BAG4/SODD protein contains a short BAG domain.

BAG (Bcl-2-associated athanogene) proteins are molecular chaperone regulators that affect diverse cellular pathways. All members share a conserved motif, called the BAG domain (BD), which binds to Hsp70/Hsc70 family proteins and modulates their activity. We have determined the solution structure of BD from BAG4/SODD (silencer of death domains) by multidimensional nuclear magnetic resonance methods and compared it to the corresponding domain in BAG1 (Briknarová, K., Takayama, S., Brive, L., Havert, M. L., Knee, D. A., Velasco, J., Homma, S., Cabezas, E., Stuart, J., Hoyt, D. W., Satterthwait, A. C., Llinás, M., Reed, J. C., and Ely, K. R. (2001) Nat. Struct. Biol. 8, 349-352). The difference between BDs from these two BAG proteins is striking, and the structural comparison defines two subfamilies of mammalian BD-containing proteins. One subfamily includes the closely related BAG3, BAG4, and BAG5 proteins, and the other is represented by BAG1, which contains a structurally and evolutionarily distinct BD. BDs from both BAG1 and BAG4 are three-helix bundles; however, in BAG4, each helix in this bundle is three to four turns shorter than its counterpart in BAG1, which reduces the length of the domain by one-third. BAG4 BD thus represents a prototype of the minimal functional fragment that is capable of binding to Hsc70 and modulating its chaperone activity.

Each of the human BAG proteins binds to Hsp70/Hsc70 and modulates their chaperone activity. The conserved BD is necessary and sufficient for this interaction (1,11). 2 Here we report the solution structure of the BD of BAG4 and its comparison with the BD of BAG1 (3,15). The BD in BAG4 is significantly shorter than its counterpart in BAG1 and may define a minimal structural unit that binds Hsp70/Hsc70. Our comparison reveals two subfamilies of BAG proteins that are structurally and evolutionarily distinct.

EXPERIMENTAL PROCEDURES
NMR Spectroscopy and Structure Calculation-BAG4 BD (residues 376 -457) was expressed in Escherichia coli as a glutathione transferase (GST) fusion construct. After initial purification of the recombinant protein by affinity chromatography on glutathione-agarose resin, the GST moiety was removed by thrombin cleavage, and the digest was separated using glutathione-agarose. Remaining impurities were eliminated by affinity chromatography on a benzamidine-agarose column and by ion-exchange chromatography on a Q-Sepharose column. NMR samples contained 1-2 mM 15 N-or 13 C/ 15 N-labeled protein, 10 mM potassium phosphate buffer, pH 7.2, 100 mM KCl, 1 mM DTT, and 1 mM EDTA in 90% H 2 O/10% D 2 O. Spectra were acquired at 30°C on Varian 500, 600, and 750 MHz spectrometers. The data were processed and analyzed with Felix 2000 (Molecular Simulations, Inc., San Diego). 1 H, 15 N, and 13 C assignments were established based on CBCA(CO)NH, HNCACB, HNCO, C(CO)NH, H(CCO)NH, HCCH-TOCSY, and threedimensional 15 N-edited NOESY. Distance restraints were obtained from three-dimensional 15 N-edited NOESY and three-dimensional 13 C/ 15 N-edited NOESY. and dihedral angle restraints were generated with TALOS (16). Structures were calculated with standard torsion angle dynamics (TAD) simulated annealing protocol implemented in CNS 1.0 (17) using restraints for 1153 interproton distances (120 long range, 5 Յ i-j , 199 medium-range, 2 Յ i-j Յ 4, 195 sequential and 639 intraresidual), 92 hydrogen bond distances, and 61 and 61 dihedral angles. The protocol consisted of high temperature TAD, followed by TAD and cartesian slow cooling stages and final minimization. The scale factor for the dihedral angle energy term was doubled throughout the calculation; otherwise, default parameters were employed. Of 28 structures, 25 structures were selected that had no distance and dihedral angle restraint violated by more than 0.5 Å or 5°, respectively. The statistics are summarized in Table I. The structure with the lowest energy was used for illustrations and discussion. All figures were generated with MOLMOL 2K.1 (18).
