Solution Structure of Human Growth Arrest and DNA Damage 45α (Gadd45α) and Its Interactions with Proliferating Cell Nuclear Antigen (PCNA) and Aurora A Kinase*

Gadd45α is a nuclear protein encoded by a DNA damage-inducible gene. Through its interactions with other proteins, Gadd45α participates in the regulation of DNA repair, cell cycle, cell proliferation, and apoptosis. The NMR structure of human Gadd45α has been determined and shows an α/β fold with two long disordered and flexible regions at the N terminus and one of the loops. Human Gadd45α is predominantly monomeric in solution but exists in equilibrium with dimers and other oligomers whose population increases with protein concentration. NMR analysis shows that Aurora A interacts through its N-terminal domain with a region of human Gadd45α encompassing the site of dimerization, suggesting that the oligomerization of Gadd45α could be a regulatory mechanism to modulate its interactions with Aurora A, and possibly with other proteins too. However, Gadd45α appears to interact only weakly with PCNA through its flexible loop, in contrast with previous and contradictory reports.

Gadd45␣ is a nuclear protein encoded by a DNA damageinducible gene. Through its interactions with other proteins, Gadd45␣ participates in the regulation of DNA repair, cell cycle, cell proliferation, and apoptosis. The NMR structure of human Gadd45␣ has been determined and shows an ␣/␤ fold with two long disordered and flexible regions at the N terminus and one of the loops. Human Gadd45␣ is predominantly monomeric in solution but exists in equilibrium with dimers and other oligomers whose population increases with protein concentration. NMR analysis shows that Aurora A interacts through its N-terminal domain with a region of human Gadd45␣ encompassing the site of dimerization, suggesting that the oligomerization of Gadd45␣ could be a regulatory mechanism to modulate its interactions with Aurora A, and possibly with other proteins too. However, Gadd45␣ appears to interact only weakly with PCNA through its flexible loop, in contrast with previous and contradictory reports.
The Gadd45 4 family of proteins consists of isoforms ␣, ␤, and ␥, with sequence identities of ϳ56% (1). Transcription of the Gadd45 genes is induced by DNA-damaging agents and other cellular stresses and is associated with growth arrest.
Human Gadd45␣ (hGadd45␣) is an acidic protein of low abundance in the cell, localized mainly in the nucleus (2). It participates in cell growth and cell cycle control, DNA repair, apoptosis, maintenance of genomic stability, and the regulation of signaling pathways (3,4). Gadd45 proteins exert these functions by their interactions with other proteins. hGadd45␣ interacts with PCNA (5), an essential factor for DNA replication and repair, the cyclin-dependent kinase inhibitor p21 WAF1 (6), and with the protein kinases cyclin-dependent kinase 1, mitogenactivated protein kinase kinase kinase 4 (MAPKKK4), and Aurora A (7,8). hGadd45␣ maintains centrosome stability by down-regulation of Aurora A kinase activity (7). The pleiotropic effects of hGadd45␣ have partly been explained by its interaction with the xeroderma pigmentosum complementing group G endonuclease and the promotion of DNA repair, which erases methylation marks relieving epigenetic gene silencing (9). However, the role of Gadd45␣ in active DNA demethylation is still controversial (10,11). Gadd45␣ interacts with the four core histones of the nucleosome and modifies the access to DNA on damaged chromatin (12). Gadd45 proteins can form oligomers, which could have a role in regulating their function (13).
We describe the NMR structure of the full-length monomeric hGadd45␣ protein, which adopts a globular ␣/␤ fold with two long disordered and flexible regions at the N terminus and one of the loops. The folded region is similar to the recent crystal structure of a construct of the murine Gadd45␥ isoform truncated at the N terminus, which forms dimers (14). We show that hGadd45␣ interacts with Aurora A through a region opposite to the location of the flexible protein chain termini and encompassing a putative dimerization site, whereas it interacts with PCNA through the flexible loop albeit very weakly.

EXPERIMENTAL PROCEDURES
Protein Production and Purification-All the human proteins were produced in Escherichia coli and purified from the soluble fractions of the lysed cells by liquid chromatography. Two different wild type hGadd45␣ protein samples were prepared, one of them without any tag, the other fused to an N-terminal His-tag, which was removed by proteolysis (15). PCNA was also purified with an N-terminal His-tag cleaved in the same way (16). Aurora A full-length and the kinase domain (residues 122-403) were produced fused to an N-terminal His-tag.
