Structure of Human Spindlin1

Spindlin1, a meiotic spindle-binding protein that is highly expressed in ovarian cancer cells, was first identified as a gene involved in gametogenesis. It appeared to be a target for cell cycle-dependent phosphorylation and was demonstrated to disturb the cell cycle. Here we report the crystal structure of human spindlin1 to 2.2Å of resolution, representing the first three-dimensional structure from the spin/ssty (Y-linked spermiogenesis-specific transcript) gene family. The refined structure, containing three repeats of five/four anti-parallel β-strands, exhibits a novel arrangement of tandem Tudor-like domains. Two phosphate ions, chelated by Thr-95 and other residues, appear to stabilize the long loop between domains I and II, which might mediate the cell cycle regulation activity of spindlin1. Flow cytometry experiments indicate that cells expressing spindlin1 display a different cell cycle distribution in mitosis, whereas those expressing a T95A mutant, which had a great decrease in phosphorous content, have little effect on the cell cycle. We further identified associations of spindlin1 with nucleic acid to provide a biochemical basis for its cell cycle regulation and other functions.

Spindlin, a major maternal transcript in Mus musculus, was named for its association and co-migration with the meiotic spindle in the first meiotic cell cycle (1). It can be periodically phosphorylated during meiosis, which modulates its ability to associate with the meiotic spindle (2). The modification of spindlin depends at least partially on the Mos/mitogen-activated protein kinase pathway (2), which is controlled by meiotic checkpoint proteins cyclin B and Cdc2 (3,4). As an ovarianspecific protein, its role in sperm development seems to be ful-filled by ssty (Y-linked spermiogenesis specific transcript) (1), a multicopy testis-specific spermatogenesis gene on the long arm of mouse Y chromosome whose dosage reduction was suggested to cause deformed sperm heads and infertility (5-7). These two genes share more than 50% identity in amino acid sequence ( Fig. 1), and together they form a new spin/ssty gene family. Homologues of spin/ssty family genes are found in Rattus norvegicus, Xenopus laevis, and Oryzias latipes but not in Drosophila melanogaster or Caenorhabditis elegans, indicating that this gene family is restricted to vertebrates (8). Furthermore, bioinformatics analysis suggests that the spin/ssty family proteins are composed of three repeats of a new protein motif ϳ50 amino acids in length (8).
Spindlin1, a human homolog of spindlin, has been shown to be related to ovarian cancer (9). Its expression is up-regulated in ovarian cancer cells but not in normal tissues and was found to localize in cell nuclei. The transfected cells, which are prone to grow as cancer cells in nude mice, showed a complete morphological change (10). Overexpression of spindlin1 might lead to variations in cell cycle distribution during mitosis, which is consistent with previous reports for other spindlin1 homologs (11).
Although previous studies indicate that members of the spin/ ssty protein family play important roles in tumorigenesis and early embryogenesis, their biochemical functions and mechanisms are largely unknown. Here we report the crystal structure of human spindlin1 at 2.2 Å of resolution, which represents the first crystal structure from the vertebrate spin/ssty gene family. The structure consists of three tandem repeats of Tudor-like domains, which represent a novel tandem repeat fold. We identify Thr-95 as important for the function of spindlin1, which is a key residue for the coordination of two phosphate ions. In addition, our preliminary biochemical data also indicate that spindlin1 associates with nucleic acid. These data provide a structural basis for further biochemical analysis of spindlin1 and the spin/ssty gene family.

EXPERIMENTAL PROCEDURES
Protein Expression, Purification, Crystallization, and Mutagenesis-The protocols for purification and crystallization of human spindlin1 have been described previously (12). The T95A mutant was generated using GCA to replace original codon by PCR and purified similar to the wild type protein.
Both wild type spindlin1 and the T95A mutant were constructed into pCDNA3.1 myc/his(Ϫ) for cell cycle analysis.
Heavy Atom Derivatization-Crystals of spindlin1 were transferred into a solution containing 35% polyethylene glycol 6000, 100 mM Tris-HCl, pH 8.0, in a stepwise manner. The mercury derivative was obtained by soaking spindlin1 crystals in this buffer supplemented with 1 mM ethyl mercuric phosphate ((C 2 H 5 HgO)HPO 2 ) for 3 days.
