An oligomeric state-dependent switch in FICD regulates AMPylation and deAMPylation of the chaperone BiP

AMPylation is an inactivating modification that matches the activity of the major endoplasmic reticulum (ER) chaperone BiP to the burden of unfolded proteins. A single ER-localised Fic protein, FICD (HYPE), catalyses both AMPylation and deAMPylation of BiP. However, the basis for the switch in FICD’s activity is unknown. We report on the transition of FICD from a dimeric enzyme, that deAMPylates BiP, to a monomer with potent AMPylation activity. Mutations in the dimer interface or in residues tracing an inhibitory relay from the dimer interface to the enzyme’s active site favour BiP AMPylation in vitro and in cells. Mechanistically, monomerisation relieves a repressive effect allosterically-propagated from the dimer interface to the inhibitory Glu234, thereby permitting AMPylation-competent binding of MgATP. Whereas, a reciprocal signal propagated from the nucleotide binding site, provides a mechanism for coupling the oligomeric-state and enzymatic activity of FICD to the energy status of the ER. Impact Statement Unique amongst known chaperones, the endoplasmic reticulum (ER)-localized Hsp70, BiP, is subject to transient inactivation under conditions of low ER stress by reversible, covalent modification – AMPylation. The enzyme responsible for this modification, FICD, is in fact a bifunctional enzyme with a single active site capable of both AMPylation and deAMPylation. Here we elucidate, by biochemical, biophysical and structural means, the mechanism by which this enzyme is able to switch enzymatic modality: by regulation of its oligomeric state. The oligomeric state-dependent reciprocal regulation of FICD activity is, in turn, sensitive to the ATP/ADP ratio. This allosteric pathway potentially facilitates the sensing of unfolded protein load in the ER and permits the transduction of this signal into a post-translational buffering of ER chaperone activity.


4
Introduction contain a conserved active site motif, HPFx(D/E)GN(G/K)R1xxR2, and many possess 80 a glutamate-containing inhibitory alpha helix (inh) responsible for auto-inhibition of 81 their canonical AMPylation activity (Engel et al, 2012;Goepfert et al, 2013). FICD is 82 a class II Fic protein (with its inh N-terminal to its Fic domain) and an ER-localised 83 type II, single-pass transmembrane protein, with a short cytoplasmic portion and a large 84 luminal-facing catalytic domain (Worby et al, 2009;Bunney et al, 2014). 85 Crystal structures of FICD and other Fic domain proteins suggest that engagement of 86 Glu234 (of the inh) with Arg374 (R2 of the Fic motif) prevents binding of MgATP in 87 a conformation conducive to catalysis (Engel et al, 2012;Goepfert et al, 2013;Bunney 88 et al, 2014;Truttmann et al, 2016). Moreover, in vitro, modification of BiP by purified 89 FICD requires mutation of Glu234; an observation suggesting that an AMPylation 90 repressed state is favoured by wild-type FICD. Remarkably, the Fic domain of FICD is 91 also responsible for BiP deAMPylation; an activity that depends on Glu234 (Preissler 92 et al, 2017a;Casey et al, 2017) and magnesium (Veyron et al, 2019). These findings 93 point to deAMPylation as the default activity of the bifunctional enzyme and implicate 94 Glu234 in a functional switch between the two antagonistic activities of the Fic active 95

site. 96
The Fic domain of human FICD forms a stable back-to-back asymmetric dimer via two 97 dimerisation surfaces (Bunney et al, 2014;Truttmann et al, 2016) and a monomerising 98 mutation in the dimer interface of Drosophila FICD did not block BiP deAMPylation 99 in vitro (Casey et al, 2017). Nonetheless, distantly related bacterial enzymes hint at a 100 possible regulatory role for Fic dimerisation: a mutation in Clostridium difficile Fic 101 (CdFic) dimer interface increased auto-AMPylation (Dedic et al, 2016) and changes in 102 oligomeric state affected the activity of the class III Fic protein from Neisseria 103 meningitidis (NmFic) (Stanger et al, 2016). 