Mitosin/CENP-F as a Negative Regulator of Activating Transcription Factor-4*

Mitosin/CENP-F is a human nuclear matrix protein with multiple leucine zipper motifs. Its accumulation in S-G2 phases and transient kinetochore localization in mitosis suggest a multifunctional protein for cell proliferation. Moreover, its murine and avian orthologs are implicated in myocyte differentiation. Here we report its interaction with activating transcription factor-4 (ATF4), a ubiquitous basic leucine zipper transcription factor important for proliferation, differentiation, and stress response. The C-terminal portion of mitosin between residues 2488 and 3113 bound to ATF4 through two distinct domains, one of which was a leucine zipper motif. Mitosin mutants containing these domains were able to either supershift or disrupt the ATF4-DNA complex. On the other hand, ATF4, but not ATF1–3 or ATF6, interacted with mitosin through a region containing the basic leucine zipper motif. Moreover, overexpression of full-length mitosin repressed the transactivation activity of ATF4 in dual luciferase-based reporter assays, while knocking down mitosin expression manifested the opposite effects. These findings suggest mitosin to be a negative regulator of ATF4 in interphase through direct interaction.

Mitosin/CENP-F is a large protein of 3113 residues possibly with multiple functions during the cell cycle. In M phase, it is a protein located at the outer kinetochore plate (1)(2)(3)(4) and is dynamically transported from kinetochores to spindle poles along microtubules by cytoplasmic dynein (5). At the end of mitosis, it is rapidly degraded (3). It is both hyperphosphorylated and farnesylated in M phase (3,6). Farnesylation of mitosin/CENP-F is critical for G 2 /M progression and its postmitotic degradation (7). In interphase, mitosin/CENP-F is ubiquitously expressed mainly from S to G 2 phases as a nuclear matrix protein (1)(2)(3). Such a property has been utilized as a proliferation marker for a variety of human malignancies (8 -10). The physiological significance of nuclear mitosin, however, is not clear. Studies on its chicken and murine orthologs, CMF1 and LEK1, respectively, have provided clues for its roles in interphase. CMF1 is highly expressed in differentiating chicken heart (11,12), with subcellular localization shifting from the nucleus to cytoplasm following skeletal myoblast differentiation (13). Its expression repression by antisense oligos diminished myosin expression in differentiating myoblasts (11,13). CMF1 is thus implicated in early events of cardiac and skeletal muscle differentiation. LEK1 is also implicated in muscle differentiation (12,14).
ATF4, also named CREB2, is a member of the ATF 1 /CREB family (15). Despite this, ATF4 lacks typical sites for protein kinase A critical for cAMP-dependent signaling and is thus unlikely a substrate of this kinase (16). However, human ATF4 is phosphorylated at Ser-245 by a growth factor-regulated kinase, RSK2, and the phosphorylation is implicated in skeletal development (17). In addition, phosphorylation of ATF4 on the DSGXXXS motif by an unknown kinase provokes its interaction with the SCF ␤TrCP ubiquitin ligase, leading to attenuated stability (16). ATF4 is important for several stress responses, in which the translation of ATF4 mRNA is specifically promoted, despite a general translational inhibition caused by phosphorylation of the translation initiation factor eIF2␣ (18 -21). Consistently, ATF4-deficient cells show defects in expressing genes involved in amino acid import, glutathione biosynthesis, and oxidative stress resistance (20). ATF4 also functions in long term memory and synaptic plasticity (22)(23)(24). In addition, ATF4-deficient mice show severe anemia, microphthalmia, and bone defects (17,(25)(26)(27), suggesting its critical roles in cell proliferation and differentiation.
ATF4 has been shown to interact with a variety of proteins through its C-terminal bZip region, implying its functional diversity and complex regulation. Through leucine zippers, it dimerizes with members either in the ATF/CREB family (28,29) or in other bZip transcription factor families, for instance c-Fos (29,30) and c-Jun (30) in AP-1 family and C/EBP␤ (31) and IGEBP1 (32) in C/EBP family. Moreover, heterodimers with other bZip family members often retain the CRE binding specificity (29,31). In addition, the Tax protein of the human T-cell leukemia virus and the ␥-aminobutyric acid type B receptor also bind ATF4 to potentially modulate its activities (33,34).
One of the striking structural characteristics of mitosin/ CENP-F family proteins is the richness in leucine zipper motifs (2,3,14). These motifs may mediate protein-protein interactions with other proteins. In an attempt to explore the role of mitosin in interphase, we found evidence for its functional relationship with ATF4.

