Inter-domain dynamics in the chaperone SurA and multi-site binding to its unfolded outer membrane protein clients

The periplasmic chaperone SurA plays a key role in outer membrane protein (OMP) biogenesis. E. coli SurA comprises a core domain and two peptidylprolyl isomerase domains (P1 and P2), but how it binds its OMP clients and the mechanism(s) of its chaperone action remain unclear. Here, we have used chemical cross-linking, hydrogen-deuterium exchange, single-molecule FRET and molecular dynamics simulations to map the client binding site(s) on SurA and to interrogate the role of conformational dynamics of the chaperone’s domains in OMP recognition. We demonstrate that SurA samples a broad array of conformations in solution in which P2 primarily lies closer to the core/P1 domains than suggested by its crystal structure. Multiple binding sites for OMPs are located primarily in the core domain, with binding of the unfolded OMP resulting in conformational changes between the core/P1 domains. Together, the results portray a model in which unfolded OMP substrates bind in a cradle formed between the SurA domains, with structural flexibility between its domains assisting OMP recognition, binding and release.


Introduction
Chaperones play vital roles in multicomponent proteostasis networks, ensuring that proteins fold and avoid aggregation in the crowded cellular milieu, and that misfolded proteins which cannot be rescued by chaperones are targeted for degradation 1,2 . It is now established that many chaperones are in rapid dynamic exchange between co-populated conformations, and that this conformational plasticity is key to their functional mechanisms 3 . In the case of ATPdependent chaperones, e.g. the Hsp60 chaperonins GroEL and TRiC, and the Hsp90 and Hsp70 families, ATP binding and/or hydrolysis promotes conformational changes that facilitate the folding and/or release of their clients 1,2,4,5,6,7,8 . However, some chaperones are not dependent on energy from nucleotide binding/hydrolysis, and instead their intrinsic structural flexibility is proposed to be key to their function 3,9,10,11,12 . The functional mechanisms of these ATP-independent chaperones, including how they bind and release their substrates in a controlled manner, is generally not well understood.
SurA is an ATP-independent chaperone involved in the biogenesis of outer membrane proteins (OMPs) in the periplasm of Gram-negative bacteria 13 -18 . This protein is thought to be the major chaperone responsible for protecting OMPs from aggregation in the periplasm 13 -18 and facilitating OMP delivery to the β-barrel assembly machinery (BAM) for folding and insertion into the outer membrane (OM) 13,14,19,20,21 . Deletion of SurA leads to OMP assembly defects, the induction of stress responses, and increased sensitivity to antibiotics and detergents 15,17,22,23,24 . Further, ∆surA strains show reduced assembly of virulence factors, such as pili and adhesins, and exhibit reduced pathogenicity in a number of species 22,25,26 . E. coli SurA has a three domain architecture, consisting of a core domain which is composed of its N-and C-terminal regions, and two parvulin-like peptidylprolyl isomerase (PPIase) domains (P1 and P2) (Fig. 1a) 27 . However, despite the availability of its crystal structure [27], how SurA binds its unfolded OMP clients and the molecular mechanism(s) of SurA function remain unknown. A substrate binding crevice was proposed based on examination of molecular packing interactions in crystals of SurA (Fig. 1b), but the location of OMP binding regions and the roles of the PPIase domains (which are not essential for in vivo or in vitro function, at least for some clients 24,28,29 ) in folding and binding its varied OMP clients remained unknown.
In the crystal structure of full-length SurA 27 , an extended conformation is observed in which the core and P1 domains are in contact (Fig. 1b, Supplementary Fig. 1a-c), while P2 is separated from this globular region via a linker (Fig. 1b, Supplementary Fig. 1d).
Examination of the molecular packing interactions in the crystal lattice revealed multiple contacts between all three domains and neighbouring molecules, which may stabilise the elongated architecture observed (Supplementary Fig. 2). SurA homologues and domain deletion variants have been crystallised in conformations with a variety of domain orientations (Supplementary Fig. 3) suggesting that SurA may have a dynamic structure.
Further, tethering of the P1 and core domains via a disulfide bond resulted in impaired OMP assembly in vivo 29 . However, the precise nature of these conformational dynamics and how they are linked to OMP binding have remained elusive.
Here, we sought to determine the conformational properties of full-length E. coli SurA in solution in an effort to better understand its conformational dynamics and how inter-domain motions may be exploited or modified by client binding. Combining mass spectrometric (MS) methods (chemical crosslinking (XL) and hydrogen-deuterium exchange (HDX)), with singlemolecule FRET (smFRET) and molecular dynamics (MD) simulations we show that SurA adopts conformations in solution that differ substantially from its crystal structure 27 .
Specifically, the P1 domain samples open and closed states relative to the core domain, and P2 is primarily located closer to the core/P1 domains than observed in the crystal structure.
We also show that multiple sites on the SurA surface, predominantly located in the core domain, are involved in client binding. OMPs bind to these specific sites in different orientations, consistent with a dynamically bound state, and the conformations adopted by the chaperone alter in response to OMP binding. Combined, our results portray a model in which the three domains of SurA form a cradle around its OMP clients, protecting them from misfolding and aggregation on their journey through the periplasm, with the conformational dynamics of the domains presumably facilitating their delivery to BAM for folding into the outer membrane.

