Structure of Ustilago maydis Killer Toxin KP6 α-Subunit

Ustilago maydis is a fungal pathogen of maize, some strains of which secrete killer toxins. The toxins are encoded by double-stranded RNA viruses in the cell cytoplasm. TheU. maydis killer toxin KP6 contains two polypeptide chains, α and β, having 79 and 81 amino acids, respectively, both of which are necessary for its killer activity. The crystal structure of the α-subunit of KP6 (KP6α) has been determined at 1.80-Å resolution. KP6α forms a single domain structure that has an overall shape of an ellipsoid with dimensions 40 Å × 26 Å × 21 Å and belongs to the α/β-sandwich family. The tertiary structure consists of a four-stranded antiparallel β-sheet, a pair of antiparallel α-helices, a short strand along one edge of the sheet, and a short N-terminal helix. Although the fold is reminiscent of toxins of similar size, the topology of KP6α is distinctly different in that the α/β-sandwich motif has two right-handed βαβ split crossovers. Monomers of KP6α assemble through crystallographic symmetries, forming a hexamer with a central pore lined by hydrophobic N-terminal helices. The central pore could play an important role in the mechanism of the killing action of the toxin.

Toxins are protein molecules that disrupt cell functions in a number of ways, some by making ion channels in cell membranes and others by interacting with membrane channels and/or receptors. Three-dimensional structures of several of these toxins have been determined, including cardiotoxin (1, 2), ␦-endotoxin (3), hemolysin (4), anthrax toxin (5), colicins (6), and diphtheria toxin (7). The tertiary structures of toxins vary widely from being almost entirely ␤-sheet, as in cardiotoxin, to nearly all ␣-helical, such as colicins. Ustilago maydis is a fungal pathogen of maize and is one of a number of fungi that secrete cellular killer toxins (8). These killer toxins are encoded by double-stranded RNA viruses in the cell cytoplasm (9). In the absence of an immunity or resistance gene, these toxins are lethal to the organism of origin and sometimes to closely related organisms but not to plant or animal cells. The U. maydis strains P1, P4, and P6 secrete toxins KP1, KP4, and KP6, respectively, all of which are low molecular weight (ϳ100 amino acids) polypeptides (10 -12). The crystal structure of the U. maydis KP4 toxin (a single polypeptide chain of 105 amino acids) has been determined and shown to possess a ␣/␤-sandwich fold (13) having a central five-stranded anti-parallel ␤-sheet and two ␣-helices (14). It appears to kill sensitive cells by blocking Ca 2ϩ channels (14). The SMK 1 secreted by the halo tolerant yeast Pichia farinosa also has a similar ␣/␤-sandwich structure (15). Williopsis mrakii killer toxin, having a fourstranded antiparallel ␤-sheet structure similar to that of ␤␥ crystalline (16,17), inhibits ␤-glucagon synthesis.
The U. maydis KP6 killer toxin gene has been cloned, sequenced, and expressed, and the protein was characterized (10). The toxin is unique in that two polypeptide chains, KP6␣ and KP6␤, are necessary for its killer activity (18). The 1234base pair double-stranded RNA P6M2 codes for a pretoxin of 219 amino acids, which is post-translationally cleaved by two endopeptidases, yielding ␣and ␤-subunits having 79 and 81 amino acids, respectively, as determined by cDNA sequence, N-terminal protein sequence analysis, and by mass spectroscopy. 2 It has been proposed that the KP6 toxin may act by either interfering with the K ϩ channel or binding to its own membrane receptor (10,19), thus depleting cellular K ϩ levels and eventually killing the cells. Here we present the crystal structure of the ␣-subunit of KP6 toxin determined by the isomorphous heavy atom replacement method and refined at a 1.80-Å resolution. We show that the symmetry-related trimeric or hexameric assembly of the subunit creates a central pore that could have functional implications.