Structure Determination and Refinement-Phases for the mercury-spindlin1 crystal were initially determined by the single-wavelength anomalous diffraction technique using CNS (14). The phasing power was calculated as 2.4, and three heavy atom sites were independently located by the heavy atom search routine, yielding an overall figure of merit of 0.41 after calculation of initial single-wavelength anomalous diffraction phases at 3.0 Å. The resolution was extended to 2.3 Å using CNS. After solvent flipping, the quality of the initial electron density maps was greatly improved. Initial manual model building and fitting were carried out using 2.2 Å-resolution native data in O (15). Positional refinement, B-factor refinement, and water molecules were added using CNS. Data collection, processing, phasing, and refinement statistics are given in Table 1.
Cell Cycle Analysis-The cell cycle phase distribution of HeLa cells was examined by flow cytometry using FACScan and Cell Quest software (BD Biosciences). 1 ϫ 10 6 cells were cotransfected with pBB14 (green fluorescent protein) and pCDNA3.1 myc/his(Ϫ) vector, spindlin1 wild type, or spindlin1 point mutants. After incubation in full media for 20 h, HeLa cells were harvested and prefixed using 0.5% paraformaldehyde. The cells were washed with phosphate-buffered saline and fixed in phosphate-buffered saline/ethanol for 1 h. The cells were stained with propidium iodide (50 g/ml) for 30 min after RNase digestion and analyzed.
Gel Shift Assays-For DNA binding assays, purified samples of about 30 bp of double-stranded DNA (dsDNA) 4 probe, 30 bp 4 The abbreviation used is: dsDNA, double-stranded DNA. of DNA primer, and a constructed spindlin1 vector of ϳ2.5 kilobases were employed. Binding reactions were conducted in 20 mM Tris, pH 8.0, 150 mM NaCl, 0.2 mM EDTA, 10%(v/v) glycerol, 1 mM dithiothreitol, 0.05 mg/ml bovine serum albumin, and 0.05% Triton-100 to final sample volumes of 10 l. After 20 min of incubation at room temperature, 1 l of loading buffer containing 0.05% bromphenol blue was added to the reaction mixtures. 1% agarose gels were run at room temperature at a constant voltage of 12 V⅐cm Ϫ1 in 1ϫ Tris borate EDTA and afterward stained with ethidium bromide.

RESULTS
Overall Structure of Spindlin1 Folds into Three Similar Domains-The crystal structure of recombinant spindlin1 was determined by single-wavelength anomalous diffraction from a single crystal soaked with mercury. The initial model was built into an electron density map calculated to 2.3 Å resolution, and data from a native crystal allowed model refinement to 2.2 Å. The two spindlin1 molecules in one asymmetric unit were traced in two (residues 25-170 and 181-235 in molecule A) (Fig. 2a) and four (residues 27-91, 104 -115, 125-168, and 184 -234 in molecule B) fragments totaling 237 residues, respectively. No electron density was evident for the other residues. Spindlin1, with dimensions of ϳ40 ϫ 40 ϫ 35 Å, exhibits an all-␤ structure that consists of 14 ␤-strands and two short ␣-helices. The structure is folded into three structural domains that are composed mainly of ␤-barrel-like structures. The three domains, which correspond to residues 25-90, 101-168, and 191-234, consist of ␤1-␤5, ␤6 -␤10, and ␤11-␤14, respec-tively. The ␣-helices are located in domain II, ahead of its first and fifth ␤-strands respectively (Fig. 2, a and b).
Sequence alignment among the three domains shows a total sequence identity of 32%, with high conservation of some aromatic residues and nonpolar residues (Fig. 2d). These residues either contribute to the hydrophobic core of each domain or for the hydrophobic interactions among the three domains. The conserved glycines and some asparagine residues are important for the protein fold. Gly-30, Gly-109, and Gly-190 are significant for the formation of the first ␤-strand in each domain. Asn-35-Gly-36, Asp-120 -Gly-121, and Asp-199 -Gly-200 form ␤-turns between the first and second ␤-strands in each domain, whereas the flexibility of Gly-43, Gly-128, and Gly-205 should be crucial for the bends of strands ␤2, ␤7, and ␤11.
