, leukemia inhibitor factor; TSS, transcriptional start site.

To understand which gene category was regulated by each 2i component, and how β-catenin may participate in this regulation, we performed ChIP-qPCR for β-catenin in 2i-adapted Ctnnb1-BioFLneo;BirA ESCs cultured with dimethyl sulfoxide (DMSO), PD03, CHIR, 2i, or 2i+LIF for 24 hours (Fig. 5B). A strong enrichment of β-catenin was observed in CHIR-, 2i-, and 2i+LIF-treated cells at the same pluripotency-related gene regions as those bound by β-catenin in CM+CHIR (Fig. 5C, upper panel); the binding was dependent on the activation of canonical Wnt signaling, since it was lost within 24 hours of the removal of CHIR (Fig. 5C; DMSO and PD03 in upper panel). Thus, stabilization of β-catenin leads to similar interactions at the DNA level in quite different culture regimens. Thus, β-catenin likely contributes to expression of pluripotency-related genes under 2i conditions by the direct association with cis-regulatory modules governing expression of the pluripotency network. Interestingly, the MEK inhibitor PD03 did not affect β-catenin association with differentiation-related gene regions (Fig. 5C, lower panel).

Differentiation Genes Fail to be Upregulated by CHIR in 2i As in CM

Predominant activation of a mesendoderm lineage differentiation program by CHIR in CM (Fig. 4B) and the association of β-catenin with differentiation-related gene regions in 2i (Fig. 5C) raised a question about how differentiation is inhibited in 2i. We performed microarray analysis in 2i-adapted Ctnnb1-BioFLneo;BirA ESCs cultured with DMSO, PD03, CHIR, 2i, and 2i+LIF according to the same experimental scheme as in Figure 5B (supporting information Table S5). In this, the effect of CHIR is distinguished by comparing the relative gene expression level in the ESCs cultured in 2i to those cultured in PD03 where CHIR is absent (2i/PD03). Comparing CM+CHIR over CM+XAV (CM+CHIR/CM+XAV) provides another metric of CHIR activity, while a comparison of 2i+LIF over 2i (2iLIF/2i) sheds light on any direct effect of LIF on gene expression. Using Kohonen self-organizing maps (SOM) [66], we identified groups of genes that shared similar patterns of expression changes between such comparisons (Fig. 5D) and visualized target gene expression change in a series of heat map (Fig. 5E). Genes that showed a greater than twofold differential expression in at least one comparison were clustered (Fig. 5D): the six cluster associated gene names are presented in supporting information Figure S5. In Figure 5D, each hexagonal map unit represents a SOM node, which has an underlying vector of a pair-wise fold change under different conditions. In Figure 5E, the gradient of each pair-wise fold change is displayed separately while maintaining the same topological structure as in Figure 5D. On comparing Figure 5D, 5E, clusters-1,-3, and -4 represented genes strongly upregulated by CHIR (2i/PD03), LIF (2i LIF/2i), and PD03 (2i/CHIR), respectively, in 2i. Cluster-2 represented genes moderately upregulated by CHIR (2i/PD03), LIF (2i LIF/2i), and PD03 (2i/CHIR). Genes in cluster-6 were upregulated by CHIR in serum conditions (CM+CHIR/CM+XAV), and those in cluster-5 were downregulated in all comparison groups.

There are three key insights from this analysis. First, CHIR, PD03, and LIF upregulated different sets of genes in 2i, as shown in cluster-1, cluster-3, and cluster-4, respectively. Second, CHIR upregulated different sets of genes in 2i from those in CM (cluster-1 vs. cluster-6). Several differentiation-related genes, such as T, Cdx2, and Cdx1 appear in cluster-6 that were downregulated in 2i compared to CHIR (see 2i/CHIR) (supporting information Fig. S5). We hypothesize that the lineage promoting action of canonical Wnt signaling as observed in CM may be suppressed through the deprivation of potential collaborating factors for β-catenin in 2i or through secondary modification of transcriptional complexes without the downregulation of β-catenin binding to cis-regulatory control regions for differentiation-associated targets (Fig. 5C). In support of the former, an Ets motif was significantly enriched in β-catenin binding regions—Ets factors provide the transcriptional output to MEK/ERK signaling (Fig. 5F, 5G; see Discussion). Third, the effect of 2i over DMSO (2i/DMSO, left bottom) appears to be an additive result of the effect of CHIR (2i/PD03, left top) and PD03 (2i/CHIR, left middle). Similarly, the effect of 2i+LIF over DMSO (2i LIF/DMSO, right middle) appears to be an additive result of CHIR (2i/PD03, left top), PD03 (2i/CHIR, left middle), and LIF (2i LIF/2i, right top).

