Crystal Structure of a Schistosoma mansoni Septin Reveals the Phenomenon of Strand Slippage in Septins Dependent on the Nature of the Bound Nucleotide*

Background: Septins are filament-forming proteins involved in membrane-remodeling events. Results: Two crystal structures of a septin with the highest resolution to date reveal the phenomenon of β-strand slippage. Conclusion: A novel mechanistic framework for the influence of the nature of the bound nucleotide and the presence of Mg2+ in septins is proposed. Significance: Identification of strand slippage might contribute to elucidating the mechanism of septin association with membranes. Septins are filament-forming GTP-binding proteins involved in important cellular events, such as cytokinesis, barrier formation, and membrane remodeling. Here, we present two crystal structures of the GTPase domain of a Schistosoma mansoni septin (SmSEPT10), one bound to GDP and the other to GTP. The structures have been solved at an unprecedented resolution for septins (1.93 and 2.1 Å, respectively), which has allowed for unambiguous structural assignment of regions previously poorly defined. Consequently, we provide a reliable model for functional interpretation and a solid foundation for future structural studies. Upon comparing the two complexes, we observe for the first time the phenomenon of a strand slippage in septins. Such slippage generates a front-back communication mechanism between the G and NC interfaces. These data provide a novel mechanistic framework for the influence of nucleotide binding to the GTPase domain, opening new possibilities for the study of the dynamics of septin filaments.

that these results have provided useful insight into how a septin filament forms from its component monomers, they are limited as a consequence of all being mammalian in origin, and therefore little is known about the diversity of structures across different phyla.
Schistosoma mansoni is one of the major species responsible for the neglected tropical disease schistosomiasis, which affects over 230 million people in 77 countries worldwide (17). The genomic data recently published for this flatworm has opened up a series of novel opportunities for the study of unique features of its metabolism and evolution. Furthermore, it brings with it also the possibility of identifying new drug and/or vaccine targets (18). Recently, four different septins were identified and described in this organism (19), and these have been classified into three of the four existing subgroups (20 -22).
Here, we describe the structure of the GTPase domain of one of these S. mansoni septins (SmSEPT10) as determined by x-ray crystallography. As such, this is the first report of a structure for a non-mammalian septin. The high resolution achieved for SmSEPT10G (the GTPase domain of SmSEPT10) bound to GDP (1.93 Å) and to GTP (2.1 Å) allows for an unprecedented and detailed analysis of septin structure. This provides surprising insights into their dynamics, which are expected to be relevant to a fuller understanding of the ways in which septins interact with other cellular components, such as membranes.

EXPERIMENTAL PROCEDURES
Expression and Purification of Recombinant SmSEPT10 and SmSEPT10G-SmSEPT10 cDNA was amplified from RNA extracted from adult worms and cloned into the pET28a(ϩ) vector as described previously (19). SmSEPT10G was obtained using the previous construct as template to a new PCR with primers flanking the GTPase domain (residues 39 -306) of SmSEPT10. SmSEPT10G was also subcloned into the pET28a(ϩ) vector, which introduces a His tag at the N terminus of the protein.
All constructs were expressed in Escherichia coli Rosetta (DE3) strain. Bacterial cells were grown at 37°C in Luria-Bertani (LB) medium supplemented with kanamycin (50 g/ml) until they reached an OD of 0.6 -0.8. The cell suspension was cooled to 18°C for 20 min, and isopropyl ␤-D-1-thiogalactopyranoside was added to a final concentration of 0.4 mM. Protein expression was performed overnight at 18°C with continuous shaking.
The cells were harvested by centrifugation at 10,000 ϫ g at 4°C for 30 min and suspended in 50 mM Tris-HCl, pH 8.0, 800 mM NaCl, 5 mM 2-mercaptoethanol, and 10% glycerol (suspension buffer). After lysis by sonication (10 cycles of 25-s bursts followed by 35 s of rest), the lysed cells were centrifuged at 20,000 ϫ g at 4°C for 30 min. The supernatant containing the soluble proteins was loaded onto a column packed with Ni 2ϩnitrilotriacetic acid resin (Qiagen), pre-equilibrated with suspension buffer. The column was incubated for 30 min and washed with suspension buffer, followed by a wash step with standard buffer (50 mM Tris-HCl, pH 8.0, 300 mM NaCl, 5 mM 2-mercaptoethanol, and 10% glycerol) and a wash step with standard buffer supplemented with 10 mM imidazole. The proteins were eluted in standard buffer containing 0.5 M imidazole.
Further purification was carried out by size exclusion chromatography on a Superdex 200 column coupled to an ÄKTA purifier system (GE Healthcare), from which they were eluted in standard buffer. The integrity and purity of the proteins were assessed by SDS-PAGE.
