Transport through recycling endosomes requires EHD1 recruitment by a phosphatidylserine translocase

Abstract

P₄-ATPases translocate aminophospholipids, such as phosphatidylserine (PS), to the cytosolic leaflet of membranes. PS is highly enriched in recycling endosomes (REs) and is essential for endosomal membrane traffic. Here, we show that PS flipping by an RE-localized P₄-ATPase is required for the recruitment of the membrane fission protein EHD1. Depletion of ATP8A1 impaired the asymmetric transbilayer distribution of PS in REs, dissociated EHD1 from REs, and generated aberrant endosomal tubules that appear resistant to fission. EHD1 did not show membrane localization in cells defective in PS synthesis. ATP8A2, a tissue-specific ATP8A1 paralogue, is associated with a neurodegenerative disease (CAMRQ). ATP8A2, but not the disease-causative ATP8A2 mutant, rescued the endosomal defects in ATP8A1-depleted cells. Primary neurons from Atp8a2^−/− mice showed a reduced level of transferrin receptors at the cell surface compared to Atp8a2^+/+ mice. These findings demonstrate the role of P₄-ATPase in membrane fission and give insight into the molecular basis of CAMRQ.

Introduction

Membrane-bound organelles of eukaryotic cells have distinct lipid compositions that influence the function of organelles (Behnia & Munro, 2005). Membrane lipids not only serve as a physical barrier, but also interact with a wide variety of integral and peripheral membrane proteins to regulate their localization and activity. Moreover, phospholipids can be distributed asymmetrically between the two leaflets of the lipid bilayer (Devaux, 1992), which provides two distinct surfaces for protein interactions. For example, the cytosolic leaflet of the plasma membrane (PM) is enriched in the aminophospholipids PS and phosphatidylethanolamine (PE), whereas the extracellular leaflet of the PM is enriched in sphingolipids. A low PS concentration in the extracellular leaflet of the PM is crucial for cell life, since phagocytes recognize and engulf cells by the binding to the PS in the surface of target cells (Balasubramanian & Schroit, 2003). A high PS concentration in the cytosolic leaflet of the PM is essential to facilitate various signaling events through membrane translocation and activation of various kinases, such as protein kinase C (Newton & Keranen, 1994) and Akt kinase (Huang et al, 2011). In addition to the PM, the membranes of endocytic organelles have asymmetric transbilayer distributions of PS (Fairn et al, 2011).

The transbilayer asymmetry of phospholipids is generated, in part, by the selective transport of phospholipids across the membranes (Graham, 2004; Holthuis & Levine, 2005). The P4 subfamily of P-type ATPases (P₄-ATPases) are responsible for the flipping of aminophospholipids to the cytosolic leaflet (Sebastian et al, 2012; Coleman et al, 2013). The human genome contains 14 P₄-ATPases, and mutations in some P₄-ATPases cause inherited genetic diseases. Mutations in ATP8B1 are associated with familial intrahepatic cholestasis type 1 (Bull et al, 1998), and mutations in ATP8A2 are associated with a severe neurological disorder characterized by cerebellar ataxia, mental retardation, and disequilibrium syndrome (CAMRQ) (Onat et al, 2013). In mice, mutations in ATP8A2 cause axonal degeneration (Zhu et al, 2012) and mutations in ATP11C cause B-cell deficiency syndrome (Siggs et al, 2011; Yabas et al, 2011). Despite the mounting evidence showing the physiological importance of P₄-ATPases, how their dysfunction causes diseases remains unclear.

REs function in several membrane transport pathways (Mellman, 1996; Maxfield & McGraw, 2004; Taguchi, 2013): the recycling of internalized molecules back to the PM, the exocytic trafficking of several cargoes, and the retrograde transport of internalized molecules to the Golgi. REs also serve as a source of membranes for autophagosomes (Longatti et al, 2012; Knaevelsrud et al, 2013; Puri et al, 2013). We previously showed that evectin-2, an RE protein with a pleckstrin homology (PH) domain, is essential for retrograde traffic (Uchida et al, 2011). The PH domain of evectin-2 selectively binds PS in vitro and is required for the localization of evectin-2 to REs. The PH domain of evectin-2 faces the cytosol, indicating that PS is present in the cytosolic leaflet of RE membranes. The C2 domain of lactadherin tagged with GFP (GFP-lact-C2) is a widely used PS probe (Yeung et al, 2008; Fairn et al, 2011). When GFP-lact-C2 was overexpressed in the cytosol, it accumulated at the REs and suppressed the retrograde traffic from REs to the Golgi, indicating that PS at cytosolic leaflet has a critical role in endosomal membrane traffic (Uchida et al, 2011). How PS is enriched in the cytosolic leaflet of RE membranes remains unknown.

Eps15 homology domain-containing protein 1 (EHD1) belongs to the EHD-containing or receptor-mediated endocytosis (RME)-1 family, whose members are highly conserved eukaryotic ATPases that have been implicated in membrane remodeling in the endosomal system (Daumke et al, 2007; Grant & Caplan, 2008). The human genome contains four EHDs (EHD1-4), one of which (EHD1) is essential for recycling membrane traffic from REs to the PM (Grant et al, 2001; Lin et al, 2001; Caplan et al, 2002). EHD1 is also required for the retrograde membrane transport of Shiga toxin from REs to the Golgi (McKenzie et al, 2012). In EHD-1 depleted cells, CD59-positive endosomes became tubulated (Cai et al, 2012, 2013). In vitro, EHD1 facilitates tubulation or fission of liposomes that contain anionic phospholipids, suggesting that EHD1 functions in the formation of membrane carriers from REs (Pant et al, 2009).

Here, we show that a P₄-ATPase (ATP8A1) localizes at REs, regulates the transbilayer asymmetry of PS in the RE membranes, and is essential for the RE localization of EHD1. Together with the observations that EHD1 binds PS in vitro and the membrane localization of EHD1 is lost in cells that are defective in PS synthesis, we propose that the PS flipped to the cytosolic leaflet by ATP8A1 is essential for the EHD1 recruitment to REs, thereby regulating the membrane traffic through REs. ATP8A2 is a tissue-specific ATP8A1 paralogue and is associated with a neurodegenerative disease (CAMRQ). ATP8A2, but not the disease-causative ATP8A2 mutant, rescued the endosomal defects in ATP8A1-depleted cells. Primary neurons from Atp8a2^−/− mice had nearly the same abundance of TfnR as wild-type (WT) neurons, but showed a lower expression of TfnR at the PM. This suggests that ATP8A2 in neurons is essential for recycling membrane traffic and defects in this pathway provide a molecular mechanism for CAMRQ.

Results

ATP8A1 is essential for membrane traffic through REs

Four P₄-ATPases (ATP8A1, ATP9A, ATP11A, and ATP11B) are ubiquitously expressed and localize at early endosomes (EEs) and/or REs (Takatsu et al, 2011; Kato et al, 2013). Because the ATPase activity of ATP8A1 is strongly activated by PS (Ding et al, 2000; Paterson et al, 2006; Soupene et al, 2008) and because Drs2 (the yeast homologue of ATP8A1) can translocate PS (Zhou & Graham, 2009), we reasoned that ATP8A1 is involved in the enrichment of PS in the cytosolic leaflet of RE membranes.

We first examined the subcellular localization of ATP8A1 in COS-1 cells. COS-1 cells are a cell line with distinctly separate organelles, therefore are suitable for the precise localization of endosomal proteins (Misaki et al, 2007) (Supplementary Fig S1). ATP8A1 forms a heteromeric complex with CDC50A, a protein that is essential for the transport of newly synthesized ATP8A1 from the endoplasmic reticulum (ER) (Bryde et al, 2010; van der Velden et al, 2010). When co-expressed with CDC50A, ATP8A1 co-localized with the RE markers, Rab11 and TfnR, but did not co-localize with other organelle markers [TGN46 (trans-Golgi network), GM130 (cis-Golgi), VPS26 (EEs), CD63 (late endosomes), and LAMP2 (lysosomes)] (Fig1A). These results showed that ATP8A1 is predominantly localized at REs.

Figure 1.

ATP8A1 regulates endosomal membrane traffic at REs

• A COS-1 cells co-transfected with GFP-ATP8A1 and myc-CDC50A were stained for myc-tag and TfnR, TGN46, GM130, VPS26, CD63, or LAMP2. mCherry-Rab11 was expressed in a cell in the top row. Nuclei were stained with DAPI (in blue). Pearson's coefficient between GFP-ATP8A1 and each organelle marker is shown in the merged image (mean ± SD, n > 20 cells from two experiments). Scale bars, 10 μm.
• B COS-1 cells were treated with ATP8A1 siRNAs for 72 h. Cell lysates were prepared and then immunoblotted with anti-ATP8A1 antibody. α-tubulin was used as a loading control.
• C, D COS-1 cells were treated with ATP8A1 siRNA1 for 72 h. siRNA-resistant GFP-ATP8A1 (WT or E191Q) and myc-CDC50A were transfected 48 h after siRNA transfection. Cells were pulsed with Alexa 594-Tfn and chased in medium with unlabeled Tfn. During the chase, cells were imaged every 6 min for 54 min. The fluorescence intensity of Alexa 594-Tfn is shown as percentage of that at 0 min (mean ± SEM, n = 4–6; ***P < 0.001; Student's t-test). See Supplementary Fig S2 for the representative images.
• E, F COS-1 cells were treated with ATP8A1 siRNA1 for 72 h. Cells were pulsed with Alexa 594-CTxB for 5 min and chased for 15 or 60 min. Cells were then fixed and stained for GM130. In (E), representative images are shown. In (F), the Pearson's coefficient between CTxB and GM130 after 60-min chase is shown. Data represent mean ± SD (n > 26 cells from three independent experiments). ***P < 0.001; two-tailed Student's t-test. Scale bars, 10 μm.

