Figures and data
Increasing membrane curvature reduces BmrA ATPase activity.
(A) Schematic representation of BmrA in Bacillus Subtilis (top left) and reconstituted in opposite orientation in liposomes (bottom) in open or apo (inward-facing in bacteria) and closed conformations (outward-facing). The green and red disks indicate the labeling position of sCy3 (donor) and sCy5 (acceptor) for smFRET experiments. The conformational change between apo and closed states occurs actively upon ATP hydrolysis ATP (top box); ATP and Vi covalently trap BmrA in the closed state (middle) while non-hydrolyzable ATP analog, AMP-PNP, binds to NBDs causing reversible protein closing (bottom). (B) Left: Cryo-EM images of EPC:bPS (90:10 w/w) apo BmrA proteoliposomes reconstituted at different temperatures: large vesicles (LV) at 20°C) and small vesicles (SV) at 4°C (scale bar: 50 nm). Right: Size distribution of LV apo BmrA, average diameter ø=123 ± 41 nm (orange, n = 594) and SV with ø=29 ± 6 nm, regardless of BmrA conformation: open (apo (light blue, n=262) or closed (ATP-Vi, dark blue, n=249). (C) Protein orientation: Fraction of remaining Alexa 488-labeled BmrA after cleavage of the NBD by trypsin, detected with TIRF microscopy. BmrA was reconstituted in LV and SV at 2 different PLR (number of molecules analyzed from left to right: n = 135, 157, 175 and 168, from 20 movies). (D) Effect of temperature on BmrA ATPase activity (in μmol of ATP consumed per min per mg of protein), in detergent (Triton X-100) (3 measurements). (E) Effect of liposome size on BmrA ATPase activity measured at 20°C and 37°C with 5 mM ATP (3 measurements).
Two-gaussian fit parameters of BmrA smFRET histograms for the 2 vesicle populations (LV and SV), at 20 °C (RT) et 33°C (HT).
Amplitude a, mean <Eapp>, and width parameter w for the Low-FRET (LF) and High-FRET (HF) peaks
Fractions of proteins in closed (Pcl) states for the two vesicle populations LV and SV, at 20 °C (RT) et 33°C (HT).
Pcl corresponds to the non-corrected High-FRET fraction calculated from the parameters of Table 1, 
smFRET shows that BmrA closed conformation depends on membrane curvature in the presence of non-hydrolysable ATP.
Experiments performed at 20°C (RT) (A,B) Cumulated histograms of apparent FRET efficiency Eapp upon addition of (A) ATP-Vi 5 mM in LV (top) and SV (bottom) (B) 5 mM AMP-PNP in LV (top) and SV (bottom). The thick line corresponds to the two-gaussian fit. Nall: number of time points in histograms, Nprot: corresponding number of proteins. (A top) Nall = 18559 and Nprot = 525. (A bottom) Nall = 7552 and Nprot = 325. (B top) Nall = 5109 and Nprot = 157. (B bottom) Nall = 5094 and Nprot = 141. For each condition, data come from two independent experiments. Error bars are calculated using bootstrapping. (C,D) Typical time traces of proteins in LV: (C) with ATP-Vi 5mM in closed state and (D) in open state. Top panel: donor (green) and FRET (red) intensities over time. Plain and dashed line represent a fit of the FRET switches. Bottom panel: intensity of acceptor (A) only (black).
Effect of curvature on conformation of BmrA.
(A,B) Cumulated histogram of Eapp for BmrA in apo conformation in LV (top A) and SV (top B) at RT (20°C), with 5 mM Mg-ATP in LV (middle A) and SV (middle B) at RT and with 5 mM Mg-ATP in LV (bottom A) and SV (bottom B) at HT (33°C). The thick lines show a two-Gaussian fit with the high-FRET peak set to <Eapp>HF in ATP-Vi. Top A: Nall = 10451, Nprot = 310. Middle A: Nall = 7743, Nprot = 226. Bottom A: Nall =2991, Nprot =1050. Top B: Nall = 2797, Nprot = 115. Middle B: Nall = 3959, Nprot = 130. Bottom B: Nall = 3235, Nprot =1093. Data come from two independent experiments and three for SV apo. Error bars are calculated using bootstrapping. (C) Cumulated histogram of Eapp for BmrA in LV (top) and SV (bottom) with 5 mM AMPPNP at HT. (top) Nall = 2759, Nprot = 823. (bottom) Nall = 2935, Nprot = 980 (D) Probability for proteins to be in closed state 
Theoretical 2-states model describing the effect of membrane mechanics on protein shape transition.