GTP Hydrolysis Assay-The hydrolytic activity assay was performed with 15 M SmSEPT10 in standard buffer. The samples were incubated with a 3-fold excess of GTP (45 M) in the presence of 5 mM MgCl 2 . Aliquots were removed after different time intervals and flash-frozen in liquid nitrogen. Nucleotides were extracted from the protein samples according to the method described by Seckler et al. (23) with minor modifications. Ice-cold HClO 4 (final concentration 0.5 M) was added to the protein samples, and after a 10-min incubation period, the protein pellet was separated by centrifugation at 20,000 ϫ g at 4°C for 10 min. The supernatant was buffered and neutralized with ice-cooled solutions of KOH 3 M (one-sixth volume), K 2 HPO 4 1 M (one-sixth volume), and acetic acid (0.5 M) (final concentration). The nucleotides were analyzed by HPLC after a centrifugation step (20,000 ϫ g at 4°C for 10 min). The GTP was separated from GDP by anion exchange chromatography on a Protein Pack DEAE 5 PW 7.5 mm ϫ 7.5 cm column (Waters) driven by a Waters 2695 chromatography system. The column was equilibrated in 25 mM Tris at pH 8.0, and 200 l of each sample were loaded into the system and eluted with a linear NaCl gradient (0.1-0.45 M in 10 min) at a flow rate of 1 ml/min at room temperature. The absorbance was monitored at 253 nm. The retention times of each guanine nucleotide were determined using a mixture of 15 M GDP and 15 M GTP in the same sample buffer after treatment with HClO 4 .
Isothermal Titration Calorimetry (ITC)-The nucleotide binding affinity of SmSEPT10G was determined at 18°C using a VP-ITC 4 calorimeter (MicroCal). Measurements were performed in standard buffer in the presence or absence of 5 mM MgCl 2 . The sample cell was loaded with 15-25 M SmSEPT10G, and the guanine nucleotide (GTP or GDP at a concentration of 2-3 mM) was titrated in a series of 45 injections of 5 l each to achieve a complete binding isotherm. All ITC experiments were repeated at least twice, and the heat of dilution (obtained by titration of the nucleotide into the buffer) was subtracted from the binding curve. The dissociation constant (K d ) and enthalpy were obtained with the software Microcal ITC Origin TM . Curve fitting was performed assuming a one-binding site model, and the reaction stoichiometry was fixed (n ϭ 1) to enable proper curve fitting due to the low affinity observed (24). 31 P NMR Spectroscopy-1 mM SmSEPT10G in 50 mM Tris-HCl, pH 7.5, 300 mM NaCl, 7% glycerol, 6 mM DTT, 10% D 2 O, and 1 mM GDP was subjected to 31 P NMR analysis either in the absence of magnesium or in the presence of different MgCl 2 concentrations (1 and 20 mM of MgCl 2 ) to evaluate the binding of the Mg 2ϩ ion to the SmSEPT10G-GDP complex. The binding of Mg 2ϩ ion to the complex can be identified by a change in the chemical shift of the GDP ␤ phosphate. 31 P{ 1 H} NMR spectra were recorded in a Bruker Avance III HD-600 NMR spec-trometer operating at 242.94 MHz ( 31 P frequency). The measurements were performed at 290 K (17°C), using 5-mm NMR tubes, 30°flip angle, 1.0 s relaxation delay, 3.0 s acquisition time, and 3000 scans. The 31 P spectra were acquired using 262,144 data points and processed with an exponential line broadening function of 15 Hz. The spectra phase and baseline were corrected automatically. A 85% phosphoric acid 1-mm capillary tube was used as an external reference for 31 P chemical shifts.
Crystallization, Data Collection, and Structure Determination-The hanging drop vapor diffusion method was used to obtain crystals of SmSEPT10G. Drops of 2 l of the purified protein in standard buffer (3.5 mg/ml protein in the presence of 2 mM GDP or GTP) were mixed with 2 l of the reservoir solution (0.2 M sodium acetate, 25% PEG 3350) at 20°C. After 24 h, the crystals were briefly transferred to the cryoprotective solution and flash-frozen in liquid nitrogen. The x-ray diffraction data were collected at the Diamond Light Source using the beamlines I24 (SmSEP10-GDP) and I04-1 (SmSEP10-GTP). The data were collected up to 1.93 and 2.14 Å for the GDP and GTP complexes, respectively. These resolutions are the highest obtained for any septin reported to date. The data were indexed, integrated, and scaled using the package Xia2. The structure of the S. mansoni Septin10 GTPase domain in complex with GDP was solved by molecular replacement with the program Phaser (25), using the GTPase domain of human SEPT2 modified by the Chainsaw program (26) as the search model (Protein Data Bank entry 2QNR). The two proteins share 64% sequence identity.