We examined whether ATP8A1 is involved in two membrane transport pathways through REs: the recycling pathway that transports cargoes, such as Tfn/TfnR, from REs to the PM and the retrograde pathway that transports cargoes, such as cholera toxin B subunit (CTxB), from REs to the Golgi. Among three siRNAs designed against human ATP8A1 (siRNA1-3), siRNA1 was the most effective at suppressing ATP8A1 expression (Fig1B) and was used for the following experiments. For the recycling assay, cells were first pulsed with Alexa 594-Tfn and then chased in media without Alexa 594-Tfn. During the chase, cells were imaged every 6 min for 1 h. After the pulse, Alexa 594-Tfn accumulated at REs in both control and siRNA-treated cells. In control cells, the fluorescence intensity of Alexa 594-Tfn in REs decreased by 50% after a 30-min chase and decreased by 70% after a 50-min chase (Fig1C and Supplementary Fig S2). In contrast, in ATP8A1-depleted cells, the fluorescence intensity of Alexa 594-Tfn in REs decreased by 10% after a 30-min chase and decreased by 20% after a 50-min chase, suggesting that ATP8A1 is required for efficient recycling of Tfn from REs to the PM. We also performed the rescue experiment by the expression of an siRNA-resistant wild-type ATP8A1. As shown in Fig1D and Supplementary Fig S2, reinsertion of the wild-type ATP8A1 restored Tfn recycling (not significant versus ‘Control’ in Fig1C by Student's t-test).

For the retrograde assay, cells were pulsed with Alexa-594 CTxB and then chased for 15 or 60 min. As was shown previously (Uchida et al, 2011), CTxB reached REs after a 15-min chase (Fig1E) and then reached the Golgi after a 60-min chase. In ATP8A1-depleted cells, CTxB reached REs after a 15-min chase as in control cells, but CTxB did not reach the Golgi after a 60-min chase. The colocalization between CTxB and GM130 after a 60-min chase was significantly lower in ATP8A1-depleted cells than in control cells (Fig1F). Thus, ATP8A1 is required for the retrograde traffic of CTxB from REs to the Golgi.

The endosomal traffic of epidermal growth factor (EGF) from the PM to lysosomes does not pass through REs (Mellman, 1996). After a 5-min pulse/60-min chase of EGF, a portion of EGF co-localized with LAMP1 in control and ATP8A1-depleted cells. Although the LAMP1-positive lysosomes appeared swollen in ATP8A1-depleted cells, no significant difference in the Pearson's coefficient between LAMP1 and EGF was observed (Supplementary Fig S3A and B). This suggests that knockdown of ATP8A1 did not affect the traffic of EGF to LAMP1-positive lysosomes.

Depletion of ATP8A1 induced the formation of aberrant endosomal tubules

Tfn is recycled as a complex with TfnR (Yamashiro et al, 1984). The impaired recycling traffic of Tfn in ATP8A1-depleted cells led us to examine the steady-state distribution of TfnR. In control cells, an endogenous TfnR stain showed a bright perinuclear structure and a number of dispersed puncta, which correspond to REs and EEs, respectively (Misaki et al, 2007) (Fig2A). In ATP8A1-depleted cells, although the RE localization of TfnR was basically maintained, TfnR also localized at an extensive array of fine tubules that emanate radially from REs. The tubules were even more obvious in the intensified image. The lengths of some TfnR-positive tubules exceeded 20 μm, nearly reaching the PM (Supplementary Fig S4). The treatment of ATP8A1-depleted cells with nocodazole, a microtubule-depolymerizing agent, disassembled the tubules, indicating that the emergence of TfnR-positive tubules depends on microtubules (Supplementary Fig S4). The staining pattern of VPS26, CD63, or syntaxin 5 (Golgi) was indistinguishable between control and ATP8A1-depleted cells (Fig2B), suggesting that this tubulation of membranes occurred specifically at REs.

Figure 2.

ATP8A1 activity is essential for fission of TfnR-positive membrane carriers

• A COS-1 cells were treated with ATP8A1 siRNA1 for 72 h and stained for TfnR. Each image represents the maximum intensity z-projection of four confocal slices. Intensified images are shown in the bottom row. Red dashed lines and arrowheads indicate nuclei and TfnR puncta, respectively. Scale bars, 10 μm.
• B COS-1 cells were treated with ATP8A1 siRNA1 for 72 h and stained for the indicated proteins. Scale bars, 10 μm.
• C COS-1 cells were treated with ATP8A1 siRNA1 for 72 h. TfnR-GFP was transfected 48 h after siRNA transfection. Cells were then imaged with an epifluorescent microscope. Selected stills from Supplementary Movie S1 are shown. The asterisks indicate the tips of the tubules of interest. Scale bars, 5 μm.
• D The effect of increasing levels of 1,2-dioleoyl-PS on the ATPase activity of purified ATP8A1 (WT and E191Q)-CDC50A complex. The K_m and V_max for the WT complex were 59.1 ± 3.5 μM and 38.7 ± 0.8 μmol ATP/min/mg protein, respectively.
• E COS-1 cells were treated with ATP8A1 siRNA1 for 72 h. siRNA-resistant ATP8A1 (WT or E191Q), and myc-CDC50A were transfected 48 h after siRNA transfection. Cells were then fixed and stained for TfnR. The graph shows percentages of cells showing aberrant TfnR-positive tubules as mean ± SD of three independent experiments. At least 50 cells were counted per experiment.

The distribution of TfnR was examined in live cells with the expression of TfnR tagged with GFP. In control cells, TfnR-GFP was mainly found as the mixture of small vesicles and short tubules at REs (Supplementary Fig S5). In contrast, in ATP8A1-depleted cells, TfnR-GFP localized to a number of aberrant tubules, as expected from the observation with endogenous TfnR. Using time-lapse imaging, the dynamics of the TfnR tubules were monitored (Supplementary Movie S1). The tubule marked with a red asterisk retracted over the time monitored and was finally absorbed into REs (Fig2C). The tubule marked with a green asterisk remained associated with REs. At the time of 0.075 min, small stub with a black asterisk budded from the middle of the tubule but was absorbed by the time of 0.084 min. Few fission events that released TfnR-positive tubules from REs into the cytoplasm were observed.

To determine whether ATP8A1 activity is required for the normal RE distribution of TfnR, we generated an ATPase activity-deficient ATP8A1 mutant. To do this, we mutated Glu191 in human ATP8A1 to Gln because the corresponding Glu198 in bovine ATP8A2, a paralogue of ATP8A1, was found to be essential for its ATPase and flippase activities (Coleman et al, 2012). The recombinant Glu191Gln (E191Q) protein complexed with CDC50A was purified to apparent homogeneity (Supplementary Fig S6). The purified WT ATP8A1 showed ATPase activity in a PS-concentration-dependent manner, whereas the E191Q mutant showed only marginal ATPase activity, even when 1 mM PS was included (Fig2D). We then expressed an siRNA-resistant WT ATP8A1 or E191Q mutant in ATP8A1-depleted COS-1 cells. The number of cells that had TfnR-positive tubules was reduced by the expression of the siRNA-resistant WT ATP8A1 but not by the expression of the siRNA-resistant E191Q mutant (Fig2E). The expression of E191Q mutant also could not restore Tfn recycling (Fig1D). Together, these results suggest that normal RE distribution of TfnR and Tfn recycling depend on the ATPase activity of ATP8A1.

The RE localization of EHD1 depends on ATP8A1

The emergence of aberrant TfnR-positive tubules that appear resistant to fission suggests that endosomal membrane fission machinery is dysfunctional in ATP8A1-depleted cells. EHD1 regulates membrane traffic through REs (Grant & Caplan, 2008; McKenzie et al, 2012) and facilitates tubulation or fission of liposomes that contain the anionic phospholipids, PS and phosphoinositides (PIPs) (Pant et al, 2009). We therefore reasoned that ATP8A1 knockdown would impair EHD1 function.

We first examined the effect of EHD1 knockdown on membrane traffic through REs. siRNA1 designed against EHD1 effectively suppressed EHD1 expression (Fig3A) and was used for the knockdown experiments. EHD1 knockdown significantly suppressed the recycling of Tfn from REs to the PM (Fig3B and C) and the retrograde transport of CTxB from REs to the Golgi (Fig3D; Supplementary Fig S7A). Strikingly, EHD1 knockdown, like ATP8A1 knockdown, resulted in the emergence of aberrant TfnR-positive tubules from REs (Fig3E; Supplementary Fig S7B). Time-lapse images of EHD1-depleted cells that express TfnR-GFP also showed aberrant TfnR-positive tubules, which appear resistant to fission (Supplementary Fig S7C and Supplementary Movie S2). In summary, EHD1 knockdown phenocopied the ATP8A1 knockdown, in that both impair membrane traffic through REs and generate aberrant TfnR-positive tubules.

