GRG5/AES Interacts with T-Cell Factor 4 (TCF4) and Downregulates Wnt Signaling in Human Cells and Zebrafish Embryos

Transcriptional control by TCF/LEF proteins is crucial in key developmental processes such as embryo polarity, tissue architecture and cell fate determination. TCFs associate with β-catenin to activate transcription in the presence of Wnt signaling, but in its absence act as repressors together with Groucho-family proteins (GRGs). TCF4 is critical in vertebrate intestinal epithelium, where TCF4-β-catenin complexes are necessary for the maintenance of a proliferative compartment, and their abnormal formation initiates tumorigenesis. However, the extent of TCF4-GRG complexes’ roles in development and the mechanisms by which they repress transcription are not completely understood. Here we characterize the interaction between TCF4 and GRG5/AES, a Groucho family member whose functional relationship with TCFs has been controversial. We map the core GRG interaction region in TCF4 to a 111-amino acid fragment and show that, in contrast to other GRGs, GRG5/AES-binding specifically depends on a 4-amino acid motif (LVPQ) present only in TCF3 and some TCF4 isoforms. We further demonstrate that GRG5/AES represses Wnt-mediated transcription both in human cells and zebrafish embryos. Importantly, we provide the first evidence of an inherent repressive function of GRG5/AES in dorsal-ventral patterning during early zebrafish embryogenesis. These results improve our understanding of TCF-GRG interactions, have significant implications for models of transcriptional repression by TCF-GRG complexes, and lay the groundwork for in depth direct assessment of the potential role of Groucho-family proteins in both normal and abnormal development.

In vertebrates, the Groucho family can be divided into two distinct structural subgroups.The first includes the "long" proteins, termed GRG (1)(2)(3)(4), for Groucho-related gene (or TLE (1)(2)(3)(4), for Transducin-like enhancer of split, in humans), which have five distinct domains (Figure 1B): the highly conserved Q (glutamine-rich) and WD (containing WD40 repeats, and involved in most protein interactions) in the N-and C-termini respectively, separated by the much more variable GP (glycine/ proline-rich), CcN (with putative casein Kinase II/cdc2 phosphorylation sites and a nuclear localization signal), and SP (serine/proline-rich) domains.They are generally accepted as co-repressors of multiple transcription factors with crucial roles in diverse processes, such as segmentation, sex determination, embryo patterning and organogenesis [47,[59][60][61].The second subgroup encompasses the proteins sometimes referred to as the "short Grouchos", including A) The human TCF4 gene consists of 17 exons (top), some of which are subject to alternative splicing (black exons).The TCF4 variant used (Ex1-17) includes isoform specific sequences such as the LVPQ domain due to the use of an alternative splice donor site at the end of exon 7 (arrowhead).A TCF4 fragment lacking the C terminus (a.a.s 7-387), was fused to Gal4's DNA-binding domain (Gal4-BD) and used as bait in a two hybrid screen.B) The GRG family comprises two distinct classes of proteins, based on their domain constitution: the "long Grouchos", GRG/TLE 1-4, consisting of five different domains (Q, GP, CcN, SP and WD40); and the "short Grouchos", such as the alternative splice-product Grg1-S or the GRG5/AES subfamily members, which contain only the Q and GP domains.doi: 10.1371/journal.pone.0067694.g001alternative splicing variants of the long forms, such as Grg1-S (a variant of mouse Grg1) and distinct members expressed from their own loci, such as GRG5 (also referred to as AES, for Amino-terminal enhancer of split).Proteins in this subgroup contain only the first two conserved domains (Q and GP) (Figure 1B) and have sometimes been suggested to act as dominant negatives of the long forms [49,62].However, this view has been contradicted by the demonstration of transcriptional repression by several short forms, including Grg1-S, AES197 (a "truncated" sea-urchin Groucho homologue) and GRG5/AES [63][64][65][66][67][68][69].
Here, we characterize the TCF4-GRG5/AES molecular interaction, map the minimal interacting region in TCF4 to a 111-amino acid stretch and show that, in contrast to other Grouchos, GRG5/AES-binding depends on the 4-amino acid motif LVPQ.Interestingly, both this motif and the 111-amino acid core binding region are present only in some TCF4 isoforms.We further demonstrate that GRG5/AES acts as an efficient repressor of TCF-β-catenin signaling both in human cells and zebrafish embryos, capable of counteracting the effects of activated β-catenin both in axis duplication and DV patterning during zebrafish embryogenesis.These results broaden our understanding of the physical and functional interactions between TCFs and Groucho-family proteins, will help develop more accurate models of Wnt-signaling regulation by the latter, and pave the way for a detailed in vivo analysis of the role played by TCF-GRG complexes in intestinal development, homeostasis and tumorigenesis.
Myc-tagged TCF4 and HA-tagged GRG5/AES mammalian expression plasmids were generated by subcloning of partial and full-length human cDNAs into CMV-promoter driven plasmids (Clontech), in frame with the respective epitopes.CMV-βcat (WT) was generated by removing β-catenin cDNA from pGAD10-βcat and subcloning into pCMV.CMV-βcatT41A (coding for a mutant β-catenin found in human tumors) was derived from CMV-βcat by site directed mutagenesis.Expression vectors for Grg1-S and Grg1-L [63] were obtained from Addgene (Addgene plasmids 11065 and 11067, respectively).CMV-G4VP was generated by subcloning the Gal4-VP16 fusion [70] into a CMV-promoter driven mammalian expression vector; G4BD.GRG5 and G4BD.Grg1-L were generated by removing the VP16 coding sequences from CMV-G4VP (yielding CMV-G4BD) and replacing them with those of GRG5/AES and Grg1-L.
TCF-Luc (containing three TCF consensus binding sites -TREs) and TCF*-Luc (with three mutant TCF binding sites) were previously described [71].For pB15UT-Luc, 3 copies of the 5 Gal4 upstream activation sequences (UAS G ) present in UAS G -βgal [70] were concatamerized and used to replace the TREs in TCF-Luc.For 15UTSV-Luc, the SV40 promoter from "trigger P5" [70] was PCR amplified and cloned into pB15UT-Luc between the UAS G and the luciferase gene.
All plasmids were sequenced to confirm that they had been accurately constructed.More detailed information about the constructs is available upon request.

