GNL3L inhibits activity of estrogen-related receptor gamma by competing for coactivator binding.

Guanine nucleotide binding protein-like 3 (GNL3L) is the closest homologue of a stem cell-enriched factor nucleostemin in vertebrates. They share the same yeast orthologue, Grn1p, but only GNL3L can rescue the growth-deficient phenotype in Grn1-null yeasts. To determine the unique function of GNL3L, we identified estrogen-related receptor γ (ERRγ) as a GNL3L-specific binding protein. GNL3L and ERRγ are coexpressed in the eye, kidney and muscle, and co-reside in the nucleoplasm. The interaction between GNL3L and ERRγ requires the intermediate domain of GNL3L and the AF2-domain of ERRγ. Gain-of- and loss-of-function experiments show that GNL3L can inhibit the transcriptional activities of ERR genes in a cell-based reporter system, which does not require the nucleolar localization of GNL3L. We further demonstrate that GNL3L is able to reduce the steroid receptor coactivator (SRC) binding and the SRC-mediated transcriptional coactivation of ERRγ. This work reveals a novel mechanism that negatively regulates the transcriptional function of ERRγ by GNL3L through coactivator competition.


Introduction
Nucleostemin and its homologues, guanine nucleotide binding protein-like 3 (GNL3L) and Ngp-1 (hereafter referred to as Ngp1), constitute a subfamily of GTP-binding proteins featured by their nucleolar distribution and a unique domain of circularly permuted GTP-binding motifs, where the G4 motif is located N-terminally to the G1, G2, and G3 motifs (Daigle et al., 2002;Leipe et al., 2002). Nucleostemin is enriched in the embryonic, mesenchymal and neural stem cells, in adult testes and several types of human cancers (Baddoo et al., 2003;Kafienah et al., 2006;Tsai and McKay, 2002). It plays a role in maintaining the continuous proliferation of neural stem cells (Tsai and McKay, 2002) and in regulating the protein stability of telomeric-repeat binding factor 1 (TRF1) (Zhu et al., 2006). Targeted deletion of nucleostemin leads to early embryonic lethality in homozygous nucleostemin-null embryos (Beekman et al., 2006;Zhu et al., 2006) and to premature senescence of heterozygous nucleostemin-null mouse embryonic fibroblast cells (Zhu et al., 2006).
Phylogenetically, nucleostemin is most closely related to GNL3L in vertebrates. They share the same yeast orthologue: Grn1p in Schizosaccharomyces pombe and Nug1p in Saccharomyces cerevisiae. Grn1p is involved in the processing of 35S pre-ribosomal RNA (rRNA), the nuclear export of Rpl25, and the maintenance of cell growth (Du et al., 2006). Mutation of Nug1 inhibits the export of 60S subunit from the nucleolus (Du et al., 2006;Kallstrom et al., 2003). Although the yeast orthologue of nucleostemin and GNL3L displays general activities in growth and ribosome biogenesis, rodent nucleostemin and GNL3L are distinctively expressed in different tissues. Furthermore, only human GNL3L, but not nucleostemin, can rescue the Grn1-deficient growth phenotype in fission yeasts (Du et al., 2006). These results suggest that nucleostemin and GNL3L have evolved specific properties in vertebrates, and become functionally diverged from each other and from Grn1p. By comparison, GNL3L retains more characteristics of Grn1p than does nucleostemin.
GNL3L bears 28% identity and 39% similarity to nucleostemin in mice. Very little is known about its function in vertebrates. To delineate the distinct activity of GNL3L, we looked for proteins that interact with GNL3L but not with nucleostemin. We first identified estrogen-related receptor ␥ (ERR␥) as a GNL3L-binding protein by a yeast two-hybrid screen, and confirmed this interaction by affinity-binding and coimmunoprecipitation assays. ERR␥ belongs to a subfamily of the nuclear receptor superfamily. The ERR gene family consists of three members, ERR␣, ERR␤, and ERR␥, that most resemble estrogen receptor ␣ (ER␣). Like ER␣, the ERR proteins contain functionally separable structures that include an AF1 domain (or A/B region), a DNA-binding domain (DBD, or C region) with two zinc fingers, a hinge region (D), and a ligand-binding domain (LBD, or E/F region) with an AF2 domain at the C-terminal end. The LBD is involved in ligand binding, receptor dimerization, and coactivator binding. The AF2 domain is required for the ligand-dependent activation function. ERR␣ and ERR␥ are found in the brain, muscle, Guanine nucleotide binding protein-like 3 (GNL3L) is the closest homologue of a stem cell-enriched factor nucleostemin in vertebrates. They share the same yeast orthologue, Grn1p, but only GNL3L can rescue the growthdeficient phenotype in Grn1-null yeasts. To determine the unique function of GNL3L, we identified estrogen-related receptor ␥ (ERR␥) as a GNL3L-specific binding protein.