Figure 3.

EHD1 localizes at REs in an ATP8A1 activity-dependent manner and is required for fission of TfnR-positive membrane carriers

• A COS-1 cells were treated with EHD1 siRNAs for 72 h. Cell lysates were prepared and then immunoblotted with EHD1 antibody. α-tubulin was used as a loading control.
• B, C Cells were treated with EHD1 siRNA1 for 72 h. Cells were pulsed with Alexa 488-Tfn and chased in medium with unlabeled Tfn. During the chase, cells were imaged every 3 min for 57 min. In (B), representative images of Alexa 488-Tfn in control or EHD1-depleted cells are shown. In (C), the fluorescence intensity of Alexa 488-Tfn is shown as percentage of that at 0 min (mean ± SEM, n > 3 cells from three independent experiments). ***P < 0.001; two-tailed Student's t-test. Scale bars, 10 μm.
• D COS-1 cells were treated with EHD1 siRNA1 for 72 h. Cells were pulsed with Alexa 594-CTxB for 5 min and chased for 60 min. Cells were then fixed and stained for GM130. The Pearson's coefficient between CTxB and GM130 is shown (mean ± SD, n > 30 cells from two independent experiments). ***P < 0.001; two-tailed Student's t-test. See Supplementary Fig S7A for the representative images.
• E COS-1 cells were treated with EHD1 siRNA1 for 72 h and stained for TfnR. The image represents the maximum intensity z-projection of four confocal slices. Scale bar, 10 μm.
• F COS-1 cells co-transfected with GFP-ATP8A1 and myc-CDC50A were stained for EHD1. Scale bar, 10 μm.
• G Cells were treated with ATP8A1 siRNA1 for 72 h and stained for EHD1. For the rescue experiments, siRNA-resistant GFP-ATP8A1 (WT or E191Q) and myc-CDC50A were transfected 48 h after siRNA transfection. Asterisks indicate cells with GFP-ATP8A1 expression. The graph represents percentage of cells with perinuclear localization of EHD1 as mean ± SD of three independent experiments (at least 141 cells were counted for each condition). Scale bars, 10 μm.

We next examined the effect of ATP8A1 knockdown on the subcellular localization of EHD1. Endogenous EHD1 mainly localized perinuclearly and co-localized with GFP-ATP8A1 (Fig3F), confirming the RE localization of EHD1 as previously described (Lin et al, 2001). In ATP8A1-depleted cells, EHD1 was not localized at the REs but diffusively distributed in the cytoplasm. The RE localization of EHD1 in ATP8A1-depleted cells was restored by the expression of siRNA-resistant WT ATP8A1, but not by the expression of siRNA-resistant E191Q mutant (Fig3G). These results suggest that the ATPase activity of ATP8A1 is required for the RE localization of EHD1.

Cellular PS level is critical for EHD1 localization

We reasoned that EHD1 is recruited to REs by phospholipids that are flipped by ATP8A1. The ATPase activity of ATP8A1 is most strongly activated by PS among the various phospholipids (Ding et al, 2000; Paterson et al, 2006) and ATP8A1 is suggested to flip PS in isolated chromaffin granules (Zachowski et al, 1989; Tang et al, 1996). However, flipping of PE at the PM is impaired in ATP8A1-depleted cells (Kato et al, 2013). The phospholipid specificity of ATP8A1 flippase activity has not been examined in reconstituted systems with purified components.

The ATPase activity of purified recombinant ATP8A1 complexed with CDC50A (Supplementary Fig S6) was assayed in the presence of various lipids. The ATPase activity of ATP8A1 was maximally activated by PS (Fig4A), in agreement with previous findings (Ding et al, 2000; Paterson et al, 2006), and to a lesser extent by PE. The flippase activity of ATP8A1 was then measured in liposomes consisting of 97.5% PC and 2.5% NBD-labeled PS or PE. ATP8A1 flipped both NBD-PS and NBD-PE to the same degree (Fig4B), suggesting that ATP8A1 is not a PS-specific flippase. As expected, the PS flippase activity of the ATP8A1 E191Q mutant that lacks ATPase activity was completely abolished.

Figure 4.

EHD1 localization is dependent on cellular PS levels

• A The effect of various lipids on the ATPase activity of the purified ATP8A1-CDC50A complex. The ATPase activity was measured in the presence of 5 mM ATP and 10 mol % of one of the following lipids: 1,2-dioleoyl-phosphatidycholine (PC), 1,2-dioleoyl-PS, 1,2-dioleoyl-PE, 1,2-dioleoyl-phosphatidylglycerol (PG), 1,2-dioleoyl-phosphatidylinositol (PI), 1,2-dioleoyl-phosphatidic acid (PA), sphingomyelin (SM), cholesterol (Chol), or brain polar lipid (BPL) and 90% PC. The activity of the ATP8A1 (E191Q)-CDC50A in the presence of 1,2-dioleoyl-PS was also determined.
• B NBD-labeled PS or PE transport (flippase) activity of ATP8A1 (WT and E191Q)-CDC50A complex. The percentage of NBD-lipid flipped was shown.
• C Schematic of the biosynthetic pathway of PS in mammalian cells. The impaired pathways in PSA-3 and R-41 cells are indicated.
• D, E PSA-3 and R-41 cells were transiently transfected with GFP-EHD1 and then cultured under the indicated conditions for 72 h. Cells were then fixed, and the nuclei were strained with DAPI. Arrowheads indicate the perinuclear RE localization of GFP-EHD1. Scale bars, 10 μm.

Data information: In (A, B), the data represent the mean ± SD of n = 3 independent experiments.

Given that ATP8A1 can flip PS and PE, we examined whether the cellular levels of PS and/or PE affect EHD1 localization. In mammals, PS is synthesized by two distinct base-exchange enzymes, PS synthase-I (PSS-I) and PS synthase-II (PSS-II) (Kuge & Nishijima, 1997). PSS-I exchanges serine for the choline of PC, whereas PSS-II exchanges serine for the ethanolamine (Etn) of PE (Fig4C). A PS-auxotrophic mutant of CHO cells, PSA-3, lacks the activity of PSS-I (Kuge et al, 1986). When PSA-3 cells are cultured with dialyzed fetal bovine serum (FBS), which is depleted of Etn, cellular levels of PS and PE decrease by around 30% compared to PSA-3 cells maintained in FBS plus Etn (Lee et al, 2012). Under these conditions, the cytoplasmic PS levels drastically decrease, in particular, in REs (Lee et al, 2012). EHD1 localization was examined in PSA-3 cells under PS/PE-deficient or PS/PE-rich conditions. Since anti-EHD1 antibody, which was used to stain EHD1 in COS-1 cells, did not stain endogenous EHD1 in CHO cells, GFP-EHD1 was transiently expressed. Unlike in COS-1 cells, GFP-EHD1 expressed in PSA-3 cells under PS/PE-rich conditions localized at discrete tubular compartments as well as perinuclear REs (Fig4D, arrowhead). In CHO cells, from which PSA-3 cells were derived, EHD1 localized at perinuclear REs and some peripheral tubules (Lin et al, 2001). The tubular compartments resemble ‘EHD1-positive tubular compartments’ that were observed in HeLa cells (Caplan et al, 2002). The EHD1-positive tubular compartments are suggested to be a way-station for recycling cargoes from the perinuclear REs to the PM (Grant & Donaldson, 2009; Sharma et al, 2009). The different morphologies of EHD1-positive compartments among cell lines may be due to how cells develop post-RE compartments. Despite these differences, under PS/PE-deficient conditions, GFP-EHD1 was evenly distributed into the cytoplasm and not localized to the tubules and perinuclear REs (Fig4D). These results suggest that the decrease of PS and/or PE levels in cells influences the membrane localization of EHD1.

To dissect the roles of PS and PE in EHD1 localization, another CHO mutant cell line (R-41) was exploited. R-41 cells are defective in the synthesis of PE from PS (Emoto & Umeda, 2000). When R-41 cells are cultured with dialyzed FBS, in a condition where PE synthesis from Etn is suppressed, PE levels decrease by around 40%, whereas PS levels are unchanged (Emoto & Umeda, 2000; Lee et al, 2012). Between the two culture conditions: dialyzed FBS plus Etn (PS/PE rich) or dialyzed FBS only (PS rich, PE deficient), the tubular and perinuclear RE localization of EHD1 was indistinguishable (Fig4E). These results suggest that PE is not an essential determinant for the membrane localization of EHD1. These findings, together with the findings utilizing PSA-3 cells, indicate that PS is critical for the membrane localization of EHD1.

EHD1 binds PS-containing liposomes

To clarify the roles of PS in EHD1 binding, we performed the binding assay with PC-based liposomes that contain different amounts of PS. In brief, His-tagged EHD1 was mixed with liposomes for 15 min, and the mixture was spun at 100,000 g for 30 min. The resultant supernatant and pellet were subjected to SDS–PAGE, and the gels were then stained with Coomassie blue for the presence of EHD1 and phospholipids (Boucrot et al, 2012). The recombinant EHD1 did not bind to liposomes that do not contain PS (Fig5A and B). When the PS concentration in liposomes increased, the binding of EHD1 increased sigmoidally (Fig5C). A sharp increase of the binding of EHD1 was observed from 40 to 70% PS with half-maximal binding at 53% PS. EHD1 did not bind to liposomes that contain 60% PE (Fig5A and B), but bound to liposomes that contain 60% PI(4,5)P₂ (Supplementary Fig S9A), suggesting that the binding of EHD1 to PS-liposomes is not necessarily through the structure-specific interaction, but rather through the electrostatic interaction.

