The protease GtgE from Salmonella exclusively targets inactive Rab GTPases

Salmonella infections require the delivery of bacterial effectors into the host cell that alter the regulation of host defense mechanisms. The secreted cysteine protease GtgE from S. Typhimurium manipulates vesicular trafficking by modifying the Rab32 subfamily via cleaving the regulatory switch I region. Here we present a comprehensive biochemical, structural, and computational characterization of GtgE in complex with Rab32. Interestingly, GtgE solely processes the inactive GDP-bound GTPase. The crystal structure of the Rab32:GDP substrate in complex with the inactive mutant GtgEC45A reveals the molecular basis of substrate recognition. In combination with atomistic molecular dynamics simulations, the structural determinants for protein and activity-state specificity are identified. Mutations in a central interaction hub lead to loss of the strict GDP specificity. Our findings shed light on the sequence of host cell manipulation events during Salmonella infection and provide an explanation for the dependence on the co-secreted GTPase activating protein SopD2.

C entral regulatory elements of intracellular signaling networks are small GTPases (guanosine triphosphate hydrolases, also referred to as G-proteins). G-proteins function as binary molecular switches and thereby inhibit or stimulate cellular networks via spatio-temporal protein-protein interactions 1 . In order to function in this capacity, the small GTPases can exist in the following two states: (1) they are inactive when complexed to guanosine diphosphate (GDP), and (2) they become active once bound to guanosine triphosphate (GTP). In the active state, signals are propagated by the recruitment of effector proteins that function as signal transmitting molecules. The activity states are regulated by dedicated regulatory enzymes that facilitate activation and inactivation of theG-proteins: guanine nucleotide exchange factors (GEFs) activate small GTPases by stimulating GDP-release and facilitating GTP-binding. In contrast, GTPase activating proteins (GAPs) accelerate inactivation via increasing the rate of intrinsic GTP-hydrolysis, converting the G-protein back to the inactive state. The activity states of small GTPases are manifested in specific conformations of two highly important regulatory loop regions that are referred to as switch I and switch II: the switch-regions are structurally disordered in the inactive state, but adopt a well-defined structure in the active form. The Rab-family of small GTPases is involved in regulating intracellular vesicular trafficking temporally and spatially 1 . These proteins are part of the so-called Rab-cycle that is characterized by a series of peripheral membrane association and cytosolic localization linked to nucleotide exchange. In order to attach to a membrane, Rab-proteins are post-translationally modified with one or two geranylgeranyl lipids at their C-termini. In the active state, these lipids confer strong membrane affinity and localize the Rab-protein to the cytosolic surface of intracellular compartments. In the inactive state, however, lipidated Rabs are targeted with high affinity by the GDP-dissociation inhibitor (GDI) that solubilizes the GTPase and thereby causes cytosolic localization.
Due to the pivotal role of small GTPases in intracellular signaling and the maintenance of cellular homeostasis, they are frequently targeted by bacterial pathogens in order to promote the infection 2 . An intriguing example in this respect is the bacterium Salmonella enterica. This bacterial pathogen is composed of a number of serovars that can infect diverse vertebrate species 3,4 . The Salmonella enterica Typhi (S. Typhi) causes typhoid fever by infecting only humans, whereas the Salmonella enterica Typhimurium (S. Typhimurium) can infect a broad host range. The molecular basis of this host range selection is not entirely clear. Recently, however, Galán and coworkers discovered that a protein encoded by the gene gtgE in S. Typhimurium contributes to host selection 5 . Consequently, expression of gtgE in S. Typhi counteracts host restriction and enables colonization of mice, thus, broadening host specificity. The protein GtgE has been identified as a cysteine protease that targets members of the Rab-family [5][6][7][8] . In particular, action of GtgE results in specific proteolytic cleavage of Rab29, Rab32, and Rab38 in vitro. The cleavage site is located in the regulatory switch I region of these Rabs and it takes place between G59 and V60 (amino acid numbering refer to Rab32) 6 . Rab32 is involved in controlling the biogenesis of lysosome-related organelles (LRO) [9][10][11][12][13][14] . It is believed that GtgE inactivates Rab32 proteolytically and thereby prevents the delivery of antimicrobial factors that would impair Salmonella infection. Since Salmonella is taken up by phagocytosis and evades destruction in the Salmonella-containing vacuole (SCV), which also constitutes an LRO, the Rab32-inactivating activity of GtgE may be an important component of the infectious mechanism 5 .
Individual atomic structures of GtgE have been reported recently 7,8 . However, the molecular basis of the recognition of Rab32 by GtgE and the biochemical details of binding have not yet been addressed. Here we present the crystal structure of the Rab32:GtgE complex. Furthermore, by using a combination of biochemical characterization, molecular dynamics simulation, and binding studies, we demonstrate that GtgE shows exclusive selectivity for the inactive GDP-bound state of Rab32. The biochemical characterization of the GDP-state specificity of GtgE explains the previously reported necessity of Rab32:GTP inactivation by the secreted Salmonella GAP SopD2 15 .
We next asked whether GtgE is selective for Rab-proteins inthe active (GTP) or inactive (GDP) state. To this end, the Rabproteins were prepared in defined nucleotide states with GDP, GTP, or the non-hydrolyzable GTP-analog GppNHp. SDS-PAGE based monitoring of GtgE-mediated Rab-cleavage indicates that only the inactive GDP-bound, but not the active GppNHp-loaded GTPases, are processed by GtgE (Fig. 1b). Additionally, active Rab32 was purified from expression in Escherichia coli bound to GTP to~90% (Supplementary Fig. 2A). Upon incubation with GtgE, Rab32:GTP was proteolytically cleaved significantly slower than the GDP-loaded form ( Fig. 1b aand Supplementary Fig. 2B). In order to investigate whether Rab32:GTP or the Rab32:GDP produced via intrinsic GTP-hydrolysis are the target of GtgE, we separately quantified GtgE-mediated Rab32:GTP-proteolysis and GTP-hydrolysis. Rab32 has an exceptionally slow intrinsic GTPase-activity with an hydrolysis constant determined to k hydr = 6.5·10 −6 s −1 (Fig. 1c, Supplementary Fig. 2A). The rate of Rab32:GTP-proteolysis in the presence of GtgE is identical to the rate of GTP-hydrolysis with an observed rate constant of k obs,prot = 6.7·10 −6 s −1 (Fig. 1c, Supplementary Fig. 2B). This demonstrates that the proteolytic turnover of Rab32:GTP is limited by the rate of intrinsic GTP-hydrolysis. Consequently, only the inactive (GDP-bound), but not the active (GTP-bound) Rab, is the target of GtgE.
