Structural basis of the P4B ATPase lipid flippase activity

P4 ATPases are lipid flippases that are phylogenetically grouped into P4A, P4B and P4C clades. The P4A ATPases are heterodimers composed of a catalytic α-subunit and accessory β-subunit, and the structures of several heterodimeric flippases have been reported. The S. cerevisiae Neo1 and its orthologs represent the P4B ATPases, which function as monomeric flippases without a β-subunit. It has been unclear whether monomeric flippases retain the architecture and transport mechanism of the dimeric flippases. Here we report the structure of a P4B ATPase, Neo1, in its E1-ATP, E2P-transition, and E2P states. The structure reveals a conserved architecture as well as highly similar functional intermediate states relative to dimeric flippases. Consistently, structure-guided mutagenesis of residues in the proposed substrate translocation path disrupted Neo1’s ability to establish membrane asymmetry. These observations indicate that evolutionarily distant P4 ATPases use a structurally conserved mechanism for substrate transport.

with purified protein yet. However, temperature-sensitive alleles of NEO1 were found to cause a loss of PE and PS asymmetry in the plasma membrane but with a greater influence on PE exposure 38 . Depletion of TAT-5, the Caenorhabditis elegans ortholog of Neo1, leads to exposure of only PE in the extracellular leaflet, suggesting the P4B-ATPases play a more substantial role in establishing PE asymmetry 39 . Although it lacks a β-subunit, Neo1 is regulated by Dop1 (Dopey ortholog), Mon2 (which is a relative of large Arf guanine nucleotide exchange factors), the small GTPase Arl1, and Any1 34,[40][41][42][43] . In fact, Neo1, Mon2, Arl1, and Dop1 assemble into a membrane remodeling complex 40,41 . These regulatory proteins are crucial for yeast growth and likely interact with Neo1 only transiently with low binding affinity, but little is known about their functions. The disruption of ANY1, which encodes a PQ-loop membrane protein, can rescue the growth deficiency of a Δneo1 or Δdrs2 strain 38,43,44 .
Recently, several P4A ATPase structures have been reported, including the S. cerevisiae Drs2-Cdc50 45,46 , Dnf1-Lem3 32 , Dnf2-Lem3 32 , the C. thermophilum Dnf1-Cdc50 47 , the human ATP8A1-CDC50A 25 , and ATP11C-CDC50A 48 . These structural studies revealed a conserved architecture among the P4 ATPases, including the 10-helix TMD, the cytosolic A domain, N domain, and P domains, and a conserved ATP-dependent lipid transport cycle. Several of the P4A ATPase structures contain substrate lipid-bound in a lumenal or exoplasmic "entry site" formed from residues that had previously been implicated in substrate recognition through mutagenesis studies 32,[48][49][50] . A second substrate lipid-binding site on the cytosolic side of the membrane, an exit site, has been identified only in the Dnf1 and Dnf2 structures 32,47 . Surprisingly, this large exit gate extends 10 Å out of the bilayer, causing the membrane to dimple into the cytosol, and it uses an Arg residue from the N-terminal cytosolic domain of the β-subunit to help coordinate the lipid headgroup 32 . However, it has been unclear whether the monomeric lipid flippase Neo1 functions by a similar mechanism or whether the cytosolically exposed exit gate is a conserved aspect of the substrate translocation path for P4 ATPases that lack a β-subunit.
An important question is how the substrate specificity is achieved for the P4B ATPases. Inactivation of Neo1 perturbs the PS and PE asymmetry of the plasma membrane, but whether or not Neo1 interacts directly with these lipids as transport substrates are unclear. By comparison, Drs2-Cdc50 primarily transports PS, while Dnf1-Lem3/Dnf2-Lem3 primarily transports glucosylceramide (GlcCer), PC, and PE, and a small amount of PS [51][52][53] . The lipid substrate specificity of P4A ATPases appears to be defined by the physiochemical properties of substrate binding pockets as demonstrated structurally and functionally 25,32,[45][46][47]50,52,54 . However, it is not clear if the mechanism of substrate recognition and translocation is conserved for all P4-ATPases, including the evolutionarily distant P4B-APases. To address these questions, we purified the yeast Neo1; examined its substrate-stimulated, in vitro ATP hydrolysis activity; determined cryo-EM structures of Neo1 in three intermediate states, and performed extensive structure-guided mutagenesis and functional assays. Our work identifies unique residues responsible for substrate specificity and provides insights into the conservation of the cytosolic binding site despite the lack of a β-subunit.