β-Catenin Complexes with Oct4 and Tcf3 at Oct-Sox Motifs in 2i Cultured mESCs

Binding of Oct4 and Sox2 to pluripotency-related gene regions was unaltered under 2i condition in contrast to the effect of CHIR on cells grown in CM (see earlier). To investigate the physical interaction of β-catenin and Tcf3 with these two pluripotency determinants, we performed coimmunoprecipitation (Co-IP) analysis with NE from mESCs cultured in 2i+LIF and 2i with antibodies specific to β-catenin, Tcf3, Sox2, and Oct4. Oct4 and β-catenin were coimmunoprecipitated with anti-Oct4 antibodies (Fig. 6A); the failure of a reciprocal IP likely reflects epitope masking of the epitope recognized by the β-catenin antibody in an Oct4 complex [34]. Tcf3 was associated with β-catenin and Oct4, but not Sox2 (Fig. 6A). Consistently, Sox2 only pulled down Oct4, not β-catenin or Tcf3. These results suggest that two regulatory complexes exist under 2i conditions: one containing β-catenin, Tcf3, and Oct4 and another with Sox2 and Oct4.

Figure 6.

Physical association of β-catenin, Oct4, Sox2, and Tcf3 and in vitro binding properties of Oct4, Sox2, and Tcf3 to the Oct-Sox composite motif. (A): Coimmunoprecipitation analysis of β-catenin, Tcf3, Oct4, and Sox2 complexes in mouse embryonic stem cell (mESCs). Asterisk indicates heavy chains of antibody used in IP. (B): Sequence of oligonucleotide probes used in EMSA. Oct motif is underlined; Sox and Lef/Tcf motifs are bolded. Mutations are shown in lowercases. (C): Cooperative bindings of Oct4, Sox2, and Tcf3 to the Oct-Sox (OS) composite motif, determined by EMSA. The indicated combinations of NE isolated from 293T cells overexpressing Pou5f1 (Oct4), Sox2, or Tcf7l1 (Tcf3) were analyzed by EMSA using DIG-labeled OS probes. Bands A, B, C, and D denoted with arrows indicate Oct4-binary, Sox2-binary, Oct4-Sox2-ternary, and Oct4-Tcf3-ternary complexes with the Oct/Sox probe, respectively. Asterisks indicate nonspecific bands. Data here is extracted from supporting information Figure S7B which provides more extensive competitor experiments and antibody supershift experiments to identity each shifted band. (D): Schematic of luciferase reporter constructs used in (E) and (F). Pou5f1 distal enhancer region containing the Oct-Sox composite motif drives the luciferase gene with a minimal TATA-box promoter element under pGL4 vector backbone. Each mutation corresponds to mutant motifs in EMSA analysis. (E): Luciferase reporter assay using Pou5f1 distal enhancer region in NIH3T3 cells and 2i-cultured mESCs (v6.5). Forty-eight hours after transfection with reporter constructs, cells were subjected to the assay. mESCs were maintained under 2i+LIF condition for 11 passages prior to the assay. (F): Luciferase reporter assay in mESCs (v6.5) cultured under 2i condition (left) or CM (right). mESCs were maintained under 2i+LIF condition for 11 passages prior to the assay. Upon transfection with reporter constructs, cells were switched into basal media of 2i culture (mixture of neurobasal media, DMEM/F12, N2, and B27 supplements) with PD03 or 2i (left), or CM in the presence or absence of CHIR (right). The assay was performed 48 hours after transfection. Abbreviations: Ab, antibodies; α-O4, anti-Oct4 antibody; α-T3, anti-Tcf3 antibody; Con, extracts from mock-transfected cells; Comp, unlabeled competitors; DMSO, dimethyl sulfoxide; IB, immunoblotting; IP, immunoprecipitation; LIF, leukemia inhibitor factor; NE, nuclear extracts; O4, extracts from Oct4-overexpressing cells; RLU, relative light unit; S2, extracts from Sox2-overexpressing cells; S2S, smaller amount of extracts (0.2 μg) from Sox2-overexpressing cells; S2L, larger amount of extracts (2 μg) from Sox2-overexpressing cells. T3, extracts from Tcf3-overexpressing cells.

