Cryo‐EM structure of MsbA in saposin‐lipid nanoparticles (Salipro) provides insights into nucleotide coordination

The ATP‐binding cassette transporter MsbA is a lipid flippase, translocating lipid A, glycolipids, and lipopolysaccharides from the inner to the outer leaflet of the inner membrane of Gram‐negative bacteria. It has been used as a model system for time‐resolved structural studies as several MsbA structures in different states and reconstitution systems (detergent/nanodiscs/peptidiscs) are available. However, due to the limited resolution of the available structures, detailed structural information on the bound nucleotides has remained elusive. Here, we have reconstituted MsbA in saposin A–lipoprotein nanoparticles (Salipro) and determined the structure of ADP‐vanadate‐bound MsbA by single‐particle cryo‐electron microscopy to 3.5 Å resolution. This procedure has resulted in significantly improved resolution and enabled us to model all side chains and visualise detailed ADP‐vanadate interactions in the nucleotide‐binding domains. The approach may be applicable to other dynamic membrane proteins.


Introduction
Lipid asymmetry between the two leaflets of a biomembrane is fundamental for cellular life [1,2] and often requires trans-bilayer movement of lipids facilitated by specific flippases [3,4]. MsbA is an ATPbinding cassette (ABC) exporter located in the cytoplasmic membrane of Gram-negative bacteria, where it acts as a lipid flippase, translocating lipid A, glycolipids and lipopolysaccharides (LPS) from the inner to the outer leaflet [5,6]. In addition, it can also function as a multidrug-resistance transporter by exporting various hydrophobic small molecules [7].
MsbA is a prototypical homodimeric ABC transporter composed of two transmembrane domains (TMD; containing six transmembrane helices each) and two nucleotide-binding domains (NBD) [8,9]. The conformational cycle of MsbA follows a "power stroke" mechanism [10]. It is triggered by ATP binding to the NBDs, followed by NBD dimerisation, ATP hydrolysis and subsequent NBD dissociation [9], all of them coupled to movements of the transmembrane helices switching between inward-facing, occluded, and outward-facing conformations. The conformational flexibility of MsbA has been demonstrated by a variety of biophysical techniques, including luminescence resonance energy transfer [11], electron spin resonance spectroscopy [12], crosslinking [13], electron microscopy [14] and molecular dynamics simulations [15]. The well-characterized structural transitions of MsbA occur in the millisecond timescale, making MsbA a model system for time-resolved structural studies [small-angle X-ray scattering (SAXS)] of integral membrane proteins (IMPs) [16,17].
Several MsbA structures have been determined in different conformational states, using various reconstitution systems, including crystal structures in detergent [8,[18][19][20] as well as single-particle cryogenic electron microscopy (cryo-EM) structures in MSP1D1 nanodiscs [9] or peptidiscs [21]. While these structures provide ample insights into LPS recognition and conformational transitions during the lipid A transport pathway as well as potential inhibition modes of MsbA, details about nucleotide coordination are limited.
Saposin-lipid nanoparticles (Salipro) have been recently introduced as a flexible reconstitution system for IMP [22]. The system uses the lipid-binding protein Saposin A (SapA) to form a discoidal scaffold to provide a lipidic environment for incorporated IMPs [22][23][24]. Salipro reconstitution works for most lipids, is independent of the size of the incorporated IMP, often results in high activity of the IMP and has been successfully used in a number of cryo-EM structures [25][26][27][28][29].
Here, we report the 3.5 A resolution cryo-EM structure of Salipro-reconstituted MsbA (from Escherichia coli) in an ADP-vanadate(Vi)-bound state. The structure shows MsbA in an occluded conformation with dimerized NBDs and allows clear visualization of the bound nucleotides.