Inter-domain conformational flexibility in SurA
We first investigated the structure and dynamics of apo-SurA in solution using XL-MS, which provides distance information in the form of spatial restraints, and enables comparison of the solution conformation(s) of the protein with structural data 30 . For this purpose, we used the bifunctional reagent disuccinimidyl dibutyric urea (DSBU), which primarily crosslinks Lys residues 31 . DSBU has been shown to crosslink residues within a straight line distance (SLD) between their Cα atoms of ca. 27-30 Å 32 . More recently it has been shown that considering the solvent accessible surface distance (SASD) between residues may more reliably predict structural models 33 (a Cα− Cα distance between crosslinked residues of up to ca. 35 Å is considered feasible for DSBU). For monomeric SurA, a total of 13 intra-domain (core-core, P1-P1 and P2-P2) (Supplementary Fig. 4 Fig. 1d). By contrast, only four (K105-K278, K278-K394, K251-K278, K252-278) of the 19 inter-domain crosslinks (core-P1, core-P2 or P1-P2) are compatible with the SurA crystal structure based on the SASD (Fig. 2b- 34 . However, we noted that the vectors of the SLDs for these crosslinks pass directly through the P1 and core domains, highlighting the importance of considering SASDs in judging crosslink violation 33,35 . Taken together, the data show that SurA populates structures in solution in which P2 is closer to both the core and P1 domains than portrayed by its crystal structure, as well as conformations in which the orientation and/or distance of P1 relative to the core is distinct from that observed in the crystal structure of the protein 27 . As an independent validation of the crosslinking results, we investigated the inter-domain distances in SurA by smFRET 36 . We selected non-conserved residues in the core, P1 and P2 domains (Q85, N193, and E301, respectively) to substitute with Cys, and constructed three variants containing Cys substitutions at two positions (core-P1, core-P2 and P1-P2) ( Fig. 3a-c, Supplementary Table 2). Each SurA variant was stochastically labelled with Alexa 488 and Alexa 594 dyes (R 0 = 60 Å), enabling inter-domain distance distributions to be monitored in a pairwise fashion. Samples containing ~50 pM of labelled SurA were interrogated using confocal fluorescence detection and alternating laser excitation (ALEX) (see Methods). Fluorescence intensities in the donor and acceptor channels yielded FRET efficiencies (E FRET ) for the passage of each single molecule through the confocal volume (fluorescence burst). These were collated into FRET efficiency histograms and compared with distributions predicted for each labelled SurA double Cys variant calculated from the SurA crystal structure (see Methods) ( Fig. 3d-f) 37,38 .
The predicted E FRET distribution for the core-P1 SurA variant calculated from the crystal structure of E. coli SurA 27 has a single maximum at ~0.6 ( Fig. 3d). In marked contrast with this prediction, the measured E FRET distribution of the core-P1 labelled SurA is (at least) bimodal, with one population centred on E FRET ~0.6, and a second, smaller, population centred on E FRET ~0.2 (Fig. 3d). This suggests that SurA populates at least two distinct conformations in solution, one (~60 %) in which P1 is located close to the core domain with an inter-domain distance similar to that in the crystal structure (core-P1 closed ), and one (~40 %) in which the P1 and core domains are further apart (core-P1 open ). The observed E FRET distributions for the labelled core-P2 and P1-P2 variants have maxima at ~0.4 and ~0.3, respectively (Fig. 3e,f), both in marked contrast with the low predicted values in the crystal structure (~0.1 and ~0.02, a spatial separation of ~85 Å and ~115 Å, for core-P2 and P1-P2, respectively). This indicates that, in the vast majority of molecules, P2 is located closer to the core and P1 domains than suggested by the SurA crystal structure 27 , consistent with the XL-MS data (Fig. 2b,c,d). Burst variance analysis (BVA) 39 showed that inter-domain motions involving each pair of domains occurs on the timescale of diffusion through the confocal volume (<1 ms) (Supplementary Fig. 5a-c). Together, these data indicate a dynamic chaperone structure in which sub-ms motions involving all three domains are occurring, in particular at the core-P1 interface which interconverts between core-P1 closed and core-P1 open states. In addition, the data show that P2 spends most of its time closer to the core and P1 domains than suggested by the SurA crystal structure. Given that this simulated annealing approach will drive SurA to adopt compact states that satisfy the maximum number of restraints within a single structure, more extended states of SurA that are significantly populated in solution, as shown by the smFRET data (Fig. 3), will not be captured by this method. Indeed, as shown by the smFRET and unrestrained MD simulations, the dynamic nature of SurA makes it challenging to define its precise conformational landscape, wherein a broad repertoire of conformations in dynamic exchange on a msec exchange are formed. Notably, no single structure can possibly satisfy the broad distributions observed by smFRET (Supplementary Fig. 7, Supplementary Table 6), providing further evidence that the crosslinks observed cannot all result from a single SurA conformation, but result from different rapidly interconverting states. Together, the unrestrained all-atom MD and simulated annealing simulations demonstrate that the three domains of SurA are able to move independently of each other as rigid bodies, facilitated by the flexible linker regions between them (Supplementary Fig. 1d). This results in chaperone structures with a broad range of inter-domain distances and orientations.