Peptide Synthesis and Binding Study-The peptide corresponding to helical region Asn-256 -Cys-267 of the ATPase domain of human Hsc70, which was predicted to bind to BAG4, was synthesized in a helixnucleated form to stabilize helicity (19) and characterized as described previously (15). 1 H-15 N HSQC spectra were recorded for 15 N-labeled BAG4 solutions containing varying concentrations of peptide, and most pronounced resonance shifts of BAG4 amides were noted and mapped onto the structure.
Computer Modeling-The homology models of BDs from BAG3 and BAG5 were created with SwissModel (22-24) using the sequence alignment shown in Fig. 2c. The coordinates were used to visualize charge and hydrophobicity distribution on molecular surfaces.   (17) bonds (
Comparison of BAG Domains from BAG1 and BAG4 -To compare the BDs of BAG4 and BAG1, we first superimposed their structures using only residues that are identical in the most conserved helix ␣3. Once ␣3 was overlaid, structurally equivalent residues were matched in the whole domain. The resulting structure-based sequence alignment is shown in Fig.  2c. As the alignment illustrates, BAG4 BD, relative to BAG1, contains a deletion of 19 residues between ␣1 and ␣2. This deletion results in shortening of ␣1 and ␣2 by 2.5 and 3.5 turns, respectively. BAG4 BD also lacks three turns of the ␣-helix at the C-terminal end of ␣3. All three helices in BAG4 BD are truncated at one end of the structure, making the BD of BAG4 significantly shorter (26 ϫ 12 ϫ 11 Å) than the BD of BAG1 (37 ϫ 14 ϫ 10 Å).
After the two domains were superimposed, ␣1, ␣2, and ␣3 in BAG4 BD matched closely with their BAG1 counterparts (3,15). The root mean square deviation (rmsd) of BAG4 backbone atoms in the three ␣-helices from equivalent atoms in murine and human BAG1 BD is 1.7 and 1.4 Å, respectively. (For comparison, rmsd between the same atoms of human and murine BAG1 is 1.2 Å). Also, the conformation of the loop between helices ␣2 and ␣3 is very similar in BAG4 and BAG1, even though it contains a single insertion, Met-169, 3 in the latter. Overall, the structure of BD from BAG4 closely resembles the upper two-thirds of the BAG1 BD (Fig. 2b).
Hsc70 Binding Interface on BAG4 BD-Most BAG1 residues that interact with Hsc70 (3) are conserved or are conservatively substituted in BAG4 BD (Fig. 2c). Hence, it can be predicted that BAG4 binds Hsc70 in a manner analogous to BAG1 (3,15). This was tested experimentally by NMR-monitored titrations of BAG4 BD with a synthetic peptide corresponding to an ␣-helix (residues Asn-256 -Cys-267) from the BAG1-binding interface on the ATPase domain of Hsc70. This helix contributes several predominantly basic residues to the intermolecular contacts with BAG1. In particular, Arg-258, Arg-261, Arg-262, and Thr-265 from Hsc70 form salt bridges with Glu-157, Glu-164, and Asp-167 from ␣2 and Asp-197 from ␣3 of BAG1 3 The sequence numbering for murine BAG1 is the same as that used in our previous NMR study reporting the solution structure of mBAG1 (15). Human BAG1 is numbered accordingly for clarity. When the crystal structure of human BAG1 was published (3), a different numbering scheme was used based on the sequence of BAG1M, a longer isoform of BAG1. For direct comparison with the sequence numbers for human BAG1 in the crystal structure, the reader should add 55 to the BAG1 sequence numbers used in the present study. Tube models of C␣ atoms from human BAG4 (red, residues 376 -457), murine BAG1 (blue, PDB ID 1I6Z, residues 99 -210), and human BAG1 (green, PDB ID 1HX1, residues 99 -205) are shown 3 ; the structures were superimposed using BAG4 backbone atoms located in ␣-helices (residues 380 -399, 407-423, 432-456) and their counterparts in BAG1 proteins. Orientation of the molecules is the same as in panel a. c, structure-based sequence alignment of BAG domains. BDs from human BAG4 (hBAG4), human BAG3 (hBAG3), human BAG5 (hBAG5/1, hBAG5/2, hBAG5/3, hBAG5/4, and hBAG5/5), human BAG1 (hBAG1), and murine BAG1 (mBAG1) were aligned as described in the text. Homology is highlighted by the default ClustalX coloring scheme (27). Positions of residues in hBAG1 that interact with N-and C-terminal lobes of Hsc70 are indicated with open and filled circles, respectively. hBAG4 and mBAG1 residue numbers are marked above or below the respective sequences. Secondary structure of hBAG4 and mBAG1 BDs is outlined on top.