Results and Discussion
Overall structure of MsbA (ADP-Vi) in Salipro MsbA from E. coli was purified from E. coli C43 cells in dodecyl maltoside (DDM) detergent and reconstituted into Salipro using 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (16:0-18:1 PC) (POPC) as lipid (Fig. 1A). This procedure resulted in a monodisperse sample with significantly higher ATPase activity (and stability) compared to the detergent sample and similar activity compared to MsbA reconstituted in MSP1D1 nanodiscs [9,30], indicating that a lipidic environment as provided by nanodisc or Salipro carrier systems is beneficial for structural studies of MsbA.
Orthovanadate can be used to mimic the transition state after ATP hydrolysis (but before phosphate release) by trapping the Mg 2+ -ADP-Vi complex in the catalytic sites. In this transition state, the stability of Salipro-reconstituted MsbA is significantly increased compared to apo MsbA. While the melting transition midpoint is 53°C for apo MsbA, it increases to 62°C for the Mg 2+ -ADP-Vi-bound complex and to 71°C for MsbA (ATPcS) (Fig. 1B). Saliproreconstituted MsbA also retained its specific activity (Fig. 1C).
Using SAXS we have detected significant differences between apo and ADP-vanadate bound MsbA in Salipro in the mid-q region from 0.5 to 3 nm À1 (Fig. 1D), which are similar to the X-ray scattering differences of MsbA in nanodiscs [17]. The distance distribution plots show very similar curve progressions and also the values of the radius of gyration (R g ) and maximum dimension (D max ) are almost identical (Table 1), indicating that the differences in the scattering reflect internal structural rearrangements. These results show that Salipro-reconstituted MsbA can indeed be trapped in a distinct conformational state by adding Mg 2+ -ADP-Vi.
Based on the increased thermal stability of Mg 2+ -ADP-Vi bound complex incorporated in Salipro, we set out to determine its structure using cryo-EM. We hypothesized that this structure would result in a high resolution and highlight key residues involved in nucleotide binding, thus filling a gap in our knowledge of this important ABC transporter. We selected Salipro-reconstitution with POPC as lipid to collect cryo-EM data on the basis of the high stability and activity of the sample [30] and the potential to combine them with stealth carrier SANS data [31]. In order to remove junk particles picked from the micrographs, we performed two rounds of 2D classifications ( Fig. 1E). This was followed by two rounds of 3D classifications to obtain a homogenous set of particles for the final 3D reconstruction. After further refinements and masking out the Salipro disc, a final map with a resolution of 3.5 A at gold standard FSC 0.143 (Figs 1F,G, 2 and 3) was obtained.
The local resolution filtered map is presented in Fig. 3. At this resolution, it was possible to resolve the structure for the transmembrane helices as well as the NBDs including the side chains. Overall, the transition state structure of MsbA (with Mg 2+ -ADP-Vi) in Salipro presented here resembles an occluded conformation similar to the structure in nanodiscs [9] but significantly different from the outward open conformation observed in the crystal structure [8] (Fig. 1F, G).  Salipro reconstitution resulted in cryo-EM structure determination of MsbA to higher resolution With the ability to reach a higher resolution in the Salipro system, it is possible to determine the detailed coordination of Mg 2+ -ADP and vanadate between the two NBDs of MsbA. While in the previous cryo-EM structures of MsbA many side chains could not be modelled due to limited resolution [9], the quality of our map at 3.5 A resolution has allowed us to unambiguously model the ADP, vanadate and Mg 2+ molecules in their binding sites located in between both NBDs (Fig. 4). The binding pocket for ADP (A) is formed by D117 (chain A), Q485 (chain B), S380 (chain A), K382 (chain A) S383 (chain A) and T384 (chain A). The aromatic residue Y351 is forming p-p stacking interactions to the adenosine. A similar interaction could be described for MsbA in liposomes with Starting with 10 302 movies, the preprocessing, including motion correction and CTF estimation, was done in Scipion, followed by particle picking in crYOLO resulting 679 185 picked particles. (B) The picked particles were subject to two rounds of 2D classification (T = 13) to remove the obvious junk particles. (C) Selected particles were further cleaned up by a round of 3D classification (T = 5). (D) Particles that fall within the best 3D classes (class 1 and 2) were further refined either separately or merged together. Here, using the reconstruction from particles that belong to class 1 reached 7.4 A; reconstruction with particles class 2 reached 4.0 A and the merged particles from both classes reached 4.2 A. (E) Afterwards, a second 3D classification was performed using class 2 particles, which yielded a final set of 83 278 particles. (F) After several refinement steps, the final 3D reconstruction map reached a resolution of 3.46 A (GSFC 0.143). Structure figures were prepared using UCSF Chimera [52]. The improved resolution also allowed the modelling of a lipid bound in a conserved binding pocket between transmembrane helices TM3, TM4, TM6 and the Nterminus (Fig. 5). Interestingly, TM4 and TM6 are important for switching from apo to Vi-trapped state [33]. MsbA often co-purifies with several lipids as observed by native MS [34]. The odd positioning of the lipid could imply a stabilising role within the cavity, rather than the POPC we added during Salipro reconstitution. The stability and activity of MsbA requires the presence of directly bound lipids [30]. Interestingly, the recent cryo-EM reconstruction of MsbA with first-generation inhibitors [35] also contains several unmodelled lipid-like densities in many pockets.
Several structures of MsbA have been previously reported in ADP-vanadate trapped states. One was the cryo-EM structure of MsbA with ADP-vanadate in nanodiscs (pdb:5TTP), determined at 4.8 A resolution   9]. In this structure, neither amino acid side chains nor ADP-Vi could be modelled due to limited resolution. Apart from the improved resolution and occupied nucleotide binding site, we see just minor differences to our structure in Salipro (RMSD: 2.4 A; Fig. 6A). The other models include a crystal structure of ADPvanadate trapped MsbA from Salmonella typhimurium determined to 4.2 A, which only allowed modelling of the protein backbone as dummy atoms (pdb:3B5Z), and the crystal structure of AMPPNP-bound MsbA (pdb:3B60) [8]. For these structures the protein was crystallised in undecyl-ß-D-maltoside detergent. Interestingly, these crystal structures show significant differences in the orientation of the transmembrane helices in particular on the periplasmic side (Fig. 6A) compared to both cryo-EM structures in lipidic environment   For efficient nucleotide binding and hydrolysis in ABC transporters the Walker A and signature motifs from opposing NBDs have to come together [36]. The C-alpha distances between S482 (Walker A motif) and S378 (signature motif) in the opposing NBD are 7.9 A  [9], while the addition of ADP-vanadate leads to an occluded state of the protein (5TTP) [9]. MsbA in detergent in complex with AMPPNP adopts an outward open state (3B60) [8]. MsbA in Salipro and ADP-vanadate (this study) also adopts an occluded state. (C) View from the cytoplasm to the ADP-Vi-bound MsbA structure in Salipro with C-alpha distances between S482 (Walker A motif) and S378 (signature motif) in opposing NBD. Furthermore, the C-alpha distance between Y351 and Q485 from the different chains is shown. Structure figures were prepared using PYMOL.

