FoxG1 and TLE2 act cooperatively to regulate ventral telencephalon formation

The vertebrate forebrain is a complex neural network consisting of a wide variety of highly specified neurons and glia. The correct localisation and timed differentiation of a variety of neuron subtypes requires the coordinated action of a large number of signalling cascades and transcriptional regulators (Hebert and Fishell, 2008). One factor involved in the timing of neuronal differentiation and in the specification of subdomain identity is the transcriptional repressor FoxG1, a member of the Fox/Forkhead family of winged-helix transcription factors [previously known as BF-1 and renamed Foxg1 for mouse, FoxG1 for other chordates and FOXG1 for human (Kaestner et al., 2000)]. FoxG1 expression is conserved throughout vertebrates in dividing progenitors of the ventricular zone and early postmitotic neurons in the telencephalic neuroepithelium (Bourguignon et al., 1998; Murphy et al., 1994; Tao and Lai, 1992; Zhao et al., 2009). Loss- and gain-of-function experiments in different species have established FoxG1 as a maintenance factor for neural progenitors. Knockout of Foxg1 leads to premature differentiation in the dorsal telencephalon and to a drastic increase in early-fate CR neurons accompanied by a loss of late progenitors (Hanashima et al., 2004; Xuan et al., 1995). Conversely, in gain-of-function experiments in frogs, a high dose of FoxG1 expands the progenitor population (Bourguignon et al., 1998; Hardcastle and Papalopulu, 2000), whereas in chicken it causes overgrowth of the neural tube (Ahlgren et al., 2003). siRNA knockdown experiments have demonstrated a role for the level of FoxG1 in controlling the timing of cortical layer differentiation (Shen et al., 2006), suggesting a concentration-dependent activity.

A second important role of FoxG1 is during the specification of the ventral telencephalon (subpallium) and its delineation against the dorsal telencephalon (pallium) and basal diencephalon (hypothalamus). In the mouse Foxg1 knockout, all markers for the ventral telencephalon are lost from an early stage, suggesting that the ventral telencephalon is never specified, while dorsal markers spread ventrally (Martynoga et al., 2005). In fish, morpholino (MO) knockdown experiments demonstrated a similar loss of the subpallium, accompanied by a ventral expansion of dorsal telencephalon markers and a ‘slippage’ of ventral telencephalic cells into the hypothalamic territory of the diencephalon (Danesin et al., 2009). Recently, it has been shown that Foxg1 is required cell-autonomously in the acquisition of ventral (subpallial) telencephalic identity in the mouse (Manuel et al., 2010).

How does FoxG1 exert these functions? At the molecular level, FoxG1 acts mainly as a transcriptional repressor by direct and indirect mechanisms (Bourguignon et al., 1998; Dou et al., 2000; Li et al., 1995; Seoane et al., 2004; Yao et al., 2001). FoxG1 shows direct repression on reporter constructs (Li et al., 1995; Yao et al., 2001), and in vivo it directly represses p27Xic and wnt8b (Hardcastle and Papalopulu, 2000; Danesin et al., 2009). However, it can also negatively regulate TGFβ signalling by binding to Smad and FoxO transcription factors (Dou et al., 2000; Seoane et al., 2004). The proliferation-promoting effect of mouse Foxg1 is independent of its DNA-binding ability (Dou et al., 2000; Hanashima et al., 2002). Transcriptional repression by FoxG1 is partly mediated by recruiting transcriptional co-repressors of the Groucho/Transducin-like enhancer of split (TLE) and AT-rich interaction domain (ARID) families (Sonderegger and Vogt, 2003; Tan et al., 2003; Yao et al., 2001). These co-repressors, in turn, recruit chromatin-modifying enzymes such as histone deacetylases to the transcription factor complex (Chen et al., 1999) or possess demethylation activity themselves (Yamane et al., 2007).

The Groucho/TLE family of transcriptional co-repressors is utilised by a large number of transcription factors to confer repressor activity. TLEs can enhance the activity of active repressors, turn inactive transcription factors into repressors, or even convert transcriptional activators into repressors. Groucho/TLEs are involved in the regulation of a variety of signalling pathways, including Notch, Wnt, TGFβ superfamily and EGF signalling, and show partially overlapping expression in various developing tissues (Buscarlet and Stifani, 2007; Chen and Courey, 2000; Gasperowicz and Otto, 2005; Hasson and Paroush, 2006; Zamparini et al., 2006). Expression of TLE1 and TLE3 has been reported in the ventricular zone of the murine telencephalon (Dehni et al., 1995; Leon and Lobe, 1997) and overexpression of TLE1 causes a delay or inhibition of neurogenesis in vivo and in cultured cortical neurons (Nuthall et al., 2004; Yao et al., 2000). Even though the full mechanism of TLE inhibition of neurogenesis in the developing forebrain still has to be resolved, TLE1 has been shown to bind a group of transcription factors involved in telencephalon patterning and differentiation, including Hes1, FoxG1 and members of the Six family (Kobayashi et al., 2001; Nuthall et al., 2002; Yao et al., 2001). Transfected FoxG1 acts as a repressor of cortical neurogenesis and this effect can be enhanced by TLE1 or reversed by the distant family member TLE6 (Marcal et al., 2005). Since FoxG1 is co-expressed with members of the Groucho/TLE family (Yao et al., 2001), we hypothesised that there might be a specific requirement for individual TLEs for the activity of FoxG1 in forebrain development.