GNL3L and ERR␥ are coexpressed in the eye, kidney and muscle, and co-reside in the nucleoplasm. The interaction between GNL3L and ERR␥ requires the intermediate domain of GNL3L and the AF2-domain of ERR␥. Gain-ofand loss-of-function experiments show that GNL3L can
inhibit the transcriptional activities of ERR genes in a cellbased reporter system, which does not require the nucleolar localization of GNL3L. We further demonstrate that GNL3L is able to reduce the steroid receptor coactivator (SRC) binding and the SRC-mediated transcriptional coactivation of ERR␥. This work reveals a novel mechanism that negatively regulates the transcriptional function of ERR␥ by GNL3L through coactivator competition. GNL3L represses ERR activities heart, kidney and adipose tissues (Bonnelye et al., 1997;Giguere et al., 1988;Hong et al., 1999). ERR␤ is expressed in the eye, heart, kidney, cerebellum and testis (Bookout et al., 2006). The functions of the ERR family genes are implicated in many aspects of embryogenesis and tumorigenesis. Mice deficient in ERR␣ exhibit reduced body weight and peripheral fat deposit, and are resistant to obesity induced by a high-fat diet (Luo et al., 2003). ERR␤-null mice display placentation defects, consistent with its role in the proliferation and differentiation of trophoblastic cells (Luo et al., 1997). In humans, the expression level of ERR␣ correlates with poor prognosis in ovarian and breast cancers (Ariazi and Jordan, 2006;Suzuki et al., 2004). Conversely, ERR␥ is a favorable indicator for human ovarian tumors (Sun et al., 2005). Although the ERR family genes are capable of binding the estrogen response element (ERE), they are different from ER␣ in that their transcriptional activity and coactivator binding do not require ligand binding (Giguere et al., 1988;Hong et al., 1999;Kallen et al., 2004), which leaves open the question of whether the activities of the ERR family genes are constitutively active or dynamically regulated.
Here, we uncover a GNL3L-mediated pathway that regulates the transcriptional activity of the ERR family genes. We show that only GNL3L, but not nucleostemin or Ngp1, can interact with ERR family genes. Coexpression of GNL3L inhibits the transcriptional activities of the ERR family genes. Conversely, knocking down the endogenous expression of GNL3L increases the ERR-mediated transactivation. Furthermore, GNL3L can compete with steroid receptor coactivators (SRCs) for their ERR␥ binding and block the SRC-mediated coactivation of ERR␥. Our study reveals a GNL3L-mediated mechanism that modulates the transcriptional activities of ERR proteins.

GNL3L interacts with ERR family genes
To determine the unique function of GNL3L in vertebrates, we searched for proteins that interact with GNL3L but not with nucleostemin. A yeast two-hybrid approach was employed, where full-length GNL3L was fused to a GAL4 DNA bindingdomain, and used to screen a mouse E17.5 brain cDNA library. From a total of 5 million clones screened, two positive clones were identified. They encoded the same in-frame partial sequence of ERR␥ (Clone 43, residues 27-458). The interaction between GNL3L and clone 43 or full-length ERR␥ was confirmed by affinity-binding assays in which agarosebound GST fusion proteins of clone 43 and full-length ERR␥, but not the GST backbone protein, were able to retain hemagglutinin (HA)-tagged GNL3L specifically ( Fig. S1A in supplementary material). We use coimmunoprecipitation assays to verify the interaction between GNL3L and ERR␥ and to determine whether nucleostemin or Ngp1 can bind ERR␥. HEK293 cells were co-transfected with Myc-tagged ERR␥ and one of the nucleostemin family genes tagged with an HAepitope. Protein complexes were precipitated with anti-Myc or anti-HA antibodies, and immunodetected for the HA-tagged or Myc-tagged proteins (Fig. 1A). We found that only GNL3L, but not nucleostemin or Ngp1, was co-purified with ERR␥ by anti-Myc immunoprecipitation (Fig. 1A, row 1). Consistently, ERR␥ was detected only in the GNL3L protein complex, but not in the nucleostemin or Ngp1 protein complexes, precipitated by anti-HA antibody (Fig. 1A, row 3). Like GNL3L, ERR␥ belongs to a family of three genes. To determine whether GNL3L interacts with only ERR␥ or with multiple members of the ERR gene family, HEK293 cells were transfected with HA-tagged GNL3L and Myc-tagged ERR␣, ERR␤, or ERR␥ expression plasmids. Our results show that GNL3L and all members of the ERR gene family can be copurified in the same protein complexes precipitated by either anti-Myc or anti-HA antibodies (Fig. 1B, rows 1 and 3). By contrast, no physical interaction between ERR␣ (or ERR␤) and nucleostemin (or Ngp1) was detected by affinity-binding assays ( Fig. S1B in supplementary material). Based on these results, we conclude that only GNL3L of the nucleostemin family can form stable protein complexes with the ERR family genes.
ERR␥ colocalizes with GNL3L in the nucleoplasm To determine whether the interaction between GNL3L and ERR␥ is physiologically possible, we examined the expression patterns of GNL3L, ERR␤ and ERR␥ by multi-tissue northern blot analyses. GNL3L was expressed most abundantly in the neural tissues, including the brain and eye, and was also detected in the muscle and kidney at low levels ( Fig. 1C). Parallel blots showed that ERR␤ was expressed most abundantly in the kidney, followed by the eye, testis, heart and muscle. High-level expression of ERR␥ was seen in the heart and eye, followed by the brain, kidney and muscle. To decide where within the cell this interaction might occur, green fluorescent protein (GFP)-tagged GNL3L or ERR␥ were expressed in U2OS cells, which express both GNL3L and ERR␥ endogenously. We found that GNL3L was distributed both in the nucleolus and in the nucleoplasm, whereas ERR␥ was localized exclusively in the nucleoplasm (Fig. 1D). These data show that the tissue expression pattern of GNL3L correlates better with that of ERR␥ than that of ERR␤. Within the cell, ERR␥ and GNL3L colocalize in the nucleoplasm. In the nucleolus, only GNL3L is found.