Figure 5.

EHD1 binds PS in vitro

• A His-tagged EHD1 was mixed with PC-based liposomes containing the indicated % (mol/mol) of PS or PE. After 15 min, the mixture was spun at 100,000 g for 30 min, and the resultant supernatant (S) and pellet (P) were subjected to SDS–PAGE. The proteins and lipids were then stained with Coomassie blue.
• B, C The intensities of individual bands in (A) were quantified with ImageJ, and the percentage of bound proteins was calculated. In (B), the data represent the mean ± SD of n = 3 independent experiments and were analyzed with one-way analysis of variance (ANOVA) followed by Tukey–Kramer post hoc test. *P < 0.01; NS, not significant. In (C), each point represents the mean of two independent experiments with an error bar showing the range of values obtained. The sigmoidal fitting was applied.
• D, E His-tagged EHD1 was mixed with PC-based liposomes (20, 40, or 60% PS) of decreasing size produced by extrusion through filter pores of the size indicated. After 15 min, the mixture was spun at 100,000 g for 30 min, and the resultant supernatant (S) and pellet (P) were subjected to SDS–PAGE. The proteins and lipids were then stained with Coomassie blue. In (E), the intensities of individual bands were quantified (mean ± SEM of three independent experiments). See Supplementary Fig S8 for the size distributions of the prepared liposomes.

We next examined the effect of liposome curvature on EHD1 binding. Liposomes with three different diameters were prepared (Supplementary Fig S8). As shown in Fig5D and E, when the curvature of liposomes increased (i.e., the diameter decreased), more EHD1 bound to liposomes at three PS concentrations examined (20%, 40%, and 60%). These results show that the membrane curvature influences the EHD1 binding to the membranes. The PS concentration required for EHD1 binding to membranes could be lowered by the membrane curvature.

EHD1 is shown to bind to liposomes that contain 10% of PIPs and 10% PS (Blume et al, 2007; Pant et al, 2009). As Blume et al reported, EHD1 bound to liposomes [35% PC, 35% PE, 10% PS, 10% cholesterol, and 10% PI(4)P or PI(4,5)P₂] under our experimental conditions (Supplementary Fig S9B and C). We also examined the binding of EHD1 to liposomes that lack PS [45% PC, 35% PE, 0% PS, 10% cholesterol, and 10% PI(4)P or PI(4,5)P₂] and found that EHD1 did not bind to these membranes (Supplementary Fig S9B and C). These results indicate that PS is required for EHD1 binding to the membranes. Given that EHD1 did not bind to liposomes with a low concentration of PS in the absence of PIPs (Fig5A and C), other lipid factors may affect the EHD1 binding to the membrane containing PS. As has been reported (Behnia & Munro, 2005; Uchida et al, 2011), PIPs were not identified in the RE membrane (Supplementary Fig S10), which suggests that the contribution of PIPs for the EHD1 recruitment to the RE membranes may be minimal.

ATP8A1 knockdown increases PS levels in the luminal leaflet of RE membranes

We hypothesized that ATP8A1 controls the transbilayer distribution of PS in RE membranes. To test this hypothesis, we generated a new PS-specific probe based on the PH domain of evectin-2. The PH domain of evectin-2 specifically binds PS, but with low affinity (Uchida et al, 2011). Two evectin-2 PH domains were tandemly connected, hereafter named 2xPH, expressed in Escherichia coli, and purified. The recombinant 2xPH retained the PS specificity and had a high affinity for PS (Fig6A). More than 95% of 2xPH was sedimented with liposomes containing 5% PS, whereas about only 40% of single PH was sedimented with liposomes containing even 20% PS (Uchida et al, 2011). Recombinant 2xPH was then applied to cells that had been fixed and treated with 0.1% saponin. Saponin was used to permeabilize RE membranes based on the observations that RE membranes are enriched in cholesterol (Gagescu et al, 2000; Mondal et al, 2009). Under these conditions, the antibody that recognizes the luminal part of TfnR stained REs, ensuring the permeabilization of RE membranes (Supplementary Fig S11A). 2xPH primarily stained a perinuclear area that co-localized with Tfn (Fig6B), indicating that RE membranes are enriched in PS compared with other intracellular organelles. It should be noted that the stain of 2xPH under these conditions reflects the presence of PS in both the cytosolic and luminal leaflets of the membranes.

Figure 6.

ATP8A1 knockdown causes an increase in RE luminal PS levels

• A His-tagged 2xPH was mixed with liposomes that were composed of 75% (mol/mol) PE, 20% (mol/mol) PC, and 5% (mol/mol) of PS, PA, PI, PIPs, or sulfatide. In the case of PE/PC, liposomes of 80% (mol/mol) PE and 20% (mol/mol) PC were used. After 15 min, the mixture was spun at 100,000 g for 30 min, and the resultant supernatant (S) and pellet (P) were subjected to SDS–PAGE. The gels were stained with Coomassie blue.
• B COS-1 cells were incubated in medium containing Alexa 488-Tfn for 40 min, washed, fixed, and stained with recombinant 2xPH. Magnified image is shown in the right panel. Scale bar, 10 μm.
• C COS-1 cells were transiently transfected with GFP-lact-C2 and stained for Rab11. Scale bar, 10 μm.
• D, E COS-1 cells were transiently transfected with GFP-lact-C2 and stained with recombinant 2xPH and anti-GM130 antibody. The asterisk indicates a cell expressing GFP-lact-C2, and the arrowhead indicates a non-expressing cell. The graph in (E) shows fluorescence intensity per pixel of recombinant 2xPH stain in RE area, which was delimited by GM130. Data are normalized to the 2xPH intensity in GFP-lact-C2 (−) cells (mean ± SD, n = 13 cells for each condition). Scale bar, 10 μm.
• F-H COS-1 cells stably expressing GFP-lact-C2 were treated with ATP8A1 siRNA1 or EHD1 siRNA1 for 72 h and stained with recombinant 2xPH and anti-GM130 antibody. The intensities of 2xPH staining and GFP-lact-C2 in RE area were measured. The intensity of 2xPH (G) and the ratio of 2xPH to GFP-lact-C2 intensities (H) are shown. In (H), data are normalized to that of control cells. Data represent mean ± SD and were analyzed with two-tailed Student's t-test (n > 26 cells from three independent experiments). **P < 0.01; ***P < 0.001; NS, not significant. Scale bars, 10 μm.

The C2 domain of lactadherin tagged with GFP (GFP-lact-C2) is a widely used PS probe (Yeung et al, 2008). GFP-lact-C2 that is expressed in the cytosol binds PS in the cytosolic leaflet, but not in the luminal leaflet of the membranes. As was shown previously (Uchida et al, 2011), the cytosolic expression of GFP-lact-C2 strongly stained REs (Fig6C), indicating that the cytosolic leaflet of the RE membranes is the most enriched with PS.

Using these two PS-specific probes, we examined how PS is distributed between the two leaflets of the RE membranes. Cells were transiently transfected with the GFP-lact-C2 plasmid, then fixed with PFA, permeabilized with saponin, washed, and stained by 2xPH. The image (Fig6D) showed two representative cells, one that did not express GFP-lact-C2 in the cytosol (indicated by arrowhead) and the other that expressed it (indicated by asterisk). In a cell that did not express GFP-lact-C2, the post-stain by 2xPH clearly visualized REs, as shown above (Fig6B). In contrast, in a cell that expressed GFP-lact-C2, the post-stain by 2xPH was much fainter than that in a cell that did not express GFP-lact-C2. The intensity of 2xPH at REs in cells with GFP-lact-C2 expression decreased to about 30% of that in cells without GFP-lact-C2 expression (Fig6E). As mentioned above, GFP-lact-C2, which is expressed in the cytosol, binds PS in the cytosolic leaflet, but not in the luminal leaflet of the membranes. Given that GFP-lact-C2 in the cytosol masks all of the PS in the cytosolic leaflet of the membranes in fixed cells and the fluorescence of the post-stain with 2xPH represents the amount of unmasked PS in the luminal leaflet (Supplementary Fig S11B), this result suggests that 70% of the PS in REs localized in the cytosolic leaflet and the other 30% localized in the luminal leaflet. If any PS in the cytosolic leaflet could not be masked by GFP-lact-C2 and was post-stained by 2xPH, the amount of PS in the cytosolic leaflet should be higher than 70%.

We then examined whether ATP8A1 depletion affects the post-stain by 2xPH at REs. In this experiment, we used COS-1 cells that stably express GFP-lact-C2 in the cytosol. The stain of 2xPH at REs in control cells was as faint as the stain in cells that transiently expressed GFP-lact-C2 (Fig6D), whereas the stain of 2xPH in ATP8A1-depleted cells was bright with 1.5-fold increase in the fluorescence intensity (Fig6F and G). The ratio of the fluorescence intensity of 2xPH to that of GFP-lact-C2 was significantly higher in ATP8A1-depleted cells than in control cells (Fig6H). This result suggests that ATP8A1 flips PS from the luminal to the cytosolic leaflet in RE membranes. On the other hand, EHD1 knockdown did not increase the intensity of post-stain of 2xPH, suggesting that EHD1 did not affect the transbilayer distribution of PS in REs.