Two molecules related by non-crystallographic symmetry were found in the asymmetric unit consistent with the Matthews coefficient. Structure refinement was carried out using Phenix (27) and Coot for model building (28), using a -weighted 2F o Ϫ F c and F o Ϫ F c electron density maps. The GDP ligands were automatically placed using the Find Ligand routine of Coot, and water molecules were located using a combination of COOT and Phenix. The complex with GTP was also solved employing Phaser using the coordinates of the previously refined GDP complex. The behavior of R and R free were used as the principal criteria for validating the refinement protocol, and the stereochemical quality of the model was evaluated with Procheck (29) and Molprobity (30). The data collection and processing parameters can be visualized in Table 1.

RESULTS AND DISCUSSION
The GTPase Domain of SmSEPT10 Is Able to Bind both GTP and GDP but Does Not Display Hydrolytic Activity-Recombinant full-length SmSEPT10 and a construct named SmSEPT10G (comprising the 268 amino acids of the GTPase domain of SmSEPT10, residues 39 -306) were successfully expressed in E. coli. After cell lysis and removal of insoluble material, strong protein bands of the expected molecular masses were detected in the soluble fraction for both samples on SDS-PAGE (data not shown). These fractions were submitted to nickel affinity chromatography followed by size exclusion chromatography for purification of the recombinant proteins.  MARCH 14, 2014 • VOLUME 289 • NUMBER 11

JOURNAL OF BIOLOGICAL CHEMISTRY 7801
GTPase activity assays were performed by incubating GTP with SmSEPT10 for different times and evaluating the GTP and GDP content of the sample by HPLC (Fig. 1). It is possible to note that even after 24 h of incubation with GTP, there is no noticeable production of GDP, indicating that the protein displays no or very low levels of catalytic activity. Such a result was not unexpected because it has been previously reported that members of the SEPT6 subgroup are catalytically less active than the remaining three subgroups (13,31), and the closest mammalian homologue to SmSEPT10 is SEPT10 itself, which belongs to this subgroup.
In order to characterize the binding of GTP and GDP, ITC experiments were performed with the SmSEPT10G construct, because it displayed a greater stability at the high concentrations necessary to perform ITC measurements. The purified SmSEPT10G had no detectable nucleotide carryover as assessed by anion exchange chromatography (data not shown), enabling the direct use of the purified protein in the ITC measurements without the need to displace any bound nucleotide. The analysis of the binding isotherms obtained for SmSEPT10G titrated with GTP ( Fig. 2A) and GDP ( Fig. 2B) reveals that the reactions were exothermic, and the dissociation constants for both nucleotides were determined ( Table 2). The affinity for both nucleotides was much lower than that determined for human SEPT2 and SEPT3, which display K d values in the micromolar range (16,32). The influence of MgCl 2 on the binding of GTP and GDP was also eval-  uated. The presence of MgCl 2 was essential for the interaction with GTP, and no binding was detected in the absence of the metal ion (data not shown). On the other hand, the binding of GDP was independent of the presence of MgCl 2 .
Considering that the Mg 2ϩ ion is not essential to the formation of the SmSEPT10G-GDP complex, we performed 31 P NMR spectroscopy to infer the relevance of Mg 2ϩ binding to this complex at physiological magnesium concentrations. Fig. 3 displays the obtained 31 P NMR spectra of SmSEPT10G-GDP in the absence and presence of different Mg 2ϩ concentrations: resonances corresponding to the ␣and ␤-phosphates of free GDP (the sharp lines at Ϫ9.87 and Ϫ5.88 ppm) and GDP bound to the protein (broad lines at Ϫ8.97 and 2.82 ppm), respectively (Fig. 3A). The sharp lines indicate that GDP tumbles freely in solution, and the broad lines show reduced mobility of GDP molecules when bound to protein. Fig. 3, B and C, shows the spectra of free and bound GDP but in the presence of 1 mM (B) and 20 mM (C) Mg 2ϩ . Comparing the spectra using Mg 2ϩ at saturating concentrations (Fig. 3C) and in the absence of the ion (Fig. 3A), it is possible to note a marked change in the chemical shift of the GDP ␤-phosphate bound to the protein. At 1 mM Mg 2ϩ , which represents the high end of physiological free Mg 2ϩ concentrations (33), the chemical shift of the GDP ␤-phosphate bound to the protein is similar to that observed in the absence of Mg 2ϩ (Fig. 3B). This suggests that within the cellular context, the SmSEPT10G-GDP complex should be predominantly in the unbound state. Consequently, during the crystallization assays, MgCl 2 was added to the purified protein only when GTP was present, and the structure obtained for the GDP-bound complex did not contain Mg 2ϩ .