The intermediate domain of GNL3L interacts with the AF2 domain of ERR␥
To gain insight into the functional importance of the GNL3L-ERR␥ interaction, we first identified the interacting domains of these two proteins using a panel of truncated GNL3L and ERR␥ mutants ( Fig. 2A,D). To define the ERR␥-binding domain in GNL3L, agarose-bound GST-ERR␥ fusion protein was used to pull down the wild-type and mutant GNL3L proteins (Fig. 2B). Our results show that ERR␥ can bind GNL3L mutants that lack the BC-domain (dBC) or the Gdomain (dG), as well as the N166I mutant that contains an Asn166 to Isl mutation in the G4 domain, which abolishes the GTP-binding capability of GNL3L (our unpublished data). Notably, the GST-ERR␥ fusion protein failed to retain mutants without the intermediate (I)-domain (dI and G3l-G, Fig. 2B), indicating that the I-domain is necessary for the ERR␥ binding of GNL3L. Different GNL3L mutants displayed distinctive subcellular distribution patterns not related to their ERR␥binding abilities. The dBC and G3l-G mutants were localized in the nucleoplasm and cytoplasm (Fig. 2C1,C2). The dG and dI mutants were localized more in the nucleolus than in the nucleoplasm (Fig. 2C3,C4), whereas the N166I mutant was diffusely distributed in the nucleus (Fig. 2C5). The ERR␥ protein structure consists of the AF1, DBD, LBD and AF2 domains (Fig. 2D). GST fusion proteins of the full-length ERR␥, the AF1-domain deletion mutant (dAF1) and the LBD deletion mutant with an intact AF2-domain (dLBD) were able to bind wild-type GNL3L. By contrast, GST fusion proteins of the AF2-domain deletion mutant (dAF2), the last 245 residues containing the LBD and AF2 domains (LBD-AF2) and a mutant deleted of the LBD and AF2 domain (AF1-DBD) were unable to retain GNL3L, demonstrating that the AF2 domain is necessary but not sufficient for the binding of ERR␥ to GNL3L (Fig. 2E). Based on these results, we conclude that the interaction between GNL3L and ERR␥ requires the I-domain of GNL3L and the AF2 domain of ERR␥, and is independent of the GTP binding and nucleolar localization of GNL3L.
Overexpression of a nucleolar form of GNL3L brings ERR␤ and ERR␥ into the nucleolus Given that GNL3L, but not ERR␥, is localized in the nucleolus, we tested the idea whether coexpression of GNL3L can bring ERR␥ into the nucleolus. Using confocal analysis, we determined the distribution of ERR␥ when coexpressed with wild-type GNL3L (WT), a nucleolar form of GNL3L (NoG3l), or an I-domain mutant of GNL3L fused to an SV40 nuclear localization sequence (NLS) (nls-I). NoG3l was created by Journal of Cell Science 120 (15) replacing the BC-domain of GNL3L with the BC-domain of nucleostemin (indicated by the grey bar in Fig. 3A) because nucleostemin has a stronger nucleolar localization capability than GNL3L and does not bind ERR␥. The I-domain mutant was fused with an SV40 NLS because it lacks endogenous NLS of its own. Affinity-binding assays confirmed that both NoG3l and nls-I mutants were capable of binding ERR␥ (Fig.  3B). Whereas ERR␥ by itself was distributed outside of the nucleolus (Fig. 3C1,C1Ј), overexpression of wild-type GNL3L (Fig. 3C2,C2Ј) and NoG3l (Fig. 3C3,C3Ј) increased the ERR␥ fluorescence signal in the nucleolar region compared with cells expressing ERR␥ by itself. Notably, in cells overexpressing the NoG3l mutant, the ERR␥ signal accumulated in the nucleolar region, particularly in the periphery of the nucleolus. By contrast, the nls-I mutant failed to alter the ERR␥ distribution ( Fig. 3C4), and distributions of wild-type GNL3L and NoG3l were unaltered by coexpression of ERR␥ ( Fig. 3C2 vs C5, and C3 vs C6). Overexpression of GNL3L and its mutants exerts the same effects on the distribution of ERR␤. Whereas ERR␤ by itself displayed a nucleoplasmic distribution (Fig.  3D1,D1Ј), both the wild-type GNL3L protein (Fig. 3D2,D2Ј) and the NoG3l mutant (Fig. 3D3,D3Ј) were able to increase the nucleolar intensity of ERR␤. By comparison, NoG3L had a stronger effect on bringing ERR␤ into the nucleolus than wild-

Fig. 1. GNL3L interacts with ERR␣, ERR␤ and ERR␥. Protein interactions between ERR␥ and nucleostemin family genes (A) and between
GNL3L and ERR family genes (B) were examined by in vivo coimmunoprecipitation assays. HEK293 cells were co-transfected with (A) Myctagged ERR␥ and HA-tagged nucleostemin, GNL3L or Ngp1 expression plasmids, or (B) HA-tagged GNL3L and Myc-tagged ERR␣, ␤ or ␥ expression plasmids. Lysates were immunoprecipitated (IP) with anti-Myc antibody (rows 1 and 2, ␣-Myc) or anti-HA antibody (rows 3 and 4, ␣-HA). The co-purified proteins (rows 1 and 3) and the immunoprecipitates (rows 2 and 4) were immunodetected with the antibodies indicated on the right. Our results show that ERR␥ interacts only with GNL3L, not with nucleostemin or Ngp1, and that GNL3L binds all ERR family proteins. (C) Tissue distribution of GNL3L, ERR␤ and ERR␥ in adult mice is shown by multi-tissue northern blots. GNL3L mRNA is expressed primarily in the neural tissues, including the brain and eye, and at lower levels in the kidney and muscle. The expression levels of GNL3L and ERR␥ match in kidney, muscle and eye, but differ in brain and heart. (D) In U2OS cells, the intensity of GFP-tagged GNL3L (GNL3L-gfp) is higher in the nucleolus than in the nucleoplasm. GFP-tagged ERR␥ (ERR␥-gfp) is localized exclusively in the nucleoplasm. The nucleolar regions are labeled by anti-fibrillarin (Fib) antibody in the right panels. Dashed lines demarcate nucleo-cytoplasmic boundaries. Bars, 10 m. GNL3L represses ERR activities type GNL3L. By contrast, overexpression of wild-type GNL3L or NoG3l had little or no effect on changing the distribution of ERR␣ (Fig. 3E). These results demonstrate that overexpression of a nucleolar form of GNL3L can change the distribution of ERR␤ and ERR␥ in living cells, suggesting the possibility that nucleolar sequestration underlies the regulation of ERR␤ and ERR␥ by GNL3L.