Here, we describe the cloning and characterisation of Xenopus tropicalis TLE2, about which little was known. We show that a conserved N-terminal eh1 motif is necessary for the physical interaction of FoxG1 with TLE2, whereas the C-terminal domain, which has previously been suggested to contain a TLE binding motif, is unnecessary. Nevertheless, we find that the C-terminal domain is necessary for the functional synergism of FoxG1 with TLE2, either alone or in combination with the N-terminal domain. We also show that FoxG1 is co-expressed with TLE2 in the ventral telencephalon (subpallium), and either FoxG1 or TLE2 knockdown abolishes or reduces the development of this region. Our findings suggest that a spatially restricted member of the Groucho/TLE family, TLE2, interacts with FoxG1 to specify the ventral telencephalon.

BLAST searches with human TLE sequences were performed in an EST database (http://informatics.gurdon.cam.ac.uk/online/xt-fl-db.html) and EST clusters were retrieved.

Clusters Xt7.1-ANBT192.5.5, Xt7.1-TEgg056g07.3 and Xt7.1-TGas107e13.3 contain the homologue of TLE1, TLE2 and TLE4, respectively; Xt7.1-CABD14417.5 contains the homologue of the short form AES. X. tropicalis sequence IDs: TLE1 (GU014558), TLE2 (GU014559), TLE4 (GU014560) and AES (GU014561). Human sequence IDs: TLE1 (NP_005068.2), TLE2 (AAH17364), TLE3 (AAH43247) and TLE4 (NP_008936). Additional BLAST search with the JGI X. tropicalis genome server (http://genome.jgi-psf.org/Xentr4/Xentr4.home.html) confirmed these results and did not identify any other sequences. Sequence alignments were performed by CLUSTALW (Larkin et al., 2007) on http://align.genome.jp/ using standard settings.

To generate antisense RNA probes, we used the following single EST clones: TGas001p14 for TLE1, TEgg002e09 for TLE2, TGas107e13 for TLE4 and TGas097a10 for AES.

Full-length constructs and Flag-tagged constructs were cloned from EST sequences by PCR and inserted into pCS2 using the following restriction sites: TLE1 was cloned via XbaI, TLE2 via XbaI, TLE4 via EcoRI/XbaI and AES via EcoRI/XbaI. Since there was no full-length clone available for TLE4, the sequence was first amplified from X. tropicalis stage (st.) 19 cDNA using 5′-GCATTAGGCTTACTATTAACACAAGGAGTC-3′ and 5′-TACAAGTATAATACAAATTCAGAGAATCACAA-3′. Flag epitopes (DYKDDDDK) were introduced by linker-PCR at the N-terminus directly after the start methionine. X. tropicalis FoxG1 was not found in the EST database and was PCR cloned from genomic sequence in pSC2. X. tropicalis FoxG1 has no introns in the coding sequence.

Xenopus laevis FoxG1-HA has been described elsewhere (Regad et al., 2007) and was used as a template for all mutants. FoxG1-N3A-HA was generated by mutating I20, L23 and V24 into alanines using the Stratagene QuikChange II Site-Directed Mutagenesis Kit and the QuikChange Primer Design Program (www.stratagene.com) to design oligos. Similarly, FoxG1-C4A-HA was generated by mutating L309, V311, L314 and V315 into alanines. FoxG1-HA constructs to mutate F233, F246, W255 and F260 into alanines were generated the same way. For FoxG1-ΔC-HA, sequences encoding amino acids 1-253 and 261-436 were amplified by recombinant PCR using 5′-TGGTGCAAGGAGAGCAGGGAGCCCGCTCTATCCATAAAGGT-3′ and 5′-GCGGGCTCCCTGCTCTCCTTGCACCACCCAAGGGCCAGC-3′ as inner primers and SP6 and T7 as outer primers. The final fragment was ligated, following EcoRI/XbaI digestion, into pCS2.

Xenopus embryos were injected in one cell at the two-cell stage and processed as described (Bourguignon et al., 1998). lacZ mRNA (200 pg) was co-injected as a linage tracer. All mRNA injections were performed in X. laevis, using X. laevis FoxG1 and X. tropicalis TLE2 mRNA. Antisense probes for X. laevis N-tubulin and X. tropicalis elrC have been described previously (Bourguignon et al., 1998; Carruthers et al., 2003). For other X. tropicalis antisense probes we used the following ESTs: for Dlx5, clone TNeu071c08; for TLE1, TGas001p14; for TLE2, TEgg002e09; for AES, TGAs097a10; and for emx1, TNeu056k18. emx1 was cloned by PCR into pSC2.

