A previous study reported that AnkX was distributed in a punctate pattern when produced ectopically in mammalian cells (Pan et al., 2008). Immunostaining with an early endosome marker showed that GFP tagged AnkX did not co-localize with early endosomes (Mukherjee et al., 2011). We hypothesized that AnkX's cellular destination may be, at least in part, determined by the localization of its enzymatic substrates, Rab1 and Rab35. However, AnkX did not localize at the Golgi where Rab1 is predominantly found (Mukherjee et al., 2011), and we reasoned that perhaps AnkX is co-opted at cellular locations where Rab35 is present, namely the plasma membrane and endosomes. We fused AnkX to an N-terminal GFP tag and analyzed its subcellular localization within transiently transfected COS-1 cells using immunogold transmission electron microscopy. We determined that GFP-AnkX was associated with the plasma membrane, nascent endosomes, and tubular compartments at the periphery of the cell, reminiscent of the subcellular localization of recycling endosomes (Figure 1, Supplementary Figures 1, 2). These membrane compartments shuttle cellular components to the plasma membrane and commonly assume a tubular morphology (van Ijzendoorn, 2006).

We then investigated which region of AnkX was responsible for localization to the plasma membrane. To address this question we generated truncated variants of AnkX fused to an N-terminal mCherry tag as follows: AnkX1-490 (containing the FIC domain and 4 ankyrin repeats), AnkX688-949 (containing 2 ankyrin repeats), AnkX721-949 (containing 1 ankyrin repeat), AnkX491-809 and AnkX491-949 (containing the same 8 ankyrin repeats), and AnkX810-949 (containing no ankyrin repeats; Figure 2A). These fragments were selected based on secondary structure predictions and available structural information (Campanacci et al., 2013). To establish which of these fragments localize at the plasma membrane, we transiently transfected HeLa cells with constructs carrying the mCherry-fused ankX truncations or full length ankX. We then stained the cells with Wheat Germ Agglutinin, a carbohydrate-specific dye that marks the plasma membrane, and analyzed the localization of mCherry-tagged AnkX and AnkX fragments in relation to the plasma membrane (Figure 2B). As expected based on the TEM results, mCherry-AnkX co-localized with WGA at the plasma membrane. In singly transfected cells, we did observe that mCherry-AnkX was not uniformly distributed along the plasma membrane, and instead it assumed a rather patchy distribution, favoring one side of the cell (Supplementary Figure 4). In 5% of the cells, mCherry-AnkX did not appear to be present at the plasma membrane, at least not in levels detectable by confocal microscopy (Supplementary Figure 4). Among all other constructs visualized by confocal microscopy only cells producing mCherry-AnkX491-949 or mCherry-AnkX688-949 displayed fluorescence at the plasma membrane. HeLa cells producing either mCherry-AnkX721-949 or mCherry-AnkX810-949 displayed fluorescence predominantly in the cytosol, although the latter also displayed lower signal in the nucleus. The FIC domain-containing mCherry-AnkX1-490 variant was dispersed throughout the cytosol and was also present on vesicular compartments, whereas mCherry-AnkX491-809 was present in the cytosol, and, surprisingly also in the nucleus. Therefore, the C-terminal half of AnkX possesses properties that provide AnkX with the ability to bind the plasma membrane, but also to bind nuclear components. To further analyze the cellular distribution of mCherry-tagged AnkX variants, we then performed cellular fractionation studies using HEK293T cells producing these protein variants. We chose to employ HEK293T cells for this approach due to their increased transfection efficiency with AnkX-derived constructs. Cellular fractionation studies confirmed that mCherry-AnkX, mCherry-AnkX491-949, and mCherry-AnkX688-949 were membrane-associated (Figure 2C). A lighter band was detected for mCherry-AnkX688-949 in the membrane fraction compared to cells producing mCherry-AnkX491-949. It is possible that this shortened fragment had a weaker interaction with the plasma membrane that was disrupted during ultracentrifugation. As expected based on our confocal microscopy results, mCherry-AnkX721-949 and mCherry-AnkX810-949 were found in the cytosolic fraction. Both mCherry-AnkX1-490 and mCherry-AnkX491-809 were present in the cytosolic and membrane fractions indicating that these fragments are potentially also present on endosomes.

