Tripartite assembly of RND multidrug efflux pumps

Abstract

Tripartite multidrug efflux systems of Gram-negative bacteria are composed of an inner membrane transporter, an outer membrane channel and a periplasmic adaptor protein. They are assumed to form ducts inside the periplasm facilitating drug exit across the outer membrane. Here we present the reconstitution of native Pseudomonas aeruginosa MexAB-OprM and Escherichia coli AcrAB-TolC tripartite Resistance Nodulation and cell Division (RND) efflux systems in a lipid nanodisc system. Single-particle analysis by electron microscopy reveals the inner and outer membrane protein components linked together via the periplasmic adaptor protein. This intrinsic ability of the native components to self-assemble also leads to the formation of a stable interspecies AcrA-MexB-TolC complex suggesting a common mechanism of tripartite assembly. Projection structures of all three complexes emphasize the role of the periplasmic adaptor protein as part of the exit duct with no physical interaction between the inner and outer membrane components.

ARTICLE

Received 30 Nov 2015 | Accepted 15 Jan 2016 | Published 12 Feb 2016

Laetitia Daury1,2, Franois Orange1,2, Jean-Christophe Taveau1,2, Alice Verchre3, Laura Monlezun3,Cline Gounou1,2, Ravi K.R. Marreddy4, Martin Picard3, Isabelle Broutin3, Klaas M. Pos4 & Olivier Lambert1,2

Tripartite multidrug efux systems of Gram-negative bacteria are composed of an inner membrane transporter, an outer membrane channel and a periplasmic adaptor protein. They are assumed to form ducts inside the periplasm facilitating drug exit across the outer membrane. Here we present the reconstitution of native Pseudomonas aeruginosa MexAB OprM and Escherichia coli AcrABTolC tripartite Resistance Nodulation and cell Division (RND) efux systems in a lipid nanodisc system. Single-particle analysis by electron microscopy reveals the inner and outer membrane protein components linked together via the periplasmic adaptor protein. This intrinsic ability of the native components to self-assemble also leads to the formation of a stable interspecies AcrAMexBTolC complex suggesting a common mechanism of tripartite assembly. Projection structures of all three complexes emphasize the role of the periplasmic adaptor protein as part of the exit duct with no physical interaction between the inner and outer membrane components.

DOI: 10.1038/ncomms10731 OPEN

Tripartite assembly of RND multidrug efux pumps

1 Universit de Bordeaux, CBMN UMR 5248, Bordeaux INP, IECB, Pessac F-33600, France. 2 CNRS, CBMN UMR 5248, Pessac F-33600, France. 3 Laboratoire de Cristallographie et RMN Biologiques, UMR 8015, CNRS, Universit Paris Descartes, Facult de Pharmacie, 4 Avenue de lObservatoire, Paris 75006, France. 4 Institute of Biochemistry, Goethe-University Frankfurt, Max-von-Laue-Str. 9, D-60438 Frankfurt am Main, Germany. Correspondence and requests for materials should be addressed to O.L. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 7:10731 | DOI: 10.1038/ncomms10731 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms10731

Gram-negative efux pumps of the Resistance Nodulation cell Division (RND) superfamily are exporters of biological metabolites and antimicrobial compounds, thus

playing a prominent role in the bacterial resistance, which has nowadays become a major health concern1,2. The inner membrane located RND pumps are driven by the proton motive force and as a part of a tripartite system, working in conjunction with an outer membrane factor (OMF), and a periplasmic membrane fusion protein (MFP), the latter assumed to link the RND component to the OMF3. Well-studied examples of tripartite RND systems are MexABOprM of Pseudomonas aeruginosa and AcrABTolC of Escherichia coli3. For these systems, high-resolution X-ray structures are available for the single components TolC4, AcrA5, AcrB610, OprM11,12, MexA13,14 and MexB15. OprM/TolC possesses a trimeric organization consisting in a 4-nm-long transmembrane domain comprising 12 strands that form a b-barrel and a 10-nm-long periplasmic domain comprising 12 a-helices and a mixed a/b equatorial domain11. MexB/AcrB forms a trimer in which each protomer is made of a 12 transmembrane a-helices domain and a large periplasmic part comprising a porter and a funnel domain extending 7 nm away from the inner membrane inside the periplasm69,15. MexA/AcrA is arranged in four consecutive domains, that is, membrane proximal, b-barrel, lipoyl and a-helical hairpin domains. MexA/AcrA has been shown to be anchored to the inner membrane via palmitoylation of an N-terminal cysteinyl residue13,14,16. It has been postulated that drugs are transported from the periplasmic side across the outer membrane in an energy-dependent manner via the RND protein and the OMF channel79,1521. This intriguing transport mechanism is suggested to occur via a peristaltic mode through the protomers of the trimeric RND component caused by consecutive functional cycling of the protomers through three different states (loose, tight and open or access, binding and extrusion)810,20,2224. Despite these structural and computational insights, only a few inhibitors of these efux pumps have been described thus far22,2528. These compounds function as competitive inhibitors or impair the proper binding of substrates and are in one case25 also transported by the efux pump system, albeit at a very low rate. To embark on the development of allosteric inhibitors, for example, those preventing the tripartite setup, the assembly of the tripartite system itself has to be understood.

This is particularly challenging, since these systems span two different membranes and the periplasm of the Gram-negative cell, hence studies related to the assembly mechanism face many methodological difculties. In the search of understanding the assembly mechanism, bipartite MFPRND and MFPOMF complexes have been reconstituted in vitro. AcrAAcrB and AcrATolC interactions have been conrmed with detergent-solubilized proteins2931. Recently a crystal structure of the heavy-metal CusBA transporter revealed six MFP proteins interacting with a RND trimer32. In addition, the architecture of bipartite OprMMexA complexes sandwiched between two lipid membranes studied by cryo-electron tomography revealed a 21-nm intermembrane distance33. Evidence for a direct interaction between RND and OMF relied on in vitro AcrBTolC binding30 and in vivo cross-linking studies34,35 suggesting a limited interface between these two membrane proteins16.

