Asymmetry in the homodimeric ABC transporter MsbA recognized by a DARPin

: ABC transporters harness the energy from ATP binding and hydrolysis to translocate substrates across the membrane. Binding of two ATP molecules at the nucleotide binding domains (NBDs) leads to the formation of an outward-facing state. The conformational changes required to reset the transporter to the inward-facing state are initiated by sequential hydrolysis of the bound nucleotides. In a homodimeric ABC exporter such as MsbA responsible for lipid A transport in Escherichia coli, sequential ATP hydrolysis implies the existence of an asymmetric conformation. Here we report the in vitro selection of a designed ankyrin repeat protein (DARPin) specifically binding to detergent-solubilized MsbA. Only one DARPin binds to the homodimeric transporter in the absence as well as in the presence of nucleotides, suggesting that it recognizes asymmetries in MsbA. DARPin binding increases the rate of ATP hydrolysis by a factor of two independent of the substrate-induced ATPase stimulation. Electron paramagnetic resonance (EPR) measurements are found to be in good agreement with the available crystal structures and reveal that DARPin binding does not affect the large nucleotide-driven conformational changes of MsbA. The binding epitope was mapped by cross-linking and EPR to the membrane-spanning part of the transmembrane domain (TMD). Using cross-linked DARPin-MsbA complexes, 8-azido-ATP was found to preferentially photolabel one chain of the homodimer, suggesting that the asymmetries captured by DARPin binding at the TMDs are propagated to the NBDs. This work demonstrates that in vitro selected binders are useful tools to study the mechanism of membrane proteins. Background: ABC exporters are suggested to hydrolyze ATP sequentially implying the existence of asymmetries. Results: An in vitro selected binder (DARPin) was identified which binds to homodimeric MsbA at a stoichiometric ratio of 1:1. (bNBD AviC ) of MsbA. bAcrB AviC is used as a negative control. C Modulation of basal ATPase activity of detergent-solubilized MsbA in complex with DARPin_55. DARPin_55 was added to MsbA (200 nM) in a two-fold serial dilution series ranging from 50 nM to 1.6 µM leading to a stimulation of the ATPase activity of MsbA by up to 2.1-fold. Addition of the unselected DARPin E3_5 (1 µM) to MsbA does not increase its ATPase activity. DARPin_55 (1 µM) alone does not hydrolyze ATP. D The ATPase activity of MsbA was determined in the presence of DARPin_55 (1 µM) or E3_5 (1 µM) as well as in absence of DARPins at varying ATP concentrations. Addition of DARPin_55 results in a higher maximal rate of ATP hydrolysis (V max ), while the apparent affinity for ATP (K m ) is not affected. E and F The stimulations of the ATPase activity of MsbA by substrates and DARPin_55 are Lipid A was added to detergent-purified MsbA at concentrations of 0, 0.05, and E , daunomycin was added at concentrations of 0, was measured in the absence presence

2 the outer leaflet of the inner membrane in bacteria (2) and the msbA gene is essential in E. coli (3). MsbA is homologous to the multidrug transporter LmrA of gram-positive Lactococcus lactis (4) and mammalian multidrug resistance transporters such as P-glycoprotein (5) and shows overlapping substrate specificity with them (6,7). At a molecular level, ABC transporters consist of four domains, two transmembrane domains (TMDs) and two nucleotide-binding domains (NBDs). In ABC exporters, the TMDs are composed of at least six transmembrane helices, which form the substrate transport pathway and provide specificity. The NBDs are also known as ATP-binding cassettes and drive the transport cycle by binding and hydrolyzing ATP. The motional energy of the NBDs are coupled to conformational changes in the TMDs which alternate between an inward-and an outward-facing state and thereby form a pathway to mediate unidirectional transport across the membrane (8). Current molecular understanding of ABC exporters is based on crystal structures of Sav1866 (8) solved at 3 Å in its outward-facing state, P-glycoprotein solved at 3.8 Å in its inward-facing state (5) and the structures of three homologues of MsbA solved in three different conformations at resolutions ranging from 3.7 Å to 5.5 Å (9). Based on these crystal structures, large conformational changes are expected to occur in ABC exporters upon nucleotide binding and hydrolysis, which could be confirmed by cross-linking experiments and EPR measurements (10)(11)(12)(13)(14). The mechanistic details of how the binding and hydrolysis of ATP at the NBDs is coupled to the transport process at the TMDs are still a matter of controversy. In many heterodimeric ABC transporters, one of the two