X-Gal staining and in situ hybridisation were carried out as described (Bourguignon et al., 1998).

All MO experiments were performed in X. tropicalis with MOs designed against the X. tropicalis genes. MOs were designed and produced by Gene Tools. All MOs used were FITC labelled. The Control MO is the standard Gene Tools control 25-mer (5′-CCTCTTACCTCAGTTACAATTTATA-3′). The upstream/ATG FoxG1 MO is 5′-TCACAAGGATTGGAGCCGGACACTC-3′ (–26 to –2); the TLE2 exon 1-intron 1 splice MO is 5′-ACGTCACACAACACACTTACCGGAG-3′. To detect the efficiency of the TLE2 exon 1-intron 1 splice MO, we performed semi-quantitative RT-PCR using primers that anneal in exon 1 (5′-CAAACCGAGACAGTCCCACT-3′) and exon 3 (5′-TTTCTGTTGCCAGCTTTTCA-3′) of TLE2. The predicted product size is 220 bp. Interfering with the splicing event would lead to larger products or to premature degradation of the mRNA.

HEK 293T cells from ATCC (30-2002) were maintained in DMEM supplemented with 10% FBS (DE14-801FH, Bio-Whittaker) and penicillin/streptomycin. Individual plasmid DNA constructs were transfected into cells plated in 6-well plates with Lipofectamine 2000 (Invitrogen) following the manufacturer's protocol. The following amounts were used: GFP, 1 μg; FoxG1-HA and FoxG1 mutants, 0.5 μg; TLE1-Flag and TLE2-Flag, 1.5 μg. Each transfection was supplemented with pCS2 to a total of 2 μg DNA.

Co-immunoprecipitation (co-IP) and western blotting of HEK 293T cells were performed by standard methods. Forty-eight hours post-transfection, PBS-washed HEK 293T cells were scraped using 0.5 ml IP-lysis buffer (50 mM Tris-acetate pH 7.5, 300 mM NaCl, 1 mg/ml BSA, 2% IGEPAL or NP40, protease inhibitors) and rotated for 20 minutes at 4°C followed by 10 minutes centrifugation at 16,000 g.

Supernatant was collected as cell lysate and 10 μl kept for expression analysis. Cell lysates were supplemented with another 0.5 ml IP-lysis buffer and 1.5 μg anti-Flag antibody (F1804, Sigma) and rotated for 3 hours at 4°C. Following the addition of 30 μl preincubated Protein A/G Sepharose (sc-2003, Santa Cruz), samples were rotated overnight at 4°C. After washing three times with IP-washing buffer (50 mM Tris-acetate pH 7.5, 300 mM NaCl, 0.5% IGEPAL, 0.1% SDS), Sepharose beads were incubated with 30 μl SDS-loading buffer and, after heating for 10 minutes at 95°C and centrifugation for 3 minutes at 16,000 g, the supernatant was loaded on a 10% SDS-PAGE gel.

Western blotting was performed in a semi-dry system according to the manufacturer's instructions (Hoefer). PVDF membrane (Millipore) was blocked with 5% dry milk powder in PBS containing 0.1% Tween 20 for 30 minutes followed by incubation with anti-Flag-HRP (1:1000; A8592, Sigma) or anti-HA-HRP (1:1000; 12013819001, Roche) antibodies in blocking buffer overnight. Following three washing steps in PBS containing 0.1% Tween 20, blots were developed using ECL reagents (GE Healthcare).

For X. tropicalis FoxG1 immunohistochemistry, embryos were fixed for 24 hours at 4°C in Dent's Fixative (80% methanol, 20% DMSO), washed three times for 10 minutes in PBS, transferred into 15% fish gelatine/15% sucrose for 16 hours, sectioned on a cryostat and stained following standard protocols (Regad et al., 2007).

For tissue culture immunohistochemistry, HeLa cells grown on coverslips were fixed with PBS containing 4% paraformaldehyde for 20 minutes and processed as described (Regad et al., 2007).

The following primary antibodies were used: rabbit anti-FoxG1 (1:50) (Regad et al., 2007), mouse anti-phospho-histone H3 (anti-pH3; 1:500, Upstate) and rat anti-HA (1:200, Roche).

For sectioning after whole-mount in situ hybridisation, embryos were processed as above and dried sections (2 hours at room temperature) were treated for 2 minutes with acetone followed by 2 minutes in PBS containing 0.1% Triton X-100 and then embedded in 90% glycerol.

X. tropicalis embryos were injected in one cell at the two-cell stage with 7.5 ng of the respective FITC-conjugated MO (Control, TLE2 or FoxG1 MO). At st. 28, embryos were fixed and sectioned at 12 μm using a Leica CM3050 S cryostat. TUNEL staining was performed using the TMR Red In Situ Cell Death Detection Kit according to the manufacturer's instructions (Roche). The injected side was identified using mouse anti-FITC primary (1:250, Roche) and Alexa Fluor 488 goat anti-mouse secondary (1:500, Molecular Probes) antibodies.