104 Here we report on the biochemical and structural basis of an oligomeric state-dependent 105 switch in FICD's activity, which is well suited to post-translationally regulate protein 106 folding homeostasis in the ER. 107 Gly299Ser mutation (in the secondary dimer surface) (Figure S1D-G). AUC yielded a 140 1.2 nM dimer dissociation constant (Kd) of wild-type FICD and SEC indicated a Kd in 141 the millimolar range for FICD L258D and a Kd of 9.5 µM for FICD G299S . We therefore 142 conclude that between 0.2 µM and 5 µM (concentrations at which the experiments that 143 follow were performed) the wild-type protein is dimeric, FICD L258D is monomeric, and 144 FICD G299S is partially monomeric. 145 In the presence of [α-32 P]-ATP both FICD L258D and FICD G299S established a pool of 146 AMPylated, radioactive BiP in vitro [ Figure 1F; also observed in the Drosophila 147 counterpart of FICD L258D (Casey et al, 2017)], whereas the wild-type enzyme did not, 148 as previously observed (Preissler et al, 2015b(Preissler et al, , 2017a. BiP is a substrate for 149 AMPylation in its monomeric, ATP-bound, domain-docked conformation (Preissler et 150 al, 2015b(Preissler et 150 al, , 2017b. These experiments were therefore performed with an ATPase-151 deficient, oligomerisation-defective, ATP-bound BiP mutant, BiP T229A-V461F . Thus, the 152 BiP-AMP signal is a result of the concentration of substrate (unmodified and modified 153 BiP) and the relative AMPylation and deAMPylation activities of the FICD enzyme. 154 As expected, a strong BiP-AMP signal was elicited by the unrestrained AMPylation-155 active FICD E234G (which cannot deAMPylate BiP). FICD E234G-L258D gave rise to a 156 similar, but reproducibly slightly weaker, BiP-AMP signal relative to FICD E234G . 157

Monomerisation switches FICD's enzymatic activities 158
The ability of the dimer interface FICD mutants to yield a detectable BiP-AMP signal 159 in vitro agreed with the in vivo data and suggested a substantial change in the regulation 160 of the enzyme's antagonistic activitieseither inhibition of deAMPylation, de-161 repression of AMPylation, or a combination of both. To distinguish between these 162 possibilities, we analysed the deAMPylation activities of the FICD mutants in an assay 163 that uncouples deAMPylation from AMPylation. As previously observed, wild-type 164 FICD caused the release of fluorescently labelled AMP from in vitro AMPylated BiP, 165 whereas FICD E234G did not (Preissler et al, 2017a) (Figure 2A). FICD L258D and 166 FICD G299S consistently deAMPylated BiP 2-fold slower than the wild-type ( Figure 2A  167 and S2A). The residual in vitro deAMPylation activity of FICD L258D and the absence 168 of such activity in FICD E234G is consistent with the divergent effect of expressing these 169 deregulated mutants on a UPR reporter in cells ( Figure S2B-C).
The FICD-mediated BiP AMPylation/deAMPylation cycle converts the co-substrate 171 ATP to the end products AMP and pyrophosphate (Preissler et al, 2017a). We exploited 172 this feature to quantify enzymatic activity. FICD was incubated with [α-32 P]-ATP, 173 either in the presence or absence of ATPase-deficient BiP T229A , and accumulation of 174 radioactive AMP was measured by thin layer chromatography. Only background levels 175 of AMP were generated by catalytically inactive FICD H363A or FICD E234G-H363A (Figure  176 2B). The deregulated, deAMPylation-defective FICD E234G yielded a weak AMP signal 177 that was not increased further by the presence of BiP, suggesting that the Glu234Gly 178 mutation enables some BiP-independent ATP hydrolysis to AMP. Conversely, small 179 but significant amounts of AMP were produced by wild-type FICD but in a strictly BiP-180 dependent fashion ( Figure 2B-C and Figure S2D). These observations are consistent 181 with a slow, FICD-driven progression through the BiP AMPylation/deAMPylation 182 cycle indicating incomplete repression of wild-type FICD's AMPylation activity under 183 these conditions. As expected, abundant BiP-dependent AMP production was observed 184 in reactions containing AMPylation-active FICD E234G alongside deAMPylation-active 185 wild-type FICD ( Figure 2B, lane 11). Importantly, large amounts of AMP were also 186 generated when BiP was exposed to FICD L258D and, to lesser extent, FICD G299S ( Figure  187 2C and S2D). Together, these observations suggest that the AMPylation activities of 188 the monomeric FICD mutants are significantly enhanced relative to the wild-type, 189 whilst their deAMPylation activities are more modestly impaired. 