ATPase sites deviates from the consensus sequence and is therefore called the degenerate site (15). Such heterodimeric ABC exporters mainly hydrolyze ATP at the consensus site and thus operate in an asymmetric fashion. In homodimeric ABC exporters such as BmrA as well as in P-glycoprotein which is a fused heterodimer but exhibits two consensus ATPase sites, biochemical experiments demonstrated that only one ATP molecule is trapped by vanadate in the posthydrolytic state per ABC transporter dimer (16,17). It was concluded that ABC exporters hydrolyze the two ATP molecules sequentially instead of simultaneously, possibly involving a mechanism of alternating catalytic sites (18). This mechanism implies that the NBD homodimer must switch to an asymmetric conformation before ATP is hydrolyzed at one of the two catalytic sites. In support of this notion, asymmetric homodimers have been observed in crystal structures of the ATP-bound NBDs of the ABC transporter HlyB (19). On the other hand, current crystal structures of outward-facing MsbA and Sav1866 feature two bound AMP-PNP at the NBDs (thus representing a pre-hydrolytic state) and display only minor structural differences between the two chains, that are related by two-fold noncrystallographic symmetry (9,20). Molecular dynamics simulation using the symmetric Sav1866 and MsbA structures as starting points revealed that the NBDs readily switch to an asymmetric state in which only one molecule of ATP enters the hydrolysis reaction at a time (21,22). Further work addressing the question of asymmetry in ABC exporters is clearly needed to resolve the apparent discrepancies between asymmetries found based on biochemical experiments and the symmetric crystal structures of the homodimers Sav1866 and MsbA. Designed Ankyrin Repeat Proteins (DARPins) represent a well-established randomized scaffold used as an alternative to antibody fragments (Fabs and scFvs) and have been selected against a variety of target proteins (23)(24)(25). A growing number of DARPin co-crystal structures demonstrate the successful application of such designed binding proteins as crystallization aids (24). Besides possible improvements of crystal diffraction, DARPins have the potential to trap the targeted protein in alternate conformations. DARPins selected against the multidrug transporter AcrB for example trap this protein in an asymmetric state with two DARPins bound to the homotrimer (25,26). In the absence of DARPins, AcrB was initially crystallized in its symmetric and later in its asymmetric state (27,28). Here, we describe the selection of DARPins against detergent-purified MsbA using ribosome display. We identified a DARPin which binds to the homodimeric transporter in a 1:1 stoichiometric ratio. The DARPin binding epitope was mapped to the TMDs and was found to be located away from the 2-fold symmetry axis of MsbA, suggesting that the DARPin recognizes asymmetries in MsbA. DEER measurements revealed that addition of the DARPin to MsbA does not influence the large ATPinduced conformational changes during transporter cycling. Nevertheless, DARPin binding enhances the basal ATPase activity of the transporter without impairing the substrate-induced stimulation suggesting that the catalytic cycle is accelerated when asymmetric MsbA conformations are populated.

EXPERIMENTAL PROCEDURES
Purification, biotinylation and spin labeling of MsbA and DARPin -A detailed description of the cloning, expression and purification of MsbA is given in the Supplementary Experimental Procedures. MsbA was expressed in E. coli C43 (DE3) cells from a pBAD vector and purified using β-UDM as detergent. DARPin expression and purification has been described previously (29). Enzymatic biotinylation was facilitated by fusing an Avi-tag sequence to the C-terminus of MsbA. Single cysteine mutants at positions A25C, R103C, N191C, S206C and S246C of MsbA were introduced by site directed mutagenesis taking cys-less msbA (C88S/C315S) as template. Likewise, single cysteine mutants were also introduced into the N-and Cterminal caps of DARPin_55 (which is cys-less by design) at the positions S12C, K16C, E29C, D151C and D160C. For methanethiosulfonate cross-linking and spin labeling, single cysteine mutants of MsbA and DARPin were purified by supplementing 5 mM dithiothreitol (DTT) which was removed (PD-10 column, GE Healthcare) prior to labeling with 10 times molar excess of MTSSL ( (1-Oxyl-2,2,5,5tetramethyl-∆3-pyrroline-3-methyl) methanethiosulfonate; Toronto Research Chemicals). Excess spin label was removed by size exclusion chromatography. For the EPR measurements, DARPin-MsbA complexes were made by spin labeling the freshly purified proteins individually, followed by gel-filtration of the protein complex.

