Crystal Structure of the CCAAT Box/Enhancer-binding Protein β Activating Transcription Factor-4 Basic Leucine Zipper Heterodimer in the Absence of DNA*

The crystal structure of the heterodimer formed by the basic leucine zipper (bZIP) domains of activating transcription factor-4 (ATF4) and CCAAT box/enhancer-binding protein β (C/EBPβ), from two different bZIP transcription factor families, has been determined and refined to 2.6 Å. The structure shows that the heterodimer forms an asymmetric coiled-coil. Even in the absence of DNA, the basic region of ATF4 forms a continuous α-helix, but the basic region of C/EBPβ is disordered. Proteolysis, electrophoretic mobility shift assay, circular dichroism, and NMR analyses indicated that (i) the bZIP domain of ATF4 is a disordered monomer and forms a homodimer upon binding to the DNA target; (ii) the bZIP domain of ATF4 forms a heterodimer with the bZIP domain of C/EBPβ that binds the cAMP response element, but not CCAAT box DNA, with high affinity; and (iii) the basic region of ATF4 has a higher α-helical propensity than that of C/EBPβ. These results suggest that the degree of ordering of the basic region and the fork and the dimerization properties of the leucine zipper combine to distinguish the structurally similar bZIP domains of ATF4 and C/EBPβ with respect to DNA target sequence. This study provides insight into the mechanism by which dimeric bZIP transcription factors discriminate between closely related but distinct DNA targets.

Protein-protein interactions are involved in a number of different cellular processes, including the regulation of transcription. Many, if not most, transcription factors, including the bZIP 1 proteins, bind their cognate DNA elements as dimers. Dimerization between transcription factors from the same or different families can generate considerable functional diversity using only a relatively small number of components. By binding with different specificity to distinct DNA sites within promoters, heterodimeric transcription factors facilitate the positioning and orientation of proteins on DNA, thus providing distinct surfaces for interaction with other transcriptional regulatory proteins bound to adjacent DNA sites.
Among coiled-coil proteins, bZIP family members form dimers via their characteristic leucine zipper helices, which consist of seven-residue (heptad) repeats, (abcdefg) n . Dimer formation by bZIP proteins is selective: some form only homodimers, some form only heterodimers, and others form both homo-and heterodimers. Hydrophobic interactions by leucine and other hydrophobic amino acids in positions 3 and 4 in the helix (the a and d positions of the helical wheel) form the hydrophobic core of the bZIP dimer. Leucine residues occupy most of the d positions in the leucine zipper helix. At the same time, hydrophilic amino acids located just outside of the hydrophobic core (the e and g positions of the helical wheel) also participate in stabilizing the dimer. A highly conserved asparagine amino acid in the a position in the middle of the leucine zipper helix has been shown to destabilize oligomer formation in several bZIP dimers (1). Thus, depending on the precise distribution of hydrophilic and hydrophobic residues in the bZIP helix, only selective pairs of bZIP transcription factors are able to form stable dimers, which in turn lead to distinct DNA binding propensity.
The basic DNA-binding region of the bZIP domain adopts a stable structure upon association with DNA. Binding to the cognate DNA induces a coil-to-helix transition of the basic DNA-binding region (2)(3)(4). NMR and CD studies of GCN4 and C/EBP bZIP domains have shown that in the absence of the DNA target, the basic region has some residual helical character and has been described as an ensemble of transiently formed helical structures (4 -6). The folding transition involved in the DNA binding of the GCN4 basic region results in an unfavorable contribution to the overall free energy from the loss of conformational entropy (6).
The bZIP transcription factor family, one of many dimeric protein families, participates in a number of different transcriptional activities. The DNA-binding motif, consisting of basic and leucine zipper regions, is one of the simplest DNAbinding domains. However, the bZIP transcription factors are capable of recognizing a broad range of DNA sequences, yet at the same time, discriminate sufficiently to regulate the transcription of a diverse range of genes in different promoters. Based on DNA-binding activity, the bZIP transcription factors in mammals are divided into three major families: (i) activator protein-1 (AP-1), (ii) cAMP response element-binding protein/ activating transcription factor (CREB/ATF), and (iii) C/EBP. AP-1 family members have been the most heavily studied. In particular, structural analyses of bZIP domains of AP-1 proteins such as Fos/Jun and GCN4 in the absence and presence of the AP-1 DNA target (5Ј-TGAGTCA-3Ј) and cAMP response element (CRE; 5Ј-TGACGTCA-3Ј) have led to a general understanding of bZIP DNA binding.