EXPERIMENTAL PROCEDURES
KP6␣ was purified to homogeneity as described previously (18,20). The selenomethionine derivative was prepared by growing cells in a medium supplemented with 20 mg/ml selenomethionine. Native protein crystals were obtained by the hanging drop vapor diffusion technique (21). Hanging drops containing 5 l of 10 mg/ml protein and 1 l of 85% ammonium sulfate in 10 mM MES buffer, pH 6.0, were equilibrated against 1 ml of reservoir solution containing 18 -21% ammonium sulfate in 10 mM MES buffer. Single crystals shaped as hexagonal rods having typical dimensions of 0.15 ϫ 0.30 ϫ 0.30 mm 3 were obtained in about 20 days at room temperature. The crystal of the selenomethionine-substituted protein was obtained under the same conditions as the native crystal, but it took longer to grow to a size of 0.1 ϫ 0.2 ϫ 0.2 mm 3 . The heavy atom derivative was prepared by soaking a single native crystal in 200 l of 0.5 mM PIP (di--iodobis(ethylenediamine) diplatinum(II) nitrate) solution in MES buffer, pH 6.0, for 36 days. Three room temperature data sets for native, platinum derivative, and selenomethionine-substituted protein crystals were collected with the in-house RAXIS IIc image plate area detector, receiving x-rays from a Rigaku RU-200 rotating anode generator operated at 50 kV and 90 mA. Graphite-monochromated CuK ␣ radiation was used. Crystals used for * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The data collection were mounted in glass capillaries and sealed with mother liquor. One crystal was used for each of the data sets. The image plate detector was placed at 77.9, 120.0, and 130.3 mm from native, PIP derivative, and selenomethionine protein crystals, respectively, and the limiting resolutions for the three data sets were 1.8 Å, 2.2 Å, and 2.3 Å, respectively. The space group is P6 3 22, and cell dimensions are a ϭ b ϭ 48.30 Å, c ϭ 124.22 Å, with one molecule/asymmetric unit. The specific volume of the crystal is 2.41 Å 3 /Da, and the solvent content is about 42%. Data were processed using the DENZO program and scaled and merged using the SCALEPACK program (22,23). Details of the data statistics are listed in Table I.
Three platinum and one selenium positions were determined by HASSP (24) and confirmed with the Shake-and-Bake program package (25). Cross-phasing was carried out between the two derivatives to correlate platinum and selenium heavy atom positions through a common origin. Heavy atom positions were refined using HEAVYv4 (24) by the origin-removed Patterson technique. The phases were then calcu-lated using PHASES (26) combining both isomorphous and anomalous signals from PIP and the isomorphous signal from the selenomethionine derivative. Solvent flattening was carried out before the MIRAS (multiple isomorphous replacement signals from PIP and selenomethionine derivatives plus the anomalous signal from the PIP derivative) maps were calculated. Details of phase determination are summarized in Table II. MIRAS maps for both hands of the heavy atom positions were calculated, and the correct hand was chosen based on the quality of the maps and evidence of right-handed ␣-helices. Chain-tracing and protein model building were performed manually on a SGI Elan workstation using CHAIN (27). The location of the selenium position played a key role in the tracing of the chain by identifying the only methionine in the ␤-strand near the C terminus. Identification of the N-terminal ␣-helix also helped to complete the model building process. The entire polypeptide chain of 79 residues was built into this MIRAS electron density map prior to refinement. A 2F o Ϫ F c density map of the single methi- where ͉F p ͉ is the protein structure factor amplitude and ͉F PH ͉ is the heavy atom derivative structure factor amplitude.
where ͉F H ͉ is the heavy-atom structure factor amplitude, and E is the residual lack of closure error.
onine-containing ␤-strand is shown in Fig. 1 along with the independent determination of the selenium position from the cross-difference Fourier with the phases from the PIP derivative alone.
The model was refined with XPLOR (28) using the simulated annealing procedure. The resolution was gradually extended to 1.8 Å. Positional and isotropic parameters of a total of 690 non-hydrogen atoms including 77 solvent water oxygens and 1 sulfate ion were refined in the final model, using the 6838 reflections (F Ն 2 F ) in the range 1.8 -8.0 Å. Some of these densities modeled as water oxygens could possibly be other ions present in the crystallization medium. The agreement between the final model and the 2F o Ϫ F c density map was excellent for the main chain and side chains, except for one break at the C ␤ atom of residue Asp 37 , which is located on the surface of the molecule in a ␤-turn. The C-terminal Lys 79 side chain is also not well defined, probably because of dynamical disorder.