Given the overlap in ChIP peaks between Oct4, Sox2, and Tcf3, the Oct-Sox composite motif recovered in β-catenin ChIP-seq data, sequence similarity within this consensus motif between Sox and Tcf DNA binding sites, and their structural similarities sharing an HMG DNA binding domain, we tested whether Tcf3 directly bound to the Oct-Sox composite motif in vitro by EMSA. NE were prepared from 293T cells overexpressing Pou5f1 (Oct4), Sox2, or Tcf7l1 (Tcf3) (supporting information Fig. S6) and tested for their ability to bind a double-stranded DNA probe incorporating an Oct-Sox composite motif located within a distal Pou5f1 enhancer (Fig. 6B) The complete data for all EMSA are shown in supporting information Figure S7A, S2B and key lanes extracted from this dataset to Figure 6C.

Oct4 and Sox2 alone complex with the Oct/Sox probe (supporting information Fig. S7A, lanes 2–7), whereas no binding was observed for Tcf3 (supporting information Fig. S7A, lanes 8–10). In contrast, Tcf3 complexed with a Lef/Tcf binding motif (LT probe) under the same conditions (supporting information Fig. S7A, lanes 12–20; band E). Next, we examined cobinding for cooperative interactions. Oct4 and Sox2 cobinding led to additional band (C) not seen when Oct4 (A) or Sox2 (B) bound alone (Fig. 6C, lanes 2–4; supporting information Fig. S7B, lanes 2–8 [67]). As above, no Tcf3 interaction was observed with the Oct/Sox motif and Tcf3 failed to compete with Sox2 at the Sox2 binding site (supporting information Fig. S7B, lanes 9–13). However, in the presence of Oct4 and Tcf3, additional bands were observed that likely reflect ternary complexes of Oct4 and Tcf3, with Tcf3 bound at the Sox2 site (band D in Fig. 6C, lanes 5–13).

Several lines of evidence support this view. First, the additional band was eliminated by unlabeled LT probe or anti-Tcf3 antibodies (Fig. 6C, lanes 10 and 13), but not by unlabeled mutated LT probe (Fig. 6C, lane 11). Second, the additional band was competed with the unlabeled WT Oct/Sox probe, and by one in which the Oct motif was mutated, but not by a probe containing mutations in both Oct and Sox motifs (Fig. 6C, lanes 6, 7, and 9). Finally, when Oct4 binding was competed by unlabeled Sox-mutated probe, or blocked with anti-Oct4 antibodies, the additional band disappeared (Fig. 6C, lanes 8 and 12, respectively). Together, these results suggest that Tcf3 binds to the Sox site in the Oct/Sox composite motif in an Oct4-dependent manner, whereas Oct and Sox factors can independently associate with their target sites.

To clarify possible patterns of complex formation when all the three proteins were present, we incubated the Oct-Sox composite motif with Oct4, Sox2, and Tcf3 (Fig. 6C, lanes 14 and 15). When low amounts of Sox2 were present with the other two proteins, we observed band shifts indicative of Oct4-DNA, Sox2-DNA, and Oct4-Tcf3-DNA complexes (Fig. 6C, lanes 14; bands A, B, and D, respectively). With higher concentrations of Sox2, we observed the formation of an additional Oct4-Sox2-DNA complex (Fig. 6C, lane 15, band C). A competition between Sox2 and Tcf3 has been computationally predicted by Mason et al. [68]. Together these data suggest a mutually exclusive competitive interaction for Tcf3 and Sox2 at Oct-Sox motifs where the association of Tcf3 requires a cooperative interaction with Oct factors to overcome the less favored consensus of a Sox versus a Lef/Tcf binding motif.