ATP8A2 can compensate for the loss of ATP8A1

ATP8A2, a tissue-specific paralogue of ATP8A1, is highly expressed in brain, testis, and retina (Zhu et al, 2012). ATP8A2 is the causative gene of wabbler-lethal mice, which show axonal degeneration (Zhu et al, 2012). A missense variant of the human ATP8A2 (Ile376Met) has recently been shown to be associated with neurodegenerative disease CAMRQ (Onat et al, 2013). Single knockout mice of either ATP8A1 or ATP8A2 are viable, but double knockout mice of ATP8A1 and ATP8A2 are neonatal lethal (Zhu et al, 2012), suggesting that the two flippases have redundant roles. We reasoned that ATP8A2, like ATP8A1, is a regulator of endosomal membrane traffic and examined whether ATP8A2 can compensate for the loss of ATP8A1 in COS-1 cells in which no ATP8A2 expression is observed.

GFP-ATP8A2 co-expressed with myc-CDC50A co-localized with TfnR in the perinuclear region (Fig7A), showing the RE localization of ATP8A2. Like bovine ATP8A2 (Coleman et al, 2009), the recombinant human ATP8A2 displayed ATPase activity that was induced by PS (Fig7B) and PS flippase activity (Fig7C). In contrast, the disease-associated I376M ATP8A2 mutant displayed little ATPase or flippase activity, consistent with the previous study using the corresponding bovine mutant (I364M) (Vestergaard et al, 2014). We then performed a rescue experiment in ATP8A1-depleted cells with the expression of three types of human ATP8A2 (WT, I376M, and E210Q), the latter corresponding to flippase-deficient E198Q of bovine ATP8A2. Although all three ATP8A2 proteins localized primarily at REs, only the WT restored EHD1 localization to REs and resulted in the disappearance of aberrant TfnR-positive tubules (Fig7D–F). Thus, ATP8A2 can compensate for the loss of ATP8A1 function in a flippase-dependent manner, suggesting that ATP8A2 has a role in the transbilayer distribution of PS in RE membranes.

Figure 7.

Wild-type ATP8A2, but not a disease-associated mutant, can compensate for the loss of ATP8A1 function

• A COS-1 cells co-transfected with GFP-ATP8A2 (WT) and myc-CDC50A were stained for myc-tag and TfnR. Scale bar, 10 μm.
• B The effect of increasing levels of PS on the ATPase activity of ATP8A2 (WT and I376M)-CDC50A complex.
• C NBD-labeled PS transport (flippase) activity of ATP8A2 (WT and I376M)-CDC50A complex. The percentage of NBD-lipid flipped was shown.
• D Cells were treated with ATP8A1 siRNA1 for 72 h. ATP8A2-1D4 (WT or mutants) and myc-CDC50A were transfected 48 h after siRNA transfection. The cells were then stained for EHD1 and 1D4-tag. Scale bars, 10 μm.
• E Quantification of percentages of cells with perinuclear EHD1 localization. Data represent mean ± SD of three or four independent experiments. At least 20 cells were counted per experiment.
• F Cells were treated with ATP8A1 siRNA for 72 h. GFP-ATP8A2 (WT or mutant) and myc-CDC50A were transfected 48 h after siRNA transfection. The cells were then stained for TfnR. Data represent mean ± SD of three or four independent experiments. At least 20 cells were counted per experiment.
• G, H Surface biotinylation assay with Atp8a2^+/+ or Atp8a2^−/− cortical neurons. The ratio of surface to total TfnR was quantified and normalized to Atp8a2^+/+. Data represent mean value (2 mice for each genotype).

We then asked whether endogenous ATP8A2 participates in the localization of recycling cargoes such as TfnR. Since ATP8A2 is not ubiquitously expressed (Coleman et al, 2009) but expressed in brain (Zhu et al, 2012), we used primary cortical neurons that were prepared from Atp8a2^+/+ or Atp8a2^−/− mice (Coleman et al, 2014). Although the total amounts of TfnR in Atp8a2^−/− and Atp8a2^+/+ neurons were comparable, the ratio of surface TfnR to total TfnR was dramatically reduced in Atp8a2^−/− neurons (Fig7G and H). These results imply that ATP8A2 functions in recycling membrane traffic to the PM in neurons.

Discussion

In the present study, we showed in mammalian cells that a P₄-ATPase ATP8A1 translocates PS from the luminal leaflet to the cytosolic leaflet in the RE membranes. We also found that a membrane fission protein EHD1, which binds PS in vitro, is dispersed from REs to the cytoplasm by ATP8A1 depletion. Cells depleted of ATP8A1 generate extensively tubulated membranes from REs, which appear resistant to fission. Cells depleted of EHD1 also generate tubular membranes from REs. From these results, we propose that the flipping of PS to the cytosolic leaflet by a P₄-ATPase is critical for recruiting a membrane fission protein to the RE membranes and generating transport vesicles from REs. Studies in Saccharomyces cerevisiae have shown that a P₄-ATPase Drs2, which flips PS and to a lesser extent PE, is essential for membrane traffic between the late Golgi compartment and endosomes (Sebastian et al, 2012) and that other P₄-ATPases Dnf1 and Dnf2, which flip phospholipids such as PC and PE, participate in endocytic vesicle formation (Pomorski et al, 2003). In Caenorhabditis elegans, a P₄-ATPase TAT-1 is most closely related to mammalian ATP8A1. The loss of tat-1 leads to the generation of abnormal endo-lysosomal compartments, suggesting that endocytic cargo sorting and recycling are impaired (Ruaud et al, 2009; Chen et al, 2010). In Arabidopsis thaliana, a P₄-ATPase ALA3 that localizes at the Golgi is essential for the formation of secretory vesicles (Poulsen et al, 2008). Thus, these studies, together with the present finding in mammalian cells, indicate that the function of phospholipid flippases in vesicular membrane traffic is evolutionally conserved from lower eukaryotes to mammals. In the C. elegans tat-1 mutant, an EHD1 homologue, RME-1, still localized at abnormal endo-lysosomes (Chen et al, 2010). We cannot explain at the present time the apparent discrepancy on the localization of EHD1/RME-1 in the loss of ATP8A1/TAT-1.

One hypothesis for the mechanism by which P₄-ATPases function in the vesicular membrane traffic is that P₄-ATPases generate membrane curvature through unidirectional movement of phospholipids, which facilitates membrane budding to form transport vesicles together with budding machinery (Sebastian et al, 2012). Recently, S. cerevisiae Gcs1, which is an Arf GTPase-activating protein and regulates retrograde transport from endosomes to the late Golgi, was shown to have an +ALPS-motif that binds highly curved membranes with anionic phospholipids including PS (Xu et al, 2013). Importantly, the localization of Gcs1 to the late Golgi/endosomes is regulated by Drs2. Our study in mammalian cells provides another mechanistic insight into the site of action of P₄-ATPases: PS flipped by ATP8A1 to the cytosolic leaflet recruits a membrane fission protein EHD1 that has an ability to bind PS, thereby regulating the membrane fission process.

PS comprises 5–10% of the total phospholipids in mammalian cells (Vance & Steenbergen, 2005; Lee et al, 2012). PS is fivefold more abundant in the purified RE fraction than in the cell lysate (Gagescu et al, 2000), suggesting that PS accounts for 25–50% of the total phospholipids in RE membranes. Using two PS-specific probes, we roughly estimated that 70 and 30% of PS are localized in the cytosolic and the luminal leaflets of RE membranes, respectively. Taking this PS asymmetry into consideration, PS may comprise 35% (= 25 × 2 × 0.7) − 70% (= 50 × 2 × 0.7) of total phospholipids in the cytosolic leaflet of RE membranes. PS is also enriched and asymmetrically distributed in the cytosolic leaflet of the PM (Devaux, 1992). A study using red blood cells, which lack intracellular organelles, showed that PS comprises 15% of total phospholipids (Dodge & Phillips, 1967), suggesting that PS comprises about 30% of the total phospholipids in the cytosolic leaflet of the PM. Therefore, the abundance of PS in the cytosolic leaflet of RE membranes could be comparable or even higher than that in the PM. This high concentration of PS in the cytosolic leaflet of RE membranes, which is produced by a P₄-ATPase ATP8A1, might be enough to recruit EHD1 specifically to REs, since the half-maximal binding of EHD1 to PS was observed at ∽50% PS in the liposome co-sedimentation assay.

