Structures of a P4-ATPase lipid flippase in lipid bilayers

Type 4 P-type ATPases (P4-ATPases) are a group of key enzymes maintaining lipid asymmetry of eukaryotic membranes. Phospholipids are actively and selectively flipped by P4-ATPases from the exoplasmic leaflet to the cytoplasmic leaflet. How lipid flipping is coupled with ATP-hydrolysis by P4-ATPases is poorly understood. Here, we report the electron cryo-microscopy structures of a P4-ATPase, Dnf1-Cdc50 from Chaetomium thermophilum, which had been reconstituted into lipid nanodiscs and captured in two transport intermediate states. The structures reveal that transmembrane segment 1 of Dnf1 becomes highly flexible during lipid transport. The local lipid bilayers are distorted to facilitate the entry of the phospholipid substrates from the exoplasmic leaflet to a cross-membrane groove. During transport, the lipid substrates are relayed through four binding sites in the groove which constantly shields the lipid polar heads away from the hydrophobic environment of the membranes.


Introduction 31
Phospholipid molecules are unevenly distributed in membrane bilayers of eukaryotic cells 1,2 . 32 Phosphatidylethanolamine (PE) and phosphatidylserine (PS) are concentrated in the cytoplasmic leaflet, 33 whereas phosphatidylcholine (PC) is enriched in the exoplasmic leaflet (lumenal or extracellular leaflet) 3 . 34 The asymmetric distribution of phospholipids is critical for numerous cellular processes, such as cell 35 signaling, apoptosis, and blood coagulation 4,5 . Lipid asymmetry is created and maintained by ATP-36 driven lipid translocases, such as lipid flippases that move specific lipids from the exoplasmic leaflet to 37 the cytoplasmic leaflet, and lipid floppases which move lipids from the cytoplasmic to the exoplasmic 38 leaflet. Lipid asymmetry can be disrupted by lipid scramblases which, upon activation, facilitate the rapid 39 bi-directional movement of lipids across the two leaflets. Extensive studies in the past few years revealed 40 that type 4 P-type ATPases (P4-ATPases) are the phospholipid flippases, a group of ATP-binding cassette 41 (ABC) transporters are the floppases, and transmembrane protein 16F (TMEM16F) could function as the 42 non-specific lipid scramblase 6 . Among the lipid translocases, P4-ATPases belong to the P-type ATPase 43 family. Most members in the family transport small metal ions across membranes, whereas P4-ATPases 44 specifically transport phospholipids that are at least ten times larger than the metal ions 7 . Therefore, it 45 was postulated that P4-ATPases had a distinct mechanism from other P-type ATPases for substrate 46 transport [8][9][10][11] . 47 Fig. S10c). The positive charges are mainly contributed by K174, R181, and K1121 on the TMs. The 238 highly conserved K1121 has been shown to be important for substrate binding and ATPase activity 19 . 239 Consistent with biochemical data 19 , the groove has high affinity to the lipid substrate in the E2P state as 240 evident by the strong phospholipid density, whereas the groove shows weak lipid binding to facilitate 241 lipid entry and exit in the E1-ATP state as indicated by the fragmented lipid density in the structure. 242 Similarly, a hydrophilic membrane-traversing groove is also present in the TMEM16F scramblase 36 . 243 The "hydrophobic gate" model suggests that TMs 1 and 2 move away from TMs 3 and 4 during 244 lipid transport 22 . Indeed, our structures show that TM1 and TM2 becomes flexible in E1-ATP. The key 245 residue, I364 of the "hydrophobic gate" (I554 in ctDnf1), is at the interface between TMs 1, 2, and 4 ( Fig.  246   S11). Thus the mutations of the residue may disrupt the E1-E2 equilibrium and hamper lipid flipping as 247 observed in the mutagenesis studies 22 . The "two-gate" model suggests that the flippases recognize the 248 phospholipid substrates by interacting with the head groups. Residues other than the classical ion binding 249 residues in ion-pumping P-type ATPases are involved in recognition 16,20,21,37 . Consistent with the model, 250 the four distinct binding sites in our structures mainly interact with the lipid head groups. However, the 251 sites do not seem to provide a discrimination mechanism for different phospholipid substrates. As shown 252 in the phospholipid-dependent ATPase activity assays, ctDnf1 may have different substrate specificity 253 from scDrs2 and hATP8A1. The amino acids that interact with the polar head group of the phospholipid 254 substrate at E2-site1 are similar to those in the structures of scDrs2 and hATP8A1. The corresponding 255 residues are Q549 and N550 of ctDnf12, S503 and N504 of scDrs2, and N352 and N353 of hATP8A1 256 (Fig. S1). The clamp residues of E2-site2 consist of both hydrophilic and hydrophobic residues, but are 257 not conserved among P4-ATPases (Fig. S1). The serine residues at E1-site1 are highly conserved among 258 P4-ATPases, and E1-site2 only provides a steric opening. Further studies on other intermediate states may 259 provide clues on the substrate specificity. 260 In summary, the structures of ctDnf1-Cdc50 suggest that P4-ATPases have evolved a unique 261 mechanism for lipid flipping. For ion-pumping P-type ATPases, most of TMs are involved in 262 coordinating ion movements in the membranes. In contrast, TMs 3-10 of P4-ATPases are kept in a fixed 263 conformation by Cdc50 during ATP hydrolysis. TMs 1 and 2 are the major regulators for lipid flipping. 264 The distorted lipid bilayers and the groove bordered by TMs 2, 4, and 6 may be the key factors in 265 controlling lipid flipping. It remains to be seen how the P4-ATPases select specific lipids to enter and exit 266 the flipping pathway.