We therefore wondered whether GtgE cooperates with other enzymes in order to convert the Rab-proteins into the inactive form as required for proteolysis. Interestingly, Salmonella secretes the bacterial Rab32-GAP SopD2 that produces Rab32:GDP from Rab32:GTP 15 . Indeed, proteolysis only happens in vitro when SopD2 is added catalytically to the reaction between Rab32:GTP and GtgE ( Supplementary Fig. 3). The rate of GtgE-mediated cleavage for Rab32:GTP in the presence of SopD2 was indistinguishable from the reaction of Rab32:GDP in the absence of this GAP. This indicates a cooperative deactivation mechanism of SopD2 with GtgE for effective Rab32-cleavage.
We further note that Rab32:GDP-proteolysis in vitro does not precipitate the protein although the cleavage site is located in the structurally and functionally crucial switch I region (Supplementary Fig. 4). In fact, GtgE-cleaved Rab32:GDP could be preparatively produced and purified by size-exclusion chromatography. The protein elutes indistinguishably from the uncleaved form (Fig. 1d). We thus probed whether switch I cleavage could affect the folding stability of the G-protein. Rab32:GDP and GtgE-cleaved Rab32:GDP were subjected to thermal unfolding experiments using differential scanning fluorimetry (DSF) (Fig. 1e). The melting points (T m ) were determined from fitting the fluorescence data to Boltzmann equation to T m,FL = 52°C and T m,cleaved = 48°C for uncleaved and cleaved Rab32:GDP, thus revealing that cleaved Rab32:GDP behaves like a well-folded and stable protein. In summary, we have shown that GtgE demonstrates exclusive specificity for inactive GDP-bound Rabproteins, and that the cleaved Rab32 remains a stable monomeric protein.
GtgE has high affinity for inactive Rab substrates. We next addressed the enzymatic properties of GtgE for inactive Rab32: GDP. First, time-resolved kinetics of GtgE-mediated Rab32:GDP cleavage were performed. We processed Rab32:GDP with catalytic quantities of GtgE, separated heat-inactivated samples from different reaction time points on the SDS-PAGE, and quantified Rab32-cleavage educts by densitometric quantification of Coomassie-stained bands since traditional fluorescence-based approaches did not function in this case (Fig. 2a, b and see Supplementary Fig. 5A, B). The catalytic efficiency (k cat /K m ) of GtgE was determined to 5·10 5 M −1 s −1 . For this purpose, the observed rate constant (k obs ) was obtained from fitting the dependence of Rab32 FL -cleavage over time to a single exponential curve (Fig. 2b), followed by division of k obs by the GtgE concentration. The obtained k cat /K m value for GtgE implies a high catalytic activity, and suggests that GtgE operates at the maximum proteolysis rate even at low micromolar concentrations of Rab32:GDP. This property is indicative of a high affinity enzyme-substrate-complex having a Michaelis-Menten constant (K M ) in the micro-or nanomolar range. Determination of the K M would require Michaelis-Menten-kinetics to be conducted at micromolar to submicromolar Rab32:GDP-concentrations. However, at such low substrate concentration, the quantification of cleavage educts and products by Coomassie-stained SDS-PAGE gels would be challenging. As an alternative, we aimed to determine the dissociation equilibrium constant (K D ) between Rab32:GDP and a catalytically inactive GtgE-mutant as a surrogate for K M . The amino acid residue C45 has previously been shown to be important for the proteolytic activity of GtgE, and likely constitutes the active site nucleophile of this cysteine protease 7,8 . We therefore confirmed that GtgE C45A indeed has no proteolytic activity on Rab32:GDP ( Supplementary  Fig. 6), allowing us to determine the K D by two independent approaches. First, we subjected Rab32:GDP (20 µM) to isothermal titration calorimetry (ITC) with GtgE C45A (200 µM), revealing a dissociation constant of K D,ITC = 96 ± 31 nM (Fig. 2c). The determined stoichiometry confirms that GtgE and Rab32:GDP form a heterodimeric 1:1-complex. In parallel, we utilized analytical ultracentrifugation (aUC) of Rab32:GDP (200 nM) covalently labeled at thiol groups with Atto488-Maleimide (Rab32: GDP fluor. ), and quantified complex formation with increasing concentrations of GtgE C45A by the difference in Rab32:GDP fluor. sedimentation (Fig. 2d). Assuming a 1:1 binding stoichiometry, the K D of the interaction is determined to K D,aUC = 118 ± 29 nM which is in good agreement with the ITC-experiment. Thus, our orthogonal approaches determined the K D for the interaction between Rab32:GDP and the inactive GtgE C45A mutant to be approximately 100 nM. Assuming that the K D equals the Michaelis-Menten constant K m for the enzymatic process, we obtain a turnover rate(k cat ) of k cat = 0.05 s −1 by multiplying k cat /K m (Rab32:GDP, GtgE) = 5.1·10 5 M −1 s −1 with the K D . In summary, as indicated bya low K D and a moderate k cat , our  Fig. 2A). The proteolysis was monitored by Coomassie-stained SDS-PAGE ( Supplementary Fig. 2B). Both data sets are presented as means of technical replicates ( ± SEM; n = 2). The data were fitted to a single exponential function, yielding the observed rate constants k obs,prot and k hydr .  Fig. 8) 16 . The X-ray structures of the complexes were solved by the molecular replacement technique using the coordinates of Rab32:GppCH 2 p (PDB ID: 4CYM 17 ) and truncated GtgE (PDB ID: 4MI7 7 ) as appropriate search models ( Fig. 3a) (see Supplementary Table 1 for data collection and refinement statistics). The average B-factor of the full-length complex is higher than for the truncated complex, though the entire electron density is well defined for both structures (Supplementary Fig. 9). Due to the higher resolution, the truncated Rab32 18-201 :GDP:GtgE 21-214,C45A -complex is discussed only and referred to as Rab32:GDP:GtgE C45A hereafter. The complex shows the general features of the previously determined individual subunits of Rab32 and GtgE 7,8,17 . The Cαatoms of the Rab32-molecules superimpose for 121 amino acids (excluding the structurally-flexible switch-regions, switch I aa48-61 and switch II aa82-99 ) with an RMSD of 0.52 Å 18 . In case of GtgE, 97 Cαs were aligned with an RMSD of 0.65 Å (excluding the previously unresolved amino acids . Rab32 is composed of a central β-sheet comprising six β-strands (β 1R -β 6R , where subscript R denotes Rab32) and five α-helices (α 1R -α 5R ) (Fig. 3a, b; Supplementary Fig. 1). However, no electron density is observed for the helix α 5R , which is usually present in all small GTPases. GtgE is composed of a six-membered central antiparallel β-sheet that is perpendicularly arranged in a tip-totip manner to the β-sheet of Rab32. The GtgE-β-sheet is flanked by five α-helices of which the two N-terminal α-helices and the three C-terminal helices are on opposite ends of the sheet structure ( Fig. 3c) 8 . The amino acid regions 23-44 and 60-82 appear to be folded within the active enzyme structure only after complex formation with Rab32, as these peptide stretches occupy significantly different positions in comparison to the GtgE-apostructure (Fig. 3c).
