Motion of the DNA-binding Domain with Respect to the Core of the Diphtheria Toxin Repressor (DtxR) Revealed in the Crystal Structures of Apo- and Holo-DtxR*

The diphtheria toxin repressor (DtxR) fromCorynebacterium diphtheriae is a divalent metal-activated repressor of chromosomal genes that encode proteins responsible for siderophore-mediated iron uptake and also of the gene of certain corynebacteriophages that encodes diphtheria toxin. DtxR consists of two 25.3-kDa three-domain subunits and is a member of a family of related repressor proteins in several Gram-positive bacterial species, some of which are important human pathogens. In this paper, we report on the first high resolution crystal structures of apo-DtxR in two related space groups. In addition, crystal structures of Zn-DtxR were determined in the same two space groups. The resolutions of the structures range from 2.2 to 2.4 Å. The four refined models of the apo- and the holo-repressor exhibit quite similar metal binding centers, which do, however, show higher thermal motion in the apo-structures. All four structures reported differ from each other in one important aspect. The N-terminal DNA-binding domain and the last 20 residues of the dimerization domain of each subunit move significantly with respect to the core of the DtxR dimer, which consists of residues 74–120 from both subunits. These results provide the first indication of a conformational change that may occur upon binding of the holo-repressor to DNA.

Iron is an essential nutrient for almost all living organisms including pathogenic bacteria. The availability of free iron in the mammalian host is extremely limited, since most of the extracellular iron is associated with transferrin and lactoferrin, and most of the intracellular iron is bound to heme-containing proteins (1). The extracellular Fe 3ϩ concentration is usually restricted to concentrations below 10 Ϫ18 M, which is far too low to satisfy normal bacterial growth requirements. In addition, one important mammalian host response to infection is the release of additional transferrin and ferritin, thereby further reducing the amount of available iron (1). The ability to acquire ferric iron from the mammalian host is therefore an important key element of infection. In order to solve this problem, many bacteria have developed elaborate mechanisms to capture ferric iron, including siderophore synthesis and specific membrane proteins (1,2). Furthermore, numerous virulence determinants produced by bacterial pathogens, including a variety of toxins, hemolysin, and proteins involved in the iron uptake system, are regulated by iron (3).
In Corynebacterium diphtheriae, the diphtheria toxin repressor (DtxR), 1 is a global iron-dependent negative repressor that is activated by ferrous iron. In the presence of the co-repressor Fe 2ϩ , DtxR binds its target DNA sequences as a homodimer, thereby repressing the genes controlled by the DtxR-regulated promotor (4 -9). The most important gene regulated by DtxR is the bacteriophage tox gene that encodes diphtheria toxin (10). It has also been shown that the expression of siderophores in C. diphtheriae is regulated by DtxR (11). Whereas in vivo only Fe 2ϩ acts as co-repressor, in vitro several divalent transition metal ions including Fe 2ϩ , Ni 2ϩ , Co 2ϩ , Mn 2ϩ , and Cd 2ϩ , as well as Zn 2ϩ , can also function as activators (8 -11).
DtxR also binds to the promotor regions of the iron-regulated irp1, irp2, irp3, irp4, and irp5 promotor/operators (12,13). Very little is known about the proteins encoded by the genes downstream from DtxR binding regions. The gene product regulated by irp1 has recently been identified as a 38-kDa periplasmatic lipoprotein implicated in the iron uptake system (14). In addition, the expression of a heme oxygenase is regulated by iron and DtxR (15). Several DtxR homologs have been discovered in Gram-positive bacteria, including the major human pathogens Mycobacterium tuberculosis (16) and Mycobacterium leprae (17), the soil bacteria Streptomyces pilosis and Streptomyces lividans (18), and Brevibacterium lactofermentum (19). These homologs share an overall amino acid sequence identity of about 60%. DtxR can therefore serve as a model for a family of related iron-dependent repressors in Gram-positive bacteria in a similar manner as the ferric iron uptake repressor (Fur), which controls more than 30 genes in Escherichia coli and related bacteria, is a model for a different family of iron-dependent repressors in Gram-negative bacteria (20).