Yeast two-hybrid assays
A commercial human fetal brain cDNA library in pACT2 (Clontech), was screened by PEG/LiAcetate co-transformation [72] with the bait plasmid pAS2-TCF4 into the AH109 yeast reporter strain (which contains integrated copies of ADE2 and HIS3 reporter genes under control of Gal4-dependent promoters).Primary candidates were selected for their ability to grow in SC-Leu-Trp-Ade medium (BIO101).A secondary screen was performed by selection on medium also lacking histidine and containing 30mM of the HIS3 inhibitor 3-Amino-1,2,4-triazole (3-AT), (Sigma).False positives were screened against by co-transformation of individual candidate prey plasmids with the control bait plasmid pLAM, which codes for a G4BD-laminin fusion protein.
Mapping of the GRG5/AES -binding region in TCF4 was done by individually co-transforming each of the various TCF4fragment encoding bait plasmids with a pACT2-GRG5 plasmid recovered in the library screen and selection on SC-Leu-Trp-Ade medium.For each fragment, the strength of the interaction was evaluated by spotting 60 individual colonies on media also lacking histidine and containing increasing concentrations of 3-AT.Evaluation of the effects of removing the LVPQ motif was done by co-transforming TCF4 bait constructs with either GRG4-QGP or Grg1-S encoding prey plasmids and spotting a mixture of 20 transformant colonies from each interaction tested on SC-Leu-Trp-Ade selective medium.

Prey-plasmid recovery
Putative candidates selected in the secondary screen were grown in SC-Leu-Trp and plasmid DNAs extracted by glass bead lysis and ethanol precipitation.DNA from each yeast clone (containing a mixture of the bait plasmid and one or more prey plasmids) was transformed into E. coli cells by electroporation and colonies containing different prey plasmids identified by PCR/agarose gel electrophoresis screening with bait plasmid and prey plasmid specific primers.Plasmid DNA was isolated by a standard alkaline lysis method and used to identify inserts by sequencing.
For total protein lysates, 20 µg of protein were loaded per case.Proteins were separated by 12% SDS-PAGE, transferred onto a Hybond nitrocellulose membrane (Amersham Biosciences) and subjected to Western blot analysis.Briefly, membranes were blocked for 30 min with 5% nonfat milk in PBS + 0.5% Tween-20 and incubated for 2 h with rabbit polyclonal anti-HA or mouse monoclonal anti-Myc antibodies (Clontech) diluted 1/1000.HA epitope-tagged GRG5/AES and Myc epitope-tagged TCF4 were detected using horseradish peroxidase-conjugated secondary antibodies (Santa Cruz Biotechnology), diluted 1/2000, followed by ECL detection (Amersham Biosciences).

Ethics statement
All experiments were conducted according to an animal protocol approved by the University of Pennsylvania Institutional Animal Care and Use Committee (#802394).Veterinary care was under the supervision of the University Laboratory Animal Resources of the University of Pennsylvania.

Fish and mRNA microinjections
Wild-type fish and embryos were maintained essentially as described previously [73].
Plasmids were linearized and transcribed in vitro using the SP6 Message mMachine Kit (Ambion).Synthesized mRNAs were purified by phenol: chloroform extraction and isopropanol precipitation and injected into one-cell stage embryos, except for the β-catenin-induced double axis experiments, in which mRNAs were injected into one blastomere of eight-cell stage embryos.Embryos were injected according to a previously described procedure [74].

GRG5/AES specifically interacts with TCF4
TCF4 is considered one of the most important regulatory genes in intestinal development and homeostasis in vertebrates.As part of our general interest in a detailed understanding of the functions played by TCF4 in these processes, we tried to identify novel TCF4-interacting proteins.Because we were particularly interested in potential differences among the sets of proteins that interact with TCF4 and those that bind its homologues, we decided to perform a two-hybrid screen similar to the one that led to the identification of βcatenin as a TCF1 partner [13].A truncated form of TCF4 (7-387 amino acids) fused to the Gal4-DNA-binding domain (G4BD) was therefore used as bait to screen a Gal4-activation domain (G4AD)-fused human fetal brain cDNA library (Figure 1A).This TCF4 fragment retains the N-terminal β-catenin interacting domain, as well as most of the HMG-box, but lacks the C-clamp motif found in E-tail containing isoforms [79], as well as the C-terminal CtBP-binding domain [80].Additionally, it includes alternative exons in the intermediate region and isoform-specific sequences such as the LVPQ motif, as a result of the use of an alternative splice donor site at the end of exon 7 [81,82] (Figure 1A).
Out of approximately 1.2 million independent yeast transformants initially screened, we isolated close to 300 candidate prey plasmids that were sequenced and analyzed by database comparison.The most frequently obtained cDNA, representing approximately 13% of the total and including many full-length clones, corresponded to GRG5/AES.This was similar to the results obtained with TCF1, but very surprising, in light of previous claims that mouse Grg5 binds TCF1 but not TCF4 [49].To demonstrate that the TCF4-GRG5/AES interaction is specific, the TCF4 bait plasmid (pAS2-TCF4) was reintroduced into yeast cells along with pACT2 (prey "empty vector" expressing the G4AD).Conversely, the GRG5/AES prey plasmid (pACT2-GRG5) was reintroduced with pLAM control bait (encoding a fusion of laminin protein with G4BD). Figure 2 shows that yeast cells are only able to grow in the selective medium when co-transformed with both TCF4 and GRG5/AES fusion plasmids, indicating that a functional twohybrid transcription factor is produced and therefore that TCF4 interacts with GRG5/AES.
To determine if TCF4 and GRG5/AES can also interact in mammalian cells, co-immunoprecipitation studies were carried out on protein extracts of transiently transfected HEK293 cells (Figure 3).We expressed Myc-tagged TCF4 proteins either in the presence or absence of HA-epitope-tagged GRG5/AES.Full length TCF4 or the fragment containing residues 7-387 (used in the original two-hybrid screen), were precipitated from the cell lysates with a monoclonal anti-Myc antibody, and the immunoprecipitates were analyzed for the presence of both GRG5/AES and TCF4 proteins by Western blotting.GRG5/AES was readily and specifically detected in immunoprecipitates from co-expressing cells, as shown by the presence of the predicted 23kDa size protein (Figure 3A, lanes 2 and 4), but not from cells expressing only one of the proteins, confirming that there is a physical interaction between GRG5/AES and both full-length and truncated TCF4.