(A) Definition of geometrical parameters 2a, θop and θcl used in the theoretical model. (B) Two-state model scheme: energy landscape for BmrA alone favors the open state. With nucleotides, NBDs dimerize, and adding energy (green) favors the closed state. Membrane energy adds a contribution that further disfavors the closed state, more so for SV (blue) than LV (orange). Open-to-closed transition is thermal, while closed-to-open is either thermal (AMP-PNP)) or ATP hydrolysis-triggered. (C) Model results: probability of being in open conformation (top, with ATP or AMP-PNP) and activity (bottom, ATP turnover rate), as a function of SUV curvature, at RT and at HT. Experimental data (dots) from 
ABC exporters exhibit a large range of opening in the apo form but similar NBD organization in closed conformation.
Structures from RCSB PDB (https://www.rcsb.org/) and EMDB (https://www.ebi.ac.uk/emdb/) of BmrA and ABC exporters sharing high homology with BmrA: bacterial MsbA and human and mouse Pgp. (A) In apo state. Distance L (in Å) between red-labelled residues, equivalent to those used for BmrA labelling (C436) and PDB ID. For BmrA, the 3D cryo-EM model in detergent and the 23 Å resolution cryo-electron microscopy density map in vesicles (NBDs: white, intracytoplasmic domain ICD: green, TMD and lipid bilayer: blue) are depicted. (B) In closed state in the presence of ATP for human Pgp and BmrA. (C) Assessing the protein geometrical parameters for the model. Side view of ABC transporters transmembrane domains. The V-shape angle of the open form θop is shown for apo BmrA in detergent and for ABC exporters sharing high homology with BmrA: MsbA and Pgp. Similar angles for Pgp in detergent, MsbA in detergent and MsbA in nanodiscs. The V-shape angle of the closed conformation θcl is measured on the structure of ATP-bound BmrA. Membrane is represented in beige. For simplicity, NBDs have been removed from the 3D models. (D) Top view of the transmembrane domains at the membrane middle plane. The protein diameter 2a is deduced from 
Cryo-EM images of proteoliposomes reconstituted at 20°C (A, B, C) and at 4°C (D, E, F).
Samples have been flash-frozen 20 min after reconstitution (A, D), after 1h additional incubation at 20°C (B, E), and after 30 min additional incubation at 37°C (C, F). Bars: 50 nm. (G) Violin graph of vesicles’ diameters from A (n=226), B (n=215), C (n=301), D (n=277), E (n=254), F (n=318).
Floatation assay in a sucrose gradient.
It first shows BmrA incorporation in LV (left) and SV (right). Bottom (B), middle (M), and top (T) fractions were analyzed on a gel, indicating minimal protein aggregation and proper BmrA incorporation.
Measure of the ATPase activity as a function of ATP concentration for SV (blue) and LV (orange).
Km and Vmax are deduced from the fit to the Hill equation with a coefficient 2, since 2 ATP molecules are involved in the process:
Protein labelling optimization.
NuPage gel of BmrA. solubilized in DDM. labelled with increasing amount of sCy3 and sCy5 dyes for optimizing the protein double-labelling, at 4°C overnight incubation. Condition 2 was selected as a compromise between labeling efficiency and activity preservation.