Functional relevance of canonical Wnt signaling with the Oct-Sox motif was supported by in vitro luciferase reporter assay using the Pou5f1 distal enhancer region (Fig. 6D). The region belongs to Group-B in Figure 2B--2G, and sequence of the Oct/Sox probe used in the above EMSA was derived from the region. We confirmed that the region had enhancer activity in 2i-cultured mESCs (v6.5), but not in NIH3T3 cells, in a copy number-dependent manner (Fig. 6E). 2i-cultured mESCs showed upregulation of the enhancer activity compared to PD03-treated one, suggesting the positive effect of CHIR stimulus, that is, canonical Wnt input, on the enhancer activity; the upregulation was diminished upon mutation of the Sox motif to the same extent as mutation of both Oct and Sox motifs. In contrast, when the Oct motif was mutated, we still observed elevated enhancer activity when 2i-culture was compared to PD03-treated alone (Fig. 6F, left). The similar trend was recapitulated in mESCs cultured CM (Fig. 6F, right). Together, these data support the conjecture that canonical Wnt signaling contributes to the transcription of pluripotency genes via the Sox site within an Oct-Sox composite motif.


Our transcriptional analysis of Wnt pathway action in ESCs has generated several new insights into pluripotency and differentiation networks. First, ChIP-seq analysis enabled the identification of genomic targets of β-catenin activity in mESCs. Second, a strong association is observed between β-catenin bound regions and those occupied by core pluripotency factors (NOS) and Tcf3. Third, there are marked differences in bound regions and candidate target genes between regions where only β-catenin::Tcf3 overlap, and those where β-catenin::Tcf3 also intersect with the core pluripotency networks (NOS). This is observed in motif enrichment (Lef/Tcf motif vs. Oct-Sox motif) suggesting distinct binding modes, the function of associated genes (axis specification- and mesoderm-related genes vs. stem cell- and ectoderm-related genes) suggestive of different biological outcomes, and the activity status of likely enhancer regions in mESCs (high activity for β-catenin::Tcf3::NOS and low activity for β-catenin::Tcf3). Fourth, under standard culture condition, the activation of canonical Wnt signaling elevated expression of differentiation-related genes, while activity of pluripotency-related genes was maintained. Fifth, under 2i condition, β-catenin also engaged at likely enhancers for TE lineage- and axis specification-related genes but under 2i conditions, these targets are not activated. Inhibition of MEK/ERK signaling by PD03 is critical in blocking these differentiation pathways and the enrichment of Ets motifs within differentiation-related enhancers suggests a cooperative interplay of Ets and canonical Wnt complexes for gene activity. Sixth, β-catenin, Tcf3, and Oct4 interact under 2i conditions. Finally, canonical Wnt signaling upregulated transcription in vitro via Oct-Sox motifs that could be engaged directly by Oct4-Tcf3 or Oct4-Sox2. Considering all data, we propose that under the 2i condition, canonical Wnt signaling participates in the pluripotency network via Oct4/β-catenin/Tcf3 complex formation and although Sox2 is absent, engagement of β-catenin still favors activity of pluripotency-associated genes (Fig. 7, upper).

Figure 7.

Schematic model of β-catenin-dependent regulation of pluripotency network. Oct4-Sox2 binding to Oct-Sox composite motifs maintains activity of key regulators of pluripotency. Tcf3 interaction with Oct factors at the same motif is predicted to destabilize this circuit. CHIR-mediated stabilization of β-catenin has opposing actions. Entry of β-catenin into Oct4/Tcf3 complexes abrogates Tcf3 actions thereby promoting pluripotency. However, the production of active canonical Wnt transcriptional complexes engages differentiation targets destabilizing pluripotency. PD03-mediated inhibition of MEK/ERK signaling restores a pluripotency balance blocking the activation of Wnt-dependent differentiation genes enabling culture under 2i conditions. Given the role of mitogen-activated protein kinase kinase/ERK signaling downstream of receptor tyrosine kinases in the regulation of the Ets-family of transcriptional regulators, and the enrichment of Ets motifs in predicted cis-regulatory, we propose the combined action of Wnt and RTK signaling in the differentiation of embryonic stem cells. Abbreviation: LIF, leukemia inhibitor factor.