RESULTS AND DISCUSSION
Quality of the Model-The numbers from the final cycle of refinement are provided in Table III. The random positional error estimated from the Luzzati plot (29) is less than 0.20 Å. The average B values remain low throughout most of the residues of the structure. The agreement between side chain densities and the amino acid sequence is unambiguous, except in a few cases. Residues Asp 37 and His 38 are located at a turn which is exposed to the outer surface; consequently, their average B values are relatively high. The C-terminal Lys 79 also has a high B average value probably because of dynamical disorder. Although the refinement results shown are based on the unit weight scheme, nonunit weight schemes were also attempted. The atomic coordinates and the electron density maps resulting from these two modes of refinement were virtually identical.
Secondary and Tertiary Structures of a KP6␣ Polypeptide-A monomer of the KP6 killer toxin ␣-subunit shown in Fig. 2a forms a single domain structure having an overall shape of an ellipsoid of dimensions 40Å ϫ 26Å ϫ 21Å. The structure has a single split ␤␣␤ motif that belongs to the ␣/␤ sandwich family (30). It consists of a four-(␤ 1 , ␤ 3 , ␤ 4 , and ␤ 5 ) stranded antiparallel ␤-sheet, a pair of antiparallel ␣-helices (␣ 2 and ␣ 3 ) that lies approximately 20°to the ␤-strands above one side of the sheet, a strand ␤ 2 along one edge of the sheet, and a short N-terminal helix ␣ 1 on the other side of the ␤-sheet. The twist of the ␤-sheet is left-handed, as it usually occurs (13). Among the three linkages of the four-stranded antiparallel ␤-sheet, one belongs to a hairpin connection (␤ 3 to ␤ 4 ), and the other two are right-handed ␤␣␤ split crossover connections (␤ 1 to ␤ 3 and ␤ 4 to ␤ 5 ) via ␣ 2 and ␣ 3 , respectively (13,30). All the secondary structural elements are connected through six loops. However, only two of them are ␤ turns; one is a ␤ type II turn (13) (Leu 39 , Ser 40 , Lys 41 , and Ser 42 ) connecting the helix ␣ 2 with ␤ 3 , and the other one is a ␤ type I turn (13) (Ser 66 , Ser 67 , Leu 68 , and Asn 69 ) connecting ␣ 3 with ␤ 5 . There are eight cysteines in the structure as indicated in Fig. 2, a and b, all of which form intrachain disulfide bridges linking these secondary structural elements into a compact domain. Helices ␣ 2 and ␣ 3 are linked to the ␤-sheet through disulfide bridges: ␣ 2 to ␤ 4 through Cys 35 and Cys 51 in a right-handed conformation and ␣ 3 to ␤ 1 through Cys 65 and Cys 18 in a left-handed conformation. The remaining two disulfide bridges are both right-handed, one formed between the N-terminal helix ␣ 1 and the loop connecting it to ␤ 1 (Cys 5 and Cys 12 ) and the other between the two longest antiparallel strands ␤ 1 and ␤ 5 (Cys 16 and Cys 74 ). The C ␣ -C ␣ distances between a pair of cysteines range from 4.05 to 5.86 Å, depending on the conformation of each individual disulfide bridge. Besides disulfide bonds, the hydrogen bonding network holds the structure together. The hydrogen bonds involving main chain atoms among the five antiparallel ␤-strands are schematically illustrated in Fig. 2c. A total of 44 hydrogen bonds are formed among main chain atoms, ranging from 2.67 to 3.14 Å, with an average bond length of 2.93 Å. There are 16 hydrogen bonds between main chain and side chain atoms and 7 hydrogen bonds among side chain atoms ranging from 2.60 to 3.42 Å, with an average value of 2.99 Å. All water molecules are hydrogen-bonded to protein atoms either directly or indirectly through another solvent atom. No exposed cluster of hydrophobic side chains was found in the monomer. Three regions in this molecule interact through hydrophobic side chains that appear to stabilize the tertiary structure. As shown in Fig. 2b, among the three hydrophobic clusters, two located between the two antiparallel ␣-helices and the ␤-sheet are formed by residues Ala 20 , Leu 27 , Ala 30 , and Leu 68 at the top of the molecule (opposite of the N terminus) and by Leu 39 , Phe 53 , Leu 57 , Phe 61 , and Met 72 at the lower middle section of the molecule. The third region, located near the N terminus, is comprised of residues Ala 3 , Phe 4 , and Phe 8 forming a hydrophobic pocket with neighboring symmetry-related molecules. There are four   Fig. 2b. Because the pH of the crystallization buffer was 6.0, one of four of these histidines was deprotonated, which made it possible to coordinate to a heavy metal ion. The surface of the protein is hydrophilic except for the N-terminal helix-␣ 1 .