In the complex structure, GtgE can be envisioned as a trident poking into Rab32 with three spike-like protrusions (Fig. 3a, d). Spike 1 comprises amino acid region 38-45 in conjunction with region 76-82. Spike 2 is represented by the β 2G -β 3G loop (where subscript G denotes GtgE), and penetrates deeply into the switch I-switch II cleft of Rab32. Thereby, spike 1 and 2 form intimate interactions with the switch I structure that is also the site of proteolytic cleavage. Spike 3 is formed by the α 4G -helix and a Cterminal loop extension. The shallow cleft between spike 2 and 3 is filled by the switch II region of Rab32.
Binding of GtgE to Rab32 does not result in major structural changes outside of the switch-regions when compared to the structure of Rab32:GppCH 2 p (Fig. 3b) 17 . Since GtgE is specific for the inactive GDP-bound Rab32 ( Fig. 1), this suggests that the contacts within switch-regions undergo large-scale conformational changes upon GtgE-binding. The switches are reminiscent of, but not identical to the conformations observed in Rabproteins in complex with the GDP-state-specific interaction partner GDI 19,20 (Supplementary Fig. 10). Spike 2 of GtgE acts like a wedge that drives the switch-regions further apart than observed in Rab:GDI-complexes. On the other hand, binding of Rab32 to GtgE does not affect the secondary structure elements of the protease (Fig. 3c) 7,8 . Rather, flexible loops previously unrecognized inthe GtgE-structure, become ordered in the Rab32:GtgE complex. In particular, the N-terminus (amino acids , including the position of the active sitecysteine (here the C45A substitution) is now resolved, in addition to the loop regions β 2G -β 3G (i.e., spike 2) and α 4G -β 6G (i.e., spike 3).
The complex interface buries an average solvent exposed surface area of 1535 Å 2 , as determined with the 'Protein interfaces, surfaces and assemblies' web service (PISA) 21 . It is mainly composed of the switch-and β 2R -β 3R -regions of Rab32 and the three spikes of the GtgE-trident (Fig. 3d, e). A large number of hydrophobic and polar interactions contribute to the complex interface, providing a potential molecular basis for the observed high affinity between GtgE and Rab32:GDP (Fig. 3e). In particular, the Rab32 residues adjacent to the cleavage site between G59 R and V60 R are intimately involved in interacting with the deep spike 1/spike 2 cavity of GtgE. The switch I residue I58 R forms the center of a hydrophobic cluster and interacts with W46 G , L81 G , C86 G , I141 G , L150 G , and I202 G , whereas D61 R from the protease recognition sequence forms a specific interaction with the guanidinium group of R142 G . Three further ionic interactions K42 R -E148 G , R55 R -D82 G , and R93 R -D182 G are likely to contribute to binding and enhancing specificity. The interacting polar network between the catalytic triad, C45 G , H151 G , and D169 G , is formed in the Rab32:GDP: GtgE C45A -complex once C45 G is modeled to the active site cysteine nucleophile ( Supplementary Fig. 11) [6][7][8] . Furthermore, the oxyanion hole can be unambiguously assigned to the peptide   Table 2). Interactions highlighted in dashed lines: hydrogen bonds (blue), saltbridges (cyan blue), hydrophobic (gray). L: loop, α: α-helix, β: β-sheet ( Supplementary Fig. 1). Magenta box: residues tested by mutations in a GtgEmediated cleavage assay (Fig. 4c, d). backbone NH-groups of C45 G and N44 G as well as the side chain Q33 G N ε2 . (Fig. 3f), while the thiol-group of C149 G potentially acts as the Brønsted acid to generate a proper N-terminal leaving group of V60 R (Fig. 3f).
In summary, GtgE interacts tightly with Rab32 via three binding platforms referred to as spike 1-3 that accommodate the switch regions. In particular, the resolved structure reveals that the cleavage site of Rab32 is pulled into a deep cavity formed by spike 1 and 2.
Confirmation of the interface of the Rab32:GDP:GtgE complex. The crystal structure of Rab32:GDP:GtgE C45A -complex suggests that many side chains within the switch region contribute to the interaction between the GTPase and the protease. In order to verify the significance of individual amino acids in complex formation, we investigated Rab32 point mutants for the ability to form a complex with GtgE by yeast-two-hybrid (Y2H) experiments using the full-length proteins (see Fig. 3e for positions of mutants). We first established an Y2H-based interaction assay by combining the proteolytically inactive mutant C45A G of GtgE as prey and the Rab32-subfamily together with Rab23 as bait (Fig. 4a). Complex formation is detected by growth on histidinelacking media only for the C45A G mutant, but not the wild-type GtgE, indicating an interaction exclusively between the GTPase substrates and the proteolytically inactive GtgE. Consistent with our biophysical analyses we observe that this interaction is of high affinity for Rab32 since growth was also observed under more stringent conditions in the presence of 10 mM 3′-AT (Fig. 4a). The interaction with Rab38 is also detected, albeit the positive signal is more susceptible to the presence of 3′-AT, indicating a weaker interaction. Rab29 is self-activating in the Y2H assay (left panel) and thus cannot be used for monitoring GtgE-binding.
Next, a library of Rab32 switch-region mutants was explored by Y2H-based interaction studies in order to verify the interface observed in the complex crystal structure. In the switch I region, the amino acid mutants T52A R (P3), I58 R (P2), G59 R (P1), D61 R (P2′), and F62 R (P3′) show a strong growth defect, confirming that they are involved in forming the Rab32:GDP:GtgE C45Acomplex (denominations in parentheses correspond to the Schechter-Berger-nomenclature for positions in the protease substrate cleavage site) (Fig. 4b, top). In particular, mutation of I58 R , identified as a hub for hydrophobic interaction in the crystal structure, does not support growth in Y2H, thereby supporting the complex interface and the significance of this residue.