Analytical Ultracentrifugation and Gel Filtration-Sedimentation velocities were measured at 25°C at a speed of 42,000 rpm in buffer: 20 mM sodium phosphate, pH 7.4, 100 mM KCl, 2 mM EDTA, 2 mM dithiothreitol. Analytical size-exclusion chromatography of hGadd45␣ proteins was done at room temperature on a calibrated Superdex-75 10/300 column using the same buffer.
NMR Spectroscopy, Structure Calculations, and Measurement of Interactions-NMR measurements were performed at 25°C on Gadd45␣ samples with protein concentrations in the range 160 -450 M. Distance, dihedral angle, and residual dipolar coupling restraints were measured and used to calculate an ensemble of 20 models that were refined by restrained energy minimization in explicit water. 1 H-15 N HSQC spectra of U-15 N Gadd45␣ in the absence and presence of Aurora A or PCNA were recorded sequentially and under identical conditions on samples dialyzed simultaneously against the same buffer. The same procedure was used to prepare the samples and record the spectra of U-2 H-15 N PCNA (with protonated amides) in the absence and in the presence of Gadd45␣, except that the measurement temperature was 35°C. Detailed experimental methods are provided in the supplemental information methods section.

Human Gadd45␣ Is Predominantly Monomeric in Solution-
Sedimentation velocity measurements show that hGadd45␣ is predominantly monomeric up to concentrations of a few hundred M, in equilibrium with dimers and other oligomeric species ( Fig. 1). Although the theoretical molecular mass of the monomer is 18.3 kDa, analytical gel filtration yields an apparent value of 29 kDa (supplemental Fig. S1). This is probably due to the long flexible segments of the molecule (see below), which accelerate the advance of the molecule through the column with respect to a compact and globular one. The line widths of most of the NMR signals of hGadd45␣ are also consistent with a protein of the size of the monomer (15). The average R 2 relaxation rates measured for the backbone 15 N nuclei in the ordered regions of the protein is 20 Ϯ 0.1 s Ϫ1 , which is consistent with monomeric hGadd45␣. Two-dimensional diffusion ordered spectroscopy yielded a diffusion coefficient slightly larger than the value measured for lysozyme (14.6 kDa) and smaller than the value measured for maltose binding protein (42 kDa; supplemental Fig. S2). These results are in agreement with published gel filtration data (providing that a smaller elution volume is not interpreted as a size larger than the monomer) and with chemical cross-linking data (13), but not with the measured electrophoretic mobility which, in addition to monomers, trimers, and tetramers, suggest a predominantly dimeric species (13). At concentrations in the low mM range, the NMR signals of hGadd45␣ broaden, and their intensity decreases with time, eventually becoming invisible except for a set of nondispersed signals from the flexible regions of the molecule. These observations can be explained by the irreversible selfassociation of hGadd45␣ into large soluble aggregates, which sequester the monomeric species and render them invisible by NMR. The tendency to aggregate could be related to the low global and local structural stability of the protein. Thus, the thermal denaturation monitored by circular dichroism is less cooperative than expected for a protein of its size (data not shown), and the solvent exchange of the amide protons is very fast (supplemental Fig. S3). These observations indicate a complex dynamic behavior of hGadd45␣, which could be of functional relevance.