There are a number of other conserved glutamate, aspartate, asparagine, glutamine, and serine residues that are distributed on the surface of the protein, but their structural roles are less clear (Fig. 2d).
Domains I, II, and III Possess a New Tudor-like Tandem Fold-A DALI search for structural similarity to the individual domains I, II, and III retrieved ϳ30 structural homologs for different fragments and suggested Tudor domains and malignant brain tumor repeat domains as closest matches. The spindlin domains and Tudor domains all compose two sheets that are made up of five anti-parallel ␤ strands. The portions of the second strands of these domains all participating in the two sheets are delineated by a kink that changes the direction of the polypeptide (Fig. 3a). Furthermore, residues that assist in forming the Tudor-like domain fold are highly conserved between domains I, II, and III of spindlin1 and the Tudor domain (Fig.  3b). However, a DALI search did not yield any match for the overall structure, suggesting the spindlin1 structure is a novel arrangement of Tudor-like domains.
reflections that was selected before refinement calculations and not included in the refinement. c Ramachandran plots were generated using PROCHECK.
Although spin/ssty repeats and Tudor domains share similar folds, there are some significant differences between them. First, superposition of these domains showed some significant disparity in the backbones of the five-strand structure. The orientation of the fifth strands of the spin-dlin1 domains is completely diverse from that of the Tudor domains. In addition, comparison of the anterior five ␤-strands gives a root mean square deviation of about 3.1 ϳ 4.5 Å, indicating considerable differences between the two kinds of domains. Furthermore, previous studies have shown that the ␤1-␤2 loop and the ␤3-␤4 loop together with the regions between these two loops are of great importance for its function, but the residue organization in the spindlin domains is not essentially conserved with the known Tudor domains (Fig. 3, a, c,  and d). All of these observations suggest that spin/ssty repeats might have a different function or at least function in a different mode than Tudor domains.
The Interactions between Spin-dlin1 Molecules in Crystal Structure and Solution-From the crystal structure, there are two molecules in one asymmetric unit that are essentially identical with a root mean square deviation of 0.4 Å for all C ␣ atoms. Examination of the crystal packing reveals the occurrence of a major interface between the two spindlin1 molecules. Each spindlin1 molecule offers two surfaces (surface A and surface B) for association with another molecule. Surface A is defined by strands ␤11, ␤14, and the loop ␤10 -␤11, whereas surface B is defined by strand ␤2 and the loops ␤1-␤2, ␤3-␤4, and ␤13-␤14.
The two spindlin1 molecules associate through an interface that involves surface A of one molecule and surface B of another molecule, the interactions of which include van der Waals contacts, steric complementarily, and hydrogen bond contacts. First, the interaction between surface A and B buries about 624 Å 2 of solvent-accessible area, which would help to stabilize the dimer architecture. Meanwhile, the first strand (␤11) of domain III in molecule A forms an anti-parallel ␤-sheet with strand ␤2 of domain I in molecule B. As a result, residues Pro-43 and Thr-45 in molecule B participate in the formation of the hydrophobic core of domain III of its neighboring molecule. It should be noted that domain III is composed of only four ␤ strands, which is one strand less than domain I and II, and yet this ␤-barrel-like structure is closed by strand ␤2 from the neighboring molecule, in a similar way to strands ␤5 and ␤10 in the first two repeats. Thus, the "missing" fifth strand of this spin/ssty repeat is compensated at least in part by molecular interactions. The hydropho- , which should also contribute to the interaction (Fig. 4, a and b).
Consistent with our structure, spindlin1 maintains a dimeric state in solution, as confirmed by dynamic light scattering and size exclusion chromatography (data not shown). Cross-linking assays using ethylene glycol succinate as linker also showed that spindlin1 exists largely as a homodimer in solution (Fig. 4c).