Among bZIP transcription factors, the CREB/ATF family appears to have a higher selectivity for DNA target sites. As homodimers, CREB/ATF proteins recognize CRE, a palindrome of two inverted identical half-sites (TGAC⅐GTCA), which differs from the AP-1 target site by 1 base pair in the center. The AP-1 target is also a palindrome consisting of two inverted identical half-sites with 1 overlapping base pair in the middle (TGACTCA) (7,8). CREB/ATF proteins in heterodimers formed between CREB/ATF proteins and AP-1 (9 -13) or C/EBP proteins (14 -16) retain their affinity to bind to the CRE half-site; at the same time, they lead their partner to bind to CRE or CRE-related half-sites with high affinity. Members of the C/EBP family bind to a relatively broad range of DNA sequences that satisfy the (A/G)TTGCG(C/T)AA(C/T) consensus CCAAT box (17).
Although a number of three-dimensional structures of bZIP domains in the presence of DNA targets have been studied (18 -20), the three-dimensional structure of a bZIP homo-or heterodimer containing the basic regions has not been studied in the absence of a DNA target. To obtain a better understanding of dimerization properties and the conformational changes to the dimer upon DNA binding, we determined the crystal structure of the bZIP heterodimer formed between ATF4 and C/EBP␤. Furthermore, we used circular dichroism and NMR to show that the bZIP domain of ATF4 does not form homodimers, but does form heterodimers with C/EBP␤, a member of the C/EBP family. The heterodimer does not bind to the CCAAT box, but binds to CRE with high affinity. Unexpectedly, the crystal structure of the C/EBP␤⅐ATF4 heterodimer revealed that the basic region of the bZIP domain of the ATF4 subunit is an ␣-helix, whereas the basic region of C/EBP␤ is disordered. These results suggest that a degree of ordering of the DNA-binding region and the dimerization properties of the bZIP domains are important in DNA recognition by bZIP transcription factors.

EXPERIMENTAL PROCEDURES
Protein Purification-The NdeI-BamHI fragments containing the coding sequences for residues 280 -341 of human ATF4 (8) and residues 224 -285 of mouse C/EBP␤ (21) were generated by polymerase chain reaction from the full-length cDNAs and cloned into a pET11a plasmid. Both proteins were separately overexpressed in Escherichia coli. The bZIP domain of C/EBP␤ was purified from the insoluble fraction. The inclusion bodies were solubilized in Buffer A (20 mM Tris (pH 7.5), 1 mM EDTA, and 20 mM ␤-mercaptoethanol) containing 6 M urea and applied to an SP-Sepharose column (Amersham Pharmacia Biotech). The column was washed with Buffer A containing 6 M urea (10 column volumes) and then with Buffer A (10 column volumes). Renatured C/EBP␤ bZIP domain was eluted with a gradient of 0 -1 M NaCl in Buffer A. Protein was precipitated on ice and pelleted by centrifugation. The resulting pellet was dissolved in water and used for crystallization trials.
The ATF4 bZIP domain was purified from the soluble fraction. The cells were frozen and thawed in 20 mM Tris (pH 8.0), 0.1 M NaCl, 1 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 5 mM ␤-mercaptoethanol, and 10 mM Na 2 S 2 O 3 . The soluble fraction was incubated for 20 min at 70°C; the precipitate was removed by centrifugation, whereas the supernatant was applied to a Q-Sepharose column (Amersham Pharmacia Biotech). The protein was eluted with a linear gradient of 20 -800 mM NaCl in Buffer A. The fractions containing the ATF4 bZIP domain were diluted 5-fold and applied to an SP-Sepharose column. The protein was eluted with a linear gradient of 20 mM to 1 M NaCl in Buffer A. The protein peak was pooled and concentrated further in a Centricon-10 concentrator (Amicon, Inc.). For CD and NMR spectroscopy studies, Cys-310 of ATF4 was substituted with Ser to prevent the cysteine oxidation and formation of high molecular mass aggregates.
Oligonucleotides and Oligopeptides-Oligonucleotides were synthesized at the Vanderbilt Core Facility. Oligopeptides were synthesized by Alpha Diagnostic International Inc. (San Antonio, TX).
Circular Dichroism Spectroscopy-CD spectra were recorded at 20°C using a Jasco-720 spectropolarimeter. All spectra were run using a 1-mm path length quartz cuvette. Protein samples contained 10 M protein, 10 mM Tris (pH 7.5), 100 mM NaCl, 5 mM MgCl 2 , and 0 -50 M DNA. The synthetic peptide samples contained 50 M peptide, 10 mM Tris (pH 7.5), 100 mM NaCl, 5 mM MgCl 2 , and 0 -80% 2,2,2-trifluoroethanol (TFE). The CD spectra were averaged from three wavelength scans, blanked, smoothed, and analyzed using the manufacturer's software. The concentrations of the proteins and synthetic peptides used for experiments were determined by amino acid analysis.