284
The authors declare no competing interests. 285

286
Protein expression and purification 287 The genes of ctDnf1 and ctCdc50 were cloned from the cDNA library of Chaetomium 288 thermophilum (var. thermophilum strain: DSM1495, a gift from Dr. Stefan Schoebel). Superfolder green 289 fluorescence protein (sfGFP) 38 , a Twin-Strep tag and a 3C protease cleavage site were fused to the N-290 terminus of ctDnf1. sfGFP, a His 9 tag, and a 3C protease cleavage site were fused to the N-terminus of 291 ctCdc50. The expression plasmids pRS426-sfGFP-twinStrep-3C-CtDnf1 and pRS424-sfGFP-His 9 -3C-292 CtCdc50 were co-transformed into S. cerevisiae strain BJ5465 using the LiAc/SS carrier DNA/PEG 293 method 39 . Yeast cells were cultured in synthetic drop-out medium supplemented with 2% raffinose at 294 30 °C for about 24h to reach an optical density (OD 600 ) of about 5. The culture was induced by the 295 addition of 2% galactose and continued for 20 h at 25°C. The cells were harvested and stored at −80 °C 296 until use. 297 The cells were suspended in the membrane extraction buffer (20 mM Tris-HCl pH 7.4, 150 mM 298 NaCl, 5mM MgCl 2 ,1mM DTT, and protease inhibitor cocktails) and lysed by high pressure 299 homogenization. The crude lysate was clarified by centrifugation (20,000×g, 25 min, 4°C). The 300 membrane fraction was pelleted by ultracentrifugation (200,000×g, 1 h, 4°C) and washed once with the 301 membrane extraction buffer. The membrane pellets were solubilized in 2% lauryl maltose neopentyl 302 glycol (LMNG, Anatrace) in the membrane solubilization buffer (20 mM Tris-HCl pH 7.4, 150 mM 303 NaCl, 5mM MgCl 2 ,1mM DTT, 10% glycerol, and protease inhibitor cocktails). After incubation at 4 °C 304 for 1 h, the solution was clarified by ultracentrifugation (200,000×g, 1 h, 4°C). The supernatant was 305 mixed with avidin (Sigma) and loaded onto a column pre-packed with StrepTactin resin (IBA 306 Lifesciences). The eluents were concentrated and incubated with 3C protease at 4°C overnight. The 307 protein solution was then loaded onto a Superdex 200 10/300 column (GE Healthcare). The peak fractions 308 were pooled and concentrated (Fig. S2). The purified protein was either reconstituted into nanodiscs or 309 flash-frozen in liquid nitrogen and stored at −80 °C. 310 The purified protein was mixed with MSP1D1 40 and yeast polar lipids (Avanti Lipids, 40 mg/ml 311 dissolved in 1% DDM) at a molar ratio of 1:2:25. Bio-beads SM2 (Bio-Rad) were then added to the 312 mixture and incubated at 4 °C overnight to remove detergents. The complex was further purified by size-313 exclusion chromatography on a Superdex 200 10/300 column. The peak fraction had a protein 314 concentration of 1.0 mg/ml (Fig. S2). It was immediately used for cryo-EM sample preparation without 315 concentrating. 316