To date, there are only few studies reporting on the assembly of the tripartite complex. In 2011, AcrABTolC assembly immobilized on a surface has been monitored by plasmon resonance surface30. And very recently, single-particle electron microscopy (EM) models, in one case including a fourth partner36, AcrZ37, have been described, where the detergent-solubilized

tripartite AcrABTolC38 or tetrapartite AcrABZTolC36 setup was stabilized by genetic fusion constructs of the complex components (and chemical cross-linkers36).

Here we report the reconstitution of native MexABOprM, AcrABTolC and interspecies AcrAMexBTolC complexes using nanodisc (ND) technology39. The visualization by single-particle EM reveals tripartite complexes made of the inner and outer membrane protein components linked together via the periplasmic adaptor protein emphasizing its role as part of the exit duct with no physical interaction between the inner and outer membrane components.

ResultsProtocol of tripartite assembly using NDs. The rationale for the reconstitution of tripartite complexes was based on the insertion of the integral membrane proteins (that is, OMF or RND) into NDs. On detergent removal, the membrane proteins (MexB, AcrB, OprM and TolC) were inserted into a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC)-containing NDs whose size is limited by the membrane scaffold protein (MSP)40 wrapped around the hydrophobic core of the lipids (Fig. 1a). The control of the assembly process relied on the insertion of a single molecule per ND, which necessitated the use of two MSP differing in size (MSP1D1 or MSP1E3D1) because of the respective diameters of the transmembrane domains of the RND and OMF proteins (RNDE80 , OMFE4055 (refs 7,11)).

Subsequently, the separately ND-reconstituted efux components were mixed with native lipidated MFP (AcrA or MexA, Fig. 1b).

This two-step reconstitution protocol was successfully applied to MexABOprM and AcrABTolC. In the following sections, the reconstitution is detailed by rst characterizing OprM or MexB into ND, followed by the process of whole-tripartite assembly reconstitution. Moreover, reconstitution of cognate AcrABTolC and non-cognate AcrAMexBTolC complexes are presented highlighting the generic approach of our protocol. The ND-reconstituted native tripartite complexes were visualized by

OMF

a

MSP

MFP

Detergent

b

RND

Figure 1 | Tripartite assembly based on RND and OMF inserted into nanodiscs. (a) After detergent removal, the integral membrane proteins are reconstituted into a small lipid bilayer wrapped by two MSPs (purple) forming the nanodisc. Lipids are red/yellow and detergent is grey.(b) Self-assembly of RND (blue) and OMF (orange) in nanodiscs in the presence of native lipid-modied MFP (green) leading to the tripartite complex in lipid membrane.

2 NATURE COMMUNICATIONS | 7:10731 | DOI: 10.1038/ncomms10731 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms10731 ARTICLE

EM and single-particle analysis, resulting in elongated structures of 33 nm along their main axis.

OprM and MexB molecules inserted into NDs. The ND reconstitution of OprM was achieved with construct MSP1D1 and POPC lipids reported to form 10-nm-diameter sized NDs. Analysis of these OprMNDs by EM was done on negatively uranyl acetate-stained samples. At a MSP:lipid:OprM molar ratio of 1:36:0.4, OprMND mainly contained one OprM molecule per ND, with their long axis preferentially oriented parallel to the carbon support (Fig. 2a). EM observations were consistent with the trimeric assembly of OprM41. An average image (from 446 particles) of the OprMND revealed a 11 nm in diameter ND spanned by a duct formed by the OprM b-barrel domain (visible due to the presence of uranyl acetate within the b-barrel),

followed by the 10-nm-long OprM periplasmic domain including the equatorial domain (Fig. 2a inset). Note that MSP1E3D1 construct produced larger ND leading to the insertion of two OprM molecules (Supplementary Fig. 1).

The formation of MexBND was achieved using POPC and MSP1E3D1 as scaffold resulting in 12- to 14-nm-diameter sized NDs (that is, a diameter larger than the transmembrane domain of MexB). At a MSP:lipid:MexB molar ratio of 1:27:1, EM revealed side views of MexBND containing one molecule per ND (Fig. 2b) in accordance with the trimeric organization of MexB (Protein Data Bank entry: 2V50). Clearly visible is also the exposed periplasmic domain that protrudes 7 nm away from the lipid-containing ND. Averaging 341 single particles revealed a continuous layer of electron density of the ND, including the 36-transmembrane helix domain of trimeric MexB. The periplasmic part of MexB exhibited furthermore two clearly distinguishable layers of density, assigned to the porter domain and to the more distal funnel domain (Fig. 2b inset).

Formation of a tripartite complex. Tripartite complex formation was achieved by mixing OprMND, MexBND and lipidated MexA in a 1:1:10 molar ratio (Fig. 1b). Formation of tripartite complexes was visualized using native PAGE resulting in an electrophoretic mobility shift on complex formation (Fig. 3a). ND-reconstituted MexB and OprM migrated as a single band strongly stained by silver (Fig. 3a, lanes 1 and 2). MexBND migration was less than OprMND due to its larger size and hydrodynamic radius. Mixing of MexBND with OprMND in a 1:1 molar ratio yielded two separate stained bands (Fig. 3a, lane 3) with corresponding electrophoretic mobilities of the two single ND-reconstituted components (Fig. 3a, lanes 1 and 2). However, in a mixture containing MexBND, OprMND and lipidated MexA (39 kDa monomer; 1:1:10 molar ratio), a signicant upshifted band was observed after staining with silver (Fig. 3a, lane 6). This extra band was only observed when all three components were mixed. When lipidated MexA was mixed with either MexBND (Fig. 3a, lane 4) or with OprMND (Fig. 3a, lane 5), no upshift of bands could be observed. Our

a b

1

3

2

4

5

6

Figure 2 | TEM observations of OprM and MexB reconstituted into nanodiscs. (a) Field of view of OprMND showing side views of isolated molecules. The average image (inset) reveals characteristic features: The OprM b-barrel in the ND (1) and the OprM periplasmic domain composed of the equatorial domain (2) and the tip of the a-barrel (3) protruding from the ND. (b) Field of view of MexB-ND showing isolated molecules. Black arrows indicate side views. On the average image (inset), a side view of MexB exhibits the periplasmic part organized in two layers (funnel (4) and porter (5) domains) protruding from the ND (6). Scale bars, 50 nm and5 nm for the inset. a.u., arbitrary unit.