In order to understand the activation of DtxR by metals, it is essential to accurately determine the three-dimensional structures of the repressor with and without metal. Crystal structures of wild-type DtxR in complex with different divalent transition metals have been determined at 2.8-Å resolution (21) and of apo-DtxR at 3.0-Å resolution in a different space group (22). Recently, the resolution for Co-DtxR, Mn-DtxR, and Zn-DtxR has been extended to 1.85, 2.2, and 2.4 Å, respectively (23,24). These structures reveal an N-terminal domain of residues 1-73 that includes the DNA-binding helix-turn-helix motif and a dimerization domain (residues 74 -140) that contains two metal binding sites. Binding site 1 in all structures elucidated so far contains a metal ion with high occupancy. The metal at this site is tetrahedrally coordinated by the side chains of His 79 , Glu 83 , His 98 and by an oxygen of a sulfate anion. This site has also been designated the "cation-anion binding site" (23). The metal at site 2 is coordinated by the side chains of Glu 105 , His 106 , the carbonyl oxygen of Cys 102 , and a water molecule. This site had only partial occupancy in the Cd-DtxR and Mn-DtxR crystal structures (23). A similar site was also found to be occupied by Ni 2ϩ in the crystal structure of the Cys 102 3 Asp DtxR variant determined by Ding et al. (25). The third domain, comprising residues 148 -226, was found to adopt an Src homology 3-like conformation (23). However, this domain is highly flexible in all crystal structures determined so far, and its function remains unclear.
In this study, we report the first high resolution crystal structures of apo-DtxR at a resolution of 2.2 Å. Crystal structures were determined in two space groups. In crystal form I the molecule lies on a crystallographic two-fold axis with one monomer in the asymmetric unit, whereas in the related crystal form II the two-fold axis is noncrystallographic, and there is one protein dimer in the asymmetric unit. We also determined the crystal structures of DtxR in complex with zinc in the same two crystal forms to a resolution of 2.3 Å. The comparison of these four crystal structures provides the first evidence of the conformational changes that might be essential for the regulation of the repressor by its co-repressor. It turns out that the mechanism presumably involves the motion of the DNA-binding domain with respect to the metal binding domain rather than a rigid body motion of the monomers with respect to each other as suggested by Schiering et al. (22) on the basis of their 3.0-and 3.8-Å structures of apo-and Ni-DtxR, respectively.

EXPERIMENTAL PROCEDURES
Protein Expression, Purification, and Crystallization-DtxR was cloned and overexpressed in E. coli as described previously (5). The protein was purified using a nickel-nitrilotriacetic acid affinity column followed by anion exchange chromatography (11). DtxR in complex with zinc and sulfate was crystallized by vapor diffusion from 1.8 -2.0 M ammonium sulfate and 10 mM Zn 2ϩ using the hanging drop method as described by Qiu et al. (21). Crystal dimensions were typically 0.5 ϫ 0.3 ϫ 0.3 mm. The metal-free form was crystallized in the presence of 1 mM EDTA in all buffers. Prior to crystallization, the protein was dialyzed twice against a buffer of 10 mM Tris, pH 8, 50 mM NaCl, 10 mM DTT, and 1 mM EDTA to ensure the removal of all residual metal ions. The two crystal modifications were obtained under the same conditions, sometimes with both crystal forms in the same drop.
Data Collection and Processing-The data set of form II apo-DtxR was collected from two different crystals at room temperature mounted in capillaries. All other data sets were collected at cryogenic temperatures using crystals frozen in rayon loops (26). Before cryo-cooling, the crystals were transferred into a drop containing 1.2-1.5 M ammonium sulfate and 15-20% glycerol (24). The data sets for apo-DtxR and for crystal form I of Zn-DtxR were collected on an RAXIS-II imaging plate detector using monochromatic CuK␣ radiation and focusing mirrors. The data set of Zn-DtxR in form II was collected at beam line 14B at NSLS (Brookhaven, NY) at a wavelength of 1.28 Å. In all cases, data were processed using DENZO and Scalepack (27). Further key information is summarized in Table I.