Results and discussion
Neo1 ATPase activity is stimulated by PE and PS. We overexpressed Neo1 in S. cerevisiae using a multicopy plasmid with a strong GAP promoter and an N-terminal triple FLAG tag. The detergent dodecyl maltoside (DDM) was used to solubilize the membrane and we purified Neo1 with an anti-FLAG affinity column followed by size-exclusion chromatography, during which DDM was replaced by lauryl maltose neopentyl glycol (LMNG) and cholesteryl hydrogen succinate (CHS) to stabilize the membrane protein (Fig. 1a, b). Neo1 has been implicated in the transport of PS and PE in vivo 38 . We used substratestimulated ATP hydrolysis activity as an indicator to investigate whether the purified protein was active and whether PE and PS directly interacted with Neo1. Dnf1-Lem3 displays a preference for lyso-phospholipid substrate 50 , so we also tested lyso-PS as a potential substrate. We found that the Neo1 ATPase activity was stimulated by PE (K m = 97.1 ± 28.6 μM), PS (K m = 62.9 ± 15.2 μM), and lyso-PS (50.3 ± 8.8 μM), with a maximum enhanced rate V max of 0.560 ± 0.059 nmol/min/μg by PE, 0.559 ± 0.042 nmol/min/μg by PS, and 0.585 ± 0.029 nmol/min/μg by lyso-PS, respectively (Fig. 1c, d). While the extent of stimulation from these substrates  was comparable, lyso-PS had the lowest K m that was about half of PE, suggesting a preference for lyso-PS. In contrast, PC and sphingomyelin did not stimulate the ATPase activity above the low basal level detected in the absence of phospholipid. Prior studies of the influence of Neo1 on membrane asymmetry suggested PE was the preferred Neo1 substrate 35 , but both PS and PE effectively stimulated ATPase activity. These data showed that the purified Neo1 was active and that the PS, PE, and lyso-PS are likely native substrates of the flippase.
Cryo-EM structure of Neo1 in the E2P state. We performed single-particle cryo-EM on purified samples either directly or after incubation with different inhibitors to capture structures in different states. We obtained cryo-EM 3D maps of Neo1 in the E1-ATP state stabilized by AMPPCP at 5.6-Å resolution, in the E2P-transition state of dephosphorylation stabilized by AlF 4 − at the 3.1-Å resolution, and in the E2P state as stabilized by BeF 3 − at 3.2-Å resolution (Supplementary Table 1, Supplementary  Figs. 1-4). The bound AlF 4 − and BeF 3 − mimic the planar pentavalent transition state of phosphoryl transfer and the covalent phosphorylated E2P ground state, respectively 14 . We found that 3D maps of Neo1 in the E2P-transition and E2P states were superimposable ( Supplementary Fig. 5). This is consistent with the previous observation that the Dnf2-Lem3 structures in the two states are nearly indistinguishable 32 . We first built atomic models into the 3.2-Å resolution map in the E2P state and into the 3.1-Å resolution map in the E2P-transition state. The 3D maps had excellent main-chain connectivity and side-chain densities except for four disordered loops: the N-terminal tail (residues 1-153), the N-domain loop (542-559), the P-domain loop (804-821), and the C-terminal tail (1142-1151) (Fig. 1e). We next used the E2P structure to help build an atomic model for the lower-resolution 3D map of Neo1 in the E1-ATP state. All these models fit well with the 3D maps and were refined to good statistics (Supplementary Table 1).
As shown in the E2P state structure, Neo1 contains all conserved P-type ATPase domains: a 10-TM helix (TMH1-10) TMD; the cytosolic A domain inserted between TMH2 and TMH3; and the cytosolic P and N domains that are both inserted between TMH4 and TMH5 ( Fig. 1e-g). The Neo1 structure aligns well in both TMD and soluble domains with the α-subunits of heterodimeric P4 ATPases Dnf1-Lem3, ATP8A1-CDC50, ATP11C-CDC50, and Drs2-Cdc50 ( Supplementary Fig. 6). Note that TMH1, TMH4, and TMH6 of Neo1 are kinked just as they are in other P4 ATPases whose structures are known 25,32,[45][46][47][48] . This conserved structural feature is essential for the substrate transporting activity. The Neo1 TMH4 kink is enabled by the highly conserved Pro-456.