Solution Structure of Human Gadd45␣-The solution structure of hGadd45␣ is represented by an ensemble of 20 refined NMR models whose superposition ( Fig. 2A) shows three disordered regions: residues 1-16, 105-118, and 164 -165. The high root mean square deviations in these regions are a consequence of their high flexibility, as confirmed by small local heteronuclear { 1 H}-15 N Overhauser effects (supplemental Fig. S4). The R 2 relaxation rates of the backbone 15 N nuclei in these mobile regions also are smaller than those of the ordered regions; however, the low signal to noise ratio in these experiments resulted in large errors in the estimation of the values of R 2 preventing a reliable full quantitative relaxation study of the protein.  The ordered regions of the chain are defined with an root mean square deviation of 0.95 and 1.50 Å for the backbone and all the heavy atoms, respectively (supplemental Table S1 and supplemental Fig. S5), and the root mean square deviation is even smaller for the secondary structure elements (0.75 Å for the backbone heavy atoms). The structure shows a segregated ␣/␤ fold with a five-stranded mixed ␤-sheet at the core and five surrounding helices (Fig. 2B). Helices ␣2 and ␣3 pack on one side of the ␤-sheet, whereas helices ␣1, ␣4, and ␣5 pack on the opposite side. The ␤2-␣3 connecting loop contains one turn of 3 10 helix. The two long flexible regions at (the N terminus and the loop connecting ␣4 with ␤4), are not completely random, but show a propensity to form local helical structures in fast exchange with other conformations. These nonrandom structures can be detected by the deviations of the corresponding backbone chemical shifts from tabulated random coil values (Supplemental Fig. S6), which are small but larger than the range observed in denatured polypeptide chains. These conformational propensities could be relevant for the interactions of hGadd45␣ with other proteins (see below) as well as for its monomeric state. In this regard we note that the published (17) analytical gel filtration chromatography on the human Gadd45␤ isoform (performed on the full length construct) is consistent with a monomeric molecule providing that the elution volume, larger than expected for a globular protein of its size, is due to the N-terminal and ␣4-␤4 loops being flexible as are the homologous regions of mGadd45␣. Therefore, it may be that the dimerization equilibrium of hGadd45␥ is shifted toward the dimer when the N-terminal segment is eliminated. Indeed, it is tempting to speculate that, in solution, Gadd45 proteins in general exist in dynamic equilibria with oligomeric populations, which can be modulated by the proteins concentration, mutations, deletions, and ligand binding.
The structure of the ordered regions of hGadd45␣ is similar to the reported crystal structure of a deletion mutant of mGadd45␥, superimposing with a root mean square deviation of 1.7 Å for the C ␣ atoms ( Fig. 3A and supplemental Fig. S7). This construct of the murine ␥-isoform was found to form dimers in solution, with the dimerization interface formed by helices ␣2 and ␣3. The possibility that other Gadd45 proteins dimerize in the same way is suggested by the high conservation of residues located in helices ␣2 and ␣3 (supplemental Fig. S8). However, earlier studies with deletion mutants and peptide binding suggested a different oligomerization site for hGadd45␣ (13).
Our hydrodynamic results on the monomeric nature of hGadd45␣ are confirmed by the amide residual dipolar couplings, an independent and purely orientational measurement that constrains the arrangement of the possible oligomers (18,19). For several residues of the well defined regions of the protein, the residual dipolar couplings are incompatible with a dimer analogous to the reported mGadd45␥ (r ϭ 0.88 and 0.82, each for one of the two chains in the dimer), but compatible with the set of 20 solution structures of monomeric hGadd45␣ (r ϭ 0.94 Ϯ 0.01; see also supplemental Fig. S9). Still, a small population of dimeric hGadd45␣ exists in equilibrium with the monomers (Fig. 1 and supplemental Fig. S1). The dimer population is much smaller or even undetectable (depending on the technique used) when the mutation L77E is introduced in helix ␣3. This mutation is analogous to the mutation L80E in mGadd45␥, which strongly destabilizes the dimer and confirms that the dimerization interface is formed by helices ␣2 and ␣3 rather than the other three possible interfaces observed in the mGadd45␥ crystal structure. These results suggest that a similar dimerization surface may exist for both proteins, although the oligomerization equilibrium is shifted toward monomeric hGadd45␣ and to dimeric mGadd45␥. The different populations of dimeric and monomeric species may be explained by two facts. First, residues differ at three positions within the ␣2 and ␣3 helices in the two isoforms ( Fig. 3B and supplemental Fig. S8). Second, interhelical packing differs, with an angle (between the longitudinal axes of both helices) of 12 Ϯ 4°a nd 29 Ϯ 2°for the hGadd45␣ NMR structures and for the mGadd45␥ crystal monomers, respectively (Fig. 3B). This different orientation results from the reorganization of the packing at the hydrophobic core, due to the different size of the Ala 43 and Leu 46 side chains in hGadd45␣ compared with the corresponding residues in mGadd45␥ (Ser 46 and Val 49 , respectively), and to the burial of the Ser 46 hydroxyl group of mGadd45␥. This polar group would destabilize the structure were it not hydrogen bonded with the carbonyl of G 42 . A third factor that may affect the association properties of the Gadd45 proteins is the flexibility of the N-terminal region, which is absent in the reported mGadd45␥ construct. This segment of the chain would lie at the opposite side of the proposed dimerization interface, but considering its length and its high mobility, it cannot be excluded that it influences the tendency of two molecules to associate.