SurA binds its OMP substrates at multiple interaction sites
We next investigated how SurA binds its OMP clients, and how this affects conformations crosslinked SurA-OmpX complexes could be observed by SDS-PAGE (Fig. 4a), and following in-gel digestion a total of 26 unique inter-molecular crosslinked peptides were detected (Fig. 4b, Supplementary Table 7). Rather than revealing a unique binding site, the data show that the same residues of OmpX contact different sites on the chaperone surface, consistent with multiple substrate binding modes. Half (13/26) of the crosslinks observed are between OmpX and the SurA core domain (Fig. 4c, Supplementary Table 7). Four Lys residues in P2, one in P1, and 7 in the core also crosslinked to OmpX, but no crosslinks were detected for the remaining 11 Lys residues in P1 and P2 indicating a localised binding surface (Fig. 4b,c). Importantly, several crosslinks were detected from the same residue in OmpX to several residues on SurA (e.g residue 82 of OmpX crosslinks to 13 different residues in SurA spanning all four regions of the chain, Supplementary Table 7). Similarly, the same site on SurA crosslinked to multiple sites on OmpX (e.g. residues 135, 294, 389 and 395 in SurA each crosslink to three residues (50, 71 and 82) in OmpX, Supplementary Table 7) consistent with a flexible and dynamic OmpX in the bound state (Fig. 4b). Note that no crosslinks were observed between SurA and Lys112 and Lys122 in OmpX, suggestive of some specificity in the binding interaction, and that since the C-terminal 49 amino acids in OmpX lack Lys residues, no information on whether these regions interact with SurA could be obtained using this crosslinker.
To probe the organisation of the SurA-OmpX complex in more detail we next exploited the ability of the photoactivatable cross-linker MTS-diazirine (Supplementary Fig. 9a) to react rapidly (within ns 45 ), and non-specifically with any residue within ~15 Å of the diazirine moiety (Cα-Cα Euclidean distance) 46 . This "tag transfer" method was developed specifically to enable detection of weak and transient protein-protein interactions 46 Table 8) and P1 (two crosslinked sites), indicating that these regions form the heart of the binding epitope (Fig. 5b,c, Supplementary Fig. 9c).
Notably, no crosslinks were detected between OmpX and the SurA P2 domain or C-terminal region, despite the highly promiscuous photoactivatable crosslinker employed. This differs from the SurA-OmpX crosslinks detected with DSBU, probably because a much longer crosslinking time (45 min) required for cross-linking with DSBU. Overall, therefore, the results suggest that OmpX adopts a range of likely interconverting conformations upon binding SurA, in which multiple specific interactions are formed predominantly with the Nterminal region of the chaperone core domain.