190 To directly assess the AMPylation activities of bifunctional FICDs we exploited the 191 high affinity of the catalytically inactive FICD H363A for BiP-AMP, as a "trap" that 192 protects BiP-AMP from deAMPylation ( Figure 2D). To disfavour interference with the 193 FICD enzyme being assayed we engineered the trap as a covalent disulfide linked dimer 194 incapable of exchanging subunits with the active FICD being assayed. A cysteine 195 (Ala252Cys) was introduced into the major dimerisation surface of the trap. To 196 preclude aberrant disulphide bond formation, the single endogenous cysteine of FICD 197 was also replaced (Cys421Ser). After purification and oxidation, this protein (S-198 SFICD A252C-H363A-C421S ; the trap) formed a stable disulphide-bonded dimer ( Figure S2E-199 F) that tightly bound BiP-AMP with fast association and slow dissociation kinetics 200 ( Figure S2G-H). Moreover, the binding of the trap to unmodified BiP was, in 201 comparison, negligible ( Figure S2G). We reasoned that adding the trap in excess to 202 reactions assembled with BiP, ATP and FICD would sequester the BiP-AMP product 203 and prevent its deAMPylation, enabling the comparison of AMPylation rates in 204 isolation from the deAMPylating activity. 205 In presence of the trap, wild-type FICD produced a detectable BiP-AMP signal; but not 206 in the absence of the trap (compare Figures 1F and 2E). Importantly, presence of the 207 trap revealed that AMPylation of BiP was greatly accelerated by FICD monomerisation 208 (> 19-fold compared to the wild-type) ( Figure 2E). As expected, BiP AMPylation by 209 FICD E234G was even faster. 210 If the enhanced AMPylation activity of the dimerisation-defective mutants, observed 211 above, truly represents divergent enzymatic activities of different FICD oligomeric 212 states, it should be possible to reveal this feature by diluting the wild-type enzyme to 213 concentrations at which an appreciable pool of monomer emerges. In AMPylation 214 reactions set up with [α-32 P]-ATP a detectable signal from radiolabelled BiP-AMP was 215 noted at enzyme concentrations near the Kd of dimerisation (between 10 and 2.5 nM; 216 Figure 3A, left). The inverse relationship of enzyme concentration to the BiP-AMP 217 signal likely reflects the opposing activities and relative populations of AMPylation-218 biased FICD monomers and the deAMPylation-biased FICD dimers in each reaction. 219 This counter-intuitive relationship of enzyme to product is resolved in the presence of 220 the AMPylation trap; the BiP-AMP signal increased in a time-and enzyme 221 concentration-dependent manner, as expected from a reaction which is proportional to 222 the absolute concentration of monomeric enzyme ( Figure 3A, right). In the presence of 223 the trap the shift in the peak of the BiP-AMP signal, after 16 hours, towards lower 224 concentrations of FICD, likely reflects incomplete protection of AMPylated BiP by the 225 trap and its enhanced susceptibility to deAMPylation at higher concentrations of 226 If monomerisation significantly enhances AMPylation activity, constitutive FICD 228 dimers that are unable to dissociate should have low AMPylation activity and fail to 229 produce modified BiP even under dilute conditions. To test this prediction, we created 230 a disulphide-linked wild-type FICD (S-SFICD A252C-C421S ), which, after purification and 231 oxidation, formed a covalent dimer ( Figure S3A). Moreover, its SEC profile was 232 indistinguishable from wild-type FICD or the cysteine-free counterpart, FICD C421S 233 ( Figure S3B). In the presence of the BiP-AMP trap, oxidised S-SFICD A252C-C421S 234 produced significantly less AMPylated BiP than either wild-type or FICD C421S at 235 similar concentrations ( Figure 3B, lane 8 and S3C).
