Guanidine hydrochloride reactivates an ancient septin hetero-oligomer assembly pathway in budding yeast

Septin proteins evolved from ancestral GTPases and co-assemble into hetero-oligomers and cytoskeletal filaments. In Saccharomyces cerevisiae, five septins comprise two species of hetero-octamers, Cdc11/Shs1–Cdc12–Cdc3–Cdc10–Cdc10–Cdc3–Cdc12–Cdc11/Shs1. Slow GTPase activity by Cdc12 directs the choice of incorporation of Cdc11 vs Shs1, but many septins, including Cdc3, lack GTPase activity. We serendipitously discovered that guanidine hydrochloride rescues septin function in cdc10 mutants by promoting assembly of non-native Cdc11/Shs1–Cdc12–Cdc3–Cdc3–Cdc12–Cdc11/Shs1 hexamers. We provide evidence that in S. cerevisiae Cdc3 guanidinium occupies the site of a ‘missing’ Arg side chain found in other fungal species where (i) the Cdc3 subunit is an active GTPase and (ii) Cdc10-less hexamers natively co-exist with octamers. We propose that guanidinium reactivates a latent septin assembly pathway that was suppressed during fungal evolution in order to restrict assembly to octamers. Since homodimerization by a GTPase-active human septin also creates hexamers that exclude Cdc10-like central subunits, our new mechanistic insights likely apply throughout phylogeny.

evidence that slow GTPase activity by Cdc12 directs the choice of incorporation of Cdc11 vs Shs1 23 into septin complexes. It was unclear why many septins, including Cdc3, lack GTPase activity. We 24 serendipitously discovered that the small molecule guanidine hydrochloride (GdnHCl) rescues 25 septin function in cdc10 mutants by promoting assembly of non-native Cdc11/Shs1-Cdc12-Cdc3-26 Cdc3-Cdc12-Cdc11/Shs1 hexamers. We provide evidence that in S. cerevisiae Cdc3 guanidinium 27 ion (Gdm) occupies the site of a "missing" Arg sidechain that is present in other fungal species in 28 which (i) the Cdc3 subunit is an active GTPase and (ii) Cdc10-less hexamers co-exist with octamers 29 in wild-type cells. These findings support a model in which Gdm reactivates a latent septin assembly 30 pathway that was suppressed during fungal evolution in order to restrict assembly to hetero-31 octamers. Given recent reports that septin hexamers made natively in human cells also involve 32 bypass of Cdc10-like central subunits via homodimerization of an active GTPase, our results 33 provide new mechanistic details that likely apply to septin assembly throughout phylogeny. 34

INTRODUCTION 53
Septin proteins are found in nearly every eukaryotic lineage, with the exception of land plants  specific residues and structural motifs in septin hetero-oligomer assembly, suggest a series of key 114 events during fungal septin evolution that enforced incorporation of Cdc10 subunits in hetero-115 oligomers, and provide what is, to our knowledge, the first evidence that guanidinium (Gdm) has the 116 potential to functionally replace Arg residues in vivo. 117

RESULTS 119
GdnHCl restores high-temperature septin function to cdc10 mutants in an Hsp104-120 independent manner 121 We previously reported interactions between an ATPase-dead mutant of Hsp104 and a 122 temperature-sensitive (TS) mutant of Cdc10, in which overexpression of the mutant Hsp104 123 inhibited the proliferation of cdc10(D182N) cells at an otherwise permissive low temperature (27ºC) 124 (30). We expected that the addition of 3 mM GdnHCl to the culture medium would mimic these 125 inhibitory effects in cells with wild-type (WT) Hsp104. Instead, we found complete rescue by GdnHCl 126 of the cdc10(D182N) TS proliferation defect at all temperatures tested (up to 37ºC) ( Fig 1A). GdnHCl 127 did not have the same effect on cells carrying TS alleles of other septins, although in its presence 128 a cdc12(G247E) strain grew slightly worse at 30ºC and slightly better at 37ºC (Fig 1A). To our 129 surprise, GdnHCl rescue of cdc10(D182N) did not require Hsp104 ( Fig 1B). Others have suggested 130 that a mitochondrial Hsp104 homolog, Hsp78, might share functions with Hsp104 (34), and Hsp78 131 and Hsp104 share the same residues within