Conformational changes in SurA upon OMP binding
Next, we examined the conformational changes induced by OMP binding to SurA using differential HDX-MS analysis ( Fig. 6a-f and Supplementary Fig. 10 To determine whether deprotection at the core-P1 interface occurs in the presence of other OMPs, the effects of binding the larger substrate, OmpF (16-stranded), on the HDX properties of SurA was examined. In the presence of OmpF, residues in the N-and Cterminal regions of the core domain were also protected from exchange, consistent with shared OmpX and OmpF binding sites. However, in marked contrast with the results for OmpX in which residues 46-72 of SurA were deprotected from exchange upon substrate binding, these residues were instead protected from exchange in the presence of OmpF, suggesting that the larger OMP binds to, or occludes, a greater surface area on the core ( Fig. 6c,d, Supplementary Fig. 10b). Importantly, as for OmpX, deprotection was observed in the P1 domain at the core-P1 interface, suggesting that structural reorganisation of this interface also occurs upon OmpF binding. Notably, the hinge region between P1 and P2 (residues 266-286) was also deprotected in the presence of OmpF, suggesting that binding of the larger substrate may also alter the conformational dynamics at locations more distal to the core.
To decouple the phenomena of protection arising as a result of OmpX/OmpF binding and deprotection as a result of conformational changes in SurA, we also compared the levels of deuterium uptake of SurA in the presence of a 7-residue peptide known to bind to the P1 domain (WEYIPNV, K d 1-14 µM) 49, 50 ( Fig. 6e,f, Supplementary Fig. 10c,d). Interestingly, extensive deprotection at the core-P1 interface (residues 39-74, 142-160 and 381-422) was observed in the presence of WEYIPNV, while protection was only observed in P1 at the known peptide binding site (residues 212-243 50 ) (Supplementary Fig. 10c,d). Combined, these results demonstrate that the OMP substrate binding surface is more extensive in OmpX/OmpF compared with WEYIPNV, but for all three cases binding triggers a structural reorganisation between the core and P1 domains.
To further study the effects of substrate binding on the conformations of SurA adopted in solution we used smFRET to examine the inter-domain distances of SurA bound to OmpX, OmpF or WEYIPNV (Fig. 7). Consistent with the HDX data ( Fig. 6), binding of OmpX to SurA resulted in changes at the core-P1 interface. Instead of the ca. bimodal E FRET distribution observed for the apo-SurA (E FRET centred on ~0.2 and ~0.6, Fig. 3d), a single maximum at an E FRET value (~0.5) between that of the open and closed states was observed for the SurA-OmpX complex (Fig. 7a). By contrast, the E FRET distributions for the core-P2  Fig. 11b,c). Similar effects on the E FRET distributions were observed when OmpF was added to SurA, (Fig. 7d-f). In marked contrast with the rather modest effects on the EFRET distributions on OmpX/OmpF binding, the addition of the P1binding peptide WEYIPNV had a profound effect on the core-P1 and core-P2 inter-domain E FRET distributions, inverting the populations of the core-P1 open and core-P1 closed distributions to favour core-P1 open (Fig. 7g), and decreasing the modal E FRET between core-P2 from ~0.34 to ~0.18 (Fig. 7h), without changing the P1-P2 E FRET distribution (Fig. 7i). Consistent with the HDX data, these results suggest that binding of WEYIPNV promotes the release of the P1 domain from the core. BVA on the SurA-substrate complexes indicated that all complexes remained dynamic on the sub-millisecond (sub-ms) timescale ( Supplementary   Fig. 12a-i), although the dynamics of the larger OmpF-SurA complex were dampened relative to those of SurA-OmpX (Supplementary Fig. 12d-f).