Structure Solution and Refinement-The coordinates of Co-DtxR at 1.85-Å resolution (24) were used as a starting point for crystallographic refinement of form I crystals using simulated annealing and conjugate

.2-Å apo-DtxR form I (this study) and several holo-DtxR structures in crystal form I (Å)
Results are given after a least-squares superposition of 213 C ␣ atoms of residues 4 -140, 148 -197, and 201-225 of one subunit that are present in all crystal structures. The larger differences with Mn-DtxR may be due to the fact that this structure is the only one in the table that was determined at room temperature. The 4th, 5th, and 6th column give the r.m.s. deviations of domains I, II, and III after superimposing all 213 C ␣ atoms of one protein monomer. The average B-factors for domains 1 and 2 are 30 Å 2 for apo-DtxR, 24 Å 2 for Zn-DtxR, and 24 Å 2 for Co-DtxR. Domain 3 has significantly higher B-factors than the first two domains with average values for all atoms of 59 Å 2 for the apoand Zn-DtxR structures and 57 Å 2 for Co-DtxR.  (29) with a dimer comprising residues 4 -140 of the Co-DtxR structure (24) as a search model. Using data in the resolution range between 8 and 3.5 Å, the rotation function gave a 5.7 peak for the correct solution (second best 3.4 ). The translation function re-

FIG. 1. Stereo views of simulated annealing omit-maps of the anioncation binding sites.
The anion and all water molecules were omitted prior to one round of simulated annealing refinement to avoid any model bias. The F o Ϫ F c electron densities are contoured at the 3.0 level. A, apo-DtxR in crystal form I. The two peaks shown are 5.9 (Wat1), and 3.7 (Wat2), respectively. The 5.9 peak is the second highest water peak in the difference electron density. B, Zn-DtxR in crystal form I. The peak at the sulfate position has a height of 11.2 , and the water peak is 8.3 . The final coordinates of the refined model including the sulfate and the water are superimposed onto the density. C, apo-DtxR in crystal form II, monomer A. The water peak shown has a height of 3.9 . D, apo-DtxR in form II, monomer B; the water peaks shown are 4.5 (Wat1 and Wat2) and 4.2 (Wat3), respectively. All water positions depicted have reasonable distances to hydrogen bond donors and/or acceptors of the protein side chains and/or main chain atoms. This figure was prepared using the program O (35). sulted in a 17.8 peak with a correlation coefficient of 0.598 and R-factor of 49.3%. Rigid body refinement reduced the R-factor to 42.5% with a correlation coefficient of 0.626. At this point, the resulting electron density for the third domain was rather weak. In order to improve the density for the third domain, noncrystallographic averaging combined with histogram matching and solvent flattening was attempted using DM (30,31). However, this did not improve the resulting electron density, and the map correlation coefficient for the region of domain 3 was only 0. 35. Fitting an incomplete model of domain 3 into the A -weighted electron density in monomer A resulted in a drop of both R and R free by more than 2%. In contrast, including the third domain to monomer B did not result in any significant improvement in R or R free . These results indicated that in monomer B domain 3 is disordered to a much larger degree than in monomer A, and this domain was subsequently left out from the model. Inclusion of low angle diffraction data with a bulk solvent correction did not improve the resulting density. In all four refinements, water molecules were added using the maxima in the F o Ϫ F c difference electron density that were at least 3.0 above the mean and had reasonable hydrogen bond geometry. In all cases, 5% of the data were used to calculate the free R values (32,33). All model building and inspecting of A -weighted (34) 2F o Ϫ F c and F o -F c electron density maps were done using the program O (35). Further refinement statistics are summarized in Table II. The atomic coordinates have been deposited with the Protein Data Bank, Brookhaven National Laboratory (Upton, NY).

RESULTS
Quality of the Models-Crystal form I of both apo-and Zn-DtxR contains one monomer per asymmetric unit. The functional dimer is generated by the crystallographic two-fold axis in space group P3 1 (Table II). All residues fall within the allowed regions of the Ramachandran diagram (data not shown).