Complex

MexBND

OprMND

ND

ND

a b

MexBND

OprMND

OprMND+MexBND

MexBND+MexA

1,236

720 480

242

OprMND+MexBND+MexA

OprMND+MexA

Absorbance at A 280(a.u.)

MexABOprM(1.19 ml)

A10 A12 B2 B4 B6

Elution volume (ml)

146

66

c

250

150

100

75

50 37

25

20 15

MexB

OprM

MexA D1

E3D1

0 1 2 3 4 5 6

80

60

40

20

0

0.0 0.5 1.0 1.5

Figure 3 | Native PAGE analysis and purication of the tripartite MexABOprM assembly. (a) Electrophoretic mobility shift assay of individual and mixed components. OprMND and MexBND were mixed in the presence and in the absence of MexA. An extra band was observed when the three components were present in the sample. Lane 0, ND; lane 1, MexBND; lane 2, OprMND; lane 3, MexBND and OprMND; lane 4, MexBND and MexA; lane 5, OprMND and MexA; lane 6, MexBND, OprMND, and MexA. Proteins were separated by native PAGE and stained with PlusOne Silver Staining Kit. (b) Analytical size-exclusion chromatography (SEC) analysis of the mixed components. (c) SDSPAGE analysis of the indicated SEC peak fractions. The molecular mass of each marker protein (in kilodalton) is indicated on the right (a) and on the left (b).

NATURE COMMUNICATIONS | 7:10731 | DOI: 10.1038/ncomms10731 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms10731

interpretation is that MexBND and OprMND do not form bipartite complexes and only when the three components of the tripartite complex (that is, MexBND, OprMND and lipidated MexA) are present in the sample, a larger complex is formed.

To get structural details on the assembly of the efux pump, the (1:1:10) mixture of OprMND, MexBND and lipidated MexA was analysed by EM (Supplementary Fig. 2). Strikingly, within the population of single particles, ca. 10% were elongated structures of 33 nm at their longest expansion. Clearly different from isolated MexBND and OprMND (Fig. 2a,b), these new structures likely correspond to complex formation evidenced on the silver-stained native polyacrylamide gel (Fig. 3a, lane 6). To improve the yield of tripartite complex, the assembly formed with OprMND, MexBND and lipidated MexA in a 1:1:20 molar ratio was puried by size-exclusion chromatography (SEC) and analysis of the fractions by SDSPAGE (Fig. 3c) revealed the presence of the three partners, in particular, in the rst peak (Fig. 3b, fraction A12). Note that OprMND and MexBND were eluted in B4 and B3 fractions, respectively when applied alone on the same column (Supplementary Fig. 3). EM analysis of A12 fraction exhibited a vast majority of elongated structures viewed from their sides (Fig. 4a). The majority of class averages (125 over 200 classes) from single-particle average image analysis revealed an edice of protein densities at both ends resembling ND densities B23 nm apart (Fig. 4b). The upper part of the complex resembled OprMND with its central duct, including the ND-surrounded b-barrel domain. Adjacent to the ND density, a bulky knot is visible, which we interpret as the equatorial domain (Fig. 4b,c). At the other end of the elongated particle, we observe the recognizable features of MexB, that is, the ND-embedded transmembrane domain, and the protruding MexB porter and funnel domains (Fig. 4b,c). In between the OprMND and MexBND densities, additional densities are present, which we assign to connecting MexA molecules. Isocontours of OprMND and MexBND (derived from the average images of Fig. 2), overlaid on the isocontours of the putative tripartite complex average image, showed a good match of densities of ND-embedded OprM and MexB molecules (Fig. 4c). The 6-nm long, non-overlapping densities most likely corresponded to MexA molecules that are interacting with both OprM and MexB

(blue contours in Fig. 4c). Hence, in accordance with this analysis, it can be concluded that the tripartite MexBMexA OprM complex was successfully formed using NDs reconstitution. Comparison of the intermembrane distance (23 nm, Fig. 4c) with the known dimensions of the periplasmic domains of MexB (7 nm (ref. 15)) and OprM (10 nm (refs 11,12)) and the observed non-overlapping density of B6 nm suggest that the

RND (MexB) and OMF (OprM) components are not in direct contact. This observation is in accordance with the results recently obtained with genetically fused RNDMFP constructs36,38, supporting the role of MexA as a part of the periplasmic duct formed by the MexBMexAOprM complex, bridging the gap between MexB and OprM (Supplementary Fig. 4).

Interestingly, few class averages (about 400 particles) showed that the contact between MexA and OprM are apparently different. OprM appears to contact the MexAB complex over MexA via small links (black arrows Fig. 4b). These extremities arising from the a-barrel of OprM and engaged the interaction with MexA resemble those observed in images of isolated OprM, which in an isolated state is present in a closed a-barrel conformation11 (Fig. 2). The complexes showing the closed OprM engaged in a presumably looser manner with the MexAB complex may correspond to intermediate steps in the formation of the tripartite system, even though we cannot exclude that these particles represent dissociated complexes due to their interaction on the EM grid.

Tripartite complex of AcrABTolC in NDs. A similar reconstitution procedure was applied to the AcrBND, TolCND and lipidated AcrA. An average image of TolC side views (average over 444 images) revealed elongated protein densities protruding from the ND and forming a tunnel/duct similar to that observed for OprM (Supplementary Fig. 5a). Side views of AcrB molecules inserted into NDs (average over 294 images) showed two main layers of protein densities resembling those of MexB molecules, that is, ND/transmembrane, porter and funnel domains (Supplementary Fig. 5b). After mixing TolCND, AcrBND and AcrA in a 1:1:10 ratio, gel ltration was performed to purify the

a b c

5 nm 5 nm

23 nm

6 nm

1

2

3

4

5

6

Figure 4 | TEM analysis of tripartite MexABOprM assembly. (a) Field of view revealing elongated complexes when OprM-ND and MexB-ND were mixed in the presence of MexA. Scale bar, 30 nm. (b) Gallery of ve class average side views of a 33-nm-long tripartite MexABOprM complex delineated by two nanodiscs (157, 192, 185, 99, 151 images, respectively) and one top view class average (56 images). Lower row, two average classes of atypical complexes showing faint contacts between OprM and MexAB (black arrows). Scale bar, 10 nm. (c) Isocontours of MexBND and OprMND (red) overlaid on isocontours of tripartite complex (blue). Characteristic features are displayed: OprM b-barrel and ND (1); equatorial domain (2), tip of a barrel (30),

MexB funnel (40) and porter (50) domains anchored to ND (6). The remaining blue densities correspond to MexA that linked OprM to MexB and interacts with the domains marked with (0). The tripartite assembly ND (6) has a smaller size compared with ND of MexB probably because of the detergent carried with MexA that may extract some lipids.