Despite the increase of PS in the luminal leaflet in ATP8A1-depleted cells, PS was still present in the cytosolic leaflet of RE membranes as the PS probe expressed in the cytosol significantly stained REs in ATP8A1-depleted cells. The presence of PS in the cytosolic leaflet of RE membranes was also supported by the observation that evectin-2, a protein that requires PS in the cytosolic leaflet for its RE localization (Uchida et al, 2011), localized at REs in ATP8A1-depleted cells (Supplementary Fig S12). Thus, although ATP8A1 contributes to the enrichment of PS in the cytosolic leaflet of RE membranes, its knockdown does not result in the complete disappearance of PS at the cytosolic leaflet. Scramblases that mediate bidirectional transbilayer movement of phospholipids or other P₄-ATPases may participate in the exposure of PS at the cytosolic leaflet in ATP8A1-depleted cells. PS is synthesized in the cytosolic leaflet of the ER and likely moves via non-vesicular routes (Maeda et al, 2013) and/or vesicular routes to the PM and other organelles. These processes (Holthuis & Menon, 2014) may also contribute to the localization of PS in the cytosolic leaflet of the RE membranes.

EHD proteins contain an amino-terminal ATP-binding G-domain, followed by a helical domain and a carboxy-terminal EH domain (Grant & Caplan, 2008). The crystal structure of full-length EHD2 was solved (Daumke et al, 2007). Dimerization of G-domain creates a putative membrane-binding surface with a polybasic stretch within the helical domain (residues 285–400). EHD2 mutants (F322A, K324D, K327D, K328D, or K329D) display a completely cytoplasmic distribution, showing the critical role of these amino acids in the helical domain in membrane localization of EHD2 (Daumke et al, 2007). EHD1 (K324D) lost its membrane localization in COS-1 cells similar to EHD2 (K324D) and did not bind to PS-liposomes in vitro (Supplementary Fig S13), which supports the idea that PS is essential for binding of EHD1 to the membranes. We noticed that when expressed in COS-1 cells, GFP-EHD1 localized at perinuclear REs and cytoplasmic tubular structures (Supplementary Fig S13). The latter was not obvious when endogenous EHD1 was immuno-stained (Fig3). The tubular structures may correspond to ‘tubular recycling compartments’ in HeLa cells (Caplan et al, 2002). EHD1 that lacks EH-domain (ΔEH) still localized at perinuclear REs, but not at cytoplasmic tubular structures (Supplementary Fig S13). This observation is consistent with the previous report that ΔEH did not localize at tubular recycling compartment in HeLa cells (Caplan et al, 2002). However, ΔEH mutant bound to PS-liposomes in vitro (Supplementary Fig S13). These results suggest that PS is not the sole determinant for EHD1 localization at cytoplasmic tubular structures. Recruitment of EHD1 to tubular structures in HeLa cells requires the interactions between EH domain and NPF-containing proteins, such as MICAL-L1 (Sharma et al, 2009) and Pacsin (Giridharan et al, 2013). It is also proposed that MICAL-L1 and Pacsin bind to phosphatidic acid in the tubular structures (Giridharan et al, 2013). It is tempting to speculate that there are two distinct recycling compartments, perinuclear REs to which EHD1 is recruited primarily by PS and tubular recycling compartment to which EHD1 is recruited cooperatively by phosphatidic acid and PS.

Recent studies have revealed the significance of PS in the cytosolic leaflet of endosomes (reviewed in Bigay & Antonny, 2012; Holthuis & Menon, 2014). As mentioned above, S. cerevisiae Gcs1 has a +ALPS-motif that binds highly curved membranes with anionic phospholipids that include PS (Xu et al, 2013). Gcs1 localizes at the late Golgi/endosomes and its localization is regulated by a P₄-ATPase Drs2, suggesting that PS flipped by Drs2 is essential for Gcs1 recruitment. K-Ras is a major proto-oncogene product (Hancock, 2003). K-Ras mainly localizes at the PM, but some of it localizes at endosomes (Lu et al, 2009; Gelabert-Baldrich et al, 2014). K-Ras has a polybasic stretch at the carboxy-terminus that electrostatically interacts with PS (Yeung et al, 2008). This interaction was recently shown to be essential for K-Ras trapping at REs, from where vesicular transport delivers K-Ras to the PM (Schmick et al, 2014). Evectin-2 that localizes at REs has a PH domain that binds PS but not PIPs. This PH domain is essential for the RE localization of evectin-2 and the regulation of the retrograde traffic from REs to the Golgi (Uchida et al, 2011). These findings, together with the present study, indicate that PS in REs has multifaceted functions in membrane traffic, such as cargo collection, membrane budding, and membrane fission. Other membrane systems, including PM, EEs, LEs, and the Golgi, exploit PIPs to exert their own functions through a variety of PIP effectors (Behnia & Munro, 2005; Di Paolo & De Camilli, 2006). Importantly, the levels of PIPs are tightly regulated by phosphorylation and dephosphorylation. As far as we know, the metabolic conversion of PS is confined to the ER and/or mitochondria in mammalian cells (Vance & Steenbergen, 2005). Therefore, in other cellular membranes, cells may exploit the lipid flipping system by P₄-ATPases to regulate the local levels of cytosolic PS. It is also noted that Drs2p in the trans-Golgi network is activated by PI(4)P (Natarajan et al, 2009), suggesting that PIPs could control the PS content in the cytosolic leaflet of the membranes.

ATP8A2 is a paralogue of ATP8A1, and mutations of ATP8A2 cause axonal degeneration in mice (wabbler-lethal mice) (Zhu et al, 2012) and are associated with neurodegenerative disease CAMRQ in humans (Cacciagli et al, 2010; Onat et al, 2013). However, cellular mechanism of ATP8A2 has been unclear. The present study showed that ATP8A2 localized at REs and could compensate for the loss of ATP8A1 when expressed in COS-1 cells. In contrast, the disease-causative human ATP8A2 mutant (I376M), which lacked ATPase and flippase activities, could not rescue the loss of ATP8A1. Furthermore, primary cortical neurons isolated from Atp8a2^−/− mice showed a drastic reduction of TfnR present in the PM, with the total amount of TfnR being comparable to that in primary cortical neurons isolated from Atp8a2^+/+ mice. These findings suggest that ATP8A2 functions in endosomal membrane traffic through REs. ATP8A2 in photoreceptor cells is localized both in the perinuclear compartment of the inner segment and in the disk membranes of the outer segment (Coleman et al, 2009, 2014). ATP8A2 in the perinuclear compartment may produce the lipid asymmetry in the bilayer to facilitate vesicle trafficking to the disk membrane, which could explain a drastic reduction in the length of outer segments in ATP8A2-deficient mice (Coleman et al, 2014). Since disk membranes have a symmetrical distribution of PE and PS (Wu & Hubbell, 1993; Hessel et al, 2000), presumably due to significant scramblase activity of rhodopsin (Menon et al, 2011), ATP8A2 does not appear to play a role in generating global lipid asymmetry in disk membranes. Some portion of ATP8A2 may be transported to the disk membranes as a result of vesicle trafficking from the perinuclear compartment.

Mutations in the very low-density lipoprotein receptor (VLDLR) is also linked to CAMRQ (Ozcelik et al, 2008; Turkmen et al, 2008). VLDLR is a receptor for Reelin, an extracellular protein that guides neuronal migration in the cerebral cortex and cerebellum (Herz & Chen, 2006) and circulates between the PM and endosomes by endocytic recycling (Duit et al, 2010). Therefore, recycling traffic of VLDLR to the PM might be impaired in neurons without functional ATP8A2. Available antibodies failed to detect VLDLR protein in primary cortical neurons used in this study. Further studies are required to address this issue.

Materials and Methods

Antibodies

The following antibodies were purchased from the manufacturers as noted: rabbit anti-ATP8A1 and rabbit anti-ARF1 (Proteintech); mouse anti-1D4 (Millipore); mouse anti-α-tubulin, rabbit anti-GM130, and mouse anti-myc (9E10) (Sigma); mouse anti-PI(4)P (Echelon); mouse anti-GM130 and mouse anti-LAMP1 (H4A3) (BD Biosciences); mouse anti-human TfnR (H68.4; Zymed Laboratories); rabbit anti-EHD1 (epitomics); mouse anti-LAMP2 (Santa Cruz); mouse anti-CD63 (Cymbus Biotechnology); rabbit anti-myc (Upstate); goat anti-VPS26 (Everest biotech); mouse anti-6xHis (Wako); rabbit anti-syntaxin 5 (Synaptic Systems); mouse anti-TfnR antibody (13E4) (Abcam); Alexa 488-, 594-, or 647-conjugated secondary antibodies (Invitrogen); and sheep anti-mouse IgG antibody-HRP and donkey anti-rabbit IgG antibody-HRP (GE Healthcare).

Reagents

The following reagents were purchased from the manufacturers as noted: cholesterol (C8503), sulfatide, PA, ethanolamine, ATP and n-octyl-β-D-glucopyranoside (Sigma); and PI (Echelon); di16:0-PIPs (CellSignals); PS, PE, and PC (porcine brain) for liposome cosedimentation assay, 1,2-dioleoyl-PC, 1,2-dioleoyl-PE, 1,2-dioleoyl-PS, 1,2-dioleoyl-PI, 1,2-dioleoyl-PG, 1,2-dioleoyl-PA, brain polar lipids (porcine), sphingomyelin (porcine brain), cholesterol (ovine wool), C6-NBD-PS and C6-NBD-PE (Avanti Polar Lipids); dithionite (Fisher); CHAPS (Anatrace); and Alexa 594-Tfn, Alexa 594-CTxB, Alexa 555-EGF (Invitrogen). To prepare Alexa 488-Tfn, human holo-Tfn (Sigma) was conjugated with Alexa 488 using Alexa Flour succinimidyl ester (Invitrogen) and then purified by PD-10 desalting columns (GE Healthcare).