Discussion
Despite its key role in OMP biogenesis and bacterial virulence 51, 52 , how SurA binds its OMP substrates both specifically, but weakly 28,44,49 , and how it is able to protect its clients from aggregation and deliver them to BAM for folding into the OM, remain poorly understood in molecular detail. Previous NMR studies have shown that OmpX, tOmpA and FhuA are dynamically disordered when bound to SurA 40 -42 . However, precisely how SurA binds its OMP clients and how OMP binding alters the conformation(s) adopted by SurA in solution have remained unknown. Here, we have exploited XL, HDX-MS, MD and smFRET, to analyse the conformational dynamics of apo-SurA and to investigate how this is modulated by substrate binding. Further, we have identified the regions of SurA involved in substrate binding for both small (OmpX) and larger (OmpF) clients. The combined data presented are consistent with a model in which specific, yet multi-site, binding by a dynamically disordered substrate is accomplished within a cradle-like conformation of SurA that is very different to that observed in its crystal structure (Fig. 8).
SurA: a dynamic ensemble of states primed for OMP capture  Fig. 5). We propose that such rapid conformational changes are likely to be important for SurA to be able to bind its clients in the periplasm and to release substrates to BAM for folding into the OM. Notably, a similarly broad range of inter-domain motions on comparable timescales to those observed here for SurA was observed previously in MD simulations for the homologous chaperone TF 5 .
Our combined XL-MS, HDX-MS and smFRET data show that the P1 domain of SurA is not statically bound to the core domain in solution. Instead, these domains are in a dynamic equilibrium between core-P1 open and core-P1 closed states, suggesting a role for core-P1 dynamics in regulating access of the client OMP to the core chaperone domain and perhaps for access to P1 itself, providing client specificity and enhancing binding affinity 50 . Previous studies revealed that tethering the core and P1 domains by creation of a disulfide bond impairs OMP assembly in vivo, and that destabilising SurA can rescue OMP assembly defects in BAM-compromised strains 29,63 . These results can now be explained by the opening and closing motions between the core and P1 domains revealed here by smFRET and HDX-MS. We also show here that the P2 domain commonly populates conformations in close proximity to the core and P1 domains in solution (Figs. 1,2) in marked contrast with the orientation of P2 in the SurA crystal structure in which it is distal to P1/core, most likely as a consequence of crystal packing (Supplementary Fig. 2) 27 . Thus, our data are consistent with SurA frequently adopting a compact, possibly cradle-like, structure (Fig. 8) that may act as the acceptor state for client binding. However, an alternative model in which more extended SurA structures initially capture the unfolded OMP, followed by compaction of the complex, is also possible. In either scenario, the result of binding is a compact SurA in which the client is bound to the core domain, protecting the OMP from aggregation within the dynamic complex. Such a dynamic structure could provide a mechanism for release of bound OMPs to BAM for folding into the OM without the requirement for ATP binding/hydrolysis to drive client release.

Dynamic interplay between SurA and its substrates
Previous reports based on crystal contacts proposed that SurA may bind its substrates via a binding crevice in the N-terminal domain (Fig. 1) 27 . By contrast, the results presented here show that SurA binds its OMP clients at multiple sites 64 , predominantly involving the core domain as indicated by tag-transfer XL and HDX. The additional binding capacity in the P1 and P2 domains may be employed in a substrate-specific manner, as suggested by the finding that P2 is required to suppress the aggregation of the 10-stranded OmpT, but not the 8-strand tOmpA 28 . Previous results have also shown that deletion of P2 can perturb SurA function in vivo, as measured by a decrease in the amount of assembled LamB in a BAMcompromised strain 29 . These assembly defects are rescued by further deletion of the P1 domain, suggesting that P2 may also play a role in regulating interactions between the P1 and core domains that inhibit SurA chaperone function 29 17 , including OmpX and OmpF which were studied here. Further work will be needed to understand the relay of interactions between SurA, its clients, other chaperones, and BAM, and to discern whether/how the mechanism of OMP delivery to the OM is dependent on the identity of the OMP client.  Fig. 13). Instead, they cluster to regions that cluster within the cradle which forms by docking of the three domains of SurA and in which the OMP is sequestered and binds predominantly to the core domain (Fig. 8). Such a model is consistent with in vivo data showing that the SurA core domain alone can largely (but not wholly) complement deletion of wild-type SurA 24,29,67 . Specific client interaction sites have also been identified in TF using the substrate PhoA, which are also located in a cradle formed between its domains 68 . In the presence of substrate, an intermediate state between core-P1 open and core-P1 closed is observed by smFRET, suggesting that clamp-like motions between these domains may be utilised by SurA to bind and sequester its substrates. This is reminiscent of the lid motions in Hsp70 that entrap its substrates 69 , the sequestration of OMPs within the cavity of the chaperone Skp 55,56 , and the conformational flexibility that has been suggested to be important for substrate capture/release by Spy 9 .
SurA plays multiple roles in OMP biogenesis, including sequestration of OMPs in the periplasm to prevent their toxic aggregation 70 and delivery of OMPs to the BAM complex to enable folding into the OM 71 . The results presented here demonstrate a role for SurA interdomain dynamics in OMP binding, notably the reorganisation of the core and P1 domains, and dynamic localisation of P2 close to these domains. Such structural plasticity may also be important for facilitating binding of SurA:OMP complexes to BAM, assisting BAM catalytic activity, or priming the OMP for membrane insertion by pre-selecting favourable conformations for folding, thereby smoothing the energy landscape of folding 42,71,72 .
Understanding the interplay between the conformational dynamics of SurA and those of BAM, in particular the communication and coordination between SurA and different BAM subunits, will be essential in unravelling the molecular mechanism of OMP biogenesis. The model of SurA action presented here, whereby a compact, dynamic and responsive chaperone structure is responsible for client binding, represents a first key step in this endeavour. This adds to the growing body of data suggesting that all components of the OMP assembly line, including SurA, Skp 55, 56 and BAM 73 -77 have intrinsic conformational dynamics which, in combination, may be key to achieving efficient OMP biogenesis in the absence of ATP.