The N-and C-terminal peptides play regulatory roles and share low sequence homology among the P4 ATPases. For example, the C-terminal peptides of the yeast Drs2 and the corresponding human homolog ATP8A1 are autoinhibitory 25,45,46,55,56 . The N-terminal peptide of Neo1 interacts with the cargo-selective sorting nexin Snx3 to mediate Neo1 trafficking 36 . The N-terminal peptide preceding the TMH1 is 183 residues long, of which the first 153, containing the Snx3 binding site, were disordered in our structure. The 30-residue ordered region of the N-terminal peptide (Glu-154 through Val-183) formed a short β-strand and a short αhelix that binds to the catalytic A domain which is essential for ATP turnover. This interaction enables signal transmission from the A domain to the substrate transporting path in the TMD (Fig. 1g). Unlike other P4 ATPases, the Neo1 C-terminal peptide following the TMH10 (Tyr-1142 to Pro-1151) is short and disordered. Therefore, it is unlikely that the Neo1 C-terminal tail is an autoinhibitory domain comparable to Drs2 or ATP8A1.
Neo1 structure in the E1-ATP state and the conserved ATPdependent transport mechanism. In the 5.6-Å resolution 3D map of Neo1 in the E1-ATP state determined in the presence of 1 mM AMPPCP (a nonhydrolyzable ATP analog), the AMPPCP density is clear and the molecule stabilizes the interface between the N and P domains (Fig. 2a, Supplementary Fig. 4). Comparison of the Neo1 structures in the E1-ATP and the E2P states shows that the whole TMD and the P domain are superimposable, while the N and A domains have undergone dramatic conformational changes (Fig. 2b). In the E1-ATP structure, the N domain is packed tightly on P domain through the AMPPCP ligand. The A domain is also ordered, although it has no interaction with the P domain and interacts only with a short helixturn-helix motif (Lys704 to Leu736) in the N domain. The long TMH2 bends upwards at Lys-236 away from the P domain. Transitioning from the E1-ATP state to the E2P state, ATP is cleaved, ADP is released from the N domain, and the phosphate is transferred onto Asp-503 of the P domain. The phosphorylated aspartate is mimicked by the BeF 3 − moiety in our Neo1 structure. To enter into the E2P state, both the N and A domains swing away from the membrane and pack tightly together. Accompanying this transition, the TMH2 becomes straightened, pointing downwards to allow the A domain movement. These large conformational changes are likely propagated to the substratebinding sites and alter the substrate-binding affinity, leading to the transport of a lipid molecule across the membrane. Conversely, the transition from E2P→E2→E1-ATP should close the exit site and release substrate into the cytosolic leaflet.
We next compared the Neo1 E1-ATP structure with the published structure of Dnf1-Lem3 in the same state and found that the TMD and the cytosolic domains are positioned similarly (Fig. 2c). Furthermore, the conformation and the substrate transport path of Neo1 in the E2P state highly resemble those of the Drs2, Dnf1, ATP8A1, and ATP11C flippases in the same E2P state ( Supplementary Fig. 6). The lack of TMH3-10 movement through the P4 ATPase transport cycle was proposed to be conferred by the tightly associated β-subunit 25 , but Neo1 retains this property in the absence of a stabilizing β-subunit. This observation strongly suggests that all P4-ATPases-whether functioning as a single subunit in the case of Neo1 or with a partner β-subunit in the majority of cases-employ a conserved substrate transport mechanism. A recent functional study of the Neo1 substrate entry site supports such conserved transport mechanism 57 . Therefore, the β subunit may not be primarily functioning to constrain the conformational changes in TMD, as previously proposed 25 .
