Mechanism of FtsZ assembly dynamics revealed by filament structures in different nucleotide states

Treadmilling protein filaments perform essential cellular functions by growing from one end while shrinking from the other, driven by nucleotide hydrolysis. Bacterial cell division relies on the primitive tubulin homolog FtsZ, a target for antibiotic discovery that assembles into single treadmilling filaments that hydrolyse GTP at an active site formed upon subunit association. We determined high-resolution filament structures of FtsZ from the pathogen Staphylococcus aureus in complex with different nucleotide analogues and cations, including mimetics of the ground and transition states of catalysis. Together with mutational and biochemical analyses, our structures reveal interactions made by the GTP γ-phosphate and Mg2+ at the subunit interface, a K+ ion stabilizing loop T7 for co-catalysis, new roles of key residues at the active site and a nearby crosstalk area, and rearrangements of a dynamic water shell bridging adjacent subunits upon GTP hydrolysis. We propose a mechanistic model that integrates nucleotide hydrolysis signalling with assembly-associated conformational changes and filament treadmilling. Equivalent assembly mechanisms may apply to more complex tubulin and actin cytomotive filaments that share analogous features with FtsZ.


32
FtsZ is an assembling GTPase that plays a key role during bacterial cell division [1]. This 33 widely-conserved protein polymerizes in the presence of GTP and metal cations into polar 34 filaments that gather at the centre of dividing cells to form a dynamic Z-ring [2]. FtsZ filaments 35 associate with a variable set of partner proteins into the divisome, which coordinates membrane 36 invagination and cell wall peptidoglycan synthesis during cytokinesis [3,4]. As such, bacterial 37 cell division proteins are targets for discovering new antibiotics [5]. 38 The head-to-tail treadmilling dynamics of FtsZ filaments, observed both in vitro [6,7] and in 39 vivo [8-10], depends on two major properties of this protein (see scheme in Fig 1A). First, GTP 40 hydrolysis only occurs within the filament, as the complete catalytic site is formed at the 41 interface between adjacent FtsZ monomers [11,12]. Second, assembly of the FtsZ filament 42 relies on the existence of two conformations that are independent of the bound nucleotide 43 [13,14]. The relaxed (R) conformation present in free monomers exhibits low affinity for the 44 filament, while the tense (T) conformation is acquired by filament subunits. Switching between 45 these two conformations explains cooperative assembly of single stranded filaments [15,16] 46 and creates different pairs of encountering interfaces at each filament end, thus enabling kinetic 47 polarity for treadmilling [14,17]. GTP-bound FtsZ monomers associate to the filament growing 48 end and, once inside the filament, are retained in the T conformation due to simultaneous 49 contact with subunits on both of its sides, even after GTP hydrolysis. The subunit at the 50 shrinking end dissociates due to weakened contact with its only neighbouring GDP-bound 51 subunit ( Fig 1A). FtsZ filaments shrink from the exposed nucleotide end, as deduced from 52 mutational studies [17,18]. A recent numerical model generates populations of treadmilling 53 filaments that accurately reproduce experimental results, thus unifying the described FtsZ 54 properties [19]. 55 56 that promote GTP hydrolysis of the nucleotide placed in the adjacent subunit within the filament 83 [11,12]. While loop T7 exhibits remarkable plasticity in FtsZ structures in the R conformation 84 [13,14,[22][23][24], it adopts a defined arrangement by coordinating a metal ion in the T 85 conformation [25]. Concomitant rotation of the neighbouring GAD enables the formation of 86 straight filaments where the nucleotide is buried from the solvent. The antibacterial compound 87 PC190723 and chemically-related FtsZ inhibitors bind in a cleft between NBD and GAD of 88 subunits in the T conformation, thus blocking filament disassembly [25][26][27]. A reduced 89 interaction area in the R conformation leads to monomers [14] or pseudofilaments where the 90 nucleotide is partly exposed to the solvent [13,23,24]. Crystal structures of FtsZ filaments in 91 the T conformation are only available in the presence of GDP, GTP with a truncated non-92 catalytic T7 loop, or the non-hydrolysable GTP analogue GTPgS [13,25,28,29]. Therefore, 93 mechanistic information on GTP hydrolysis and its implication in filament assembly dynamics 94 is currently largely missing. 