TCF4 130-240 is sufficient for strong interaction with GRG5/AES and other Grouchos
To map the GRG5/AES-interaction domain of TCF4, we next fused successive truncations of TCF4 to G4BD (Figure 4A) and analyzed their ability to bind GRG5/AES by yeast two-hybrid (Figure 4).We first divided the original TCF4 bait fragment into two overlapping regions, TCF4 7-260 and TCF4 130-387 (Figure 4A), both of which were found to mediate the interaction (Figure 4B, lanes 4 and 5).We next tested the overlapping fragment, TCF4 130-260 , and found that this region is sufficient to mediate specific interaction with GRG5/AES (Figure 4B, lane 6).It was also shown to be necessary, as no interaction was detected with the flanking regions: TCF4 261-387 failed to activate the reporter gene, as did TCF4 7-129 , despite its ability to interact with β-catenin (Figure 4B, lanes 7 and 8).
To gain further insight into this interaction, we next attempted to determine the minimal interacting region.Additional TCF4 truncations were thus generated in fusion with G4BD (Figure 4A) and tested by two-hybrid for adenine reporter activation (Figure 4B).Because the yeast strain AH109 is HIS-and contains an integrated copy of the HIS3 gene under the control of a Gal4-dependent promoter, we were also able to more accurately evaluate the interaction ability of TCF4 fragments.For each interaction experiment, 60 transformant colonies were Yeast cells containing both TCF4 and GRG5/AES grew in the selective media.pLAM and pACT2 plasmids were used as bait and prey negative controls respectively, while the β-catenin prey plasmid constitutes a positive control for the interaction assays.doi: 10.1371/journal.pone.0067694.g002therefore assayed for growth on dropout medium plates lacking histidine and containing increasing doses of 3-aminotriazol (3-AT), an inhibitor of HIS3.With increasing doses of 3-AT, more and more of the enzyme is inhibited, selecting for stronger gene activation and more stable interactions (Figure 4C).
Because the 253-FRQ/HPY-257 motif in TCFs has been suggested to be involved in binding to Groucho family members [48,83,84], we tested a TCF4 fragment lacking this pentapetide (TCF4 130-240 ).As shown in Figure 4B and 4C, transformant yeast cells grew well, both in medium lacking adenine and in medium lacking histidine, even at the highest 3-AT concentration tested (30mM), demonstrating that the motif is not required for this specific interaction.In contrast, TCF4 130-220 completely lacked GRG5/AES-binding (Figure 4B), showing that amino acids 220-240 are required for the interaction.We further determined that fragments 151-260 and 172-260 are still able to interact with GRG5/AES and activate the adenine reporter (Figure 4B).However, the binding affinity was reduced, as shown by the limited growth of transformants in medium lacking histidine at high 3-AT doses (Figure 4C).Additionally, TCF4 172-240 showed dramatically reduced GRG5/ AES-binding and the interaction was only observed under low stringency 3-AT selection (Figure 4C).Altogether, these data indicate that the GRG5/AES-interacting domain of TCF4 resides within amino acids 130-240 and that this region is sufficient for a strong TCF4-GRG5/AES interaction.Importantly, strong interaction of the TCF4 130-240 fragment was also observed with Grg1-S and GRG4-QGP (compare Figures 4B and 5B), suggesting it defines a general core region for TCF4-GRG interactions.

The LVPQ motif is critical for TCF4's interaction with GRG5/AES but not with other Grouchos
The 20-amino acid segment of TCF4 we determined to be essential for GRG5/AES-binding (amino acids 221-240) encompasses the LVPQ motif, which is present only in some TCF4 splice variants, and invariantly found in TCF3 [81,82].Because this motif has been implicated in repression mediated by these particular TCFs [81,82,85], we investigated whether it plays a role in TCF4-GRG interactions, by testing the effects of deleting it from TCF4 130-240 in yeast two-hybrid assays (Figure 5).As shown in Figure 5B, absence of the LVPQ motif completely abrogates interaction with GRG5/AES, implying that it plays a critical role in TCF4-GRG5/AES binding.However, this cannot be generalized to other GRGs, as the LVPQ deletion had no effect on TCF4's interaction with Grg1-S and only mildly hampered its binding to GRG4-QGP, a "truncated" GRG4 that, like GRG5/AES or Grg1-S, includes only the Q and GP domains (Figure 5B).These results show that Groucho proteins bind differently to TCF4 and suggest its interaction with GRG5/AES might be modulated by alternative splicing.