the ATP binding pocket that contact Gdm in the 132 bacterial Hsp104 homolog ClpB (Fig 1C), but Hsp78 was also dispensable for GdnHCl rescue of 133 cdc10(D182N) high-temperature growth (Fig 1B). Inhibition of Hsp104 (i.e., curing of prions) or 134 Hsp78 (i.e., induction of cytoplasmic petites) by GdnHCl in vivo requires concentrations in the 135 medium of ≥1 mM (35), but GdnHCl was sufficient to provide partial rescue of 37ºC growth by 136 cdc10(D182N) cells at concentrations <0.1 mM (Fig 1D,E). We conclude that GdnHCl suppresses 137 septin defects in cdc10(D182N) mutants via a mechanism that does not involve Hsp104 or its 138 homolog Hsp78. 139 140

GdnHCl drives exclusion of mutant Cdc10 molecules from higher-order septin structures 141
We previously noted that the cdc12(G247E) mutation appears to decrease the levels of WT Cdc10 142 (30). Cells tolerate loss of Cdc10 by assembling filaments via hexameric building blocks in which 143 Cdc3 forms a non-native G homodimer, but cdc10∆ cells are TS due to thermal instability of the 144 non-native Cdc3 homodimer (23). Rescue of the TS defects at 37ºC of the cdc10(D182N) and 145 cdc12(G247E) mutants by GdnHCl could reflect stabilization of Cdc10-less hexamers, rather than 146 any specific effect on the cdc10(D182N) allele. Consistent with this idea, GdnHCl also fully 147 suppressed the TS phenotype resulting from cdc10(G100E) mutation (S1 Fig), as well as that of 148 cdc10∆ (Fig 2A). Clear but partial GdnHCl rescue was also observed for cdc10(G44D) cells (S1 149 was restricted to the cytoplasm (Fig 2B). WT Cdc10-GFP, on the other hand, continued to localize 158 to the bud neck despite the presence of GdnHCl (Fig 2B). Similarly, in diploid cells co-expressing 159 Cdc10-mCherry and Cdc10(D182N)-GFP, the addition of GdnHCl specifically eliminated 160 Cdc10(D182N) incorporation in septin rings, with no obvious effect on WT Cdc10 (Fig 2C). Our 161 results suggest that when mutations within the Cdc10 G interface perturb the ability of Cdc10 to 162 acquire a conformation that binds tightly to Cdc3, GdnHCl allows other Cdc3 molecules to 163 outcompete the mutant Cdc10 proteins for occupancy of the Cdc3 G interface, ultimately co-164 assembling with Cdc11, Cdc12, and Shs1 into Cdc10-less hexamers capable of robust septin 165 functions. As predicted from this model, GdnHCl was unable to rescue cdc10∆ growth when Cdc3 166 carried a mutation (W364A) previously shown (23) to block Cdc3 homodimerization (Fig 2D). is able to promote Cdc10-less septin filament assembly outside of yeast cells, we co-expressed 173 Cdc3, Cdc11, and hexahistidine-tagged Cdc12 (6xHis-Cdc12) in E. coli, which normally lacks 174 septins, and purified trimeric Cdc3-6xHis-Cdc12-Cdc11 complexes using high-salt buffers, as 175 described previously (12,15,23). We then used centrifugal sedimentation of septin filaments 176 followed by densitometry of Coomassie-stained bands on SDS-PAGE gels (Fig 3) to ask if these 177 complexes could form hexamers and polymerize into filaments in vitro upon salt dilution. We 178 observed some truncation of the C terminus of Cdc11 during expression/purification, as well as co-179 purification of the E. coli chaperonin GroEL, which binds indiscriminately to affinity beads (37) and 180 co-migrates with Cdc3 (Fig 3). Hence we paid most attention to the amounts of 6xHis-Cdc12. 181 182 Previous studies found that ~35% of Cdc10-less complexes sediment in these conditions, compared 183 to ~80% of Cdc10-containing octamers (12), consistent with poor hexamer (and thus filament) 184 formation in the absence of Cdc10. We found similar results (Fig 3). To test effects of GdnHCl, we 185 either grew the bacteria in the presence of 100 mM GdnHCl and kept GdnHCl in all buffers 186 thereafter, or we only introduced the GdnHCl during polymerization. The presence of GdnHCl during 187 expression, purification, and polymerization was associated with an increase in septin complex 188 sedimentation (Fig 3). Adding GdnHCl just prior to polymerization had a subtler effect (Fig 3). 