(3). The peptide interacted with BAG4 BD; the most pronounced 1 H/ 15 N chemical shift changes in BAG4 BD induced by peptide binding were localized in ␣2 (Fig. 3, a and c). This is consistent with the expected peptide contact sites and with the results of NMR titrations of BAG1 with the same peptide (15).
To further define the binding interface in BAG4 BD across the conserved helices ␣2 and ␣3, we used site-directed mutagenesis of the predicted contact residues within these helices. The results indicated that Glu-414 and Asp-424 from ␣2, as well as Arg-438, Lys-439, and Gln-446 from ␣3 are important for the interaction of BAG4 with Hsc70, because mutating these residues to alanine abolished or weakened the binding (Fig. 3, b and c). The equivalents of these residues in BAG1 each make direct contact with Hsc70 (3). As expected, mutations in ␣1 (E388A, E389A) had no effect, because this helix does not interact with Hsc70. Thus the results indicate that the binding surface for the heat shock chaperone is similar in BAG4 and BAG1, involving ␣2-␣3 and extending through an area of ϳ10 ϫ 30 Å across one face of the domain.
Whereas most intermolecular interactions between BAG4 and Hsc70 are identical to those in the BAG1⅐Hsc70 complex (3), the contributions and relative importance of some residues differ. For example, the BAG4 E453A mutant still interacted with Hsc70, even though mutation of the equivalent residue in BAG1 (D197A, made as a double mutant with Q201A) resulted in a failure to bind Hsc70 (15). In BAG1, Asp-197 forms a salt bridge with Arg-258 in Hsc70 whereas Gln-201 is not involved in direct contact (3). Aspartic or glutamic acid is found at this position in all human BDs. It is possible that the longer aliphatic side chain of glutamic acid precludes formation of a stable intermolecular salt bridge as seen in the BAG1⅐Hsc70 complex, and consequently, this residue is not critical for BAG4⅐Hsc70 recognition.
Structure-based Sequence Alignment of Human BAG Domains-The sequence similarity between BDs of BAG4 and BAG1 is strongest in ␣3 (30% identical) and somewhat weaker in ␣2 (24% identical), so the alignment was clear for these two helices. However, until now, low levels of homology for the sequences in ␣1 obscured alignments of this region, even with knowledge of the BAG1 structure (1-3). In some cases, this segment of BDs was excluded from sequence alignments. Now, with the structures of two BDs in hand, it is possible to evaluate the conserved folding pattern of ␣1 and to more accurately predict sequence homology for the BDs. A structure-based sequence alignment is presented in Fig. 2c. The most significant correction to previous alignment attempts is the relative position of ␣1; BAG4 ␣1 spans residues 380 -399, corresponding to structurally equivalent residues 104 -123 in BAG1.
Whereas the sequence alignment of more distantly related BAG4 and BAG1 BDs required comparison of their structures, sequences of BDs from BAG3, BAG4, and BAG5 can be aligned with a high degree of confidence (Fig. 2c, compare with Refs. 2 and 3). BDs from BAG3 and BAG4 are the most closely related among human BDs, being 60% identical. BAG5 was shown previously to contain four BDs (20 -40% identical to BAG3 and BAG4) (1, 2), arranged in tandem and accounting for most of the protein. Now, our alignment has revealed that another segment in BAG5, located immediately after the first BD and predicted to contain three ␣-helices, is yet another BD, which we termed BAG5/BD2. For the alignment in Fig. 2c, we renumbered the remaining BDs from the identifiers cited previously (2, 10) to BAG5/BD3, BAG5/BD4, and BAG5/BD5. The role of these linked BDs in BAG5 is not yet understood.