GNL3L suppresses the transcriptional activities of ERR family genes
To investigate whether GNL3L can modulate the transcriptional activity of the ERR genes, an in vivo cell-based luciferase assay system was set up where CV-1 cells were cotransfected with a firefly luciferase reporter construct driven by three repeats of a consensus palindromic estrogen response element (ERE, see Materials and Methods), an ERR␥ expression plasmid, and a pRL-null reporter construct. The ERE-specific transcriptional activity was determined by the ratio between the Firefly and Renilla luciferase activities in the same sample, which represent ERE-driven and basal activities, respectively. This dual luciferase assay system was used to eliminate variations caused by transfection and non-specific effects on the common transcription-translational machinery. The Firefly-to-Renilla luciferase activity ratio for each experiment was expressed as the fold increase over the negative sample not transfected with ERR␥. Our results show that ERR␥ can increase the ERE-specific transcriptional activity six times higher than that of the control sample (6.0±0.5, mean±s.e.m.), and coexpression of GNL3L attenuates ERR␥dependent increase by 50% (2.9±0.2) (Fig. 4A1, WT). This ERR␥ inhibitory effect of GNL3L requires its ERR␥interacting domain, because samples that coexpress the dI mutant fail to show such a repressive activity. To determine whether this inhibition is caused by nucleolar sequestration of ERR␥ by GNL3L, a nucleolar form of GNL3L (NoG3l) or a nucleoplasmic mutant of GNL3L (dBC) was coexpressed with ERR␥. Our results demonstrate that both NoG3l and dBC can inhibit ERR␥-mediated transcriptional activity more than or as much as the wild-type GNL3L. To test whether this transcriptional repressive effect of GNL3L can act on other members of the ERR family, we set up the same transactivational assay for ERR␤ and ERR␣. Our data show that GNL3L reduces the transcriptional activity of ERR␤ from 4.7 times (±0.3) to 2.1 times (±0.1) over the control sample in an I-domain-dependent manner, and both NoG3l and dBC inhibit ERR␤ as much as the wild-type protein (Fig. 4B1). Compared with ERR␤ and ERR␥, the ERR␣-dependent increase of ERE-driven transactivation is less (2.7±0.1). Although the wild-type GNL3L and the NoG31 mutant can reduce the ERR␣-mediated transcriptional activity, the dBC mutant fails to do so (Fig. 4C1). The GNL3L effect on the transcriptional activities of the ERR family genes is specific, because GNL3L does not suppress estradiol (E2)-induced ER␣-mediated transactivation when using the same reporter assay system (P=0.27, Fig. 4D). Finally, we confirm that these different effects of wild-type and mutant GNL3L on the transcriptional activities of ERR genes are not caused by different expression levels of the GNL3L or ERR proteins (Fig.  4A2,B2,C2).
GNL3L competes with SRC1 and SRC2 (GRIP1) for ERR␥ binding To look for a mechanism other than nucleolar sequestration in order to explain the GNL3L-mediated inhibition of ERR transactivation, we examined the possibility that GNL3L binding of ERR␥ prevents ERR␥ from accessing coactivators, such as SRC1 and SRC2 (GRIP1), which have been shown to also bind the AF2 domain of ERR␥ (Hong et al., 1999). To test this idea, GST fusion proteins of ERR␥ were used to pull down whole-cell lysates that contained a fixed amount of SRC1 or SRC2, mixed with increasing amounts of GNL3L (Fig.  6A1,B1). In each sample, whole-cell proteins were adjusted to the same concentration. Western blot analyses of the agarosebound protein fractions showed that less SRC1 or SRC2 proteins were retained by GST-ERR␥ as more GNL3L protein was bound by ERR␥ [ Fig. 6A1,B1, (R)]. The ability of GNL3L to compete with SRC1 and SRC2 for ERR␥ binding is abolished by a deletion of its ERR␥-interacting I-domain (Fig.  6A2,B2). Conversely, to determine whether SRC1 or SRC2 can displace GNL3L from the ERR␥ protein complex, GST-ERR␥ fusion proteins were used to pull down a fixed amount of GNL3L in the presence of increasing amounts of SRC1 or SRC2. Our results show that both SRC1 (Fig. 6C) and SRC2 (Fig. 6D) can reduce GNL3L bound by GST-ERR␥ in a dosedependent manner. These data demonstrate that binding between ERR␥ and GNL3, and between ERR␥ and SRC1 or SRC2 are mutually exclusive, and suggest that blocking ERR␥ from accessing SRC1 and SRC2 may be responsible for the inhibitory transcriptional activity of GNL3L.