Selection and Screening of DARPin specific to
MsbA -The N3C DARPin library (23) was used for binder selection by ribosome display as described (30,31). Four rounds of selections were carried out following the solution panning strategy (29). 0.05% of β-UDM were used to maintain MsbA in active form. Single clones of the enriched pool of DARPins from the 4 th selection round were expressed from the vector pQE30 myc5 (29) in E. coli XL-1 Blue yielding DARPins carrying an Nterminal RGS-His 6 tag (M-R-G-S-H-H-H-H-H-H) and a C-terminal Myc 5 -tag (5 times amino acid sequence M-E-Q-K-L-I-S-E-E-D-L-N-E). Single DARPin clones were analyzed for binding to bMsbA AviC by crude cell extract ELISAs as described previously (29). Cysteine cross-linking of MsbA and DARPin_55 mutants -For cross-linking studies, two thiolspecific homobifunctional cross-linkers namely M5M (1,5-Pentanediyl Bismethanethiosulfonate, Toronto Research Chemicals) and M11M (3,6,9-Trioxaundecane-1,11-diyl-bismethanethiosulfonate, Toronto Research Chemicals) with spacer arms of 9.1 Å and 16.9 Å were used (32). The MsbA and DARPin_55 mutant proteins were mixed at ratios of 1:3 and incubated with 0.5 mM of cross-linker for 15 minutes at 4 °C (if not stated otherwise). The reactions were stopped by addition of SDS sample buffer and separated by non-reducing SDS-PAGE. The ability of DARPin_55_29C to cross-link to MsbA_191C was also tested using inside-out membrane vesicles (ISOVs) containing overexpressed MsbA_191C (with detergent purified MsbA_191C, this cross-link is readily formed). To this end, MsbA_191C containing ISOVs (6 mg in 3ml) were mixed with DARPin_55_29C (35 mg in 7 ml) and M11M (1 mM) was added. After incubation for 30 min, N-ethylmaleimide (NEM, 10 mM) was added to react with the remaining free cysteines for another 30 min. Subsequently, MsbA was purified according to the standard protocol and subjected to non-reducing SDS-PAGE (Fig. S8).

SEC of isolated
ATPase assay -The ATPase activity of purified protein was measured from the release of P i from 4 ATP in a colorimetric malachite green assay as described (33). Briefly, purified protein (200 nM) in 50 mM K-HEPES buffer pH 7.4 containing 0.05 % β-UMD, 2.5 mM MgCl 2 and 2 mM ATP was incubated at 30 °C in a total reaction volume of 50 µl for 10-15 minutes. MsbA was inactivated by heating at 80°C for 2 min and free P i was measured using malachite green solution. The OD 600 of the samples was measured in a Microplate reader Infinite M100 (TECAN) and compared with the OD 600 of standards containing 0.5-4 nmol P i . Readings were corrected for P i contamination originating from the ATP and buffer. DARPin_55stimulated ATPase activity of MsbA was performed by adding DARPin_55 at concentrations as indicated to MsbA (200 nM) 15 min prior to the assay start. All cysteine mutants used in this study showed ATPase activity comparable to the wild-type. Spin labeling did not affect the activity of MsbA (Supplementary Table S1).