4 NATURE COMMUNICATIONS | 7:10731 | DOI: 10.1038/ncomms10731 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms10731 ARTICLE

a

b

c d e

Figure 5 | TEM analysis of tripartite AcrABTolC and AcrAMexBTolC assemblies. (a) Field of view revealing puried AcrABTolC assemblies. Scale bar, 30 nm. (b) Gallery of one top view class average (137 images) and four class average side views of a 33-nm-long tripartite AcrABTolC complex (223, 119, 196, 178 images, respectively) and. (c) Enlarged average image of tripartite AcrABTolC assembly. (d) Isocontours of tripartite AcrABTolC assembly (red) overlaid on isocontours of tripartite MexAB OprM assembly (blue). (e) An average of tripartite AcrAMexBTolC assembly. Scale bar, 10 nm for c,d,e.

tripartite complex for further EM analysis (Fig. 5a and Supplementary Fig. 6). Average classes of side and top views showed that a complex was composed of TolCND (upper part), AcrBND (lower part) and connecting those, densities assigned to AcrA molecules were observed in the negative-stained averaged image (Fig. 5b,c). Interestingly, the intermembrane/ND distance was similar to that of MexABOprM tripartite complex, likewise suggesting that there is no direct contact between TolC and AcrB (Fig. 5d). Clearly visible from this comparison, the tripartite assemblies of AcrABTolC and of MexABOprM share similarities suggesting a common mechanism for tripartite assembly.

Hybrid tripartite complex of AcrAMexBTolC in NDs. Thus far, we used cognate components for our analysis of tripartite assembly. Since the image analysis suggested a common mechanism of tripartite formation and complex assembly, we also analysed mixtures of non-cognate TolCND, MexBND and AcrA as well as OprMND, AcrBND and MexA (1:1:10 ratio) by EM. Surprisingly, for the MexBAcrATolC mixture, we also encountered complexes albeit less frequent (236 complexes), with an overall appearance similar to the cognate tripartite complexes

(Fig. 5e and Supplementary Fig. 7). Even for the non-cognate mixture of OprMND, AcrBND and MexA tripartite complex particles were observed, but at a very low frequency (o1%). The formation of a lower amount of hybrid (non-cognate) complexes compared with genuine ones suggests that the components most likely exhibit much lower binding afnities.

DiscussionWe have devised a generic approach allowing the formation of tripartite RND efux pumps as characterized by native gel and EM analysis. By means of a ND toolkit, integral membrane components of two tripartite RND efux systems were incorporated into a lipid membrane ND and were further assembled after adding puried lipidated MFPs. Similar tripartite assemblies were obtained for the two different pumps from E. coli and fromP. aeruginosa, that is, AcrBAcrATolC and MexBMexA OprM, respectively. With an overall height of 33 nm, the tripartite complexes were composed of an OMF molecule and an RND molecule facing each other with their periplasmic domains without being in direct contact, linked by MFP molecules. These self-assembled tripartite complexes strongly resembled the AcrABZTolC edice obtained using protein fusion36 and suggests a 3:6:3 stoichiometric assembly of the components. In this report, the tripartite assemblies comprised native proteins and self-assembly of the three components occurred without any additional protein(s) or chemical cross-linking. Interestingly, both tripartite setups based on native (this work) or fused36,38 components indicated a common organization of the MFP connecting OMF and RND components at this level of detail (Supplementary Fig. 4). The three-dimensional reconstruction of MexABOprM (Supplementary Fig. 4a) at an estimated resolution of 25 (Supplementary Fig. 4b) displays the distinct features of the tripartite system (that is, OMF, MFP-hairpin, -lipoyl, -b-barrel and -membrane proximal domains, in addition to the RND transporter), which is in good agreement with the cryo-EM structure by Du et al.36 (Supplementary Fig. 4c). This organizational resemblance also suggests that our tripartite pump systems were assembled with similar stoichiometry, that is, 3:6:3 as has been previously proposed for other tripartite pumps as well32,42. It is worth noting that by mixing MexBND and OprMND, complex formation was not observed on native gel or on EM grid suggesting that OprM does not strongly bind to MexB in solution as was previously observed by isothermal titration calorimetry for the homologous partners TolC and AcrB31.

Our results also provided evidence that RND tripartite systems were able to self-assemble in a mixture of non-cognate components. When MexB was interchanged with AcrB in the self-assembling process, tripartite AcrAMexBTolC complexes were observed but with a lower occurrence compared with the formation of cognate complexes. The formation of hybrid tripartite complexes has a signicant meaning since it has been observed that AcrAMexBTolC hybrid system could confer resistance to E. coli for only a limited subset of drugs normally transported by the cognate efux systems43. The fact that we observed only few hybrid AcrAMexBTolC complexes may also explain the partial resistance observed in an E. coli DacrB background complemented with heterologous expressed mexB. MexA or AcrA are known for their high exibility most likely facilitating tripartite formation44. The spectrum of antibiotic resistance of the non-cognate efux systems could be extended by side chain substitutions in the helical hairpin of AcrA, which interacts with TolC or by substitutions in the AcrA b-barrel domain interacting with MexB43. The use of variants of AcrA43, TolC45 or MexB46 may provide a better t with the two other

NATURE COMMUNICATIONS | 7:10731 | DOI: 10.1038/ncomms10731 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms10731

components. Since similar tripartite assemblies were formed with cognate and non-cognate components, as schematically shown in Fig. 5e, we suggest that there might be a general mechanism triggering tripartite assembly, the formation efciency being dependent on the binding afnities of the MFPs, therefore modulating drug efux.