The analysis of canonical Wnt pathway mutants sheds additional light on this process. First, β-catenin activity is essential under 2i conditions for the derivation and maintenance of mESCs [32]. Second, our work, and that of others indicate that Tcf proteins are required for occupancy at Oct4-dependent promoter [[32-34]] and engagement of β-catenin can enhance Oct4 promoter activity [[32-34]]. Tcf3 is a key component in the transcriptional complex; Tcf3 actions are linked to groucho-mediated gene silencing [69]. Tcf3 inhibits pluripotency and the removal of Tcf3 can functionally substitute for the actions of β-catenin under 2i conditions [[70, 71]]. Thus, the main action of β-catenin appears to be to neutralize the destabilizing activity of Tcf3 in the pluripotency network. Interestingly, mutant forms of β-catenin lacking the transcriptional activating domain are effective in maintaining pluripotency [32] suggesting that β-catenin acts by abrogating Tcf3 silencing rather than forming a β-catenin-dependent activation complex. The conclusion that Tcf3 and β-catenin do not form an active transcriptional complex is supported by recent studies of Tcf3 mutant mouse embryos [72].

Wnt actions in maintaining a state of pluripotency have also been linked to the control of telomerase activity through direct regulation of Tert promoter activity in mESCs. β-Catenin binding was reported to be enriched around the Tert gene in ChIP analysis of ESCs and binding further enhanced by Wnt3a treatment, or expression of a stabilized form of β-catenin [73]. In contrast, we see no enrichment at the Tert locus in our whole genome analysis (data not shown).


ESCs are in an inherently unstable state wherein endogenous Tcf3 activity antagonizes the core pluripotency network through competition at Oct/Sox motifs for Sox2 binding. Surprisingly, analysis of Sox2, Nanog, and Oct4 binding suggests that Tcf3 is not a transcriptional target of this network (data not shown). Under standard culture conditions of serum and LIF, β-catenin plays no significant role in maintaining pluripotency: its activity is not essential [30, 32] and stimulation of the pathway favors differentiation through engagement at “classic” Lef/Tcf motifs around differentiation-associated target genes. Presumably, LIF actions are dominant and the levels of activation of target genes are not sufficient to trigger widespread differentiation. In contrast, in 2i medium Tcf3 actions are critical and β-catenin is essential to overcome Tcf3's inhibitory effects within the pluripotency network. What remains to be determined is how the inhibition of MEK/ERK signaling blocks the differentiation promoting arm of Lef/Tcf::β-catenin-directed gene regulation. Our data on motif recovery and β-catenin engagement suggest a cooperative role for MEK/ERK-directed Ets factors independent of Lef/Tcf::β-catenin binding to activate target differentiation promoting genes consistent with the critical actions of Fgf and Wnt signaling in promoting lineage commitment in early mammalian development (Fig. 7, lower).


We thank Drs. Philippe Soriano, Alan B. Cantor, and Taku Saito for provision of experimental materials (specified in supporting information Methods); Laurie Chen, Joe Vaughan, Jill McMahon, Christian Daly, Jennifer Couget, Qianzi Tang, and Genome Modification Facility, Harvard Stem Cell Institute for providing technical assistance. We are also grateful to Dr. Ung-il Chung for critical comments and helpful discussions. Work in A.P.M.'s laboratory was supported by a grant from the NIH (DK056246). X.S.L. was supported by a grant from the NIH (2R01HG4069). S.O. was supported by Grants-in-Aid for young scientist (#23689079) from the Japan Society for the Promotion of Science (JSPS), Daiwa Securities Health Foundation Grant, and Nakatomi Foundation Research Grant. K.A.P. and A.P.M. are currently affiliated with the Department of Stem Cell Biology and Regenerative Medicine, Broad-CIRM Center, W.M. Keck School of Medicine, University of Southern California, CA.


The authors indicate no potential conflicts of interest.