Alignment of the putative BDs from BAG2 and BAG6/Scythe was not straightforward, and therefore we did not include their sequences in Fig. 2c. Even though both proteins bind to Hsc70 and inhibit its refolding activity (1,9), they are obvious outliers and their structures are likely to be different from other BDs. In BAG2, the C terminus of the protein itself terminates the putative BD shortly after the segment with homology to ␣2. Thus, if the folding pattern were retained for the first part of the domain, BD in BAG2 would lack most of ␣3. In contrast, the BD of BAG6 may lack the structural equivalent of ␣1 because the corresponding sequence contains several prolines that are likely to disrupt the helix. Only half of the residues whose counterparts in BAG1 contact the chaperone are conserved in Sites of alanine substitution, which are described in panel b, are depicted as spheres. Black, gray, or white spheres indicate mutations that abolished, weakened, or did not affect binding to Hsc70, respectively. The chemical shifts and mutations that affect binding are located in ␣2-␣3, marking an interaction interface that is closely similar to that seen in BAG1 (3,15). Orientation of the molecule is the same as in Fig. 2. BAG6. The region with ␣2-␣3 homology is necessary for the binding of BAG6 to Hsc70 (9). It remains to be seen whether this segment forms a structural domain per se or is part of a novel larger domain, together with the preceding proline-rich region or other parts of the protein.
Implications for Human BAG Proteins-To consider conserved structure/function relationships in the human BAG family, we used BAG4 BD as a template to construct threedimensional homology models for the BDs from BAG3 and BAG5. The sequence identity is 60, 39, 29, 22, 23, and 43% for BAG3 BD, BAG5/BD1, BAG5/BD2, BAG5/BD3, BAG5/BD4, and BAG5/BD5, respectively. The BDs from BAG2 and BAG6 were excluded from this comparison because they lack the clear homology seen in the other members of the family. Thus, a "gallery" of BDs was generated consisting of three experimentally determined structures (BAG4, murine BAG1, Ref. 15 and human BAG1, Ref. 3) and six homology models (Figs. 4 and 5). This gallery was used to compare the overall globular shape as well as surface features and protein interaction interfaces across the family. It should be noted for the homology models that the main chain conformation is likely to be accurate, but the positions of the side chains are less well defined. Our models provide a reasonable representation of overall surface feature distribution and can be useful in the absence of experimental structures of other members of the BAG family. BAG1, BAG3, and BAG4 interact with Hsc70 (1,2,6,11,20), and, consistent with that, the Hsc70-binding faces (helices ␣2 and ␣3) of BDs from these family members are very similar with respect to charge distribution (Fig. 5, top row). Acidic residues (Glu-413 and Glu-414 from ␣2, Asp-424 and Glu-427 from the connecting loop, and Glu-450 and Glu-453 from ␣3 in BAG4) dominate diagonally across the surface (Fig. 4a). The N-terminal portion of ␣3 presents a cluster of basic residues (Arg-435, Arg-438, and Lys-439 in BAG4), seen in the upper left corner of the domain (Fig. 4a). Some of the acidic residues and the basic cluster on the ␣2-␣3 face are shared by all the BDs. Similarly, a central hydrophobic region (Leu-420 and Leu-421 from ␣2 and Val-442, Ile-445, and Leu-449 from ␣3 in BAG4) is conserved through the family (Fig. 5).
Among the BAG5 BDs, BAG5/BD5 is the most canonical whereas BAG5/BD2 is the least. In fact, the charge distribution in BAG5/BD2 is rather unusual, resulting in a large dipolar moment on the ␣2-␣3 face of the domain. In BAG5/BD3 and BAG5/BD4, the cluster of acidic residues on ␣2 and ␣3 is not conserved, which could impede the interaction with Hsc70. In particular, the equivalents of the critical contact residue Glu-414 in BAG4 are cysteine in BAG5/BD3 and threonine in BAG5/BD4. When this glutamic acid is mutated to alanine in BAG4 or BAG1, binding to Hsc70 is abolished. The putative Hsc70 contact interfaces in BAG5/BD5 and also BAG5/BD1 are more closely similar to that in BAG4. In BAG5/BD5, this critical glutamate residue is conserved but an arginine is present at this site in BAG5/BD1. Interestingly, both of these BDs bind Hsc70 in in vitro assays. 2 Future studies are needed to reveal how BAG5 is organized, whether its five BDs act independently or interact with one another, and to identify the molecular targets of BAG5/BDs that do not bind Hsc70.