The substrate transport path and the substrate specificity of Neo1. Structures of six P4 ATPases have been reported so far; four are in complex with substrate lipids, including S. cerevisiae Dnf1-Lem3, Chaetomium thermophilum Dnf1-Cdc50, human ATP8A1-CDC50, and human ATP11C-CDC50 25,32,45-48 (Supplementary Fig. 6). These structures reveal a conserved flippase architecture, similar conformational changes as the enzymes go through the transport cycle, and a very similar substrate translocation path, despite having different substrate specificities. A structure-based homology search revealed that Neo1 in the E2P state aligned well with all known P4 ATPase structures in that state, but the best match was with Dnf1 (Fig. 3a). Yet the superposition of Neo1 and Dnf1 revealed a conspicuous difference between these flippases: the ECL2, 4, and 5 of Dnf1, which are primarily responsible for binding the β-subunit Lem3, are much longer than those in Neo1 (Fig. 3a, b). These extracellular loops are similar in length between Neo1 and cation transporters. This observation led us to hypothesize that P4A ATPases evolved into two-subunit ATPases by the lengthening of ECL2, 4, and 5, thereby acquiring the ability to bind a β-subunit. It's also possible that the two subunits bind first, and then the α-subunit evolved these longer loops to form the stabilized complex.
Previous reported Dnf1 structure contained phospholipid molecules in both the substrate entry and exit sites, but we did not capture any endogenous lipid molecules in the Neo1 substrate sites. By aligning the Neo1 and Dnf1 structures, we narrowed down the substrate translocation path of Neo1 to a groove in the membrane domain between TMH1-4 and TMH6 (Fig. 3a, c). This putative substrate path can be divided into three sections: the top (lumenal) substrate entry site composed of several polar residues (Gln-209, Ser-221, Tyr-222, Ser-452, and Thr-453); the middle transport path between lipid-binding sites that contains the hydrophobic gate (Phe-202, Leu-226, Val-229, Val-457, and Val-461) and flanking polar residues (Gln-193, Thr-233, Arg-460,  (Fig. 4a). To identify key residues responsible for Neo1 substrate-transporting activity and specificity, we next carried out extensive structure-guided mutagenesis and functional assays (Fig. 4b, Supplementary Figs. 7 and 8). Most of these residues were mutated to either alanine or to residues present at the same position in other P4 ATPases.
Neo1 is essential for cell viability, and so we used a plasmid shuffle complementation assay to test whether Neo1 variants could support the viability of a neo1Δ strain (Fig. 4b, Supplementary Fig. 7). The parental neo1Δ strain harboring wild-type (WT) NEO1 on a URA3-marked plasmid was transformed with HIS3-marked plasmids harboring either WT Neo1, an empty vector, or the Neo1 mutants. The strains were replica-plated on SD medium to select for both plasmids, or on SD plus 5-FOA to counter-select against the URA3 plasmids. Because neo1Δ cells are inviable, the empty vector control strain failed to grow on 5-FOA, while the WT copy of Neo1 fully supported growth. We observed that all the Neo1 mutants supported viability comparably to WT Neo1 except for Neo1 P456G, which failed to support growth (Fig. 4b). Pro-456 is near the center of M4, and nearly all P-type ATPases have a proline in this position that is crucial for unwinding the M4 helix. We found that Neo1 P456G totally abolished the Neo1 activity in vivo. Surprisingly, Neo1 P456A fully supported viability, suggesting that maintenance of hydrophobicity with alanine in this region is essential (Fig. 4b).
Previous studies indicated that there is partial functional redundancy between Neo1 and Drs2. Therefore, we also tested the Neo1 mutants for their ability to support the growth of a drs2Δ neo1Δ double mutant over a range of temperatures, which provides a more sensitive assay for loss of Neo1 function (Fig. 4c,  Supplementary Fig. 8). drs2Δ single mutants grow well at 30°C, but are cold-sensitive and grow slowly at 25°C and fail to grow at 20°C. We found that the exit-site mutations Neo1 S488W and Neo1 R247L were lethal when combined with drs2Δ and failed to grow at 30°C. Many of the other Neo1 entry-site and transportpathway mutations substantially reduced growth in this background relative to WT Neo1. These results indicate that the Neo1 variants represent an allelic series, ranging from complete loss of function (P456G) to apparent WT activity, with respect to cell viability (P456G < S488W = R247L < S221L < Q193A < S452Q < E237D < R247A < P456A = T453S < S488A = Q209G).