Coexpression of GNL3L increases the mobility and decreases the SRC1 component of the ERR␥ DNAprotein complex
To determine whether GNL3L forms a high-order DNAprotein complex with ERR␥, electrophoretic mobility shift assays (EMSAs) were conducted using a radiolabeled probe containing a canonical ERE sequence (TCAGGTCA -CTGTGACCTGA) and cell extracts expressing the indicated Fig. 3. Overexpression of GNL3L brings ERR␤ and ERR␥ into the nucleolus. (A) To generate a nucleolar form of GNL3L (NoG3l), we replaced the N-terminal nucleolus-targeting domain of GNL3L with the corresponding region of nucleostemin (grey bar), which has a stronger nucleolus-targeting activity than GNL3L but lacks the ability to bind ERR␥. To create a nucleoplasmic form of GNL3L (nls-I), we fused the I-domain of GNL3L to an SV40 nuclear localization signal (oval). (B) Affinity-binding assays show that both nls-I and NoG3l maintain the ability to bind ERR␥. To measure the effect of GNL3L overexpression on the distribution of ERR␥, U2OS cells were transfected with Myctagged ERR␥ alone (C1), Myc-tagged ERR␥ and HA-tagged wild-type GNL3 (C2) or mutant GNL3L (C3,C4) or HA-tagged GNL3L constructs alone (C5 and C6). Double-transfected cells were labeled with anti-Myc (red) and anti-HA (green) antibodies, and visualized by confocal analyses. Single-transfected cells were immunostained with anti-fibrillarin antibody (Fib) and anti-Myc or anti-HA antibody. The ERR␥ (red) fluorescence intensities are measured quantitatively along the lines indicated by arrows, shown in the right panels of (C1Ј-C3Ј), and the nucleolar regions (No) are indicated by the increase of green fluorescence. Compared to cells transfected with only ERR␥, the fluorescence intensity of ERR␥ in the nucleolus is increased in cells cotransfected with NoG3l or the wild-type GNL3L. By contrast, the nls-I mutant does not change the distribution of ERR␥ (C4). Neither does ERR␥ alter the distribution of GNL3L (C5) or NoG3l (C6). The same analyses were performed using ERR␤ (D1-D3) and ERR␣ (E1-E3). Our results showed that, when coexpressed with wild-type GNL3L (D2) or NoG3l (D3), ERR␤ begins to accumulate in the nucleolus. GNL3L overexpression has little or no effect on the distribution of ERR␣ (E2,E3). Bars, 10 m. proteins (Fig. 7A). Compared with the probe alone (lane 1) and vector-transfected cell lysate samples (lane 2), the Myc-tagged ERR␥-transfected sample (lane 3) yields an ERE-ERR␥specific DNA-protein complex (Fig. 7A, arrow b). The signal of this complex can be competed by excess unlabeled probes (lane 4), and supershifted by anti-Myc antibody (lane 5, arrow a). When ERR␥ was coexpressed with GNL3L, two fastmoving DNA-protein complexes were identified (arrows d and e, lane 6), and no additional slow-moving complexes were seen. The mobility of the fast-moving complexes can be retarded by anti-Myc antibody (lane 7, arrow c), but not by anti-HA antibody (lane 8), suggesting that they contain ERR␥ but not GNL3L. GNL3L alone fails to interact with the ERE probe (lane 9). The intensity of the fast-moving complex, complex d, was reduced when ERR␥ was coexpressed with a GNL3L mutant lacking its ERR␥-binding I-domain (dI), indicating that the appearance of this fast-moving complex d depends on the interaction between GNL3L and ERR␥.
The increased mobility of the ERR␥ DNA-protein complex by GNL3L may be caused by cleavage of the ERR␥ protein or by changes in the protein conformation or components of the ERR␥ DNA-protein complex; failure of this fast-moving Journal of Cell Science 120 (15) complex d to be supershifted by anti-HA antibody indicates that it does not contain GNL3L or the HA-epitope of GNL3L is masked in this particular protein conformation. To address these different possibilities, we retrieved the fast-moving and slow-moving protein complexes (complex d and complex b, respectively) from the EMSA gel and analyzed the protein amount and size of ERR␥, GNL3L, SRC1, and SRC2 in these two complexes by western blottings (Fig. 7B). Anti-Myc western analysis shows that the size of the ERR␥ protein remains the same in both complexes, excluding the possibility that the increased mobility is a result of ERR␥ protein cleavage. Anti-HA western blotting detects no GNL3L protein in the retrieved protein complexes, consistent with the idea that GNL3L does not bind the DNA-bound ERR␥. Notably, we are able to detect SRC1 in the slow-moving complex b but not in the fast-moving complex d, demonstrating that coexpression of GNL3L reduces SRC1 binding to the ERR␥ DNA-protein complex. Although the SRC2 protein is present in both complexes, the amount of SRC2 relative to ERR␥ is reduced in the fast-moving complex d compared with the slow-moving complex b, which suggests that SRC2 binding to ERR␥ is also diminished by GNL3L coexpression. Fig. 4. Overexpression of GNL3L inhibits transcriptional