95 We report a dozen filament structures of the FtsZ core from the pathogen S. aureus, hereafter 96 SaFtsZ, in complex with various GTP mimetics and metal ions. The high resolution of these 97 structures enables description of key interactions at the interface of adjacent monomers forming 98 the catalytic site with unprecedented detail. Together with structure-based mutational analysis 99 of critical residues around the active site, our results shed light on the mechanisms of FtsZ 100 filament assembly, GTP hydrolysis and treadmilling. 101 103 We analysed the effect of several GTP mimetics on SaFtsZ (residues 12-316) assembly, 104 monitoring polymer formation with light scattering and sedimentation assays (Fig 1B and 1C). 105

Stabilization of FtsZ polymers by GTP mimetics and cations
SaFtsZ assembly required GTP plus MgCl2 and polymers disassembled upon nucleotide 106 exhaustion. BeF3is a chemical mimetic of phosphate that acts on G-proteins [30,31] and 107 tubulin [32,33] as a reversibly-binding analogue of the GTP g-phosphate. We found that 108 addition of BeF3stabilizes SaFtsZ polymers against disassembly, suggesting that BeF3 -109 replaces the g-phosphate in SaFtsZ filaments following GTP hydrolysis. Fast depolymerization 110 can be induced by Mg 2+ chelation with EDTA. BeF3with GDP and Mg 2+ was ineffective for 111 polymer assembly. Negative-staining electron microscopy showed that similar bundles of 112 SaFtsZ filaments ( Fig 1D) form with GTP, GTP plus BeF3and GMPCPP, a slowly-hydrolysing 113 analogue of GTP that induces robust SaFtsZ assembly [29]. The non-hydrolysable GTP 114 analogues GMPPCP and GTPgS fail to induce SaFtsZ assembly in these experiments, similarly 115 to GDP and its analogue GMPCP, highlighting their inability to mimic GTP function on SaFtsZ 116 ( Fig 1C and 1D). We confirmed that BeF3stabilizes the assembly of full-length SaFtsZ in a 117 qualitatively similar manner to the SaFtsZ core, as well as that of distant homologs from 118

Structure determination of FtsZ filaments with GTP mimetics 130
Our attempts to determine the structure of SaFtsZ with bound GTP in the T conformation 131 employing X-ray crystallography yielded filament structures where density in the active site 132 only accounted for GDP, like in reported structures [25,29]. Therefore, we solved twelve crystal 133 structures of SaFtsZ in complex with different GTP mimetics, analogues and metal ions, at 134 resolutions ranging from 1.4 to 1.9 Å (S1 Table), enabling unambiguous assignment of ligand 135 densities (S3 Fig). We complemented this study with four additional structures of single-residue 136 mutants complexed to GDP (S1 Table). In all cases, SaFtsZ adopts the T conformation forming 137 straight filaments with minor but relevant differences between them, as detailed below. 138 Subsequent description assumes a filament orientation where nucleotide-binding loops T1-T6 139 in the NBD face upwards, while the T7 synergy loop in the GAD looks downwards (Fig 2A). The filament structure of SaFtsZ complexed to GDP and BeF3in the presence of Mg 2+ reveals 156 that BeF3adopts a tetrahedral configuration with beryllium lying 2.7 Å apart from GDP (Fig  157   3). Two fluorine atoms contact the bottom FtsZ monomer through hydrogen bonds (H-bond) 158 with residues A71 and A73 in loop T3, G108 in loop T4, and T109 in helix H4 (Fig 3A). The 159 third fluorine atom coordinates Mg 2+ , together with the β-phosphate of GDP. In a canonical 160 pre-catalytic state bound to GTP, the positive charge of Mg 2+ is expected to deprive in electrons 161 the link between the β-and γ-phosphates in GTP, thus providing the main activating charge.   