X. tropicalis embryos were injected at the one-cell stage with 15 ng X. tropicalis FoxG1 MO. At st. 25, embryos were injected in the brain ventricle with 30 μl of a mixture of 1 μg/μl CMV-β-gal and 100 ng/μl CMV-FoxG1 DNA (X. laevis) or mutated versions thereof (note that the X. tropicalis MO does not recognise X. laevis FoxG1 as it is targeted to an upstream region that is not present in the X. laevis construct). Immediately after injection, embryos were electroporated using home-made platinum electrodes and an SD9 stimulator (Grass Technologies) at the following settings: 20V, 50-millisecond pulse duration, 1-second interpulse space, six pulses. Embryos were fixed and analysed for β-gal and Nkx2.1 expression at st. 31-32.

We scanned the largest available X. tropicalis EST collection (http://informatics.gurdon.cam.ac.uk/online/xt-fl-db.html) for homologues of the six human members of the Groucho/TLE family (Gasperowicz and Otto, 2005). Three of the resulting four hits could easily be assigned as the homologues of human TLE1, TLE4 and AES based on sequence homology. The last sequence had only 64%and 67% identity to human TLE2 and TLE3, respectively. However, synteny analysis revealed that the sequence has the same neighbouring chromosomal genes as human and mouse TLE2/Tle2 (Fig. 1). We therefore decided to name this gene X. tropicalis TLE2, a decision supported by phylogenetic tree analysis (see Fig. S1 in the supplementary material). The adjacent gene to TLE2 in the human, murine and Xenopus genome is AES/Grg5, a truncated member of the TLE family that has been generated by tandem gene duplication (Bajoghli, 2007) and shows dominant-negative activity (Roose et al., 1998). Also, TLE1 and TLE4 are arranged adjacent to each other in both the human and Xenopus genome. X. tropicalis TLE1, TLE2, TLE4 and AES were fully sequenced and TLE1, TLE2 and TLE4 were shown to have the typical Groucho family domain structure (Fig. 2). Xenopus laevis TLE2, TLE4 and AES have been cloned before and named ESG1, Xgrg-4 and AES/Xgrg-5, respectively (Choudhury et al., 1997; Roose et al., 1998).

At the neural plate stage (st. 16), all three TLEs (TLE1, TLE2 and AES) were expressed throughout the anterior neural plate (ANP). Expression of TLE1 appeared to be stronger at the anterior-most neural fold, similar to FoxG1 expression in the presumptive telencephalic area (Fig. 3, left column). For TLE2 and AES, the strongest expression was in a small horseshoe-shaped area in a more medial part of the ANP that is fated to give rise to the ventral diencephalon (Eagleson and Harris, 1990). At late neurula stage (st. 22-23) TLE1, TLE2 and AES were partially co-expressed, showing areas of stronger and weaker expression along the neural tube as well as in the eye primordium. In their anterior-most expression they all extended into the ventral forebrain region and possibly overlapped with FoxG1 (Fig. 3, middle column). At early tadpole stage (st. 28), the expression of the individual TLEs became more specific (Fig. 3, right column). TLE1 showed strong expression in the telencephalon, the eye, mid- and hindbrain and the branchial arches and weaker expression along the neural tube. TLE2 appeared to be expressed in two stripes in the midbrain/diencephalon area and in the eye, weakly along the neural tube and diffusely in the branchial arch region. However, in whole-mounts, TLE2 appeared to be absent from the telencephalon (Fig. 3, red arrowhead). AESshowed expression in the forebrain, eye, branchial arches, otic vesicle and along the neural tube, similar to the expression of AES in X. laevis (Molenaar et al., 2000). The expression of TLE4 was low and diffuse (data not shown), unlike X. laevis TLE4 (Xgrg-4), which is widely expressed but is higher in neural tissues (Molenaar et al., 2000). FoxG1 was expressed throughout the telencephalon, the olfactory placodes and the cranial nerves as well as in the anterior retina. This expression pattern is identical to that of X. laevis FoxG1 and the two gene products are 96% identical at the amino acid level (see Fig. S2 in the supplementary material).

In cleared or sectioned whole-mount embryos at st. 22 (Fig. 4D), X. tropicalis TLE1 and TLE2 were expressed in the forebrain and also in more posterior parts of the CNS. Within the telencephalon, TLE1 showed a substantial overlap with FoxG1 (Fig. 4B,G), whereas TLE2 overlapped with FoxG1 only in the ventral-most telencephalon (Fig. 4B,K,E). FoxG1 and TLE2, but not TLE1, were also co-expressed in the anterior optic chiasm, which is the ventral boundary of the telencephalon with the diencephalon (Fig. 4C,H,L) (Bachy et al., 2002).