activity of ERR proteins independently of nucleolar distribution. (A1) Estrogen response element (ERE)-specific transcriptional activities were measured in CV-1 cells by the ratio between the ERE-driven firefly luciferase activity and the Renilla-null luciferase activity. ERR␥ elicits a six-fold increase in the ERE-specific transcriptional activity. Coexpression of wild-type GNL3L (WT) leads to a 50% reduction in the ERR␥-mediated transcriptional activity. This decrease is reversed by a deletion of the ERR␥-binding I-domain of GNL3L (dI). Coexpression of either the nucleolar form (NoG3l) or the nucleoplasmic form (dBC) of GNL3L suppresses the ERR␥ transcriptional activity more than or to the same extent as the wild-type GNL3L protein. (B1,C1) Using the same approach, we show that this inhibitory activity of GNL3L can also work on (B1) ERR␤ and (C1) ERR␣ with the exception that the dBC mutant has little effect on the ERR␣-mediated transactivation. Error bars represent the standard error of the mean (± s.e.m.). ***P<0.0001. (A2,B2,C2) Expression levels of wild-type and mutant GNL3L proteins and ERR proteins in the experimental samples are compared in western blots sideby-side using anti-HA and anti-Myc antibodies, respectively; ␣-tubulin (␣-Tub) was used as a loading control. (D) GNL3L fails to suppress the estradiol (E2)-induced transcriptional activity of ER␣ on the ERE-driven promoter in the same cell-based reporter system. GNL3L represses ERR activities GNL3L suppresses the SRC-mediated transcriptional coactivation on ERR␥ Next, we addressed the issue whether GNL3L interferes with the function of SRC proteins as coactivators for ERR␥. In a cell-based reporter system similar to that described in Fig. 4, coexpression of ERR␥ and SRC1 is able to produce an eightfold increase (8.0±0.3) in the ERE-specific transcriptional activity compared with the control sample -which is 1.7 times higher than the sample expressing only ERR␥ (4.8±0.3) (Fig.  8A). When coexpressed with GNL3L, the luciferase activity is Fig. 5. Endogenous GNL3L suppresses transcriptional activity of ERR family genes. (A) To confirm the GNL3Lmediated negative regulation of ERR activities from a loss-of-function angle, a small interfering RNA (siRNA) approach was used to knock down the endogenous expression of GNL3L. Compared to the control knockdown sample (siNEG), the protein knockdown efficiencies of GNL3L-specific siRNA duplexes (siGNL3L-1 and siGNL3L-2) in HEK293 cells stably expressing HAtagged GNL3L are estimated to be 83% and 84%, respectively. (B) Consistent with our overexpression data, the transcriptional activity of ERR␥ is increased 2.5 times by siGNL3L-1 and siGNL3L-2 treatment as compared with the siNEG-treated sample. (C,D) GNL3L knockdown has the same effect on the ERR␤-and ERR␣-mediated transactivation, although their increase is less dramatic than the increase in the ERR␥-mediated transactivation. Error bars represent the standard error of the mean (± s.e.m.). **P<0.001; ***P<0.0001. Fig. 6. GNL3L competes with SRC1 and SRC2 for ERR␥ binding. Agarose-bound GST fusion proteins of ERR␥ (1 g) were used to pull down whole-cell lysates containing a fixed amount of SRC1 (A) or SRC2 (B), mixed with increasing amounts of the wild-type GNL3L (A1,B1) or the dI mutant lacking the ERR␥-interacting domain (A2,B2). Whole-cell proteins in each sample were adjusted to the same amount. In the agarose-retained portions (R), the interaction between GNL3L and ERR␥ can reduce the amount of SRC1 and SRC2 bound by ERR␥ in a dosedependent manner, but the dI mutant fails to do so. Conversely, when GST-ERR␥ fusion proteins were used to pull down the same amount of GNL3L in the presence of increasing amounts of SRC1 (C) or SRC2 (D), SRC1 and SRC2 were able to reduce the amount of GNL3L bound by ERR␥ in a dose-dependent way as well. Proteins in the agarose-bound fraction and in the supernatant are indicated by (R) and (S), respectively. reduced to 3.6 times (± 0.2, for 100 ng of GNL3L) and 2.5 times (± 0.2, for 200 ng of GNL3L) that of the control sample, representing respectively a 55% and 70% reduction compared with the sample expressing both ERR␥ and SRC1. Again, a deletion of the ERR␥-interacting domain (dI) abolishes the suppressive activity of GNL3L (P=0.17). The ability of GNL3L to inhibit the transcriptional coactivator activity can also work on SRC2 (Fig. 8B). Here, we show that GNL3L, but not the dI mutant, can reduce the ERR␥-SRC2-mediated 7.4fold increase (7.4±0.2) in the luciferase activity down to 3.4fold (± 0.2, for 100 ng of GNL3L) and 2.1-fold (± 0.2, for 200 ng of GNL3L) compared with the control sample. The dI mutant fails to exhibit this inhibitory activity (P=0.58). These data demonstrate that GNL3L can suppress the SRC1-Journal of Cell Science 120 (15) mediated and SRC2-mediated coactivation of ERR␥ in an Idomain-dependent manner.