Neo1 is primarily found in the Golgi, and a loss of its flippase activity should increase substrate lipids in the luminal leaflet of the Golgi and their ultimate exposure in the extracellular leaflet of the plasma membrane. Therefore, to test the influence of Neo1 mutations on substrate transport, we employed toxin sensitivity assays to measure the exposure of PS and PE on the plasma membrane extracellular surface (Fig. 5a). Papuamide A (Pap A) is a potent cytotoxic agent that binds specifically to PS, and duramycin binds specifically to PE. Both toxins produce pores in the membrane that kill sensitive cells on which their respective targets are exposed. WT yeast cells are relatively resistant to Pap A and duramycin because most of their target lipids are restricted to the inner cytosolic leaflet of the plasma membrane. The neo1Δ strain expressing a WT copy of Neo1 was resistant to duramycin and papuamide relative to the neo1Δany1Δ strain, which exposed both lipids and was hypersensitive to both toxins. The any1Δ mutation suppresses neo1Δ lethality but does not suppress the loss of PS and PE asymmetry caused by loss of Neo1 function 44 .
Relative to these two control strains, many of the Neo1 mutants displayed a significant loss of membrane asymmetry. We further explored the Neo1 mutants that had PE exposure and no apparent PS exposure by using higher concentrations of papuamide A to assess how well they maintained PS asymmetry (Fig. 5b, c). Neo1-S221L displayed a slight hypersensitivity to papuamide A, but the rest of the mutants showed sensitivity indistinguishable from that of the WT.
The Neo1 mutants fell into three classes regarding substrate recognition (Fig. 5d). PS-specific mutants that expose PS but maintain near normal PE asymmetry (Q209G, S452Q, and T453S), PE-specific mutants that only expose PE (Q193A, S221L, R247A, and R247L), and mutants that expose both lipids. These substrate-specific, separation-of-function mutations strongly imply that the targeted residues are directly involved in substrate binding. For example, entry-site residues known to bind PS in ATP8A1-CDC50 correspond to the PSspecific Neo1 mutants. These results define residues crucial for PE recognition in Neo1, and these residues are in the entry site (S221), transport path (Q193), and exit site (R247). We note Mutations that decrease the transport of both PS and PE are not likely causing misfolding of Neo1 because these alleles fully support the growth of neo1Δ cells (Fig. 4b). We tagged such Neo1 variants with GFP and examined their localization relative to mKate-Aur1, a Golgi marker. Mutations that significantly alter the structure of membrane proteins in the secretory pathway typically cause retention of the mutant protein in the endoplasmic reticulum (ER) 32,58 . However, all of the Neo1 variants localized normally to the Golgi complex. In addition, we did not detect any significant difference in expression levels by immunoblotting for the GFP-tagged Neo1 variants (Supplementary Fig. 9). Thus, these mutations likely cause a loss of either substrate recognition or transport rather than a structural defect.
A comparison of the substrate-binding sites of lipid flippases. According to previous functional studies, each type of P4 ATPase may transport several different lipid substrates, but the preference is usually not equal 35,51,52 . We hypothesize that the substrate preference is likely determined by the chemical and electrophysiological properties of the substrate-binding sites. Having defined the substrate-binding sites in Neo1 via mutagenesis and functional assays, we compared the substrate entry sites of Neo1, Dnf1, Drs2, ATP8A1, and ATP11C (Fig. 6a), which are all surrounded by TMH1, TMH2, TMH4, and TMH6. We found that they all contained polar and non-charged residues such as Ser, Thr, Asn, and Gln. Structure-based sequence alignment showed that key residues at structurally equivalent positions were not well conserved in the primary sequence (Fig. 6b). The Neo1 functionally important dipeptide motif 209-QA-210 in TMH1 is analogous to GA in Dnf1 and Dnf2, QQ in Drs2 and ATP8A1, and the single residue Q in ATP11C. The Neo1 dipeptide motif 221-SY-222 in TMH2 is analogous to TT in Drs2 and ATP8A1, TS in ATP11C, LS in Dnf1, and FA in Dnf2. The Neo1 dipeptide motif 452-ST-453 becomes NN in ATP8A1, SN in Drs2, NF in ATP11C, and QS in both Dnf1 and Dnf2. Finally, the Neo1 residue Ala (A978) becomes an asparagine in all the other P4 ATPases compared here. The idea that these residues in the entry sites of the P4 ATPases play an important role in the lipid-transporting preference is underscored by previous The red heat map displays papuamide A dose responses; growth inhibition at 1.5-2 μg/mL indicates aberrant exposure of PS. The data represents growth relative to WT cells in absence of the drug. Two-way ANOVA was performed to test the variance and comparisons with WT Neo1 were made with Tukey's post hoc analysis (n = 3, ±standard deviation (SD)). *p < 0.05, **p < 0.01, ***p < 0.001 (red for papuamide A and blue for duramycin). Neo1 (Y222S) is a gain-of-function mutation that weakly suppresses drs2Δ 57 . b, c Neo1 mutants hypersensitive to only duramycin in (a) were tested at higher concentrations of papuamide A (up to 6 μg/mL). All the values are an average of at least three biological replicates ± standard deviation (n = 3, ±SD). d Table summarizing the effects of Neo1 mutations on membrane asymmetry.