We also conducted a Rab32 switch II mutational analysis by employing the Y2H-screening (Fig. 4b, bottom). However, none of the mutants directly involved in the complex interface demonstrate a detectable growth defect, indicating that switch II mutations may be tolerated due to the high complex affinity. In contrast, the mutants W80A R and D81A R in the interswitchregion show severe growth inhibition. The side chain of W80 R is a constituent of the important Rab hydrophobic triad that is relevant for binding to most interaction partners 22   W80 R side chain interacts with F62 R , the W80A R mediated growth defect can be explained by an influence on the positioning of the switch I F62 R side chain that is important for GtgE binding (Fig. 4b, top). In addition to the Y2H-assays, we probed the structure-activity relationship of the complex with mutants in all interacting regions (Rab switch-regions and GtgE-spikes, Fig. 3e) by studying GtgE-mediated Rab32-proteolysis (Fig. 4c, Table 3). All designed mutations in Rab32 were either constructed based on the structural analysis of the Rab32:GtgE complex (K43A R , D61A R , R93A R ) or inspired by the alignment between Rab32 and the non-GtgE substrate Rab23 (A56K R , L64E R , R87E R ) (Fig. 4c, Supplementary Fig. 1). The tested Rab32-mutants display at least a 10-fold activity reduction, supporting their significance for the substrate recognition and complex formation. Mutants D61A R and L64E R show a 30-fold decrease in activity with GtgE, emphasizing the importance of the salt bridge between D61 R and R142 G . The L64E R -mutation most likely induces a steric hindrance between F34 G and L41 G , which further give insights into the Rab-substrate restriction. The mutants K43A R and R93A R abrogate salt bridge formation with E148 G and D182 G , respectively, hence confirming the relevance of the corresponding ionic interaction for complex formation. In addition, the decrease in activity for the R87E R mutant supports the importance of the arginine side chain in binding to GtgE spike 3 via a backbone H-bridge with K194 G . The mutant A56K R presumably causes steric clashes with the side chains of L150 G and P143 G that form the basis of the important spike 2 loop of GtgE (Fig. 3e), thereby corroborating the relevance of this region for recognition of Rab32 by the protease.
Moreover, all GtgE-mutants except for E148A G show a significant drop in activity (Fig. 4d). In particular, Q33A G decreases the activity 200-fold, confirming its key role in the oxyanion hole. The importance of the D61 R -R142 G salt bridge as a central interaction site mentioned before is once more confirmed using the R142A G mutation. The C149A G mutant reduces the activity by a factor of 2. Its localization near the active site suggests a role as a proton donor for the newly formed Nterminus during catalysis. The salt bridge formed by D182 G with the switch II residue R93 R shows a greater impact in the complex formation than the one constituted by E148 G with K43 R . The residue K186 G is indirectly involved in the complex interface, stabilizing spike 3 by forming two intramolecular polar contacts with Q178 G and E153 G near the catalytic base H151 G , rendering K186 G an important organizing element in the switch II/spike 3interface.
The Rab32:GDP:GtgE complex structure was also used to obtain insights into the Rab-selectivity. On the basis of the amino acid sequence alignment between the GtgE-substrates Rab32, 29, 38 and the non-GtgE-substrate Rab23, we identified four amino acids that are located in the protein-protein interface and may be incompatible with binding of Rab23 to the protease (Supplementary Fig. 1). Therefore, we tested whether Rab23 is converted into a GtgE-substrate by introducing the Rab32 analogous substitutions K40A, E48L, Q50V, and E70R in vitro. However, this Rab23 mutant was not cleaved by GtgE, indicating a multifactorial selection mode that cannot be reduced to a few specific amino acid side chains (Supplementary Fig. 14).
Thus, the mutational analysis confirms the Rab32-GtgEinterface observed in the complex crystal structure.
Insights into the Rab32:GDP specificity of GtgE. We next addressed the molecular basis for the strict preference of GtgE for the inactive GDP-bound Rab32 by analyzing the Rab32:GDP: GtgE C45A -complex crystal structure (Fig. 1b). Three potential hubs on Rab32 were identified that could hypothetically contribute to the targeting of GtgE to the inactive GTPase: hub 1 is constituted by the switch II residue E86 R , hub 2 by the switch I amino acid Y54 R , and hub 3 by the switch II side chain of F88 R together with K194 G and E114 G (see below, Fig. 5a, b).
Close inspection of a structural superposition of the complex with active, GppCH 2 p-bound Rab32 suggests a potentially important role of the hub 1 side chain of E86 R in activity-state discrimination by GtgE 17 . In the GtgE-bound form, the side chain of E86 R protrudes into a Rab32-cavity that would be occupied by the γ-phosphate of GTP or GppCH 2 p in the active state (Fig. 5a). Consequently, the conformational rearrangement of switch II by GtgE-binding would result in a steric and electrostatic clash of E86 R with the γ-phosphate, thereby making binding of Rab32: GTP to GtgE incompatible. In order to test this hypothesis, we analyzed the GtgE-mediated proteolysis of the E86A R -mutation of Rab32 with the full-length proteins. Surprisingly, the E86A Rmutation did not change the activity-state preference of GtgE since only GDP-bound but not GppNHp-loaded Rab32-E86A R was a substrate as monitored by SDS-PAGE based proteolysis (Fig. 5c). Hence, additional factors contribute to the Rab32:GDP specificity of GtgE.
As a further potential decisive element of the nucleotide discrimination, we focused on the switch I amino acid Y54 R as hub 2. The Y54 R -position is conserved among small GTPases and forms a polar interaction with the GTP γ-phosphate in the active state. In the Rab32:GDP:GtgE C45A structure, Y54 R is turned away from the phosphates, pointing towards the guanine base. This observation could imply that Y54 R is unable to support GTPcoordination in the Rab32:GtgE complex and therefore does not permit binding of GtgE to Rab32:GTP. Thus, the full-length mutants Y54F R and Y54A R were loaded with GDP or GppNHp and tested for their substrate properties towards GtgE WT (Fig. 5d and Supplementary Fig. 15A). Surprisingly, we find that again only the inactive GDP-bound Rab32-mutants were cleaved, and even after long incubation time no hydrolysis could be detected for the GppNHp-bound mutants, rendering the Y54 R unimportant in the nucleotide-state selection.
