Structural insight into the membrane targeting domain of the Legionella deAMPylase SidD

AMPylation, the post-translational modification with adenosine monophosphate (AMP), is catalyzed by effector proteins from a variety of pathogens. Legionella pneumophila is thus far the only known pathogen that, in addition to encoding an AMPylase (SidM/DrrA), also encodes a deAMPylase, called SidD, that reverses SidM-mediated AMPylation of the vesicle transport GTPase Rab1. DeAMPylation is catalyzed by the N-terminal phosphatase-like domain of SidD. Here, we determined the crystal structure of full length SidD including the uncharacterized C-terminal domain (CTD). A flexible loop rich in aromatic residues within the CTD was required to target SidD to model membranes in vitro and to the Golgi apparatus within mammalian cells. Deletion of the loop (Δloop) or substitution of its aromatic phenylalanine residues rendered SidD cytosolic, showing that the hydrophobic loop is the primary membrane-targeting determinant of SidD. Notably, deletion of the two terminal alpha helices resulted in a CTD variant incapable of discriminating between membranes of different composition. Moreover, a L. pneumophila strain producing SidDΔloop phenocopied a L. pneumophila ΔsidD strain during growth in mouse macrophages and displayed prolonged co-localization of AMPylated Rab1 with LCVs, thus revealing that membrane targeting of SidD via its CTD is a critical prerequisite for its ability to catalyze Rab1 deAMPylation during L. pneumophila infection.


Introduction
Altering the function of host proteins through post-translational modification is a popular strategy among microbial pathogens [1]. While some modifications like ubiquitination or phosphorylation are almost universal among eukaryotic proteins, others such as phosphocholination and AMPylation are less frequent yet equally fascinating. AMPylation (also known as adenylylation) was first described in 1967 for the Escherichia coli enzyme GS-ATase (glutamine synthetase adenylyl transferase; GlnE) which modifies the glutamine synthetase with adenosine monophosphate (AMP) [2,3]. More than 40 years passed before AMPylation activity was rediscovered in translocated effectors from Vibrio parahemolyticus (VopS) [4], Histophilus somnii (IbpA) [5], and L. pneumophila (SidM/DrrA) [6]. In each of these newer cases, the AMPylated host targets were Rho or Rab family small guanine nucleotide binding proteins (GTPases) that regulate a wide variety of processes in cells.
The activation state of Rab GTPases is determined by the type of guanine nucleotide they are bound to [7]. The GDP-bound form is inactive, whereas the GTP-bound conformation is active. Activation of Rab GTPases is stimulated by GDP/GTP exchange factors (GEFs), while GTPase-activating proteins (GAPs) stimulate the intrinsic GTP hydrolysis activity of Rabs, thereby promoting conversion of GTP to GDP and, consequently, Rab inactivation [7]. In their active form, Rab GTPases are anchored to membranes via geranylgeranyl lipid groups that are covalently attached to C-terminal cysteine residues, whereas inactive GTPases are chaperoned to the cytosol by GDP dissociation inhibitor (GDI) [7,8].
L. pneumophila is a facultative intracellular pathogen that, upon inhalation of contaminated water droplets, can enter the human lung and cause Legionnaires' pneumonia, a potentially fatal disease that primarily impacts individuals with a weakened immune system [9]. Upon phagocytosis by monocytes such as alveolar macrophages, L. pneumophila translocates close to 300 effector proteins into the host cytosol to bypass cellular defense mechanisms and establish a safe replication compartment, the Legionella-containing vacuole (LCV) [10,11]. Effector translocation requires a functional type IV secretion system (T4SS) called Dot/Icm [12,13], and interference with this process by disrupting or mutating dot/icm genes renders L. pneumophila avirulent, underscoring the importance of the effectors for host cell manipulation.
During infection, L. pneumophila acquires material from various membrane trafficking pathways in order to convert the originally plasma membrane-derived vacuole into a camouflaged endoplasmic reticulum (ER)-like compartment [14]. Rab GTPases control vesicle trafficking within eukaryotic cells and are, thus, frequently manipulated by intracellular pathogens, including L. pneumophila [15,16]. Rab1 plays a key role in the transport of early secretory vesicles from the ER to the Golgi [17]. The effector SidM (or DrrA) recruits Rab1 to the LCV surface early during infection, activates it by catalyzing GDP/GTP exchange, and AMPylates it [6,[18][19][20][21][22]. AMPylated Rab1 is protected from inactivation by both bacterial and host GAPs and, therefore, remains in an active GTP-bound form [6]. Accumulation of Rab1 on LCVs peaks around 2 hours post infection [18,19]. Inactivation and removal of Rab1 is initiated upon translocation of the effector SidD which deAMPylates Rab1, thereby priming it for inactivation by the L. pneumophila GAP LepB and causing its gradual removal from the LCV 4-6 hours after bacterial uptake [23,24]. L. pneumophila mutants lacking SidD (ΔsidD) fail to deAMPylate Rab1, resulting in a prolonged co-localization of AMPylated Rab1 with LCVs [23,24]. The same observation has been made for a L. pneumophila strain lacking the Rab1 GAP LepB (ΔlepB), showing that both SidD and LepB cooperate in the deAMPylation and subsequent de-activation of Rab1 during L. pneumophila infection [23,24].
SidD is a protein of 507 amino acids with no significant sequence homology to other entries in the database. Using protein crystallography, we recently succeeded in obtaining structural information about the N-terminal deAMPylase domain of SidD spanning residues 37 to 350 [25]. This domain, which is sufficient to catalyze deAMPylation in vitro, possesses noticeable similarity to metal-dependent protein phosphatases of the PPM family, most notably human PP2Cα and bacterial PstP [25]. The catalytic site of SidD is comprised of a binuclear metal center with strong dependence on magnesium ions, which are coordinated in part by three aspartate residues (D92, D110, D326) [25]. Substitution of any of the three aspartate residues with alanine abolished the deAMPylation activity of SidD [25]. As a consequence, a L. pneumophila strain producing SidD(D92A) phenocopied a ΔsidD strain and displayed prolonged co-localization with Rab1 during infection [25].
The function of the carboxy-terminal domain of SidD (CTD) has remained unclear. Truncated SidD variants lacking a complete CTD failed to localize to Golgi membranes within transiently transfected COS-1 cells, suggesting that this domain contributes to SidD localization [25]. In this study, we present the crystal structures of both the CTD and full length SidD and reveal how the CTD mediates membrane targeting of SidD, a process that we find here to be critical for the protein's ability to deAMPylate its target during infection.