As discussed above, the MFP component possesses a high adaptability in the maintenance of interactions between the OMF and RND components. It has been proposed that during drug export, conformational changes of the MFP component, triggered by substrate and/or H binding and/or release to/from the RND component could catalyse the assembly/disassembly of the tripartite complex47. The role of the MFP as a central element with conformational change capability might indicate that the tripartite assembly could be a multi-stage process. The observed assemblies in this study and in Du and Luisis work36 might represent the nal stage formation of a fully functional tripartite setup. However, we cannot exclude that the observed setup might be an intermediate formation that, under in vivo conditions, is subsequently subjected to a yet unknown conformational step leading towards fully functional tripartite complex, where the OMF and RND components are in direct contact. This might explain the previously reported direct cysteine disulde cross-linking between AcrB and TolC34 and would possibly indicate an energy requirement (for example, substrate-binding energy and/or proton motive force) for functional tripartite complex formation.

The reconstitution of the tripartite assembly opens new perspectives for the development of efux pump inhibitors (EPIs), precluding antibiotic efux and resulting in an increase of intracellular antibiotic concentration. Efforts have been concentrated up till now on the search for compounds that could compete for antibiotic-binding sites and/or block drug translocation48,49. Nevertheless, such inhibitors are particularly difcult to identify owing to the broad variety of substrates that the multidrug pumps can accommodate. Another strategy would be to target the assembly of the tripartite system. The generation of a leak in the channel by preventing the protein/protein recognitions or by hampering their structural adaptability could also affect the pump efciency. Compounds having such effect would represent a class of EPIs acting on the assembly process of efux pumps, instead of acting on drug translocation. Interestingly, unlike RND proteins, OMFs have often been observed to be shared among different tripartite systems, independent of the nature of the inner membrane protein or periplasmic adaptor component50. In the case of P. aeruginosa, OprM can operate with MexAB, MexXY51 and MexMN52. An inhibitor compound precluding OprMMexA interaction could therefore also inhibit the activity of the other pumps.

Our protocol opens the way towards a better understanding of the tripartite assembly of AcrABTolC and MexABOprM. RND and OMF components inserted into NDs are able to self-assemble in a tripartite complex in the presence of a lipidated MFP component. We provide evidence for a common mechanism of tripartite assembly with cognate and non-cognate components, whereby MFP molecules link their cognate OMF partners to cognate and non-cognate RND components. From the single-particle analysis, there is clear evidence that the RND and OMF components are not in direct contact within the tripartite assembly. This approach can be extended to a vast number of RND pump systems and open the eld for further structural analysis at atomic level. In addition, the development of an EPI class targeting the tripartite assembly process can be approached more systematically. Although our protocol is not suitable for a large screening of compounds, it is a proof of concept for inhibitor compound analysis.

Methods

Materials and reagents. POPC was purchased from Avanti Polar Lipids (USA), sodium cholate hydrate, octyl-poly-oxyethylene, n-octyl-b-D-glucopyranoside (OG) and n-Dodecyl b-D-maltoside (DDM) were purchased from Sigma. SM2 Bio-beads and 415% Mini-PROTEAN TGX gel were obtained from Bio-Rad. Superose 6 3.2/300 column and PlusOne Silver Staining Kit were purchased from GE Healthcare. Cu 300 mesh grids were obtained from Agar Scientic.

Lipid preparation. POPC lipids were dissolved in methanol/chloroform (v/v), dried onto a glass tube under steady ow of nitrogen and followed by exposure to vacuum for 1 h. The lipid lm was suspended in the reconstitution buffer (100 mM NaCl, 10 mM Tris/Cl, pH 7.4) and subjected to ve rounds of sonication for 30 s each. Lipid concentration was quantied by phosphate analysis.

Protein preparation. Two MSPs, MSP1D1 and MSP1E3D1 (genetic constructs available from AddGene) expressed and puried from bacteria40, were used to make OprM/TolC and MexB/AcrB NDs, respectively. The acrA and tolC genes were individually cloned into pET24a after amplication with primers acrA_forward 50-GATTCGGGGCCCAACAAAAACAGAGGGTTTACG-30, acrA_reverse 50-ATAATAGGATCCTTAAGACTTGGACTGTTCAGGCTG-30, tolC_forward 50-AAAACATATGAAGAAATTGCTCCCC-30 and tolC_reverse 50-AAAACTCGAGGTGGTGGTGGTGGTTACGGAAAGGGTTATGACCGTT-30 containing NdeI and XhoI restriction sites, analogous to the cloning of acrB as described previously53. AcrA, AcrB and TolC were separately produced in E. coli C43(DE3)DacrAB and subsequently puried as previously described for AcrB53. In brief, cells were grown in TB medium at 37 C at 180 r.p.m. till OD600 of 1.3 before induction with 0.5 mM IPTG. After induction, the cells were grown overnight at 20 C at 180 r.p.m. After harvest, cells were suspended at 5 g cells wet weight per ml in lysis buffer (Tris/Cl 20 mM pH 7.5, NaCl 500 mM, MgCl2 2 mM, 100 mM

PefaBloc (Sigma), and trace amounts of DNAse I and lysozyme). Cells were lysed by three passages through a Constant Systems cell disruptor at 22 kpsi and centrifuged for 30 min at 18,000g at 4 C. The supernatant was subsequently centrifuged for 1 h at 180,000g at 4 C to collect the membranes. Membranes were solubilized (10 ml per g wet weight membranes) in lysis buffer containing 10 mM imidazole, pH 7.5, and 2% DDM for 1 h at 4 C while gently stirring. Samples were centrifuged for 1 h at 180,000g and the supernatant was subjected to Ni-NTA afnity chromatography. The proteins were eluted in a 10 ml buffer (Tris/Cl20 mM, pH 7.5, NaCl 150 mM, imidazole 200 mM pH 8.0, and 0.03% DDM). AcrA and AcrB were then subjected to SEC using a Superpose 6 HR column (GE Healthcare) with Tris/Cl 20 mM pH 7.5, NaCl 150 mM and 0.05% DDM as running buffer. The TolC sample was passed through a desalting column (NAP-10, GE Healthcare) using Tris/Cl 20 mM pH 7.5, NaCl 150 mM and 1.5% OG as buffer.