Discussion
In this manuscript, we identify a GNL3L-mediated mechanism that suppresses the transcriptional activity of ERR family genes by coactivator competition (Fig. 9). We show that ERR binding is specific to GNL3L, but not to nucleostemin or Ngp1. GNL3L and ERR␥ colocalize in the nucleoplasm, and their interaction requires the I-domain of GNL3L and the AF2-domain of ERR␥. Gain-of-function and loss-of-function studies reveal that GNL3L has the ability to suppress the transcriptional activity of ERR genes, and does not require the nucleolar localization of GNL3L to do so. We also demonstrate that GNL3L can compete with SRC1 and SRC2 for their binding to ERR␥, resulting in an increased electrophoretic mobility of the DNA-bound ERR␥ complex and the inhibition of the SRC1 and SRC2 coactivator function on ERR␥. The AF2-domain binding, SRC competition and transcriptional inhibition activities suggest that GNL3L represents a new class of transcriptional co-repressor for nuclear receptors. However, GNL3L fails to form a stable complex with the DNA-bound ERR␥, and the I-domain of GNL3L -which is necessary and sufficient for ERR␥ binding -lacks the LxxLL motif found in most transcriptional coactivators and co-repressors that interact with the AF2domain (Hentschke et al., 2002;Huss et al., 2002;Lee et al., 1998;Rosenfeld and Glass, 2001;Webb et al., 2000;Zhang et al., 2000). These data support the role of GNL3L as a novel regulator for the ERR gene family and argue against its role as a classical transcriptional co-repressor. It is worth noting that the absence of interaction between GNL3L and the DNAbound ERR␥ in vitro does not exclude the possibility that these two proteins still coexist in the same DNA-bound complex in the native chromatinized context, because the binding of transcription factors to the core response element in vivo is aided by a number of cofactors as well as by histone proteins involved in chromatin remodeling. Furthermore, the ability of GNL3L to compete for SRC binding and to inhibit the transcriptional function of ERR␥ might depend on specific chromatin structures or involve cofactors other than SRC proteins.
GNL3L and ERR proteins are colocalized in the nucleoplasm, but only GNL3L is also found in the nucleolus. The nucleolar localization of GNL3L might have created two potential mechanisms that affect its activity in the nucleoplasm. First, GNL3L can enter or exit the nucleolus by itself. In this case, signals that promote the nucleolar accumulation of GNL3L may cause a disinhibition of the ERR activities, and signals that release the nucleolus-bound GNL3L into the nucleoplasm may allow more GNL3L to bind ERR proteins. Alternatively, GNL3L may carry some ERR proteins with it when entering the nucleolus, in which case nucleolar sequestration of ERR proteins may account for some of the inhibitory activity of GNL3L. Although overexpression of a nucleolar form of GNL3L and, to a less extent, the wild-type GNL3L increase the nucleolar intensity of ERR␤ and ERR␥, the non-nucleolar dBC mutant is able to suppress ERR␤ and ERR␥ activities as much as wild-type GNL3L. These results demonstrate that the suppressive effect of GNL3L on the ERR activity is not mediated by a nucleolar sequestration GNL3L itself cannot bind the ERE probe (lane 9). The intensity of the fast-moving complex d is reduced by a deletion of the ERR␥binding I-domain of GNL3L (lanes 10-12). (B) The fast-moving complex d and the slow-moving complex b were retrieved from the EMSA gel, fractionated in SDS-denaturing PAGE and analyzed for their ERR␥ (␣-Myc), GNL3L (␣-HA), SRC1, and SRC2 protein components by western blotting. Our results indicate that the increase in the electrophoretic mobility of the ERR␥-DNA complex by GNL3L coexpression is due to a loss of SRC1 binding (arrow) and diminished SRC2 binding, rather than by protein cleavage of ERR␥. GNL3L represses ERR activities mechanism and that, under physiological conditions, the ERRbinding and transcriptional inhibition events of GNL3L take place in the nucleoplasm. Nevertheless, the fact that overexpression of NoG3l can bring ERR␤ and ERR␥ into the nucleolus supports the notion that GNL3L can interact with these two proteins in vivo.
Unlike ER␣ whose activity is controlled by hormone binding, no ligand has so far been identified for ERR family genes and their transactivation works in a ligand-independent manner. This GNL3L-mediated inhibition on the activities of ERRs provides one mechanism to regulate their functions in a cell-context-dependent and dynamic way. At the transcriptional level, although the relative abundance of GNL3L matches that of ERR␥ in most tissues we examined, they are distinctively different in the brain and heart. GNL3L is expressed highly in the brain but little in heart, whereas ERR␥ is found at high levels in heart but not in brain. The differences between the expression levels of GNL3L and ERR␥ in those organs indicate that tissues that express the same level of ERR␥ may exhibit differential ERR activities depending on their GNL3L expression. In the adult brain where little ERR␥ is found, GNL3L might have other regulatory  Fig. 4, we show that the ERE-specific transcriptional activity in cells coexpressing ERR␥ and SRC1 (8.0±0.3) is 1.7 times higher than that of the ERR␥-expressing sample (4.8±0.3). When coexpressed with the wild-type GNL3L (WT), this ERR␥ and SRC1-mediated ERE-specific transcriptional activity is reduced by 55% and 70% compared with the sample expressing both ERR␥ and SRC1 in a dose-dependent manner. This inhibitory effect of GNL3L on the SRC1-mediated coactivation of ERR␥ requires the I-domain of GNL3L because deletion of this domain (dI) fails to suppress the transcriptional activity of ERR␥ and SRC1 (P=0.17). (B) Using the same approach, we show that GNL3L can also suppress the coactivator function of SRC2 on the ERR␥-dependent transcriptional activity in a dose-dependent (54% reduction for 100 ng of GNL3L and 71% reduction for 200 ng of GNL3L) and I-domain-dependent (P=0.58) manner. Error bars represent the stand error of mean (± s.e.m.). ***P<0.0001. Fig. 9. GNL3L inhibits the transcriptional activities of ERR family genes by coactivator competition. Our data reveal a novel mechanism that regulates the activity of ERR family genes by the nucleolar GTP-binding protein GNL3L. GNL3L decreases the transcriptional activity of ERR proteins. This event takes place in the nucleoplasm and does not require the nucleolar localization of GNL3L. The interaction between GNL3L and ERR␥ displaces coactivators such as SRC1 and SRC2 from the ERR␥ complex. The SRC-depleted ERR␥ protein binds DNA without GNL3L, resulting in transcriptional inhibition. In this model, the nucleolar accumulation of GNL3L does not appear to affect its ability to suppress the transcriptional function of ERR proteins (grey arrows targets. At the post-translational level, GNL3L is partitioned between the nucleolus and the nucleoplasm by a dynamic process (L.M. and R.Y.L.T., unpublished data). Like nucleostemin (Meng et al., 2006;Tsai and McKay, 2005), the nucleolar accumulation of GNL3L is controlled by its GTP binding and a N-terminal basic domain  (L.M. and R.Y.L.T., unpublished data). Notably, the nucleolar residence of GNL3L is significantly shorter than that of nucleostemin (L.M. and R.Y.L.T., unpublished data). The transient residence of GNL3L in the nucleolus might explain why nucleolar compartmentalization of GNL3L does not seem to play a role in its ability to suppress the ERR transcriptional function.
In conclusion, it is known that the ERR family genes are transcriptionally active without the ligand, but it is unclear if and how their activities can be controlled in a dynamic manner. Our work unravels a GNL3L-mediated mechanism that modulates the transcriptional activity of ERR␥ by coactivator competition. Given the important role of the ERR genes in embryogenesis and tumorigenesis, the differential regulation of their activities by GNL3L can provide us with new insight into these two processes in a cell-type-specific manner.