100 189 mM is well below GdnHCl concentrations that begin to unfold globular proteins (38), and the other 190 septins remained stoichiometrically associated with 6xHis-Cdc12 in the presence of 100 mM 191 GdnHCl (Fig 3), which we take as evidence that sedimentation reflected authentic filament 192 polymerization, rather than GdnHCl-induced partial denaturation and subsequent non-specific 193 aggregation. We conclude from these data that GdnHCl is able to promote septin filament assembly 194 by recombinant Cdc10-less septin complexes produced in a heterologous system that lacks septin-195 regulatory pathways. Thus GdnHCl likely acts directly on septin proteins, and seems to work best 196 when present during septin folding/assembly. Using an unbiased genetic approach, we previously identified a spontaneous mutation in the Cdc3 201 G interface, G261V, that improves the proliferation of cdc10∆ cells, and concluded that the 202 introduction of Val in this position stabilizes a conformation of the G interface that is more competent 203 for self-association than the major conformation populated by WT Cdc3 at high termperatures (23). 204 We reasoned that Gdm might bind in the Cdc3 G interface and similarly promote the 205 homodimerization-competent conformation. For clues as to where Gdm might bind, we first used 206 the structure of human SEPT2 bound to the non-hydrolyzable nucleotide analog 5'-guanylyl 207 imidodiphosphate (GppNHp) (39) to generate a homology model of the globular domain of 208 monomeric Cdc3, and then performed unbiased in silico docking simulations to ask for the lowest-209 energy predicted sites of Gdm binding. Four sites had binding free energies of ≤-3 kcal/mol. One 210 site, between the sidechains of Thr302 and His262, lies in the G interface, and had the second-211 lowest predicted free energy of binding ( Fig 4A). 212 213 As a parallel approach, we considered that (i) Gdm mimics the distal end of an Arg sidechain, and 214 (ii) a number of previous in vitro studies of diverse non-septin proteins -including T4 lysozyme (40), 215 β-galactosidase (41), and carboxypeptidase A (42) -demonstrated that Gdm is able to functionally 216 occupy molecular vacancies that are created by the replacement of an Arg residue with residues 217 having smaller sidechains (e.g. Ala). We thus hypothesized that one or more Arg residues might 218 contribute to G interface contacts for other septins that are, unlike S. cerevisiae Cdc3, capable of 219 robust G homodimerization. 220

221
To identify candidate Arg residues, we performed a phylogenetic analysis of Cdc3 homologs from 222 different species, which we separated into two categories based on the severity of phenotypes 223 accompanying deletion of the CDC10 homolog. If non-Saccharomyces species possess an Arg that 224 favors G homodimerization by Cdc3, then we predicted this group of species should easily form 225 Cdc10-less septin hexamers and be better able to tolerate deletion of the Cdc10 homolog. Such 226 functional information is available for non-Saccharomyces species from eleven distinct fungal 227 genera. In Aspergillus (fumigatis (43) or nidulans (44)), Candida albicans (45), Cryptococcus 228 neoformans (46), Fusarium graminearum (47), Neurospora crassa (48), Schizosaccharomyces 229 pombe (49), and Ustilago maydis (50), deletion of the Cdc10 homolog has distinctly milder 230 phenotypic consequences compared to deletion of the Cdc3 homolog. In Magnaporthe oryzae, 231 septin ring assembly is perturbed by deletion of any septin, but higher-order, filamentous structures 232 persist in cells lacking the Cdc10 homolog Sep4, whereas in the other mutants septin localization 233 is almost exclusively diffuse (51). Finally, in Coprinopsis cinerea, a UV-induced mutant defective in 234 fruiting body development was rescued by a gene encoding CcCdc3, but the expression of CcCdc10 235 remained about 100-fold decreased in the mutant following rescue, suggesting that the mutant was 236 defective in both CcCdc3 and CcCdc10 expression, and that rescue of CcCdc3 expression was 237 sufficient to restore function (52). We interpret these results as evidence that in C. cinerea, as well, 238 loss of Cdc10 is better tolerated than loss of Cdc3. 