FoxG1 affects neuronal differentiation in a dose-dependent manner when it is ectopically overexpressed in Xenopus embryos (Bourguignon et al., 1998). A low dose of FoxG1 cell-autonomously enhances neurogenesis, whereas a higher dose cell-autonomously suppresses endogenous neurogenesis. Thus, FoxG1 misexpression forms a convenient assay to test for synergism of a factor with FoxG1. To examine whether FoxG1 and TLE2 work together, we injected TLE2 mRNA together with a low dose of FoxG1. Low-dose injection of FoxG1 mRNA alone caused ectopic neurogenesis (97%, n=39) on the injected side (Fig. 5A,G), whereas high-dose injection resulted in suppression of the lateral stripe of primary neurogenesis and in ectopic neurons adjacent to the injected region (90%, n=39) (Fig. 5B,C,G), as previously described (Bourguignon et al., 1998). Injection of 200 pg TLE2 mRNA had no effect (82%, n=40) (Fig. 5D,G) or resulted in minor ectopic neurogenesis (18%, n=40), but co-injection with 50 pg FoxG1 resulted in a predominant high-dose phenotype (58%, n=38) (Fig. 5E-G). This shows that the presence of the co-repressor TLE2 can change the activity of FoxG1. We were not able to test TLE1 in this assay as it was very poorly expressed in the embryo from the injected mRNA (data not shown).

We next asked whether TLEs physically interact with FoxG1. Co-immunoprecipitation (co-IP) and western blotting of transfected HEK 293T cells showed that both TLE2 and TLE1 interact with FoxG1. The interaction with TLE1 might be weaker, as consistently less FoxG1 was immunoprecipitated with TLE1 than with TLE2 (Fig. 5H).

The binding of TLEs to their target transcription factors is, in most cases, mediated via their WDR domain. This domain recognises two conserved motifs in the proteins that TLEs interact with: a WRPW motif and an engrailed homology 1 (eh1) domain, FXIXXIL (reviewed by Buscarlet and Stifani, 2007; Gasperowicz and Otto, 2005), which is present in eight classes of the homeodomain protein superfamily and in more than 50% of all members of the Fox family (Buscarlet and Stifani, 2007; Gasperowicz and Otto, 2005; Yaklichkin et al., 2007b). An interaction between human TLE1 and murine or chicken Foxg1 has been mapped to a C-terminal domain of Foxg1 (Sonderegger and Vogt, 2003; Yao et al., 2001), comprising amino acids 276-336 of mouse Foxg1 (which corresponds to amino acids 229-292 in Xenopus FoxG1). This domain contains a peptide that shows structural similarity to WRPW motifs and is conserved in Xenopus FoxG1 [Y254WPMSPFLSLH (Yao et al., 2001)].

To test the potential of each motif to contribute to the interaction between FoxG1 and TLE2, we changed either the three conserved hydrophobic amino acid residues of the N-terminal eh1 to alanines (FoxG1-N3A-HA) or deleted the C-terminal motif (amino acids 254 to 260, FoxG1-ΔC-HA; see Fig. 6A,B). In transfected HEK 293T cells, the C-terminal mutation did not have an effect on TLE2 binding, whereas FoxG1-N3A-HA lost the interaction with TLE2 (Fig. 6C). A double mutant, FoxG1-N3A-ΔC-HA, also did not interact with TLE2 (Fig. 6C). FoxG1 contains a double CK1-phosphorylation motif near the N-terminus that is required for its nuclear localisation (Regad et al., 2007). This phosphorylation site overlaps with the N-terminal eh1 motif and the lack of interaction between FoxG1-N3A-HA and TLE2 could be due to a failure of FoxG1 to localise to the nucleus. However, in transfected HEK 293T cells, both wild-type and mutant FoxG1 were nuclear (Fig. 6E, right panel).

A computational analysis of eh1 motifs within the Fox family of transcription factors revealed that in addition to the fully conserved eh1 motif at the N-terminus (F18SINSLV), there is a ‘remnant’ eh1 motif in the C-terminal region of FoxG1 that lacks the highly conserved phenylalanine [L309SVDRLV (Yaklichkin et al., 2007b)]. Mutation of the four hydrophobic amino acid residues of the remnant C-terminal eh1 motif to alanines (FoxG1-C4A-HA) did not affect TLE binding (Fig. 6C) and was not investigated further. In parallel, we tested the ability of all FoxG1 mutants to bind TLE1, and the results were very similar to those obtained with TLE2 (data not shown).

To address the functional activity of the FoxG1 mutants, we assessed the potential of the mutants to convert a low-dose FoxG1 mRNA injection into a high-dose phenotype. Injection of 400 pg TLE2 mRNA alone had only a minor effect on neurogenesis, showing a mild low-dose phenotype (30%, n=61) (Fig. 6D). Injection of wild-type FoxG1 together with TLE2 showed a strong conversion to a high-dose phenotype (74%, n=46); however, the FoxG1-N3A-HA and the double mutant FoxG1-N3A-ΔC-HA produced a very low percentage of the high-dose phenotype when co-injected with TLE2 (8%, n=50 and 11%, n=45, respectively). Surprisingly, the deletion mutant FoxG1-ΔC-HA was unable to efficiently convert a low-dose injection into a high-dose phenotype in the presence of TLE2 (5%, n=66) (Fig. 6D). All three FoxG1 mutants were able to induce a low-dose or high-dose phenotype when overexpressed alone, although the double mutant showed a reduced low-dose phenotype, in the sense that the number of ectopic neurons observed was lower (data not shown). This suggest that all mutants are expressed and are functionally active and, in addition, that the double mutant is more severe that the single mutants. Taken together, these results suggest that the N-terminal eh1 motif within FoxG1 plays a crucial role in the physical and functional interaction with TLEs. The C-terminal motif does not appear necessary for interaction with TLE2, but it is required for functional synergism, which could be indirect.