The topology of KP6␣ along with the other two recently determined killer toxins SMK (15) and KP4 (14) is illustrated schematically in Fig. 2d. Although all of them have striking similarity in folding motifs, i.e. ␣/␤ sandwich motifs, there are distinct differences among them; KP6␣ has two right-handed ␤␣␤ split crossovers, whereas both SMK and KP4 have two left-handed crossovers. A survey (30) has shown that there are more known right-handed ␣/␤ sandwich proteins than the lefthanded ones. Whether the handedness of connections has any significance in the biological function is not clear at this moment, but it is evident that all resulting structures are unique in their specific functions. Although the KP4 and KP6 killer toxins are both produced by U. maydis, the difference in their folding connections may indicate that they are not functionally similar.
Assembly of KP6␣ Molecules-As described above, the hydrophobic region with exposed phenyl rings of Phe 4 and Phe 8 of the monomer is located in the N-terminal ␣ 1 -helix. Three molecules of the KP6␣ monomer assemble through this region by a crystallographic 3-fold rotational symmetry forming a trimer. Two trimers further associate by a crystallographic 2-fold rotational symmetry axis perpendicular to the 3-fold axis forming a hexamer with the hydrophobic phenol basket in the middle of the assembly, as shown in Figs. 4 and 5. The trimeric association is stabilized by intermolecular salt bridges and hydrogen bonds; the bond lengths range from 2.77 to 3.00 Å with an average of 2.88 Å. The salt bridges are formed between Glu 15 and Asp 45 of one monomer and Arg 47 of another. The Ser 78 side chain and main chain carbonyl at the C terminus of one monomer form hydrogen bonds with side chains of Arg 28 and Tyr 29 of the neighboring monomer. Intermolecular interactions within a trimer are shown in Fig. 3. The hexamer is stabilized by hydrophobic interactions between the N termini of the two trimers, especially the Gly 7 -Phe 8 -Gly 9 sequence. FIG. 2. a, the overall structure is an ␣/␤ sandwich composed of a four-stranded antiparallel ␤-sheet (␤ 1 , ␤ 3 , ␤ 4 , and ␤ 5 ) with two antiparallel ␣-helices (␣ 2 and ␣ 3 ) on one side of the sheet, a ␤-strand (␤ 2 ) along one edge of the sheet, and a short N-terminal ␣-helix (␣ 1 ) on another side of the ␤-sheet. Disulfide bridges are drawn in yellow. The secondary structure elements have the following terminal residues ␣ 1 (2-8), ␤ 1 (14 -20), ␤ 2 (25-27), ␣ 2 (28 -33), ␤ 3 (42-45), ␤ 4 (50 -53), ␣ 3 (58 -67), and ␤ 5 (70 -77). The figure was drawn with SETOR (33). b, stereo view of a KP6␣ protein structure shown as a C ␣ trace. Residues in purple are involved in hydrophobic contact within the molecule. Four histidine residues are also shown with blue labels. Four disulfide bridges are shown in yellow; black labels indicate the locations of the cysteines. c, the main chain hydrogen bond network in the ␤-sheet. Only the backbone atoms of the polypeptide chain are shown. Hydrogen bonds were drawn with dashed lines. d, a comparison of the topologies of KP6␣, SMK, and KP4 toxins. The triangles and the circles represent ␤-strands and ␣-helices, respectively. Minor elements were shown in pale colors. The secondary structure assignment for the KP6␣ toxin was performed by the algorithm of Kabsch and Sander, as implemented in the program PROCHECK (34).
The network of intermolecular salt bridges and hydrogen bonds links the monomers to form a solvent-filled funnel-like structure in which the ␤-sheet of each molecule spreads on the inner surface and the N-terminal ␣ 1 -helix sits close to the tip of the funnel (Fig. 4, a and b). The funnel is 9 -10 Å across at the top, has a 6-Å diameter opening in the middle and narrows to a 4.2-Å diameter barrel bordered by three symmetry related phenyl rings arising from the Gly 7 -Phe 8 -Gly 9 string of residues at the N-terminal ␣ 1 -helix. More than 50 well ordered water molecules have been located in the funnel. Fig. 4, a and b shows the KP6␣ trimer with phenyl rings at the tip of the funnel and water oxygens inside of the funnel. The assembly of the hexamer brings the two narrow funnel tips face to face, thus forming an hourglass shaped structure as shown in Fig. 5. The distance between two planes of intratrimer salt bridges is about 34 Å.