Structure Solution and Refinement-Considering that much previous structural work on human septins has also been performed on the isolated GTPase domains (14,15,34), the structures described here for SmSEPT10G allow for direct comparison. Despite the fact that it is expected that SmSEPT10 would be bound to GTP when present within heterofilaments (as is the case for SEPT6 within the SEPT2/6/7 complex), we have been successful in solving crystal structures for both the GTP-and GDP-bound forms. Both complexes crystallized in space group C2 with two monomers in the asymmetric unit and were successfully solved by molecular replacement. The structures have been refined to 1.93 Å (GDP) and 2.1 Å (GTP), respectively, the highest resolutions reported to date for any septin structure, yielding final R/R Free values of 18.51/21.70% for SmSEPT10G-GDP and 18.68/22.12% for SmSEPT10G-GTP. The structures reported for SmSEPT10G (residues 39 -306 of the full SmSEPT10 sequence) consist of 3,965 protein atoms, 129 water molecules, and two molecules of GDP in the case of the GDP complex and 4,019 protein atoms, 193 water molecules, two molecules of GTP, and two magnesium ions in the case of the GTP complex. However, in both cases, some residues presented no interpretable electron density and are absent from the final models. In the GDP complex, these correspond to residues 39, 69 -78, 103-110, and 246 -248 from subunit A and residues 39, 71-78, and 103-108 from subunit B, whereas in the GTP complex they correspond to residues 39, 69 -77, 107-110, and 245-248 from subunit A and 39, 70 -77, and 108 from subunit B. Despite the difference in the ␤3 strand register that will be discussed below, the monomer structures are very similar in both complexes, showing a root mean square deviation that varies between 0.58 and 0.63 Å (upon automatically overlaying C ␣ s). These values compare with 0.62 Å when the dimers are simultaneously superposed, suggesting the relative orientation of the two subunits to be effectively identical in both structures, consistent with the crystals being isomorphous. When compared with other septin monomers, the mean RMS deviation is 1.4 Å.
Overall Description of the Structures-SmSEPT10G has the typical septin fold, based on that of small GTPases (Fig. 4). In addition to the central six-stranded ␤-sheet (␤1-␤6), septins possess a C-terminal extension known as the septin unique element, which corresponds to ␤-strands 9, 10, 7, and 8 (in that order) and ␣-helices 5 and 6 (13). Helix 6 possesses a characteristic ␣-aneurism (35,36), which is conserved in all known structures (residues 290 -294). ␤2 runs antiparallel to the remainder of the sheet and is poorly defined in all previously described structures. Here we observe clear density for the entire strand in both the GDP-and GTP-bound forms due to the significantly higher resolution of the present structures. The helix ␣5Ј  has been seen in two rather different orientations in previous structures, and here it is observed to be in the more common of the two, as observed in SEPT2 (13) and SEPT7 (14,15), but different from that seen in SEPT3 (16). The septin unique element contributes to two interfaces, known as G and NC, leading to the formation of a filament within the crystal lattice. In the present structures, this filament is essentially identical in both the GDP-and GTP-bound forms and also identical to that observed previously for the only known crystal structure of a septin heterocomplex, that composed of septins 2, 6, and 7 (13). Furthermore, similar filaments are also observed when individual septin GTPase domains are crystallized separately (13)(14)(15)(16). This is the case for SEPT3, SEPT7, and SEPT2 (when bound to GDP). The only exception to this observation described to date is that of SEPT2 bound to a GTP analog, GppNHp (12), and this will be discussed below. The filament observed in the present structures is neither foreshortened nor expanded with respect to the heterocomplex, unlike those observed for SEPT3, SEPT2, and SEPT7. In the structures of both complexes, the nucleotide is found bound to its canonical binding site at the G interface. Fig. 5 shows the specific interactions made in the B subunit of the GTP complex. The higher resolution of the present structures allows us to comment on details of the binding site that have not been described previously. The P-loop (residues 48 -55) (37) provides the most important interactions stabilizing the nucleotide phosphates. In septins, the loop extends to include the side chain of Thr 56 , which coordinates the ␣-phosphate of the nucleotide in both the GDP-and GTP-bound complexes. Lys 54 lies between the ␤and ␥-phosphates in the GTP complex whereas Thr 50 interacts solely with the latter via water Wat 60 (w60 in Fig. 5), itself hydrogen-bonded to the main chain amide of the invariant Gly 103 from the DXXG motif (G3 motif) of the switch II region. Interestingly, the aspartic acid of this motif is in fact a glutamic acid in SmSEPT10, and this position is generally not well conserved in the SEPT6 subgroup to which SmSEPT10 belongs. Furthermore, Gly 103 does not interact directly via its backbone amide with the ␥-phosphate in the classical arrangement for small GTPases but does so via a water molecule (Wat 60 ).