Crystal structure of SidD and its CTD
To investigate the function of the SidD CTD at a molecular level, we initiated the structural characterization of the full-length protein by X-ray crystallography. Crystals that diffracted to 3.6 Å were obtained with a slightly truncated variant comprised of residues 37-507 (SidD  ). At this stage, crystal derivatization using heavy-atom soaking procedures or isomorphous replacement by SeMet yielded very poor diffraction data. Further attempts at finding a molecular replacement solution using the structure of the SidD 37-350 catalytic domain (PDB ID code 4IIP) as a search model provided a clear solution with 6 molecules per asymmetric unit that showed partial-difference electron density for the remaining C-terminal region of SidD. However, the lack of high-resolution data precluded the iterative refinement and model-building steps required to interpret the structural information. Thus, we focused on obtaining a higherresolution crystal structure of the CTD alone. The construct encompassing amino acids 350-507 (SidD 350-507 ) crystallized readily and diffracted well. The structure was solved by singlewavelength anomalous diffraction using selenium as the anomalous scatterer. The model was refined to a resolution of 2.5 Å (R work 26.4%, R free 28.7%) with good stereochemistry. Residues 350-367 and residues 496-507 have been excluded from the final model as they were located in areas of poor-quality electron density and were most likely disordered. The statistics for data collection and refinement are summarized in S1 Table. The structure of SidD 350-507 revealed a compact helical domain where a central helix (α9) forms the hydrophobic core that is surrounded by five helices that are tilted with respect to the central one and form an antiparallel bundle (Fig 1A). The numbering of secondary structure elements (helices, loops) within the CTD is a continuation of the nomenclature previously introduced for the N-terminal domain of SidD [25]. Despite extensive in silico analyses, we were able to only find domains or regions with weak structural homology to SidD 350-507 (for details see Supporting Material and S9 Fig to S13 Fig). Automated ion placement in PHENIX. REFINE identified a hexa-coordinated Mg 2+ with the classical octahedral geometry formed by the O δ atoms of T417, T421, the main chain carbonyl oxygens of L411, V412, and D414, and one water molecule (Fig 1B). In addition, upon closer inspection of this region we found a second metal ion, which despite not having a recognizable coordination shell, was identified as Zn 2+ based on the clear Zn peaks in anomalous difference Fourier maps from data collected at the zinc edge ( Fig 1B and S1 Fig). Given their selective retention during protein production in and purification from E. coli, we cannot exclude the possibility that they contribute to structural stability of SidD or possibly even a yet undefined catalytic activity (see Supplementary  Information).
With the 2.5 Å resolution structure of SidD 350-507 on hand, we used each of the individual structures (SidD 37-350 , PDB ID code 4IIP and SidD 350-507 , current work) as separate ensembles for the molecular replacement-based solution of the aforementioned 3.6 Å SidD 37-507 structure. Residues 85-88, 269-276, 365-382 and 495-507 were not well-defined in the electron density map and could not be modeled. A summary of the data collection and refinement statistics is given in S1 Table. The six SidD 37-507 molecules present in the asymmetric unit displayed identical structures with an L-like shape formed by two domains in nearly perpendicular orientation ( Fig 1C). We used small-angle X-ray scattering (SAXS) to evaluate the overall structure of SidD in solution. The ab initio molecular envelope calculated from the SAXS data revealed a similar L-shaped structure but with a maximum diameter D max of 115 Å, which differed from the calculated value of 89.2 Å for the crystal structure. Indeed, the SidD 37-507 crystallographic structure exhibited a low fit with the observed solution scattering curve (χ 2 = 2.48) supporting the hypothesis of interdomain flexibility (Fig 2A and 2B). Given that the scattering curve might represent the average of a mixture of conformations coexisting in solution, we explored this possibility by multi-state modeling with SAXS profiles (MultiFoXS) [26]. This approach allowed the deconvolution of the scattering pattern of SidD 37-507 into two subsets of conformations that best reproduced the SAXS profile. A superposition of these structures through the CTD showed significant tumbling of the N-terminal deAMPylase domain, suggesting that SidD is not a rigid molecule (Fig 2C and 2D).

An exposed hydrophobic loop mediates membrane association of SidD
We previously found that the CTD is responsible for the localization of exogenously produced GFP-SidD to the perinuclear region, primarily membranes of the Golgi [25]. The molecular mechanism by which the CTD recognizes and binds to Golgi membranes, but not to other cell organelles, has remained unclear. Unlike many other L. pneumophila effectors, purified SidD showed no binding to phospholipids [25] (S2 Fig), suggesting the existence of a different mode of membrane targeting. Upon closer examination of the CTD sequence, we noticed a stretch of hydrophobic residues between helices α7 and α8 (370-FLGIYGFFT-378; loop CTD ) that was

PLOS PATHOGENS
Membrane targeting of SidD Crystallographic structure of the CTD of SidD, residues 350-507, shown in cartoon ribbon format in three orthogonal orientations. Nomenclature for helices continues the layout previously described for the N-terminal domain of SidD [25]. The zinc and magnesium ions are represented as grey and green spheres, respectively. (B) Final electron density map (2F obs -F calc contoured at 1.5σ, blue mesh) within the area of the two metal ions showing the coordination sphere of the magnesium ion. (C) Crystallographic structure of SidD, residues 37-507, shown in cartoon ribbon format in two orientations. The Nterminal domain (NTD) is colored in slate and the CTD in yellow. https://doi.org/10.1371/journal.ppat.1008734.g001