Reconstitution of MsbA into liposomes -
Detergent-purified MsbA was reconstituted into lipid membrane vesicles made of polar E. coli lipids and egg phosphatidylcholine mixed at a ratio of 3:1 (34). For EPR and DARPin binding experiments, the reconstitution buffer was 20 mM Tris/HCl pH 7.4, 150 mM NaCl, for ATPase activity measurements it was 50 mM K-HEPES pH 7.0. MgCl 2 was added only where indicated shortly before performing ATPase activity or EPR experiments in order to avoid aggregation of the liposomes. The amount of reconstituted MsbA was determined by the Biorad D C Protein Assay according to manufacturer's protocol and compared BSA protein concentration standards. Binding of DARPin_55 to the reconstituted MsbA was tested by incubating the DARPin_55 with reconstituted MsbA (3:1 ratios) or empty liposomes for 15 minutes at 4°C. Unbound DARPin_55 was removed by three buffer washes after centrifugation at 20000 g for 20 minute at 4 °C. The pelleted liposomes were dissolved in SDS buffer and were separated by SDS-PAGE (15% acrylamide gel) followed by SYPRO ruby staining (Fig. S6B).
Distance measurements by pulse EPR experiments -Dipolar time evolution data were acquired using the four-pulse DEER (Double Electron Electron Resonance) experiment (35). DEER traces were measured in a home-made high power Q band spectrometer equipped with a homemade oversized resonator (36). In Q band all pulse lengths were set to 12 ns and deuterium modulation artefacts were averaged (8 times d1 time increment of 16 ns). The pump frequency was placed at the maximum intensity of the nitroxide spectrum, and the observer frequency 80 MHz lower. All samples were prepared independently from different batches and the DEER traces were found to be reproducible. Samples contained 10% (v/v) d8-glycerol and were shock-frozen in liquid nitrogen before measurement. Data analysis was performed with the software DeerAnalysis 2010 (37). The simulation of the possible spin label rotamers attached at selected positions in MsbA was performed using the Matlab program package MMM (freely available at http://www.epr.ethz.ch/software/index) based on a rotamer library approach (38).

8-N 3 -[α-32 P]ATP photolabeling -MsbA191C
(1.5 µM) and DARPin_55_29C (6 µM) were crosslinked with M11M (1.5 mM) in 20 mM Tris pH 7.5, 150 mM NaCl, 0.05 % β-UDM containing 3 mM MgCl 2 . 8-N 3 -[α-32 P]ATP (1 µM, 0.6 Ci/mmol) was added to 10 µl protein sample (quadruplicates) and the reaction mixtures were incubated at 25 °C for 5 minutes followed by UV irradiation (100 millijoules, UV cross-linker, Hoefer) on ice. Labeled crosslinked MsbA-DARPin complex was analyzed by SDS-PAGE and photolabeling was quantified using a phosphor imager (Molecular Dynamics, Storm 840). To analyze the amount of cross-linking, four aliquots were supplemented with cold 8-N 3 -ATP followed by UV radiation, SDS-PAGE and staining with SYPRO Ruby (Invitrogen). The fluorescent signal was read with the LAS-3000 imaging system and the bands were quantified using the Aida software (Raytest). secondary-active multidrug transporter AcrB (both prepared as biotinylated Avi-tagged proteins) were used (Fig. S3). This assay revealed 37 MsbAspecific DARPins that were further analyzed by size-exclusion chromatography. 27 of these DARPins displayed various degrees of oligomerization (soluble aggregates) and 6 of them showed monomer/dimer equilibrium and were therefore excluded. The remaining 4 MsbA-specific DARPins were monomeric but only one DARPin (named DARPin_55) formed a stable complex with MsbA that could be isolated by size exclusion chromatography (Fig. S4). The affinity of DARPin_55 for MsbA was estimated by competition ELISA (23), in which the binding of bMsbA AviC to immobilized DARPin_55 myc5 was competed by pre-incubation with increasing amounts of free DARPin_55. The dissociation constant K D derived from these experiments was 80 nM (Fig.  S5). Hence, from 2000 potential DARPin binders analyzed, we could identify one specific highaffinity binder (Fig. 1A).