NMR Spectroscopy-Lyophilized proteins were dissolved in 0.5 ml of 90:10 (v/v) H 2 O/D 2 O at 1.5 mM. The pH was adjusted to 6.5 without correction for deuterium isotope effect. A series of homonuclear twodimensional NMR experiments were recorded at 15, 25, 35, 45, and 55°C. Experiments were done on a DRX600 (14.2 T) instrument (Bruker) operating at 600.13 MHz. Sweep widths were adjusted to 7183 Hz. The size of acquired free induction decay was 2048 complex points. The carrier frequency was set on the H 2 O. Thirty-two transients were acquired for each of 1024 t 1 values. In NOESY experiments (22), time proportional phase incrementation was used (23). Total correlation spectroscopy spectra (24) were recorded with a DIPSI-II mixing pulse sequence (50 ms) (25) using the States-TPPI method (26). Pulsed field gradient water suppression was used in all experiments (27).
DNA Binding Assay-The DNA-binding activities of the bZIP domain of C/EBP␤ and the bZIP domain of ATF4 were examined by electrophoresis mobility shift assay using oligonucleotides containing CRE or the CCAAT box. bZIP proteins in a micromolar concentration range were mixed with [ 32 P]DNA (5 nM final concentration) in 10 l of binding buffer (40 mM Tris (pH 7.5), 5 mM MgCl 2 , 100 mM NaCl, 0.01% Nonidet P-40, 1 mM dithiothreitol, and 20 mg/ml poly(dA-dT)/poly(dG-dC) (1:1, w/w)). The reaction mixtures were incubated for 30 min on ice and analyzed by native electrophoresis on 20% polyacrylamide gel (Phast system, Amersham Pharmacia Biotech). The positions of radiolabeled DNA on the gel were determined by autoradiography.
Crystallization, Data Collection, and Processing-Equimolar quantities of purified, concentrated bZIP domains of C/EBP␤ and ATF4 were mixed to form the heterodimer. Crystals were obtained by vapor diffusion at 15°C from a solution containing 0.3 mM C/EBP␤⅐ATF4 heterodimer, 1 M ammonium sulfate, 50 mM sodium cacodylate, and 5% dioxane. Hanging drops were equilibrated over well solutions containing twice the concentration of salts and dioxane. Plate-like crystals grew to final dimensions of 1.0 ϫ 0.3 ϫ 0.1 mm over the course of 3 weeks. The crystals belong to the orthorhombic P222 1 space group with unit cell dimensions of a ϭ 72.628, b ϭ 81.152, and c ϭ 35.317 Å, with one heterodimer in one asymmetric unit. Using 30% glycerol as a cryoprotectant, a data set was collected at Ϫ172°C from one crystal to a resolution of 2.6 Å on an R-AXIS II mounted on a Rigaku RU-200 x-ray generator. The data set was indexed, processed, and scaled using the programs DENZO and SCALEPACK from the HKL package (28) ( Table I).
Structure Determination, Refinement, and Model Quality-The molecular replacement solution was found using X-PLOR (29). The polyalanine model of the Fos⅐Jun complex in the crystal structure of the Fos⅐Jun⅐AP-1 site DNA complex (Protein Data Bank code 1FOS) was used as a search model. The electron density map phased by the molecular replacement model allowed us to build the ␣-helical backbone of the protein: residues 289 -339 for ATF4 and residues 239 -284 for C/EBP␤. Cycles of rebuilding using the program O (30) alternating with rigid body refinement, simulated annealing, and positional refinement against 12 to 3.0-Å data with X-PLOR allowed most of the side chains to be modeled and reduced the free R factor to 36.2% and the crystallographic R factor to 26.5%. The model that included residues 286 -341 for ATF4 and residues 239 -284 for C/EBP␤ was further refined by torsion angle molecular dynamics alternating with conjugate gradient minimization refinement and individual B factor refinement against 8 to 2.6 Å data using the program CNS (31). The final model contains residues 286 -341 for ATF4 and residues 239 -285 for C/EBP␤. Since the electron density for side chains of ATF4 Glu-286, Gln-287, and Lys-289, and C/EBP␤ Lys-242 was not apparent in a 1.5 2F o Ϫ F c map, these side chains were excluded from the model. The ␤-mercaptoethanol moiety, three Fe 3ϩ ions, and 86 water molecules were included at the final steps of the refinement. Although no iron-containing compounds were used for crystallization, they were included in refinement as possible contaminants of ammonium sulfate and dioxane. The final refinement converged to an R factor of 21.7% and an R free of 27.3% and good stereochemistry judged by the program PROCHECK (32) ( Table I).