PLOS PATHOGENS
Membrane targeting of SidD unstructured in the crystal structure of SidD37-507, indicative of intrinsic flexibility, and that could constitute a potential membrane anchor for SidD ( Fig 3A).
To confirm the importance of loop CTD for SidD membrane association within mammalian cells, a loop CTD deletion mutant (GFP-CTDΔ loop ) was generated, and its localization was determined in transiently transfected COS-1 cells (Fig 3B and 3C). In comparison to GFP-tagged CTD which showed the aforementioned co-localization with the Golgi marker giantin [25], GFP-CTDΔ loop failed to localize to the perinuclear Golgi region and, instead, displayed a cytosolic distribution pattern similar to that of GFP (control). To further understand the molecular details of how the loop CTD mediates membrane association, we studied the importance of individual amino acids for this process. Notably, four of the nine loop CTD residues had bulky aromatic side chains (underlined in 370-FLGIYGFFT-378) (Fig 3A) capable of associating with lipid bilayers by inserting into their hydrophobic core made from hydrocarbon tails. While phenylalanine tends to deeply penetrate into the hydrophobic core of membranes, tyrosine and tryptophan assumes a saddle-like distribution preferentially at the lipid-water interface [27,28]. Sequence alignment between SidD from L. pneumophila and its only two other known homologs from L. rowbothamii (Lr) and L. belliardensis (Lb) revealed that the hydrophobic character of the loop is conserved in all three of them ( Fig 3A).
To test if the aromatic residues F370, Y374, F376, and F377 within loop CTD play a role for CTD membrane binding, we replaced each of them individually with either alanine (a small nonpolar residue), serine (a polar residue), or a related aromatic residue (tyrosine with phenylalanine or vice versa), and determined the localization of these SidD variants in transiently transfected COS-1 cells by fluorescence microscopy (Fig 3D and 3E). We found that substitution of F370, F376, or F377 with either alanine or serine severely disturbed Golgi localization of the mutant proteins and rendered them predominantly cytosolic. A more conservative replacement of F370, F376, or F377 with tyrosine, on the other hand, had no noticeable effect on the proteins' Golgi targeting capabilities, emphasizing the critical importance of an aromatic residue at these three positions (Fig 3D and 3E). In contrast, Y374 was dispensable for membrane binding, and its substitution with phenylalanine, alanine, or even serine had no effect on Golgi targeting of GFP-CTD (Fig 3D and 3E). Overall, these results demonstrated that the hydrophobic loop within the CTD of SidD functions as membrane-targeting determinant, and that residues with bulky hydrophobic side chains at positions 370, 376, and 377 are crucial for membrane binding.
To further confirm the contribution of the loop CTD to membrane binding, we studied the association of purified recombinant SidD  or SidDΔ loop with synthetic membranes in a co-flotation assay ( Fig 4A). Upon sucrose gradient centrifugation, the majority of SidD  was detected in the liposome-containing fraction, whereas SidDΔ loop was undetectable ( Fig  4B). Notably, upon cryo-electron microscopy analysis, liposomes that were incubated with SidD 37-507 appeared fully decorated by SidD 37-507 , whereas liposomes that were incubated with SidDΔ loop showed a clean bilayer as did control liposomes ( Fig 4C). Finally, using surface plasmon resonance spectroscopy (SPR), we demonstrated a specific SidD-membrane interaction which yielded a dissociation constant of 4.4 μM, whilst SidDΔ loop , as expected, exhibited no detectable binding (S3 Fig). Importantly, circular dichroism and SAXS measurements showed that the deletion of the hydrophobic loop of SidD neither affected the structure of SidD, nor significantly altered the interdomain flexibility (S4 Fig). Together, these results demonstrate that the hydrophobic loop CTD plays an integral role for association of SidD with biological membranes.

A carboxyl-terminal pair of α-helices determines Golgi specificity of SidD
While hydrophobic interactions can significantly contribute to membrane anchoring of proteins, they provide little selectivity for a particular membrane compartment or organelle. Given the preference of exogenously produced GFP-SidD for Golgi membranes, we