In order to obtain additional insights into the nucleotide-state specificity of GtgE, we performed atomistic molecular dynamics (MD) simulations. We first considered calculating the structure of a hypothetical complex between active Rab32:GTP and GtgE. Therefore, the only currently available structure of active Rab32 bound to the non-hydrolysable GTP-analog GppCH 2 p was taken from its complex with the effector protein VARP (PDB ID: 4CYM) 17 . Replacement of the structure of the GppCH 2 p-bound Rab32 from its complex with VARP resulted in steric clashes of Rab32 with GtgE despite the global structure of Rab32 closely resemble each other in the two X-ray structures ( Supplementary  Fig. 16). Instead of probing the formation of Rab32:GTP:GtgE complex by extensive free energy calculations, we computationally replaced GDP with GTP in the nucleotide binding pocket of Rab32 and probed the structural determinants for GtgEdissociation from the complex. This approach is justified since small GTPases can form complexes with their binding partners in both activity states albeit with differing affinities 23 . It is common, however, that protein-protein affinities are mainly determined by the rate of complex dissociation (k off , 10 −5 -10 3 M −1 s −1 ) rather than by the second-order rate constant of complex association (k on , 10 6 -10 7 M −1 s −1 ). The MD-simulation, therefore, mimics the initial molecular steps of GtgE dissociating from Rab32:GTP, and aims to probe determinants that would contribute most to dissociating the Rab32:GTP:GtgE complex once formed. Notably, these computations do not address the rate of complex association, since they would require extensive free energy simulations, and are outside the scope of the present study.
A principal component analysis based on the MD-trajectories of the Rab32:GtgE complex confirms that the Rab32:GtgE complex is preserved in the GDP-state. However, replacement of the nucleotide for GTP alters the complex behavior and reveals loop-motions of Rab32 and GtgE into opposite directions ( Supplementary Fig. 17). Consequently, the analysis suggests that Rab32 is binding to GtgE in the GDP-state but dissociates from the protease in the GTP-state.
In our ca. 200 ns MD-simulations, the global structure and dynamics of both Rab32 and GtgE remain similar, but we observe an increase in distance between F88 R and K194 G in the GTPbound complex (Fig. 5e, bottom and Supplementary Fig. 18A). The MD-simulations reveal that a third putative hub (hub 3) in the complex interface between switch II and spike 3 could be involved in the Rab32-GtgE recognition, comprising residues F88 R , K194 G , and E114 G (Fig. 5b). In the MD-simulations with GTP, we observe an increase in the F88 R -K194 G distance, which could lead to a weaker Rab32-GtgE interaction, consistent with the observed lack of Rab32:GTP-binding to GtgE in the in vitro experiments (Supplementary Fig. 18B). In order to probe the Right: van-der-Waals contacts of hub 3 indicated by yellow dashed lines for the GDP (top) and the GTP-complex (bottom), respectively (see Supplementary Fig. 18). c, d Processing of the hub mutants hub 1 E86A R (c), and hub 2 Y54A R (d) by GtgE in the GDP and GTP states, analyzed with a gel shift activity assay (see also Supplementary Fig. 15A). e Selected atom pair distances in a GDP-bound and hypothetical GTP-bound Rab32:GtgE complex from 190 ns MD-simulations (left panels) and histograms of respective distance occurrences (right panels), with an ion-pair distance threshold (< 4.5 Å) highlighted in gray. Top: E86 R and R87 R form a stable ion-pair after ca. 120 ns in the hypothetical GTP-bound complex (in green), initiated by an electrostatic repulsion from the γ-phosphate of GTP (see also Supplementary Fig. 18C). In contrast, no stable ion-pair between E86 R and R87 R is observed in the GDP-bound structure (in gray). Bottom: Distances between F88 R of Rab32 and K194 G of GtgE from MD-simulations of the GDP-(in gray) and GTPbound (in green) Rab32:GtgE complexes. In the GDP-bound form, K194 G forms an interaction with F88 R , while in the GTP-bound complex, K194 G dissociates from F88 R , which may trigger the dissociation of GtgE from Rab32 (see also Supplementary Fig. 18A). The MD-simulations suggest that F88 R may form a decisive element for the GDP-state preference of GtgE toward Rab32. f Mutations in Rab32 hub 3 change GtgEs nucleotide-state selectivity leading to cleavage of Rab32:GppNHp. Rab32 F88G loaded with GDP or GppNHp was treated with cleared E. coli lysate (or 1:100 diluted lysate) overexpressing wild-type GtgE and analyzed by an gel shift assay (see also Supplementary Fig. 18B) function of F88 R in Rab32, we replaced this residue with a glycine (F88G R ), and re-initiated the MD-simulations. Interestingly, the MD-simulation of the F88G R demonstrate that the GTP-and GDP-states behaves more similar in this mutant ( Supplementary  Fig. 18C). Therefore, to validate the significance of hub 3 for the GtgE selectivity towards GDP over GTP, we produced the Rab32 F88G R -mutant. To this end, we purified the full-length Rab32 wild type and the F88G R -mutant with GDP or GppNHp, challenged it with cleared E. coli cell lysates containing overexpressed full-length wild-type or GtgE-mutants (E114A G , K194A G , E114A G /K194A G ), and analyzed Rab32-proteolysis using SDS-PAGE based gel shift ( Fig. 5f and Supplementary  Fig. 15B). Interestingly, although Rab32 F88G R generally becomes a worse substrate for GtgE, the F88G R mutation indeed results in a loss of the strong GDP-state preference of GtgE for Rab32 (Fig. 5f). The decrease in GtgE-efficiency for Rab32 F88G R is expected due to the involvement of F88 R in the complex interface. However, the F88G R -mutant is now cleaved both in the GDPand GTP-states in contrast to the wild-type GTPase (Fig. 5f). Thus, the F88 R -moiety is obviously important in the switch II rearrangement during complex formation, as indicated by the MD-data. In contrast, the alanine substitutions of GtgE E114 G and R194 G involved in hub 3 show opposite effects and are neutralizing each other in the double mutant when combined with Rab32 F88G R ( Supplementary Fig. 15B, right). The increased space in the F88G R mutant may be required to form a catalytically competent Rab32:GTP:GtgE complex. We speculate that E114A G decreases the proteolytic activity emphasizing its vital role as electrostatic anchor for K194 G . When not fixed by the carboxylate function of E114 G , the ε-amino group of K194 G is free to establish other intramolecular salt bridges or polar contacts with the neighboring D193 G or Y195 G , and could thereby disturb the organization of the hub 3 interface. The Rab32 switch II interaction with spike 3 may thus be a crucial part in the nucleotide selection.
In conclusion, we provide insights into a multifactorial selection mechanism since no single tested hub was able to abolish the nucleotide substrate preference of GtgE entirely. Nevertheless, the computationally derived and experimentally verified F88G R -mutation results in the proteolysis of both active and inactive Rab32.

Discussion
In this work, we have biochemically and structurally characterized the Rab32:GDP:GtgE complex. We have established that GtgE specifically acts on inactive, GDP-bound Rab32.
Furthermore, based on combined atomistic molecular simulations and in vitro experiments, we could demonstrate that the F88 R position in the switch II regulatory region of Rab32 contributes to a multifactorial activity-state discrimination mechanism by GtgE and that it forms a basis for the GDP-state specificity of the protease.