To identify the molecular mechanism underlying AnkX localization at the plasma membrane, we analyzed AnkX's ability to bind phosphoinositide lipids. Multiple Legionella effectors were reported to specifically recognize and bind phosphoinositide lipids (Hilbi et al., 2011). Most of the phosphoinositide binding effectors identified thus far bind PI(4)P and/or PI(3)P (Hilbi et al., 2011). Given that the LCV is enriched in PI(4)P, phosphoinositide binding is thought to serve as an anchoring mechanism for Legionella effectors that attach to the cytosolic surface of the LCV after translocation into the host cell. Additionally, this mechanism can facilitate targeting of particular host membrane compartments (Pizarro-Cerda et al., 2014). To evaluate AnkX's ability to bind phosphoinositide lipids, we performed a protein-lipid overlay assay using purified GST-AnkX and a commercially available nitrocellulose membrane spotted with phosphatidylinositol and all seven phosphoinositide species (Figure 3A). Through this approach we determined that AnkX bound mainly PI(3)P and PI(4)P (Figure 3B). To identify which region of AnkX was responsible for phosphoinositide binding, we generated two GST-tagged AnkX variants: GST-AnkX1-490 and GST-AnkX491-949 (Supplementary Figure 3). Our protein-lipid overlay assay showed that the C-terminal region of AnkX was capable of phosphoinositide binding while the N-terminal region was not (Figure 3B). We concluded that the C-terminal half of AnkX harbors the elements necessary to mediate association with the plasma membrane.

AnkX's localization at the plasma membrane and its ability to phosphocholinate Rab35 suggested that AnkX could target Rab35-positive recycling endosomes. To determine whether AnkX associates with these endosomes, we transiently co-transfected HeLa cells with a plasmid encoding GFP-Rab35 and a plasmid encoding either mCherry-AnkX or mCherry as a control. By confocal microscopy, we observed that the GFP-Rab35 signal was present in the cytosol, membrane compartments, the plasma membrane, and in some cells the nucleus (Figure 4A); nuclear localization of GFP-Rab35 has previously been observed, but this is believed to be caused by nuclear translocation of GFP alone (Kouranti et al., 2006). In agreement with our electron microscopy data, mCherry-AnkX was found to be present at the plasma membrane and on intracellular membrane compartments. When the two proteins were co-produced they co-localized on vacuolar compartments that resembled enlarged endosomes. Similar structures were observed in cells producing GFP-Rab35 in the absence of AnkX (Figure 4A).

Rab35 regulates recycling of protein cargo between the plasma membrane and endosomes, and therefore, we next sought to determine whether AnkX targets Rab35-positive recycling endosomes. The major histocompatibility Class I (MHC-I) enters cells through clathrin-independent endocytosis and then either returns back to the plasma membrane via recycling tubules or is degraded by lysosomes (Caplan et al., 2002). Loss of Rab35 activity has been shown to inhibit MHC-I recycling, indicating that its recycling back to the plasma membrane is, at least in part, dependent on Rab35 (Allaire et al., 2010). To determine whether AnkX targets recycling tubules that are carrying MHC-I, we performed an antibody uptake assay, whereby HeLa cells producing mCherry or mCherry-AnkX were incubated with MHC-I antibody conjugated with Alexa 488 fluorophore to mark endosomes that internalize MHC-I. Using confocal microscopy, we observed that mCherry-AnkX co-localized extensively with MHC-I at the plasma membrane and on tubular compartments (Figure 4B). These results support the hypothesis that AnkX targets Rab35-positive recycling endosomes in human cells.