In crystal form II, the asymmetric unit contains a dimer for both apo-and Zn-DtxR. The dimer is the result of a noncrystallographic two-fold axis in space group P3 2 21. The first two domains are well defined in both monomers. Domain 3, however, appears to be severely disordered in one of the monomers where virtually no electron density was visible.   Fig. 4, the C ␣ atoms of only subunits A from the two DtxR dimers being compared are used for calculating the rotational and translational components of the superposition operation. This superposition operation is subsequently applied to both the A and the B subunit of the "second" dimer in the comparison. Ramachandran plot. Further information on the crystallographic refinements is given in Table II. Since the apo-and Zn-DtxR form I crystals diffract to higher resolution and the model is also more complete in these structures, they will mainly be used for the detailed comparison of the metal binding sites of the apo-and holo-repressor described below. The overall structures of apo-DtxR and of DtxR bound to its divalent metal ion co-repressor are very similar; the r.m.s. deviations of C ␣ -atoms range from 0.3 to 0.5 Å (see Table III). However, a careful analysis of the structures reveals significant domain motion of the DtxR subunits, as will be described below after comparing the metal binding sites in apo-and holo-DtxR. Metal Site 1-As mentioned before, metal binding site 1 has been observed to be occupied in all high resolution crystal structures of DtxR determined in the presence of divalent metal ions so far (21,23,24). The metal was found to be tetrahedrally coordinated by N ⑀2 of His 79 , O ⑀1 of Glu 83 , N ␦1 of His 98 , and a sulfate (or phosphate) ion. In apo-DtxR the metal binding site 1 is clearly not occupied (Fig. 1A). Yet no significant conformational changes appear to occur near site 1 upon metal removal. The r.m.s deviations of all 29 atoms of the metal-coordinating residues His 79 , Glu 83 , and His 98 range between 0.1 and 0.2 Å when comparing the apo-and holo-DtxR subunits in all four structures with only minor changes in side chain conformations (Fig. 2). The average B-factors for residues His 79 , Glu 83 , and His 98 are 22 Å 2 for apo-DtxR and 11 Å 2 for Zn-DtxR (see Table II). The side chains of the coordinating residues have apparently more degrees of freedom in the apostructure than in the metal-containing repressor.
The presence of a sulfate anion as the fourth ligand in Zn-DtxR was confirmed by the height (11 ) of the peak at the anion site in the F o Ϫ F c electron density and the shape of the density, which strongly suggests the presence of a tetrahedral anion (see Fig. 1B). These peaks have a relative height of 9.9 and 9.1 , respectively, in the two subunits of Zn-DtxR form II (data not shown), a clear indication that the anion binding site is occupied in all three independent Zn-DtxR subunits in crystal forms I and II. On the other hand, the anion binding site is empty in all three apo-DtxR subunits determined, and water peaks (of 5.9 in apo-DtxR form I and 3.9 and 4.5 in the two subunits of form II) appear about 2.6 Å from the center of the sulfate anion observed in the metal-containing structures (Fig.  1, C and D). The absence of the anion in all three apo-DtxR subunits in the two crystal forms suggests that the metal cofactor is necessary for anion binding.
Metal Site 2-In wild type DtxR, metal binding site 2 has so far only been observed to be partially occupied in the 2.8-Å Cd-DtxR (21) and the 2.2-Å Mn-DtxR structures (23). In both of those cases, the metal appears to be coordinated by O ⑀1 of Glu 105 , N ⑀2 of His 106 , the carbonyl oxygen of Cys 102 , and a well ordered water molecule. A comparison of the apo-DtxR and Mn-DtxR structures shows no significant conformational changes of this binding site (Fig. 3) with a r.m.s. deviation for all non-hydrogen atoms of the coordinating residues of 0.2 Å. The average B-factors of these residues are also very similar in the apo-and the manganese-containing structures: 21 Å 2 for apo-DtxR and 22 Å 2 for Mn-DtxR.