EXPERIMENTAL PROCEDURES
Plasmid Constructs-To express mitosin mutants fused to Gal4 BD in yeast, cDNA fragments encoding mitosin TG from amino acids 2488 -3113 and mitosin Core from 2792 to 2887 (40) were cloned in-frame into pAS2-1 (Clontech) (Fig. 1A). A cDNA coding for the full-length mitosin was merged from cDNA fragments (3) and cloned into pUHD30F (35) to express FLAG fusion in mammalian cells. For vector-based RNA interference assay, a synthetic DNA fragment containing a 22-bp inverted repeat corresponding to nucleotides 200 -221 of human mitosin cDNA (GenBank TM accession NM-016343) was ligated into pBS/U6 vector (36) to create pBS/U6/Mi-1.
To express other family members of ATF4 in yeast, cDNAs containing the entire open reading frames from pATF1 (kindly provided by Dr. Y. Zhang), RSV-ATF2 (from Dr. M. Castellazzi), pCG-ATF3 (from Dr. T. Hai), and pCGN-ATF6 (from Dr. R. Prywes) were cleaved with appropriate restriction enzymes and cloned into pACT2.
For protein expression in Escherichia coli, the full-length ATF4 cDNA was cloned into pGEX-2T to express GST fusion. A mitosin cDNA fragment coding for amino acids 2484 -3113 was isolated from pMALmitosin mt10 (3) and inserted into pFLAG1 (IBI) to express FLAGmitosin mt10 . The construction of pMAL-mitosin 10/NB and pMAL-mitosin 10/KN and expression of MBP fusions have been previously described (3). pMAL-mitosin 10/NBZ was constructed to mutate codons for Leu-2571 and Leu-2578 in the leucine zipper motif within mitosin 10/NB into Ala by PCR-based mutagenesis. All the PCR-amplified sequences were confirmed by sequencing.
Yeast Two-hybrid Screen-The screen was performed following manufacturer's protocols for the Matchmaker 2 system (Clontech). Briefly, pAS-mitosin TG was cotransformed with a human placental cDNA library (Clontech) into yeast strain Y190 using the lithium acetate method. Colonies grown on S.D./-Leu/-Trp/-His/3-AT agar medium were assayed for ␤-galactosidase activity. Plasmid DNA was prepared from candidate positive clones and transformed into the E. coli strain TG1 to propagate. A second round of cotransformation with bait plasmids was then performed to confirm the interaction. Plasmids having passed the confirmation were subjected to sequencing and further analysis.
In Vitro Binding Assay-Bacterial lysates containing GST or MBP fusion proteins in lysis buffer A (10 mM Tris-HCl, pH 7.4, 150 mM NaCl, 0.25% Nonidet P-40, 10 mM EDTA, 1 mM dithiothreitol, and protease inhibitors) were incubated at 4°C for 30 min with 40 l 50% slurry of glutathione-agarose beads (Sigma) or amylose resin (New England BioLabs), respectively. Each aliquot of the glutathione beads was then incubated with 50 l of bacterial lysate containing FLAG-mitosin mt10 at 4°C for 2 h, followed by three times of wash with 500 l of lysis buffer A. The bound proteins were then analyzed by immunoblotting or Coomassie Blue staining after SDS-PAGE. Similarly, each aliquot of the amylose beads was incubated with 50 l of HEK293T lysates containing GFP-ATF4 in lysis buffer B (50 mM HEPES-KOH, pH 7.8, 500 mM NaCl, 1% Nonidet P-40, 5 mM EDTA, 3 mM dithiothreitol, and protease inhibitors) (37) at 4°C for 2 h, followed by five rounds of wash using phosphate-buffered saline (PBS). The bead-bound MBP fusion proteins were eluted using 50 l of 10 mM maltose in PBS and analyzed by Western blotting or Coomassie Blue staining after 8% SDS-PAGE.
EMSA-GST-ATF4 and GST were purified from bacterial lysates using glutathione-agarose beads and eluted with 5 mM glutathione in PBS. MBP and its fusion proteins with mitosin were purified using amylose resin and eluted with 10 mM maltose in PBS.