A series of biochemical approaches have shown that phosphocholination prevents activation of Rab35. Therefore, we hypothesized that during infection AnkX would have the same effect, and thus disrupt the normal dynamics of recycling endosomes. To determine whether AnkX affects endocytic recycling during macrophage infection, we performed a transferrin-recycling assay. Transferrin is routinely used as a marker for endocytic recycling; it remains bound to its receptor throughout the recycling pathway and is released outside the cell upon returning to the cell surface (Mellman, 1996). U937 macrophages were infected with Lp01, Lp01ΔdotA, Lp01ΔankX, Lp01ΔankX complemented with a plasmid carrying wild-type ankX, or Lp01ΔankX complemented with a plasmid carrying ankX_H229A, a point mutant in the catalytic motif of the FIC domain that renders the protein catalytically inactive (Mukherjee et al., 2011). At 1 h post-infection, macrophages were subjected to a pulse-chase sequence with fluorescently labeled transferrin followed by incubation with unlabeled transferrin (Figure 5A). After the pulse step, U937 cells retained comparable levels of transferrin regardless of which Legionella strain they were infected with (Supplementary Figure 6). We then quantified the amount of fluorescently labeled transferrin remaining after a 30-min chase with unlabeled transferrin (Figure 5B). Macrophages infected with Lp01 displayed the highest level of fluorescence, indicating increased retention of transferrin within these cells. This result supports the idea that endocytic recycling of transferrin was blocked at an early stage of infection. In contrast, macrophages infected with the Lp01ΔdotA translocation-deficient mutant showed the lowest fluorescence signal among the five conditions, indicating that endocytic recycling of transferrin was robust in the absence of translocated Legionella effectors. Macrophages infected with the Lp01ΔankX strain displayed levels of fluorescence that were significantly lower than those of cells infected with Lp01, and, thus, in the absence of AnkX, labeled transferrin along with its receptor were recycled back to the plasma membrane. We also noted that, although significantly reduced in Lp01ΔankX-infected cells, the amount of labeled transferrin was 1.6 times higher than that detected in macrophages infected with the Lp01ΔdotA mutant (Figure 5B). When macrophages were infected with the ankX complemented Lp01ΔankX strain, retention of labeled transferrin increased to a level similar to that of the Lp01-infected cells. Interestingly, macrophage infection with the Lp01ΔankX ectopically producing the catalytically inactive AnkX^H229A resulted in diminished retention of labeled transferrin compared to macrophages infected with the Lp01ΔankX strain producing active AnkX (Figure 5B). This change was not attributable to a difference in the production of AnkX or AnkX_H229A within the Lp01ΔankX background since we verified that the two proteins were produced at comparable levels prior to macrophage infection (Supplementary Figure 5). Based on these results we concluded that the phosphocholination activity of AnkX was responsible for disruption of endocytic recycling. Notably, transferrin was not present at the LCV, consistent with previous findings that this compartment lacks transferrin (Joshi et al., 2001).

The Lp01ΔankX mutant has previously been shown to be impaired in its ability to escape phagosome maturation early following macrophage infection (Pan et al., 2008). Our results above indicated that the phosphocholination activity of AnkX is important for exerting its biological function. Therefore, we hypothesized that phosphocholination could play a key role in the avoidance of phagolysosome maturation. To test this hypothesis, we infected U937 macrophages with Lp01, Lp01ΔdotA, Lp01ΔankX, Lp01ΔankX complemented with a plasmid carrying wild-type ankX, and Lp01ΔankX complemented with a plasmid carrying ankX_H229A. At 2 h post-infection, macrophages were fixed, immunostained with an antibody against the lysosomal marker LAMP-2, and quantified with regard to the number of LAMP-2 positive LCVs. U937 cells infected with wild-type displayed 18(±5)% of LAMP-2 positive vacuoles, whereas cells infected with Lp01ΔdotA and Lp01ΔankX displayed 77.3(±9)% and 63(±4.3)% LAMP-2 positive vacuoles, respectively (Figure 5C). These results are in agreement with findings reported by Pan et al. (2008). In addition, we showed that when macrophages were infected with the Lp01ΔankX strain complemented producing AnkX the percentage of LAMP-2 positive LCVs decreased to 31.6(±6)%. In contrast, infection with the Lp01ΔankX strain complemented with catalytically inactive AnkX^H229A did not rescue the phenotype and showed that 70.6(±7.3)% LCVs were LAMP-2 positive, similar to the percentage observed in cells infected with the Lp01ΔankX or the Lp01ΔdotA strains. Therefore, we concluded that AnkX's phosphocholination activity plays a critical role in avoidance of phagosome maturation.