Structure of the C/EBP␤⅐ATF4
Complex-The three-dimensional structure of the C/EBP␤⅐ATF4 bZIP heterodimer solved at 2.6 Å resolution is similar to those of other bZIP dimers determined in the complex with DNA targets. The heterodimeric coiled-coil is formed by the more curved C/EBP␤ helix wrapping around the approximately straight ATF4 helix (Fig. 1A). The most prominent structural feature of the C/EBP␤⅐ATF4 heterodimer is that most of the basic region of the ATF4 bZIP domain is a continuous ␣-helix. The ␣-helix starts at Asn-288 of ATF4 in the N terminus, which is conserved among bZIP domains. In the three bZIP domain⅐DNA complex structures determined (18,20,33) this asparagine residue has been shown to contact DNA. In contrast, the entire basic region of C/EBP␤ is disordered.
The helical conformation of the bZIP domain is maintained in part by several intrahelical interactions (Fig. 1, B and C). In particular, intrahelical interactions between ATF4 Tyr-295 and Lys-299 as well as between ATF4 Gln-297 and Arg-300 contribute to the stability of the ␣-helical conformation in the fork and the basic region of ATF4. Both pairs of residues are bridged by well ordered water molecules (Figs. 1B and 2A). Interestingly, the Gln-297 and Arg-300 pair in ATF4 have been shown to be important for the discrimination between CRE and AP-1 sites by AP-1 and CREB/ATF proteins. The GCN4 point mutant containing only these two substitutions from ATF1 has been shown to have a preference for CREB/ATF sites over the AP-1 site (35). The second pair of interaction, Tyr-295 and Lys-299, appears to be unique to ATF4 because position 295 in other bZIP domains is usually occupied by serine, cysteine, or phenylalanine. ATF4 Arg-296 is conserved in the basic region and the crystal structures of GCN4 bound to either an AP-1 site or a CRE DNA target show that ATF4 Arg-296 interacts differently with the central base pair(s) of the DNA target (18,19). In our structure of the non-complexed bZIP dimer, ATF4 Arg-296 is sandwiched between Tyr-295 and Lys-299 in the same turn of the ␣-helix and is unable to adopt many alternative conformations. These four residues are involved in important intrahelical interactions that may increase a degree of ordering of the basic region and the fork of ATF4.
Ala-291, Ala-292, Arg-294, and Tyr-295 in the basic region of ATF4 participate in crystal lattice interactions by making a backbone-backbone contact with the same residues in the symmetrically related ATF4 molecule (ATF4*) ( Fig. 2A). The successive alanine residues in these regions of the helices allow the main chain atoms to approach at the distance of 3.0 -3.5 Å to make strong van der Waals contacts, resulting in two ATF4 helices packing at an angle of ϳ80°. These lattice contacts may contribute to the stability of the ␣-helical conformation of the basic region of ATF4 in the crystal structure. The crystallization of the heterodimer in this particular conformation reflects that the ␣-helical propensity of the ATF4 basic region is higher than that of C/EBP␤. Evidence that the basic region of ATF4 may have some ␣-helical propensity also comes from circular dichroism studies performed on two synthetic 26-residue peptides comprising both the basic region and the fork of ATF4 or C/EBP␤. In aqueous solution, both peptides lack helicity. The ␣-helical content increased upon addition of TFE (Fig. 3C), which is known to stabilize the helical conformation of a peptide and is used as a quantitative probe for helical tendency in polypeptides (36). In both peptides, the mean residue ellipticity ([] 222 ) exhibited sigmoidal dependence on TFE concentration and showed a sharp transition as TFE concentration was increased above 10%. For the ATF4 peptide, [] 222 was higher than for the C/EBP␤-derived peptide at TFE concentrations above 20% and approached Ϫ31,800 degrees cm 2 dmol Ϫ1 , which corresponds to ϳ89% helicity assuming a value of Ϫ35,800 degrees cm 2 dmol Ϫ1 for 100% helicity of a 26-residue peptide at 20°C (37). For the C/EBP␤ peptide, [] 222 approached Ϫ24,700 degrees cm 2 dmol Ϫ1 , which corresponds to 69% helicity and is 20% below the value shown by the ATF4 peptide. Under these conditions, the basic region of ATF4 has a significantly greater helical propensity than the same region of C/EBP␤ and lends support to the notion that the ␣-helical basic region of ATF4 seen in our structure is not just reflection of crystal packing forces.