MexB, MexA and OprM membrane proteins were separately produced in E. coli C43DacrB and puried as described for MexB54, MexA13 and OprM12,55. In brief, cells were grown in TY medium at 37 C at 200 r.p.m. till OD600 of 0.6 before induction with 1 mg ml 1 IPTG. After induction, the cells were grown overnight at 20 C at 200 r.p.m. After harvest, cells were suspended in lysis buffer (Tris/Cl20 mM pH 8, NaCl 200 mM). Cells were lysed by two passages through a Constant Systems cell disruptor at 35 kpsi and centrifuged at 10,000g for 25 min at 4 C. For MexB and MexA, the supernatants were submitted to an ultracentrifugation at 100,000g for 1 h at 4 C. For OprM, the supernatant is solubilized with 2% n-octylpoly-oxyethylene before ultracentrifugation to get rid of the inner membranes. Membranes were subsequently submitted to an overnight detergent solubilization in Bis-Tris 10 mM pH 7.4, glycerol 20%, imidazole 10 mM and NaCl 500 mM and at a 2:1 and 40:1 (w/w) detergent-to-protein ratio for MexB and MexA/OprM, respectively (protein concentration is determined using the Bicinchoninic acid test from Sigma). Samples were subjected to an ultracentrifugation at 100,000g for1 h at 4 C. Protein purications were performed by afnity chromatography on a HisTrap HP column followed by a gel ltration (superose 6 HR, 10/300 GE Healthcare). After purication, protein buffers contained 0.03% DDM for MexA and MexB and 0.9% OG for OprM.

Preparation of MexB/AcrB and OprM/TolC in NDs. MexB and OprM were inserted into NDs according to the standard protocol39,40. Briey, to obtain MexBND, MexB solution was mixed with POPC solution and MSP solution at a nal 27:1:1 lipids/MSP/protein molar ratio in a 10 mM Tris/Cl, pH 7.4, 100 mM NaCl with 0.009% DDM and 15 mM Na-cholate solution. Detergent was removed by the addition of SM2 Bio-beads into the mixture shaken overnight at 4 C. For OprMND, OprM solution was mixed with POPC solution and MSP solution at a nal 36:1:0.4 lipids/MSP/protein molar ratio in a 10 mM Tris/Cl, pH 7.4, 100 mM NaCl, 30 mM OG and 15 mM Na-cholate solution. Like for MexBND, detergent was removed with SM2 Bio-beads overnight at 4 C. The same protocol was used for AcrB and TolC proteins.

Purication of the tripartite complex. To enrich the tripartite complexes from the initial mixed samples, we subjected the MexABOprM (or AcrABTolC) containing samples to SEC using a Superose 6 column pre-equilibrated with reconstitution buffer, which was also used as running buffer. Peak fractions were

6 NATURE COMMUNICATIONS | 7:10731 | DOI: 10.1038/ncomms10731 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms10731 ARTICLE

collected and subjected to SDSPAGE analysis. The resulting gel was stained with PlusOne Silver Staining Kit.

Native gel analysis. For native gel electrophoresis, samples (0.5 to 2 ml) were separated on 415% continuous gradient polyacrylamide Mini-PROTEAN TGX gel in sample buffer (500 mM Tris/Cl pH 6.8, 30% glycerol, 0.05% bromophenol blue). Electrophoresis was performed at constant voltage of 150 V for 90 min, in Tris-Glycine running buffer. The gel was stained with PlusOne Silver Staining Kit.

Electron microscopy acquisition and analysis. For EM grid preparations, a diluted mixture of the samples suspension was applied to a glow-discharged carbon-coated copper 300 mesh grids and stained with 2% uranyl acetate (w/v) solution. Images were recorded under low-dose conditions on electron microscope (Tecnai 12 and F20 FEG, FEI) using a FEI Eagle 4k 4k and a Gatan USC1000

2k 2k cameras. Image alignment and two-dimensional averages were performed

with SPIDER using a reference-free alignment procedure. For MexABOprM and for AcrABTolC, a total of 22,979 and 28,863 particles, respectively, were automatically selected and processed for class averaging.

For assessing the occurrence of complex formation, OprMND, MexBND and lipidated MexA were mixed at a ratio of 1:1:10, and sampled after 1, 6, 12, 24, 48, 96 and 168 h (1 week) and from there on every 168 h up to 6 weeks. For each sample, two negatively stained grids were prepared and subjected to transmission electron microscope. Twenty-ve micrographs were randomly collected per grid (total 50 micrographs per sample). The occurrence of the complexes was expressed as a ratio of complexes present on the grid corresponding to the number of complexes determined by a manual picking divided by the number of objects (ND alone; MexBND, OprMND and complexes) calculated by automatic picking based on the nding of maximum values in a Difference of Gaussian ltered micrograph.

References

1. May, M. Drug development: time for teamwork. Nature 509, S4S5 (2014).2. Li, X.-Z., Plsiat, P. & Nikaido, H. The challenge of efux-mediated antibiotic resistance in Gram-negative bacteria. Clin. Microbiol. Rev. 28, 337418 (2015).

3. Nikaido, H. Multidrug resistance in bacteria. Annu. Rev. Biochem. 78, 119146 (2009).

4. Koronakis, V., Sharff, A., Koronakis, E., Luisi, B. & Hughes, C. Crystal structure of the bacterial membrane protein TolC central to multidrug efux and protein export. Nature 405, 914919 (2000).

5. Mikolosko, J., Bobyk, K., Zgurskaya, H. I. & Ghosh, P. Conformational exibility in the multidrug efux system protein AcrA. Structure 14, 577587 (2006).

6. Eicher, T. et al. Transport of drugs by the multidrug transporter AcrB involves an access and a deep binding pocket that are separated by a switch-loop. Proc. Natl Acad. Sci. USA 109, 56875692 (2012).

7. Murakami, S., Nakashima, R., Yamashita, E. & Yamaguchi, A. Crystal structure of bacterial multidrug efux transporter AcrB. Nature 419, 587593 (2002).

8. Murakami, S., Nakashima, R., Yamashita, E., Matsumoto, T. & Yamaguchi, A. Crystal structures of a multidrug transporter reveal a functionally rotating mechanism. Nature 443, 173179 (2006).