In vitro selection and identification of
The ATPase activity of MsbA is stimulated upon complex formation with DARPin_55 -The effect of DARPin binding to MsbA was studied with respect to the ATPase activity of the transporter. The basal ATPase activity of detergent-purified MsbA (200 nM) was stimulated up to 2.1 fold upon complex formation with DARPin_55 (serial dilution from 50 nM to 1.6 µM) in a dose-dependent manner reaching its maximum at a concentration of 400 nM DARPin_55 (Fig. 1C). Addition of the control DARPin E3_5 (39,40) did not alter the ATPase activity of MsbA and DARPin_55 alone did not hydrolyze ATP. By determining the Michaelis-Menten kinetics of ATP hydrolysis it was found that DARPin_55 binding increases the maximal velocity V max (increase from 162±9 to 255±8 nmole P i /min/mg MsbA) while the apparent K m of the ATPase activity remains unchanged (102±22 and 117±14 µM ATP for MsbA and the MsbA/DARPin_55 complex, respectively) (Fig.  1D). Lipid A and daunomycin are known substrates of MsbA and were found to stimulate the ATPase activity of the detergent-purified transporter about two fold in the absence (41) and 3-4 fold in complex with DARPin_55. This implies that drug-induced and DARPin_55-mediated stimulations of the basal ATPase activity of MsbA are additive ( Fig. 1E and F). Therefore, DARPin binding does not appear to interfere with MsbA's capability to bind its cargo substrates. MsbA and the MsbA_E506Q (42) mutant (whose ATPase activity is abolished) were reconstituted in proteoliposomes (34). The addition of lipid A and daunomycin stimulate the ATPase activity of reconstituted MsbA about two-fold as described previously (Fig. S6) (41). In contrast to detergent-purified MsbA the basal ATPase activity of reconstituted MsbA was not stimulated by the addition of DARPin_55, because DARPin_55 does not bind to membrane-embedded MsbA as discussed below (Fig. S7).
Binding of DARPin does not impair MsbA's ability to undergo conformational changes -To study the possible structural effects induced by DARPin binding to the transporter, single cysteine mutants of MsbA were labeled with MTSSL (methanothiosulfonate spin label) and interspin distances between the two labeled sites were measured by Double Electron Electron Resonance (DEER) in the apo-and AMP-PNP-bound states. The DEER analysis on two representative spinlabeled sites (position 191 in the TMDs and position 539 in the NBDs) is presented in Fig. 2. The 191-191 distance in the apo-state centered at 5.8 nm was found to be in line with the apo crystal structure (43), showing a C α -C α distance of 4.7 nm. The distance decreased to 4.1 nm upon addition of AMP-PNP and MgCl 2 ( Fig. 2A), which fully agrees with the interspin distance simulated with the program MMM (38) based on the AMP-PNP structure of MsbA (43). Upon complex formation with DARPin_55, MsbA showed the same distance distributions, indicating that the detergentsolubilized MsbA retained the ability to accomplish the conformational changes induced by nucleotide binding (Fig. 2A).
The AMP-PNP-state exhibited a sharp 1.8 nm distance between the two 539 labels located in the NBDs, both in the absence and presence of DARPin_55 (Fig. 2B). The obtained distance is again in perfect agreement with the crystallographic data, strongly indicating that DARPin binding does not lead to major structural changes in MsbA. The apo-state of the 539 mutant of MsbA showed a peak at about 6 nm in line with the inverted V-shaped apo crystal structure (43), which is characterized by a C α -C α distance of 6.4 nm. A second broad peak between 2 and 4 nm was present which might represent intermediate conformations adopted during NBD closure or disengagement. However, these short distances were only observed in this particular mutant and they were strongly suppressed after reconstitution into liposomes (Fig. S6). Interestingly, adding DARPin_55 to the apo-state of the 539 mutant of MsbA diminished the short distances but did not affect the 6 nm peak suggesting a stabilization of the open-apo structure by DARPin_55 in analogy to the reconstitution into liposomes (Fig. 2B, Fig. S6). Taking into consideration all tested mutants, the inverted Vshaped crystal structure and the outward-facing AMP-PNP bound structure (9) appear to describe the conformational changes of detergent-solubilized MsbA appropriately.