Dimerization of C/EBP␤ and ATF4 -As expected, the CD spectra of C/EBP␤ showed two negative peak at 208 and 222 nm, indicating the presence of ␣-helix in the leucine zipper region, and a negative peak near 200 nm, indicating a random conformation (38) in the basic region (Fig. 3A). The CD spectra of the bZIP domain of ATF4, however, revealed a negative peak near 200 nm and no strong negative peak at 208 and 222 nm. This indicates not only that the basic region is a random coil, but also that the leucine zipper region is also disordered, which suggests that the bZIP domain of ATF4 may not form a coiledcoil or a typical stable bZIP homodimer in the absence of DNA.
When the bZIP domain of ATF4 was titrated with the CREcontaining DNA, two negative peaks at 208 and 222 nm appeared in the CD spectrum, which are characteristic of an ␣-helical structure. The ␣-helical content is estimated by [] 222 . The end point of the titration does not exceed Ϫ18,200 degrees cm 2 dmol Ϫ1 , which corresponds to 48% helicity and suggests that the leucine zipper region of ATF4 may not necessarily where F c and F o are the calculated and observed structure factor amplitudes for reflection hkl, respectively. d R free is the same as R, but calculated over a 10% fraction of the reflection data that was randomly selected and not included in the refinement.
form a stable coiled-coil even in the presence of the DNA target.
In contrast, the CD spectra for the bZIP domain of C/EBP␤ showed the helical content of ϳ50%, which increased to 81% upon binding to CCAAT box DNA (Fig. 3A). When equal amounts of the bZIP domains of C/EBP␤ and ATF4 were mixed together, the CD spectrum showed two strong negative bands at 206 and 222 nm (Fig. 3B), leading to an estimation of the helical content of ϳ85%. This exceeds not only that of the bZIP domain of ATF4 in the presence of specific DNA (CRE), but also that of the bZIP domain of C/EBP␤ without DNA. In fact, it is comparable to that of the bZIP domain of C/EBP␤ in the pres-ence of the DNA target (CCAAT box). This indicates that the formation of the heterodimer between the bZIP domains of ATF and C/EBP␤ induces an ␣-helix in the bZIP domain of ATF4 even in the absence of specific DNA. Proteolytic digestion of the bZIP domains confirmed our CD observation (Fig. 4). Proteolysis by endoproteinases, V8 (Glu-C) and trypsin showed that the basic region of the C/EBP␤ bZIP domain is digested in a short period time (Ͻ10 min under the condition we used), whereas the leucine zipper region remained uncut for a longer period of time. However, the basic region of the ATF4 bZIP domain is more stable than that of C/EBP␤, with only about one-third of the basic region from the N terminus digested in the same period time. Surprisingly, there was an additional cleavage at the middle of the leucine zipper region of the ATF4 bZIP domain, suggesting that the leucine zipper of ATF4 may not be able to maintain an ␣-helical conformation in the absence of specific DNA.
NMR experiments performed for ATF4 and C/EBP␤ at protein concentration of 1.5 mM at different temperatures are in good agreement with the results of the CD experiments. Fragments of NOESY spectra showing H N -H N cross-peaks indicative of the potential ␣-helical conformation for both proteins are shown in Fig. 5. C/EBP␤ shows a significantly larger variation of chemical shifts. The dispersion observed for ATF4 is rather typical for a marginally structured polypeptide. At 35°C, very few non-intra-residue cross-peaks were observed for ATF4. In contrast, the C/EBP␤ spectra showed a number of broad crosspeaks in the H N -H N area at 15°C that narrowed significantly on increasing the temperature to 35°C. A number of other non-intra-residue nuclear Overhauser effect cross-peaks were found for C/EBP␤, although many of them were broad. This behavior is consistent with a 15.2-kDa dimer with an aspect ratio of ϳ4:1. As the temperature was further increased, the intensity of these cross-peaks diminished, albeit at a much slower rate than in the case of ATF4, most likely due to the increased rate of exchange with solvent. The abundance of non-intra-residue H N -H ␣ cross-peaks is characteristic of an ␣-helical conformation (data not shown) (39). The comparison of total correlation spectroscopy and NOESY spectra at 35°C (data not shown) revealed no non-intra-residue H N -H ␣ crosspeaks for ATF4. The same comparison for C/EBP␤ showed ϳ30 cross-peaks.