2965
The for our (occluded state) structure in Salipro (Fig. 6C), 6.2 A for the ADP-Vi-bound (occluded state) cryo-EM structure in nanodiscs (pdb:5TTP) [9], 6.0 A for the ADP-Vi-bound and 7.5 A for the AMPPNP-bound (outward-open) crystal structures, respectively, in detergent (pdb:3B5Z/3B60) [8]. Similarly, the distance of the C-alpha atoms of Y351 (A) and Q485 (B) is 16.3 A in Salipro (Fig. 6C), 15.8 A for 5TTP [9] and 18.0/16.8 A for 3B5Z/3B60 [8], indicating a similar and optimal orientation of the NBDs for nucleotide binding, details of which could now be visualized with our improved resolution obtained by Salipro-reconstitution. The comparison with the AMPPNP-bound crystal structure of MsbA [8] (pdb:3B60) in outward-open conformation reveals only minor differences with respect to the NBDs (RMSD: 1.0 A). The side chain of Glu424 coordinates the Mg 2+ in our structure, while it adopts a different rotamer in pdb:3B60. Overall, the nucleotide-binding pocket is slightly more closed in our structure compared to AMPPNP-bound state, due to a different orientation of loop T350-P357. The similarity between NBD distances of different states indicates that the switch in the transmembrane region between occluded and outwardopen conformations of ADP-Vi-bound states is not accompanied by major rearrangements of the NBDs.
The reason why our reconstitution in Salipro resulted in a significantly improved resolution for the ADP-Vibound state of MsbA compared to other reconstitution systems is still not understood. One possibility would be that MsbA is less dynamic in Salipro (when trapped with ADP-Vi). As these effects, however, are difficult to predict our results further illustrate that many reconstitution systems should be tried for improving the resolution of any membrane protein of interest.
Saposin A in a pNIC28-Bsa4 vector was expressed with an N-terminal His 6 -tag and tobacco etch virus (TEV) cleavage site in the E. coli strain Rosetta-gami 2 [22] and grown in terrific broth (TB) media at 37°C until an OD 600 of 1.5 was reached. The temperature was lowered to 20°C, and 0.1 mM IPTG was added for an overnight induction. After harvesting the cells, the pellet was resuspended in lysis buffer (20 mM sodium phosphate (pH 7.4), 300 mM NaCl, 5% glycerol, 15 mM imidazole) and the cells were lysed via sonication. The cell suspension was heated to 70°C for 20 min to precipitate all thermolabile components. After a centrifugation step (16 000 g, 20 min, 4°C) an immobilized metal affinity chromatography (IMAC) step was performed to purify SapA. Pre-equilibrated Ni-NTA resin was added to the supernatant of the centrifugation step and gently mixed for 90 min at 4°C. The resin was washed, and the protein eluted with buffers with an increasing imidazole concentration (20 mM sodium phosphate (pH 7.4), 300 mM NaCl, 5% glycerol, 30/400 mM imidazole). TEV-protease was added to the eluted protein and dialysed overnight at 4°C against dialysis buffer (20 mM sodium phosphate (pH 7.4), 300 mM NaCl, 5% glycerol, 1 mM DTT). In a second IMAC step, the TEV-protease and the cleaved His 6tag were removed. The cleaved protein was concentrated and applied to an S75 16/600 (GE Healthcare, Chicago, IL, USA) column [size-exclusion chromatography (SEC) buffer: 20 mM HEPES, pH 7.4, 200 mM NaCl] to perform a SEC to yield the pure SapA. The protein was concentrated and stored at À80°C.