If FoxG1 and TLE2 interact in vivo in the developing forebrain we would expect common aspects in their respective knockdown phenotypes. FoxG1 is encoded by an intronless gene and was knocked down by an ATG/upstream MO, whereas TLE2 was knocked down by injection of a splice MO targeted to the exon 1-intron 1 boundary. The knockdown of FoxG1 protein was determined by immunohistochemistry with anti-FoxG1 antibody (Fig. 7A), and the effectiveness of the TLE2 splice MO was assessed by aberrant splicing as detected by PCR (Fig. 7B,C).

To assess possible similarities in the loss-of-function phenotype we injected Control, FoxG1 or TLE2 MO into X. tropicalis embryos and performed whole-mount in situ hybridisation for the ventral telencephalon marker Nkx2.1. Nkx2.1 is also expressed in the diencephalon, but within the telencephalon it is good marker for the ventral-most part of the subpallium: the pallidum or medial ganglionic eminence (Bachy et al., 2002). In FoxG1 MO-injected embryos, the subpallial expression of Nkx2.1 was entirely missing (100%, n=11), consistent with previous data from mouse and fish (Martynoga et al., 2005; Manuel et al., 2010; Danesin et al., 2009), and in TLE2 MO embryos it was greatly reduced (70%, n=10) (Fig. 7G).

To assess whether FoxG1 knockdown also affects neurogenesis in the Xenopus forebrain, we studied the effect of FoxG1 and TLE2 MO knockdown on elrC expression, which is one of the earliest markers of neuronal differentiation, being expressed in cells undergoing the transition from proliferation to differentiation (Carruthers et al., 2003; Good, 1995; Perron et al., 1999). We observed a reduction in the expression domain of elrC in the forebrain of st. 19 embryos after either FoxG1 or TLE2 MO injection (Fig. 7H-K). Loss of elrC expression could indicate a failure of early neural specification, increased cell death or a failure to exit the progenitor compartment. We did not observe a statistically significant increase in cell death with either MO (Fig. 7L,M,N; see Fig. S3E in the supplementary material). Failure to exit the progenitor compartment should result in an increase in proliferation. However, we found a significant reduction (by 40%) in pH3-positive cells in the FoxG1 MO-injected side as compared with the control side in both st. 19 and st. 28 embryos (n=12, P=0.003) (Fig. 7O; see also anti-pH3-stained green nuclei in Fig. 7A), suggesting that the loss of elrC is due to a loss of neural specification in the ventral forebrain, rather than to any prolonged maintenance of cells in the progenitor state.

To address whether the loss of ventral telencephalic fate is accompanied by an increase in dorsal markers, we assessed the expression of emx1, a dorsal telencephalic (pallial) marker (Bachy et al., 2002). emx1 expression was expanded ventrally in the FoxG1 MO embryos, but not in the TLE2 MO embryos (see Fig. S3 in the supplementary material), suggesting that FoxG1 knockdown, but not TLE2 knockdown, causes a widespread respecification.

Finally, to test for functional synergism of FoxG1 with TLE2 in vivo, we tested the ability of wild-type and mutated FoxG1 to induce Nkx2.1 expression in a FoxG1 MO background (Fig. 7P-T). We found that wild-type FoxG1 and the N-terminal and C-terminal mutants (FoxG1-N3A and FoxG1-ΔC) were able to induce Nkx2.1, suggesting that in the forebrain each motif is able to compensate for the absence of the other. However, the double mutant, in which both N-terminal and C-terminal motifs are affected (FoxG1-N3A-ΔC), was unable to induce Nkx2.1. This finding supports the notion that cooperation between the TLE2-binding eh1 motif and the C-terminal WRPW-like motif is necessary for ventral telencephalon specification.

In this study, we have cloned and characterised the expression of four X. tropicalis TLEs: TLE1, TLE2, TLE4 and AES. We have shown that TLE1, TLE2 and AES are highly expressed in the nervous system; however, they did not show identical expression patterns. We focused on the expression of TLE1 and TLE2 in the developing forebrain, specifically the telencephalon. Expression of TLE2 has previously been described only by northern blotting (Choudhury et al., 1997).