Table). This solvent shell is altered in a 188 structure lacking Mg 2+ , where bridging waters B2-B4 shift away from R143 by ~1 Å and BeF3 -189 tilts away from the pre-catalytic water by 6°, likely weakening BeF3binding (S4C and S4D 190  presence of Na + in the T7 loop alters the configuration of ion-coordinating residues, especially 206 V203. Notably, Q48 lies outside the Na + coordination sphere whereas water K2 lies within and 207 forms an H-bond with Q48. These changes, likely associated with increased filament stability, 208 correlate with a tenfold reduction in GTPase activity upon substitution of KCl by NaCl (S3 209   Table). 210 We also sought to solve the structure of SaFtsZ complexed to GMPCPP, as it induces filament 211 formation with Mg 2+ (Fig 1C and 1D). However, the resulting structure only showed density 212 for GMPCP, which binds SaFtsZ as GDP does shift towards R143 by 0.6 and 0.8 Å respectively, while water B3 is absent. Notably, the 235 catalytic water (Fig 4B, red star) approaches AlF4by 1.4 Å as compared to the BeF3structure, 236 which is accompanied by a 30º rotation of catalytic residue D210. A squared bipyramid is, thus, 237 formed between AlF4 -, the linking oxygen in the GDP β-phosphate and the catalytic water. In 238 addition, the geometry of K + coordination is rearranged on both sides of the subunit interface. 239 On one hand, the N208 sidechain contacts the Mg 2+ coordination sphere while keeping direct 240 contact with K + . On the other hand, the Q48 sidechain shifts 2.1 Å away from its K + 241 coordination position in the BeF3structure, while D46 moves 0.8 Å away from the Mg 2+ ion. 242 As a result, D46 loses an H-bond with the Mg 2+ coordination sphere, compensated by H-bond 243 formation with a new solvent molecule that also contacts top residue N208 (Fig 4C). position, while the neighbouring top GAD rotates by ~5º towards top helix H7 (S1 and S2 261 Movies). This movement is coordinated with opening of bottom monomer motifs that surround 262 the nucleotide, including helix H2, loop T3, loop T5, and the N-terminal end of helix H7. As a 263 result, the interaction between top strand S9 and bottom loop T5 (Region-B, Fig 2B) is 264 reinforced, while the contact between top loop T7 and bottom helix H2 (Region-D, Fig 2B) is 265 weakened (S7A Fig).  Table). Second, in contact Region-D 272  and D213N, all suppress SaFtsZ assembly as monitored by light scattering and polymer 288 sedimentation (Fig 5A and 5B), supporting a prominent role in filament formation. In 289 accordance, the crystal structures of mutants D46A and D210N exhibit altered configurations 290 at Region-D (Fig 5C). In the D46A mutant, bottom residue Q48 shifts 3 Å away from the K + 291 site, while the whole top T7 loop is distorted. In the D210N mutant, the top T7 loop lacks K + 292 and moves 2 Å away from the bottom monomer, while the bottom T3 loop is severely 293 rearranged. Interestingly, mutants Q48A and R143K supported assembly, forming polymers that are 311 stabilized against disassembly by GTP consumption as compared to the wild-type protein (Fig  312  5A and 5B). This correlates with reduced GTPase activity by about six-fold in both mutants 313 (S3 Table). The structure of the Q48A mutant shows that its K + coordination position is 314 occupied by water K2 (Fig 5C), indicating a role for this residue in securing K + within the T7 315 loop for efficient catalysis. In the structure of the R143K mutant, K143 occupies an equivalent 316 position to that of R143 in the wild-type protein but its positive charge lies 3 Å further from the 317 GDP a-phosphate ( Fig 5C). As the R143Q mutant is incompetent for polymerization and 318 catalysis, our results indicate that a positively-charged residue in this position is required for 319 GTP hydrolysis. The polymers formed by the Q48A and R143K mutants were mostly 320 filamentous ribbons (Fig 5D), rather than the bundles observed for the wild-type protein (Fig  321   1D). Because the employed concentration of 8 mM Mg 2+ slows polymer nucleation (notice lag 322 times in Fig 5A), the effects of the mutations on assembly were confirmed at 2 mM Mg 2+ (S8 323

329