PLOS PATHOGENS
Membrane targeting of SidD hypothesized that the CTD, in addition to the aforementioned membrane-targeting determinant (loop CTD ), contained a specificity determinant that mediated preferential Golgi localization. One feature within the CTD that could constitute such a membrane specificity determinant were two patches of charged residues; a positively charged patch composed of residues K416 and K433 adjacent to the loop CTD , and a negatively charged patch containing D464 and E467 in helix α12 (S5A Fig). Electrostatic interactions often provide selectivity to a protein's membrane localization and orientation by pairing with oppositely-charged lipid head groups, thus favoring interaction with specific types of membranes [29,30]. To test if the charged patches near the loop CTD were responsible for SidD's preference for Golgi membranes, we performed charge inversion through site-directed mutagenesis, replacing either K416 and K433 with glutamates (KK/EE) or D464 and E467 with arginine residues (DE/RR). The localization of these GFP-tagged CTD mutants was examined in transiently transfected COS-1 cells. Notably, charge inversion had no noticeable effect on the extent to which the signal of either GFP-CTD(KK/EE) or GFP-CTD(DE/RR) overlapped with that of the Golgi marker giantin (Fig 5A, S5B Fig), suggesting that neither of the two charged patches made any major contribution to the enrichment of SidD on Golgi membranes or to the protein's general membrane binding capacity.
Next, we systematically shortened SidD-CTD from its C-terminal end ( Fig 5A) and determined the localization pattern of these truncated variants in transiently transfected COS-1 cells in order to identify a possible membrane specificity determinant. Despite containing the complete loop CTD region, the CTD variants ending at position 387, 406, or 435, showed a cytosolic distribution pattern similar to GFP (Fig 5C), suggesting that they no longer bound to membranes, most likely because of folding or stability issues. In contrast, CTD(322-450), which lacked the carboxy-terminal two alpha-helices, α12 and α13 (Fig 5B), still associated with cellular membranes (Fig 5C, S6A Fig). Interestingly, in addition to its enrichment on Golgi membranes, CTD(322-450) was also found on other membrane compartments, including tubular organelles, that did not stain positive for giantin, suggesting that these membranes were not the Golgi. Upon co-production of GFP-tagged CTD(322-450) with a variety of organellar markers in transiently transfected COS-1 cells, we detected a substantial colocalization between CTD(322-450) and the mitochondria marker Mito-RFP (Mito-red fluorescence protein) (Fig 5D). These results suggested that the C-terminal two alpha-helices (α12 and α13) that are missing in CTD(322-450) were responsible for the selective enrichment of SidD on membranes of the Golgi, while the loop CTD functioned as general membrane anchor that is insufficient to distinguish lipid bilayers of different composition. Consistent with this model, we found that substitution of F370, one of the residues critical for loop CTD membrane binding (Fig 3D), with serine resulted in the failure of CTD(322-450; F370S) to associate with either Golgi or mitochondria membranes (Fig 5C and 5D and S6A Fig). Similar results were obtained in transiently transfected Hela cells, both by microscopy and subcellular fractionation, where F370S substitution rendered CTD(322-450) mostly cytosolic (S6 Fig). Taken together, our data revealed that the CTD mediates localization of SidD to membranes using a dual-binding mode that combines a general membrane-binding determinant in form of the loop CTD with a specificity determinant represented by a helix-turn-helix unit at the C-terminus.
Despite extensive efforts using a variety of protein-protein interaction approaches, we were unable to identify any proteinaceous host cell factor that stably interacted with CTD. Since lipid-protein overlay assays had already excluded charged phospholipids as possible SidD ligands [25] (S2 Fig), we performed sequence alignment among the known SidD homologs and identified five conserved residues within the C-terminal two-helix bundle (L469, K472, D484, I485, L491) that are surface-exposed and available to aide in possible ligand interaction. Site-directed mutagenesis of a cluster of four of these residues, either individually or combined, had no obvious effect on Golgi localization of CTD within transiently transfected cells (S7 Fig), suggesting that membrane selectivity may involve interactions through additional CTD residues.

SidD localizes to the surface of Legionella vacuoles
A L. pneumophila ΔsidD mutant is defective for the timely removal of Rab1 from LCVs [23,24], suggesting that during infection the surface of the LCVs is a site of SidD function. Nonetheless, earlier efforts to detect translocated SidD on LCVs by immunofluorescence microscopy have been unsuccessful [23,24], most likely because the levels of endogenously produced SidD were below the detection limit. To experimentally confirm the selectivity of SidD for LCV membranes, we exogenously produced GFP-tagged CTD, CTDΔ loop , or CTD(322-450) in transiently transfected COS-1 cells and then challenged these cells with L. pneumophila. If LCV membranes can be recognized by the CTD, then GFP-tagged SidD should localize to the surface of these compartments. To reduce possible interference by endogenously produced SidD, a L. pneumophila ΔsidD strain was used for these studies, while the T4SS-deficient mutant strain (Lp03) that cannot establish a replication compartment and, instead, is delivered to lysosomes, was included as control (Fig 6). While GFP-CTD accumulated as a ring-shaped green halo about LCVs containing L. pneumophila ΔsidD but not LCVs containing the avirulent strain Lp03 (Fig 6), no enrichment of GFP-CTDΔ loop was observed around either type of LCV (Fig 6), showing that SidD has the capacity to recognize and bind to LCV membranes containing virulent L. pneumophila in a process that requires the loop CTD as membrane binding determinant. Notably, despite its ability to associate with host membranes, GFP-CTD (322-450) failed to accumulate on LCVs (Fig 6), suggesting that the C-terminal α-helix bundle determines membrane specificity of SidD during Legionella infection.

Localization to Legionella vacuoles is critical for SidD function during infection
With the newly acquired ability to disturb SidD membrane binding, we set out to address another important question-whether membrane localization is required for the function of SidD during infection. Earlier studies revealed that failure of L. pneumophila to deAMPylate Rab1 results in a prolonged colocalization of Rab1 with LCVs [23][24][25]. This kinetic defect in Rab1 removal resulted from the inability of AMPylated Rab1 to be efficiently deactivated by the Rab1 GAP LepB. Due to the lack of molecular probes capable of directly binding to and visualizing AMPylated Rab1within cells, monitoring Rab1 dynamics has been used as a surrogate assay for examining the Rab1 AMPylation status and, thus, the activity of SidD [23][24][25].
To investigate the importance of the loop CTD for the biological function of SidD during L. pneumophila infection, the dynamics of Rab1 was analyzed in bone marrow-derived A/J mouse macrophages infected with L. pneumophila containing a chromosomal copy of either sidD (strain Lp02), ΔsidD, or sidDΔ loop . Consistent with our earlier findings [25], Rab1