DARPin_55 binds to the transmembrane domain of detergent-purified MsbA -To delineate the binding epitope of DARPin_55 on full-length MsbA we used purified sub-domains of MsbA, namely the TMD and the NBD. For this purpose, the domains were purified as Avi-tagged biotinylated proteins (bTMD AviC and bNBD AviC ) and binding to DARPin_55 was assessed by the same ELISA setup as used to identify MsbA-specific DARPins described above (Fig. 1B, Fig. S2). This ELISA experiment revealed that DARPin_55 binds to the transmembrane domain (bTMD AviC ) but not to the nucleotide binding domain (bNBD AviC ) of MsbA. Binding of DARPin_55 to bTMD AviC resulted in a comparatively smaller ELISA signal than binding to full length bMsbA AviC . This could be due to partially unstructured regions on the TMDs in the absence of the NBDs. Binding of DARPin_55 was further tested using membrane vesicles containing overexpressed MsbA and MsbA reconstituted in proteoliposomes (Fig. S7). In contrast to detergentpurified MsbA, DARPin_55 did not bind to the membrane-embedded transporter indicating that the binding epitope on MsbA cannot be accessed by DARPin_55 because it is (at least partially) masked by the lipid bilayer.
Cysteine cross-linking of DARPin and MsbA mutants -In order to further specify the binding epitope of DARPin_55 on MsbA, single cysteines were introduced into MsbA and DARPin_55 and proximities between these cysteines were detected using the thiol-specific homobifunctional crosslinkers M5M and M11M. A total of nine single cysteine mutants were made at different positions in the cytoplasmic part of the TMD of MsbA and in DARPin_55 (four in MsbA at positions 103, 191, 206 and 246 and five in DARPin_55 at positions 12, 16, 29, 151 and 160). Using ELISA, each mutated DARPin was confirmed to bind to every single cysteine MsbA mutant (data not shown). All possible combinations of MsbA and DARPin_55 cysteines mutants were then cross-linked as described in the experimental procedures. The single cysteine mutants of MsbA alone did not form any cross-linked products in the presence of M11M and M5M (Fig. 3A) nor could any of the DARPin mutants be cross-linked to cys-less MsbA (Fig. S8). Cysteines introduced at the N-terminus of DARPin_55 (at positions 12, 16 and 29) were efficiently cross-linked to MsbA_191C and to a lesser extent to MsbA_206C and MsbA_246C with both cross-linkers (Fig. S8). MsbA_103C formed only faint cross-linking products with all DARPin mutants tested. Likewise, the cysteines at the Cterminus of DARPin_55 (at positions 151 and 160) were only poorly cross-linked with the four MsbA single cysteine mutants with either cross-linker. These faint cross-linking products seemed promiscuous and were therefore considered as background. In summary, the N-terminus of DARPin_55 is in close proximity to the MsbA_191C position whereas the orientation of the C-terminal part of DARPin_55 could not be resolved by only this analysis.
One molecule of DARPin_55 binds to the MsbA homodimer -The maximal degree of DARPin crosslinking to MsbA never seemed to exceed 50 % of the MsbA chains (Fig. S8). To further investigate this phenomenon, cross-linking between DARPin_55_29C and MsbA_191C was carried out in the absence of nucleotides as well as in the presence of AMP-PNP or ATP-vanadate using increasing amounts of DARPin (Fig. 3). Interestingly, cross-linking was saturated at a stoichiometry of approximately one DARPin binding to homodimeric MsbA in its apo-state (Fig.  3A). The exact stoichiometry was quantified by SYPRO Ruby staining to 0.73 ± 0.01 DARPins cross-linked to MsbA (Fig. 4A). The presence of AMP-PNP or ATP-vanadate appears to diminish the cross-linking reaction ( Fig. 3B and C), which might be caused by a longer distance between the thiolgroups in this state (see EPR data below) and/or a decreased binding affinity of the DARPin for the nucleotide-bound state(s) of MsbA. Despite multiple attempts, we were not able to perform surface plasmon resonance (SPR) measurements which would have allowed for the determination of binding constants of DARPin_55 recognizing different conformational states of MsbA. The ATPase activity of the DARPin_55_29C/MsbA_191C complex is stimulated by a factor of two compared to MsbA_191C alone and remains unchanged upon cross-linking using M11M (Fig.3D). This indicates that the DARPin_55 remains bound to MsbA during its entire catalytic cycle. Binding stoichiometries were further studied by gel filtration experiments. The MsbA/DARPin_55 complex was separated from excess DARPin_55 by size-exclusion chromatography (Fig. S4) followed by a quantitative analysis of DARPin_55 and MsbA using the protein chip technology (Agilent Technologies) (Fig. S9). In agreement with the cross-linking results, the ratio between DARPin_55 and MsbA dimer was quantified to 0.8. The binding stoichiometries in the gel filtration assay as well as in the cross-linking experiment were slightly less than one DARPin bound per MsbA dimer. For the gel filtration assay, this is likely due to partial dissociation of DARPin_55 from MsbA during the size-exclusion run whereas in the cross-linking experiment, some of the thiol-moieties of the cysteines might have undergone oxidation during the purification and as a consequence would not react with the methanethiosulfonate cross-linker. At this point, two scenarios can explain the binding stoichiometry of one DARPin bound per homodimeric MsbA: i) Binding of the first DARPin molecule to MsbA sterically prevents binding of the second or ii) the DARPin recognizes asymmetries in MsbA.