We have shown that these genes are co-expressed at the neural plate stage, but show a bias in the strength of expression towards different future subdomains of the forebrain. Later on, these differences are amplified, as TLE1 is expressed throughout the telencephalon whereas TLE2 is restricted to the ventral-most part, i.e. the subpallium and the optic chiasm, which forms the border between telencephalon and diencephalon and where TLE1 is not expressed. TLE1 and TLE2 are expressed in more-posterior regions in distinct patterns, which we have not detailed here. The mouse homologue of TLE1 is expressed in the embryonic telencephalon (Dehni et al., 1995; Xuan et al., 1995; Yao et al., 1998). TLE2 exhibits very limited expression in the mouse forebrain (Grbavec et al., 1998) and is also expressed in the developing mammalian pancreas and in chicken feather buds (Hoffman et al., 2008; Houghton et al., 2003).

Whether TLEs have distinct or overlapping (redundant) functions is an open question and there are data to support either possibility (Gasperowicz and Otto, 2005; Yao et al., 1998). Our expression analysis of Xenopus TLE1 and TLE2 supports the view that these transcriptional co-factors have distinct functions; their neural expression profiles only partially overlap, providing the potential to interact with different transcription factors.

Previous studies in chicken and mouse that analysed the interaction between TLE1 and Foxg1 used large deletions to identify potential binding sites (Sonderegger and Vogt, 2003; Yao et al., 2001). Both groups identified a polypeptide of ~65-70 amino acids C-terminal of the Foxg1 DNA-binding domain as crucial for the interaction with TLE1 in in-vitro assays. Within the binding region, the short peptide motif YWPMSPFLSLH was reported to show similarity to WRPW motifs (Yao et al., 2001). Another study has shown that the anti-neurogenic activity of TLE1 is due to binding of TLE1 to Hes1 and other WRPW motif-containing proteins, rather than to eh1 motif-containing binding partners, such as FoxG1 (Buscarlet et al., 2008). Therefore, it was unclear whether the co-repressor TLE1, or the eh1 motif of Foxg1, is important for the activity of Foxg1 in the developing telencephalon.

However, in co-IP assays performed in cell culture between FoxG1 and TLE2, we could not detect any effect on TLE2 binding to FoxG1 when the C-terminal motif is deleted. The mutation of single aromatic amino acids (crucial components of both known motifs) in this region also had no effect on binding (data not shown). By contrast, we identified an N-terminal eh1 motif in FoxG1 as the main binding site for TLE2. Although the co-IP assay does not allow us to conclude with certainty that the interaction between FoxG1 and Groucho/TLE is direct, the eh1 motif shows the strongest conservation with known TLE binding sites (Yaklichkin et al., 2007a; Yaklichkin et al., 2007b).

How can the differences to previous reports be explained? One possibility is that FoxG1 interacts with TLE1 and TLE2 via different domains. However, this is not supported by our data, as we have not found any differences between TLE1 and TLE2 in co-IP assays (except for a lower amount of immunoprecipitated FoxG1 with TLE1). Another possibility is that the differences are due to the assays used. For example, the work on TLE1 used in vitro assays with bacterially derived proteins, which might lack important modifications necessary for binding to the eh1 motif. Indeed, a previous study has shown that the phosphorylation state of S/T residues next to the initial phenylalanine in an eh1 motif has an influence on the interaction with Groucho (Goldstein et al., 2005) and we have previously described a double casein kinase I phosphorylation site that overlaps with the N-terminal eh1 motif (Regad et al., 2007). TLEs can also be post-translationally modified, for example by hyperphosphorylation, which is also induced by the presence of FoxG1 (Buscarlet et al., 2008; Nuthall et al., 2002; Nuthall et al., 2004), or by sumoylation (Ahn et al., 2009). Our analysis bears the advantage of having co-factors and protein modifications present and, under these conditions, which better resemble those in vivo, we detect the N-terminal eh1 motif of FoxG1 as the main binding motif for TLE2.

We have used a Xenopus neurogenesis assay (Bourguignon et al., 1998) to assess the cooperation of FoxG1 with TLE2 in the embryo. Since the expression of FoxG1 in the posterior neural plate and lateral ectoderm is ectopic, this assay can only be used to show functional synergism, rather than to draw any conclusions on the specific activity of FoxG1 in the developing forebrain. With this assay, we have shown that TLE2 modulates the activity of FoxG1. When expressed together with TLE2, a low dose of FoxG1 suppresses endogenous neurogenesis, which is otherwise only observed when FoxG1 is overexpressed at a high dose. We interpret this finding to mean that the presence of TLE2 allows FoxG1 to function more efficiently as a transcriptional repressor. This is analogous to the observation that co-expression of TLE1 enhances the repressor activity of FoxG1 on a FoxG1-sensitive reporter construct (Yao et al., 2001).

Mutating the N-terminal eh1 motif (FoxG1-N3A-HA) leads to a loss of functional cooperation with TLE2, consistent with the loss of TLE2 binding by this mutant in co-IP experiments. The FoxG1-ΔC-HA mutant, which represents a deletion of the C-terminal YWPMSPF motif, also shows a lack of functional cooperation with TLE2 in the neurogenesis assay. This is intriguing because deletion of this site does not abolish interaction with TLE2 in co-IP assays. Therefore, we suggest that even though this site is not necessary for TLE2 binding, it is important for the functional cooperation of FoxG1 with TLE2. One explanation is that a third factor exists, which binds FoxG1 at this motif and has an influence on FoxG1/TLE activity. As the motif YWPMSPF also contains a MAP kinase consensus site (P-X-S/T-P), it could serve as a possible protein-protein interaction site that is dependent on phosphorylation.