Interaction of hGadd45␣ with PCNA and Aurora A Kinase-Among the few proteins reported to interact directly with hGadd45␣, we have tried to structurally characterize the binding of human PCNA and Aurora A kinase because both of them can be expressed in E. coli and can be purified in the amount and with the purity necessary for NMR studies (20,21). In the case of PCNA, its backbone NMR spectrum has been assigned previously (16), making it possible to map the interaction site on both hGadd45␣ and PCNA.
The interaction with PCNA is the earliest and most documented interaction of Gadd45 proteins. It has been reported that hGadd45␣ binds to human PCNA (6), as do mGadd45␤ and mGadd45␥ (22), though none of them contain the typical PCNA interacting protein (PIP) box motif (23). The PIP box is a conserved sequence that binds to three equivalent sites on the PCNA homotrimeric ring, with one PIP sequence for each PCNA monomer (24,25). Thus, binding of Gadd45 proteins to a site on PCNA different from the PIP binding site might be compatible with the binding of PIP-containing proteins to PCNA, whereas these other proteins compete among them (26). However, the experimental evidence for the identification of the hGadd45␣/PCNA interaction sites is contradictory (27).
Yeast two-hybrid data reported by two independent groups map the interaction site to different nonoverlapping Gadd45 regions, and inconsistent results also were obtained for the regions of PCNA binding to hGadd45␣ (6,22). We have analyzed in detail this interaction by NMR observing the changes in the signals of hGadd45␣ in the presence of the homotrimeric 87-kDa PCNA ring. A reduction in the signal intensities would be expected as a result of the large size of the complex, with a larger decrease for those residues at the interface. We could observe significant changes only for a few residues in the flexible ␣4 -␤4 loop and at the C terminus (Fig. 4). Binding through these regions are consistent with one set of reported yeast two-hybrid data (27), but not with the other set (6) because opposite results were reported for hGadd45␣-(1-136) binding to PCNA. The available assignment of the NMR spectrum of PCNA (16) allowed us to examine the interaction from the other side, observing the changes in the spectrum of PCNA in the presence of hGadd45␣. Because of the large size of PCNA, it must be perdeuterated to observe the NMR signals of its protonated amide groups. The size increase on binding to Gadd45␣ (18 kDa) should not reduce much their intensities, but should cause chemical shift perturbations on those PCNA residues at the binding site. However, the values that could be measured were all small and below the experimental error (supplemental Fig. S10). Based on the absence of significant chemical shift perturbations, we estimate that the dissociation constant of the possible complex would be larger than 2.2 mM (assuming a 1:1 PCNA:hGadd45␣ monomer ratio) or larger than 0.64 mM (if we assume a 3:1 ratio, one molecule of hGadd54␣ per PCNA ring). It has been proposed that the hGadd45␣/PCNA interaction coordinate the cell cycle and the DNA damage response (6), reducing the negative control of cell growth and reducing apoptosis (27). We have measured here that a direct physical interaction between the two pure proteins in solution is, at best, transitory and very weak, although it could be enhanced by other factors present in the cell nucleus.