EPR-distances between DARPin_55 and MsbA -
In order to rule out possible steric clashes between the two DARPins in the complex, the binding epitope of DARPin_55 on MsbA was mapped by EPR. Various combinations of five spin-labeled single cysteine mutants spanning the MsbA structure (positions 25,65,103,191,482) and two spinlabeled DARPin_55 (positions 29 or 160) were measured to extract interspin distances by DEER.
Due to the 1:1 stoichiometric ratio between DARPin_55 and MsbA, binding of one spin-labeled DARPin to the homodimeric spin-labeled transporter gives rise to a three-spin complex. This was confirmed by measuring the spin concentration in the MsbA-DARPin complexes using continuous wave EPR (data not shown). To disentangle the intra-MsbA from the MsbA-DARPin distances, we measured the MsbA alone and in the presence of the DARPin. To reduce the known effects of three-spin artefacts in the distance distributions (44), the intra-MsbA and the MsbA-DARPin distances should be as distinct as possible.
Among all combinations tested, only two (MsbA_191R1-DARPin_55_29R1 and MsbA_65R1-DARPin_55_160R1, with R1 denoting the MTSSL spin-labeled side chain) resulted in extractable MsbA-DARPin distances (Fig. 5) (all DEER distances are presented in Fig. 6). The intra-MsbA_191R1 distances in the apo-and AMP-PNPstates are shown in Fig. 5A. Addition of DARPin_55_29R1 resulted in the appearance of a short distance peak at 1.5-2 nm confirming the close proximity of the N-terminal region (N-cap) of the DARPin_55 to the position 191 in the TMDs already observed by cross-linking (Fig. S8). Interestingly, the distance between these two positions becomes slightly longer (~2 nm) in the presence of AMP-PNP, whereas the intra-MsbA_191R1 distance remains unchanged as seen above with unlabeled DARPin_55 (Fig. 5A, lower panel and Fig. 2A). The longer distance observed between the two nitroxide labels in MsbA and DARPin_55 might explain the reduced degree of M11M mediated cross-linking observed upon addition of nucleotides ( Fig. 3B and  C).
Whereas the position of the N-cap of DARPin_55 with respect to the TMD could be readily mapped by cross-linking and EPR, the orientation of the DARPin's C-cap was challenging to find, also because no cross-link could be established between the 151 and 160 positions of DARPin_55 and any of the MsbA mutants used (Fig. S8). After screening by DEER a number of complexes with DARPin_55 labeled at the C-cap (position 160), we found a clear distance constraint between the C-cap and the extracellular region of the transporter (position 65). According to the structural changes expected from the X-ray data, addition of AMP-PNP to isolated MsbA resulted in an increase of the 65-65 distance from 4 to 6 nm (Fig. 4B). When DARPin_55_160R1 was analyzed in complex with MsbA_65R1, we found a distinct MsbA-DARPin_55 distance centered at 3.9 nm in the AMP-PNP state, whilst the overlap with the intra-MsbA distance prevented a clear MsbA-DARPin_55 distance assignment in the  (Fig. 6). Based on these observations we could rule out that the C-cap is oriented towards the NBDs, thus the transmembrane domain of MsbA seems to be covered at least partially by the DARPin_55. Importantly, the DARPin binding epitope was found to be far away from the 2-fold symmetry axis of MsbA, thus ruling out the possibility of steric clashes between two DARPins as the cause for the stoichiometric ratio observed.