Thr 50 is unique to the SEPT6 subgroup of septins, being a serine in the remaining subgroups. The Mg 2ϩ ion, which is only present in the GTP complex, is directly bound by Ser 55 from the P-loop, two oxygens from the ␤and ␥-phosphates, two water molecules, and Glu 100 from the C-terminal end of ␤3, which replaces the Asp typically found in the DXXG motif. In the GDP complex, due to the lack of Mg 2ϩ , this glutamic acid is released into the switch II loop region, as described below. Similarly, the water molecule Wat 60 is lost due to both the absence of the ␥-phosphate and a concomitant change in conformation to Gly 103 (which no longer shows significant electron density).
Interestingly, switch I does not participate in Mg 2ϩ binding. This is different from what is observed in the SEPT2-GppNHp and SEPT3-GDP complexes in which switch I is directly involved via the side chain of Thr 78 or its homologue, Thr 102 , respectively. The SEPT6 subgroup is unique in lacking this oth- The nomenclature for the secondary structure elements follows that adopted by Sirajuddin et al. (12) and is shown on the right. erwise conserved threonine and in having a five-residue deletion prior to it. As a consequence, SmSEPT10G does not follow the otherwise universal "loaded spring" mechanism (38), which involves the participation of switch I not only in Mg 2ϩ binding but also in anchoring the ␥-phosphate directly via the threonine backbone amide (Fig. 6). Indeed, the homologue to the switch I threonine is disordered in the GTP complex, and the resulting coordination of the Mg 2ϩ ion is significantly different from that described for other septins (Fig. 7). Not only is one of the protein ligands different (Glu for Thr), but its coordination position is also different, due to an exchange with a water molecule.
Lack of Catalytic Activity-The minimal hydrolytic activity of the SEPT6 subgroup is well known and has been further demonstrated here for SmSEPT10. In addressing the structural basis for this lack of activity, it has been speculated (12), based on the structure of SEPT2, that Thr 78 from switch I may be essential for catalysis. Here we provide some direct support for this proposal. This threonine normally performs three important functions. It directly coordinates the Mg 2ϩ ion, it donates a hydrogen bond to the ␥-phosphate, and it secures a water molecule via a main chain carbonyl, ready for in-line nucleophilic attack during catalysis. The absence of this threonine in SmSEPT10 means that this water is not observed in the structure we report here, making catalysis impossible. Either of the competing proposals for the catalytic mechanism, be it associative (39) or dissociative (40), requires the presence of such a water molecule. Therefore, our structure strongly supports the notion that it is the differences in the switch I region (and particularly the absence of the threonine) that are responsible for the lack of activity observed for this subgroup (Fig. 6). However, it is important to add that the structure reported here does not represent the transition state for catalysis, and therefore a degree of caution should be applied when trying to provide a  MARCH 14, 2014 • VOLUME 289 • NUMBER 11 mechanistic explanation for the experimentally observed lack of catalytic activity.

Strand Slippage in a Schistosome Septin
Water Structure and the G Interface-Several water molecules appear to be important for binding both GDP and GTP. Two of these interact directly with the N7 and N3 positions of the base, and a further one interacts with the O3Ј of the ribose. The water that is anchored to N7 also interacts with Gly 53 of the P-loop and with a second water that connects to Ile 52 and Lys 184 (of the G4 GTPase signature sequence, AKXD). The guanine base lies sandwiched between this lysine and Arg 253 . The latter is a septin-specific residue (20) coming from ␤7 and shows some degree of disorder in the electron density maps. It forms a salt bridge with Glu 192 from the neighboring subunit (also septin-specific), suggesting this interaction to be an important component of the G interface. Tyr 255 from the loop between ␤7 and ␤8 slots into the G interface interacting with Arg 253 in a manner identical to that seen previously for other septin structures. This residue has recently been shown to be essential for the formation of an intact G interface (16). Finally, Gly 238 donates a hydrogen bond to O6, and Asp 186 (from the AKXD G4 signature sequence) secures N1 and N2 of the guanine base as normally observed in small GTPases.
Besides the salt bridge with Arg 253 described above, Glu 192 is also normally observed interacting with the ribose base either directly or via a water molecule. Furthermore, N2 and N3 of the base also form cross-interface hydrogen bonds with the main chain oxygen of Thr 187 , the former directly and the latter via a water molecule. His 159 from the neighboring subunit may interact with the nucleotide phosphates, but the density is poorly defined in both subunits of both of the structures reported here, indicating, at best, a weak interaction. In the only other structure of a septin-GTP (or GTP analog) complex, that of SEPT2 with GppNHp (3FTQ), this histidine (His 158 in SEPT2) is reported to form a salt bridge with Asp 106 (Asp 107 in SEPT2, part of the switch II region). This is not observed in either of our structures or in that of SEPT3 and may be SEPT2-specific. Indeed, different from the SEPT2-GppNHp complex, the switch II region is not completely structured in either of the SmSEPT10 structures described here. When ordered, it also takes a course different from that observed in either SEPT3 (16) or SEPT7 (14,15). However, it should be pointed out that G interfaces observed in homofilaments encountered in crystal structures are likely to be promiscuous rather than physiological. Nevertheless, the significantly higher resolution of the structures described here makes them useful models for better understanding such interfaces.