Differential scanning fluorimetry (nDSF)
Thermal protein unfolding was measured by differential fluorescence fluorimetry using a nanoDSF instrument (Prometheus, NanoTemper Technologies, Munich, Germany) that monitors intrinsic tryptophan fluorescence. Protein and ligand concentrations were 0.5 mgÁmL À1 and 0.5 mM, respectively, in 10 mM HEPES, pH 7.4, 150 mM NaCl, 5 mM MgCl 2 buffer. The ADP-Vi state was prepared in a forward reaction by mixing ATP and vanadate and incubating the sample for 2-3 h. A thermal gradient from 20 to 90°C with a heating rate of 1°CÁmin À1 was applied. All measurements were performed in duplicates. The ratio was calculated from fluorescence intensities measured at 350 and 330 nm.

Small-angle X-ray scattering
The analysis of MsbA incorporated into Salipro nanoparticles in the apo and the ADP-Vi state by SAXS was performed at the Bio-SAXS beamline P12 [41] on the storage ring PETRA III (DESY, Hamburg, Germany). All measurements were performed at 10°C in 20 mM HEPES, pH 7.4, 200 mM NaCl, with protein concentrations of 1-5 mgÁmL À1 . The normalization and the background subtraction were performed by the automatic procedures on the beamline [42]. The radii of gyration were extracted by the Guinier approximation. The program GNOM [43] was used to calculate the distance distribution function (P (r)) and the maximal protein dimension (D max ) from the scattering curve.

Cryo-EM grid preparation
To the purified MsbA sample, 1 mM vanadate, 1 mM ATP and 1 mM MgCl 2 were added and incubated for 3 days at 4°C. The sample was prepared for cryo-EM by applying 4 µL of the sample (0.6 mgÁmL À1 ) to a glow-discharged Quantifoil holey carbon grid R2/2 (CU, 200 mesh). The grids were blotted with filter paper and plunge-frozen in liquid ethane using an FEI Vitrobot Mark IV (Thermo Fisher Scientific Ltd, Waltham, MA, USA) with zero blot force, 6 s blot time at 4°C and 95% humidity.

Cryo-EM data acquisition
Cryo-EM data were collected on the Titan Krios microscope at the ESRF [44] operated at 300 kV using EPU Movies were recorded in counting mode at a nominal magnification of 165 0009, corresponding to a pixel size of 0.827 AÁpixel À1 at the specimen level. The defocus range was set from À0.8 to À3.0 µm. The exposure time was 5 s for each movie at a dose rate of 5.42 e À Ápixel À1 Ás À1 , accumulating to a total dosage of 39.55 e À Á A À2 over 40 frames. A total of 10 302 movies were recorded. Data collection parameters are summarized in Table 2.

Image processing
Pre-processing was carried out in Scipion [45] using scripts implemented on CM01 at the ESRF, which perform beam induced motion correction [46] and contrast transfer function (CTF) estimation [47]. Particle picking and initial 2D classification were carried using crYOLO [48] and RELION [49], respectively, as implemented in Scipion [50]. The cleaned-up set of particles were exported and used for further processing in cryoSPARC [51]. Particles were further clean-up in a 3D classification follow by a heterogeneous refinement step to remove bad particles. The best classes were used for a non-uniform refinement (C1) and a second round of 3D classification was performed. All 3D classifications and refinements were carried out in cryoSPARC. The best resolved class was selected for particle CTF refinement followed by non-uniform refinement (C2). The final map consists of a set of 83287 particles. Post-processing and local resolution estimation were done in cryoSPARC. Refer to Fig. 2 for the cryo-EM data processing pipeline.

Model building and refinement
The coordinates of MsbA that were available from the Protein Data Bank under accession number 5TTP were used as starting model. This model was placed in the cryo-EM map using UCSF Chimera [52]. The cryo-EM map was autosharpened using PHENIX prior to model building (B factor: À94 A 2 ). The model was rebuild and refined iteratively using Coot [53], ISOLDE [54] and the PHENIX suite [55]. The later was also used for real space refinement and model validation [56,57]