Only two other proteins are known to have exclusive affinity for inactive-state Rabs. Here, the functionally and structurally related regulatory factors Rab escort protein (REP) and GDI are involved in the prenylation and membrane delivery or membrane cycling of Rabs, respectively, and interact with low nanomolar affinity with Rab:GDP [24][25][26] . In particular, the physiological role of GDI is to effectively solubilize lipidated Rabs from intracellular membranes after the protein has returned to the inactive form. Even though the complexes of Rab-proteins with REP and GDI have been extensively investigated on a structural level, the molecular basis for the exclusive interaction with inactive Rabs remains elusive 20,27,28 . Our structural and molecular dynamics analysis of the Rab32-GtgE interactions, however, revealed that the side chain of F88 R in Rab32 forms an important structural determinant for discriminating the active and inactive Rab GTPase.
It is interesting to note that the Rab-recruitment to intracellular membranes is dependent on membrane-localized GEFs. Therefore, Rab-targeting and activation are directly coupled 29 . Consequently, GEF-mediated Rab32 recruitment to the SCV would result in the creation of active Rab32:GTP that is not a GtgEsubstrate and consequently would not be subject to proteolysis. Interestingly, Salmonella secretes another protein that ensures inactivation of Rab32:GTP, making Rab32:GDP accessible to GtgE. During infection, the Salmonella Rab32-GAP SopD2 is released and catalyzes the conversion of Rab32:GTP to Rab32: GDP 15 . Apparently, Salmonella causes effective GtgE-mediated Rab32-cleavage by ensuring the prior conversion of active into inactive Rab32.
The question arises as to why Salmonella has evolved with a GDP-state specific Rab-protease making it dependent on a second secreted protein (i.e. SopD2). We hypothesize, however, that the presence of SopD2 ensures time-efficient access to GDPloaded Rab32 since the latter would be blocked by tightly binding to GDI. We therefore suggest the following order of events: The Rab-protein is targeted to and activated at the membrane by a corresponding GEF 29 . This successive pathway of GEF-mediated Rab32 activation/recruitment and SopD2-mediated Rab32-inactivation results in direct access of GtgE to Rab32 (Fig. 6). In the hypothetical case where GtgE was to attack Rab32:GDP without prior action of GEFs and SopD2, it would need to directly The resulting SCV-bound GTP-loaded GTPase is then rapidly converted by the action of the GAP SopD2. Eventually, Rab32:GDP is cleaved quickly by GtgE. GEF guanine-nucleotide exchange factor, GDI GDP-dissociation inhibitor compete with GDI for Rab32-binding in the cytosol. Since GDI and GtgE occupy the identical binding site on Rab32 and since the interaction of Rabs with GDI are of high nanomolar affinity, GDI would effectively shield Rab32 from GtgE 20,26,27 . The high affinity of the Rab32:GDP-GtgE interaction (indicated by a nanomolar K D , K D, Rab32:GDP:GtgE = 100 nM) would not be sufficient to displace GDI (K D,Rab:GDI < 5 nM) in the cytosol and would possibly not allow time-efficient Rab32-cleavage. In contrast, the combination of SopD2 and GtgE Salmonella at the SCV membrane during infection guarantees fast Rab32inactivation as a results of a series of rapid enzymatic steps (Fig. 6): Rab32:GDP is likely recruited to the SCV via a GEF (BLOC-3) from its cytosolic complex with GDI promptly. The resulting pool of active Rab32:GTP is then immediately targeted by the GAP SopD2, leading to GTP-hydrolysis and the production of membrane-localized Rab32:GDP. Finally, GtgE is attacking Rab32:GDP at the SCV and the proteolyzed GTPase is formed quickly. Therefore, the cooperation of SopD2 and GtgE may overall be the most efficient way for quantitative Rab32proteolysis (Fig. 6). An efficient Rab-inactivation therefore appears to be crucial for the progressing SCV maturation.
Together with previous findings, our present results contribute to the understanding of the order of molecular events at the SCV: the maturation of the SCV is dependent on the small GTPases Rab7 and Rab9 [30][31][32] . After activation, Rab9 may recruit the Rab32-GEF BLOC-3 to the surface of the SCV 33,34 . BLOC-3 subsequently recruits Rab32 to the SCV from its complex with GDI and activates the G-protein by loading with GTP. In order to stably establish the infection, Salmonella counteracts Rab32dependent membrane pathways with SopD2 and GtgE, leading to Rab32 conversion into the GDP-state and irreversible proteolysis in the switch I region, respectively 5,15 .
It is interesting to note that despite of Rab32 recruitment to the SCV, the protein is not stably accumulated at the membrane in Salmonella Typhimurium (containing the GtgE and SopD2 proteins), indicating that GtgE-cleaved Rab32:GDP is lost from the vacuole in an active or passive process 5 . Likely, the Rab32proteolysis is not affecting intrinsic membrane affinity since the C-terminal lipids remain unaffected, making a direct GtgEmediated Rab membrane release improbable. Since our results demonstrate that cleaved Rab32 remains a stable protein that alsocan be purified in vitro, active processes to remove the Rabprotein from membrane may be ongoing. Here, ubiquitination in conjunction with proteasomal degradation or membrane extraction by GDI are conceivable mechanisms. Cleaved Rab32 may still interact with and be extracted by GDI, leading to effective removal of the GTPase. GDI is only weakly interacting with the switch I regions of Rab-proteins and therefore presumably tolerating the proteolytic modification 20 .
We speculate that once extracted by GDI, the GtgE-cleaved Rab32 is presumably not recruited by BLOC-3 to the SCV again. Although the 3D-structure of BLOC-3 is unknown, its sequence similarity to components of the Ypt7-GEF-complex Mon1/ Ccz1 suggests that the interaction with Rab32 happens in a mechanistically identical manner 35 . In contrast to the Rab-GDI interactions, the switch I loop intimately interacts with the GEF. Consequently, Rab32-proteolysis is potentially incompatible with activation by BLOC-3, and thus re-recruitment of GDI-bound Rab32 would be prohibited.
In conclusion, GtgE is a GDP-state specific Rab32-protease having a high affinity for its substrate. Structural studies and MDsimulations identify an amino acid hub that contributes to the selective binding of the inactive state of Rab32. The analysis provides insights into the order of protein interactions involved at the SCV and imply a requirement for the Salmonella GAP SopD2 in order to permit GtgE-mediated Rab32-proteolysis. It will be interesting to investigate how the GtgE-processed Rab32 contributes further to the infection process.