The ATPase sites of asymmetrically stabilized MsbA display different nucleotide binding properties -Cross-linking DARPin_55_29C to asymmetric MsbA_191C provides the unprecedented advantage that the conformationally unequal chains of homodimeric MsbA can be separated by SDS-PAGE, thus questions as whether the two ATPase sites of asymmetric MsbA display differences with respect to nucleotide binding can be for the first time addressed. The cross-linked DARPin-MsbA complex was incubated with 8-N 3 -[α-32 P]ATP followed by UV cross-linking of the nucleotide to the two chains of MsbA. Subsequently, the proteins were separated by SDS-PAGE and the amount of cross-linking was determined by autoradiography (Fig. 4). The ratio between the bands of DARPin_55 cross-linked to MsbA and MsbA alone was quantified to be 0.73 ± 0.01 by SYPRO Ruby staining. Interestingly, the MsbA chain cross-linked to DARPin_55 showed a stronger photolabeling with 8-N 3 -[α-32 P]ATP than the MsbA chain alone (ratio between MsbA-DARPin55 to MsbA of 1.39 ± 0.07). This indicates that the MsbA monomer to which the DARPin was cross-linked to was photolabeled 1.9 times stronger than the free MsbA chain.

DISCUSSION
Asymmetries in the NBDs of ABC exporters arising from trapping of nucleotides by vanadate have been discussed since the mid-nineties when Urbatsch and Senior first described the catalytic cycle of P-glycoprotein and later were also found in the homodimeric transporter BmrA (16,17). Using an in vitro selected DARPin binder, we provide here novel evidence supporting considerable asymmetries to be found in a well-studied ABC exporter.
Our biochemical analysis shows that DARPin_55 binds to the TMDs of MsbA in a stoichiometry of one DARPin per homodimeric transporter. The binding epitope was mapped to the TMDs of MsbA and therefore, the possibility that binding of the first DARPin sterically hinders binding of a second DARPin to the symmetry-related epitope on MsbA could be excluded. In further support of this notion, DARPin binding does not affect the closure of the NBDs and the transition to the outward-facing state of MsbA upon addition of AMP-PNP. The DARPin therefore recognizes asymmetries in MsbA.
Asymmetric homodimers are not unusual. The epidermal growth factor receptor from Drosophila has been recently crystallized in an asymmetric state in complex with two growth factor ligands binding in a low-and a high-affinity mode (45). Caspase-9 is activated by dimerization and was crystallized as asymmetric homodimer exhibiting one active and one inactive catalytic site (46). However, despite serious efforts we did not succeed to solve the cocrystal structure of the MsbA-DARPin_55 complex and therefore the molecular details of the asymmetries in MsbA remain elusive.
Epitope mapping revealed that the DARPin binding site includes membrane-spanning portions of the TMD (Fig. 6). There are two observations which further support this unexpected finding. First, DARPin_55 only binds to MsbA in its detergentpurified but not in its membrane-bound form (Fig.  S7). In a detergent-purified MsbA sample, free detergent monomers, protein-free detergent micelles and detergent bound to the membrane protein are in equilibrium (47). There is constant rapid exchange of detergents molecules between these pools. Lipid bilayers in contrast are more rigid (48). It is therefore likely that DARPin_55 partly displaces the detergent micelle, but not the relatively stiffer lipid membrane to bind to its epitope. Second, in agreement with the positioning of DARPin_55 with respect to the membrane boundary, the randomized positions in the β-turns of the second and the third DARPin repeat exclusively harbor hydrophobic amino acids (Fig. S10). These residues could mediate hydrophobic contacts with the transmembrane helices of MsbA, whereas in the first repeat, there are two glutamates which could interact with the cytoplasmic portion of the TMD. Importantly, DARPin_55 only recognizes MsbA, but states of the transporter from an ensemble of structurally similar conformations which exhibits increased basal ATPase activities and is fully compatible with its function to bind lipid A and drugs.         MsbA482-DARPin160 > 7 nm n.d.

nm
n.a. not assigned; n.d. not determined.