239 240 By contrast, only the Shs1 homolog is non-essential for proliferation in Kluyveromyces lactis (53). 241 In Ashbya gossypii, deletion of any mitotic septin prevents septin ring assembly; loss of the Cdc10 242 homolog is as severe as loss of the Cdc3 homolog (54). According to our hypothesis, if there is a 243 key Arg residue that promotes Cdc3 homodimerization, and S. cerevisiae lacks it, then K. lactis and 244 A. gossypii should also lack it. above. As ScCdc3 is predicted to be unable to hydrolyze GTP (39), we also included for comparison 248 human SEPT2 and SEPT9, because for these septins crystal structures are available of non-native 249 G homodimers in the "GTP state" (bound to the non-hydrolyzable GTP analogs GppNHp or GTP S, 250 respectively). Only sixteen residues were distinctly different between a group including S. cerevisiae 251 and the two species that cannot tolerate Cdc10 loss, and the group of species that tolerate CDC10 252 deletion. For six of these residues, the changes reflect substitutions with strong predicted effects on 253 amino acid properties (polarity, charge, size, etc.). One of these non-conservative variants lies in 254 the C-terminal extension, which is disordered in septin crystal structures. Based on the crystal 255 structures of SEPT2•GppNHp and SEPT9•GTP S, all of the other five non-conservative variants 256 cluster near the G interface (Fig 4B), and one of these residues -corresponding to Thr302 in 257 ScCdc3 -is an Arg in eight of the nine species that tolerate Cdc10 loss (S2 Fig). Moreover, in the 258 SEPT2•GppNHp structure the Gdm group of the corresponding Arg in SEPT2 is located in the same 259 place where our in silico docking results predicted that Gdm binds to Cdc3, between Thr302 and 260 His262 (Fig 4). Together, these findings strongly support the idea that Gdm promotes Cdc3 261 homodimerization by occupying the same site in the Cdc3 G interface that in other septins is 262 occupied by the Gdm moiety of an Arg sidechain. This Arg residue was likely substituted to Thr 263 during the evolution of the fungal lineage that includes S. cerevisiae, A. gossypii, and K. lactis, 264 concomitant with loss of Cdc10-less hexamers in these yeasts. 265

GdnHCl promotes higher-order septin assembly in A. gossypii cells lacking Cdc10 267
Our phylogenetic comparisons predicted that Gdm should act similarly on the A. gossypii Cdc3 268 homolog to promote AgCdc3 homodimerization in the absence of Cdc10. To test this prediction, we 269 exposed A. gossypii cells lacking CDC10 to GdnHCl and assessed higher-order septin assembly 270 by monitoring the localization of Shs1-GFP expressed in the same cells. As reported previously 271 (54), in the absence of GdnHCl Shs1-GFP fluorescence was cytoplasmic and diffuse ( Fig 5). 272 Addition of GdnHCl restored the localization of Shs1-GFP to branch points and hyphal tips (Fig 5), To better understand the molecular details of Gdm "rescue" of Arg-substituted mutants, earlier in 279 vitro Gdm studies also tested other small molecules with similar chemical properties, including urea 280 and the GdnHCl derivatives aminoguanidine (Pimagedine, here "aGdnHCl") and N-ethylguanidine 281 ("eGdnHCl") (55). We undertook an analogous approach to study GdnHCl rescue of the cdc10 TS 282 phenotype. 283 284 First, we tested urea, which differs from Gdm by only a few atoms (Fig 6A). We grew WT or cdc10 285 cells on solid rich medium containing 0.64 M urea, a concentration we found to slow, but not prevent, 286 growth of WT cells, and saw no cdc10 rescue (Fig 6A-B). To test a gradient of concentrations, we 287 also placed filter disks soaked with GdnHCl or urea on lawns of WT or cdc10 cells. We saw a clear 288 zone of rescue around the GdnHCl disk, but not with urea ( Fig 6C). We next exposed WT or cdc10 289 cells to 0.375 mM GdnHCl, aGdnHCl, eGdnHCl, or combinations of GdnHCl plus eGdnHCl or 290 GdnHCl plus aGdnHCl (0.375 mM each). At 0.375 mM GdnHCl provided only a partial rescue (Fig  291  6D), allowing us to detect subtle effects of the GdnHCl derivatives when combined with GdnHCl. At 292 22ºC, 0.375 mM of any single drug had no noticeable effect on growth; slight growth inhibition of 293 both WT and mutant cells was observed when the total concentration of GdnHCl plus derivative 294 was 0.75 mM (Fig 6D). At 37ºC, in addition to the expected partial cdc10 rescue by GdnHCl, there 295 was a very slight rescue by eGdnHCl, and an even less pronounced rescue by aGdnHCl (Fig 6D). 