GRG5/AES represses Wnt signaling in human cells and cooperates both with dnTCF4 and Grg1-L for efficient transcriptional repression
GRG5/AES has variously been reported to act either as a dominant negative, that interferes with the function of the long GRGs [49,62] or, like these, as a co-repressor of multiple transcription factors, including members of the TCF family [66][67][68]86].Together with our observation of an interaction between GRG5/AES and TCF4 this prompted us to examine the ability of GRG5/AES to counteract TCF-β-catenin-mediated transcription (Figure 6).As previously observed in other cell lines [71], co-transfection of HEK293 cell line with a TCF-Luc construct (containing three TCF consensus binding sites upstream of the firefly luciferase cDNA) together with β-catenin resulted in a remarkable increase of luciferase activity compared to TCF-Luc expression alone (Figure 6A).Addition of increasing amounts of GRG5/AES plasmid reduced luciferase reporter activation in a dose-dependent manner (Figure 6A).At the highest dose tested, GRG5/AES suppressed TCF promoter activity 12-fold, similar to the repressive effect of a dominant negative form of TCF4, a wellaccepted antagonist of TCF-β-catenin signaling [18].As expected, co-transfection with the control reporter TCF-Luc* (with mutated TCF binding sites) resulted in no alteration of luciferase activity.
To further evaluate the putative repressive function of GRG5/ AES, we compared its activity to those of other Groucho-family members, broadly accepted as general co-repressors of the TCF/LEF-family of proteins: GRG4, Grg1-L and Grg1-S.As shown in Figure 6B, GRG5/AES was found to repress TCF-βcatenin-dependent transcription in human cells similarly to other family members tested -and indeed more efficiently than GRG4 (see also Figure S1).Surprisingly, in this specific cell line, the "long Grg1" (Grg1-L) was not capable of repressing transactivation of the TCF-Luc reporter gene (Figure 6B).This might have resulted from a lower protein expression level in the Figure 4. Mapping of TCF4's GRG5/AES-interacting region using yeast two hybrid assays.A) Schematic diagram of the Gal4BD-TCF4 constructs tested for interaction with GRG5/AES.B) GRG5/AES-interacting TCF4 fragments activate an ADE2 reporter gene, conferring to yeast transformants the ability to grow in selective medium lacking tryptophan, leucine and adenine.C) Transformants were also tested for activation of a HIS3 reporter, as determined by growth in medium lacking histidine and with increasing doses of 3-aminotriazol (3AT).As defined by both reporter assays and summarized on the right panel of the diagrams, the region encompassing residues 130-240 is sufficient for the interaction, with residues 220-240 being strictly necessary.Two previously described fragments were included as internal controls: dnTCF4, that lacks the N terminal β-catenin binding domain [13]; and TCF1 176-359 [49]; n.d.: not done.doi: 10.1371/journal.pone.0067694.g004cells (Figure S1), or rather be another indication of the complex and context-dependent nature of Groucho proteins' functions.To determine whether GRG5/AES is actively repressing TCFdependent transcription as has been demonstrated for "long Grouchos" [87], or rather functioning by blocking the formation of TCF-β-catenin active complexes, we tested the ability of GRG5/AES to interfere with basal transcriptional activity when tethered to DNA.Gal4-fusion constructs encoding either GRG5/AES or full-length Grg1 (Grg1-L) were therefore transfected into HEK293 cells and their ability to repress the activity of a strong, Gal4-responsive, promoter was compared.As shown in Figure S2, GRG5/AES is able to repress basal transcription to the same degree as Grg1-L, implying that GRG5/AES harbors a functional repression domain comparable to those of the long Grouchos.These results demonstrate that GRG5/AES can effectively repress endogenous TCFs and down-regulate TCF-β-catenin-mediated transcription.
Because HEK293 cells express all TCF/LEF-family members [62], we could not determine whether this repression was Co-transformant yeast cells were selected in medium lacking tryptophan and leucine and assayed for the indicated interactions using the adenine reporter gene (in Trp-Leu-Ade-medium).The same 111-amino acid region (residues 130-240) is sufficient for strong interaction with all three GRGs tested, and amino acids 221-240 are required for it, whereas removal of the LVPQ motif abrogates binding to GRG5/AES, but not the other GRGs, although it has some effect on the TCF4-GRG4 interaction.The ACT2 plasmid was used as a negative control.doi: 10.1371/journal.pone.0067694.g005mediated through TCF4.To address this limitation, we cotransfected HEK293 cells with both GRG5/AES and dnTCF4 expression plasmids, to test whether they act cooperatively or independently in repressing TCF-Luc reporter transactivation.As shown in Figure 6C, this resulted in 22.5-fold transcriptional repression, significantly higher than the 6.3-fold additive effect of GRG5/AES-and dnTCF4-mediated repressions.This implies that GRG5/AES and dnTCF4 indeed act cooperatively in transcription repression, suggesting that the TCF4-GRG5/AES interaction contributes to GRG5/AES's repressor function.Furthermore, we asked whether GRG5/AES could similarly cooperate with Grg1-L, using the same reporter gene Figure 6.GRG5/AES represses TCF-β-catenin mediated transcription and co-operates with both dnTCF4 and Grg1-L for transcriptional repression.HEK293 cells were transfected with either the reporter construct TCF-Luc (containing three TCF consensus binding sites upstream of the firefly luciferase cDNA), or the control reporter TCF*-Luc (with mutations in the TCFbinding sites) and the indicated combinations of expression plasmids.A) Addition of GRG5/AES plasmid decreases the luciferase activity induced by transfected βcat and endogenous TCFs in a dose dependent manner.At the highest dose (1.2 μg/μl), GRG5/ AES-mediated repression is comparable to the effect of dominant negative TCF4 (dnTCF4), a previous established repressor of the canonical Wnt pathway [13].B) Repression of TCF-β-catenin mediated transcription by various Groucho-family proteins.In contrast to "long" Groucho Grg1 (Grg1-L), GRG5/AES, GRG4 and Grg1-S (a short GRG1 isoform [63]) effectively repressed TCF-dependent luciferase activity.1.2 µg were used for transfection of each Groucho expression plasmid.C) Co-transfection of GRG5/AES with either dnTCF4 or Grg1L (0.4µg of each plasmid) results in higher repression of TCF-β-catenin dependent luciferase activity then transfections with each effector plasmid individually.β-galactosidase assays were performed as an internal control for transfection efficiency.All the assays were done in triplicate in three independent experiments.doi: 10.1371/journal.pone.0067694.g006assay (Figure 6C).Interestingly, although Grg1-L alone was unable to repress TCF-mediated transactivation in this assay, its co-transfection with GRG5/AES significantly increased the latter's repression activity, showing that GRG5/AES can cooperate with the "long Grouchos" in downregulating Wnt/βcatenin signaling.