Thr-95 Is an Important Residue for Phosphate Ion Binding-Two phosphate ions maintained by hydrogen bonds in the loop between domain I and II (P loop) were found in the refined structure of spindlin1 and named c301 and c302, and their position in the overall structure is shown in Fig.  2a. Phasing with heavy atom assigned as phosphorous gives strong anomalous difference electron density at the phosphate binding site, confirming the existence of phosphates (Fig. 5a). Atomic emission spectroscopy revealed that spindlin1 associates with phosphate ions in solution, thus giving a characteristic line for phosphorus in phosphate free buffers.
Several arginine and lysine residues, but no aspartate or glutamate residues, are located in the phosphate binding surfaces, which form a positively charged surface region. The electrostatic interactions between these cationic residues and anionic phosphate groups should contribute to the phosphate binding. For further stabilization, the phosphate groups also form a number of hydrogen bonds with spindlin1.
There are 10 hydrogen bonds between c301 and spindlin1. The O1 oxygen of phosphate c301 forms three hydrogen bonds; two hydrogen bonds are formed directly with the OG1 and N atoms of Thr-95, and another hydrogen bond is mediated by the water molecule S40 with the main chain oxygen of Val-93. The O2 oxygen also involves three hydrogen bonds; two hydrogen bonds with the NH1 and NH2 atoms of Arg-92 and another hydrogen bond mediated by water molecule S40 with Val-93. The O3 and O4 oxygens of phosphate c301 both form two hydrogen bonds with the polypeptide. The OG1 and N of Thr-139 form two hydrogen bonds with O3, whereas the last two hydrogen bonds are generated between O4 and the N atoms of Thr-95 and Asn-138, respectively (Fig. 5b). The detailed length of each hydrogen bond is listed in Table 2.
Compared with c301, the interactions for the c302 phosphate ion are weaker with a total of eight hydrogen bonds. The O1 oxygen of c302 involves the formation of three hydrogen bonds; two hydrogen bonds with the NH2 and NE atoms of Arg-97 and one mediated by water molecule S4 with the O atom of Ile-98. The O2 oxygen of c302 forms two hydrogen bonds with NH1 of Arg-133 and OD1 of Asn-138. The O3 oxygen contributes one hydrogen bond with NH2 of Arg-133, whereas O4 forms one hydrogen bond with NH2 of Arg-97 and one mediated by water S98 with the ND2 atom of Asn-138 (Table 1, Fig. 5c).
From our structure, the long P loop (residues 90 -100) is relatively stable (Fig. 2a) with an average B factor of 29.7 for main chain atoms. There are five residues in this loop that interact with the phosphate ions either directly (Arg-92, Thr-95, and Arg-97) or indirectly via hydrogen bonds mediated by water molecules (Val-93 and Ile-98). Thus, the presence of phosphate c301 and c302 should greatly benefit the stabilization of this loop.
Of these hydrogen bonds to the two phosphate ions, the interactions associated with the phosphate O1 atom are the strongest, as evidenced by the electron density between the phosphate O1 atom and OG1 of Thr-95 seen clearly even at a contour level of 3. From atomic emission spectroscopy experiments, the phosphorus content in wild type spindlin1 is 62 Ϯ 2 g/mol, whereas only 24 Ϯ 3 g/mol could be measured for the T95A mutant, indicating a greater decrease of the phosphate content in this mutant. This mutant could not be crystallized with the native crystallization conditions, which indicates that the structure of spindlin1 is at least partially influenced by the presence of phosphate. Thus, Thr-95 is a critical residue for the binding of phosphate ions.
The Loop between Domain I and II Is the Key Site for Cell Cycle Regulation Activity-To examine the potential effects of the phosphate ions on the function of spindlin1, the cell cycle distribution of control HeLa cells were compared with fluorescence-activated cell-sorted green fluorescent protein-positive HeLa cells expressing either wild type spindlin1 or the T95A mutant. All cells were incubated at 37°C for 24 h after transfection with their respective vectors and were then applied to fluorescence-activated cell sorter to analyze the proliferation states of these samples.