Oligos containing the consensus CRE site were synthesized according to sequences provided by Santa Cruz Biotechnology, Inc. The wildtype sequences were 5Ј-AGAGATTGCCTGACGTCAGAGAGCTAG-3Ј and 5Ј-GTGCTAGCTCTCTGACGTCAGGCAATCT-3Ј. The mutant sequences were 5Ј-AGAGATTGCCTGTGGTCAGAGAGCTAG-3Ј and 5Ј-CTAGCTCTCTGACCACAGGCAATCTCT-3Ј (CRE sites are shown in bold; the mutated bases are underlined). The wild-type probes were labeled using Klenow DNA polymerase and [␣-32 P]dCTP (10 Ci/l). EMSA was performed as described (32) with minor modifications. Briefly, purified protein(s) were incubated with 0.5 pmol of labeled probe in a 20-l solution containing 25 mM Tris-HCl, pH 8.0, 50 mM KCl, 0.5 mM EDTA, 2.5 mM dithiothreitol, 1 mg/ml bovine serum albumin, 10% glycerol, and 1 g of poly(dG-dC) for 20 min at room temper-ature in the absence or presence of cold competitor oligos of 100-fold molar excess. Protein-DNA complexes were resolved with 5% PAGE followed by autoradiography.
Dual Luciferase Reporter Assay-The firefly luciferase reporter construct pCRE-ATF4X2 contains two artificial CRE sites upstream of a minimal promoter and was a gift from Dr. T. Hai (Department of Molecular and Cellular Biochemistry, Ohio State University). The reporter construct pBScycAluc (provided by Dr. K. Oda, Department of Biological Science and Technology, Science University of Tokyo, Japan) contained the firefly luciferase gene driven by rat cyclin A promoter (38).
HEK293T cells in 6-well plate were cotransfected with 6 g of plasmid mixture per well, including reporter construct (1 g) and pRL-TK (0.1 g for each transfection in Fig. 5C and 0.2 g in Fig. 5D) for constitutive expression of Renilla luciferase (Promega) as an internal control. p15-1 (2 g), which is essential for activation of the tetracycline-responsive promoter in the pUHD vector (39), was included in the cocktails. Other cotransfected plasmids are indicated in Fig. 5. Luciferase assays were performed using a dual-luciferase reporter assay system (Promega) 48 h after transfection. The ratio of firefly luciferase activity to Renilla luciferase activity was presented in arbitrary units as the relative luciferase activities.

RESULTS
Interaction of ATF4 with Mitosin-We have previously shown that mitosin TG , a mutant containing amino acids 2488 -3113, was capable of kinetochore localization (40). This mutant contains three leucine zipper motifs potential for protein-protein interactions (3). To screen for its interacting proteins, yeast two-hybrid screen was performed using mitosin TG as bait. A positive clone was identified, coding for amino acids 185-351 of the transcription factor ATF4. Although ATF4 was not a kinetochore protein (data not shown), it might link mitosin to transcription regulation in interphase since mitosin orthologs in mice and chicken appear to be involved in differentiation (11)(12)(13)(14). Further assays were therefore performed.
The interaction between mitosin TG and ATF4 185/351 was specific, since mitosin core containing two leucine zipper motifs between residues 2792 and 2887 failed to bind ATF4 in yeast (Fig. 1, A and B). Moreover, full-length GST-ATF4 immobilized on glutathione beads pulled down FLAG-mitosin mt10 , a mutant expressed in E. coli and containing residues 2484 -3113 (Fig.  1C, lane 2), while GST alone did not (Fig. 1C, lane 1), further suggesting a direct interaction between mitosin and ATF4. Deletions were made to further map the mitosin-binding domain of ATF4. We found that deleting residues 296 -351 or 185-196 from ATF4 185/351 disrupted the binding activity to mitosin TG in yeast two-hybrid assays ( Fig. 2A). ATF4 thus bound mitosin through a C-terminal region including the bZip domain responsible for the dimerization and DNA binding (41).
We then examined whether mitosin interacted with other members of the ATF family. The homology among the bZip domains of the ATF family members is compared in Fig. 2B. When fragments of ATF1, ATF2, ATF3, and ATF6 covering regions roughly equivalent to residues 185-351 of ATF4 were expressed, no interaction with mitosin TG was observed in yeast (Fig. 2C). Therefore, mitosin appears to associate with only ATF4.
Binding of Mitosin to the ATF4-DNA Complex-ATF4 mediates transcriptional activation by binding to the cAMP-responsive elements (CRE) (15,28). We then tested whether mitosin interfered with the DNA binding activity of ATF4 using EMSA. Despite some degradation (Fig. 1C), purified GST-ATF4 formed a uniform complex with radioactively labeled DNA probes containing a CRE site (Fig. 3A, lanes 2 and 3). The complex was competed by excess amount of the cold probes (Fig. 3A, lane 4), but not by the mutant probes with two base substitutions in the CRE site (lane 5). Moreover, GST alone failed to bind the labeled probe (Fig. 3A, lanes 1, 6, 7).