296 By contrast, mixing GdnHCl with aGdnHCl or eGdnHCl to a total concentration of 0.75 mM resulted 297 in full or nearly full rescue of cdc10 37ºC growth, respectively ( Fig 6D). We interpret these findings 298 to mean that aGdm and eGdm (the guanidinium ion derivatives) occupy the position between Cdc3 299 Thr302 and His262 differently than does Gdm. They may not provide the appropriate molecular 300 contacts across the G dimer interface, or they may bind less well, or even too well (if Gdm acts only 301 transiently during Cdc3 folding). Indeed, aGdm and eGdm were predicted by in silico modeling to 302 bind in the same location as Gdm but with lower free energies, with the additional moieties projecting 303 in various directions ( Fig 6E) in ways that might further alter Cdc3 conformation. It is also possible 304 that yeast cells are less permeable to these GdnHCl derivatives. Urea does not rescue at all, 305 possibly because it cannot form a cation. Finally, as would be expected if only small molecules like 306 Gdm can fit into the pocket provided by Thr302 and surrounding residues, 5 mM arginine 307 hydrochloride in the medium provided no cdc10 rescue ( Fig 6F). 308 309

Mutating Cdc3 Thr302 to Arg prevents cdc10∆ rescue by GdnHCl 310
To test if the absence of Arg at ScCdc3 position 302 is sufficient to explain the inability of S. 311 cerevisiae cdc10∆ cells to proliferate at 37ºC, we replaced Thr302 with Arg. The double-mutant 312 cdc10∆ cdc3(T302R) cells were, like cdc10∆ CDC3 + , temperature-sensitive, but unlike cdc10∆ 313 CDC3 + the addition of GdnHCl failed to restore proliferation at 37ºC (Fig 7). By contrast, when we 314 replaced Thr302 with an amino acid with similar properties, Val, we saw rescue by Gdm (Fig 7). We 315 interpret these data as evidence that Gdm binds to ScCdc3 near Thr302 (or Val302) in order to 316 promote homodimerization, but does so in a way that is not recapitulated by an Arg sidechain. 317 Instead, Arg302 blocks functional Gdm binding. We conclude that the T302R substitution alone 318 cannot "reverse evolution", yet GdnHCl does. How? 319 320 Arg or Gdm at ScCdc3 position 302 likely "tunes" a key His residue within the G homodimer 321 interface 322 ScCdc3 Thr302 is predicted to lie within the septin ⍺4 helix, which is located near both the nucleotide 323 binding pocket and the G interface (Fig 4). To begin to understand the molecular mechanism by 324 which Gdm at this position promotes Cdc3 homodimerization, we first examined thirteen available 325 septin crystal structures and asked which other residues are nearby the residue in the position 326 equivalent to Thr302. The only non-⍺4-helix residue within 5 Å in every structure was a highly 327 conserved His (Table 1)  In SEPT2•GppNHp, His158 is positioned in cis by contacts (~3 Å) with a Glu residue (Glu202) (39), 339 which lies one turn away from Arg198 in the ⍺4 helix ( Fig 4B). The Gdm moeity of SEPT2 Arg198 340 (residue corresponding to Thr302 in Cdc3) is also within ~3 Å of the backbone amide carbonyl of 341 * Residues at positions corresponding to ScCdc3 Thr302 were identified based on structure-guided alignments. SEPT2, Arg198; SEPT3, Arg224; SEPT7, Gln210; SEPT9, Arg442; SEPT10, Arg199; Cdc11, Leu187. ** Residues from the other protomer across the G interface. ***Dimer in solution is mediated by the C-terminal extension, and is unaffected by G interface mutations, thus contacts across the G interface are presumably crystal-induced.