L. pneumophila strains Lp01 (hsdR rpsL) and Lp01ΔdotA (T4SS⁻) are derivatives of L. pneumophila strain Philadelphia-1 (Berger and Isberg, 1993). The wild-type Lp01, Lp01ΔankX mutant, Lp01ΔdotA, and the pJB1806-ankX plasmid were a generous gift from Dr. Craig Roy. All bacterial strains used in this study are listed in Supplementary Table 1. The pJB1806-ankX_H229A shuttle vector was generated by quick-change mutagenesis using oligonucleotides listed in Supplementary Table 2 using a standard protocol. Legionella strains were grown in AYE liquid media or on CYE solid media and maintained as described (Feeley et al., 1979). The pJB1806-ankX and pJB1806-ankX_H229A shuttle vectors were transformed into the Lp01ΔankX strain to complement the deletion strain with catalytically active or inactive AnkX, respectively. Bacterial growth media was supplemented with 25 μg/ml chloramphenicol to maintain pJB1806 plasmids in the Lp01ΔankX strain and 1 mM IPTG was added to the medium to induce gene expression. HeLa tissue culture cells were maintained in RPMI 1640 medium supplemented with 2 mM L-glutamine and 10% FBS, and incubated in 5% CO₂ at 37°C. COS-1 and HEK293T cells were grown in DMEM medium supplemented with 2 mM L-glutamine supplemented with 10% FBS. U937 cells were grown in RPMI 1640 medium supplemented with 10% FBS and were differentiated by supplementing the medium with 10 ng/ml 12-O -tetradecanoylphorbol-13-acetate (TPA) for 16 h.

Antibodies were purchased from BioLegend (MHC-I, clone w6/32 IgG2a), ThermoFisher Scientific (Goat anti-Rat Alexa 350, A-21093; Goat anti-Rat Texas Red, T-6392; Goat anti-Mouse Alexa 488, A-11001; Mouse monoclonal anti-GST, MA4-004; Rabbit polyclonal anti-mCherry, Pa5-34974, HRP-conjugated anti-mouse antibody, NA931), Abcam (Rabbit polyclonal anti-LAMP-2; ab37024; Rabbit polyclonal anti-GFP antibody, ab6556), Enzo (Rabbit polyclonal Calnexin, ADI-SPA-860), ProteinTech (Mouse Monoclonal Alpha Tubulin, 66031-1-Ig), and Jackson ImmunoResearch (Rabbit IgG ChromePure, 011-000-003). The rat polyclonal anti-Legionella antibody and the rabbit polyclonal anti-AnkX used for detection of purified AnkX in Figure 3 were a kind gift from Dr. Matthias Machner (NIH). We also thank Dr. Zhao-Qing Luo (Purdue University) for the rabbit polyclonal anti-AnkX antibody used for detection of AnkX in Legionella lysate (Supplementary Figure 5).

Full-length ankX and truncations were amplified using oligonucleotides in Supplementary Table 2 and were inserted by recombination into pDONR221™. The inserted fragments were then recombined into the mammalian expression vector 362-pCS-Cherry-DEST (Addgene plasmid #13075) or the pcDNA6.2/N-EmGFP-DEST vector (ThermoFisher Scientific) using the Gateway™ cloning technology (ThermoFisher Scientific). Plasmids encoding GST-tagged AnkX variants were generated by using the pDONR221-ankX₁₋₄₉₀ and pDONR221-ankX_491−949 donor vectors to recombine ankX into the destination plasmid pDEST15™ (ThermoFisher Scientific). The pEGFP-C1-Rab35 plasmid was a kind gift from Dr. Julie Donaldson (NIH). Plasmids used in this study are listed in Supplementary Table 1.

HeLa cells grown on 12 mm cover glass (Fisherbrand™) were transiently transfected with plasmids encoding fluorescently-tagged constructs using the Lipofectamine 3000 reagent (ThermoFisher Scientific). At 14 h post-transfection cells were washed with PBS and fixed with 4% paraformaldehyde in PBS for 20 min at room temperature. To visualize the plasma membrane, cells were treated with a Wheat Germ Agglutinin Alexa Fluor 488 conjugate (ThermoFisher Scientific) at 5 μg/ml for 10 min at 37°C. Coverslips were mounted using ProLong Diamond anti-fade reagent (ThermoFisher Scientific) and image acquisition was performed using a ZEISS 710 confocal microscope, a 63× Plan-Apochromatic oil immersion objective (numerical aperture of 1.4), and the ZEN 2012 software (Carl ZEISS MicroImaging).