Cryo-EM sample preparation and data collection 317
The freshly prepared samples were incubated with 1mM BeF 3 or 1mM AMPPCP on ice for 318 30min before vitrification. The cryo-grid preparation was performed at 4 °C and 100% humidity in an FEI 319 Vitrobot Mark IV. 4 µl sample was applied to each freshly glow-discharged grid (Quantifoil, R1.2/1.3). 320 The grids were then plunge-frozen in liquid ethane. The cryo-grids were screened with a 200 kV FEI 321 Talos Arctica microscope equipped with a FEI Ceta camera. The data were collected on a 300 kV FEI 322 Titan Krios TEM with a K2 summit camera and GIF Quantum energy filter (Gatan). The images were 323 collected at a magnification of 130,000× with a calibrated pixel size of 1.055 Å. The dose rate was set at 324 8 e − /s/pixel and the exposure time was 8 s, corresponding to a total dose of 57.5 e − /Å 2 . Movie stacks (32 325 frames each) were recorded with the software SerialEM 41 under low-dose conditions with defocuses 326 ranging from -1 to -2 μm. 327

Image processing 328
The movie stacks were subject to motion correction and electron-dose weighting by using 329 MotionCor2 42 (Fig. S3a, S4a). The program Gctf 43 was used to estimate the contrast transfer function 330 (CTF) parameters. Images of high quality were selected for further image processing on the basis of the 331 CTF power spectra of the corrected images. The following calculations are performed with RELION3.0 44 . 332 Particles of high quality were selected according to 2D classification (Fig. S3b, S4b) and 3D classification 333 results. The selected particles were subject to several rounds of CTF refinement and polishing. After 334 mask-based post-processing, the final maps had resolutions of 3.40 Å and 3.48 Å for the AMPPCP and 335 BeF 3 samples, respectively (Fig. S3, S4). All the resolution estimations were based on gold-standard 336 Fourier Shell Correlation (FSC) 0.143 criteria. 337 The model for the E2P (BeF 3 -) structure was built manually in Coot 45 , with the guidance of the scDrs2 338 structures. The model was refined in real space using Phenix 46 . For the model building of E1-ATP 339 (AMPPCP), the E2P model was fit in the E1-ATP density map. Each domain is subject to rigid body 340 refinement. Due to the local resolution limits, the A and N domains were not refined further. The rest 341 parts of the E1-ATP model were refined in real space with Phenix. Model validation was done with 342 ATPase activity assay 344 The ATPase activity assays were carried out by using BIOMOL® Green (Enzo) to measure the 345 free phosphate concentrations. The reaction solutions consisted of 0.05mg/ml protein, 0.01% LMNG, 346 0.02% C 12 E 9 (Anatrace), 150 mM NaCl, 20 mM HEPES-NaOH pH 7.5, 5mM MgCl 2 , 1mM DTT, 2.5mM 12 ATP, and lipids at the indicated concentrations. The reactions were carried out at 30 °C for 20 min, and 348 then immediately diluted 10 times for color development. 100 µl reagent was added to 50 µl sample and 349 the mixture was incubated at room temperature for 20 min. The absorbance at 650 nm was measured in a 350 microplate reader (BioTek Cytation5 Fig. 1 Structures of ctDnf1-Cdc50. a, Cryo-EM map of ctDnf1-Cdc50 with BeF 3 -. The A, N, P, and M domains and Cdc50 are labeled and colored yellow, red, blue, tan, and pink, respectively. b, Cryo-EM map of ctDnf1-Cdc50 with AMPPCP. Colors are the same as in a. c, Ribbon diagram of ctDnf1-Cdc50 with BeF 3 -. TM1, TM2, and L1/2 are labeled. d, Ribbon diagram of ctDnf1-Cdc50 with AMPPCP. TM2 is labeled.