The interaction between hGadd45␣ and the mitotic kinase Aurora A has been observed by yeast two-hybrid, immunoprecipitation, and pulldown assays (7,11). Mouse embryonic fibroblast cells deficient in mGadd45␣ exhibit genome instability, centrosome amplification, and aberrant mitosis, with multiple spindles and failed cytokinesis (3). Since ectopic expression of hGadd45␣ in Chinese hamster ovary cells antagonizes the centrosome amplification induced by Aurora A (7), The spectra were acquired in 20 mM sodium phosphate, pH 7.4, 100 mM KCl, 2 mM dithiothreitol, 2 mM EDTA at 25°C. Inset, relative hGadd45␣ peak intensities with versus without PCNA plotted against the residue number; the experimental error is indicated by the error bars. An overall reduction of ϳ25% in the cross-peak intensities in the sample with PCNA is caused by the dilution of the mixed sample during dialysis with respect to the sample with only hGadd45␣ (because of the higher osmotic pressure). For some residues, a quantitative measurement was not possible, and the corresponding position is left empty in the plot. and hGadd45␣ inhibits Aurora A in HeLa cells, it can be hypothesized that hGadd45␣ maintains centrosome stability by down-regulation of Aurora A kinase activity. Direct interaction with Aurora A in solution is confirmed by the analysis of the hGadd45␣ 1 H, 15 N HSQC spectra (Fig. 5A), with a large reduction in its signal intensities as a result of the large molecular mass of the complex formed (67 kDa or larger). The intensity reduction is not uniform along the sequence, with the residues changing the most located in a region of the molecule opposite to the location of the flexible chain termini and encompassing helices ␣2 and ␣3 (Fig. 5B). The location of the binding site is confirmed by the same measurements on hGadd45␣ mutant L77E (in helix ␣3), whose NMR signals show smaller changes (and mostly uniform along the sequence) in the presence of Aurora A (supplemental Fig. S11). A truncated mutant of Aurora A containing only the kinase domain (residues 122-403) causes also a smaller and uniform reduction in the hGadd45␣ signal intensities, indicating that hGadd45␣ interacts principally with the N-terminal domain of Aurora A (supplemental Fig. S11). This domain is a nonconserved region that could be the binding site to regulatory proteins. Because the Aurora A binding site of hGadd45␣ overlaps with the Gadd45␣ dimerization site, it is likely that the two interactions compete with each other. Indeed, the dimerization interface, mapped by an analogous procedure (differential decrease in the intensity of hGadd45␣ signals in spectra at 200 and 1,300 M protein concentration), overlaps with the Aurora A binding site (data not shown). The NMR data do not rule out a possible interaction of Aurora A with hGadd45␣ dimers, binding on an accessible surface in the proximity of the dimer interface. However, this alternative mode of binding is less probable because a larger number of perturbed residues on the surface of hGadd45␣ than those shown in Fig. 5 would be expected if both dimerization and binding to the much bigger Aurora A molecule were happening. Dynamic light scattering measurements on a sample identical to that used for the NMR measurements did not clarify this issue as the estimated diameters of the species present in the sample were quite variable over consecutive measurements (data not shown). In any case, the hGadd45␣/Aurora A interaction that can be measured by NMR should not be interpreted in terms of a single species. In solution, the complex coexists with different populations of free hGadd45␣ and free Aurora A protein, in addition to a certain amount of dimeric hGadd45, all of them in dynamic equilibrium.

CONCLUSIONS
Self-association of hGadd45␣ has been detected in mammalian cells by co-immunoprecipitation of the protein expressed with different tags, as has formation of hetero-oligomers of Gadd45 isoforms (13). However, it is unclear whether this association is direct or mediated by other cellular components. The relevance of hetero-oligomerization is also unclear, as the FIGURE 5. Interaction of Gadd45␣ with Aurora A. A, overlay of the 1 H-15 N HSQC spectra of ϳ65 M Gadd45␣ in the absence (black) and in the presence (red) of Aurora A (1:1 molar ratio). The spectra were acquired in 20 mM Tris, pH 8.0, 200 mM NaCl, 2 mM dithiothreitol, 0.5 mM EDTA at 25°C. Inset, relative hGadd45␣ peak intensities with versus without Aurora A, plotted against the residue number. Error bars indicate the estimated experimental error. B, surface representation of hGadd45␣ with residues colored according to relative intensity (bound versus free hGadd45␣): larger than 0.3 (or not possible to analyze) in yellow, between 0.3 and 0.2 in pink, between 0.2 and 0.1 in magenta, and smaller than 0.1 in red. The left-hand view has the same orientation as Fig.  3B, and the right-hand view resulted from a 180°rotation along the indicated axis. Helices ␣2 and ␣3 are indicated on the left-hand view. expression of each gene is induced by different signals, human Gadd45␣ being the only one up-regulated by p53 (28). Oligomeric hGadd45␣ does not protect UV-irradiated nucleosomes from nuclease digestion, indicating that it loses the ability to modify DNA accessibility on damaged chromatin (13); thus, the oligomeric state of hGadd45␣ may affect its function. Our results show that hGadd45␣ is primarily monomeric, in dynamic equilibrium with minor populations of oligomeric forms. Competition and shifting of this equilibrium could be the basis for a temporal/spatial regulatory mechanism of the interactions with different effectors in the nucleus. The direct interaction with Aurora A through an overlapping site is consistent with a regulatory role of hGadd45␣ self-association, which could release active Aurora A.