Switch Regions-There are few differences between the GDPand GTP-bound complexes in the switch I region, which is largely disordered in both cases. This might be unexpected were it not for the fact that the SEPT6 group is catalytically less active, as mentioned above. The only significant difference is that Lys 81 in the GDP complex interacts with Ser 55 , which is a ligand to the Mg 2ϩ in the GTP complex. In the case of switch II, neither structure is completely ordered (in the GDP complex, FIGURE 6. SmSEPT10 lacks catalytic activity. Both p21 H-ras (A) and SEPT2 (B) display the classical "loaded spring" mechanism by which main chain amide groups from both switch regions (SW1 and SW2) form direct hydrogen bonds to the ␥-phosphate. In the case of switch 1, this is provided by a threonine residue that is not conserved in SmSEPT10. This threonine also secures a water molecule (red sphere in A and B), which is poised for in-line attack during catalysis. In SmSEPT10 (C), no direct hydrogen bonds are formed with the ␥-phosphate by residues from the switch regions. SW2 interacts via a water molecule (small red sphere in C), and SW1 is completely absent due to the lack of the threonine. As a consequence, there is no water adequately poised for catalysis. In SmSEP10, a standard 2F o Ϫ F c electron density map for this region is also shown. density is lacking for residues 103-110 (subunit A) and 103-108 (subunit B), whereas in the case of the GTP complex, this applies to residues 107-110 (subunit A) and residue 108 alone (subunit B). Furthermore, there is no interaction between neighboring switches II across the G interface as seen in SEPT7 and SEPT3. As mentioned above, in the GTP complex, switch II is unstructured at least to some extent in both monomers and is two residues shorter due to strand slippage (described below). Consequently, the ordered parts of switch II and its associated water structure show significantly different conformations in the two complexes.
Strand Register-For the first time, we are able to describe with confidence the register of residues within certain parts of the septin structure that until now have been poorly defined due to the limits of resolution. This applies particularly to ␤2, which lies at the edge of the main ␤-sheet and is the only strand to run antiparallel to the remainder. The register of this strand has been interpreted differently in the several septin crystal structures reported to date. Although this could potentially be explained by alternative crystal packing, it seems far more likely to be due to the ambiguities inherent in the interpretation of poorly defined electron density.
The most striking difference between the GDP-and GTPbound forms of SmSEPT10 affects strand ␤3, which lies between ␤2 on one side and the parallel strand, ␤1, on the other. Upon comparing the two complexes, it becomes clear that the ␤3 strand is shifted with respect to both of its neighbors by exactly two residues (ϳ6 Å) (Fig. 8). The consequence is to increase the size of the ␤2/␤3 hairpin loop in the GTP complex and simultaneously reduce that of the loop following ␤3, which leads into switch II. A total of 18 hydrogen bonds on both sides of ␤3 in the GTP complex are replaced by 19 with a shifted register in the GDP complex (Fig. 9). As a consequence, ␤2 and ␤3 are extended toward the NC interface in the presence of GTP. On the other hand, a rearrangement of switch II at the G interface means that ␤3 forms additional hydrogen bonds with ␤1 in the GDP complex. Despite not having been commented on by the authors, a similar phenomenon appears to occur in human SEPT2 (12). Upon comparing the crystal structures of the GDP-bound form (2QA5 or 2QNR) with that with the GTP analog GppNHp (3FTQ), the ␤3 strand is also observed to be shifted by two residues. This shift is clearly evident with respect to ␤1, but the poor definition of the density of the ␤2 strand due to limited resolution made it impossible to exactly define the correct relative register. This is emphasized by the different interpretations present in Protein Data Bank files 2QA5 and 2QNR. Here, we are able to unambiguously affirm that it is the ␤3 strand that alone is shifted with respect to both ␤1 and ␤2 and therefore with respect to the rest of the sheet.
This two-residue slippage appears to be facilitated by a Lys-Leu-Lys-Leu repeat within the ␤3 strand sequence (Fig. 8). Similar observations have been made for other systems that use strand slippage as an activation mechanism, as is the case for factor VIIa and Arf (41,42). Furthermore, similar phenomena have also been reported in T-cell receptor ␣ subunits (43), amyloidogenic transthyretin mutants (44), and the ␤-amyloid peptide at different pH values (45). In the latter, all residues are shifted by two to the right bring Glu 100 into a position for coordinating the Mg 2ϩ ion. Slippage appears to be facilitated by the existence of a Lys-Leu-Lys-Leu repeat and the presence of many small ␤-branched amino acids.