Fig. 2 Comparison of ctDnf1-Cdc50 structures in the E1-ATP and E2P states. a,
Overlay of the E1-ATP and E2P structures by superimposing TMs 3-10 of ctDnf1 and ctCdc50 (grey). The A domain is yellow in E1-ATP and blue in E2P. The N domain is red in E1-ATP and cyan in E2P. TM2 is purple in E1-ATP and TMs 1 and 2 are green in E2P. b, same as a, except the N domains are omitted for clarity. The motion distance of the A domain between the E1-ATP an E2P states is labeled. c, same as a, except the A domains are omitted. The movement of the N domain between the E1 and E2 states is indicated. d, same as a, except rotating by 90 degrees and the A and N domains are omitted to show the movements of TMs 1 and 2 between the E1-ATP and E2P states. a, Cryo-EM density map of the E2P state with the lipid nanodisc. The map is low pass filtered to 6 Å. Colors are as in Fig. 1, with the addition of lipid nanodisc density (grey). b, 90-degree rotation of a, showing the nanodisc density around TMs 1 and 2 (green dashed circle). The A domain is omitted for clarity. c, Cut-away top view of the nanodisc. TMs are labeled. d, Cryo-EM density map of the E1-ATP state with the lipid nanodisc. The curved dashed line indicates the depression that is required to fit the A domain on the surface of the membranes. e, 90-degree rotation of d, showing the local membrane thinning. f, Cut-away top view of the nanodisc. g, Zoom-in view of e. The surface of ctDnf1 is colored according to the electrostatic potential. The black dashed frame highlights the distorted membranes and the negatively charged patch. Fig. 4 Lipid substrate binding sites a, Lipid binding sites in E2P. ctDnf1 is shown as ribbon. The density of the lipids is shown as grey meshes at 1.5 σ. Two modeled PC and their interacting residues are shown as sticks. The left panel shows the overview and the right panels show the magnified views of each site. TMs, residues, and binding sites are labeled. b, Electrostatic potential surfaces of the flippase in E2P, showing the environment of the lipid binding sites. The electrostatic potential surfaces are calculated using APBS with the default setting in PyMOL. The membrane cartoon is colored grey. The left panel shows the overview and the right panels show the magnified views of each site. c-d, same as a and b, except showing the binding sites in the E1-ATP structure. A phosphate head group and a lyso-PC are modelled in site 1 and 2, respectively.

Fig. 5 Conformational changes of the lipid substrate binding sites between E1-ATP and E2P
a, Superimposition of E1-ATP (cyan cylinders) and E2P (tan cylinders), showing the groove bordered by TMs 2, 4, and 6. Q549 and N550 that interact with the lipid substrate are shown as sticks. The movement of TM2 during the E2-E1 transition is indicated by a red arrow. The lipid binding sites are marked. b, Side view of the groove. The dashed line outlines E2-site1 which is disrupted and occupied by Q549 and N550 in E1-ATP. The membrane is colored grey. The lipid molecules with light green heads are arranged to show the distortion of bilayers. The lipid molecule with the dark green head represents a substrate that is entering the transport pathway via E1-site2. b, Cartoon drawing of E2P. The lipid substrate is trapped in E2-site1. c, As in a, with the lipid bound at E1-site1. d, As in b, but the lipid substrate has been flipped to the cytosolic leaflet and waits at E2-site2 to be released.       PCTPKSWNISDDLGQVEYIFSDKTGTLTQN  PCTPKSWNISDDLGQIEYIFSDKTGTLTQN  PAMARTSNLNEELGQVKYLFSDKTGTLTCN  AAMARTSNLNEELGQVKYIFSDKTGTLTCN  PAMARTSNLNEELGQVKYLFSDKTGTLTCN  GPLVNTSDLNEELGQVEYIFTDKTGTLTEN  GALVNTSDLNEELGQVDYVFTDKTGTLTEN  PAKARTTTLNEQLGQIHYIFSDKTGTLTQN  PAEARTTTLNEELGQVEYIFSDKTGTLTQN  QLQCRALNITEDLGQIQYIFSDKTGTLTEN  ASYAANTAISEDLGQVEYILTDKTGTLTDN  PAQARTSNLNEELGMVDTILSDKTGTLTCN   LLLRGCHLRNTEWALGVVVFTGHDTKIMMNA  LLLRGATLRNTPWIHGVVVFTGHETKLMRNA  MILRGATLRNTAWIFGLVIFTGHETKLLRNA  VLLRGCTLRNTKWAMGVVMFTGGDTKIMLNS  LLLRGCTLRNTKWAMGMVIFTGDDTKIMINA  ILLRGTQLRNTQWGFGIVVYTGHDTKLMQNS  ILLRGAQLRNTQWVHGIVVYTGHDTKLMQNS  ILLRGTQLRNTQWVFGIVVYTGHDTKLMQNS  LLLRGATLKNTEKIFGVAIYTGMETKMALNY  LLLKGATLKNTEKIYGVAVYTGMETKMALNY  ILLRGCVIRNTDFCHGLVIFAGADTKIMKNS  MLLRGCVLRNTEWCFGLVIFAGPDTKLMQNS  LLLRGCTLRNTDAVVGIVIYAGHETKALLNN  TLLQSCYLRNTEWACGVSVYTGNQTKLGMSR  ILLRDSKLRNTEYVYGAVVFTGHDTKVIQNS   VME  QME  IME  VME  VME  IMN  VMQ  IMN  NME  SME  IMT  IMV  KMV  KMI  -