Repression of AMPylation was imposed specifically by the covalent dimer, as non-237 oxidised FICD A252C-C421S elicited a conspicuous pool of BiP-AMP -more than the wild-238 type enzyme ( Figure 3B, lane 9 and S3C) -an observation explained by the weakening 239 of the FICD dimer imposed by the Ala252Cys mutation ( Figure S1D-E). Similarly, in 240 absence of the trap, the ability of pre-oxidised S-SFICD A252C-C421S to establish a pool of 241 AMPylated BiP was greatly enhanced by diluting the enzyme into a buffer containing 242 DTT. FICD C421S , by contrast, produced similar amounts of modified BiP under both 243 non-reducing and reducing conditions ( Figure 3C). 244 DeAMPylation activities of oxidised and non-oxidised FICD A252C-C421S were 245 comparable and similar to wild-type FICD ( Figure 3D-E, S2A and S3D), pointing to 246 the integrity of these mutant enzymes. Together, these observations argue that covalent 247 S-SFICD A252C-C421S dimers selectively report on the enzymatic characteristics of wild-248 type FICD in its dimeric state. This protein therefore serves to help validate the 249 conclusion that a low concentration of wild-type FICD favours the formation of 250 monomers, whose AMPylation activity is de-repressed, to promote BiP modification. 251 An AMPylation-repressive signal is transmitted from the dimer interface to the 252 active site 253 The crystal structure of dimeric FICD suggests the existence of a hydrogen-bond 254 network, involving the side-chains of Lys256 and Glu242, linking the dimer interface 255 with the enzyme's active site, impinging on the AMPylation-inhibiting Glu234 ( Figure  256 4A). To test this notion, we mutated both putative dimer relay residues. FICD K256S and 257 FICD E242A formed stable dimers, as assayed by SEC, with dimer Kd values under 400 258 nM ( Figure 4B and S1D-E). In vitro both mutants established a pool of modified BiP 259 ( Figure 4C and S4A). This remained the case even at FICD concentrations in which 260 negligible amounts of monomer are predicted (2 and 10 µM; Figure S4A). De-261 repression of AMPylation by these dimer relay mutations was also evidenced by the 262 enhanced BiP-dependent AMP production, relative to wild-type FICD ( Figure 4D  and S4C). These observations suggest that residues connecting the dimer interface and 268 the active site contribute to repression of AMPylation and that mutating these residues 269 uncouples a gain-of-AMPylation activity from the oligomeric state of FICD. 270 Transmission of a repressive signal via a network of intramolecular interactions is also 271 supported by the correlation between de-repression of BiP AMPylation and the 272 negative effect of various mutants on the global stability of FICD. Differential scanning 273 fluorimetry (DSF) revealed an inverse relationship between the AMPylation activity 274 and the melting temperature (Tm) of FICD mutants ( Figure 4E and S4D). These 275 differences in flexibility were observed despite the fact that the DSF assays were 276 conducted at relatively high protein concentrations (2 µM) that would favour 277 dimerisation of all but the most dimerisation-defective mutants. 278 Nucleotide binding stabilises all FICD variants ( Figure S4D), a feature that is 279 conspicuous in case of the AMPylation de-repressed FICD E234G (Bunney et al, 2014). 280 However, monomerisation imposed by the Leu258Asp mutation, did not significantly 281 increase ATP-induced stabilisation of FICD (∆Tm) ( Figure 4F and S4E). Interestingly, 282 although AMPylation activity correlated with increased FICD flexibility this was not 283 reflected in an appreciably altered propensity to bind ATP. This suggested that the 284 variation in enzyme activity of different FICD mutants may arise not from variation in 285 their affinity for nucleotide but from their particular mode of ATP binding. To explore 286 this possibility, we set out to co-crystallise FICD variants with MgATP. 287

Monomerisation favours AMPylation-competent binding of MgATP 288
High-resolution X-ray crystal structures of monomeric and dimeric FICD were 289 obtained in various nucleotide bound states (Table 1). The tertiary structure of the Fic 290 domain of both the monomeric FICD L258D and the dimeric relay mutant FICD K256S 291 deviated little from that of the nucleotide-free wild-type dimer structure (FICD:Apo; 292 PDB: 4u04) ( Figure 5A and S5A). Moreover, co-crystallisation of FICD L258D , 293 FICD K256A or the wild-type dimer with ATP or an ATP analogue (AMPPNP) also 294 resulted in no significant Fic domain conformational change from FICD:Apo ( Figure  295 5A and S5A). Accordingly, the greatest root-mean squared deviation (RMSD) between 296 the Fic domain of the FICD:ATP structure and any other monomeric or dimer relay 297 FICD structure is 0.53 Å (observed between FICD:ATP and FICD L258D :Apo; residues 298 213-407). The only conspicuous change in global tertiary structure occurred in the TPR 299 domain of FICD L258D co-crystallised with ATP or AMPPNP, in which the TPR domain 300 is flipped almost 180° from its position in other FICD structures ( Figure 5A). Notably, 301 in all FICD structures the inh remains firmly juxtaposed to the core Fic domain. 