In both Zn-DtxR structures described in this paper, site 2 does not appear to be occupied by a metal ion. There is no density present except for a water molecule (data not shown). It has been suggested that the lack of metal binding at site 2 is due to the formation of a persulfide or a mixed disulfide at cysteine 102 (23,24). In all metal-bound DtxR crystal structures investigated so far, including the two Zn-DtxR structures described in this paper, the F o Ϫ F c difference electron density had its strongest peak at approximately 2 Å from the S ␥ of Cys 102 , which, based on geometric criteria, can be interpreted as a sulfur atom bound to the S ␥ (23,24). However, in the present study, this peak is significantly weaker in apo-DtxR crystal form I and not present in crystal form II of apo-DtxR. The latter structure was determined from fresh crystals and crystallized with 10 mM dithiothreitol present at all times to prevent oxidation. Nevertheless, in the apo-structures, the side chain of Cys 102 adopts the same conformation as in the metalbound DtxR structures determined previously. These results may suggest that this might be the biologically relevant conformation for the cysteine side chain in the apo-as well as the holo-repressor. However, there is no steric hindrance that would prevent the cysteine side chain from adopting a conformation that would bring the S ␥ closer to metal site 2. Such a motion would be in agreement with reports that Cys 102 plays an important role in the mechanism of activation as evidenced by site-directed mutagenesis studies, which showed that a replacement by all amino acids but Asp results in a loss of metal-activated DNA-binding in in vivo assays (9). Elucidation of the role of Cys 102 in metal binding might have to await the structure determination of the fully reduced form of DtxR with both metal binding sites occupied.
Comparison of Apo-and Zn-DtxR Structures: Domain Motion with Respect to the Core-The crystal structure of DtxR in complex with different divalent cations was originally solved in our laboratory in crystal form I (21), whereas crystal form II was initially reported by Schiering et al. (22). Both forms were crystallized from similar conditions and are closely related. In form I, the protein dimer lies on a crystallographic two-fold axis, and there is one monomer in the asymmetric unit. In form II, the crystallographic two-fold axis is slightly shifted and becomes a noncrystallographic two-fold axis. Consequently, the c axis is doubled, and the space group changes from P3 1 21 (form I) to P3 2 21 (form II). Schiering et al. (22) obtained a structure of apo-DtxR at 3.0 Å resolution in a crystal modification similar to form II (R ϭ 0.233, R free ϭ 0.413) and suggested on the basis of a comparison with a Ni-DtxR structure at 3.8 Å in form I (R ϭ 0.193, R free not reported) that binding of the metal would cause a small change in quaternary structure between the different crystal forms, thereby activating the repressor.
A comparison of the four high resolution structures of apo-DtxR and Zn-DtxR in the same two space groups described in the present paper shows that not differences in quaternary but rather in tertiary structure are most likely to be involved in repressor activation. For this analysis, we superimposed, as a first step, the 137 C ␣ atoms of the first two domains of monomer A of the DtxR dimer onto each other. Next, we evaluated, overall as well as per residue, how much subunits A and B deviate from each other after this "A on A" superposition. The results clearly show that the B subunits deviate in each case significantly more than the A subunits (Table IV). This suggests that the B subunits, or parts thereof, undergo a motion with respect to subunit A. A more detailed analysis of the most deviating pair of protein dimers, apo-DtxR form II versus Zn-DtxR form I, provides clear evidence for a domain motion within the subunit (see Fig. 4A). The key observation is that residues 74 -120 of the B subunits superimpose as well or even better onto each other than any part of subunit A, although none of these atoms was used in calculating the parameters for this "A on A" superposition. Evidently, residues 74 -120 of the B subunits do not move with respect to the A subunit. The deviations for B subunit residues 4 -74 that form the DNA binding domain and for residues 121-140 at the C terminus of the second domain are much larger than for the B subunit residues 74 -120 (Fig. 5). As residues 74 -120 of subunits A and B make numerous interactions with each other, we can define an immobile "core" of the DtxR dimer consisting of these residues in both subunits. The rest of each subunit can move with respect