His158 ( Fig 4B, Table 1), and thus is equally well located to position His158. In CrSEPT and 342 SEPT9•GTP S, the Arg198 equivalent instead contacts the Glu202 equivalent ( Fig 4B). Thus in 343 some but not all septins the Gdm moiety of the ⍺4 Arg is ideally positioned to contact the trans loop 344 1 His and thereby potentially bias partner choice during septin G dimerization. By contacting His262, 345 Gdm bound near Thr302 in Cdc3 could bias Cdc3 towards homodimerization and away from 346 To look for additional evidence in support of this model, we next asked which other residues are 349 located in the vicinity (≤5 Å) of the trans loop 1 His in thirteen available septin crystal structures. The 350 ⍺4 helix was the only region to meet this criterion in every structure, and, as expected, when Arg 351 was present at the position corresponding to ScCdc3 Thr302, it was always within 5 Å (Table 1). 352 Apart from the ⍺4 helix, the Switch II loop, and adjacent trans loop 1 residues, residues within 5 Å 353 of the His fell within only two other regions: the P-loop and the Switch I loop (Table 1). Remarkably, 354 these are precisely the regions wherein we found non-conservative G interface substitutions 355 between fungal species that tolerate CDC10 deletion and those that do not ( Fig 4B). These we proposed that the conformation of the Switch II loop in Cdc12 biases choice of its G-dimer 366 partner (Cdc11 or Shs1)(58). In that study, we showed that a Switch II mutation in Cdc12 is sufficient 367 to bias partner choice. In an independent study (29), we found that a spontaneous mutation (D210G) 368 in the equivalent Switch II residue in Cdc3 restores the ability of Cdc3 to interact with a nucleotide-369 free mutant Cdc10 at high temperatures. Like Gdm, the cdc3(D210G) mutation rescues the TS 370 phenotype of cdc10(D182N) cells, but unlike Gdm it does so by restoring, rather than bypassing, 371 incorporation of the mutant Cdc10 subunits; in other words, the Switch II mutation D210G biases 372 Cdc3 partner choice towards nucleotide-free Cdc10, whereas Gdm biases partner choice away from 373 it. Our model predicts that in cdc3(D210G) cells Gdm should be less able to exclude the mutant 374 Cdc10 than in cells expressing WT Cdc3, because the mutant Cdc3 Switch II will be unable to 375 accomplish the molecular contact(s) with the Cdc3 trans loop 1 that Gdm promotes. Indeed, Gdm 376 slightly reduced, but did not eliminate, Cdc10(D182N)-GFP localization to septin rings in 377 cdc3(D210G) cells ( Fig 8A). These data are consistent with an important role for contacts between 378 the Switch II loop and the trans loop 1 during Cdc3 homodimerization promoted by Gdm. 379 380 Since Cdc12 is an active GTPase, and the Switch II changes conformation upon GTP hydrolysis, 381 we interpreted our previous results with Switch II-mutant Cdc12 as evidence that the Switch II 382 conformation normally "communicates" across the G interface the phosphorylation state of Cdc12's 383 bound nucleotide (58). We further bolstered this argument by mutating in Cdc12 a Thr residue in 384 the Switch I loop that is required for septin GTPase activity (39,58). Notably, Cdc3 lacks Thr in this 385 position. How could Gdm operate via Switch II conformation on Cdc3, a "dead" GTPase? We noticed 386 in our phylogenetic analysis of Cdc3 homologs in other fungal species that many possess the 387 "catalytic Thr" (S2 Fig). In fact, there was a perfect correlation between the presence of the catalytic 388 Thr (or Ser) and the ⍺4 Arg whose Gdm group is presumably mimicked by Gdm in Cdc3 (S2 Fig). Analogous to our model of Cdc12 G-partner choice via slow GTP hydrolysis (58), we reasoned that 390 in a species with an active "Cdc3" GTPase, a transient "Cdc3"•GTP molecule might prefer to 391 dimerize with "Cdc10" in that species, and "Cdc3"•GDP might prefer to form a homodimer, 392 bypassing "Cdc10" incorporation into septin complexes. proliferate at 37ºC (Fig 8C). We hypothesized that this defect reflects misfolding at high temperature 408 of the Cdc3-AspB chimera to a conformation incapable of interacting with itself or with Cdc10, and 409 searched for spontaneous suppressors of the TS phenotype. We obtained a suppressor in