The bZIP domains of ATF4 and C/EBP␤ are compared with an ideal ␣-helix in Fig. 2B. The bZIP domain of ATF4 is nearly a perfect (root mean square for C-␣ of 0.8 Å) ␣-helix, whereas the bZIP domains of C/EBP␤ in the C/EBP␤⅐ATF4 heterodimer and in the C/EBP␤ complexed with the CCAAT box are significantly curved. Interestingly, the C/EBP␤ in the heterodimer is more curved than that in the C/EBP␤⅐CCAAT box complex. This suggests that the bZIP domain of C/EBP␤ is flexible and that, depending on the partner molecule, it may adapt readily to form a stable coiled-coil. The flexibility of the bZIP domain of ATF4 awaits the structural analysis of the ATF4 in the homodimer complexed with the DNA target (CRE). The fact that ATF4 does not form a stable homodimer (it is monomeric in FIG. 2. bZIP ␣-helices. A, electron density of the ATF4 basic region from the 2F o Ϫ F c map contoured at 1.0. Intrahelical hydrogen bonds in the ATF4 forkbasic region junction formed between ATF4 Tyr 295 and Lys 299 as well as between ATF4 Gln 297 and Arg 300 contribute to the stability of the ␣-helical conformation in the fork and the basic region. Dashed lines indicate the distances between water molecules and respective atoms of the side chains. The symmetryrelated ATF4* molecule is shown in red. B, comparison of ␣-helices. The ideal (gray) ␣-helix was generated using in-sightII. Each bZIP domain was superimposed on the ideal ␣-helix. The ATF4 bZIP domain is indicated in red, the C/EBP␤ bZIP domain in the heterodimer in green, and the C/EBP␤ bZIP domain in the C/EBP␤⅐DNA complex in yellowgreen. C, the potential disulfide bond in the ATF4 bZIP homodimer. The second subunit of the ATF4 bZIP dimer was generated by superimposing the bZIP domain of ATF4 in the heterodimer on the bZIP domain of C/EBP␤ in the same heterodimer structure. solution) and that, even in the presence of the DNA target, it forms a relatively weak homodimer suggests that ATF4 may maintain its straight ␣-helical conformation, i.e. it fails to adapt the intertwining conformation necessary for the establishment of a stable coiled-coil. Our recent CD data (data not shown) indicate that when ATF4 (bZIP domain) is oxidized, it has a higher ␣-helical content in the presence of CRE than the unoxidized form or the Cys-310 3 Ser mutant. The covalent disulfide bond may hold two bZIP domains of ATF4 together, maintaining the more stable bZIP dimer, which subsequently binds to the DNA target with higher affinity.
DNA-binding Activity-The functional activity of the recombinant bZIP domains of ATF4 and C/EBP␤ was also examined by an electrophoretic gel shift assay using CRE-or CCAAT box-containing DNA probes (Fig. 6). Both proteins bound their specific DNA elements (ATF4 to CRE and C/EBP␤ to the CCAAT box) with high affinity, resulting in formation of protein⅐DNA complexes with distinct mobility on native polyacrylamide gel. As expected, the ATF4 bZIP domain did not bind to CCAAT box DNA (compare lanes 3 and 9), whereas the C/EBP␤ bZIP domain bound to both CRE and CCAAT box probes with comparable affinity (compare lanes 2 and 8), indicating a broad range of sequence specificity. The addition of higher concentrations of the ATF4 bZIP domain did not affect the binding of C/EBP␤ to the CCAAT box (lanes 4 -6). Only a small amount of the CCAAT box-bound C/EBP␤⅐ATF4 heterodimer appeared in the gel as a minor band with higher mobility. In contrast, the addition of even a small amount of the ATF4 bZIP domain to the C/EBP␤ bZIP domain resulted in the preferential formation of the heterodimeric C/EBP␤⅐ATF4⅐CRE complex (lanes 10 -12). These results clearly indicate that the ATF4 bZIP domain has a preference for binding to CRE, whereas the C/EBP␤ bZIP domain accommodates both binding sites with comparable affinity. While the CCAAT box DNA failed to attract the ATF4 bZIP domain even in the presence of the C/EBP␤ bZIP domain, CRE was able to accommodate both bZIP domains (lanes 4 -6). Thus, the heterodimer formed between C/EBP␤ and ATF4 displays the same DNA-binding preferences as ATF4 does. Apparently, the heterodimer forms a more stable complex on CRE than the ATF4 homodimer does.