Methods
DNA cloning. For all cloning tasks, either XL1 blue or Mach1 E. coli strains were used. Site and ligation independent cloning (SLIC) approach was applied for all molecular cloning. The gtgE gene from Salmonella enterica serovar Typhimurium str. SL1344 (CBW17091.1) was synthesized with an optimized sequence for E. coli codon usage provided in a standard cloning vector (Life Technologies). The sopD2 gene (WP_001145561.1.) was amplified from genomic DNA from the same strain. Both Salmonella genes were subcloned for heterologous expression into a modified pMAL-c2X vector (Addgene ID 75286) harboring an N-terminal His 6 -MBP tag (MBP: maltose binding protein) fused through a TEV protease cleavage site between the 3′-NdeI and 5′-HindIII restriction sites. Additionally, truncated Rab32  and GtgE 21-214,C45A were generated in the pMAL vector as described above and used for protein crystallization and structural studies, only. Recombinant protein expression and purification. All proteins were recombinantly expressed in E. coli BL21(DE3)-RIL chemically competent cells (Agilent) growing in LB broth supplemented with ampicillin. The expression culture was inoculated with 2-5% (v/v) from a pre-culture and grown at 37°C shaking at 180 r. p.m.. Expression was induced at an OD 600 of 0.5 by addition of 0.3 mM IPTG (final). After induction the culture rested at 4°C for 30 min and was grown 16-20 h at 22°C and collected at 5900 × g for 25 min. The pellet was washed with phosphate-buffered saline (1xPBS) and collected again under same conditions. Cell pellets were snap frozen in liquid nitrogen and stored at −20°C until further use. Pellets of GtgE and SopD2 proteins were resuspended in 10 mL Buffer A (50 mM Hepes pH 7.5, 500 mM NaCl, 2 mM β-mercaptoethanol (BME)) for each gram cells and lysed mechanically with one passage through a fluidizer at 2.0 kbar (Constant systems). Lysate was cleared by centrifugation 48,000 × g for 30 min and the supernatant was applied to a 5 mL Ni 2+ -immobilized metal-affinity chromatography column (IMAC) column (HiTrap, GE Healthcare) previously equilibrated with Buffer A. The column was washed with 6 column volumes (CV) 5% Buffer B (Buffer A supplemented with 500 mM imidazole) and protein of interest was eluted applying a gradient up to 50% Buffer B in 10 CV. Fractions containing the His-MBP tagged protein of interest were pooled and dialyzed over night at 4°C against Buffer C (20 mM Hepes pH 7.5, 50 mM NaCl, 2 mM BME) and subjected to Tobacco Etch Virus (TEV)-protease cleavage. Cleaved protein mixture was applied a second time on an IMAC column to remove the TEV protease and His-MBP tag. The highly pure protein was collected in the flow through, concentrated, snap frozen and stored at −80°C.
Purification of Rab GTPases: The procedure was the same as described for the GtgE purification above except all buffers were supplemented with 1 mM MgCl 2 and 10 µM GDP and after cell lysis 1 mM phenylmethylsulfonyl fluoride (PMSF; final) was added before centrifugation. To reach homogeneity after the IMAC Rab GTPases were further subjected to size-exclusion chromatography (SEC) on a Superdex 75 (16/600) (GE Healthcare) pre-equilibrated with SEC-buffer (20 mM HEPES, pH 7.5, 50 mM NaCl, 1 mM MgCl 2 ,10 µM GDP, and 1 mM DTT). Fractions containing monomer were concentrated, snap frozen and stored at −80°C. The identity of all proteins was verified by mass spectrometry using an electronspray-ionization mass spectrometer coupled to a liquid chromatography (LC-ESI-MS; LCQ fleet; Thermo Scientific).
For the in vitro complex formation, 8 µM Rab GTPase was incubated with 8 µM GtgE C45A in SEC-buffer for 30 min at room temperature. For analysis 50 µL of the mixture was subjected to an analytical gel filtration (Superdex 75 13/30, GE Healthcare) pre-equilibrated with SEC-buffer with a flow of 0.5 mL/min recording the absorption at 280 nm. As negative controls, the respective monomers were analyzed identically.
Yeast-Two-hybrid interaction studies. For the Y2H assay, pLexA-GTPase plasmids (bait) encoding for full-length Rab proteins were transformed with MAT-a yeast strain Y187 adopted from a previous method 36 and plated on selection dropout medium lacking tryptophan (SD-W) 37 . In brief, 700 µL of stationary yeast culture was collected at 375 × g for 2 min and vortexed with 100 µL sterile one-step buffer (0.2 M lithium acetate, 40% PEG 3350, and 100 mM DTT). Subsequently 100-500 ng of respective plasmid DNA was added and incubated at room temperature for 15 min before heat shock was applied for 30 min at 45°C. The pGAD-gtgE C45A plasmids (prey) encoding for inactive full-length GtgE protein were transformed the same way with MAT-α S. cerevisiae L40ΔGal4 and were grown on SD-L plates (lacking leucine) at 30°C for 3-4 days. For yeast mating, single colonies of a prey and a bait clone were grown separately in appropriate selection medium for 2 days while shaking. To 100 µL YAPD media 37 in a 96-well plate (Sarstedt) 50 µL of bait and prey culture were added sequentially and incubated 22-24 h at 30°C with 180 r.p.m. Cells were resuspended and 5 µL were spotted on SD-LW plates as mating control and SD-LWH (lacking histidine) for phenotypic read out. Plates were analyzed after 3-4 days of incubation at 30°C. 3-Aminotriazol (3′-AT, Sigma Aldrich), an inhibitor of the histidine biosynthesis, was added to SD-LWH agar plates in the range of 0-10 mM probing the interaction intensity qualitatively.
Isothermal titration calorimetry. Interaction studies by isothermal titration calorimetry (ITC) were conducted on a ITC200 microcalorimeter (MicroCal). Measurements were performed in SEC-buffer at 37°C. Rab32 FL :GDP was diluted to 20 µM and GtgE FL , C45A was introduced to the injection unite with 200 µM. Injection volume was set to 1.5 µL per injection and the heat power was recorded over time until binding saturation was obtained. The binding isotherms were integrated, corrected for the offset, and the data were fitted to a one-site-binding model using the ITC analysis software provided by the manufacturer (MicroCal), yielding the equilibrium-dissociation constant (K D ).
Analytical ultracentrifugation. Interaction studies by analytical ultracentrifugation (AUC) were conducted in a Beckman ProteomeLab XL-A analytical ultracentrifuge (Beckman Coulter) equipped with a fluorescence detection system (Aviv AU-FDS; Aviv Biomedical) as described previously 38 . A constant concentration of 200 nM Rab32 FL -Atto488 in SEC-buffer was incubated with varying molar ratios of GtgE FL,C45A as indicated for 2 h in the dark at room temperature, ensuring equilibrium conditions prior to measurement. The data were analyzed with SedFit software 39 . The sedimentation coefficients and molecular weights of the present species were obtained by solving the Lamm equation. The fraction of labeled Rabproteins bound to GtgE C45A was plotted against the GtgE C45A concentration and data were fitted with a hyperbolic function for calculation of the K D .