mutational studies. In one study, mutations of the TMH1 key motif GA and the TMH4 motif QS of Dnf1/Dnf2 were shown to alter their activity for PC, PE, and GlcCer, and mutation of the TMH6 key reside N1226 of Dnf1 abolished the transport of all three substrates 52 . In another study, the TMH1 key motif QQ of Drs2 was shown to influence the specificity for PS transport 49 .
The substrate-binding exit site initially defined in Dnf1-Lem3 and Dnf2-Lem3 was unanticipated because it extends substantially out of the membrane plane into the cytosolic domains of b a  TM2   TM4   TM6   TM1   N504  Q237   TM2   TM4   TM6   TM5   TM1   TM2   TM4   TM6   TM1   TM2   TM4   TM6   TM5   TM1   TM2   TM4   TM6   TM5   S611   A231  S243  N1226 Fig. 6). It is formed from cytosolic extensions of TMH2 and TMH4 and a loop emanating from TMH4 that forms the membrane-proximal region of the P domain. We show in the current study that this cytosolically exposed lipid-binding site is functionally required for substrate transport in Neo1, a P4B ATPase. Comparison of the exit sites suggests a mechanism for how Neo1 functions without a β-subunit. The most membrane-distal residue forming the exit site in Dnf1-Lem3 is R51 in the cytosolic N-terminal tail of Lem3, analogous to R247 in Neo1 on the TMH2 side of the cleft (Fig. 6c). The β-subunit N-terminal tail interacts with the loop from TMH4 that bends back towards the membrane to form part of the exit site (W652 in Dnf1), before starting the P-domain where the phosphorylated Asp resides. This TMH4 loop is much longer in the P4A ATPases than in the P4B ATPases and is nearly absent in P2 ATPases (e.g., SERCA) ( Supplementary Fig. 10).
In summary, we have determined the structure of a singlesubunit lipid flippase and we suggest that P4B ATPases such as Neo1 arose first from cation transporters and gave rise to the P4A group with their new β-subunit requirement. The overall architecture and the substrate entry site of Neo1 are remarkably conserved in structure but, interestingly, not in amino acid sequence. Key differences between the P4A ATPases and P4B ATPases reflect regions important for β-subunit interaction, including the shorter extracellular loops in Neo1, a shorter TMH4 cytosolic loop, and the ability to form the exit site in the absence of the β-subunit. Our extensive mutational studies identified key residues responsible for Neo1's substrate preference. Structural comparison of Neo1 with other P4 ATPases revealed that the basic ATP-dependent lipid transport mechanism is highly conserved throughout evolution (Fig. 7). The capability and preference of the ATPases for transporting a multitude of lipid species are likely conferred by varying the amino acids of the key motifs in the otherwise structurally similar substrate entry and exit sites. Our work has advanced the mechanistic understanding of a large and often essential family of the P4 ATPases.