GRG5/AES counteracts β-catenin dependent dorsal specification in zebrafish
In zebrafish, as in Xenopus, nuclear accumulation of βcatenin in the early embryo is required for activation of dorsalizing genes and establishment of the dorsal-ventral (DV) axis [56,88,89].Furthermore, both β-catenin and Wnt overexpression result in embryo dorsalization and secondary axis formation [14,20,54,[90][91][92].Our observations that GRG5/AES interacts with TCF4 and behaves as a consistent repressor of TCF-β-catenin dependent transcription in human cells therefore raised the possibility that GRG5/AES may also regulate DV patterning during early zebrafish embryogenesis.
To test for this possibility, we started by microinjecting grg5/aes mRNA into 1-cell stage embryos (Figure 7).Morphologically, the vast majority of injected embryos were apparently normal at 10hpf, with the characteristic thickening of the neural plate along the dorsal side.However, some embryos displayed reduction of the dorsal side (Figure 7B).In contrast, and as previously reported, overexpression of an active βcatenin (T41A) resulted in dorsalized embryos, easily recognized by an animal-vegetal elongated morphology and, in some embryos, formation of ectopic dorsal axes [54] (Figure 7C, D).To further characterize the effect of grg5/aes microinjection, we investigated the expression pattern of the dorsal marker chordin by performing in situ hybridizations at mid-gastrulation in grg5/aes-injected embryos.As shown in Figure 7, chordin expression was clearly reduced, suggesting that GRG5/AES indeed influences DV axis patterning by inhibition of dorsal-specific gene expression.
Next, we sought to evaluate GRG5/AES's ability to antagonize β-catenin dorsalizing activity in zebrafish embryos.When we microinjected β-catenin into a single blastomere of 8cell stage embryos, severe disruption of DV patterning occurred.Forty-five percent of injected embryos were dead at 24hpf, likely as a consequence of extreme dorsalization (Figure 8), while 62% of the remaining embryos displayed axis duplication.While a complete secondary axis was obtained on a few occasions, partial axis duplications were more common.Nevertheless, to avoid ambiguity, we grouped both types in the same class, scoring embryos as having double-axis whenever there was notochord duplication (Figure 8B).In contrast, coinjection of β-catenin and grg5/aes mRNAs into a single blastomere of 8-cell stage embryos resulted in a much higher survival rate (Figure 8A).The number of embryos dead at 24 hours was reduced 9 fold, and dorsalized phenotypes were more moderate compared with those produced by β-catenin injection alone (data not shown).Furthermore, there was a consistent decrease in the number of embryos with a duplicated axis, indicating that GRG5/AES antagonizes βcatenin signaling in zebrafish embryos.Comparable results were observed in embryos co-expressing β-catenin and dnTCF4 (Figure 8A), a truncated version of TCF4 that lacks the β-catenin binding domain and is similar to constructs previously shown to inhibit axis formation in both Xenopus and zebrafish [13,14].
The ability of GRG5/AES to counteract β-catenin in vivo was further explored by examining the expression of markers for the DV character of the injected embryos (Figure 9).During gastrulation, chordin is expressed in the prospective dorsal region of the embryo [75] and, by 70% epiboly, it encompasses approximately half of the circumference of the margin (Figure 9A).Conversely, gata2 expression is restricted to the ventral region (Figure 9I).As expected, dorsalized β-catenin-injected embryos displayed completely circular chordin expression in all of the tested embryos (Figure 9B,Q) and, consistently, a complete suppression of gata2 (Figure 9J, R).However, when grg5/aes mRNA was co-injected, β-catenin-induced dorsalization was efficiently rescued as shown by restoration of normal marker gene expression patterns.The embryos were placed into four classes based on DV extent of marker gene expression (Figure 9Q).All the co-injected embryos tested for chordin showed reduction of the circular expression induced by β-catenin.In 48% of the embryos, the chordin expression domain was restricted to less than 40% of the margin cells or was completely absent (class 1, Figure 9C, Q), indicating loss of dorsal-specific gene expression caused by GRG5/AES.In 33% of the embryos tested, the normal pattern of chordin expression was restored (class 2, Figure 9D, Q); 19% still showed ventrally expanded expression when compared to uninjected controls, but to a lesser degree than embryos injected only with β-catenin (class 3, Figure 9E, Q).GRG5/AES was also able to rescue gata2 expression in 68% of the total embryos, either fully (41%) or partially (27%) (Figure 9I-M, R).Similar results were observed in control experiments of βcatenin and dntcf4 co-injections, with clear suppression of βcatenin induced dorsalization and strong evidence of a ventralizing effect (Figure 9F-H, N-P).

GRG5/AES antagonizes Wnt/β-catenin signaling during early zebrafish embryogenesis
Sp5l is a direct target of canonical Wnt signaling whose expression reflects Wnt involvement in mesoderm and neuroectoderm patterning during gastrulation [93,94].We therefore predicted that, if indeed GRG5/AES antagonizes βcatenin activity, it should be able to rescue, at least partially, sp5l upregulation induced by β-catenin.As expected, injection of activated β-catenin mRNA resulted in greatly expanded sp5l expression at the animal pole of 70% epiboly embryos (Figure 10B) and this overexpression was clearly reduced upon coinjection of grg5/aes mRNA.The embryos analyzed by in situ hybridization were classified as above, in different classes of rescue efficiency and a representative example of each class is shown (Figure 10C-E).In 75% of the co-injected embryos, GRG5/AES repressed β-catenin-induced overexpression, from which 45% were partially rescued, while in 27% the normal expression pattern was completely restored.Co-injection of dntcf4 control mRNA similarly suppressed or completely abolished sp5l overexpression (Figure 10I).

TCF4 binds to GRG5/AES in yeast and human cells
Despite intensive work to describe how the TCF-β-catenin pathway operates in a variety of developmental processes, a detailed understanding of the specific molecular mechanisms that fine-tune basic signaling is still lacking.Although β-catenin regulation has been extensively studied, much less is known about the regulation of TCFs and how individual members of this family dictate Wnt signaling output in different cell types.We were particularly interested in TCF4 because it plays pivotal roles in a variety of processes, including central-nervous system patterning [95,96], skin homeostasis and wound repair [97] and particularly normal development and pathology of the intestine [17,18,29,98].We therefore performed a yeast twohybrid screen for new putative TCF4-interacting proteins and demonstrated that TCF4 physically interacts with GRG5/AES, a distinct member of the GRG family that exemplifies a subclass of "truncated" proteins, commonly referred to as the "short Grouchos".This observation was unexpected, because it had been previously reported that mouse Grg5 did not interact directly with TCF4, although it did with other TCFs [49].Interestingly, in the same study, mouse Grg5 was shown to interact with Xenopus TCF3 but not with the orthologous mouse protein.This might be due to sequence divergence among different species, affecting binding specificities.However, there is a growing body of evidence for the existence of multiple isoforms of both GRG5/AES and TCF4, resulting from extensive alternative splicing and/or alternative promoter usage [79,82,99,100].As would be expected from first principles, and has been confirmed by several reports, this structural variation results in specificity in protein-protein interactions, with profound functional implications [81,99,101].It is therefore possible that apparently conflicting results stem from the use of different isoforms, as it is not always clear in the literature which particular one was used.At any rate, our work clearly demonstrates the existence of an interaction between TCF4 and GRG5/AES, confirmed by coimmunoprecipitation and transcription repression assays in human cells.This result is particularly important because it was obtained in an unbiased way -in a cDNA library screen for TCF4 interacting proteins where GRG5/AES was the most frequently obtained candidate -and not through candidate testing.Additional support for an in vivo interaction between TCF4 and GRG5/AES comes from the overlapping expression patterns of these two proteins observed in several vertebrates [86,[102][103][104][105] and the evidence for their critical roles during pituitary gland development [102].