COS-1 cells transiently producing GFP-AnkX were washed briefly with PBS and fixed overnight with 4% paraformaldehyde and 0.5% glutaraldehyde in 0.1 M sodium cacodylate buffer pH 7.4 at 4°C. Cells were then washed with 0.1 M sodium cacodylate buffer pH 7.4, scraped, and pelleted. The cell pellet was enrobed with 4% low melting point agarose and cut into small 1–2 mm³ cubes. The cells were washed with 0.1 M sodium cacodylate buffer pH 7.4, dehydrated in an ascending ethanol series and infiltrated with LR White resin. Samples were embedded in gelatin capsules and polymerized by UV at 4°C for 2 days. The samples were sectioned on a Reichert-Jung Ultracut E ultramicrotome, and 60–70 nm thick sections were collected onto formvar/carbon coated 200 mesh nickel grids. Immunogold Labeling—Samples were blocked with 0.05 M glycine for 15 min and Aurion goat blocking solution for 30 min before being incubated on drops of anti-GFP antibody diluted to 2.8 μg/ml in Aurion 0.1% BSA-c 7.4 for 1 h. Control grids were incubated on drops of ChromePure rabbit IgG (Jackson ImmunoResearch, Cat No. 011-000-003) diluted to 2.8 μg/ml in Aurion 0.1% BSA-c 7.4. Grids were washed on six drops of Aurion 0.1% BSA-c and incubated on drops of Aurion goat anti-rabbit IgG conjugated to 10 nm gold diluted 1:20 in Aurion 0.1% BSA-c for 2 h. Grids were washed on six drops of Aurion 0.1% BSA-c, 3 drops of PBS, fixed on drops of 2% glutaraldehyde in PBS, and then washed on five drops of Nanopure water. The grids were then post-stained with 2% uranyl acetate in 50% methanol and Reynolds' lead citrate. The samples were examined with a ZEISS Libra 120 transmission electron microscope operating at 120 kV, and images were acquired with a Gatan Ultrascan 1000 CCD camera.

HEK293T cells producing mCherry-AnkX or -AnkX variants were scraped, washed, resuspended in PBS supplemented with protease inhibitors, and lysed by 30 passes through a 27″ gauge needle. The post-nuclear supernatant (PNS) was obtained via centrifugation at 15,000 × g for 10 min at 4°C. To obtain the cytosolic fraction, the PNS was spun at 51,000 rpm for 45 min at 4°C using a TLA-100 rotor in a Beckman ultracentrifuge. PBS was added to the pellet and spun again to wash away cytosolic contaminants. The pellet was resuspended with an equal volume of 2% NP-40 in PBS and collected as the membrane fraction. The PNS, cytosolic fraction, and membrane fraction were analyzed via SDS-PAGE and immunoblot.

U937 macrophages were challenged with Legionella strains at a multiplicity of infection (MOI) of 50 and spun at 200 × g for 5 min and then incubated at 37°C. At 2 h post-infection cells were washed three times with PBS and fixed in 4% paraformaldehyde for 20 min. To distinguish extra- and intracellular bacteria a two-step immunostaining procedure was performed. Without macrophage permeablization, Legionella was detected with an anti-Legionella rat primary antibody (1:3,000) followed by Cascade Blue-conjugated Goat anti-Rat IgG antibody (1:1,000). Cells were then permeabilized with ice-cold methanol for 20 s and internalized bacteria were detected anti-Legionella rat primary (1:3,000) and Texas Red-conjugated Goat α-Rat IgG (1:1,000).

For detection of LAMP-2-positive LCVs, infected U937 cells were immunostained with anti-LAMP-2-rabbit primary antibody (1:150) washed three times with PBS and fluorescently labeled with Alexa 488-rabbit secondary antibody (1:1,000). Coverslips were mounted using ProLong Diamond anti-fade reagent (ThermoFisher Scientific). The percentage of LAMP-2-positive LCVs was determined by scoring 100 cells per coverslip with three replicates for each specific condition. The unpaired two-sided Student's t-test was performed with a p-value cut off of < 0.05 to determine where there was a statistically significant difference in localization to the LCV.