9. Seeger, M. A. et al. Structural asymmetry of AcrB trimer suggests a peristaltic pump mechanism. Science 313, 12951298 (2006).

10. Sennhauser, G., Amstutz, P., Briand, C., Storchenegger, O. & Grtter, M. G. Drug export pathway of multidrug exporter AcrB revealed by DARPin inhibitors. PLoS Biol. 5, e7 (2007).

11. Akama, H. et al. Crystal structure of the drug discharge outer membrane protein, OprM, of Pseudomonas aeruginosa: dual modes of membrane anchoring and occluded cavity end. J. Biol. Chem. 279, 5281652819 (2004).

12. Phan, G. et al. Structural and dynamical insights into the opening mechanism of P. aeruginosa OprM channel. Structure 18, 507517 (2010).

13. Akama, H. et al. Crystal structure of the membrane fusion protein, MexA, of the multidrug transporter in Pseudomonas aeruginosa. J. Biol. Chem. 279, 2593925942 (2004).

14. Higgins, M. K., Bokma, E., Koronakis, E., Hughes, C. & Koronakis, V. Structure of the periplasmic component of a bacterial drug efux pump. Proc. Natl Acad. Sci. USA 101, 99949999 (2004).

15. Sennhauser, G., Bukowska, M. A., Briand, C. & Grtter, M. G. Crystal structure of the multidrug exporter MexB from Pseudomonas aeruginosa. J. Mol. Biol. 389, 134145 (2009).

16. Symmons, M. F., Bokma, E., Koronakis, E., Hughes, C. & Koronakis, V. The assembled structure of a complete tripartite bacterial multidrug efux pump. Proc. Natl Acad. Sci. USA 106, 71737178 (2009).

17. Eicher, T. et al. Coupling of remote alternating-access transport mechanisms for protons and substrates in the multidrug efux pump AcrB. eLife 3, e03145 (2014).

18. Fernandez-Recio, J. et al. A model of a transmembrane drug-efux pump from Gram-negative bacteria. FEBS Lett. 578, 59 (2004).

19. Husain, F., Humbard, M. & Misra, R. Interaction between the TolC and AcrA proteins of a multidrug efux system of Escherichia coli. J. Bacteriol. 186, 85338536 (2004).

20. Pos, K. M. Drug transport mechanism of the AcrB efux pump. Biochim. Biophys. Acta 1794, 782793 (2009).

21. Tal, N. & Schuldiner, S. A coordinated network of transporters with overlapping specicities provides a robust survival strategy. Proc. Natl Acad. Sci. USA 106, 90519056 (2009).

22. Ruggerone, P., Murakami, S., Pos, K. M. & Vargiu, A. V. RND efux pumps: structural information translated into function and inhibition mechanisms. Curr. Top. Med. Chem. 13, 30793100 (2013).

23. Ruggerone, P., Vargiu, A. V., Collu, F., Fischer, N. & Kandt, C. Molecular dynamics computer simulations of multidrug RND efux pumps. Comput. Struct. Biotechnol. J. 5, e201302008 (2013).

24. Schulz, R., Vargiu, A. V., Collu, F., Kleinekathfer, U. & Ruggerone, P. Functional rotation of the transporter AcrB: insights into drug extrusion from simulations. PLoS Comput. Biol. 6, e1000806 (2010).

25. Lomovskaya, O. et al. Identication and characterization of inhibitors of multidrug resistance efux pumps in Pseudomonas aeruginosa: novelagents for combination therapy. Antimicrob. Agents Chemother. 45, 105116 (2001).

26. Nakashima, R. et al. Structural basis for the inhibition of bacterial multidrug exporters. Nature 500, 102106 (2013).

27. Opperman, T. J. et al. Characterization of a novel pyranopyridine inhibitor of the AcrAB efux pump of Escherichia coli. Antimicrob. Agents Chemother. 58, 722733 (2014).

28. Vargiu, A. V., Ruggerone, P., Opperman, T. J., Nguyen, S. T. & Nikaido, H. Molecular mechanism of MBX2319 inhibition of Escherichia coli AcrB multidrug efux pump and comparison with other inhibitors. Antimicrob. Agents Chemother. 58, 62246234 (2014).

29. Tikhonova, E. B., Dastidar, V., Rybenkov, V. V. & Zgurskaya, H. I. Kinetic control of TolC recruitment by multidrug efux complexes. Proc. Natl Acad. Sci. USA 106, 1641616421 (2009).

30. Tikhonova, E. B., Yamada, Y. & Zgurskaya, H. I. Sequential mechanism of assembly of multidrug efux pump AcrAB-TolC. Chem. Biol. 18, 454463 (2011).

31. Touz, T. et al. Interactions underlying assembly of the Escherichia coli AcrAB-TolC multidrug efux system. Mol. Microbiol. 53, 697706 (2004).

32. Su, C.-C. et al. Crystal structure of the CusBA heavy-metal efux complex of Escherichia coli. Nature 470, 558562 (2011).

33. Trpout, S. et al. Structure of reconstituted bacterial membrane efux pump by cryo-electron tomography. Biochim. Biophys. Acta 1798, 19531960 (2010).

34. Tamura, N., Murakami, S., Oyama, Y., Ishiguro, M. & Yamaguchi, A. Direct interaction of multidrug efux transporter AcrB and outer membrane channel TolC detected via site-directed disulde cross-linking. Biochemistry 44, 1111511121 (2005).

35. Weeks, J. W., Celaya-Kolb, T., Pecora, S. & Misra, R. AcrA suppressor alterations reverse the drug hypersensitivity phenotype of a TolC mutant by inducing TolC aperture opening. Mol. Microbiol. 75, 14681483 (2010).

36. Du, D. et al. Structure of the AcrAB-TolC multidrug efux pump. Nature 509, 512515 (2014).

37. Hobbs, E. C., Yin, X., Paul, B. J., Astarita, J. L. & Storz, G. Conserved small protein associates with the multidrug efux pump AcrB and differentially affects antibiotic resistance. Proc. Natl Acad. Sci. USA 109, 1669616701 (2012).