PLOS PATHOGENS
Membrane targeting of SidD colocalized with Lp02-containing LCVs early but not late during infection (49±1% positive at 2 h, 28±3% at 4 h and 15±1% at 6 h) (Fig 7A), whereas LCVs containing Lp02ΔsidD showed a prolonged colocalization with Rab1 compared to Lp02 (52±2% vs 15±1%) at 6 h after bacterial uptake (Fig 7B). Similar to Lp02ΔsidD, vacuoles containing Lp02sidDΔ loop showed a pronounced delay in Rab1 removal from LCV membrane 4 h post infection compared to Lp02 (62% vs 28%) (Fig 7B). Rab1 removal deficiencies were even more evident 6 h post infection for Lp02sidDΔ loop , with 49% LCV staining positive for Rab1 compared to only 15% for Lp02 (Fig 7A and 7B). A similar defect in Rab1 removal was observed in bone marrow-derived macrophages challenged with Lp02sidD(D92A), a strain that produced a deAMPylation-inactive protein and that is known to exhibit prolonged Rab1 retention [25] (Fig 7B). Immunoblot analyses confirmed that production of SidDΔ loop had no effect on the expression level of sidM in L. pneumophila (S8A Fig). A beta-lactamase reporter-based translocation assay also verified that the deletion of the loop did not interfere with the delivery of SidDΔ loop during infection (S8B Fig). The finding that a SidDΔ loop variant phenocopied the Rab1 removal deficiency of the deAMPylation defective variant SidD(D92A) indicates that membrane association via the loop CTD was of critical importance for SidD to perform its biological function, namely the deAMPylation of Rab1 surrounding Legionella vacuoles.

Discussion
In this study, we provided insight into the structure and function of the L. pneumophila effector SidD, the first and thus far only known translocated effector to exhibit deAMPylation activity, and discovered an important role for the CTD as a membrane targeting module.

Membrane targeting of SidD
By solving the crystal structure of SidD 37-507 , we discovered that the protein assumed an Llike shape within the crystal lattice. The N-terminal deAMPylase domain possesses a protein phosphatase fold, while the CTD is formed by a bundle of anti-parallel alpha-helices with no apparent similarity to known structures ( Fig 1A). The two domains are connected via a flexible linker and are positioned almost perpendicular to each other in the crystal lattice ( Fig 1C). Our SAXS analyses subsequently revealed a relative mobility between the two domains, with 76% of the particles existing as L-shaped molecules in solution while the other 24% of the particles assumed a more extended morphology (Fig 2C). If and how this interdomain flexibility between the two domains is regulated is currently unclear, but it could facilitate the recognition of AMPylated Rab1 during infection. Rab1 is connected to membranes via its~25 amino acid long flexible C-terminal linker termed the hypervariable domain which allows the GTPase domain to be elevated 100Å or more above the lipid bilayer [31,32]. Having a high degree of flexibility between the catalytic domain and the localization domain could increase the sampling radius of membrane bound SidD from a two-dimensional plane to a three-dimensional space (S1 Video).
A notable feature within the CTD that caught our attention was the loop CTD (region 370-FLGIYGFFT-378), which is conserved among SidD homologs ( Fig 3A) and critical for membrane binding of L. pneumophila SidD (Figs 3 and 4). Deletion of the loop CTD strongly attenuate the ability of SidD to associate with membranes both in vitro (Fig 4) and within cells

PLOS PATHOGENS
Membrane targeting of SidD ( Fig 3B). Moreover, substitution of any of the three phenylalanine residues at position 370, 376, or 377 with small or non-hydrophobic residue (alanine or serine) abolished the ability of GFP-CTD to robustly localize to the Golgi apparatus in transiently transfected COS-1 cells (Fig 3D and 3E), indicating that large hydrophobic side chains play a critical role for anchoring SidD to membranes. There are a wide variety of strategies used by bacterial effector proteins to associate with membranes [33], ranging from transmembrane domains (YlfA/LegC7 and YlfB/LegC2) [34], membrane curvature-sensing domains (L. pneumophila RavZ) [35] and phosphoinositide binding domains (SidM, SidC, LidA, RavD, or RavZ to name a few) [35][36][37][38][39][40] to host-mediated ubiquitination (Salmonella SopB) [41] and lipidation (farnesylation, geranylation, S-palmitoylation) [42][43][44][45]. The loop CTD of SidD is much shorter than a typical transmembrane domain (~20 residues long) and, thus, unlikely to cross the entire lipid bilayer. Instead, our data suggest that the loop might form an amphipathic helix or segment that inserts laterally into membranes, with F370, F376, and F377 being buried within the hydrophobic core made from the lipids' hydrocarbon chains. Replacement of either one of these aliphatic phenylalanines by tyrosine did not noticeably affect Golgi localization of the CTD, showing that the amphipathic nature of tyrosines was tolerated at those positions, probably because their hydroxyl group was facing out of the lipid bilayer towards the polar head groups. Serine residues, on the other hand, were not tolerated at amino acid positions 370, 376, and 377 likely due to their hydrophilic character, and neither were alanine residues which may not penetrate the membrane deep enough in order to provide sufficient adhesion. Short stretches of hydrophobic residues have previously been found to function as membrane localization domains in several bacterial effectors, including ExoS and ExoT (both Pseudomonas aeruginosa), YopE (Yersinia pseudotuberculosis), and SopB (Salmonella) (reviewed in [46]). The~22 residue long leucine-rich membrane localization region in ExoS, ExoT, and YopE is predicted to form an amphipathic helix, with charged residues clustered on one side, hydrophobic residues on the opposite face of the helix. In ExoS, substitution of the conserved leucine residues with asparagine phenocopied the membrane targeting defect of a deletion mutant lacking the entire membrane localization domain, whereas substitutions of charged residues abolished plasma membrane localization but not perinuclear localization of ExoS [47,48]. Thus, like ExoS, SidD relies on the combined effort of a membrane binding determinant and a specificity determinant for intracellular targeting.
Not surprisingly, deletion of the membrane localization domain from ExoS or YopE limited the ability of P. aeruginosa to efficiently ADP-ribosylate host cell Ras GTPases or of Y. pseudotuberculosis to exhibit maximum virulence [47,49]. In macrophages infected with L. pneumophila, we made a similar observation where a mutant strain producing SidDΔ loop displayed the same kinetic defect in Rab1 removal from LCVs as a ΔsidD strain, demonstrating that proper membrane binding of SidD was critical for its function and that failure to stably associate with LCVs interfered with the ability of SidD to efficiently deAMPylate Rab1 prior to its removal (Fig 7). Our earlier studies already hinted at the importance of membrane binding for the function of SidD, where the cytotoxicity of SidM overproduction in transiently transfected COS-1 cells was counteracted by full-length SidD but not by truncated SidD lacking the CTD membrane-targeting domain [25]. We also confirmed that exogenously produced GFP-CTD, but not GFP-CTDΔ loop targeted to LCVs of virulent L. pneumophila but not to those of the T4SS-defective mutant Lp03 which is delivered to lysosomes (Fig 6), indicating that SidD can distinguish different types of vacuolar membranes.
During infection, the vacuole containing virulent L. pneumophila is gradually converted into an ER-like compartment that evades endolysosomal fusion [14,50]. Part of that transformation process is the acquisition of certain phospholipids such as PI(4)P and the removal of others (PI(3)P) [51]. Several L. pneumophila effectors have been shown to directly modify phosphoinositides, including the effector LepB (phosphatidylinositide 4-kinase), SidF (phosphoinositide phosphatase), or SidP (phosphoinositide phosphatase) [52][53][54]. Other effectors such as SidM and SidC possess phosphoinositide-specific binding domains for stable association with the membrane of the LCV or surrounding organelles [37,38,51,55,56]. Although SidD did not stably interact with phospholipids in vitro [25], the finding that it localizes to LCVs during infection or Golgi membranes in transiently transfected cells, both organelles that contain PI(4)P, suggests that this phospholipid and/or a protein common to both compartments could contribute to the selectivity of SidD for these membranes. The fact that vacuoles containing Lp03 do not maintain a stable pool of PI(4)P may explain why they failed to attract GFP-CTD (Fig 6).
By producing truncated CTD variants in transiently transfected COS-1 cells, we discovered that the selectivity of SidD for specific membranes required a C-terminal helix-loop-helix motif (residues 451-507) of the CTD, and that deletion of α12 and α13 resulted in the failure of CTD(322-450) to selectively target the Golgi (in transiently transfected cells) (Fig 5) or the LCV (during infection) (Fig 6). Instead, the protein localized to organelles other than the Golgi, including mitochondria ( Fig 5D). We hypothesize that the α12 and α13 bundle forms an interaction platform for a yet-to-be determined host cell ligand that is enriched on the LCV during infection (Fig 6) or the Golgi within transiently transfected cells (Fig 5), and that deletion of this platform causes SidD to indiscriminately insert into membranes via its hydrophobic loop. While the host ligand that directs the CTD to specific membranes awaits identification, our data provide evidence for a previously unknown dual mode mechanism for membrane targeting by a L. pneumophila effector, where the combined action of a general membrane-binding determinant and a specificity determinant directs SidD to a particular type of membrane. We speculate that this dual targeting mode allows proper SidD localization within a wide variety of amoebal species that the bacterium encounters in its natural freshwater habitat, thus providing the organism with a fitness advantage. Future studies will reveal if additional effectors from L. pneumophila or related pathogens use a similar targeting strategy within host cells.