Nucleotide loading of GTPases. Nucleotide exchange of small GTPases with GDP or GTP (Carbosynth) was performed in SEC-buffer supplemented with 10 mM EDTA and a 40-fold molar excess of nucleotide over protein and incubation at 15°C for 5-16 h before performing size-exclusion chromatography (Superdex 75 16/ 60; GE Healthcare). Nucleotide exchange with the non-hydrolyzable analog GppNHp (Jena Bioscience) was performed by addition of equimolar amounts to the GTPase with 5 U Antarctic Phosphatase (NEB) per mg protein in nucleotide exchange buffer (50 mM Tris HCl pH 8.0; 200 mM (NH 4 ) 2 SO 4 ; 10 µM ZnCl 2 ). Accidentally precipitated protein was removed by centrifugation and a buffer exchange was performed on a NAP 5 desalting column (GE Healthcare) according to the manufacturer's instructions with SEC-buffer containing 1 µM GppNHp. Protein containing fractions were pooled, concentrated, snap frozen, and stored at −80°C. Nucleotide loading efficiency was tested by ion-pairing reversed-phase high performance chromatography (RP-HPLC; Prontosil C18, Bischhoff Chromatography) in 50 mM potassium phosphate buffer pH 6.6; 10 mM tetrabutylammonium bromide and 12% acetonitrile (v/v). Protein samples (25 µM, 30 µL) were heat precipitated at 95°C for 5 min and cleared by centrifugation for 10 min at 48,000 × g. Supernatant was subjected to chromatographic separation and nucleotide peaks were integrated and normalized to the total amount of nucleotides detected set to 100%. Nucleotide retention times were determined with the respective nucleotide standard in a separate run.
Kinetics of Rab32 cleavage. Cleavage of Rab GTPases by GtgE was monitored by SDS-PAGE gel shift assay. Typically 8 µM of Rab FL was cleaved by 8 nM GtgE FL in SEC-buffer supplemented with 10 µM GDP, 10 µM GTP or 1 µM GppNHp respectively at 25°C. At different time points, 10 µL samples (ca. 2 µg) were taken an the reaction was quenched immediately with pre-heated Laemmli buffer and 3 min boiling at 95°C. For quantification purpose 2 µM, MBP was added as internal standard into the reaction mixture prior to start of the experiment. After Coomassie staining (Brilliant blue R-250, Thermo Scientific), band intensities were determined by densitometry from gray values with image studio lite software (LI-COR). Rab-bands were normalized with the respective MBP-band intensity and plotted over time. The apparent rate constant (k obs ) was obtained by fitting the data exponentially and k cat /K M resulted from dividing k obs by GtgE concentration [E 0 ]. For rapid qualitative analysis of GtgE-mutant activity, the method was adopted as follows: GtgE variants were recombinantly expressed in 50 mL. Cells equal to 5 OD units were resuspended in 0.5 mL SEC-buffer lacking nucleotides and lysed by ultrasonication (Digital Sonifier® S-450, Branson Ultrasonics). Protein expression levels were monitored from the cleared lysate by SDS-PAGE and total protein amounts were determined with the OD at 280 nm. Subsequently, the lysate was introduced directly to the assay as described above (ca. 2 µg in 10 µL reaction) or diluted 1:100 in nucleotide free SEC-buffer prior to use. with COOT 42 , rigid body refinement and preliminary positional refinement with phenix.refine 43 , the GtgE subunit was re-traced. Application of feature-enhanced maps using PHENIX in the early and intermediate stages of model building and refinement aided greatly in extending the model to finally include residues 23-213 of GtgE and residues 22-195 of Rab32, along with the GDP and magnesium ion cofactors of the latter protein 44 . Model refinement was performed in phenix.refine and REFMAC 45 . Eventually, water molecules were placed automatically using ARP/wARP 46 . The final model of the Rab32 18-201 :GDP:GtgE 21-214,C45A complex, with refined Translation/Libration/Screw (TLS) tensors and individual B-factors, converged to R work and R free values of 20.8% and 24.5%, respectively. Coordinate geometries were confirmed via MOLPROBITY to possess good stereochemistry and small bond-length and angle RMSDs, with a single glycine residue 59 of Rab32 -adjacent to the site of proteolytic cleavage by GtgE-in the disallowed region of the Ramachandran plot 47 . Magnesium ion coordination was verified with the CheckMyMetal server 48 .
The full-length Rab32:GDP:GtgE C45A structure was solved by molecular replacement, applying the refined coordinates of the Rab32 18-201 :GDP:GtgE 21-214, C45A heterodimer as the initial search model. A single copy of the complex constituted the asymmetric unit of the crystals belonging to space group P6 5 22. In this case, model building was straight forward, eventually featuring residues 23-217 of GtgE and residues 21-195 of Rab32. TLS refinement yielded converging values for R work = 20.1% and R free = 24.0% along with low bond-length and angle RMSDs. Inspection of the Ramachandran plot showed good stereochemistry, with glycine residue 59 of Rab32 again the sole outlier. Structural representations of proteins were prepared with PyMol 49 or Visual Molecular Dynamics (VMD) software 50 .
Molecular dynamics simulations. MD-simulations were performed based on our resolved X-ray structure of the truncated Rab32:GDP:GtgE C45A -complex (i.e., Rab32 18-201 :GDP:GtgE 21-214,C45A ) by replacing Ala-45 with Cys in order to model the wild-type GtgE. The protein complex was embedded in a water-ion environment with a 150 mM NaCl concentration, and the complete simulation setup comprised 98,980 atoms. The wild type and F88 G truncated Rab32:GtgE complexes were simulated in presence of both GTP and GDP for 190 ns at T = 310 K using a 1 fs time step, by using the CHARMM36 force field 51 , and treating the long-range electrostatics using the Particle Mesh Ewald (PME) approach 52 . The MDsimulations were performed using NAMD 53 , trajectories were analyzed using VMD 50 , and principal component analysis based on the MD-trajectories were performed using ProDy 54,55 .
Data availability. Structure factors and model coordinates for the full length and truncated Rab32:GDP:GtgE C45A -complexes have been deposited in the RCSB Protein Data Bank under the accession codes 5OED and 5OEC, respectively. The data that support the findings of this study are available from the corresponding author upon request.