Methods
Expression and purification of Neo1. The NEO1 gene was cloned into pRS423 with a strong GAP promoter and an N-terminal triple FLAG tag. The construct was then transformed into S. cerevesiae. Yeast cells were grown in 2 L SD-H medium (20 g/L glucose, 6.7 g/L yeast nitrogen base without amino acids, 0.04 g/L adenine hemisulfate salt, and 1.92 g/L yeast synthetic drop-out medium supplements without histidine) for about 20 h, then transferred to 18 L of YPD medium for another 12 h before harvest. Cells were resuspended in lysis buffer (20 mM Tris-HCl, pH 7.4, 0.2 M sorbitol, 50 mM potassium acetate, 2 mM EDTA, and 1 mM phenylmethylsulfonyl fluoride [PMSF]) and then lysed using a French press at 15,000 psi. The lysate was centrifuged at 10,000×g for 30 min at 4°C. The supernatant was collected and centrifuged at 100,000×g for 60 min at 4°C. The membrane pellet was collected and then resuspended in buffer A containing 10% glycerol, 20 mM Tris-HCl (pH 7.4), 1% DDM, 0.1% CHS, 0.5 M NaCl, 1 mM MgCl 2 , 1 mM MnCl 2 , 1 mM EDTA, and 1 mM PMSF. After incubation for 30 min at 4°C, the mixture was centrifuged for 30 min at 120,000×g to remove the insoluble membrane. The supernatant was mixed with pre-equilibrated anti-FLAG (M2) affinity gel (GenScript) at 4°C overnight with shaking. The affinity gel was then collected and washed three times in buffer B (20 mM HEPES, pH 7.4, 150 mM NaCl, 0.01% LMNG, 0.001% CHS, and 1 mM MgCl 2 ). The proteins were eluted with buffer B containing 0.15 mg/mL 3× FLAG peptide and were further purified in a Superose 6 10/300 Increase gel filtration column in buffer C (20 mM HEPES, pH 7.4, 150 mM NaCl, 0.0025% LMNG, 0.00025% CHS, and 1 mM MgCl 2 ). Finally, the purified proteins were assessed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis gel and concentrated for cryo-EM analysis. Approximately, 20 μg of Neo1 can be purified from 18 L of cells.
ATP hydrolysis assay. The ATPase activity assays were performed using BIO-MOL Green Reagent (Enzo Life Sciences, Inc.) to measure released inorganic phosphate. The lipids PC, SM, PE, lyso-PS, and PS were solubilized with 20 mM sodium cholate in 20 mM HEPES, pH 7.4, 150 mM NaCl. Each reaction contained 0.025 mg/mL protein, 0.003% LMNG, 0.0003% CHS, 20 mM HEPES, pH 7.4, 150 mM NaCl, 10 mM MgCl 2 , and 0.25 mM ATP. Reactions were carried out at 37°C for 15 min and then terminated immediately by addition of the reagent. After incubation of the mixture for 20 min at room temperature, the absorbance at 620 nm was measured using a microplate reader (SpectraMax M2e). The phosphate concentration was determined by calibration with the phosphate standard (BML-KI102).
Cryo-electron microscopy. To capture different states, the purified Neo1 was mixed with various solutions for 1 h on ice: E1-ATP, 5 mM MgCl 2 , and 2 mM AMPPCP; E2P, 5 mM MgCl 2 , 10 mM NaF, and 2 mM BeSO 4 . After incubation, 2.5-μL aliquots of purified Neo1 at a concentration of about 1 mg/mL were placed on glow-discharged holey carbon grids (Quantifoil Au R2/2, 300 mesh) and were flash-frozen in liquid ethane using an FEI Vitrobot Mark IV. Cryo-EM data were collected automatically with SerialEM in a 300-kV FEI Titan Krios electron microscope with defocus values from −1.0 to −2.0 μm. The microscope was operated with a K3 direct detector at a nominal magnification of 130,000× and a pixel size of 0.413 Å per pixel. The dose rate was 8 electrons per Å2 per s, and the total exposure time was 8 s.
Cryo-EM image processing. Program MotionCorr 2.0 59 was used for motion correction, and CTFFIND 4.1 was used for calculating contrast transfer function parameters 60 . All the remaining steps were performed using RELION-3 61 . The resolution of the map was estimated by the gold-standard Fourier shell correlation at a correlation cutoff value of 0.143.
For the Neo1 structure in the E2P state, we collected 4354 raw movie micrographs. A total of 2,315,630 particles were picked automatically. After 2D classification, a total of 2,068,707 particles were selected and used for 3D classification. Based on the quality of the four 3D classes, 1,279,510 particles were retained for further 3D reconstruction, refinement, and postprocessing, resulting in a 3.25-Å average resolution 3D map.
For the E1-ATP state, we collected 1175 raw movie micrographs. A total of 688,445 particles were picked automatically. After 2D classification, a total of 668,445 particles were selected and used for 3D classification. Based on the quality of the four 3D classes, 264,891 particles were selected for further 3D reconstruction, refinement, and postprocessing, resulting in the 5.64-Å average resolution 3D map.