Short Versus Long BAG Domains-In our comparison of BDs from human BAG family members, we have demonstrated the presence of two structurally distinct BAG domains that divide the family into two subfamilies, characterized by the presence of "short" (BAG3, BAG4, and BAG5) or "long" (BAG1) BAG domains. Gene structure of these two BD types also differs and reflects separate evolutionary history. Whereas the BDs in BAG3, BAG4, and BAG5 are neither flanked by nor contain any introns, there are several phase 0 introns in the BD region of the human BAG1 gene. The existence of short and long varieties raises numerous questions about evolution and function of BDs. Because the short BD is sufficient for binding to Hsc70 and modulating its activity, what is the advantage of having each ␣-helix extended at one end of the long BD by three turns? Does the long BD represent the original form, and did the short BD arise by elimination of the part of the molecule that was not required for its function? Alternatively, did the long BD evolve from an ancestral short BD, with the extra residues enabling it to gain a new function?
To learn more about short and long BDs, we investigated which types are present in other species. Sequences similar to BDs are found in fungi, plants, and animals (1)(2)(3)  In panels a and b, the solvent accessible surface of the domain is colored according to electrostatic potential. Areas with negative, positive, or neutral character are depicted in red, blue, or white, respectively. In panel a, the view is the same as in Fig. 2, with helices 2 and 3 and the Hsc70 binding site facing forward. In panel b, the molecule is rotated 180°around a vertical axis relative to the view in panel a, thus revealing the opposite side with helix 1 in front. In panels c and d, the surface is colored according to hydrophobicity. Yellow color intensity is proportional to increasing hydrophobic character, and the front and back views of the molecule are displayed as in panels a and b. Charged and hydrophobic residues are labeled in the appropriate panels.  (1); Neurospora crassa, CAB88563), homology is limited to the Hsc70-binding region of ␣2 and ␣3, which is common to both short and long BDs. However, the C termini of the proteins limit ␣3 to a size typical of a short BD. Consistent with this observation, secondary structure prediction indicates that the three ␣-helices span a region of approximately 90 residues. Altogether, BDs in fungi are likely to be short. Also in plant BDs (2,3), only the chaperone-binding sequence of ␣2 and ␣3 is conserved. In this case, however, protein termini do not limit the size of the BDs, and secondary structure prediction is not straightforward. Classification as short or long may therefore require knowledge of the three-dimensional structure of these proteins.
The presence of the short BD in at least two kingdoms, animals and fungi, suggests that this form of BD is of ancient origin. It is not clear if the long BD is limited only to a part of the animal kingdom, or if it will be found elsewhere as well.
Interestingly, nematodes possess an orthologue of BAG1, which contains the long BD, but the short BD has not yet been identified in the completed genome of C. elegans. Does this imply that the short BD is absent in some animal species?
Obviously, many open questions remain. The origin of short and long BDs, point of their divergence, diversity of BDs within eukaryotes, and significance of their different lengths are still obscure. Also, even though the structures of BAG domains from human and mouse BAG1 (3,15) and from human BAG4 have been determined, it is not straightforward to predict structures of BDs from distantly related BAG family members whose sequence similarity is limited to Hsc70-binding regions of ␣2 and ␣3. Once the structures of BDs from yeast and plant proteins, as well as those from BAG2 and BAG6, are known they will provide new insights into the structure-function relationship in this diverse family of molecular regulators. Acknowledgments-We thank N. E. Preece for assistance with the Varian 500 MHz spectrometer, D. Kedra for help with sequence database searches, J. C. Reed for helpful discussions and critical review of the manuscript, and S. Hammond for assistance in preparing the manuscript for publication. This research was performed in part at the Environmental Molecular Sciences Laboratory (a national scientific user facility sponsored by the United States DOE Office of Biological and Environmental Research) located at Pacific Northwest National Laboratory, operated by Battelle for the DOE.  1-9), whose sequences are aligned in Fig. 2, are colored according to electrostatic potential (rows 1 and 2) and hydrophobicity (rows 3 and 4). Residues 99 -210 of murine BAG1 (PDB ID 1I6Z), residues 99 -205 of human BAG 3 (PDB ID 1HX1), residues 376 -457 of human BAG4, and corresponding residues from homology models of BAG3 and BAG5 BDs are shown. Coloring and orientation are the same as in Fig. 4 to permit direct comparison. Front views are shown in rows 1 and 3, whereas back views are presented in rows 2 and 4.