Control experiments for smFRET
(A) Control of the single molecule regime in SV, using a double labeling of the protein and of the liposome membrane. Top: Scheme of the experiment. sCy5-labelled BmrA is reconstituted in SV that contain 0.01% fluorescent lipids (Bodipy®-FL). Colocalization of sCy5 and Bodipy®-FL signals indicates a liposome containing at least one protein. Bottom: Percentage of liposomes (per image) containing at least one BmrA, for two different PLR, 1:203000 and 1:40600. In both cases, most liposomes are empty. 120 images were analyzed. (B) Control for the cleanliness of surfaces. Surface observation after the cleaning process and surface treatment (full system but without liposomes), without any fluorescent sample. The donor signal (Exc 532 nm; Em 593 nm), the FRET signal (Exc 532 nm; Em 684 nm) and the acceptor signal (Exc 638 nm; Em 684 nm) are shown. (C) Scheme of smFRET with BmrA labelled with sCy3 and sCy5 at C436, reconstituted in a liposome immobilized with neutravidin on a coverslip passivated with PEG/PEG-biotin. (D) Control for the specificity of the liposomes tethering. Fluorescence image of 0.5% DHPE-Texas Red® liposomes incubated in the chamber for 5 min on a surface (left) treated with 20 µg/mL of neutravidin and (right) not treated with neutravidin. Scale bars: 10 µm. (E) Typical TIRF-microscopy images showing the Donor channel (Exc 532 nm Em 593 nm), the FRET channel (Exc 532 nm Em 684 nm), and the merge (Donor: green and FRET: red) (F) Control of the integrity of the liposomes after immobilization on the coverslip. Left: Liposomes containing 0.5% DHPE-Texas Red® lipids and the soluble dye pyranine in their lumen are incubated for 5min in the chamber. Potential rupture of the liposomes due to the adhesion via the PEG-biotin neutravidin interaction is detected via the fluorescence signal: intact liposomes correspond to colocalization of the green and red signals. Right: fraction of intact liposomes (per image) as a function of the PEG-biotin density on the surface, 10, 2.5, 1 and 0%. The optimal condition is obtained at 1% (*), which has been used in the whole study. 21 images were analyzed. Scale bars: 10 µm.
smFRET calibration using a DNA ruler system.
(A) Scheme of the assay using DNA. 40 base-pair (bp) double-stranded DNAs are labelled with sCy3 (Donor D) and Cy5 (Acceptor A) at different positions for sCy5 (19. 12 and 7 bps), corresponding to the distances between the dyes: LD-A = 6.7, 4.7 and 3.4 nm, respectively. (B) Distributions of the apparent FRET efficiency Eapp for the 3 DNA samples (N = 443, 460 and 538 for 19, 12 and 7 bps, resulting more than 105 time points). (C) Mean apparent FRET efficiency <Eapp> deduced from the Gaussian fit of the respective histograms (thick lines in B) as a function of the distance between the dyes (Supplementary Table S1). (D-E) Typical intensity signals of donor (green), FRET (yellow), direct acceptor (red) and corresponding apparent FRET efficiency Eapp (blue) according to time for an interdye distance (D) LD-A = 4.7 nm and (E) LD-A = 6.7 nm. Dashed grey line indicates the acceptor or donor bleaching time. Experiments were performed at 20°C.
Protein activity in the presence of ATP analogs.
(A) Normalized ATPase activity measured by spectrometry in vesicles (LV). In the presence of 5 mM ATP, BmrA ATPase activity is reduced by 80% with 2 mM Vanadate (well above Km=0.05 mM), but only by 25% with 5 mM AMP-PNP. Note that there is no ATP in our smFRET experiments with AMP-PNP (Fig. 3). (B) Protein ATPase activity as a function of temperature, measured by spectrometry in detergent (Anapoe® X-100 at 0.05%). We have checked that the protein remains inhibited by 4 mM Vanadate, independently of the temperature (blue triangles) whereas it changes significantly upon addition of 10 mM ATP (red circles). (Between 2 and 5 measurements).
2D smFRET histograms efficiency versus stoichiometry in all conditions at RT (T=20°C).
Two-dimension histograms of the FRET apparent efficiency Eapp versus the apparent stoichiometry Sapp. All instantaneous values are depicted. From top to bottom, the protein is in Apo, ATP 5 mM, AMPPNP 5 mM, ATP Vanadate 5 mM, respectively. We display data after selection process, i.e. Donor-Acceptor pairs before photobleaching of a dye. (A) Histograms for LV with the total numbers of data points (from top to bottom): Nall = 10451, 7743, 5109 and 18559. (B) Histograms for SV with Nall = 2797, 3959, 5094 and 7552.
Time average FRET efficiency Eavg per protein.