Redox and C/EBP␤⅐ATF4 -It has been shown that some bZIP transcription factors are regulated by redox tension (40,41). Most of the bZIP domains contain a cysteine amino acid in the basic region, which functionally interferes with DNA-binding activity when it is oxidized. The bZIP domain of ATF4 does not contain the cysteine residue in the basic region; instead, it contains one cysteine amino acid in the leucine zipper region, which is not conserved among other bZIP domains. The cysteine residue in the ATF4 bZIP domain is in the a position (in the helical wheel), which is in position to be a part of the hydrophobic core of the bZIP dimer. The crystal structure of the C/EBP␤⅐ATF4 bZIP heterodimer indicates that this cysteine residue in the ATF4 bZIP domain is positioned to make a disulfide bond with the same cysteine residue from the other subunit of the ATF4 bZIP domain when ATF4 forms a homodimer, presumably in the presence of the DNA target (see Fig. 2C). This cysteine residue, when oxidized, may stabilize the ATF4 homodimer by forming the covalent disulfide bond. Our study indicated that ATF4 is monomeric in the absence of the DNA target, but in the presence of the DNA target, it forms a relatively less stable homodimer on DNA. On the other hand, ATF4 forms a more stable heterodimer with C/EBP␤ even in the absence of the DNA target. The heterodimer binds to CRE with high affinity, but not to the C/EBP site. Depending on the oxidation state (due to the cellular oxygen level), ATF4 forms either a stable homodimer via the intersubunit disulfide bond or a more stable heterodimer, such as C/EBP␤⅐ATF4. As a heterodimer, it binds CRE or a CRE-related DNA element and regulates transcription of distinct genes. It has been reported (42) that both ATF4 and C/EBP␤ are induced to a higher level in anoxic tension. Although there is no direct evidence, perhaps in a low oxygen state, ATF4 may form a stable heterodimer with C/EBP␤, rather than a less stable ATF4 homodimer without the disulfide bond, and may activate the transcription of genes involved in the recovery process from anoxic stress.

Formation of a Stable bZIP
Dimer-This study on the bZIP domains of ATF4 and C/EBP␤ and the heterodimer formed by these proteins addresses the formation a stable cross-family bZIP heterodimer and the interaction with the DNA target. It has been generally understood that the hydrophobic core of the bZIP dimer is formed by hydrophobic amino acids occurring at every 4-3 position of the helix (or the a and d positions in the helical wheel) with a leucine in every seventh position. In fact, these residues are conserved throughout bZIP proteins. Nonetheless, some bZIP dimers are more stable than others, and this difference in stability depends on the distribution of polar amino acid side chains that are partially exposed (particularly in the g and e positions). These amino acids potentially present attraction or repulsion interactions within the same ␣-helix and between two parallel ␣-helices. A number of structural studies including ours have confirmed this notion. As indicated above, ATF4 does not form a stable homodimer, but C/EBP␤ does. From the potential attraction-repulsion interaction scheme based on the distribution of polar side chains on the surface of the helices of the C/EBP␤ homodimer and the C/EBP␤⅐ATF4 heterodimer, a slight preference for heterodimer may be inferred. The exclusive formation of the heterodimer from the 1:1 mixture of the two proteins was not expected. In the crystal structure of the heterodimer, there is a clear asymmetry of the ␣-helices, the nearly straight ATF4 ␣-helix and the more curved C/EBP␤ ␣-helix, which wraps around the former. To form a stable coiled-coil, two parallel or antiparallel helices need to be curved enough to intertwine in order to maintain the optimal orientation of main chain atoms necessary for side chains to participate in the maximum number of attractive interhelical interactions. Although it remains to be seen, ATF4 may maintain its straight ␣-helical conformation, preventing it from intertwining to form the stable homodimer. However, C/EBP␤ appears to be more flexible and allows necessary intertwining for the formation of a stable homodimer. In the heterodimer, the more flexible C/EBP␤ compensates for the lack of flexibility in ATF4 by wrapping around it to maintain the maximum attractive interaction between the two helices.