Isoform-specific regions of TCF4 are critical for differential binding to GRGs
All TCF/LEF family members share two highly conserved structural features: a 55-amino acid N-terminal domain responsible for β-catenin binding and a centrally located highmobility-group (HMG) box that mediates DNA-sequence recognition and bending for an architecturally favourable context for transcription control [2][3][4] (Figure 1A).Between these two domains is one of the most divergent regions among the family members (and even orthologues), to which various roles involving protein-protein interactions have been assigned [43,46,106,107].GRG binding has also been broadly assigned to this region in previous interaction studies for TCF1, Xenopus TCF3 and LEF1, though recently a second (and weaker) site in LEF1's HMG box has been described as mediating interaction with Drosophila Groucho [69].No interaction mappings were previously reported for TCF4 and, due to the low sequence conservation in this region, results previously obtained with other TCFs are not easily transposable to TCF4.We performed an exhaustive mapping of the GRG5/AES-binding region in TCF4 by yeast two-hybrid and determined that a 111-amino acid stretch (residues , located between the β-catenin binding domain and the HMG box, is sufficient for the TCF4-GRG5/AES interaction (Figure 4).Interestingly, this region largely corresponds to an alternative-splicing prone region, encompassing exons 4-6, only present in some TCF4 variants [99,101].
Although this fragment displays some similarity to the GRGbinding regions described for other TCFs (Figure S3), no highly conserved sequences are recognizable, in contrast to what has been reported for other GRG partners, which bind through specific and short motifs [108].Indeed, it had been speculated that the pentapeptide FRQ/HPY present in TCF/LEF-family members might be involved in the interaction with GRGs [48,83] because it is somewhat similar to the motifs found in other GRG partners, such as Hairy-related proteins (WRPW) [109], Runt (VWRPY) [110] and HKb (FRPW) [84], but we have demonstrated that this motif is not required for binding between TCF4 and GRG5/AES (Figure 4).Equally dispensable is the sequence corresponding to a short LEF1 stretch proposed to mediate Groucho binding (Figure S3) [69].
Our mapping refinement assays further led us to identify both convergent and divergent requirements for TCF4 binding by various GRGs.Indeed, the 111-amino acid core region for the TCF4-GRG5/AES interaction was also shown to be sufficient for TCF4 binding to GRG4 and GRG1.Similarly, the last 20 residues (221 to 240) of this region were found to be critical for all three interactions, as their removal abrogated them.On the other hand, GRG5/AES, but not GRG1 or GRG4 binding depends on the LVPQ motif, which is located in the 221-240 region and, within the TCF family, is only present in TCF4 as a result of specific alternative splicing, and invariantly in TCF3 (Figure S3).The LVPQ motif had previously been implicated in repression mediated by these particular TCFs [81,82], but its role in modulating TCF-GRG interactions was unknown.Our observations therefore support the idea that binding of GRG5/AES depends on the expressed splice TCF4 variant and that contradictory results may be explained by different splice variants of TCF4 used in different reports.More importantly, it strengthens the idea that multiple TCF4 isoforms can differentially influence tissue-specific Wnt responses.
Nevertheless, it should be noted that an additional segment (amino acids 130-171) is also required for binding.This might indicate that TCF-GRG interactions, rather than being mediated by a single short motif, involve multiple cooperative binding events.Alternatively, it is possible that more important than specific sequences is a favorable conformation of the GRG-binding region to which the various segments found to be required contribute.Consistent with this possibility, though GRG family members bind the same TCF/LEF proteins, different specific requirements seem to underlie individual interactions, the only obvious feature shared by the GRGbinding regions in TCFs being a high proline content (Figure S4).Interestingly, similar proline-rich regions have been identified in other transcription factors that interact with Groucho-related proteins and are also a feature of several transcription repressors [65,67,[111][112][113].The significance of this observation is unknown, but it might point to a requirement for a disordered structure of the TLE/GRG-binding regions.