to these 92 core residues. All other possible combinations of pairwise superpositions of DtxR dimers exhibit the same result, although the actual differences are smaller. The core deviates by only 0.15-0.21 Å, while the rest of the subunits differ by r.m.s. deviations for the DNA-binding domain up to 0.84 Å ( Table V). The DNA-binding domain is observed to be able to rotate with respect to the core and exhibits a change in the angle of the putative DNA recognition helixes H3 and H5 by 1.8°(161.5°in Zn-DtxR form I and 159.7°in apo-DtxR form II as calculated using the program EDPDP (39)). In addition, some C ␣ atoms of the recognition helix H3 are shifted significantly in the four structures determined, up to 1.7 Å of residue 39 when comparing the apo-DtxR form II and Zn-DtxR form I dimers (Figs. 4 and 5). The ability to move the recognition helices with respect to the constant core of the DtxR dimer is likely to be of critical importance to allow the repressor to adopt the optimal conformation for interacting with its cognate DNA. DISCUSSION The DtxR dimer can be described as consisting of the following elements: (i) a rigid core formed by residues 74 -120 of subunit A plus the same residues in subunit B; (ii) the Nterminal DNA-binding domain, which can rotate with respect to the core; (iii) the C-terminal residues 121-140 of the dimerization domain, which vary more in conformation than the core,  and (iv) the hyperflexible C-terminal third domain. Fig. 4B illustrates this domain organization. The core of the apo-DtxR dimer is depicted in yellow, the two DNA-binding domains in each monomer in blue, and the C-terminal residues in green.
Interestingly, our present studies do not reveal any correlation between metal content and structural differences, since the smallest structural changes are observed between apo-DtxR in form I and Zn-DtxR in form I (Fig. 6). Hence, crystal packing effects also play a role regarding the differences observed in the orientation of the DNA-binding domain in the four structures reported. Nevertheless, the motions of the DNA binding domain seen in Figs. 4 and 5 are most intriguing, since they are the type of motions one would expect to occur upon binding the co-repressor. Clearly, the DNA-binding motifs are able to move, presumably, to interact better with DNA. Even small changes in the geometry of the two DNA recognition helices with respect to each other might have a large effect on the affinity. It seems as if the motion observed in the apo-and Zn-DtxR crystals reported in this paper shows a tantalizing glimpse of what may occur in the living cell upon DNA binding. The changes seen in Figs. 5 and 6 are to a first order of approximation a motion of residues 1-73 in subunit B with respect to the rest of this subunit. It is gratifying that this amounts to an experimental determination of the border of the first two domains in DtxR, which coincides quite precisely with the assignment initially given by us in the first report on the DtxR structure (21).
In evaluating these results, it may be useful to note that the modification of Cys 102 , which is observed time and again in high resolution metal-containing wild type DtxR structures (23,24), may be preventing us from unraveling the structure of the true holo-repressor, which should contain two fully occupied metal binding sites and not one as has been the case in virtually all wild type DtxR structures reported so far. The possibility cannot be excluded that in the true holo-repressor, with an unmodified Cys 102 , the DNA binding domains are even more differently positioned with respect to the DtxR core domains than has been observed in the apo-and metal-containing DtxR structures reported in this paper. In addition, it is also possible that binding cognate DNA might lead to additional conformational changes of the holo-repressor. Further investigations will hopefully lead to answers to these intriguing questions.
It is obvious that conformational changes are not the only way the repressor might be regulated by metal ions. For instance, for the Met repressor it has been proposed that the positive charge of the co-repressor S-adenosylmethionine is a key ingredient of the stronger interaction of the holo repressor than of the apo-repressor with DNA (36 -38). In the absence of information concerning the structure of holo-DtxR in complex with cognate DNA, we refrain from discussing here further this potential electrostatic mechanism. Our studies presented in this paper not only describe the first high resolution structures of apo-DtxR but also show significant domain motions within the DtxR subunits, which are likely to be important for the functioning of the repressor.