302 When co-crystallised with MgATP or MgAMPPNP the resulting FICD structures 303 contained clear densities for nucleotide ( Figure 5B and S5B). The AMPylation-biased 304 FICD mutants also contained discernible, octahedrally coordinated Mg 2+ ions ( Figure  305 5Bii-iii and S5B). As noted in other Fic AMPylases, this Mg 2+ was coordinated by the 306 and -phosphates of ATP/AMPPNP and Asp367 of the Fic motif (Xiao et al, 2010;307 Khater & Mohanty, 2015b;Bunney et al, 2014). Interestingly, in the dimeric wild-type 308 FICD:ATP structure, crystallised in the presence of MgATP, there was no density that 309 could be attributed to Mg 2+ (Figure 5Bi). The only possible candidate for Mg 2+ in this 310 structure was a water density, located between all three phosphates, that fell in the Fic 311 motif's anion-holea position incompatible with Mg 2+ coordination (Zheng et al, 312 2017). 313 Alignment of the nucleotide-bound structures revealed that ATP or AMPPNP were 314 bound very differently by the wild-type dimer and the AMPylation-biased monomeric 315 or dimer relay FICD mutants ( Figure 5C and S5C). Concordantly, the RMSD of ATP 316 between the wild-type FICD and monomeric FICD L258D was 2.17 Å (and 2.23 Å for 317 FICD K256A 's ATP). As previously observed in other ATP-bound Fic proteins that 318 possess an inhibitory glutamate, the nucleotide in FICD:ATP was in an AMPylation 319 non-competent conformation (Engel et al, 2012;Goepfert et al, 2013) that is unable to 320 coordinate Mg 2+ ; an essential ion for FICD-mediated AMPylation (Ham et al, 2014). inhibitory glutamates (Xiao et al, 2010;Engel et al, 2012;Goepfert et al, 2013;Bunney 330 et al, 2014). As a result, in-line nucleophilic attack into the --phosphoanhydride bond of ATP would not be sterically hindered and the N2 of His363 would be well 332 positioned for general base catalysis ( Figure 5C and S5C-D). 333 The presence of ATP in both dimeric wild-type FICD and monomeric FICD L258D 334 (although in different binding modes) is consonant with the DSF data ( Figure 4F and 335 S4E). Apart from Glu234, the residues directly interacting with ATP are similarly 336 positioned in all structures (maximum RMSD 0.83 Å). However, considerable 337 variability is observed in Glu234, with an RMSD of 4.20 Å between monomeric and 338 dimeric wild-type ATP structures, which may hint at the basis of monomerisation-339 induced AMPylation competency. In ATP-bound structures the inhibitory glutamate is 340 displaced from the respective apo ground-state position, in which it forms an inhibitory 341 salt-bridge with Arg374: R2 of the Fic motif ( Figure S6A). However, the displacement 342 of the Glu234 side chain observed in the FICD:ATP structure (from its position in 343 FICD:Apo; PDB 4u0u) would be insufficient for AMPylation-competent binding of the 344 -phosphate of an ATP/AMPPNP (see distances i and ii, Figure 5C and S5C). This 345 steric clash is relieved by the side chain conformations observed in the AMPylation-346 competent structures (see iii and iv, Figure 5C and S5C). 347 The findings above suggest that the AMPylation-biased FICD mutants attain their 348 ability to competently bind MgATP by increased flexibility at the top of the inh and by 349 extension through increased Glu234 dynamism. It is notable that all the nucleotide 350 triphosphate-bound FICDs crystallised with intact dimer interfaces ( Figure S6A

ATP is an allosteric modulator of FICD 361
Given the conspicuous difference in the ATP binding modes observed between 362 AMPylation-competent FICD mutants and the AMPylation-incompetent wild-type dimeric FICD, we were intrigued by the possibility that ATP may modulate other 364 aspects of FICD enzymology and regulation. 365 In order to explore the effects of nucleotide on the different pre-AMPylation complexes 366 formed between either dimeric or monomeric FICD and its co-substrate, ATP-bound 367 BiP, we utilised BioLayer Interferometry (BLI). Biotinylated, client-binding-impaired, 368 ATPase-defective BiP T229A-V461F was made nucleotide free (Apo) and immobilised on 369 a streptavidin biosensor. Its interactions with catalytically inactive, dimeric FICD H363A 370 or catalytically inactive, monomeric FICD L258D-H363A were measured in the presence 371 and absence of nucleotides. The binding of both monomeric and dimeric FICD to 372 immobilised BiP was greatly enhanced by the pre-saturation of BiP with ATP ( Figure  373 6A and S8A). This is consistent with ATP-bound BiP as the substrate for FICD-374 mediated AMPylation (Preissler et al, 2015b). Moreover, the binding signal produced 375 by immobilised, ATP-bound BiP interacting with monomeric FICD L258D-H363A :Apo was 376 significantly stronger than that produced from the corresponding dimeric 377 FICD H363A :Apo analyte ( Figure 6A). In contrast, AMPylated BiP bound more tightly to 378 dimeric FICD H363A than to monomeric FICD L258D-H363A (forming a pre-deAMPylation 379 complex, Figure S2G). These findings align with the role of dimeric FICD in 380 deAMPylation and the monomer in AMPylation. 