GRG5/AES represses TCF-β-catenin-mediated transcription in human cells and synergizes with Grg1-L for efficient repression
The long Grouchos are generally recognized as dedicated co-repressors of a wide range of DNA-binding factors, including the TCF/LEF family [59,108,114,115].The highly conserved Qdomain mediates interaction with TCF/LEF transcription factors, as well as tetramerization, which is thought to be a prerequisite for repressor function [116][117][118].Furthermore, the GP domain interacts with histone deacetylases (HDACs) in repressor complexes, promoting a more compact and transcriptionally inactive chromatin [47,62,119,120].As for the short Grouchos, and particularly GRG5/AES, the situation is more complex.Similarly to the long proteins, human GRG5/AES was originally described as a transcriptional corepressor [67] but subsequent reports have provided evidence that it can act either as a co-repressor [67][68][69]86,121,122] or as a "de-repressor" [62,116,117].Many of the interactions between the long GRGs and different transcription factors are mediated by their C-terminal WD domains (Figure 1) and in these cases a mechanism for de-repression can easily be postulated: a short Groucho, incapable of interacting with the co-repressor protein but still able to bind other Groucho family members, might "poison" tetramers, by rending them less proficient at forming repression complexes [65,112,113].In the case of TCFs, however, the interaction is mediated by the Q domain [62], present in both the short and long Grouchos, making this mechanism unlikely.Initial reports of GRG5/AES as a TCF de-repressor therefore required an alternative explanation: the divergence between the GP domains of the long GRGs and those of GRG5/AES make the latter unable to interact with HDACs 1 and 3.However, this explanation is still not satisfactory for two reasons.First, lack of HDAC binding does not prevent GRG5/AES from acting as a co-repressor for other transcription factors [66][67][68]86], possibly through interaction with the basal transcription machinery.Second, structural studies have suggested that GRG-mediated repression of TCF-β-catenin dependent transcription is based on competition for TCF binding -at which GRG5/AES is fully proficient [123].
In this study, we demonstrate that GRG5/AES consistently behaves as a TCF co-repressor, in both human cells and zebrafish embryos.More specifically, we show that, at least in HEK293 cells, GRG5/AES is able to strongly repress TCF-βcatenin transcription in a dose-dependent manner.Additionally, we observed a similar, though less pronounced, repressive effect of GRG5/AES in DLD-1, a colorectal tumor cell line which displays constitutive TCF4-β-catenin dependent transcription due to APC mutations (data not shown).This is in agreement with a study published during the preparation of this manuscript, showing that GRG5/AES is also able to repress LEF1 in human cell lines [69], but in stark contrast to what was observed in another recent study, in which it de-repressed LEF1-mediated activity [124].Moreover, our work shows that GRG5/AES is also able to synergize with long Grouchos for strong TCF-mediated repression, thus providing additional support to the growing idea that GRG5/AES is not a general GRG antagonist and highlighting the importance of taking caution with generalizations about a protein's function and biological effects.
How can these apparently conflicting observations regarding GRG5/AES's role in repressing TCF-β-catenin dependent transcription be reconciled?Our studies strongly suggest that GRG5/AES's function is critically dependent on the specific TCF/LEF isoforms present in the cell.For example, as previously noted, the TCF4 region involved in GRG5/AES binding is affected by alternative splicing, which could, in some cells, give rise to isoforms that do not interact with GRG5/AES.A similar mechanism is behind TCF1's role as a Wnt antagonist in the intestine: because the prevailing TCF1 isoform in that tissue lacks a β-catenin interacting region, it can only form Wntsignaling repressing complexes and therefore acts as an intrinsic dominant-negative [125].The relative abundances of other GRG interactors might also be important.Indeed, some of them are well known to interact only with the long Grouchos, and it is also possible that others have a preference for the short ones, like GRG5/AES.The balance of all these partners in a cell is therefore likely to affect GRG5/AES's availability to functionally interact with TCFs.Interestingly, the same process could also limit the availability of one or another of the long Grouchos, implying that, under specific circumstances, some of these might also behave as dominant negatives and potentially explaining our observation that Grg1-L, by itself, was unable to act as a TCF co-repressor in HEK293 cells (Figure 6B,C).
In both cases -differences in TCF isoforms and differences in other partners -a de-repressor role for GRG5/AES hinges upon the notion of "tetramer-poisoning" by a functionally impaired protein.This is consistent with the notion that homotetramerization is essential for efficient repression [121,126], but opposed by our findings of cooperation between GRG5/AES and Grg1-L and at least by another study showing that a C. elegans GRG5/AES-related protein was found to cooperate with a long Groucho orthologue in co-repression [122].Alternatively, GRG5/AES's subcellular distribution might be a crucial determinant of its apparently inconsistent behavior.As it varies in different cell contexts [124], it is conceivable that a predominantly cytoplasmic distribution could contribute to retain other GRGs in the cytoplasm, thus limiting their repressive activity.

GRG5/AES represses dorsal cell fate in zebrafish and antagonizes Wnt signaling during DV patterning
Analysis of both mutant lines and overexpressing transgenics have demonstrated the indispensable role of canonical Wnt signaling in establishing the DV axis during early zebrafish embryogenesis [14,54,57,88,[127][128][129][130].Asymmetric nuclear accumulation of maternal β-catenin demarcates the specification of dorsal identity in the early embryo [56].Parallel to what is observed in Xenopus, overexpression of β-catenin leads to strongly dorsalized embryos and formation of double axes [54,89].In a similar way, and with the intrinsic limitations of all overexpression experiments, our results suggest that GRG5/AES might modulate DV patterning in the early embryo: first, GRG5/AES overexpression inhibits dorsal cell fate, effectively reducing endogenous expression of early dorsalspecific genes like chordin.Second, it can antagonize and rescue the most severe β-catenin-induced dorsalization phenotypes, including double-axis formation and expansion of dorsal domains (accompanied by a reciprocal reduction of ventral regions).Third, similar to a dominant negative version of TCF4, it downregulates expression of direct Wnt target genes and interferes with axis specification, most likely through TCF-GRG repressor complexes.These results consistently suggest that GRG5/AES can act as a repressor of Wnt signaling during early zebrafish development and concur with previous evidence for short Grouchos' repressive activity in vivo.For example, in sea urchin, fragments of LvGroucho containing only the Q and GP domains (like GRG5/AES) or even the Q domain alone, were shown to inhibit TCF-β-cateninmediated endomesoderm specification along the animalvegetal axis [64].Also, a naturally truncated version of GRG1/ TLE1 was shown to counteract β-catenin in Xenopus [63] and, more recently, a GRG5/AES-related protein was found to act as a co-repressor to control developmental decisions in C. elegans [122].
Considering its endogenous expression in several species [86,108,[131][132][133][134], including zebrafish (Costa A. M. S. et al, manuscript in preparation), we propose that GRG5/AES orthologues act endogenously as negative regulators of Wnt signaling during early patterning of the DV axis in the zebrafish embryo.It would be interesting, for example, to analyze if there is an upward gradient of TCF-GRG5/AES complexes along the DV axis, that mirrors the reported downward TCF-β-catenin gradient [88].Also, given that some Tcf4 variants are maternally and zygotically expressed during zebrafish embryogenesis [104], similarly to what we found for zebrafish Grg5 (Costa A. M. S. et al, manuscript in preparation), it is conceivable that specific Tcf4 isoforms may play a role in DV patterning.In fact, in Xenopus it was demonstrated that depletion of maternal XTcf4 resulted in reduction of chordin expression and ventralized embryos consistent with loss of organizer gene expression [135].However, such roles for TCF4 have been difficult to definitively establish, as no phenotype was obvious in the available TCF4 mutants or knockdowns studies in several species [17,19,136], perhaps due to partial functional redundancy with other TCF-family members.In this context, a Tcf4 zebrafish mutant lacking the LVPQ domain, which we found to be specifically needed for GRG5/AES-binding, could provide an invaluable tool to future studies of the physiological role of TCF4-GRG5 interaction during DV patterning and to gain further insight into how individual TCF4 splice variants may be involved in this process.An additional question is whether GRG5/AES plays an important role in the intestinal physiology of vertebrates.Since TCF4 is the most prominently expressed TCF/LEF family member in the gut, our characterization of the TCF4-GRG5/AES interaction may have important implications for normal intestinal development and tumorigenesis.