Different MBP-mitosin fusion proteins were then purified (Fig. 3, B and C). The proline-rich C-terminal portion of mitosin appeared not stable in E. coli. Mutants containing this region were heavily degraded (3), though the full-length products were still visible (Fig. 3C, lanes 2 and 4). These mutants alone failed to show any binding to the CRE in EMSA (data not shown). When MBP-mitosin mt10 was added to challenge the ATF4-DNA complex, however, it supershifted the complex in dose-dependent manners (Fig. 3D, lanes 1-4). This result was in agreement with the interaction between ATF4 and this portion of mitosin (Fig. 1) and further indicated that the association with mitosin mt10 did not alter the DNA binding activity of ATF4. MBP-mitosin 10/KN also supershifted the ATF4-DNA complex (Fig. 3D, lanes 7 and 8), whereas MBP alone had no effect (lane 9), suggesting the existence of an ATF4 binding domain between amino acids 2645 and 3113 of mitosin. MBPmitosin 10/NB , on the other hand, disrupted the ATF4-DNA complex also in dose-dependent ways (Fig. 3D, lanes 5 and 6), suggesting another ATF4 binding domain between residues 2484 and 2645 of mitosin whose association alone abolished the ATF4-DNA interaction.
Mitosin Is a Negative Regulator of ATF4-mediated Transcriptional Activation-To understand the functional relevance of the interaction between mitosin and ATF4, we examined effects of mitosin overexpression or repression on the transactivation activity of ATF4. A plasmid, pUHD-mitosin, was constructed to express FLAG-tagged full-length mitosin. The FLAG-mitosin was indeed full-length according to its comigration with the endogenous one in SDS-PAGE ( Fig. 5A; Fig. 5B,  panels 1 and 2). Furthermore, it also exhibited nuclear distribution and kinetochore localization similar to the endogenous mitosin (data not shown). On the other hand, pBS/U6/Mi-1, a plasmid for expression of a small interference RNA against mitosin mRNA, was able to significantly knockdown endogenous mitosin in HEK293T cells (Fig. 5B, lanes 3 and 4). In contrast, either mitosin overexpression or repression did not alter levels of ATF4 (Fig. 5B).
Assays were performed initially with the reporter construct pCRE-ATF4X2 in HEK293T cells (Fig. 5C). Comparing to the vector pCEP4F, cotransfection with pCEP-ATF4 increased luciferase activity by 1.9-fold on average (Fig. 5C, lanes 1 and 2). Further expression of the full-length FLAG-mitosin, however, suppressed the reporter activity to basal levels (Fig. 5C, lanes  1-3). In contrast, overexpressing mitosin alone did not significantly affect the basal activity of the reporter (data not shown). Consistently, when pBS/U6/Mi-1 and pBS/U6 were respectively cotransfected with pCEP-ATF4, the reporter activities differed by about 2.1-fold (Fig. 5B, panels 4 and 5). Transfection of pBS/U6/Mi-1 alone had little effect (Fig. 5B, lane 6), indicating that the effect of mitosin depletion is ATF4-dependent. Similar results were also seen when luciferase was expressed using a rat or human cyclin A promoter ( Fig. 5D; data not shown). In these cases, knocking down mitosin increased the reporter activity by 2.3-fold and 3.5-fold, respectively, in ATF4dependent manner (Fig. 5D; data not shown). To eliminate the possibility that any double-stranded small RNA was able to activate ATF4-mediated transcription activation, similar assays were performed using a pBS/U6 construct for silencing murine caveolin-1 (43). Transfection of this construct failed to affect mitosin expression, nor did it influence the transactivation activity of ATF4 (data not shown). Collectively, these results indicated that mitosin functioned as a repressor of ATF4 in vivo. DISCUSSION We report that mitosin serves as a negative regulator of the transcription factor ATF4 through direct interaction. First, we cloned ATF4 in yeast two-hybrid screens using mitosin mutant, mitosin TG , as bait (Fig. 1). Direct interaction was further confirmed by pull-down assays using bacterially expressed mitosin mt10 , a mutant similar to mitosin TG (Fig. 1). On the other hand, ATF4 interacted with mitosin through a C-terminal region between residues 185-351 covering the DNA binding bZip domain in two-hybrid assays (Fig. 2) (28,31). Its leucine zipper region was required, but not sufficient, for the interaction (Fig.  2). The corresponding regions in other members of the ATF/ CREB family, however, failed to interact with mitosin (Fig. 2). Second, mitosin mutants affected the DNA binding activity of ATF4. Both mitosin mt10 and mitosin 10/KN were able to supershift the ATF4-DNA complex in EMSA (Fig. 3), while the mutants themselves showed no binding to the DNA probe (data not shown). In contrast, mitosin 10/NB disrupted the ATF4-DNA complex, implying that mitosin might regulate ATF4 by regulating its DNA binding activity. Third, mitosin levels significantly affected ATF4-mediated transcription activation. Overexpression of full-length mitosin down-regulated the transactivation activity of ATF4, while mitosin depletion manifested the opposite effect (Fig. 5).