The analysis of total accessible surfaces for the monomer, the trimer, and the hexamer bears out the above description of molecular association. The details are listed in Table IV. The total accessible surface area for KP6␣ monomer is 4783 Å 2 . Because of the formation of the trimer and the hexamer, total accessible surface areas for each monomer decreased by 21.2 and 24.2%, respectively. The loss of accessible surface area/ monomer because of oligomerization is 1014 Å 2 for the trimer and 1157 Å 2 for the hexamer. These numbers are in good agreement with the values of ϳ800 Å 2 for a homodimer and ϳ1000 Å 2 for a heterocomplex of a 10-kDa protein calculated by Jones and Thorton (31) for 59 different protein-protein complexes. Among lost accessible surfaces, the loss of hydrophobic surface contributed the most, followed by the charged surface and the polar surface. As an overall effect of oligomeric association of KP6␣, the percentage of the hydrophilic surface area, consisting of charged and polar residues, increased from 73.0% for the monomer to 75.6 and 78.6% for the trimer and the hexamer, respectively, making them more soluble in an aqueous medium. By forming hexamer from trimer, the gain or loss of both accessible hydrophilic and hydrophobic surfaces are in small magnitudes, ϳ3%, indicating a weak tendency to form the hexamer from the trimer.
Implications of the Quaternary Structure-It is conceivable that the crystallographic monomer, the trimer, or the hexamer of KP6␣, in association with one or more polypeptides of KP6␤ forms a molecular assembly that interacts with either the K ϩ channel or another receptor on the cell membrane or even by itself penetrates the membrane and causes disruption of the cellular ion balance. No conclusive biochemical evidence is available yet on any of these possibilities. The KP6␣ hexamer has an inner pore opening of 4.2 Å, large enough for monovalent cations such as K ϩ to pass through and a length of 34 Å, similar to the thickness of the membrane bilayer. Although neither KP4 nor SMK structures exhibit similar oligomeric association, the distantly related all ␤-strand ␥-cardiotoxin (32) is trimeric. However, the central opening of the trimer narrows to less than 0.5 Å where interfacial atoms are virtually at van der Waals contact distances (32). It was proposed that ␥-cardiotoxin in its oligomeric state could either interact with a membrane receptor or could insert itself into the membrane forming an ion channel (32). To our knowledge, no other member of this class of toxins is known to have a central pore like the one in the KP6␣ hexamer.
Cardiotoxins are known to act by depolarizing the Ca ϩ channel. KP4 was shown to kill cells by blocking divalent cation channels but not Na ϩ or K ϩ channels (14). KP4 crystallizes as a monomer and a single polypeptide chain is known to possess the killer action (14). Structural data suggest that its participation in channel formation is unlikely. The mechanism of pH-dependent killer action of SMK is still unknown (15). The monomer containing the ␣and the ␤-peptides forms a dimer in the crystal (15). Despite the overall similarity of folds of monomers of these toxins, there appears to be diversity in the mechanism of killing action, possibly stemming from diverse oligomeric states of their functional entities. KP6 is distinctive in the sense that it kills by depleting the K ϩ , unlike any other toxin in the family. The architecture of the KP6␣ hexamer provides a pore structure that has right dimensions to function as a K ϩ channel, either outside or within the cell membrane bilayer. However, the outer surface of the hexamer as well as the monomer has polar and charged residues, accounting for the solubility of the toxin in aqueous medium. In order for the KP6 assembly to enter the membrane bilayer, the outer polar surface must be concealed. It is conceivable that the ␤-subunit plays a critical structural role in this aspect because it is known that both subunits are necessary for the killer action of the toxin. Any large rearrangement in the tertiary structure of KP6␣ on complexation with the ␤-subunit is unlikely, because  of its compact nature that is rigidly held by four disulfide bridges. Formation of higher order oligomeric states either by the KP6␣ hexamer or by an ␣-␤ complex to hide hydrophilic surfaces from the lipid environment is also a plausible scenario. Further insight into the mechanism of action of KP6 may come from the crystal structure of an active complex of ␣and ␤-subunits of the toxin.