which 410 proliferation at 37ºC was restored (Fig 8C). Sequencing the coding region of Cdc10 revealed a 411 single nucleotide change causing the amino acid substitution Q265H ( Fig 8D); the Cdc3-AspB 412 chimera was unchanged. In most septin structures, His in this position makes a critical contact 413 across the G interface with a highly conserved Trp residue in the "Sep4" motif (3) (Fig 8E), the same 414 Trp we mutated in the cdc3(W364A) mutant (see Fig 2D). Since Gln replaces His here in ScCdc10 415 (and in the other fungal species with putatively GTPase-dead Cdc3 homologs; Table 2), the 416 Cdc3•GTP-Cdc10•GDP interface must involve a different kind of interaction. We interpret these 417 findings as evidence that stable association of Cdc10•GDP with GDP-bound Cdc3-AspB, rather 418 than the Cdc3•GTP with which Cdc10 co-evolved, requires His-Trp contacts between Sep4 motifs 419 of the sort found in other dimers between two GDP-bound septins. Indeed, incorporation of Cdc10-420 mCherry into septin rings was partially restored in cdc3-aspB cdc10(Q265H) cells ( Fig 8B). Finally, 421 Gdm was unable to fully rescue the TS phenotype of the cdc3-aspB CDC10 + strain (Fig 8C), as 422 expected if the site of Gdm action is occluded by the ⍺4 Arg present in the Cdc3-AspB chimera. 423 These data provide further support for the idea that during evolution ScCdc3 lost the ability to 424 hydrolyze GTP to GDP and, consequently, the option of assembling septin complexes without a 425 central Cdc10 homodimer. 426

427
The GTPase domains of Cdc3 and AspB are <50% identical, and differ at many more positions than 428 the five identified by our phylogenetic analysis as co-varying with the ⍺4 Arg and the catalytic Thr 429 To ask if those five differences are sufficient to direct Cdc10 bypass during septin 430 assembly, we used CRISPR-Cas9 to cut the endogenous CDC3 coding sequence and, via 431 homologous recombination, replace most of it with a "recoded" gene encoding the same polypeptide 432 sequence but using numerous synonymous codons, or with a similarly "recoded" gene additionally 433 encoding the five substitutions (P127E D128S K181T T302R Q306D). Recoding allowed us to 434   Thus, although we cannot rule out the possibility that recoding with synonymous codons alters Cdc3 442 co-translational folding in some way that masks effects of the amino acid changes we introduced, it 443 appears that additional sequence changes are needed to "reverse evolution" and restore the ability provided one example of how Cdc10 adapted to interacting with a Cdc3 subunit "fixed" in the GTP-509 bound state. Others previously noted (39) that ScCdc10 is unusual among septins in that it has Lys 510 rather than His in the trans loop 1. If the residue in this position acts to promote interaction with a G 511 dimer partner in a specific nucleotide state, this may explain why Lys in this position is shared by 512 AgCdc10 and KlCdc10, but not any of the Cdc10 homologs in the fungal species in which the Cdc3 513 homolog retains the "catalytic Thr" (Table 2). Co-variation of these residues likely reflects co-514 evolution. S7A Fig illustrates a

Fungal strains and plasmids 569
The sources and genotypes for all fungal strains and plasmids are listed in the Supporting 570 Information (S1 Table). Primer and synthetic gene sequences are listed in S2 Table. Yeast were 571 transformed using the Frozen-EZ Yeast Transformation II Kit (Zymo Research). Genetic 572 manipulations were otherwise performed according to standard methods (90), except for the 573 creation of the cdc3-aspB chimera and the CDC3(P127E D128S K181T T302R Q306D) and 574 CDC3(T302R Q306D) strains and associated "recoded" control strain, which were made in yeast 575 strain JTY5397 using CRISPR-Cas9 cleavage of the CDC3 locus and repair with PCR products as 576 donor templates, following an established protocol (91) and using plasmid pEM-CDC3-CRISPR1, a 577 gift of Ed Marcotte, which encodes Cas9 and a guide RNA targeting nucleotides 1039-1069 of the 578 CDC3 ORF (5' GATATTGTAGAGAACTACAG 3'). To create the donor template for the cdc3-aspB 579 chimera, a portion of the aspB coding sequence, the aspB plasmid pRL10, which was a gift of 580 Michelle Momany