To determine efficiency of recycling, U937 cells were grown on glass coverslips in a 24-well plate and following differentiation into macrophages cells were infected with Legionella strains Lp01, Lp01ΔdotA, Lp01ΔankX, Lp01ΔankX harboring pBJ1806-ankX, and Lp01ΔankX harboring pJB18060-ankX_H229A at an MOI of 50 for 1 h. Infected macrophages were then allowed to uptake fluorescently labeled human transferrin (Tf-Alexa-488; ThermoFisher Scientific) at 5 μg/ml in RPMI for 1 h. Cells were either fixed to measure the fluorescence intensity after the pulse step or they were washed with PBS containing unlabeled human transferrin (100 μg/ml) three times and then incubated further with unlabeled transferrin at 100 μg/ml in RPMI for 30 min. Cells were fixed and immunostained to identify infected cells and visualized by confocal microscopy as described above. All parameters for imaging remained consistent throughout the experiment. Using ImageJ-FIJI software (Schindelin et al., 2012, 2015) the mean fluorescence intensity of Tf-Alexa-488 in 50 infected cells from three replicates for each condition was measured. To account for variation in cell size, the mean fluorescence intensity was determined by normalizing each cell's fluorescence intensity to the total area of the cell. A one-way ANOVA was performed with a p < 0.01 to determine if any strains showed a significant difference in recycling. Post-Hoc Tukey-Kramer test was performed with a p-value cut off of < 0.01 in order to identify which pairs of infected cells were significantly different from each other.

Recombinant HaloTag-AnkX was produced in the Single Step (KRX) competent E. coli strain (Promega), and AnkX was purified using the HaloTag protein purification system as previously described (Neunuebel et al., 2012). Briefly, E. coli cells producing HaloTag-AnkX were lysed using an LV10 microfluidizer (Microfluidics). The cell lysate was spun at 25,000 × g for 20 min, and pre-equilibrated HaloLink resin were incubated with the supernatant for 2 h at 4°C. The resin was then washed and AnkX was cleaved off the resin using His-tagged tobacco etch virus (TEV) protease (Promega) for 2 h at 4°C. HisLink resin (Promega) was used to remove the TEV protease from the supernatant.

AnkX fragments were produced as GST fusion proteins in E. coli BL21 (DE3) at 25°C overnight after induction with 0.5 mM isopropyl-β-dithiogalactopyranoside (IPTG). E. coli cells producing GST-tagged variants were harvested and resuspended in PBS supplemented with 1 mM MgCl₂ and 1 mM β-mercaptoethanol (PBS-MM) followed by lysis using the LV10 microfluidizer (Microfluidics). The cell lysate was centrifuged at 24,000 × g for 35 min, and the supernatant was incubated with pre-equilibrated Glutathione Sepharose 4B (GE Healthcare) for 2 h at 4°C. The resin was washed three times with PBS-MM, and proteins were eluted with 50 mM Tris-HCl (pH 8) containing 10 mM reduced glutathione (Sigma). Glutathione was removed by passing the eluate through a Zeba column (ThermoFisher Scientific).

Protein-lipid overlay assays were performed using commercially available phosphoinositide strips (Echelon Biosciences Inc.). Nitrocellulose membranes pre-spotted with different phospholipids were blocked with 3% non-fat milk in PBST [PBS and 0.1% Tween-20 (v/v) pH 7.5] for 1 h at room temperature. The blocked membranes were incubated with AnkX, GST-AnkX1-490, or GST-AnkX491-949 (0.5 μg/ml in blocking buffer) overnight at 4°C. Protein binding to lipids was visualized with an anti-GST antibody (1:2,000) or an anti-AnkX antibody, and an HRP-conjugated anti-mouse antibody (1:5,000) using a ChemiDoc Touch Imaging system (Bio-Rad).

HeLa cells were grown on cover slips and transiently transfected with mCherry-AnkX using Lipofectamine 2000 (ThermoFisher Scientific) reagent for 16 h. Following transfection cells were incubated with an anti-MHC-I antibody conjugated to Alexa 488 (1:100) for 30 min at 37°C and 5%CO₂. After internalization, cells were rinsed twice with PBS and washed with acid (0.5% acetic acid, 0.5 M NaCl, pH 3.0) for 20 s to remove remaining surface-bound antibody. The acid stripping was followed by a PBS wash and final rinse with RPMI. Cells were fixed with 4% paraformaldehyde in PBS for 20 min at room temperature and visualized by confocal microscopy.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.