Strains, plasmids, and reagents
All Legionella strains are derivatives of Legionella pneumophila Philadelphia-1 Lp02 (thyA, hsdR, rpsL) [57]. Lp03 is a T4SS-defective strain variant with a mutation in dotA [57], while Lp02sidD(D92A) encodes a deAMPylation-deficient mutant form of SidD [25]. Lp02sidDΔ loop has a chromosomal sidD allele with a deletion of the sequence encoding amino acids 370 to 378. L. pneumophila was cultured in liquid AYE medium or maintained on solid CYET plates as described before [58].

Membrane targeting of SidD
NcoI as restriction sites and adding 5 or 6 histidines tag at the C-terminus respectively. pHis-Parallel2-SidD Δloop was constructed by Gibson assembly using pHis-Parallel2-SidD 37-507 as template and the primes SidDΔloop-up, SidDΔloop-low, Amp-up and Ampi-low. The Clontech Mito-RFP plasmid was a gift from Richard Youle (National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD). Plasmids were introduced into L. pneumophila by natural transformation [59]. E. coli was grown in Luria-Bertani (LB) broth with antibiotics (30 μg/ml Kanamycin, 100 μg/ml Ampicilin or 100 μg/ml Chloramphenicol) where necessary.

Cell lines and immunofluorescence microscopy
COS-1 and Hela cells were grown in Dulbecco's modified minimum Eagles's medium (DMEM) supplemented with 10% FBS and incubated at 37˚C in 5% CO 2 . Bone marrowderived macrophages (BMMs) were isolated from the femurs of female A/J mice and differentiated in RPMI-1640 containing 20% FBS, 1.6 mM glutamine, 30% L-cell culture medium, and penicillin (10,000 IU/ml) and streptomycin (10 mg/ml) for one week as previously described [60].
COS-1 or Hela cells grown in 24-well plates were transfected with plasmids encoding EGFP fused SidD variants using Lipofectamine1 2000 (Life Technologies, Inc.). After overnight incubation, cells were fixed with 3.8% formaldehyde, permeabilized with cold methanol, blocked with 1% BSA and stained with anti-giantin antibody at a dilution of 1:3000. Coverslips were mounted with Prolong1 Gold antifade reagent (Thermo Fisher Scientific) and imaged with a Zeiss Axio Observer Z1 inverted microscope or Zeiss LSM800 confocal microscope.
For transfection-infection assays, COS-1 transfected with plasmids encoding GFP-tagged SidD variants were challenged with the indicated L. pneumophila strains at an MOI = 50, and spun at 200g for 5 min to enhance bacteria-cell contact. After 1 h, the cell monolayers were washed three times with warm DMEM to remove extracellular bacteria and incubated in DMEM for another 1 h. The cell monolayers were chemically fixed with 3.8% formaldehyde. Cells were permeabilized with cold methanol, blocked with 1% BSA, and stained for outside bacteria using rat anti-Legionella antibody and goat anti-rat Alexa Fluor355-conjugated antibody while intracellular bacteria were labeled with goat anti-rat Alexa TexRed-conjugated antibody.
BMMs were challenged with the indicated L. pneumophila strains at an MOI = 5, fixed with 3.8% formaldehyde at the indicated time points, and stained for extracellular and intracellular bacteria as described above. The presence of Rab1 on LCV membranes was determined using protein-specific antibody and goat anti-rabbit Alexa Fluor 488-conjugated antibody. Images were taken on a Zeiss LSM800 confocal microscope with Airyscan.