Cytomotive proteins such as actins and tubulins self-assemble into nucleotide-hydrolyzing 330 dynamic filaments that are able to perform mechanical work by themselves and also serve as 331 rails for motor proteins [39]. Treadmilling of FtsZ filaments, driven by GTP hydrolysis, is 332 essential for correct cell division in bacteria, which lack homologs of cytoskeletal motor 333 proteins. In this work, we reveal the detailed mechanism of interfacial nucleotide hydrolysis by nucleophilic attack. Second, we found that bottom residue R143, which connects to D213 and 344 the nucleotide g-phosphate through bridging water B1, contributes to catalysis and that the 345 position of its positive charge is critical for efficient GTP hydrolysis. Accordingly, R143K and 346 not R143Q mutation allows assembly but presents reduced GTPase activity (Fig 5 and S3 347 Table). The equivalent residue in MjFtsZ, R169, was suggested to stabilize the transition state 348 [12]. The role of R143 is likely equivalent to that of the arginine finger in Ras-like GTPases, 349 while in these enzymes the finger residue contacts the nucleotide directly [40]. Third, we 350 showed that a two-cation mechanism involving Mg 2+ and K + operates in FtsZ for GTP 351 hydrolysis. While Mg 2+ contacts the band g-phosphates of GTP and has a direct role in 352 catalysis, K + locates in the top T7 loop and contacts the substrate indirectly through its 353 coordinating residue N208, which forms H-bonds with solvent molecules connected to Mg 2+ 354 and the g-phosphate. Accordingly, mutation N208L abolishes polymerization. Taken together, 355 SaFtsZ can be classified as a type II Mg 2+ /K + enzyme, as the monovalent cation has an allosteric 356 role in catalysis [41]. This differs from most type I GTPases, where K + assists catalysis through 357 direct contact with the nucleotide phosphates, thus providing the activating charge that in other 358 GTPases is supplied by the arginine finger [40]. Fourth, we found that locking of K + within the 359 top T7 loop by bottom residue Q48 plays a role in catalysis. In agreement, the Q48A mutant 360 allows assembly but exhibits reduced GTPase activity (Fig 5 and S3 Table). K + is bound weakly 361 and can be evicted from its binding site by subtle changes around the T7 loop, as observed in 362 the structures of mutant Q48A and wild-type SaFtsZ complexed to GMPPCP and Mg 2+ . 363 Moreover, substitution of K + by Na + strongly reduces the GTPase activity ( S5 Fig and S3 364 Table), similarly to FtsZ from other species, with some exceptions [42]. This K + preference for 365 hydrolysis correlates with estimated concentrations in the bacterial cytosol of 200 mM K + and 366 5 mM Na + [43]. 367 A dynamic water shell mediates assembly and catalysis 368 The high resolution of the structures reported here enables location of solvent molecules at the 369 subunit interface that contribute to filament assembly and catalysis (S2 Table). Water molecules 370 M1-M4, located between the nucleotide phosphates and the bottom T2 loop, coordinate Mg 2+ 371 binding, with water M4 playing a prominent role as it also bridges the nucleotide phosphates 372 with the residue N208 in the top T7 loop (Fig 3). Notably, waters M1-M4 rearrange to the structures complexed to BeF3or AlF4 -(S1 and S2 Movies), while additional changes arise 386 in the absence of the g-phosphate (S3, S4, S5 and S6 Movies). We distinguish two major 387 interacting areas in the straight filament interface. On one hand, a pivoting area for interface 388 opening, as observed in molecular dynamics simulations of SaFtsZ filaments [44], involves 389 interface contact Regions A and B (Fig 2B). While rearranged, this area is roughly maintained 390 in R conformation structures of SaFtsZ pseudofilaments [13,23,24]. On the other hand, we 391 define a crosstalk area around the g-phosphate site, which encompasses interface contact 392 Regions C and D (Fig 2B) and is specific of straight filaments in the T conformation. The  We propose a model for the FtsZ catalytic assembly cycle (Fig 6). Our description starts with 402 a filament where all subunits are in the T conformation and contain GTP. We reason that Mg 2+ 403 requirement for FtsZ polymerization is mainly related to its shielding effect over the acidic 404 charge of the triphosphate nucleotide, while it also contributes to accurate positioning of the g-405 phosphate for catalysis (Fig 3 and S4D Fig). Additionally, K + within the filament T7 loop is 406 secured through labile coordination with residue Q48 from the neighbouring subunit. In such 407 configuration, nucleophilic attack by the catalytic water over the g-phosphate occurs through 408 an transition state