In SmSEPT10G, the register shift results in Lys 93 and Leu 94 in the GDP-bound form taking up the positions occupied by Lys 95 and Leu 96 in the GTP complex. As such, either Lys 95 or Lys 93 aligns with Asp 86 of the neighboring ␤2 strand. Furthermore, the presence of a Phe-X-Phe sequence (residues 40 -42) at the beginning of the ␤1 strand appears to be related to the same phenomenon because it allows a Leu 96 -Phe 42 hydrophobic contact in the GTP complex to be replaced by an analogous Leu 94 -Phe 42 contact for the alternative register of the ␤-strand. This can be more clearly appreciated from the hydrogen bonding map of the two states shown in Fig. 9.
One critical structural feature related to strand slippage appears to be the presence of the magnesium ion in the GTP complex. The metal is directly coordinated by Glu 100 from the sliding ␤3 strand. In the GDP complex, which does not contain Mg 2ϩ , Glu 100 is shifted into the loop following ␤3, which leads into the switch II region. Thus, metal ion coordination must certainly be one of the factors that control the register of the ␤-sheet. This is consistent with the fact that the recently solved structure of human SEPT3 (16) in complex with GDP/Mg 2ϩ presents the ␤3 strand shifted in a way reminiscent of the GTP complex described here. In this case, Glu 100 is replaced by Asp 125 (3SOP), which coordinates the magnesium ion indirectly via a water molecule. It is also consistent with the structure of SEPT6 within the heterofilament formed by human septins 2, 6, and 7 (13). In this complex, SEPT6 is unique in being bound to GTP rather than GDP, and despite the low resolution of the structure, it was interpreted to have a strand register between ␤1 and ␤3 identical to that observed here for the SmSEPT10G-GTP complex (Fig. 10).
A second consequence of the slippage is the extension of strands ␤2 and ␤3 toward the NC interface in the GTP complex. This extension has been previously noted (although not the slippage) in the case of human SEPT2 (12) and used to explain why the structure in the presence of the GTP analog does not form filaments within the crystal. This was attributed to a ϳ20 o tilt of the ␤2/␤3 region of the sheet, which would lead to a clash with the ␣0 helix of the NC interface, thus preventing filament formation. However, as shown in Fig. 9C, the extension to the sheet strands in SmSEP10G (as well as SEPT3 and SEPT6 from FIGURE 9. ␤-strand slippage. Shown are HERA hydrogen bonding diagrams for part of the ␤-sheet composed of strands 1-3. The situation in the GDP complex (A) is different to that of the GTP complex (B), but the total number of main chain hydrogen bonds is almost identical in both cases. In C, the result of slippage is seen to be the communication between the G interface to the left and the NC interface to the right. Part of a filament of the GTP complex is shown (dark blue) with a single subunit from the GDP complex superimposed at the center (light blue). The sliding (␤3) strand is shown in yellow (GTP complex) and orange (GDP complex). FIGURE 10. The consequences of strand slippage. Part of the heterocomplex of SEPT2-SEPT6-SEPT7 is shown (red, dark blue, and yellow, respectively) with the SmSEPT10G structure, as observed when complexed to GTP (light blue), superimposed on SEPT6. The consequence of strand slippage is the extension of the ␤-hairpin connecting ␤2 to ␤3 and thereby the coverage of the N-terminal helix ␣0. This region corresponds to the polybasic region known to be important for membrane association in mammalian septins. The implication is that strand slippage may affect membrane binding in a nucleotide-dependent fashion.
the heterofilament) does not lead to sheet distortion and is not incompatible with filament formation, so much so that such filaments are indeed observed in both crystal structures reported here. Indeed, this is compatible with data on yeast septins suggesting that filament formation is independent of the nature of the nucleotide (46).
It seems more likely that the observed sheet distortion in SEPT2 is in fact the result of crystal packing because this region forms extensive ␤-sheet hydrogen bonds with the equivalent region from a 2-fold related molecule. It would seem that rather than sheet distortion preventing filament formation, it is nonfilamentous crystal packing that leads to sheet distortion. Overall, there seems to be no incompatibility between strand slippage/sheet extension and filament formation.