Recombinant plasmids and mutation analyses
Full-length ERR family genes were cloned from mouse brain cDNAs by reverse transcription (RT)-PCR. Deletions and point mutations of GNL3L and ERR␥ were introduced by the stitching PCR method as described previously (Tsai and McKay, 2002;Tsai and McKay, 2005). The final PCR products were subcloned into pCIS expression vectors and confirmed by sequencing.
Yeast two-hybrid screen Full-length mouse GNL3L was subcloned in the pAS2-1 vector and used as a bait to screen an E17.5 mouse brain cDNA library in the pACT2 vector (Clontech). The bait and library plasmids were co-transformed into the Saccharomyces cerevisiae strain Y190, and selected for both histidine + and ␤-galactosidase + phenotypes. cDNA plasmids were re-transformed into Escherichia coli HB101 by electroporation and expanded for further analysis.

Affinity binding and competition assays
Full-length cDNAs of GNL3L and ERR␥ were subcloned into the pGEX4T-2 vector. GST fusion proteins were expressed in BL21/DE3 as described previously (Tsai and McKay, 2002;Tsai and Reed, 1997). Epitope-tagged proteins were expressed in HEK293 cells and extracted in phosphate-buffered saline (PBS)-Triton X-100 (1%) buffer, supplemented with protease inhibitors. Sepharose-bound GST fusion proteins (2-5 g) were incubated with whole-cell lysates for 2 hours at 4°C, washed five times with extraction buffer, including twice with high-salt buffer (extraction buffer plus 500 mM of NaCl), fractionated on 10% SDS-PAGE, and detected by western blot analyses.

Northern blot analyses
Ten micrograms of total RNA were isolated from CD-1 mice using Trizol solutions (Invitrogen), fractionated on 1% formamide denaturing agarose gels, and transferred onto Hybond XL membrane (Amersham). Filters were then hybridized with 32 Plabeled probes at 65°C overnight and washed under high stringency conditions. Plaque date was counted as embryonic day 0.5 (E0.5).

Image acquisition
Confocal images were captured on a Zeiss LSM510 confocal microscope using a 63ϫ plan-apochromat oil objective. Images were scanned using the multi-track program, a 512ϫ512 frame size, 3ϫ zoom, and <1.4 m optical thickness. Detector gain and amplifier offset were adjusted to ensure that all signals were appropriately displayed within the linear range. Fluorescence intensities were digitally quantified in Fig. 3CЈ,DЈ,EЈ using the profile display mode along the path indicated by arrow.

Dual luciferase assays
For gain-of-function experiments, CV1 cells were grown in DMEM supplemented with 5% charcoal/dextran-treated FBS. Transient transfection was performed in 24well plates using the Lipofectamine-Plus reagent. Total DNA amount in each well was adjusted to 2 g using the empty expression vector. Cell extracts were prepared 30 hours after transfection. HEK293 cells were used for siRNA knockdown experiments because of their high GNL3L expression levels. HEK293 cells were split and grown in DMEM supplemented with 5% chacoal/dextran-treated FBS. On the next day, transfections of siRNA duplexes were performed in 24-well plates using the Oligofectamine reagent (Invitrogen). Firefly (100 ng) and Renilla (10 ng) reporter plasmids, ERR (50 ng), ER␣ (100 ng), GNL3L-related genes (200 ng or as specified), and/or SRC (50 ng) expression vectors were transfected on the fourth day, and cell extracts were prepared 1 day later. For E2 stimulation (Fig. 4D), cells were treated with 100 nM of E2. Firefly and Renilla luciferase activities were measured by the Dual-Luciferase Reporter Assay System (Promega). The expression of the firefly luciferase reporter gene was driven by three repeats of a synthetic consensus palindromic estrogen response element (ERE, GGTCACTGTGACC).

EMSA and post-EMSA western blot
We thank Seokwoon Kim for his technical support in the beginning of this work. SRC1 and SRC2 expression plasmids were kindly provided by Jiemin Wong, and the ERE reporter construct was kindly provided by Stephen Safe. This work was supported by a grant from the National Cancer Institute (R01 CA113750-01) to R.Y.L.T.