where key residues and solvent molecules around the crosstalk region are 409 rearranged. GTP hydrolysis proceeds at slow rate due to unique catalytic properties of FtsZ 410 described above, followed by fast Mg 2+ release through the exit pore, which likely rearranges 411 to also liberate inorganic phosphate. This generates the assembled GDP-bound filament in the 412 T conformation, where absence of key interfacial ionic interactions mediated by the g-413 phosphate and Mg 2+ destabilizes filament contacts, which is accompanied by minor structural 414 rearrangements. We speculate that this allows detachment around the crosstalk region while 415 contacts around the pivoting area are roughly maintained. This eventually allows structural 416 relaxation into the R conformation, likely accompanied by K + release from loop T7 as density 417 for ion is absent across FtsZ structures in the R conformation, altogether leading to interfacial 418 dissociation at the filament minus end. Free FtsZ monomers in the R conformation 419 spontaneously exchange-in GTP and Mg 2+ to enter a new assembly cycle at the filament plus 420 end. The nucleotide g-phosphate and Mg 2+ likely stabilize helix H2 and loop T2 in the NBD of 421 the R monomer such that its top surface is preorganised similar to the T conformation. However, 422 the bottom surface of this R monomer is substantially different from the T conformation, 423 including a highly-flexible T7 loop, whereas the exposed T7 loop in an all-T filament [15,16]  HCl pH 8.4-9.0, and 24-28% polyethylene glycol 5000 monomethyl ether (PEG5000; Aldrich, 500 Steinheim, Germany). 501 The structure of SaFtsZ complexed with GDP and BeF3was obtained from crystals of 14 502 mg/ml SaFtsZ-GDP (28% PEG 5000 MME, pH 8.8) soaked in 10 mM BeF3during 20 h. The 503 structure of SaFtsZ complexed to GDP, BeF3and Mg 2+ was obtained by SaFtsZ-GDP co-504 crystallization with 2 mM BeF3and 10 mM MgCl2 (25% PEG 5000, pH 8.6). The structure of 505 SaFtsZ complexed to GDP, BeF3and Mn 2+ was obtained from SaFtsZ-GDP crystals (25% PEG 506 5000, pH 8.8) soaked in 10 mM BeF3and 20 mM MnCl2 during 20 h. The structure of SaFtsZ 507 complexed to GDP and Na + was obtained using SaFtsZ-GDP (15 mg/mL) purified with Na + , in 508 300 mM NaCl protein crystallization buffer (26% PEG 5000, pH 8.8). The structures of SaFtsZ-509 GDP in the presence of divalent cation chelators were obtained by soaking of SaFtsZ-GDP 510 crystals in 10 mM CyDTA or 10 mM EGTA during 23 h, and by SaFtsZ-GDP crystallization 511 in the presence of 1 mM EDTA (26% P(EG 5000, pH 8.5). The structure of SaFtsZ complexed 512 to GDP, AlF4and Mg 2+ was obtained from SaFtsZ-GDP crystals (24% PEG 5000, pH 8.6) 513 soaked in 10 mM AlF4and 20 mM MgCl2 during 23 h. The structure of SaFtsZ complexed to 514 GMPPCP was obtained by crystallization of 14 mg/ml SaFtsZ-GMPPCP-EDTA, with 1 mM 515 GMPPCP, 190 µM EDTA and 216 µM MgCl2 (22% PEG 5000, pH 8.7). The FtsZ-GMPPCP-516 Mg 2+ structure was obtained by soaking FtsZ-GMPPCP-EDTA crystals (22% PEG 5000, pH 517 8.8) in 20 mM MgCl2 during 23 h. The FtsZ-GMPPCP-Mn 2+ structure was obtained by soaking 518 FtsZ-GMPPCP-EDTA crystals (22% PEG 5000, pH 8.5) in 20 mM MnCl2 during 23 h. The 519 structure of SaFtsZ complexed with GMPCP was obtained by slow co-crystallization of 25.5 520 mg/mL apo-SaFtsZ in 240 mM KAcetate (instead of 300 mM KCl) with 0.8 mM GMPCPP 521 (26% PEG 5000, pH 8.8). The structure of SaFtsZ(D210N)-GDP was obtained from small 522 crystals grown from SaFtsZ(210N)-GTP (15.6 mg/mL) in 27% PEG 5000, pH 8.8. The 523 structure of SaFtsZ(R143K)-GDP (12.6 mg/mL) was from crystals grown in 24% PEG 5000, 524 pH 8.4. The structure of SaFtsZ(Q48A)-GDP (15 mg/mL) was from crystals grown in 26% 525 PEG 5000, pH 8.7. The structure of SaFtsZ(D46A)-GDP (8.7 mg/mL) was from crystals grown 526 in 23% PEG 5000, pH 8.4. 527

Structure determination 528
All crystals were flash-cooled by immersion in liquid nitrogen and diffraction data were 529 collected at the ALBA (Spain) and ERSF (France) synchrotrons. All data were processed using 530 XDS [53] and Aimless from the CCP4 Suite [54]. Data collection and refinement statistics are 531 presented in S1  ESRF synchrotrons with help of local staff from beamlines XALOC and ID23-1, respectively. 547