and is based on the yeast two-hybrid vector pGBKT7 (Takara Bio/Clontech), was 581 used as template for a PCR reaction with Q5 polymerase (New England Biolabs) and primers 582 Cdc3AspBfw and Cdc3AspBextend1 according to the polymerase manufacturer's instructions. The 583 resulting product, which included the sequence encoding part of AspB flanked by CDC3 sequences, 584 was used as template for a second Q5 reaction with primers Cdc3AspBfw and 585 Cdc3GTPase_extend2, in order to extend the CDC3 homology to span the site of the Cas9 cut. To 586 create the donor template for the CDC3(P127E D128S K181T T302R Q306D) strain and its 587 "recoded" control, an 870-nucleotide segment of the Cdc3 ORF corresponding to the GTPase 588 domain was synthesized (Integrated DNA Technologies) with multiple synonymous codons 589 replacing the native sequence (S2 Table), representing 67 nucleotide changes but no amino acid 590 substitutions. An otherwise identical sequence also including the P127E D128S K181T T302R 591 Q306D mutations was also synthesized (S2 Table). Donor template PCRs were done with Q5 and 592 primers G1KTTRQDfw and Cdc3recodere (S2 Table). Transformants were screened by PCR of the 593 CDC3 locus and subsequent sequencing. Low-copy, LEU2-marked CDC3 or CDC3-GFP plasmids 594 were pFM831 (39) or pML109 (26). Plasmids YCpHLcdc3(T302R)-GFP and YCpHLcdc3(T302V)-595 GFP were made via site-directed mutagenesis using template plasmid pML109 by Keyclone 596 Technologies (San Marcos, CA). Low-copy, kanMX-marked plasmid YCpK-Cdc10-1-GFP, encoding 597 cdc10 (D182N)-GFP, was described previously (30).

Electron microscopy 637
Cells cultured at 37°C in YPD medium with 3 mM GdnHCl were collected by vacuum filtration and 638 vitrified by high pressure freezing using a HPM100 (Leica Microsystems, Wetzlar, Germany) 639 apparatus. The frozen cells were cryo-substituted using a Leica ASF2 device in a medium 640 containing osmium tetroxide (1%), water (5%) and uranyl acetate (0.1%) in acetone using a protocol 641 described elsewhere (94). The cells were embedded into Epon before being sectioned into 50 nm 642 sections using an ultramicrotome UC6 (Leica) equipped with a 4.5 mm diamond knife (Diatome, 643 Hatfield, PA). The resulting sections were deposited onto copper electron microscopy grids (mesh 644 size of 100). The images were collected with a 120 kV Lab6 microscope (Technai Spirit, FEI, 645 Eindhoven, Netherlands) equipped with a CCD Quemesa camera (Olympus, Tokyo, Japan). 646 647

Fluorescence microscopy 648
All images of S. cerevisiae were captured using an EVOSfl all-in-one microscope (Advanced 649 Microscopy Group, Mill Creek, Washington) using a 60X objective and Texas Red, RFP, or GFP 650 LED/filter cubes, as described previously (30). Bud neck fluorescence was quantified using line 651 scans as described previously (30,58). Intensity values or ratios thereof were plotted using 652 GraphPad Prism 8.0, using the medium smoothing (kernel density) setting for violin plots. When 653 necessary for presentation, images were inverted and brightness-and contrast-adjusted using 654 Adobe Photoshop (Adobe Systems Incorporated, San Jose, California), always the same way for 655 every image of the same type. 656 657 Ashbya cells in minimal low-fluorescence medium were mounted onto 2% agarose gel pads and 658 the edges were sealed with Valap (a 1:1:1 mixture of vaseline, lanolin, and paraffin). Images were 659 acquired using a Zeiss Axioimager-M1 upright light microscope (Carl Zeiss, Jenna, Germany) 660 equipped with a Plan-Apochromat 63X/1.4 numerical aperture oil objective and an Exfo X-Cite 120 661 lamp. Fluorescence imaging was performed using Zeiss 38HE filter cubes (GFP). Images were 662 acquired using an Orca-AG charge-coupled device (CCD) camera driven by µManager.

Sequence alignments 715
Multiple alignments were performed using the COBALT tool at the NCBI server 716 (https://www.ncbi.nlm.nih.gov/tools/cobalt/) or, for sequences of yeast species not available via 717 NCBI, using the "Fungal Alignment" function at the Yeast Genome Database (98). In some cases, 718 sequences obtained via the Yeast Genome Database were aligned with other sequences via 719 COBALT. 720

Trp Trp
His His