Analysis of the cell cycle distribution of exponentially growing cells by propidium iodide staining revealed that 21% of cells transfected with wild type spindlin1 entered the G 2 /M phase, whereas only 11% of HeLa cells and 10% of cells transfected with vectors were in G 2 /M, indicating that overexpression of The cross-linker is ethylene glycol succinate, with a final concentration of 1 mM. Spindlin1 at a concentration of ϳ0.2, 0.4, 0.8, and 1.6 mg/ml was used for cross-linking, and the same amount of protein was loaded onto the SDS-PAGE gel. The spindlin1 dimer bands were clearly observed for all concentrations, whereas a small portion of protein is exhibited with even higher oligomer state as the increasing amount of protein.
spindlin1 would lead to a marked increase in the percentage of cells in G 2 /M (p ϭ 0.001). The results of fluorescence-activated cell sorter analysis also showed a slight increase of S stage cells when spindlin1 was overexpressed (34% of spindlin1 wild type, 26% of control cells, and 26% of cells transfected with vector, with a Student's t test difference of p ϭ 0.05). These results were in agreement with previous studies (10,11). Notably, cells expressing the spindlin1 T95A mutant showed a significant decrease in the percentage of cells in G 2 /M (9%) and S (29%) phase compared with those expressing wild type spindlin1 and were nearly identical to those of the control cells (p values are 0.51 and 0.44, respectively) (Fig. 6, a and b).
Thus, our data suggest that Thr-95 is crucial for the function of spindlin1. From our structure the loss of the phosphate ions, which are maintained by hydrogen bonds, is expected to result in the distortion of the loop P. Thus, we propose that mutation of Thr-95 should break the interaction between the loop P and strands ␤7 and ␤8 of domain II, destroying its stability. Thus, the P loop should be related to the cell cycle regulation of spin-dlin1, although further work is required to confirm this hypothesis.
Spindlin1 Molecules Possess DNA Binding Activity-Prompted by the observation that some Tudor-like domains possess nucleic acid binding activity together with the potential cell cycle regulation, tumorigenesis, and anti-apoptosis functions of spindlin1, we performed gel shift assays to assess the binding of spin-dlin1 to nucleic acid.
In this assay, mixtures of different concentrations of protein and nucleic acids were analyzed on agarose gels. Our initial binding data showed that increasing the concentration of spindlin1 would lead to a shift of the band corresponding to 30 bp of dsDNA, which suggests that spindlin1 could interact with dsDNA. This interaction was not affected by altering the dsDNA probe, indicating that the association between spindlin1 and dsDNA is nonspecific. However, similar assays using 30 bp of single-stranded DNA (ssDNA) as a probe did not show an obvious band shift, implying that spindlin1 does not bind ssDNA (Fig. 6c). Furthermore, spindlin1 also exhibited higher affinity for super helical than for open cycle dsDNA, as the super helical bands shifted much earlier than the open cycle bands when mixing spindlin1 with dsDNA vectors (Fig. 6d). Because no metal ions are found in the structure and the addition of EDTA did not inhibit the protein/DNA interaction, metal ions are not likely to be required for nucleic acid binding activity (data not shown). Our data suggest the existence of protein-nucleic acid interactions, providing a basis for investigating DNA/spindlin1 interactions and the opportunity to explore the biological consequences of this interaction in further detail.

Functional Implications of Other Tudor-containing Proteins-
Although not shown in the previous bioinformatics analysis, the spin/ssty repeats and Tudor domains exhibit similar folds. Interestingly, like spin/ssty repeats, Tudor domains are usually presented in a tandem repeat manner. TUD, a protein with 11 Tudor domain repeats, is also specifically expressed during oogenesis and early embryogenesis (16). It is essential for the germ cell formation of embryos but without an obvious somatic function (17). All of these observations imply some functional relationship between spin/ssty repeat and Tudor domains.