Our own work has also uncovered differences that depend on the assay used. In the neurogenesis assay, mutation of either the N-terminal or C-terminal motif is sufficient to interfere with the functional cooperation of exogenous FoxG1 and TLE2. However, within the forebrain, each motif seems to be able to compensate for the absence of the other. In both assays, simultaneous interference with both motifs renders FoxG1 inactive. One possible explanation is that both mutants are compromised in recruiting co-repressors, but different target genes have different threshold requirements for transcriptional repression. The presence of different additional co-factors or FoxG1/TLE modifications in the different assays might also play a role. Although we do not yet fully understand the details of FoxG1-TLE interaction, the common denominator in all the functional assays used in this work is that the double mutant is more severely affected that any single mutant. These findings suggest that both the N-terminal eh1 motif and the C-terminal WRPW-like motif contribute cooperatively to FoxG1 activity, either by interaction with TLE alone or in combination with other co-factors.

If FoxG1 and TLE2 interact in vivo one would expect that the phenotype of knocking down FoxG1 would be similar to that of knocking down TLE2. Indeed, both FoxG1 and TLE2 knockdown affected Nkx2.1 expression in the ventral telencephalon. The expression of Nkx2.1 in this area was eliminated with the FoxG1 MO and reduced with the TLE2 MO. This is consistent with the expression profile of FoxG1 and TLE2 and, particularly, with their co-expression in the ventral telencephalon (subpallium). The fact that FoxG1 knockdown has more severe effects than TLE2 knockdown in the telencephalon can be explained by the observation that FoxG1 is expressed throughout the entire telencephalon, whereas TLE2 is restricted to the ventral-most part.

The requirement for FoxG1 in the specification of the ventral telencephalon is consistent with previous findings in mouse and zebrafish (Danesin et al., 2009; Martynoga et al., 2005). Here, we suggest that this event is mediated by the specific interaction of FoxG1 with the transcriptional co-repressor TLE2, which might provide a mechanism for the cell-autonomous role of FoxG1 in this region (Manuel et al., 2010). Although we have focused on TLE2, we cannot exclude a contribution by TLE1 and finer assays will be needed to distinguish the contribution of each factor.

One of the hallmarks of the FoxG1 knockout and knockdown phenotype in mice and zebrafish is premature neuronal differentiation, particularly in the dorsal forebrain (pallium) (Danesin et al., 2009; Xuan et al., 1995), which transiently results in a greater number of neurons. In Xenopus, we have not seen this increase but instead observed a reduction in newly postmitotic neurons in the telencephalon of st. 19 tadpoles, marked by the expression of elrC. This is unlikely to reflect a delay in neurogenesis accompanied by increased proliferation, as we detected this reduction over a long period of time (st. 19 to st. 25) and we observed a reduction in mitotic cells. Therefore, we suggest that the reduction of elrC expression signifies a failure in the specification of telencephalic tissue that might precede the specification of neurogenesis. Indeed, this would be consistent with our earlier observation that FoxG1 is sufficient to induce the neuroepithelial marker Sox3 in isolated naïve ectoderm (animal caps) (Hardcastle and Papalopulu, 2000). The difference between the Xenopus phenotype and that seen in mouse and zebrafish, in which an increased number of neurons is transiently detected, could be due to phylogenetic differences in the development of the forebrain. Alternatively, partial compensation by other genes, as originally suggested for the mouse (Xuan et al., 1995), might result in different phenotypes. In this respect, it might be relevant that in zebrafish there are three Foxg1 genes: foxg1a is the main Foxg1 paralogue expressed in the telencephalon, but foxg1b, which would not be targeted by the foxg1a MO, is transiently expressed in the ventral telencephalon (Zhao et al., 2009). At present, we cannot exclude the possibility that the reduction in elrC is due to premature neurogenesis that depletes the overall number of progenitors and hence reduces the overall number of neurons generated.

In conclusion, based on the specific overlap of expression between FoxG1 and TLE2 in the ventral telencephalon, their physical and functional interaction as shown by co-IP and a neurogenesis assay, the induction of Nkx2.1 in the forebrain and the similar phenotypes resulting from MO knockdown, we suggest that FoxG1 and TLE2 synergise to specify the ventral telencephalon. It remains to be seen whether other regionally expressed co-factors, such as JARID1B (PLU-1; KDM5B) (Tan et al., 2003), mediate some of the other activities of FoxG1.

We thank Nitin Sabherwal for help with the manuscript. This work was funded by a Wellcome Trust Senior Fellowship grant to N. Papalopulu (057819/Z/05). N. Panagiotaki and B.B. are Wellcome Trust four-year PhD students. Deposited in PMC for release after 6 months.