381 Interestingly, in presence of magnesium bound nucleotide (either MgATP or MgADP) 382 the FICD H363A interaction with ATP-bound BiP was weakened ( Figure 6A). This effect 383 was considerably more pronounced for monomeric FICD L258D-H363A . To quantify the 384 effect of FICD monomerisation on the kinetics of pre-AMPylation complex 385 dissociation, BLI probes preassembled with biotinylated, ATP-bound BiP and either 386 apo dimeric FICD H363A or apo monomeric FICD L258D-H363A were transferred into 387 otherwise identical solutions ± ATP (schematised in Figure S8B). The ensuing 388 dissociations fit biphasic exponential decays and revealed that ATP binding to FICD 389 accelerated the dissociation of monomeric FICD H363A more than dimeric FICD H363A 390 ( Figure 6B and S8C). The effect of ATP was noted on both the slow dissociation phase 391 of the monomer (koff,slow; Figure 6C-D) and on the percentage of dissociation attributed 392 to the fast phase (%Fast; Figure 6D  The structural data indicates that FICD's oligomeric state can impact significantly on 398 the mode of ATP binding, and Figure 6B indicates an allosteric effect of nucleotide 399 binding on FICD. Together these observations suggested bi-directional intramolecular 400 signalling from the dimer interface to the nucleotide-binding active site and therefore 401 the possibility that ATP binding in FICD's active site may also influence the oligomeric 402 state of the protein. To investigate this hypothesis, hetero-dimers of N-terminally 403 biotinylated FICD H363A assembled with non-biotinylated FICD H363A were loaded onto 404 a BLI streptavidin biosensor. The dissociation of non-biotinylated FICD H363A from its 405 immobilised partner was then observed by infinite dilution into buffers varying in their 406 nucleotide composition ( Figure 6E and S8E, schematised in Figure 6F). ATP but not 407 ADP induced a 3-fold increase in the dimer off rate ( Figure 6G). This is suggestive of 408 a mechanism whereby changing ATP/ADP ratios in the ER may modulate the 409 oligomeric state of FICD. 410

412
This study addresses a key process in the post-translational UPR by which bifunctional 413 FICD switches between catalysis of BiP AMPylation and deAMPylation, in order to 414 match the folding capacity of the ER to the burden of unfolded proteins independently 415 of changes in gene expression. The high affinity of FICD protomers for each other 416 specifies the presence of principally dimeric FICD in the ER, shown here to restrict the 417 enzyme to deAMPylation. This is the dominant mode of FICD both in vitro and in cells 418 under basal conditions (Preissler et al, 2017a;Casey et al, 2017). However, establishing 419 a pool of monomeric FICD unmasks its potential as a BiP AMPylase and enfeebles 420 deAMPylation. The structural counterpart to this switch is the mode by which MgATP, 421 the AMPylation reaction's co-substrate, is productively engaged in the active site of the 422 monomeric enzyme. Our studies suggest that monomerisation relieves the repression 423  The inverse correlation observed between the thermal stability of FICD mutants and 440 their AMPylation activity, supports a role for enhanced flexibility in enabling the 441 enzyme to attain the conformation needed for catalysis of this reactiona role clarified 442 by the crystallographic findings (see below). The biophysical assays also suggest that 443 monomeric FICD is more allosterically sensitive to ATP binding, as it exhibits a pronounced nucleotide-dependent reduction in the affinity for its co-substrate, ATP-445 bound BiP. The observation that ATP significantly accelerated the dissociation of 446 monomeric, nucleotide-free FICD from ATP-bound BiP suggests that this feature of 447 the monomer is mediated allosterically (not by enhanced susceptibility of a destabilised 448 protein to co-substrate competition for the same active site). The lower affinity of 449 monomeric FICD for its BiP:ATP co-substrate, in the context of a quaternary pre-450 AMPylation complex, conspicuously distinguishes it from the dimer and is a feature 451 that may also enhance AMPylation rates: ground-state destabilisation has been 452 demonstrated in a number of enzymes as a means of catalytic rate enhancement, by 453 reducing the otherwise anti-catalytic tight binding of an enzyme to its substrate 454 (Andrews et al, 2013;Ruben et al, 2013). 455 A structure of the quaternary pre-AMPylation complex, that could inform our 456 understanding of the features of the monomeric enzyme, does not exist. Nevertheless, 457 important insights into the effect of monomerisation were provided by structures of 458 FICD and its nucleotide co-substrate. Dimeric wild-type FICD binds ATP (without 459 magnesium) in an AMPylation incompetent mode. This is consistent with all other 460 inhibitory glutamate containing Fic structures crystallised with ATP or ATP analogues 461 (Engel et al, 2012;Goepfert et al, 2013). In stark contrast, we have discovered that 462 despite the presence of an inhibitory glutamate, monomerisation, or mutations in 463 residues linking the dimer interface to Glu234, permit the binding of ATP with 464 magnesium in a conformation competent for AMPylation. 465 Our studies suggest that the disparity in FICD's ATP binding modes stems from a CdFic (Dedic et al, 2016) and Bacteroides thetaiotaomicron (BtFic; PDB: 3cuc), but 499 not in the monomeric Shewanella oneidensis Fic (SoFic) protein (Goepfert et al, 2013). 500 Moreover, a His57Ala mutation in dimeric CdFic (which is structurally equivalent to 501 FICD K256A ) causes increased solvent accessibility and auto-AMPylation of a region 502 homologous to the loop linking FICD's Glu242-helix and the inh (Dedic et al, 2016). 503 Despite differences in detail, these findings suggest the conservation of a repressive 504 relay from the dimer interface to the active site of dimeric Fic proteins. 505 Our biophysical observations also suggest a reciprocal allosteric signal propagated 506 from FICD's nucleotide binding site back to the dimer interface; enhanced dimer 507 dissociation was induced by ATP but not ADP. Consequently, it is tempting to 508 speculate that FICD's oligomeric state and hence enzymatic activity might be regulated 509 by the ADP/ATP ratio in the ER. Under basal conditions, low ADP concentrations 510 allow ATP to bind both the monomeric and dimeric pools of FICD, shifting the 511 equilibrium towards the monomer and favouring BiP AMPylation. Stress conditions 512 may increase ADP concentration in the ER (perhaps by increased ER chaperone 513 ATPase activity). This increase would be proportionally much greater than the 514 concomitant decrease in [ATP] (in terms of respective fold changes in concentration). 515 The increased [ADP] would therefore be able to effectively compete with ATP for the 516 monomer-dimer FICD pools and thereby shift the equilibrium back towards the BiP de-517 AMPylating FICD dimer. 518 The regulation of BiP by FICD-mediated AMPylation and deAMPylation provides the 519 UPR with a rapid post-translational strand for matching the activity of a key ER 520 chaperone to its client load. The simple biochemical mechanism proposed here for the 521 requisite switch in FICD's antagonistic activities parallels the regulation of the UPR 522 transducers, PERK and IRE1, whose catalytically-active conformation is strictly linked 523 to dimerisation (Dey et al, 2007;Lee et al, 2008). A simple correlation emerges, 524 whereby ER stress favours dimerisation of UPR effectors, activating PERK and IRE1 525 to regulate gene expression and the FICD deAMPylase to recruit BiP into the chaperone 526 cycle (possibly through an increased ER ADP/ATP ratio). Resolution of ER stress 527 favours the inactive monomeric state of PERK and IRE1 and, as suggested here, the 528 AMPylation-competent monomeric FICD ( Figure 7). 529

531
The FICD crystal structures have been deposited in the PDB with the following 532  BiP V461F -AMP FAM by the indicated FICD proteins (at 7.5 µM) as detected by a change 607 in fluorescence polarisation (FP). DeAMPylation rates calculated from independent 608 experiments are given in Figure S2A.  were detected by autoradiography, quantified, and normalised to the signal in lane 6. 638 The mean radioactive signals ± SD from three independent experiments are given. The  The elution times of protein standards are indicated as a reference. 666

C) Radioactive in vitro AMPylation reactions containing the indicated FICD proteins, 667
[-32 P]-ATP, and BiP T229A-V461F were analysed by SDS-PAGE. The radioactive BiP-668 AMP signals were detected by autoradiography and proteins were visualized by 669 Coomassie staining of the gel. See Figure S4A. as well as dimeric wild-type FICD (dFICD) and FICD E234G (dFICD E/G ) were tested. 683 ADP and ATP concentrations in mM are given in parentheses. See Figure S4E for K½ 684 quantification. 685   Figure S8D. 739 E) BLI dissociation traces of the FICD dimer at different nucleotide concentration. At 740 t = 0 the species on the biosensor is a heterodimer of N-terminally biotinylated and an 741 exchangeable, non-biotinylated FICD. Dissociation was conducted ± ligands (5 mM), 742 as indicated. A representative experiment of four independent repeats, with mono-743 exponential fits are shown. See Figure S8E for raw data. 744 F) Cartoon schematic of the BLI assay workflow used to derive data presented in (E) 745 and Figure S8E.     (as in Figure 2B). Plotted below are mean AMP values ± SD (n = 3). The elution times of protein standards are indicated as a reference. Note that wild-type 841 FICD, FICD C421S , and oxidised S-SFICD A252C-C421S co-elute as dimers. See Figure S1D-    Tm values ± SD from three independent experiments. Note that FICD K256A is more 880 stable than FICD K256S but less than wild-type FICD. Furthermore, the stabilities of 881 oxidised and non-oxidised FICD C421S-A252C relative to the wild-type correlate inversely 882 with their AMPylation activities ( Figure 3B). The same data for the wild-type FICD, 883 FICD E242A , FICD G299S , FICD L258D and FICD K256S-L258D in the Apo state are presented in 884 Figure 4E. 885 E) Plot of the increase in FICD melting temperature (∆Tm) against ATP concentration 886 as measured by DSF (derived from Figure 4F). Note the similarity in the plot of 887 FICD L258D (mFICD) and the wild-type dimer (dFICD); mFICD K½ 2.5 ± 0.6 mM and 888 dFICD K½ 3.2 ± 0.3 mM. Shown are mean ∆Tm values ± SD of three independent 889 experiments with the best fit lines for a one site binding model.  helix (see Figure S6). Note, structure averaged B-factors are comparable (see Table 1). 972 For clarity, the TPR domain (up to residue 182) is not shown. 973 and also the ligand present in the dissociation buffer (at 5 mM) if applicable. Note, 1004 probes loaded with biotinylated FICD incubated with mFICD H363A act as controls for non-specific association and dissociation signals, these were subtracted from the 1006 respective dFICD H363A traces in Figure 6E. 1007 Table S1 1008 Crystallisation conditions. Where applicable the crystallisation conditions (and seed dilution) of the crystals used for micro-seeding are also shown. 1009 Note, PEG percentage is given in w/v and EtOH percentage in v/v.

Plasmid construction 1016
The plasmids used in this study have been described previously or were generated by 1017 standard molecular cloning procedures and are listed in Table S2. Experiments were performed at cell densities of 60-90% confluence. Where indicated, 1034 cells were treated with cycloheximide (Sigma) at 100 µg/ml diluted with fresh, pre-1035 warmed medium and then applied to the cells by medium exchange. 1036 1037

Mammalian cell lysates 1038
Cell lysis was performed as described in (Preissler et al, 2015a) with modifications. In 1039 brief, mammalian cells were cultured on 10 cm dishes and treated as indicated and/or 1040 transfected using Lipofectamine LTX with 5 µg plasmid DNA, and allowed to grow the FICD protein and mCherry as a transfection marker, using Lipofectamine LTX as 1116 described previously (Preissler et al, 2015b). 0.5 µg DNA was used in Figure S2B 610/20, respectively. Data were processed using FlowJo and median reporter (in Q1 1123 and Q2) analysis was performed using Prism 6.0e (GraphPad). 1124 1125

Production of VSV-G retrovirus in HEK293T cells and infection of CHO-K1 cells 1126
In an attempt to establish BiP AMPylation in FICD -/cells ( Figure 1A), cells were 1127 targeted with retrovirus expressing FICD (incorporating the naturally-occurring 1128 repressive uORF found in its cDNA) and mCherry. HEK293T cells were split onto 6 1129 cm dishes 24 h prior to co-transfection of pBABE-mCherry plasmid encoding FICD 1130 (UK 1939; Table S2)  plate was developed with 400 mM LiCl and 10% (v/v) acetic acid as a mobile phase 1287 and the dried plates were exposed to a storage phosphor screen. The signals were 1288 detected with a Typhoon biomolecular imager and quantified using ImageJ64.  Table S1. 1460 Diffraction data were collected from the Diamond Light Source, and the data 1461 processed using XDS (Kabsch, 2010) and the CCP4 module Aimless (Winn et al, 1462(Winn et al, 2011Evans & Murshudov, 2013). Structures were solved by molecular replacement 1463 using the CCP4 module Phaser (McCoy et al, 2007;Winn et al, 2011). For the 1464 FICD L258D :Apo and FICD:ATP structures the human FICD protein (FICD:MgADP) 1465 structure 4U0U from the Protein Data Bank (PDB) was used as a search model. 1466 Subsequent molecular replacements used the solved FICD L258D :Apo structure as a 1467 search model. Manual model building was carried out in COOT (Emsley et al, 2010(Emsley et al, ) 1468 and refined using refmac5 (Winn et al, 2003). Metal binding sites were validated 1469 using the CheckMyMetal server (Zheng et al, 2017). Polder (OMIT) maps were 1470 generated by using the Polder Map module of Phenix (Liebschner et al, 2017;Adams 1471Adams et al, 2010. Structural figures were prepared using UCSF Chimera (Pettersen et al, 1472             Step