Recent work has revealed that the Tudor domain might be a methylated protein binding domain. 53bp1, a conserved checkpoint gene for DNA double-stranded breaks (18) that contains two tandem Tudor domains, was found to bind directly to Lys-79-methylated histone H3 (19,20). Furthermore, the Tudor domain in survival motor neuron, a gene responsible for autosomal recessive proximal spinal muscular atrophy, is required for its methylated partner binding activity (21,22). All of these domains display a similar five anti-parallel ␤-strand structure with spin/ssty repeats (Fig. 3a), and thus, spindlin, the gene family that contains three spin/ssty repeats, might also possess a similar binding activity, although further evidence is required to confirm this.
Some other Tudor-like domains are also involved in nucleic acid binding. Research on Tudor domains in 53bp1 showed that the residues between loop ␤1-2 and loop ␤3-4 are necessary for dsDNA binding, whereas the residues Trp-18, Asn-21, Tyr-23, Asp-43, and Glu-47 should be responsible for nucleic acid binding (19). Although sequence alignment showed that these residues are largely conserved in spin/ssty repeat as well, their spatial positions are not necessarily the same (Fig. 3, b and d).
Trp-37, Asn-41, and Asp-70 in spin/ssty repeat adopt similar positions in the structure, whereas Asp-67 is located in a different strand, and the side chain of Val-44 extends in an opposite direction compared with Tyr-23 of 53bp1. In another study, Steiner and co-workers (23) implied that the nucleic acid binding surface of Tudor-like domain in NusG is composed by loop ␤1-2 and loop ␤3-4 and portions of ␤1, ␤2, and ␤4. However, superposition of these structures shows that spindlin1 domains are greatly diverse in these loop regions, particularly in the ␤1-2 loop (Fig. 3, a and c). This loop is much longer in spin/ssty repeats than in other Tudor domains, leading to a greater distance between loop ␤1-2 and loop ␤3-4. It is not yet clear whether or not the spatial variations would give rise to different binding modes. Thus, whether spindlin1 interacts with nucleic acid in a similar manner to other Tudor domains awaits further resolution.
The Oligomer State Might Be Important for the Function of Spindlin1-Spindlin1 is a dimeric protein from gel filtration chromatography and dynamic light scattering experiments, and we observe in the crystal structure an interface between two spindlin1 molecules that buries 620 Å 2 from solvent. Based on our structure, the dimerization occurs by association of the surface A (strand ␤11, ␤14, loop ␤10 -␤11) and surface B (strand ␤2, loops ␤1-␤2, ␤3-␤4, ␤13-␤14), which results in the formation of two equivalent surfaces that are not involved in dimer formation. These two surfaces may, therefore, associate with other tandems in a strictly identical way, as occurs in the crystal. Based on this, we speculate that spindlin1 has the potential to form higher order oligomers. Chemical cross-linking studies show that spindlin1 has a strong tendency to form a homodimer, yet a small portion of trimer can also be observed (Fig. 4c). Thus, the two vacant sites are likely to be the binding sites for other spindlin1 molecules.
Similar dimeric architectures are also found in other structures. The homodimer of the phosphorylation domain in PhoP, which is maintained mainly by hydrogen bonds of the acidic residues, involves an asymmetric association similar to the spindlin1 homodimer. Birck and co-workers (24,25) also believe that the two free interacting surfaces on the tandem repeats appear as an unprecedented unit of further oligomerization, which should be important for its function. Spindlin1 is highly expressed at both the mRNA and protein level. Of all the maternal mRNA, about 0.35% of transcripts encode the Spin gene (1), whereas its protein product is repeatedly found in oocytes and early embryos and has been used as a marker protein for the cell cycle study (2,26,27). Based on the accumulation of spindlin1 to the cell nuclei and its potential for oligomerization, it is possible that spindlin1 is present in the oligomeric state under certain circumstances, which should be important for its function.
In summary, we have reported the three-dimensional structure of spindlin1, which is the first to be characterized from the spin/ssty gene family. The spindlin1 structure possesses a novel arrangement of the Tudor repeat domain fold. Thr-95 is a functionally important residue that plays roles in the binding of phosphate ions and stabilizing the loop between domain I and II and which is essential for regulating cell cycle progression. We have also identified double-stranded nucleic acid as its molecular binding partner, thus providing a structural and biochemical basis for further functional investigation of spindlin1.