The consequence of strand slippage is to generate a frontback communication mechanism between the G and NC interfaces. Strand slippage as a protein activation mechanism is not entirely unprecedented but is rare. The example most relevant to the current discussion was seen in Arf proteins. Like septins, Arf proteins are small GTPases involved in a series of membrane-remodeling events associated with intracellular vesicle transport (47,48). Membrane association depends on the exposure of an N-terminal myristoyl group that is dependent on the presence of GTP rather than GDP. Unlike septins, however, Arf proteins do not form filaments but rather are active as monomers. The best known examples are Arf1 and Arf6, which are specifically involved in the assembly of coat complexes at the Golgi and plasma membrane, respectively. Their activity is controlled by GTP hydrolysis, which leads to ␤3 strand slippage analogous to that described here. However, in this case, the ␤2 strand accompanies ␤3, and both slide as a rigid body with respect to ␤1 (42, 49 -51) (Fig. 11).
What is most intriguing is that, like septins, Arf proteins have an N-terminal ␣-helix that precedes the ␤1 strand. In septins, this helix corresponds to the polybasic region, known to be involved in membrane association (11). The orientation of this helix is radically different in the Arf complexes with GDP and GTP, being folded back on the structure in the former and exposed in the latter. This conformational change controls membrane association and involves both the N-myristoyl group and hydrophobic and basic residues from the helix (52). Based on the crystal structure of the SEPT2/6/7 heterocomplex, where helix ␣0 is present (albeit poorly defined), it is possible to assert that the strand slippage in the GTP complex of SmSEPT10G would cause steric hindrance and force the helix into a different conformation, in a fashion analogous to that seen in Arf. The similarity in both structure and function of these two families is highly suggestive of a similar mechanism of action. We therefore speculate that strand slippage is an integral part of the mechanism by which septins associate and disassociate from membranes. This implies that the importance of GTP binding and hydrolysis is related to membrane association rather than filament formation. This is consistent with results described on SEPT4 by Zhang et al. (11), who demonstrated that the GDP complex had greater affinity for membranes containing phosphotidylinositol-4,5-bisphosphate than the corresponding GTP complex.
Conclusions-In mammalian septin complexes, the SEPT6 subgroup members occupy the middle position of the heterofilament, interacting with a SEPT2 subgroup member via a G interface and with SEPT7 via an NC interface. Assuming a similar arrangement for heterofilaments in schistosomes leads to the conclusion that both of the homotypic interfaces observed in our crystal structures should be considered promiscuous because they are not anticipated to exist physiologically. With the current report, there are now examples of crystal structures for all four individual subgroups of septins, revealing that the phenomenon of promiscuity in interfacial interactions is completely generic. Discovering the structural determinants responsible for the correct assembly of a heterofilament therefore remains an important challenge in current septin research (53).
SmSEPT10 belongs to the SEPT6 subgroup, whose members are expected to be bound to GTP rather than GDP when incor- FIGURE 11. Comparison between SmSEPT10 and Arf6. Septins may operate via a mechanism analogous to that seen in Arf proteins, which use ␤-strand slippage as a means to dislodge the N-terminal helix and so influence membrane association. A, superposition of Arf6 in its GDP complex (blue) and GTP␥S complex (yellow), showing the slippage of both strands ␤2 and ␤3 with respect to ␤1, in the direction of the short N-terminal helix. B, in SmSEPT10, a similar color scheme is adopted, but in this case, ␤3 slips with respect to both ␤1 and ␤2. porated into heterofilaments, due to their extremely low catalytic activity. The manner by which GTP and its associated Mg 2ϩ are bound to SmSEPT10G therefore represents the best description presented to date for the physiological state of this subgroup. On the other hand, the relevance of the GDP-bound form of SmSEPT10 is less certain. It should be considered a good high resolution model for the remaining septin subgroups. Despite this disclaimer, it seems highly likely that the strand slippage observed in SmSEPT10G and described in detail here is indeed an important mechanistic aspect of septins in general because, although not described by the authors, it has also been observed in the case of the hydrolytically active SEPT2 (12). The mechanism is analogous to that previously described in another small GTPase, Arf, but is unique in the fact that such strand slippage promotes the breakdown of interactions from the two sides of the slipping strand.
The structures we describe here emphasize the uniqueness of the SEPT6 subgroup. Among other features, SmSEPT10 shows an unusual Mg 2ϩ coordination; a disordered switch I region even in the presence of GTP; an unusual primary structure within switch I itself (involving the lack of a normally conserved threonine that coordinates the magnesium); incomplete closure of the switch II region, which fails to interact directly with the ␥-phosphate and remains largely disordered even in the GTP complex; alterations to the P-loop (where Thr 50 , at the G interface, substitutes a serine conserved in all other subgroups); and the absence of a polybasic region prior to the GTP-binding domain. These unique features of the SEPT6 subgroup are probably interrelated, implying a specific role in filament assembly, function, or dynamics. The difference between the GDP-bound and GTP-bound forms of SmSEPT10G and, more specifically, the shift to the ␤3 strand is likely to be a more general phenomenon that applies to several or all of the septin subgroups. It seems likely that it is an important molecular switch controlling membrane association and dissociation and thereby membrane-remodeling events.