Mitosin appears to contain two distinct ATF4-binding domains between residues 2484 -3113. Because mitosin 10 cells transfected with or without pUHD-mitosin were lysed and subjected to 3-12% gradient SDS-PAGE. Immunoblotting was performed with anti-mitosin antibody (upper panel). The blot was then stripped and reprobed with anti-FLAG antibody (lower panel) to show that the mitosin band in lane 2 also contained the FLAG epitope. B, protein levels in different samples. pCEP-ATF4 was cotransfected with the indicated plasmids into HEK293T cells. Immunoblotting was performed to visualize mitosin, ATF4, and ␤-actin. C, reporter assays for pCRE-ATF4X2. The indicated plasmids were cotransfected into HEK293T cells to assess their effects on the reporter activity. The amount of each plasmid (g) is also shown. The relative luciferase activity is presented in arbitrary units as mean Ϯ S.D. from two experiments. D, reporter assays for human cyclin A promoter. Results were averaged from two experiments. able to supershift the ATF4-DNA complex in EMSA (Fig. 3), the first domain is located between residues 2645 and 3113, which covered the core region critical for kinetochore targeting (40). Nevertheless, the core region alone was not sufficient for binding ATF4 (Fig. 1). On the other hand, disruption of the ATF4-DNA complex by mitosin 10/NB suggests existence of another ATF4 binding domain between 2484 and 2645 (Fig. 3), which was further confirmed by pull-down assays (Fig. 4). In contrast to the previous binding domain, this one abolishes ATF4-DNA interaction. Mutagenesis study further suggested that the leucine zipper motif between residues 2557 and 2592 of mitosin is involved in binding to ATF4 (Fig. 4). Persistence of the interaction in high salt conditions (500 mM NaCl) (Fig. 4) further supports hydrophobic interaction between ATF4 and mitosin 10/NB , very likely by forming leucine zippers. Despite this, this second binding domain appears recessive in the presence of the first one because mitosin mt10 , which contains both domains, did not abolish the ATF4-DNA complex (Fig. 3). Although we failed to achieve coimmunoprecipitation of ATF4 with full-length mitosin, possibly because of poor solubility of mitosin/CENP-F as a nuclear matrix protein in interphase (data not shown) (2), the influences of full-length mitosin on ATF4-mediated transcription (Fig. 5) strongly suggest that their interaction is not an artifact.
What would be the physiological roles of their interaction? Nuclear matrix proteins are implicated in transcription regulations (44 -47). As a nuclear matrix protein with multiple leucine heptad repeats (2), mitosin may serve as a scaffold protein in interphase to sequester ATF4 and possibly other transcription factors. Therefore, increased expression of mitosin in S-G 2 phases (3) may alter activities of ATF4, which in turn regulates expressions of downstream genes. A possible example is cyclin A expression, which increases after G 1 /S transition (48,49) and resembles that of mitosin (3). A CRE is found in both human and rat cyclin A promoters and contributes significantly to the cell cycle-dependent expression of cyclin A (38,49). ATF2 is a more potent transactivator for rat cyclin A promoter than ATF4 (38). Moreover, Jun family transcription factors complex with ATF2 and further largely stimulate cyclin A promoter, while ATF4 inhibits such effects (38). Contradictorily, ATF4 levels increase following the cell cycle progression from G 0 (38). Therefore, it is likely that, in intact cells, ATF4 is inhibited specifically by increasing levels of mitosin during S-G 2 phase to allow elevated activation of the CRE site in cyclin A promoter by ATF2 and associated Jun family transcription factors. In our hands, although ATF4-stimulated cyclin A promoter in unsynchronized HEK293T cells, its activity may not be as strong as that of ATF2, especially in S phase. Alternatively, such effect may be due to the specific cell line.
Mitosin may also regulate gene transcription through association with ATF4 and/or other transcription factors in other cellular activities, for instance, muscle differentiation (11,13,14). In addition, ATF4 is critical for neuron functions for longterm memory and synaptic plasticity (22)(23)(24). Lack of mitosin expression in brain (data not shown) probably enables efficient activation of ATF4 in neurons. These are interesting issues deserving further investigations.