Plasmids

The following genes were amplified by polymerase chain reaction (PCR) with cDNA derived from HeLa cells using the following primers: 5′-AGCCGCCTCGAGCATGCCCACCATGCGGAGGAC-3′ (ATP8A1; sense primer, XhoI site is underlined) and 5′-AGGCTCCCCGGGTCACCATTCGTCGGGCCTC-3′ (ATP8A1; antisense primer, SmaI site is underlined); 5′-GCCTCGAGATGTTCAGCTGGGTCAGC-3′ (EHD1; sense primer, XhoI site is underlined) and 5′-GCGAATTCTCACTCATGTCTGCGCTTG-3′ (EHD1; antisense primer, EcoRI site is underlined); 5′-TGGCAGGAATTCATGATGGATCAAGCTAGATC-3′ (TfnR; sense primer, EcoRI site is underlined) and 5′-GGTATCGGATCCCAAACTCATTGTCAATGTCCC-3′ (TfnR; antisense primer, BamHI site is underlined); 5′-GGAAGACCTGGAATTCATGGCGATGAACTATAACG-3′ (CDC50A; sense primer, EcoRI site is underlined) and 5′-TAGATGCATGCTCGAGTTAAATGGTAATGTCAGC-3′ (CDC50A; antisense primer, XhoI site is underlined); 5′-GCAGATCTCATGGGCACCCGCGACGACGA-3′ (Rab11; sense primer, BglII site is underlined) and 5′-GCGTCGACTTAGATGTTCTGACAGCACT-3′ (Rab11; antisense primer). Only one isoform (NM_001105529) of ATP8A1 was amplified, which corresponds to bovine isoform β1 (Ding et al, 2000). The products encoding ATP8A1 and EHD1 were introduced into pEGFP-C3 and pEGFP-C2 (Clontech), respectively, to generate N-terminal GFP-tagged constructs. The product encoding TfnR was introduced into pEGFP-N2 (Clontech) to generate a C-terminal GFP-tagged construct. The product encoding CDC50A was introduced into pcDNA with myc-tag (Invitrogen) to generate an N-terminal myc-tagged construct. The product encoding Rab11 was introduced into pmCherry-C3 (generated by the replacement of the EGFP sequence in pEGFP-C3 with an mCherry sequence) to generate an N-terminal mCherry-tagged construct. For the preparation of siRNA-resistant N-terminal GFP-tagged ATP8A1 expression vector, six silent mutations were introduced by PCR in the ATP8A1 siRNA1 target sequence (see below) using 5′-GCACCTCGAATCTGAATGAGGAACT-3′(sense primer) and 5′-GCGCCATCGCCGCAGTGTCTGTGGG-3′ (antisense primer). ATP8A1 E191Q mutant was constructed by site-directed mutagenesis of siRNA-resistant wild-type ATP8A1. Human ATP8A2 (Coleman et al, 2009) was introduced into pEGFP-C2 using In-Fusion HD Cloning Kit (TAKARA). ATP8A2 (E210Q or I376M) and EHD1 (K324D) were constructed by site-directed mutagenesis of wild-type ATP8A2 and EHD1, respectively. GFP-EHD1 ΔEH (1-438) was generated using 5′-GCCTCGAGATGTTCAGCTGGGTCAGC-3′ (sense primer, XhoI site is underlined) and GCGAATTCTCACTCCACGTCGTCGATGC (antisense primer, EcoRI site is underlined).

Cell culture

COS-1 cells were maintained in DMEM with 10% FBS. PSA-3 and R-41 cells were cultured in Ham's F-12 medium as previously described (Lee et al, 2012).

Plasmid transfection

Cells were transiently transfected with plasmids using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions.

RNA interference

siRNA duplex oligomers were designed (Nippon EGT) as follows: GCUAUGGCUCGAACAUCUA (ATP8A1 siRNA1), CUCUGACUACUCCAUAGCU (ATP8A1 siRNA2), GGUGUAUAGGUCUCUAUAA (ATP8A1siRNA3), GUUUCCUCGCCUCUCGAAA (EHD1 siRNA1), and CCGUCACUCCAUACAGUAU (EHD1 siRNA2). A total of 20 nM siRNA was introduced to COS-1 cells using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's instruction. After 4 h, the medium was replaced by DMEM with 10% FBS and cells were further incubated for 72 h for subsequent experiments. Cells treated only with Lipofectamine RNAiMAX are defined as control cells.

Immunocytochemistry

Immunocytochemistry was performed as described previously (Uchida et al, 2011). TfnR staining was performed using anti-TfnR antibody (H68.4) unless indicated otherwise. For the stain of endogenous EHD1, cells were fixed with pre-warmed 2% para-formaldehyde in PBS at room temperature for 10 min, quenched with 50 mM NH₄Cl, and permeabilized with 0.1% Triton X-100 in PBS at room temperature for 5 min. After blocking with 3% BSA in PBS, cells were incubated with rabbit anti-EHD1 antibody. For ARF1 staining, fixed and permeabilized cells were incubated with the primary antibody in Can Get Signal immunostain solution A (TOYOBO). For LAMP1 and PI4P staining, fixed cells were permeabilized with 20 μM digitonin in PBS for 5 min at room temperature before blocking. Rab11 staining was performed as previously described (Lee et al, 2012).

Establishment of COS-1 cells stably expressing GFP-lact-C2

GFP-lact-C2 was amplified from lact-C2 in pEGFP-C2 vector (Uchida et al, 2011) using 5′-GCAGACAAGCTTCGCCACCATGGTGAGCAAGG-3′ (sense primer) and 5′-GCAGACATCGATCTAACAGCCCAGCAGCTCC-3′ (antisense primer), and subsequently cloned into pLHCX (Clontech) at HindIII/ClaI. Recombinant retrovirus harboring GFP-lact-C2 was prepared using a retrovirus packaging kit ampho (Takara-bio) according to the manufacturer's instructions. In brief, pLHCX/GFP-lact-C2 was transfected into HEK293T cells in combination with pGP (encoding gag and pol) and pE-ampho (encoding amphotropic env). The conditioned medium containing recombinant retrovirus particle was collected, then filtrated with 0.4-μm membrane filter to remove cell debris, and stored at 4°C before use. COS-1 cells were infected with EGFP-lact-C2 retrovirus in DMEM with 10% FBS and 8 μg/ml polybrene, and cells were subjected to drug-selection with 200 μg/ml hygromycin for 2 weeks. The hygromycin-resistant colonies were then picked up.

Visualization of PS at the luminal leaflet

COS-1 cells that were transiently or stably expressing GFP-lact-C2 were fixed with 3.7% formaldehyde in PBS at room temperature for 15 min and permeabilized with 0.1% saponin in PBS at room temperature for 10 min. After blocking with 3% BSA in PBS, cells were treated with recombinant 2xPH-6xHis (2 μg/ml), mouse anti-6xHis tag (1:1,000), and rabbit anti-GM130 at 4°C overnight. After washing with PBS, cells were incubated with anti-mouse IgG conjugated with Alexa 594 and anti-rabbit IgG conjugated with Alexa 647 at room temperature for 1 h, washed with PBS, and then mounted. Intensity of recombinant 2xPH-6xHis in RE area was quantified with ImageJ (NIH).

Streptolysin O treatment

Cells were fixed with 2% PFA in PBS and incubated at 4°C for 15 min with 200 units/ml Streptolysin O (Sigma) in ice-cold PBS containing 10 mM DTT. Cells were then washed with ice-cold PBS containing 10 mM DTT twice, incubated in PBS containing 10 mM DTT at 37°C for 20 min., and washed with PBS.

Confocal microscopy

Confocal microscopy was performed using a LSM510 META (Carl Zeiss Microimaging) with a 63 × 1.4 Plan-Apochromat oil immersion lens, or a TCS SP8 (Leica) with a 63 × 1.2 Plan-Apochromat water immersion lens. With a TCS SP8, excitation was done with a 65 mW Argon laser emitting at 488 nm. Emissions were collected from 500 to 600 nm using Spectral detector.

Tfn recycling assay

COS-1 cells were first incubated in serum-free medium for 30 min at 37°C and subsequently in medium containing 200 μg/ml Alexa 488 or 594-Tfn for 60 min. After washing with ice-cold PBS, cells were incubated in ‘acid-wash buffer’ (20 mM sodium-acetate buffer, 1 mM CaCl₂, 150 mM NaCl, pH 4.8) on ice for 5 min to remove surface-bound transferrin. Cells were washed with ice-cold PBS and chased in phenol red-free DMEM containing 400 μg/ml unlabeled human holo-Tfn at 37°C with 5% CO2. During the chase, cells were imaged every 3 or 6 min for 1 h with a TCS SP8. The perinuclear fluorescence intensity of Alexa 488 or 594 minus the background fluorescence intensity was quantified with ImageJ.

Live imaging for TfnR-GFP in xyt mode

COS-1 cells expressing TfnR-GFP were placed in phenol red-free DMEM/10% FBS. Cells were imaged in xyt mode at 37°C with 5% CO₂ using a TCS SP8. After the photobleach with 100% power of 488 nm laser for 10 s exposure, the images were collected every 0.5 s for 1–2 min.