Protein production and purification
Recombinant proteins for structural analyses were produced in E. coli BL21(DE3) (Stratagene) and purified similarly to previously described methods [25]. Briefly, cells were grown in LB medium to an OD 600 of 0.8, protein production was induced with 1 mM isopropyl-β-dithiogalactopiranoside (IPTG) at 20˚C overnight. Cells were harvested by centrifugation, resuspended in TBS (150 mM NaCl, 50 mM Tris-HCl, pH 7.4 supplemented with 5 mM imidazole, 10 mM β-mercaptoethanol (BME)) and lysed at 4˚C by high pressure homogenization (27 Kpsi; Constant System Ltd). All subsequent purification steps were carried out at 4˚C. Insoluble material was removed by ultracentrifugation, and His-tagged proteins were purified by affinity chromatography using 10 ml of Ni-NTA beads (QIAGEN) packed in a gravity column. After extensive washing with the same loading buffer, proteins were eluted by addition of 200 mM imidazole and dialyzed in 25 mM NaCl, 10 mM BME and 50 mM Tris-HCl pH 8.5 buffer. Tobacco etch virus (TEV) protease was included in a 1/20 protease/protein ratio during the overnight dialysis to cleave the His-tag. The protein solution was loaded onto an ion exchange chromatography column (HiTrap Q HP; 5 ml; GE Healthcare) followed by an isocratic gradient from 0.025 M to 1 M NaCl in 20 column volumes. Fractions containing SidD  or SidD 350-507 were concentrated and loaded onto a HiLoad 16/60 Superdex 200 Gel Filtration column (GE Healthcare) or a HiLoad 16/60 Superdex 75 Gel Filtration column (GE Healthcare), respectively, that had been equilibrated with TBS supplemented with 10 mM BME. Fractions containing pure protein were pooled, concentrated to 1 mg/ml and stored at -80˚C. SidDΔ loop was expressed and purified as described for SidD  .
Selenomethionine (SeMet)-substituted SidD 350-507 was produced in E. coli B834 (Stratagene), a methionine auxotroph strain, using SelenoMet medium (Molecular Dimensions) following the manufacturer's instructions. SeMet SidD 350-507 purification was carried out under the same conditions as that of the native protein. The efficiency of SeMet incorporation was evaluated by mass differences between native (unlabeled) and SeMet-labeled protein samples using MALDI-TOF mass spectrometry. The observed mass differences confirmed 100% selenomethionine incorporation at seven expected sites.

Crystallization, data Collection and Structure determination
Crystals of native and SeMet substituted SidD 350-507 were grown by hanging drop vapor diffusion at room temperature by mixing a 11 ratio of protein stock (8 mg/mL) to well solution (2.0 M NaCl, 20% glycerol, 0.1 M sodium acetate, pH 4.6). Crystals of SidD 37-507 were grown by sitting drop vapor diffusion at room temperature by mixing a 3:1 ratio of protein stock (10 mg/ mL) to well solution (2.88 M sodium formate, 0.1% Anapoe 35, 0.09 M Tris-HCl, pH 8.0). Crystals of SidD 350-507 appeared within one week whilst crystals of SidD 37-507 appeared after three months.
Data collection with the SidD 350-507 SeMet-substituted crystal was carried on a Pilatus 6M detector (Dectris) on the I02 beamline at Diamond Light Source (United Kingdom). Posterior data collection at higher resolution for native SidD 350-507 (2.5 Å) and for native SidD 37-507 (3.6 Å) was performed on a Pilatus 6M detector (Dectris) on the PXI-X06SA MD1 beamline at Swiss Light Source (Switzerland). All diffraction data sets were processed using the XDS program [61]. The Se positions in SidD 350-507 and the phase calculations were determined by direct methods using AutoSharp [62]. An initial model was generated with phenix.autobuild within the PHENIX suite [63]. The model was manually completed in Coot [64] and then refined with PHENIX with alternating rounds of manual intervention and optimization. Data collection and refinement statistics are shown in S1 Table. SAXS analysis SAXS data were collected on the BL21 beamline at the Diamond Light source (United Kingdom), using the HPLC-integrated SAXS setup with a Pilatus 2M detector over an angular range qmin = 0.004 Å -1 to qmax = 0.37 Å -1 . For SidD  or SidD Δloop , a volume of 45 μl at 11 mg/ml was injected into a pre-equilibrated (25 mM HEPES pH 7.5, 300 mM NaCl and 0.5 mM TCEP) Shodex KW-402.5 gel filtration column. Scattering was recorded over the course of protein elution with a flow rate of 0.16 ml/min at 20˚C. Data averaging and reduction was carried out with ScÅtter (Version 3.1r by Robert P. Rambo, Diamond Light Source, UK). Further analyses (radius of gyration; Rg, maximum distance; Dmax and particle distance distribution function; p(r) calculations) were carried out with the ATSAS suite package [65]. Lowresolution shape envelopes for SidD  were determined using the ab initio bead modelling program DAMMIF [66]. Modelling was performed without the use of symmetry restraints. Twenty simulations were performed, which generated similar shapes from witch an averaged filtered structure was obtained using DAMAVER [67]. Then, this structure was refined as a fixed input core in DAMMIN to generate a final model with 795 dummy-atoms. The resulting bead model was converted to a map envelope and visualized using CHIMERA [68]. The program SUPCOMB [69] was used to compare the ab initio model with the crystallographic data whilst the program CRYSOL [70] was used to compare the 1D scattering curve with the theoretical scatter of SidD 37-507 model. The Multifox webserver [26] was used to calculate a multistate model of SidD  or SidD Δloop in solution. The crystallographic structure was used as input model where residues I335 to L339 where considered flexible, and then 10.000 conformations were generated by the RRT (Rapid Random Trees) algorithm sampling. The top solution corresponded to a two-state model, and there was no significant improvement in the χ 2 scores for models of three or more states.

