Inhibition of glycogen synthase kinase-3 alleviates Tcf3 repression of the pluripotency network and increases embryonic stem cell resistance to differentiation

Canonical Wnt signalling is a key regulator of stem cells in epithelial tissues (reviewed in⁷). This pathway has also been proposed to play a major role in self-renewal of pluripotent embryonic stem (ES) cells. Wnt ligands promote nuclear accumulation of β-catenin, which associates with DNA-bound Tcf/Lef factors and activates transcription^8-10. Glycogen synthase kinase-3 (Gsk3)^{1, 11} negatively regulates Wnt signalling by phosphorylating β-catenin leading to its ubiquitination and proteolysis^{12, 13}. This is prevented by inhibitors of Gsk3 which thereby act as mimetics of Wnt stimulation. Gsk3 inhibitors such as BIO or CHIRON99021 (CH) support short term expansion of mouse ES cells^{1, 2} and this has been interpreted as evidence for canonical Wnt function in self-renewal^{1, 6, 14}. Differentiation is only partially suppressed, however, and cultures collapse upon passaging². Robust and long-term self-renewal additionally requires the cytokine leukaemia inhibitory factor (LIF), which activates the transcription factor Stat3^{15, 16}, or inhibition of the mitogen activated protein kinase (Mapk) cascade¹⁷.

Deletion of Tcf3 can delay ES cell differentiation^{18, 19} but, unlike other Tcfs, evidence that β-catenin directly activates Tcf3 target genes is lacking. Significantly, genetic analyses in the embryo^{20, 21} have not revealed a requirement for Wnt in the naïve epiblast, the counterpart of ES cells. Furthermore, Gsk3 is a negative regulator of proteins involved in metabolism, transcription, translation, cell cycle, anti-apoptosis and signal transduction⁴. Its inhibition therefore has potentially much broader effects than canonical Wnt signalling. Importantly, Gsk3 inhibition is not necessary for ES cell propagation if LIF and inhibition of the Mapk cascade are combined² or if LIF is used with serum or Bmp4²².

Selectivity is a general concern with the use of kinase inhibitors. CH has limited cross-reactivity with many other kinases²³, but similar information is not available for BIO. We tested 7 proprietary Gsk3 inhibitors (Supplementary Information, Fig. S1). These compounds have distinct chemical structures, reducing the likelihood of shared off-target effects. In combination with the Mek inhibitor PD0325901 (PD), all promoted undifferentiated ES cell expansion over several passages in bulk culture and enabled colony formation at clonal density in a dose-dependent manner (Fig. 1a). Some of these compounds are effective at nanomolar concentrations. Interestingly, at slightly higher concentrations colony formation was reduced (see compounds C-G, Fig. 1a). We have also observed this effect for CH (data not shown) and infer that incomplete inhibition of Gsk3 is optimal. To test further whether Gsk3 is the critical target we carried out a genetic perturbation. We previously showed that ES cells lacking both isoforms of Gsk3²⁴ can be maintained using a Mek inhibitor alone without CH². However, adaptation of these cells during repeated gene targeting manipulations cannot be excluded. We therefore transfected ES cells with siRNAs against Gsk3α, Gsk3β or both, and scored formation of undifferentiated ES cell colonies at low density in the presence of PD. Immunoblotting confirmed specific knockdown of Gsk3α and β (Supplementary Information, Fig. S2). In wild-type ES cells double knock down produced a small increase in colony number while single knock downs had no effect (data not shown). We then tested ES cells in which one Gsk3α and both Gsk3β alleles have been inactivated²⁴. Gsk3α siRNAs reproducibly increased undifferentiated colony formation by approximately three-fold in these cells while Gsk3β siRNAs had no effect, testifying to the specificity of the siRNA-response (Fig. 1b). Collectively these results validate the conclusion that reduced activity of Gsk3 enhances ES cell self-renewal.

To investigate the involvement of β-catenin we used ES cells carrying null and floxed alleles (βcat^fl/−). Following transient transfection with CreIresGFP, cells were sorted by flow cytometry and plated in N2B27 with PD and CH (2i) plus LIF (2i+LIF, Supplementary Information, Fig. S3a). The GFP negative (non-transfected) population produced compact colonies typical of ES cells in 2i, while the GFP fraction yielded colonies of dispersed cells (Fig. 1c). Both expanded after picking and clonal lines were maintained thereafter in 2i+LIF. Immunoblotting confirmed that the dispersed cells had undergone deletion (Supplementary Information, Fig. S3b). Interestingly, these β-catenin-null (βcat^Δ/−) ES cells re-established cell-cell contacts and normal ES cell colony morphology within 3 passages, even though β-catenin protein remained undetectable (Supplementary Information, Fig. S3b and Fig. 1d). As reported in the accompanying manuscript by Lyashenko et al., this adaptation is most likely due to compensatory up-regulation of plakoglobin²¹ which can replace β-catenin in adherens junctions.

βcat^Δ/− cells expressed Nanog and Oct4 (Fig. 1d,e; Supplementary Information Fig. S3b) and readily formed alkaline phosphatase-positive colonies at clonal density (Fig. 2a). By immunostaining, Nanog appears at a similar level in all cells, suggesting that βcat^Δ/− ES cells in 2i+LIF are uniformly undifferentiated (Fig. 1d). We compared marker gene expression in 2i+LIF to post-implantation epiblast stem cell (EpiSC)s^25-27 (Supplementary Information, Fig. S3c) and to cultures in activin and FGF2 which induces ES cell differentiation into EpiSCs (Fig. 1e). Null cells in 2i + LIF expressed ground state ES cell markers Klf4, Klf2 and Nr0b1 and lacked the early differentiation marker Fgf5, whereas after culture in activin and FGF2 they showed the reciprocal pattern characteristic of EpiSCs. Previous reports of a partly differentiated phenotype of β-catenin-null cells^{28, 29} may therefore reflect characterisation of EpiSCs rather than ES cells. Interestingly the null cells failed to up-regulate brachyury in EpiSC culture, consistent with this being a canonical Wnt target³⁰.

Wild type ES cells can be cultured from single cells in 2i + LIF or with slightly lower efficiency using any two of these components³¹. βcat^Δ/− cells in contrast formed no colonies in 2i and very few in CH+LIF. Only when both PD and LIF were present was colony formation robust (Supplementary Information, Fig. S3d and Fig. 2a). Stable transfection with a β-catenin transgene (Supplementary Information, Fig. S3e) restored clonal expansion in 2i or CH+LIF (Fig. 2a). These results establish that the effect of Gsk3 inhibition is in large part mediated via β-catenin. Nonetheless, colony yield from βcat^Δ/− cells was rather higher in 2i+LIF than in PD+LIF. This result was consistent with three alternative Gsk3 inhibitors (Supplementary Information, Fig. S4). Therefore β-catenin-independent effects of Gsk3-inhibition contribute to maximise ES cell clonogenicity.

We inserted destabilised GFP (GFPd2³²) into the Rex1 locus of βcat^fl/− ES cells and then generated βcat^Δ/− derivative clones, CreC and CreD (Supplementary Information, Fig. S5a,b). Rex1 is down-regulated at the onset of ES cell differentiation³³. In 2i+LIF or PD+LIF, both the βcat^fl/− and βcat^Δ/− ES cells were almost uniformly GFP-positive (Fig. 2b), as reported for other markers in these defined conditions³¹. In 2i or CH+LIF, however, while βcat^fl/− cells retained GFP, CreC and CreD cells lost expression. The pluripotency-associated cell surface marker, PECAM³⁴, behaved in the same manner (Supplementary Information, Fig. S6a). We then assayed colony formation after a 48h period in the different conditions. βcat^fl/− cells produced undifferentiated colonies in all cases except after culture in N2B27 alone (Fig. 2c). βcat^Δ/− cells produced colonies after 48h in 2i+LIF or PD+LIF but not following culture in 2i or CH+LIF.

We cultured Rex1GFPd2 cells for 48 hours in CH, PD or LIF alone. PD or LIF delayed down-regulation of Rex1GFP in βcat^fl/− and βcat^Δ/− cells (Supplementary Information, Fig. S5c) whereas CH maintained GFP in floxed but not in null cells (Fig. 2d). Similar results were obtained in the presence of serum (Supplementary Information, Fig. S5d). In line with Rex1GFP, down-regulation of Nanog was delayed by PD or LIF but not by CH in βcat^Δ/− cells (Supplementary Information, Fig. S6b). qRT-PCR analysis confirmed that CH maintained expression of naïve pluripotency markers and suppressed up-regulation of early differentiation markers in βcat^fl/− ES cells (Fig. 2e). In contrast, βcat^Δ/− ES cells, lost naive markers and up-regulated ectodermal markers Fgf5 and Sox3. Activated caspase-3 was not increased (Supplementary Information, Fig. S6c) indicating that apoptosis is not induced. These results indicate that Gsk3 inhibition resists exit from naïve pluripotency and that this response is dependent on β-catenin.

β-catenin interacts with Tcf/Lef transcription factors and can activate their targets via its C-terminal transcriptional activation domain¹⁰. To test the requirement for β-catenin-mediated transcriptional activation we established clonal βcat^Δ/− “rescue” lines stably expressing full-length β-catenin (RescueWT) or truncated β-cateninΔC (RescueΔC), which lacks the C-terminal transactivation domain⁵ (Fig. 3a). Both full-length and β-cateninΔC localised primarily to the cell membrane (Supplementary Information, Fig. S7). RescueWT but not RescueΔC transfectants activated TOPFlash in response to CH (Fig. 3b). We tested differentiation sensitivity by culture for 48h with CH only followed by replating. Both RescueWT and RescueΔC cells produced undifferentiated colonies indicating restored response to CH (Fig. 3c). Down-regulation of the pluripotency markers Nanog and Klf4 and up-regulation of the early differentiation marker Fgf5 were similarly suppressed by CH in RescueWT and RescueΔC cells (Fig. 3d). These data demonstrate that the canonical transcriptional activation mechanism is not required for β-catenin to increase ES cell resistance to differentiation.

β-cateninΔC cannot function as a transcriptional activator, but it retains domains for interaction with Tcf/Lef factors. Tcf3 is the predominant Tcf/Lef in ES cells and has been reported to behave primarily as a repressor¹⁸. β-catenin might relieve Tcf3-mediated repression¹⁸. This can explain why Axin2 and Cdx1, which are induced by CH in RescueWT cells are also more modestly induced in RescueΔC cells (Fig. 3e). Intriguingly Tcf3 deletion results in delayed ES cell differentiation^{18, 19}. We deployed siRNA in Rex1GFPd2 reporter cells to investigate whether the Tcf3 loss-of-function phenotype¹⁸ is dependent on β-catenin. Rex1GFP down-regulation following withdrawal from 2i+LIF was reduced by Tcf3 siRNA in both βcat^fl/− and βcat^Δ/− ES cells (Fig. 4a). Tcf3 knock down (Supplementary Information, Fig. S8a) also maintained Nanog and Klf4 expression in both βcat^fl/− and βcat^Δ/− ES cells (Fig. 4b). Expression of canonical Wnt pathway targets Axin2 and Cdx1 was increased by Tcf3 siRNA, independent of β-catenin (Fig. 4b), confirming that increased transcription of some Tcf targets does not require direct activation by β-catenin.

Tcf3-null ES cells¹⁸ self-renew robustly in LIF or PD alone and are insensitive to CH (Fig. 4c,d), suggesting that a requirement for Gsk3 inhibition is dictated by Tcf3. To test whether the effect of CH involves direct interaction of β-catenin with Tcf3 we established Tcf3-null ES cell lines expressing wild-type Tcf3 (Tcf3WT) or Tcf3 lacking the β-catenin binding domain (Tcf3ΔN) (Supplementary Information, Fig. S8b). Both Tcf3WT and Tcf3ΔN suppress activation of the TOPFlash reporter by CH (Fig. 5a). Similarly, both reduced expression of Cdx1 (Fig. 5b). Consistent with insensitivity to β-catenin, Tcf3ΔN was the more effective repressor of this endogenous Wnt target. In PD alone Tcf3-null ES cells efficiently formed undifferentiated colonies whereas both Tcf3WT and Tcf3ΔN transfectants behaved like wild type and failed to clone (Fig. 5c). Addition of CH restored colony forming efficiency by Tcf3WT cells. In contrast cells expressing Tcf3ΔN formed very few colonies. Importantly, however, they produced undifferentiated colonies at similar frequency to other cells in PD+LIF. Interestingly, the addition of CH in this condition increased colony numbers for both Tcf3WT and Tcf3ΔN cells. These data demonstrate that the response to CH is largely mediated by a direct interaction between Tcf3 and β-catenin but also confirm that additional effects downstream of Gsk3 inhibition promote maximal ES cell self-renewal.

Tcf3 binds in proximity to many gene promoters that are also bound by Oct4 and has been proposed as a component of a recursive circuit at the core of the pluripotent transcriptome^{6, 35, 36}. By chromatin immunoprecipitation (ChIP) we confirmed detection of Tcf3 at promoters of Klf2 and Nodal (Fig. 5d)³⁶. In the presence of PD and LIF we found that these and other candidate Tcf3 targets showed no consistent transcriptional up-regulation on addition of CH (data not shown). However, CH has minimal functional impact in this context (Fig. 2a,b,c). Cells cultured for 24 hours without 2i+LIF alone down-regulate naive genes preparatory to commitment. In these conditions addition of CH reproducibly induced up-regulation of Klf2 and Nodal (Fig. 5e). This effect was β-catenin-dependent but was observed in RescueΔC cells indicating that it is not mediated by canonical activation. We also found that Tcf3 target genes^{6, 36} were up-regulated when CH was added to ES cells in LIF and serum (Supplementary Information, Fig. S8c).

Canonical Wnt signalling has been proposed to support pluripotency by converting Tcf3 complexes from repressors to activators or by displacing Tcf3 with other Tcfs through which β-catenin activates transcription⁶. However, a simpler model is indicated by the findings that: (i) loss of Tcf3 phenocopies inhibition of Gsk3 even in the absence of β-catenin; (ii) transcriptionally inactive β-catenin fully restores responsiveness to Gsk3 inhibition; and, (iii) full responsiveness to CH requires interaction between β-catenin and Tcf3. We propose that β-catenin abrogates Tcf3 repression and that this is sufficient to permit transcription at Tcf3 target genes that are also bound by pluripotency factors (Fig. 5f,g).

Recently Kelly et al³⁷ proposed a model in which β-catenin acts to inhibit differentiation of ES cells independent of activation of Tcf/Lef targets. Interaction with Oct4 and modulation of Oct4 target genes is proposed to account for the effects of β-catenin stabilisation downstream of Gsk 3 inhibition. However, these authors did not consider Tcf3. The data they present are consistent with our findings and can be explained by β-catenin-mediated derepression of Tcf3 targets, including Oct4³⁶. Furthermore, the reported interaction between β-catenin and Oct4³⁷ could reflect recruitment of β-catenin by Tcf3 to promoter sites co-occupied by Oct4.

In the mouse embryo, genetic analyses have not revealed a function for canonical Wnt signalling in early epiblast cells²¹, although a later role in axis formation in vertebrate embryos is well-defined^{20, 21, 38}. Wnt signalling has been shown to promote differentiation in ES cell-derived embryoid bodies³⁹ and characterised targets induced by the canonical pathway in differentiating ES cells include mesodermal lineage specification genes such as brachyury³⁰. Exit from naïve pluripotency and entry into differentiation may be associated with up-regulation of other Tcfs and a switch in the mode of action of β-catenin from derepression of Tcf3 to direct transcriptional activation. It should also be noted that the cell adhesion role of β-catenin is crucial for differentiation, as documented in the accompanying paper from Lyashenko et al.

Characterisation of β-catenin-null ES cells establishes that β-catenin is not essential for maintenance of the pluripotent ground state. However, β-catenin does mediate the additional resistance to differentiation conferred by inhibition of Gsk3. This facultative recruitment of β-catenin is only required if ES cells are cultured without LIF. Consistent with this, Stat3-null ES cells are dependent on inhibition of Gsk3². It is important to note, however, that although the two pathways can substitute for one another they also have independent effects because ES cells cannot be stably maintained by LIF or CH alone but thrive in the presence of both³¹. Furthermore, addition of LIF to 2i or of CH to PD+LIF increases clonogenicity and for some mouse strains and the rat the combination of all three is important for robust ES cell propagation³¹.

In summary, selective inhibition of Gsk3 consolidates the ES cell ground state primarily, although not exclusively, via stabilisation of intracellular β-catenin, which eliminates the repressive influence of Tcf3 on the pluripotency network (Fig. 5 f,g). This mechanism is not essential when the core pluripotency factors Oct4, Sox2, Nanog, Klf2 and Klf4 are robustly expressed, but can complement activation of Stat3 and/or inhibition of Mapk to stabilise transcription of these core factors and their targets.