Figure 1 .
Figure 1.Schematic representation of the mammalian TCF4 and GRG domains organization.A) The human TCF4 gene consists of 17 exons (top), some of which are subject to alternative splicing (black exons).The TCF4 variant used (Ex1-17) includes isoform specific sequences such as the LVPQ domain due to the use of an alternative splice donor site at the end of exon 7 (arrowhead).A TCF4 fragment lacking the C terminus (a.a.s 7-387), was fused to Gal4's DNA-binding domain (Gal4-BD) and used as bait in a two hybrid screen.B) The GRG family comprises two distinct classes of proteins, based on their domain constitution: the "long Grouchos", GRG/TLE 1-4, consisting of five different domains (Q, GP, CcN, SP and WD40); and the "short Grouchos", such as the alternative splice-product Grg1-S or the GRG5/AES subfamily members, which contain only the Q and GP domains.

Figure 2 .
Figure 2. Physical interaction of TCF4 and GRG5/AES in yeast cells.The AH109 yeast strain was co-transformed with the indicated constructs (central panel).Transformants were selected in medium lacking tryptophan and leucine (left panel) and activation of a Gal4-dependent ADE2 gene was examined by growth on selective plates additionally lacking adenine (right panel).Yeast cells containing both TCF4 and GRG5/AES grew in the selective media.pLAM and pACT2 plasmids were used as bait and prey negative controls respectively, while the β-catenin prey plasmid constitutes a positive control for the interaction assays.

Figure 3 .
Figure 3. Physical interaction of TCF4 and GRG5/AES in human cells.A) Lysates of HEK293 cells transiently transfected with the indicated expression plasmids were subject to immunoprecipitation (IP) with anti-Myc antibody and the immunoprecipitated proteins were analyzed by Western blot (WB) with anti-HA and anti-Myc antibodies.Co-precipitation of HA-tagged GRG5/AES was detected only in the lysates of cells co-transfected with Myc-TCF4 7-387 (lane 2) or Myc-TCF4FL (full-length) (lane 4).B) 10% of each total cell lysate was loaded for WB with either anti-HA or anti-Myc antibodies.The asterisks indicate nonspecific binding to immunoglobulin light (*) and heavy (*´) chains.doi: 10.1371/journal.pone.0067694.g003

Figure 5 .
Figure 5. Convergent and divergent sequence requirements for TCF4 binding to different GRGs.A) Schematic view of TCF4 (left) and GRGs (right) fusion constructs used in yeast-two hybrid assays.B) Two-hybrid analysis of the importance of amino acids 130-240, 221-240 and the LVPQ motif for the interactions between TCF4 and GRGs.Co-transformant yeast cells were selected in medium lacking tryptophan and leucine and assayed for the indicated interactions using the adenine reporter gene (in Trp-Leu-Ade-medium).The same 111-amino acid region (residues 130-240) is sufficient for strong interaction with all three GRGs tested, and amino acids 221-240 are required for it, whereas removal of the LVPQ motif abrogates binding to GRG5/AES, but not the other GRGs, although it has some effect on the TCF4-GRG4 interaction.The ACT2 plasmid was used as a negative control.

Figure 8 .
Figure 8. GRG5/AES reduces both mortality and axis duplication in β-catenin-overexpressing embryos.A) One blastomere of 8-cell stage zebrafish embryos was injected with the indicated combinations of β-catenin (50 pg), grg5/aes (500 pg) and dntcf4 (500 pg) mRNAs.Partial and complete axis duplications were grouped to avoid ambiguity.The total number of scored embryos (n) is indicated at the lower part of the panel.B) Example of 24hpf embryo showing complete axis duplication (induced by β-catenin injection), with two discernible heads (arrows), notochords (asterisks) and tails (arrowheads).Dorsal view, anterior to the left.C) Example of 24hpf embryo injected with β-catenin showing extreme dorsalization.doi: 10.1371/journal.pone.0067694.g008

Figure 9 .
Figure 9. GRG5/AES antagonizes β-catenin in dorsal-ventral patterning during early zebrafish embryogenesis.Wholemount in situ hybridizations for chordin (A-H) and gata2 (I-P) of double-injection experiments (columns).Animal pole views, dorsal to the right, of 70% epiboly stage-embryos.Injection of β-catenin strongly dorsalizes embryos, as shown by an expanded and circumferential expression of the dorsal-specific gene chordin (B) and a total absence of the ventral-specific gene gata2 (J).Coinjection of grg5/aes (C-E; K-M) rescues β-catenin-induced dorsalization and restores normal expression domains for both marker genes.dntcf4 co-injections (F-H; N-P) were used as a control for β-catenin antagonism.Co-injected embryos were grouped in four classes of marker gene expression: C1 (C, F) -abnormal expression consistent with embryo ventralization; C2 (D, G; K,O) − restoration of normal expression patterns (A,I); C3 (E, H; L, P) -partial reversion of β-catenin-induced abnormalities; C4 (M) − abnormal expression consistent with embryo dorsalization.Q-R: distribution of the embryos in the different classes of chordin (Q) and gata2 (R) expression.GRG5/AES antagonizes β-catenin effects on chordin and gata2 in 100% and 68% of the embryos respectively.n: number of embryos tested.doi: 10.1371/journal.pone.0067694.g009

Figure 10 .
Figure 10.GRG5/AES represses Wnt signaling during early zebrafish embryogenesis.Whole-mount in situ hybridizations for sp5l, a direct target of canonical Wnt signaling.Lateral view, dorsal to the right, of 70% epiboly stage-embryos.Injection of β-catenin induces broad ectopic domains of sp5l expression (B), which are reversed by co-injection of grg5/aes (C-E) or dntcf4 (F-H).Expression patterns in co-injected embryos were grouped in four classes, as above.AES reversed β-catenin-induced misexpression in 75% of the embryos (I). n: number of embryos tested.doi: 10.1371/journal.pone.0067694.g010