Normalized histograms of the FRET efficiency averaged over time for each protein with AMP-PNP 5 mM (top) and ATP Vanadate 5 mM (bottom) in (A) LV (orange) Nprot = 157 (top) and 525 (bottom) and (B) SV (blue) Nprot = 141 and 325.
Single Gaussian fit on apo-BmrA FRET distribution.
Cumulated histogram of apparent FRET efficiency Eapp for: (A) Apo-BmrA in V125 (Nall=10451 and Nprot=310). A single Gaussian fit (thick line) gives: amplitude a = 0.091, mean FRET efficiency <Eapp> = 0.11 and width w = 0.13. (B) Apo-BmrA in V20 (Nall=2797 and Nprot= 115). The single Gaussian fit (thick line) gives a = 0.083, <Eapp> = 0.14 and w = 0.13.
2D smFRET histograms efficiency versus stoichiometry in all conditions at HT (T=33°C).
Two-dimension histograms of the FRET apparent efficiency Eapp versus the apparent stoichiometry Sapp. All instantaneous values are depicted. From top to bottom, the protein is in Apo, ATP 5 mM, AMPPNP 5 mM, respectively. We display data after selection process, i.e. Donor-Acceptor pairs before photobleaching of a dye. (A) Histograms for LV with the total numbers of data points (from top to bottom): Nall = 2726, 2991 and 2759. (B) Histograms for SV with Nall = 2289, 3235 and 2935.
Dynamics of ATP-induced conformational changes at RT (20°C).
smFRET intensity time traces collected in LV (A-C) and SV (B-D), upon addition of ATP. (A-B) On three examples, a stable high FRET signal is measured until either the donor or acceptor dye bleaches, correlating with a drop of FRET signal. Plain and dashed line represent a fit of the FRET switches. (C-D) A conformational change is detected with anti-correlated donor (green) and FRET (red) signal, when acceptor intensity (black) remains constant, demonstrating a dynamical event. These switches were observed only twice over 226 proteins for the LV and 3 times over 130 proteins for the SV.
Dynamics of ATP-induced conformational changes at HT (33°C).
smFRET intensity time traces collected in LV (A) and SV (B), upon addition of ATP. The FRET disappears rapidly (after 1 to 1.5 s) due to photobleaching. No clear conformational switch is detected over this time period.
Time average FRET efficiency Eavg per protein at RT (20°C).
Normalized histograms of the FRET efficiency averaged over time for each protein with 5 mM ATP (A) in LV (orange) Nprot = 226 (top) and Nprot = 748 (bottom) and (B) in SV (blue) Nprot = 130 (top) and Nprot = 685 (bottom).
Schematic representation of the model.
We consider a vesicle with total area A0 (shown by solid black line in A), having a circular hole of radius a (shown by dashed line in A). The full 3D shape of the vesicle can be obtained by rotating the arc through the axis of symmetry z. The vesicle can take (A) a spherical shape or (B) a deformed shape depending upon the opening angle θP.
Comparison with experimental results and effect of size distribution on the model.
(A,B) At room temperature (RT). (C,D) At high temperature (HT). (A, C) Fraction of closed population Pcl versus liposome curvature, for ATP (blue) and AMP-PNP (green). The lines are the analytical calculations, circles are the experimental data and the open boxes are the data points that are calculated analytically using the full size distribution of the liposome. (B, D) Activity normalized to the activity in a flat membrane versus curvature. Other parameters are the same as in Fig. 4, main text. We fit the experimental data with 2a=4nm, θop=π/10, θcl=-π/10, and obtain κ=27.5 kBT.
Effect of protein shape and size on the fraction of closed population.
(A) Calculated fraction of closed population Pcl versus liposome curvature for various values of θop and θcl (2a=4 nm). For higher difference in θop and θcl, the curve is sharper. (B) Fraction of closed population Pcl for different protein sizes 2a (θop =-θcl =π/10). For larger protein size, the fraction Pcl drops faster with curvature. For both the plots, we use κ=27.5 kBT.
BmrA activity (in µmol ATP/min/mg) corrected by the inward facing fractions of proteins in the different conditions (determined as in Fig. 1C, main text)