Epifluorescence microscopy imaging system based on an inverted microscope (IX81-ZDC; Olympus), 150 × NA 1.45 TIRFM objective (Olympus), motorized filter wheel (Olympus), electron multiplying charge coupled device camera (iXon; Andor), 488 nm laser line and a MetaMorph software (MDS Analytical Technologies). The emission filter 531/46 nm with 488 nm laser for acquisition of EGFP-streaming images was used.

Expression and purification of ATP8A1 and ATP8A2

Expression, purification, and reconstitution of ATP8A1 (WT and E191Q) and ATP8A2 (WT and I376M) containing the nine amino acid 1D4 C-terminal tag were performed as described previously Coleman et al (2009).

ATPase assay

ATPase activity was measured as described Coleman et al (2009). In brief, immunoaffinity-purified ATP8A1 or ATP8A2 in PC with increasing amounts of PS was assayed in 50 mM HEPES-NaOH (pH 7.5), 150 mM NaCl, 12.5 mM MgCl₂, 1 mM DTT, 5 mM ATP, and 10 mM CHAPS at 37°C. The total lipid concentration was kept constant at 2.5 mg/ml. Assays were terminated by the addition of 6% (wt/vol) SDS. The release of phosphate from ATP was determined using the colorimetric microplate method (Gonzalez-Romo et al, 1992).

Flippase assay

Flippase assays were performed as described (Coleman et al, 2009; Coleman & Molday, 2011) with ATP8A1 or ATP8A2 reconstituted in liposome consisting of 97.5% PC and 2.5% fluorescently labeled (NBD) PS or PE.

Expression and purification of recombinant proteins

WT or mutant EHD1 cDNA was subcloned into pET-45b (Novagen), and the N-terminal His-tagged protein was expressed in E. coli Arctic Express (DE3) RP cells (Agilent Technologies). Bacterial cultures in LB medium were induced with 1 mM IPTG and grown overnight at 13°C. The protein was purified using a HisTrap HP column (GE Healthcare) according to the manufacturer's instructions and then dialyzed against 20 mM HEPES-NaOH (pH 7.5) containing 300 mM NaCl, 1 mM MgCl₂, and 1 mM DTT.

The sequence encoding a tandem fusion of evt-2 PH domain (2xPH) was introduced into pET-21a (Novagen), and the C-terminal His-tagged protein was expressed in E. coli BL21 (DE3) cells. Bacterial cultures in LB medium were induced with 1 mM IPTG and grown overnight at 20°C. The protein was purified using a HisTrap HP column and then dialyzed against PBS.

Liposome co-sedimentation assay

The liposome co-sedimentation assay using His-tagged 2xPH was performed as previously described (Uchida et al, 2011) with modifications. Lipid mixtures were dried under nitrogen gas and hydrated in 25 mM HEPES-NaOH (pH 7.5) containing 100 mM NaCl and 0.5 mM EDTA for 90 min at 37°C, and vortexed briefly. To remove protein aggregates, the protein solution was subjected to centrifugation at 100,000 g for 30 min at 4°C. 10 μg of proteins were incubated with 65 nmol liposomes in the buffer (50 μl) for 15 min at room temperature, and the mixture was centrifuged at 100,000 g for 30 min at 20°C. The resultant supernatant and pellet were subjected to SDS–PAGE, and the proteins and lipids were stained with Coomassie blue. Lipid stain was observed just below bromophenol blue. As a negative control, a sample without liposomes was subjected to the same procedure.

The assay using His-tagged EHD1 was performed as described above with the following modifications: (i) dried lipid mixtures were hydrated in 20 mM HEPES-NaOH (pH 7.5) containing 100 mM NaCl, 1 mM MgCl₂, and 1 mM DTT; (ii) the protein solution was subjected to centrifugation at 100,000 g for 15 min at 4°C to remove aggregates; (iii) 7 μg of proteins were incubated with 50 nmol liposomes.

The intensities of individual bands were quantified with ImageJ. The band intensity in the pellet fraction of the sample without liposomes (a) was subtracted from that of the sample with liposomes (Ippt) to calculate the amount of bound protein (Ippt − a). The percentage of the bound protein was obtained by the following equation: %bound = 100 × (Ippt − a) / [Isup + (Ippt − a)], where Isup represents the band intensity in the supernatant fraction. The binding data from two independent experiments were analyzed by fitting to the ‘one-site specific binding with Hill slope’ equation in GraphPad Prism 5 (GraphPad Software). Y = B_max × Xh / (kh + Xh), where k and h represent the concentration of PS resulting in half-maximal binding and the Hill coefficient, respectively.

To prepare liposomes of decreasing size, dried lipid mixtures containing 1% (mol/mol) rhodamine-labelled PE (Avanti) were hydrated in 20 mM HEPES-NaOH (pH 7.5) containing 100 mM NaCl, 1 mM MgCl₂, and 1 mM DTT and then dipped into liquid nitrogen for rapid cooling. After completely frozen, the lipids were transferred to a bath at 37°C for thawing. After completely thawed, the lipids were frozen again. This freeze–thaw operation was repeated three times. The lipids were then extruded through 400 nm or 100 nm polycarbonate filters (Whatman) at room temperature using Avanti mini-extruder and stored on ice for further experiments. After centrifugation with His-EHD1, the fluorescence of each supernatant was measured, and the amount of precipitated liposomes was calculated. The percentage of precipitated liposomes was used to calculate the percentage of liposome-bound His-EHD1 in Fig5D and E. Approximately 70–95% of liposomes were precipitated.

Liposome size measurement by dynamic light scattering (DLS)

The liposome size was measured using a Zetasizer Nano ZS (Malvern) equipped with a 4 mW He–Ne laser (633 nm) and disposable cuvettes (ZEN0112, Malvern) at 25°C. The z-averaged diameter of the particles was calculated from the intensity of the scattered light using the Malvern Zetasizer software. The data of all samples met quality criteria assuming monomodal size distributions.

Western blotting

Western blotting was performed as described previously (Uchida et al, 2011).

CTxB retrograde traffic assay

CTxB retrograde traffic assay was performed as described previously (Uchida et al, 2011).

EGF traffic assay

Cells were pre-incubated in serum-free DMEM for 30 min at 37°C and then pulsed with 2 μg/ml Alexa 555-EGF in serum-free DMEM for 5 min at 37°C. After washing, cells were chased in DMEM containing 10% FBS for 60 min at 37°C, fixed, and stained for LAMP1. Pearson's coefficient between Alexa 555-EGF and LAMP1 was determined with ImageJ.

Cortical neuron culture

Cortical neurons were obtained from Atp8a2^+/+ or Atp8a2^−/− mice (embryonic day 15–17) as previously described (Koie et al, 2014). In brief, 6-well plate dishes were coated with 0.01% poly-L-lysine hydrobromide (Sigma) in PBS for overnight at 37°C and subsequently washed four times with water. Cerebral cortices were trypsinized for 5 min at 37°C, treated with 2 U/μl DNase I (Invitrogen), and dissociated by triturating with fire-polished Pasteur pipette. The neurons were cultured in Neurobasal medium (Invitrogen) containing 2% B27 supplement (Invitrogen), 2 mM Glutamax (Invitrogen), and antibiotics for 4 days.

Biotinylation assay

For surface biotinylation, cortical neurons (4 DIV) were cooled on ice, washed three times with ice-cold PBS containing 1 mM CaCl₂ and 0.5 mM MgCl₂, and then incubated with PBS containing 1 mM CaCl₂, 0.5 mM MgCl₂, and 1 mg/ml EZ-link Sulfo-NHS-SS-Biotin (Thermo) for 40 min at 4°C. Unreacted biotin was quenched by washing cells with ice-cold PBS containing 100 mM glycine, and then cells were washed twice with ice-cold PBS. Cells were harvested in lysis buffer (20 mM Tris-HCl pH 7.4, 5 mM EDTA, 100 mM NaCl, 1% Triton X-100, 0.1% SDS) with complete protease inhibitor cocktail (Roche). Homogenates were centrifuged at 13,000 rpm for 10 min at 4°C. 10% of the resulting supernatant was stored as the total fraction. The remaining 85% of the supernatant was rotated for 1 h at 4°C with NeutrAvidin agarose resin (Thermo) and centrifuged at 500 g for 1 min at 4°C. Precipitates were washed three times with wash buffer (25 mM HEPES, 150 mM NaCl, 5 mM EDTA, 0.1% Triton X-100) and stored as the surface fraction. Both total and surface protein samples were boiled for 5 min at 95°C. Western blotting was performed and band intensities were quantified with ImageJ. The ratio of surface to total TfnR or tubulin was calculated, and that of tubulin was subtracted from that of TfnR to exclude the amount of cytoplasmic protein in the surface fraction. The calculated value was then normalized to that of Atp8a2^+/+.

Statistical analysis

Statistical comparisons of means were made using two-tailed Student's t-test or one-way ANOVA. P < 0.05 was used as the criterion for statistical significance.

Author contribution

SL, YU, JW, TKi, and KM performed experiments; TM, TN, and TKo provided reagents; SL, YU, TI, RM, TT, and HA designed research; SL, YU, TT, and HA wrote the manuscript.

Conflict of interest

The authors declare that they have no conflict of interest.