Lipid overlay assay
Lipid overlay assays were performed following the manufacturer's instructions. Briefly, membrane strips (Echelon Biosciences, cat# P-6001, P-6002) with spotted lipids were blocked with TBST + 5% BSA at room temperature for 1 h, incubated with purified SidD  or SidM at room temperature for 1 h, and washed with TBST three times to remove unbound protein. The membrane strips were then incubated with anti-SidD or anti-SidM antibody, washed again with TBST three times, followed by incubation with HRP-conjugated goat anti-rabbit secondary antibody. The binding of protein to lipids on the membrane strips was detected by chemiluminescence.

Circular dichroism measurements
Circular dichroism (CD) measurements were carried out with a Jasco J-810 spectropolarimeter (JASCO, Tokyo, Japan) equipped with a Peltier temperature control. CD spectra were acquired at 25˚C with 0.2 nm data pitch, 50 nm/min scanning speed, 4 sec response, 4 nm band width and 5 accumulations, using a 0.1 cm path length quartz cuvette. The samples were measured at a concentration of 30 μM in 5 mM HEPES buffer pH 7.5, 75 mM NaCl and 0.5mM DTT.

PLOS PATHOGENS
Membrane targeting of SidD with 10 mM HEPES pH 7.5, 150 mM NaCl and 1 mM DTT at a final lipid concentration of 1 mM. Small unilamellar vesicles (SUVs) were prepared by vigorous vortexing of the lipid suspension followed by sonication for 10 minutes with cycles of 5 second pulses in 5 second intervals.
Liposome preparation for cryo-EM followed the same protocol as for the flotation assay but without the inclusion of Rhod-PE and sucrose.

Liposome flotation assay
In a standard reaction, 100 μl liposomes (1 mM) were mixed with 25 μM SidD 37-507 or Sid-D Δloop in buffer FB and incubated for 15 minutes. The sample was gently mixed with a stock solution of 80% sucrose in FB to achieve a final sucrose concentration of 30%, which was placed at the bottom of an ultra-Clear TM tubes (Beckman Coulter, cat. 344090) and overlaid with 600 μl buffer FB25 (25% sucrose in FB), and 100 μl buffer FB. Tubes were centrifuged at 240,000 g for 1 hour at 4˚C, and the top 30-50 μl fraction was collected and used for SDS-PAGE analysis.

Cryo-electron microscopy analysis
Samples containing liposomes incubated with SidD 37-507 , SidD Δloop or no protein, were loaded on freshly glow-discharged Quantifoil R2/2 grids. Vitrification was performed on Vitrobot Mark II (FEI Company, USA) maintained at 8˚C and at a relative humidity close to saturation (90% rH). Five microliters of sample solutions were absorbed onto the grid for 30 seconds and blotted with filter paper. Grids were immediately plunged into a liquid ethane bath and stored under L2 until visualization. Imaging of cryoTEM samples was made on a JEM-2200FS/CR transmission electron microscope operated at 200 kV. An in-column omega energy filter helped to record images with improved signal to noise ratio by zero-loss filtering. The energy selecting slit width was set at 9 eV. Digital images were recorded on UltraScan4000 CCD camera (Gatan Inc.) under low-dose conditions at a nominal magnification of 50,000x obtaining a final pixel size of 2.7 Å/pixel. Cross-sectional membrane intensity profiles were calculated with the image processing software ImageJ [71].

Surface plasmon resonance (SPR)
SPR data were collected using a BIAcore 3000 system (GE Healthcare) and a L1 sensor chip (GE Healthcare), which contains alkyl chains for capturing liposomes. All SPR experiments were performed in running buffer (10 mM HEPES pH 7.5, 150 mM NaCl, 1 mM DTT) at 25˚C. The L1 chip surface was conditioned with two injections of a mixture of isopropanol and 50 mM NaOH at a ratio 2:3 (5 μl at 10 μl/min). After overnight dialysis at 4˚C against the running buffer, a 2-fold serial dilution of SidD 37-507 and SidD Δloop was prepared (32 to 0.125 μM) in running buffer. For each concentration, 20 μM liposomes were immobilized over the chip (10 μl at 5 μl/min) and any excess unbound liposomes were removed by injecting 5 μl of 50 mM NaOH (10 μl/min flow rate). Uncovered sensor chip surface was blocked with a 0.2

PLOS PATHOGENS
Membrane targeting of SidD mg/ml BSA (10 μl at 5 μl/min) to avoid non-specific binding. Binding experiments were carried out by injecting SidD  or SidD Δloop (40 μl at 20 μl/min) over the liposome-coated chip. The L1 chip was regenerated and stripped of liposomes after each run by injecting a mixture of isopropanol and 50 mM NaOH at a ratio 2:3 (5 μl at 10 μl/min) and 20 mM CHAPS (5 μl at 10 μl/min). No loss of sensor chip binding capacity due to regeneration occurred, and the capture of liposomes was similar in each run. The sensograms were processed using the BIAcore 3000 BiaEvaluation software (GE Healthcare). Equilibrium dissociation constants were obtained by fitting the maximum RUs reached at each SidD concentration to the steadystate affinity model, assuming a 1:1 binding stoichiometry. Each experiment was done in triplicate.

Ethics statement
All experiments were carried out in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The animal use protocol was reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of NIH (protocol number ASP#18-084).