38. Kim, J.-S. et al. Structure of the tripartite multidrug efux pump AcrAB-TolC suggests an alternative assembly mode. Mol. Cells 38, 180186 (2015).

39. Denisov, I. G., Grinkova, Y. V., Lazarides, A. A. & Sligar, S. G. Directed self-assembly of monodisperse phospholipid bilayer Nanodiscs with controlled size.J. Am. Chem. Soc. 126, 34773487 (2004).40. Ritchie, T. K. et al. Chapter 11 - Reconstitution of membrane proteins in phospholipid bilayer nanodiscs. Methods Enzymol. 464, 211231 (2009).

41. Lambert, O. et al. Trimeric structure of OprN and OprM efux proteins from Pseudomonas aeruginosa, by 2D electron crystallography. J. Struct. Biol. 150, 5057 (2005).

42. Janganan, T. K. et al. Evidence for the assembly of a bacterial tripartite multidrug pump with a stoichiometry of 3:6:3. J. Biol. Chem. 286, 2690026912 (2011).

43. Krishnamoorthy, G., Tikhonova, E. B. & Zgurskaya, H. I. Fitting periplasmic membrane fusion proteins to inner membrane transporters: mutations that enable Escherichia coli AcrA to function with Pseudomonas aeruginosa MexB.J. Bacteriol. 190, 691698 (2008).44. Vaccaro, L., Koronakis, V. & Sansom, M. S. P. Flexibility in a drug transport accessory protein: molecular dynamics simulations of MexA. Biophys. J. 91, 558564 (2006).

45. Bokma, E., Koronakis, E., Lobedanz, S., Hughes, C. & Koronakis, V. Directed evolution of a bacterial efux pump: adaptation of the E. coli TolC exit duct to the Pseudomonas MexAB translocase. FEBS Lett. 580, 53395343 (2006).

NATURE COMMUNICATIONS | 7:10731 | DOI: 10.1038/ncomms10731 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms10731

46. Tikhonova, E. B., Wang, Q. & Zgurskaya, H. I. Chimeric analysis of the multicomponent multidrug efux transporters from gram-negative bacteria.J. Bacteriol. 184, 64996507 (2002).47. Janganan, T. K., Bavro, V. N., Zhang, L., Borges-Walmsley, M. I. & Walmsley,A. R. Tripartite efux pumps: energy is required for dissociation, but not assembly or opening of the outer membrane channel of the pump. Mol. Microbiol. 88, 590602 (2013).48. Lomovskaya, O., Zgurskaya, H. I., Totrov, M. & Watkins, W. J. Waltzing transporters and the dance macabre between humans and bacteria. Nat. Rev. Drug Discov. 6, 5665 (2007).

49. Nikaido, H. & Pags, J.-M. Broad-specicity efux pumps and their role in multidrug resistance of Gram-negative bacteria. FEMS Microbiol. Rev. 36, 340363 (2012).

50. Yum, S. et al. Crystal structure of the periplasmic component of a tripartite macrolide-specic efux pump. J. Mol. Biol. 387, 12861297 (2009).

51. Morita, Y., Tomida, J. & Kawamura, Y. MexXY multidrug efux system of Pseudomonas aeruginosa. Front. Microbiol. 3, 408 (2012).

52. Mima, T., Sekiya, H., Mizushima, T., Kuroda, T. & Tsuchiya, T. Gene cloning and properties of the RND-type multidrug efux pumps MexPQ-OpmE and MexMN-OprM from Pseudomonas aeruginosa. Microbiol. Immunol. 49, 9991002 (2005).

53. Pos, K. M. & Diederichs, K. Purication, crystallization and preliminary diffraction studies of AcrB, an inner-membrane multi-drug efux protein. Acta Crystallogr. D Biol. Crystallogr 58, 18651867 (2002).

54. Mokhonov, V. et al. Multidrug transporter MexB of Pseudomonas aeruginosa: overexpression, purication, and initial structural characterization. Protein Expr. Purif. 40, 91100 (2005).

55. Ferrandez, Y. et al. Amphipol-mediated screening of molecular orthoses specic for membrane protein targets. J. Membr. Biol. 247, 925940 (2014).

Acknowledgements

This work was supported by French national agency ANR (Assembly ANR-12BSV8-0010-03) and Conseil Rgional dAquitaine (20121301007) grants. We thank A. Luciani,T. Dayris C. Garnier and M. Dezi for technical assistance. We are grateful to the staff at Bordeaux imaging center for access facility. We thank Cordouan Technologies forthe use of Elmo. K.M.P. and R.K.R.M. are supported by the German Research Foundation (SFB 807, Transport and Communication across Biological Membranes), the

DFG-EXC115 (Cluster of Excellence FrankfurtMacromolecular Complexes) and a research grant from the Human Frontier Science Program (HFSP-RGP004/2013). L.M. was supported by a grant from Vaincre la Mucoviscidose. A.V. was supported by a grant from the Rgion Ile-de-France (DIM-Malinf 110058). The research of R.K.R.M and K.M.P was conducted as part of the Translocation Consortium (http://www.translocation.eu

Web End =http://www.transloca http://www.translocation.eu

Web End =tion.eu ) and has received support from the Innovative Medicines Joint Undertaking under Grant Agreement no.115525, resources which are composed of nancial contribution from the European Unions Seventh Framework Program (FP7/2007-2013) and EFPIA companies in kind contribution.

Author contributions

L.D., J-C.T. and O.L. designed and interpreted the experiments with I.B., M.P and K.M.P. A.V., L.M., M.P., I.B. and R.K.R.M. produced and puried membrane proteins. L.D. and F.O. produced proteins in nanodisc and performed EM characterizations. C.G. and L.D. puried and characterized the complexes. L.D. carried out native gel analysis. J-C.T. performed EM analysis. O.L. and K.M.P. wrote the manuscript with input from all authors.

Additional information

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: The authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/

Web End =reprintsandpermissions/

How to cite this article: Daury, L. et al. Tripartite assembly of RND multidrug efux pumps. Nat. Commun. 7:10731 doi: 10.1038/ncomms10731 (2016).

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Web End =http://creativecommons.org/licenses/by/4.0/

8 NATURE COMMUNICATIONS | 7:10731 | DOI: 10.1038/ncomms10731 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications