Characterization of Homo- and Heterodimerization of Cardiac Csx/Nkx2.5 Homeoprotein*

Csx/Nkx2.5 is an evolutionarily conserved homeodomain (HD)-containing transcription factor that is essential for early cardiac development. We found that the HD of Csx/Nkx2.5 binds as a monomer as well as a dimer to its DNA binding sites in the promoter of the atrial natriuretic factor (ANF) gene, an in vivotarget gene of Csx/Nkx2.5. Csx/Nkx2.5 physically interacts with each other in vitro as well as in cells, and the HD is critical for homodimerization. Lys193 and Arg194, located at the COOH-terminal end of HD, are essential for dimerization. Lys193 is also required for a specific interaction with the zinc finger transcription factor GATA4. Csx/Nkx2.5 can heterodimerize with other NK2 homeodomain proteins, Nkx2.3 and Nkx2.6/Tix, with different affinities. A single missense mutation, Ile183 to Pro in the HD of Csx/Nkx2.5, preserved homodimerization function, but totally abolished DNA binding. Ile183 → Pro mutant acts in an inhibitory manner on wild type Csx/Nkx2.5 transcriptional activity through the ANF promoter in 10T1/2 cells. However, Ile183 → Pro mutant does not inhibit wild type Csx/Nkx2.5 function on the ANF promoter in cultured neonatal cardiac myocytes, possibly due to failure of dimerization in the presence of the target DNA. These results suggest that complex protein-protein interactions of Csx/Nkx2.5 play a role in its transcriptional regulatory function.

The homeodomain (HD) 1 -containing transcription factors, characterized by their 60-amino acid DNA binding domain, play critical roles in developmental patterning and differentiation. The HD forms three ␣-helices and contacts the major groove of DNA through the third helix (1). Contrary to the highly specific biological functions of individual homeobox genes, in vitro DNA binding studies have demonstrated that most HD proteins bind to similar short consensus sequences containing the TAAT motif (1,2). This apparent discrepancy may result from target gene's specificity for each HD protein in vivo being achieved by multiple mechanisms, such as interaction with other factors (3), small differences in DNA binding affinities to target sites (4), translational regulation of homeobox gene expression (5,6), subcellular localization (7), or protein phosphorylation (8,9).
Homo-or heterodimerization of transcription factors has been proposed to regulate transcriptional activity of many transcription factors. Combinatorial use of a limited number of transcription factors allows the regulation of a larger number of biological processes, increasing both the diversity as well as the specificity of control. However, a very limited number of studies have addressed the homo-and heterodimerization of HD-containing transcription factors among more than 400 members of HD proteins from yeast to human. Homodimerization ability has been demonstrated for Oct1 (10), Paired (11), Cdx2 (12), Even-skipped (13), Mix.1 (14), and Pit1 (15). Heterodimerization was shown for HNF1␣-HNF1␤ (16), Oct1-Oct2 (15), Mix.1-Siamois (14), MCM1-MAT␣2 (17,18), and Extradenticle-Ultrabithorax (19). The monomer of HD proteins is sufficient to interact with DNA, and the DNA-bound monomer recruits other partners to the complex (11,20). Most likely, the monomer HD proteins regulate through the monomeric DNA binding site, whereas homo-or heterodimerized HD proteins regulate through the dimeric sites. These differential interactions would provide more precise gene regulation at each developmental stage (21).
The biological significance of dimerization of paired-like class HD proteins has been demonstrated in Xenopus Mix.1, which regulates dorsal-ventral patterning (14). To date, the significance of homodimerization of HD-containing transcription factors has not been well established in mammals. Interestingly, 10 heterozygous mutations of human CSX/NKX2.5 were recently identified in patients with congenital heart disease. These patients show progressive atrioventricular conduction defects, left ventricular dysfunction, atrial septal defect, ventricular septal defect, and tetralogy of Fallot (22,23). Four * This work was supported by the Charles H. Foundation and American Heart Association Massachusetts Affiliate Fellowship and Beginning Grant-in-aid (to H. K.), an American Heart Association National Grant (to A. U.), and by National Institutes of Health (NIH) Grant R01-HL51253 and a Specialized Center for Research in Atherosclerosis in Congenital Heart Disease grant from NIH Grant P50-HL61036 (to S. I.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 1 The abbreviations used are: HD, homeodomain; bp, base pair(s); ANF, atrial natriuretic factor; HA, hemagglutinin; PCR, polymerase chain reaction; MBP, maltose-binding protein; EMSA, electrophoretic mobility shift assay; PMSF, phenylmethylsulfonyl fluoride; DTT, dithiothreitol; Ab, antibody; mAb, monoclonal antibody; aa, amino acid(s); mutations are single missense mutations in the HD that result in markedly reduced DNA binding (24), raising the possibility that if Csx/Nkx2.5 forms homodimers, the mutants with DNA binding defects may dominantly inhibit CSX/NKX2.5 function in human cardiac development and maturation. In Xenopus, injections of mRNA encoding non-DNA binding mutants of Xenopus XNkx2.3 and XNkx2.5 suppressed normal heart formation and resulted in a small heart or no heart formation in the most severe cases (25). This in vivo evidence suggests that a non-DNA binding mutant of Csx/Nkx2.5 may act in a dominant inhibitory manner on wild type Csx/Nkx2.5 through protein-protein interaction. Therefore, it is critical to examine whether Csx/Nkx2.5 proteins homo-or heterodimerize to regulate their transcriptional activity.
The NK2 class of HD proteins, first described in Drosophila (26), is highly conserved from nematode to human and is characterized by a unique Tyr residue at position 54 in the third helix of the HD. The most frequently observed HD binding motif is T/AAAT, but the NK2 class HD binds to the unique T/CAAG motif. The guanine nucleotide at the fourth position is distinct from all other HD⅐DNA complexes that usually have thymidine at this position, and 54 Tyr is responsible for the unique DNA recognition (27)(28)(29)(30).
The ANF promoter has been proposed to be a direct target of Csx/Nkx2.5 (42)(43)(44)(45)(46). Csx/Nkx2.5 binds at Ϫ87 and Ϫ242 bp sites, and transactivates the ANF gene synergistically with the zinc finger transcription factor GATA4 (42)(43)(44)(45). In this study, we demonstrate the homodimerization of Csx/Nkx2.5 at ANF Ϫ242 site and determine the critical amino acid residues for homodimerization as well as heterodimerization with GATA4. In addition, we generated a single missense mutation, Ile 183 to Pro in the HD of Csx/Nkx2.5, which preserved homodimerization function, but totally abolished DNA binding. Ile 183 3 Pro mutant acted in an inhibitory manner on wild type Csx/Nkx2.5 transcriptional activity through the ANF promoter depending on the cellular context.
Nuclear extracts of neonatal cardiac myocytes were prepared as follows: cells on 10-cm plates were washed with HBS buffer (25 mM HEPES, pH 7.6, 130 mM NaCl) and soaked in 3 ml of low salt buffer (25 mM HEPES, pH 7.6, 1 mM DTT, 0.1% Triton X-100, 0.5 mM PMSF, 1 mM sodium orthovanadate, 2 mM sodium pyrophosphate) for 10 min on ice. Cells were scraped with the low salt buffer and lysed by Dounce homogenizer. Nuclei pelleted by centrifugation at 200 ϫ g for 5 min were resuspended in 100 l of extraction buffer (20 mM HEPES, pH 7.6, 450 mM NaCl, 0.5 mM EDTA, 1 mM DTT, 25% glycerol, 0.1 mM PMSF, 1 mM sodium orthovanadate, 2 mM sodium pyrophosphate) and incubated for 30 min at 4°C. After centrifugation at 15,000 ϫ g for 10 min, the supernatant was used as a nuclear extract for EMSA using the same methods described above with 66 mM NaCl in the reaction.
Protein-DNA complexes analyzed by electrophoresis in 5% native polyacrylamide gel were transferred to polyvinylidene difluoride membrane with Tris-glycine-methanol buffer (25 mM Tris, 192 mM glycine, 20% methanol). The membrane was blotted with anti-Csx/Nkx2.5 Ab (40), anti-FLAG Ab (Sigma), or anti-HA Ab (Roche Molecular Biochemicals). EMSA in which proteins were transferred to membranes contains 0.3 g/l final concentration of nuclear extract from noninfected cardiac myocytes or 0.044 g/l final concentration of nuclear extract from adenoviral infected cardiac myocytes and 0.16 pmol/l final concentration of nonradiolabeled double-stranded oligonucleotides.
Protein-DNA binding affinity (K d ) was estimated by the protein concentration at which 50% of the DNA probe has become bound (49). Molecular mass of MBP fusion protein was estimated by addition of MBP protein (molecular mass ϭ 42 kDa) and HD (aa 122-212, molecular mass ϭ 10.9 kDa) or full length (molecular mass ϭ 32.8 kDa) or ⌬C (aa 1-250, molecular mass ϭ 27.5 kDa) or ⌬N (aa 122-318, molecular mass ϭ 21.6 kDa).
Recombinant Adenoviruses-FLAG-or HA-tagged wild type or FLAG-tagged Ile 183 3 Pro mutant was inserted into the shuttle vector pADloxp vector (50), creating pADloxp-Csx/Nkx2.5(FLAG-wild) or pADloxp-Csx/Nkx2.5(HA-wild) or pADloxp-Csx/Nkx2.5(FLAG-IP). One g of plasmids was cotransfected with 1 g of ⌿5 viral DNA into Cre8 cells to produce adenoviruses according to the methods reported previously (50). For control, ⌿5 viral DNA expressing no transgene was infected to 293 cells. The viral particle number was determined by plaque assays, and 5-15 multiplicity of infection was used for infection to neonatal rat cardiac myocytes prepared as described previously (51). The expression of wild type or Ile 183 3 Pro mutant protein was determined by Western blotting and immunostaining using anti-FLAG mAb (Sigma) or anti-HA mAb (Roche Molecular Biochemicals).
To perform coimmunoprecipitation assay, 293 cells in 100-mm plates were transfected with 9 g of pcDNA3-FLAG-Csx/Nkx2.5 and/or 9 g of pcDNA3-HA-Csx/Nkx2.5 using the calcium phosphate method. Total plasmid amount was adjusted with pcDNA3 empty vector to 18 g. Cells were lysed in the lysis buffer (20 mM HEPES, pH 7.5, 100 mM NaCl, 5 mM MgCl 2 , 0.5% Nonidet P-40, aprotinin (2 g/ml), pepstatin (0.7 g/ml), 0.1 mM PMSF, 1 mM DTT) and precleared with normal mouse IgG-bound protein G. Approximately 1 mg of protein in 1 ml of lysis buffer was incubated with 3 g of anti-FLAG mAb affinity gel (Sigma), washed five times with lysis buffer, and resolved on SDS-PAGE and subjected to Western blotting using peroxidase-conjugated anti-HA Ab (Roche Molecular Biochemicals).
Reporter Gene Assays-10T1/2 fibroblast cells cultured in six-well plates were cotransfected with 1.0 g of ANF(Ϫ638)-Luc reporter construct (provided by K. R. Chien), 0.4 g of Rous sarcoma virus ␤-galactosidase (provided by B. Markham), 0.4 g of pcDNA3-Csx/Nkx2.5 with or without 0.4 g or 0.8 g of pcDNA3-Csx/Nkx2.5(Ile 183 3 Pro) using the calcium phosphate method. 0.4 g of pAT2-GATA4 expression vector (provided by B. Markham) was cotransfected together with the plasmids listed above. Total plasmid amount was adjusted to 3.0 g with pcDNA3 vector plasmid. After glycerol shock using 1ϫ HEPES buffer containing 15% glycerol, cells were cultured for another 48 h, lysed with 300 l of reporter lysis buffer (Promega), and assayed for luciferase activity (Promega) and ␤-galactosidase activity.

Csx/Nkx2.5 Forms a Homodimer on a Palindromic DNA
Sequence in the ANF Promoter-The ANF promoter contains three specific binding sites for Csx/Nkx2.5 that are located upstream of the transcription start site at positions Ϫ408, Ϫ242, and Ϫ87 bp. At position Ϫ242 bp, the promoter contains two binding sites spaced by 5 nucleotides (Fig. 1A, panel a). Because Csx/Nkx2.5 contacts DNA through its HD, we analyzed and compared the HD binding affinity for two sites in the ANF promoter at positions Ϫ242 bp (ANF Ϫ242) and Ϫ87 bp (ANF Ϫ87). Applying EMSA, we determined that the HD protein binds to the ANF Ϫ242 site with a K d ϭ 1 ϫ 10 Ϫ9 M (Fig. 1A, panel b), which is slightly lower than the previously reported K d for the related Drosophila HD protein NK2 (52), but in the affinity range for the HD protein (K d in the range of 10 Ϫ9 to 10 Ϫ10 M) (53). However, for the protein-DNA interaction with the ANF Ϫ87 site, we estimated a K d ϭ 8.2 ϫ 10 Ϫ8 M (Fig.  1A, panel c). Therefore, the HD bound to the ANF Ϫ242 site with more than 80 times higher affinity than to the ANF Ϫ87 site. Interestingly, we also observed that Csx/Nkx2.5 forms an additional specifically shifted band with a migration most likely corresponding to the occupation of the two specific DNA binding sites of the ANF Ϫ242 at higher protein concentrations (Fig. 1A, panel b, lanes 3-6). We asked whether the protein concentration at which the second band appears is physiologically relevant. Csx/Nkx2.5 forms a dimer at a protein concentration of 3.1 ϫ 10 Ϫ9 M (lane 3), which is comparable with the concentrations reported for other transcription factors (54 -56).
To confirm that the appearance of the second shifted band is a result of the simultaneous occupation of two specific binding sites of the ANF Ϫ242, we performed EMSA using an oligonucleotide in which one of the DNA binding sites was deleted from the ANF Ϫ242 site. We found that HD bound to the mutated site (converted to a single DNA binding site) as a monomer (M in Fig. 1B) without showing the slow migrating bands. The binding affinity to the monomeric DNA binding site was slightly reduced (K d ϭ 1-3.1 ϫ 10 Ϫ9 M). Furthermore, we found that when the HD was mixed with the full-length Csx/Nkx2.5, the HD and the full-length protein produced newly shifted bands migrating to an intermediate position between the HD and full-length homodimers (Fig. 1C, HDϩFull). These data indicate that the Csx/Nkx2.5 protein binds with a higher affinity to the palindromic ANF Ϫ242 site than to the mutated monomeric ANF Ϫ242 site or the monomeric ANF Ϫ87 site. Most likely Csx/Nkx2.5 forms a dimer on ANF Ϫ242 site, and dimerization stabilizes the protein-DNA interaction (see below).
Homodimerization of Csx/Nkx2.5 in Vitro and in Cells-We next examined whether Csx/Nkx2.5 proteins physically and specifically interact with each other in the absence of DNA. MBP-fused Csx/Nkx2.5 ( Fig. 2A, lane 1 To examine whether Csx/Nkx2.5 homodimerizes in cells, we cotransfected FLAG epitope-tagged Csx/Nkx2.5 expression plasmid with HA epitope-tagged Csx/Nkx2.5 expression plasmid into the human embryonic kidney carcinoma cell line 293 and confirmed that both Csx/Nkx2.5 proteins coimmunoprecipitated with anti-FLAG Ab (Fig. 2B, lane 1). Thus, Csx/ Nkx2.5 can homodimerize in solution as well as in cells, and binding to DNA is not required for this interaction.
Homodimerization of Endogenous Csx/Nkx2.5 on the ANF Ϫ242 Site-To examine whether endogenous Csx/Nkx2.5 homodimerizes on the ANF Ϫ242 site, nuclear extracts prepared from neonatal rat cardiac myocytes were used for EMSA. As shown in Fig. 3A, lane 1, endogenous Csx/Nkx2.5 forms monomers (M) as well as dimers (D). These two bands corresponded to the bands shifted by the nuclear extract from adenovirus infected rat cardiac myocytes that expressed FLAG-tagged Csx/ Nkx2.5 (Fig. 3A, lanes 2-4). Because of the high expression levels of Csx/Nkx2.5 in the adenoviral vector-infected cardiac myocytes, 20-fold dilution of the nuclear extracts was required to shift the DNA probe to a similar level compared with that of the endogenous Csx/Nkx2.5 (Fig. 3A, lane 1 versus lane 3). When the protein-DNA complex was transferred to the membrane and blotted with anti-Csx/Nkx2.5 Ab, we detected the signal at the monomeric (M) and dimeric (D) protein-DNA complex both in the uninfected and the virus infected nuclear extracts (Fig. 3B).
To ascertain whether Csx/Nkx2.5 protein forms dimers with DNA in cardiac myocytes, adenovirus-encoding FLAG-tagged Csx/Nkx2.5 and/or HA-tagged Csx/Nkx2.5 were coinfected into cardiac myocytes, and the nuclear extracts were mixed with the DNA probe for EMSA analysis (Fig. 3C). The protein-DNA complex was transferred to a membrane followed by Western blot analysis (Fig. 3D). We detected the signal at the monomeric (M) and dimeric (D) protein-DNA complex, similar to Fig.  3B. Additional slow migrating bands observed in these experiments (* in Fig. 3D) corresponded to Csx/Nkx2.5 protein unbound DNA (lanes 4 -6 in Fig. 3D). Both FLAG-tagged and HA-tagged Csx/Nkx2.5 were detected at the dimeric protein-DNA complex when both proteins were coexpressed in cardiac myocytes (Fig. 3D, lane 2), suggesting that two Csx/Nkx2.5 molecules homodimerize on DNA.
Lys 193 -Arg 194 within the HD Is Required for Dimerization-To confirm the specificity and to identify the regions that are required for dimerization, we mapped the dimerization domain of Csx/Nkx2.5 using in vitro binding assays. Initially, four [ 35 S]methionine-labeled COOH-terminal deletion mutants were mixed with MBP-HD (Fig. 4A). Two COOH-terminal deletion mutants of Csx/Nkx2.5, 1-230 and 1-199, associated with MBP-HD, whereas the further deletions to 1-159 or 1-149 abolished the association (Fig. 4B, top panel). These results indicate that amino acids between 159 and 199 are necessary for dimerization. Next, 5 amino acid serial deletion mutants from the carboxyl terminus of HD, 1-196, 1-191, 1-186, and 1-181, were examined. The 1-196 protein interacted with the HD, but the 1-191, 1-186, and 1-181 proteins did not (Fig. 4B,  middle panel). Further single amino acid deletions revealed that 1-193 dramatically reduced the interaction, and 1-192 completely abolished the interaction (Fig. 4B, bottom panel). Therefore, two basic amino acids, Lys 193 and Arg 194 are necessary for the interaction with the HD.
We further mutated Lys 193 and Arg 194 into neutral or acidic amino acids (Lys 193 to Ile, Arg 193 to Ile, and Lys 193 -Arg 194 to Ile 193 -Asp 194 ) and examined them for dimerization with the HD as well as with full-length Csx/Nkx2.5 (Fig. 4C). The Lys 193 3 Ile mutant markedly reduced the interaction with the HD, and an ϳ50% reduction was observed in Arg 194 3 Ile mutant. The interaction with the HD was undetectable when both amino acids were mutated (Fig. 4C, HD). However, we still detected a weak interaction between Lys 193 -Arg 194 mutant and fulllength Csx/Nkx2.5 (Fig. 4C, Full). These findings confirm that two amino acids Lys 193 and Arg 194 , are required for the dimerization of the HD, and additional protein domain(s) outside of HD are also likely to be involved in dimerization.
Involvement of the Region(s) Outside of HD for Dimerization-To identify the domain(s) outside of the HD of Csx/ Nkx2.5 that are involved in homodimerization on DNA, we examined DNA binding affinity of HD and full-length protein on the palindromic ANF Ϫ242 site or a mutated monomeric ANF Ϫ242 site shown in Fig. 1. The HD protein bound DNA predominantly as a monomer at a low protein concentration (Fig. 5A, panel a, lanes 1-3) and dimerized more at a higher protein concentration (Fig. 5A, panel a, lanes 5 and 6). The monomer to dimer transition was observed between lanes 4 and 5 at a protein concentration of 0.91-2.7 ϫ 10 Ϫ8 M in the HD protein (arrow in Fig. 5A, panel a). However, with the fulllength Csx/Nkx2.5 (Fig. 5A, panel b), the monomer-dimer transition occurred more abruptly between lane 2 and lane 3 (protein concentration 0.71-2.1 ϫ 10 Ϫ9 M). We tested and confirmed that HD and full-length protein bound to the mutated monomeric site with similar affinity (K d ϭ 1.0 -3.0 ϫ 10 Ϫ9 M for HD; 0.71-2.1 ϫ 10 Ϫ9 M for full-length Csx/Nkx2.5) (Fig. 5A, panels  c and d).
To further examine the regions responsible for dimeric DNA binding, we constructed two deletion mutants, a carboxyl terminus deletion mutant (1-250) and an amino-terminal deletion mutant (122-318) and examined their DNA binding on the ANF Ϫ242 site (Fig. 5B). In the COOH-terminal deletion mu-tant (Fig. 5B, panel a), the monomer-dimer transition was observed between lanes 4 and 5 (0.66 -2.0 ϫ 10 Ϫ8 M), which was similar to that of HD protein. The amino-terminal deletion showed the monomer-dimer transition between lanes 3 and 4 (2.4 -7.3 ϫ 10 Ϫ9 M) (Fig. 4B, panel b); therefore it required 3-fold lower protein concentration than that of the HD or the carboxyl terminus deletion, but still required 3-fold higher protein concentration than that of the full-length protein.
Taken together, although the HD and full-length Csx/Nkx2.5 binds the monomeric DNA binding site with a similar affinity, full-length Csx/Nkx2.5 preferentially forms dimers at ϳ13-fold lower protein concentration than the HD alone. Thus, regions outside of the HD, particularly the COOH-terminal region of Csx/Nkx2.5, seem to facilitate protein-protein interactions involved in the dimerization on DNA.
Lys 193 Is Necessary for Association with GATA4 -We and others have reported that Csx/Nkx2.5 interacts with the transcription factor GATA4 (42)(43)(44)(45). It was demonstrated that the  1-3), FLAG-tagged Csx/Nkx2.5 or HA-tagged Csx/Nkx2.5 protein migrated faster than that without DNA (* in lanes 4 -6) and was detected in the monomeric (M) or dimeric (D) protein-DNA complex. Both FLAG-tagged and HA-tagged Csx/Nkx2.5 were detected in the dimeric protein-DNA complex (lane 2) when both proteins were coexpressed in cardiac myocytes. second zinc finger of GATA4 is involved in the specific interaction with the HD of Csx/Nkx2.5, and amino acids between 182 and 199 are responsible for the direct interaction with GATA4 (43). Our data presented in Fig. 4 revealed that this domain is also responsible for homodimerization. Therefore, we examined whether GATA4 associates with Lys 193 -Arg 194 mutants (Fig.  6A). As shown in Fig. 6A, lane 1, the wild type Csx/Nkx2.5 (1-318) associated with GATA4-GST protein, whereas Lys 193 3 Ile (lane 2) and Lys 193 3 Ile/Arg 194 3 Asp (lane 4) mutants abolished the interaction. Interestingly, the Arg 194 3 Ile mutant (lane 3) associated with GATA4 with an apparent higher affinity than wild type Csx/Nkx2.5, in contrast to its lower homodimerization ability (Fig. 4C). These data demonstrate that Lys 193 in the HD of Csx/Nkx2.5, which is critical for homodimerization, is also essential for the interaction with GATA4.
Generation of an Inhibitory Mutant-To examine the effects of protein dimerization on transcriptional activity, we attempted to create a mutant protein that does not dimerize, but does bind, DNA. As shown in Fig. 4, we constructed Csx/Nkx2.5 mutants that do not dimerize to the HD. We next examined the DNA binding of these mutants using ANF Ϫ242 site and found that the Lys 193 3 Ile mutant completely abolished DNA binding (Fig. 6A). Lys 193 -Arg 194 mutant bound DNA, but the binding affinity was significantly lower than that of wild type Csx/ Nkx2.5 (Fig. 7A).
As an alternative to examine the effect of protein dimerization, we generated a converse mutant in which protein dimerization is preserved, but DNA binding is abolished. By mutating Ile 183 in the third helix of HD into Pro, DNA binding of Csx/Nkx2.5(Ile 183 3 Pro) mutant was completely abolished (Fig. 7B), but this mutant associated with MBP-Csx/Nkx2.5 protein with a similar affinity as that of wild type protein (Fig.  7C, lane 1 versus lane 2). In contrast, this Ile 183 3 Pro mutant markedly reduced the interaction with GATA4 (Fig. 7C,  lane 6).
We tested the function of the Ile 183 3 Pro mutant by transient transfection assays in 10T1/2 fibroblasts using ANF(Ϫ638)-luciferase reporter construct (ANF-Luc), which includes the Ϫ87 and Ϫ242 bp sites shown in Fig. 1A. The Ile 183 3 Pro mutant did not bind DNA (Fig. 7B), and the mutant itself did not activate or repress the ANF-Luc (data not shown). When we cotransfected the expression plasmid encoding Ile 183 3 Pro mutant protein with wild type Csx/Nkx2.5 at 1:1 ratio (0.4 g), the luciferase activity of wild type Csx/Nkx2.5 de- creased by ϳ44%. A slight further reduction (ϳ53%) of ANF-Luc activity was observed when the Ile 183 3 Pro expression plasmid was increased to 2:1 (0.8 g) (Fig. 7D). In the presence of GATA4 expression plasmid, we observed a further increase of ANF-Luc activity from the ANF promoter as reported previously (43). We found that the Ile 183 3 Pro mutant reduced luciferase activity by ϳ20% at 1:1 ratio of plasmid amount and by ϳ44% at 2:1 ratio. These data demonstrate that the non-DNA binding mutant, Ile 183 3 Pro, acts in an inhibitory manner on wild type Csx/Nkx2.5 in transient transfection assays in 10T1/2 cells.
Ile 183 3 Pro Mutant Does Not Inhibit the Csx/Nkx2.5-dependent ANF Promoter Activation in Neonatal Cardiac Myocytes-We further examined the inhibitory effect of the Ile 183 3 Pro mutant on endogenous Csx/Nkx2.5 as well as wild type Csx/Nkx2.5 in cultured neonatal cardiac myocytes. In rat neonatal cardiac myocytes, the base-line ANF-Luc activity was high. When we used the LipofectAMINE transfection method, the base-line ANF-Luc activity was approximately the same as that detected in 10T1/2 cells transfected with the wild type Csx/Nkx2.5 expression plasmid. ANF-Luc activation was suppressed by the cotransfection of Ile 183 3 Pro expression plasmid by 29%. (Fig. 8A, 183 I-P). When we cotransfected the wild type Csx/Nkx2.5 expression plasmid, ANF-Luc activity was increased by 50% (Fig. 8A, Wild). However, cotransfection of Ile 183 3 Pro expression plasmid did not inhibit the wild type Csx/Nkx2.5 function (Fig. 8A, 183 I-PϩWild). Similar results were obtained using the calcium phosphate methods (data not shown).
Transfection efficiency of primary cardiac myocytes is known to be very low when plasmid vectors are used. Therefore, we infected cardiac myocytes with adenoviral vectors, which exhibit a high efficiency of gene transfer. More than 90% of cardiac myocytes expressed either Ile 183 3 Pro mutant or wild type Csx/Nkx2.5 (Fig. 8B), and each construct expressed a similar protein amount determined by Western blotting (Fig.  8C). Twenty-four hours after adenovirus infection, ANF-Luc reporter gene was transfected into cardiac myocytes, and the transcriptional activation was measured by luciferase activity. When Ile 183 3 Pro mutant protein was expressed by the adenoviral vector, ANF-Luc activity was suppressed by 31% (Fig.  8D, 183 I-P). In contrast, wild type Csx/Nkx2.5 activated the ANF-Luc reporter by 3.5-fold (Fig. 8D, Wild), which was not suppressed by coexpression of Ile 183 3 Pro mutant (Fig. 8D, 183 I-PϩWild). When we examined DNA binding of the nuclear extract from cardiac myocytes expressing wild type alone (Fig.  8E, panel a) or wild type with Ile 183 3 Pro mutant (Fig. 8E,  panel b), there was no significant difference in DNA binding of wild type Csx/Nkx2.5 either as monomers or dimers. These data indicate that the expression of Ile 183 3 Pro mutant weakly suppresses ANF-Luc activity in cardiac myocytes; however, Ile 183 3 Pro mutant does not suppresses ANF-Luc activity or reduce DNA binding of wild type Csx/Nkx2.5 when coexpressed with wild type Csx/Nkx2.5 (see "Discussion").

DISCUSSION
The NK2 class homeobox-containing transcription factor Csx/Nkx2.5 is one of the earliest cardiogenic markers from insects to vertebrates (28,31,34,38,39,57,58). Recently, human CSX/NKX2.5 mutations were identified in patients with congenital heart disease (22,23). These patients show progressive conduction delays and in some cases left ventricular dysfunction after birth, suggesting that Csx/Nkx2.5 also functions in the later stages of heart development and maturation. While evidence is accumulating that Csx/Nkx2.5 mutations inhibit normal cardiac development and maturation in both humans and Xenopus (22,23,25), the molecular mechanisms for these phenotypes remain to be explained. In this study, we report the new finding of protein dimerization of Csx/Nkx2.5, which may yield insights into the dominant effects of CSX/NKX2.5 mutations found in humans.
Csx/Nkx2.5 binds to the palindromic Csx/Nkx2.5 consensus binding sites in the ANF promoter as a monomer as well as a dimer, and dimer formation increases the protein-DNA binding affinity. Csx/Nkx2.5 physically interacts with each other in vitro as well as in a cell. Two basic amino acids, Lys 193 -Arg 194 , located in the third helix of HD are necessary for dimerization, and Lys 193 is also indispensable for the association with GATA4, which is a cofactor for Csx/Nkx2.5 function. To examine the functional significance of dimerization, we generated a converse mutant (Ile 183 3 Pro) that does not bind DNA, but has preserved homodimerization ability. In transient transfection assays in 10T1/2 fibroblast cells, this mutant acts in an inhibitory manner on wild type Csx/Nkx2.5. Also, this mutant suppresses the ANF-Luc activity in cardiac myocytes, but does not inhibit the ANF-Luc activity that is further induced by wild type Csx/Nkx2.5.
Protein Dimerization of Csx/Nkx2.5 through the HD-By using various deletion and point mutants, we found that two positively charged amino acids, Lys 193 -Arg 194 at the COOHterminal end of HD (Lys 57 -Arg 58 in HD), are critical for proteinprotein interaction. As expected, these two amino acids were highly conserved within NK2 class HD (29). It was shown that three amino acids (Arg 58 -Val 59 -Lys 60 ) located at the carboxyl end of the HD of Pit1 are involved in the homodimerization on DNA by forming a protein-protein interface with the POUspecific domain (15). In contrast to Csx/Nkx2.5, Pit1 requires DNA to homodimerize (59).
The Lys 193 -Arg 194 mutation, located at the carboxyl end of the HD in Csx/Nkx2.5, markedly reduced DNA binding, which is consistent with the NMR structure of another NK2 class HD protein Drosophila NK-2. The third ␣-helix (helix III) of NK-2 extends up to amino acid 62 in the presence of DNA (52,60). We demonstrated that Lys 193 is required for the Csx/Nkx2.5 and GATA4 interaction as well as for homodimerization of Csx/Nkx2.5.
Regions Outside the HD Facilitate Dimerization on DNA-Cooperative dimerization of HD proteins has been characterized in paired and paired-like HD proteins (11). Paired HD proteins cooperatively bind DNA with a palindromic TAAT sequence separated by 3 bp. The presence of Arg 28 or Arg 43 prevents cooperative dimerization, and paired class HD proteins do not have Arg residues at the 28 and 43 positions (11). In contrast, 50% of HD proteins have conserved Arg 28 or Arg 43 residues among ϳ350 HD proteins (53). NK class HD proteins, as well as engrailed, bcd, POU, and msh class proteins do not have Arg 28 or Arg 43 , suggesting the possibilities for cooperative dimerization in these classes of HD proteins.
In Csx/Nkx2.5, regions outside of the HD, particularly the region carboxyl terminus to the HD (aa 251-318), appear to facilitate cooperative dimerization on DNA. Compared with the DNA binding of HD of Csx/Nkx2.5 or the COOH-terminal deletion mutant, we found that the full-length protein facilitated dimerization, and the monomer-dimer transition occurred at ϳ13-fold lower protein concentrations than that of the HD or the carboxyl terminus deletion mutant. Our observation that Csx/Nkx2.5 homodimerizes through the HD as well as outside of the HD supports the hypothesis that protein-protein interactions play important roles in cooperative dimerization. Alternatively, a full-length Csx/Nkx2.5 molecule may "bend" DNA to facilitate the binding of a second Csx/Nkx2.5 molecule to DNA, or full-length Csx/Nkx2.5 proteins may be more stable than HD proteins in a dimerized form on DNA. It is also possible that these effects may function cooperatively to regulate the transcriptional activation of target genes.
another NK2 class HD protein Vnd (61). Also, Tinman, the Drosophila homologue of Csx/Nkx2.5, binds the sequence located at Ϫ5.4 kilobases of the Dmef2 gene, which contains two NK2 binding sites separated by 165 bp (62). Mutations that disrupt either one of two Tinman binding sites caused loss of activation of the Dmef2 gene, leading to the hypothesis that the physical interaction of Tinman molecules occurs by looping of the 165-bp intervening segment (62). Although it has yet to be shown that Tinman protein homodimerizes, our data are consistent with this hypothesis.
Heterodimerization with Other NK2 Class Proteins-Several NK2 class HD proteins are coexpressed both temporally and spatially, suggesting that NK2 class HD proteins may heterodimerize. As shown in Fig. 6, Csx/Nkx2.5 and Nkx2.6/Tix associated with each other, but the association of Csx/Nkx2.5 and Nkx2.3 was significantly weaker. Although further quantitative analyses are necessary, these results suggest that NK2 class HD proteins potentially interact with each other, and the affinity of the interaction is different depending on heterodimer pairs. In this study, we did not examine the possible heterodimerization of Csx/Nkx2.5 with other classes of HD proteins. Of note, mouse Nkx2.3 is not expressed in the heart, and Nkx2.6/Tix expression is restricted to the sinus venosa and outflow tract of the mouse heart. Other NK2 class HD proteins coexpressed temporally and spatially similar to Csx/Nkx2.5 in the heart have not been described in mouse and human (29,41,63).
The Effect of Non-DNA Binding Mutant on Wild Type Csx/ Nkx2.5-Based on the studies of phenotypes caused by the non-DNA binding mutants of Csx/Nkx2.5 in patients and Xenopus (22)(23)(24)(25), and on the evidence for the formation of homodimers of Csx/Nkx2.5 in our study, non-DNA binding mutants might act in a dominant inhibitory manner. We generated a single missense mutation in the third helix of the HD, Csx/Nkx2.5(Ile 183 3 Pro), which abolishes DNA binding, but preserves dimerization ability. The interaction between GATA4 and Csx/Nkx2.5(Ile 183 3 Pro) was significantly weaker than that of wild type Csx/Nkx2.5; therefore, it is likely that Csx/Nkx2.5(Ile 183 3 Pro) will not sequester GATA4 from wild type Csx/Nkx2.5 when it is overexpressed. The transcriptional activation of wild type Csx/Nkx2.5 on ANF(Ϫ638) promoter was indeed suppressed by the coexpressed Csx/Nkx2.5(Ile 183 3 Pro) in a dose-dependent manner in 10T1/2 cells. ANF(Ϫ638) promoter activity was slightly suppressed by  (43). The Ile 183 3 Pro mutant reduced ANF-Luc by ϳ20% at a 1:1 ratio (0.4 g) and by 44% at a 2:1 ratio (0.8 g). Values are means Ϯ S.E. Csx/Nkx2.5(Ile 183 3 Pro) in neonatal cardiac myocytes where endogenous Csx/Nkx2.5 is expressed. However, when both wild type and Csx/Nkx2.5(Ile 183 3 Pro) mutants were overexpressed in cardiac myocytes, transcriptional activation by wild type was not suppressed by the Csx/Nkx2.5(Ile 183 3 Pro) mutant. Thus, unlike in 10T1/2 cells, Csx/Nkx2.5(Ile 183 3 Pro) does not seem to act as a typical dominant inhibitory mutant on the ANF(Ϫ638) promoter in cultured cardiac myocytes. The EMSA using cell lysates prepared from adenovirus-infected cardiac myocytes revealed that coexpression of Csx/ Nkx2.5(Ile 183 3 Pro) mutant does not inhibit the specific binding of wild type Csx/Nkx2.5 to the ANF Ϫ242 site (Fig. 8E). Since the Csx/Nkx2.5(Ile 183 3 Pro) mutant expressed in cardiac myocytes did not bind to the ANF Ϫ242 site (data not shown), it is possible that Csx/Nkx2.5(Ile 183 3 Pro) mutant loses the ability to form dimers with wild type Csx/Nkx2.5 on DNA and, therefore, does not inhibit the function of wild type Csx/Nkx2.5 on the ANF promoter in cardiac myocytes. It is also possible that the inhibitory effect of Csx/Nkx2.5(Ile 183 3 Pro) observed in 10T1/2 cells may occur through a mechanism in-dependent of wild type Csx/Nkx2.5. Mutant Csx/Nkx2.5 protein may squelch a transcription factor that is critical for the ANF(Ϫ638) promoter activity in 10T1/2 cells.
Protein homodimerization of Csx/Nkx2.5 yields the potential for it to precisely regulate a number of genes by utilizing monomeric and dimeric binding. It is possible that the genetically dominant effect of the human CSX/NKX2.5 missense mutations (22)(23)(24) may in part be due to an inhibitory effect of the mutant protein over the wild type protein on target genes that require dimeric binding.