␣-Helical Basic Region of ATF4 -The most prominent feature of the crystal structure of the C/EBP␤⅐ATF4 bZIP heterodimer is the ␣-helix in the basic region of the ATF4 bZIP domain, which orients most of the conserved residues in this region in position for specific DNA (CRE) binding; in contrast, the basic region of the C/EBP␤ bZIP domain is disordered. It appears that the ␣-helix in the basic region of the ATF4 bZIP domain is stabilized in part by the crystal lattice interaction between the two symmetrically related basic regions of the ATF4 bZIP domains. Although, other than the crystal structure analyzed in this study, there is no direct evidence indicating the ␣-helix in the basic region of the ATF4 bZIP domain, there are critical data complementing the crystal structure result. The proteolysis by trypsin and Glu-C suggested that the basic region of the ATF4 bZIP domain is more stable than that of the C/EBP␤ bZIP domain (Fig. 4). The CD spectra of the mixture of the ATF4 bZIP domain and C/EBP␤ also support that the ␣-helix in the basic region of the ATF4 bZIP domain shown in the crystal structure of the C/EBP␤⅐ATF4 heterodimer may not be a mere artifact resulting from crystallization. The CD spectra indicate that, in solution, the ␣-helix content of the bZIP heterodimer consisting of the ATF4 and C/EBP␤ bZIP domains in the absence of specific DNA (CRE) (Fig. 3B) is comparable to, if not higher than, that of the C/EBP␤ bZIP domain in the presence of specific DNA (CCAAT box) (Fig. 3A). In addition, the CD spectra (Fig. 3C) obtained from the 26-amino acid peptides comprising both the basic region and the fork of ATF4 (ATF4-bf) or C/EBP␤ (C/EBP␤-bf) in the presence of the helixstabilizing agent TFE suggest a higher ␣-helix propensity for ATF4-bf than for C/EBP␤-bf. Although it contains a leucine zipper region, our NMR and CD studies indicated that the entire bZIP domain of ATF4 is in equilibrium between an ␣-helix and a random coil, existing mostly as a random coil in the absence of DNA. The addition of specific DNA (CRE) shifts the equilibrium toward an ␣-helix by inducing an ␣-helix in the bZIP domain including the basic region. The heterodimerization of the ATF4 bZIP domain with the C/EBP␤ bZIP domain may also induce an ␣-helix in the ATF4 bZIP domain including the basic region, but not in the basic region of the C/EBP␤ bZIP domain probably because of its lower ␣-helix propensity. Once the ␣-helix is induced in the leucine zipper region of the ATF4 bZIP domain by the formation of the heterodimer with the C/EBP␤ bZIP domain, the intrahelical water-mediated hydrogen bonds found in the basic and fork regions of the ATF4 bZIP domain (Figs. 1B and 2A) may help extend stabilizing ␣-helical conformation from the leucine zipper region to the basic region.
With the ordered ␣-helical conformation in the basic DNAbinding region, ATF4 may be more selective in binding to the DNA target via an entropy-controlled manner. With its disordered basic region, C/EBP␤ may be able to adapt to have a broad range of DNA-binding specificities and to accommodate CRE with comparable affinity to its canonical DNA site (CCAAT box). As a result, the C/EBP␤⅐ATF4 heterodimer has a DNA-binding preference similar to that of the ATF4 homodimer, but forms a more stable complex on CRE than the ATF4 homodimer does. Whether the preformed C/EBP␤⅐ATF4 heterodimer interacts with CRE or the binding of each bZIP monomer to DNA precedes the dimerization cannot be concluded from available data. In both possible pathways, the interaction of the ATF4 basic region with the CRE half-site apparently plays a decisive role in selection of target DNA. Many of the CREB/ATF proteins form heterodimers with AP-1 (9 -13) or C/EBP (14 -16) family members that bind CRE or composite sites containing at least one CRE half-site. The data suggest that other CREB/ATF proteins may interact with DNA targets and dimerization partners using a mechanism similar to that discussed here for ATF4.
Cross-family bZIP Heterodimer-Most transcription factors are dimeric. A dimeric DNA-binding domain provides significant advantages over its monomeric counterparts for accurate regulation of transcription. A heterodimer provides a unique advantage: a small number of components can be used to generate new transcription factors that bind to distinct DNA sites with different specificity and that interact with different proteins. They can regulate transcription of different sets of genes by binding to different sites in different promoters. Estes et al. (42) showed that ATF4 and C/EBP␤ levels increase upon exposure of cells to anoxic conditions and that anoxic tension leads to enhanced binding of ATF4 to a CCAAT box/CRE (or AP-1 site) composite site. This suggests that ATF4 may form a heterodimer with C/EBP␤ and participate in a unique role in this condition of stress. More recently, Fawcett et al. (34) demonstrated that CREB/ATF family members form heterodimeric complexes with C/EBP␤ on a CCAAT box/CRE composite site that resides in the Gadd153 promoter. Gadd153, also known as CHOP and a member of the C/EBP transcription factor family, is transcriptionally activated by cellular stress signals, resulting in growth arrest or DNA damage (GADD). In response to arsenic stress, ATF4 activates Gadd153 transcription by forming the complex with C/EBP␤, whereas ATF3 represses Gadd153 transcription from the CCAAT box/CRE composite site.