For the E2P-transition state, we collected 5119 raw movie micrographs. A total of 3,037,926 particles were picked automatically. After 2D classification, a total of 2,661,522 particles were selected and used for 3D classification. Based on the quality of the four 3D classes, 1,673,321 particles were selected for further 3D reconstruction, refinement, and postprocessing, resulting in the 3.08-Å average resolution 3D map.  Fig. 7 A proposed lipid flipping mechanism by Neo1. The sketch highlights the six major states of the Post-Albers model, of which two are captured in this study. The four states not captured are labeled in paratheses. Despite being a single-subunit flippase, Neo1 follows the general mechanism proposed for the two-subunit P4 ATPases. The sketch also illustrates that the substrate lipid entry site dips into the transmembrane region and the exit site goes into the cytosolic region of the ATPase.
Structural modeling, refinement, and validation. We first built the model of Neo1 in the E2P state at 3.2-Å resolution. We generated the initial model based on the structure of Dnf1 (PDB ID 7KYC) by SWISSMODEL (https:// swissmodel.expasy.org), and then manually corrected it in COOT and Chimera 62,63 . The complete Neo1 model was refined by real-space refinement in the PHENIX program and subsequently adjusted manually in COOT. Using the model of Neo1 in E2P as a reference, the model of Neo1 in the E2P-transition state at 3.1-Å resolution was built and refined using COOT, Chimera, and PHENIX. For the Neo1 in E1-ATP state at the 5.6-Å resolution, the model of Neo1 in E2P was split into TMD, A, N, and P domains, which were then fitted and refined into the map as the rigid body using COOT, Chimera, and PHENIX. Finally, all models were validated using MolProbity in PHENIX 64,65 . Structural figures were prepared in Chimera and PyMOL (https://pymol.org/2/).
Yeast strains and plasmid construction. The primers used in this study are listed in Supplementary Table 2. The strains and plasmids used in the study are listed in Supplementary Table 3. All yeast culture reagents were purchased from Sigma-Aldrich and BD Scientific, and strains were grown in YPD or minimal selective media. Yeast transformation was performed using the standard LiAC-PEG method 66 . For plasmid shuffling assays, 10-fold serial dilutions starting from a cell suspension with a cell density of OD 600 = 1 were spotted on synthetic defined media (SD) and SD-5-FOA and incubated at 30°C or temperatures indicated for at least 2 d before imaging. 5-FOA is converted to a toxic metabolite by the Ura3 protein and therefore cells retaining the URA3-marked plasmid cannot grow on 5-FOA medium 67,68 . All images of yeast colonies are representative of three biological replicates (three independently isolated strains of the same genotype). DNA constructs and mutations were created by Gibson assembly and quick-change mutagenesis according to the manufacture instructions (using PfuTurbo; Agilent).
Toxin sensitivity assays. Papuamide A was a kind gift from Raymond Andersen from the University of British Columbia; duramycin was purchased from Sigma Aldrich. For toxin sensitivity assays, mid-log cells were diluted to 0.1 OD 600 in fresh YPD medium and 100 μL of cells were distributed to each well of 96 wellplate with or without the toxin in 100 μL of YPD. Toxin dilutions were calculated based on final concentrations in a total well volume of 200 μL. Plates were incubated at 30°C for 20 h. Concentrations of the cells were measured in OD 600 /mL with a Multimode Plate Reader Synergy HT (Biotek). Growth relative to vehicle control (no drug) was used as 100% growth. All values are an average of at least three biological replicates ±standard deviation.
Fluorescence microscopy. Strains expressing GFP-Neo1 and Aur1-mKate were grown to the mid-log phase in synthetic media. Cells were washed with fresh medium 3 times and resuspended in fresh SD medium. Cells were then mounted on glass slides and observed immediately at room temperature. Images were acquired using a DeltaVision Elite Imaging System equipped with a 100×,1.4 NA oil immersion objective lens followed by deconvolution using softWoRx software (GE Healthcare Cytiva, Pittsburgh, PA). Images were analyzed in FIJI (FIJI Is Just ImageJ) using the JaCOP (Just another Colocalization Plugin) add-on. Colocalization was quantified through the computation of the Manders' coefficient in JaCOP as the fraction of overlap between Neo1 and Aur1.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support this study are available from the corresponding authors upon reasonable request. The cryo-EM 3D maps and the corresponding atomic models of the Neo1 have been deposited at the EMDB database and the RCSB PDB with the respective accession codes of EMD-24413 and 7RD6 for E2P, EMD-24414 and 7RD7 for E2Ptransition, and EMD-24415 and 7RD8 for E1P-ATP. Source data are provided with this paper.