Cloning, expression and purification of SurA
A pET28b plasmid containing the mature SurA sequence preceded by an N-terminal 6x Histag and thrombin-cleavage site (pSK257) was a kind gift from Daniel Kahne (Harvard University, USA) 78 . The thrombin-cleavage site was mutated to a TEV-cleavage site using Cys-containing variants (Q85C, N193C, E301C, Q85C-N193C, N193C-E301C and Q85C-E301C) were generated by Q5 site-directed mutagenesis (New England Biolabs) and were purified as detailed above, except for the addition of 10 mM DTT to all buffers in the purification procedure, up until the elution step.

Expression and purification of OMPs
OMPs were purified using a method adapted from 48

Single molecule Förster resonance energy transfer (smFRET)
smFRET experiments were performed using a custom-built experimental set-up for µs ALEX as described previously 80 . Laser wavelengths and powers used were 488 nm, 140 µW and 594 nm, 120 µW, respectively. The laser alternation period was set to 40 μs (duty cycle of 40%). Samples of labelled SurA were prepared on the day of use from concentrated stocks that had been stored at -80 °C and were kept on ice and in the dark while in use. A sample Fluorescence bursts were analysed using customised Python 2.7 scripts 83 , and made use of FRETBursts, an open source toolkit for analysis of freely-diffusing single-molecule FRET bursts 84 . Functions from the FRETBursts package were used to estimate the background signal as a function of time, identify and remove artefacts due to photophysical effects such as blinking, and provide an optimal signal to noise ratio. To obtain EFRET values three correction parameters were applied as described previously 83 : γ-factor (to account for differences in the efficiency of excitation of each dye), donor leakage into the acceptor channel and acceptor direct excitation by the donor excitation laser. The data from each 10 minute acquisition was merged prior to subsequent analysis. In order to remove bursts arising from incorrectly labelled proteins, the data were filtered using ALEX-2CDE, yielding bursts with a Gaussian distribution of S values in a narrow range of dye stoichiometry (S within 0.25-0.75) 85 . Typically, ~10000 bursts were collected for each condition examined after all filters had been applied. Filtered bursts were then assembled into 1D histograms and kernel density estimation used to approximate 1D probability density functions of the EFRET values in each condition.
Burst variance analysis (BVA) 39   where r = the steady state anisotropy, r 0 = the initial anisotropy, t = fluorescence lifetime measured for the decay, and T r = the rotational correlation time.

Molecular dynamics simulations
All-atom molecular dynamics simulations of the mature sequence of SurA (residues  in explicit solvent were performed with GROMACS 5.0.2 88 using the CHARMM36 force field 89 . For simulations starting from the crystal structure for full-length SurA (PDB: 1M5Y 27 ), loop residues which are unresolved in the structure were modelled using MODELLER 90 , and the four missing N-terminal residues were added in Chimera 91 . The system was minimised (5000 steps) followed by equilibration for 25 ps, with backbone and sidechain position restraints of 400 kJ mol -1 nm -2 and 40 kJ mol -1 nm -2 , respectively, in the x, y and z directions.
The temperature reached its target value (300 K) within the first 10 ps and remained stable for the rest of the equilibration. The system contained 202 sodium ions and 198 chloride ions away from the core. The two structures were aligned on the core domain and the P1 and core domains were removed from the full-length SurA structure. Linker residues between domains were added using MODELLER 90 , and the four missing N-terminal residues were added in Chimera. The system was minimised (5000 steps) followed by equilibration for 25 ps with backbone and sidechain position restraints of 400 kJ mol -1 nm -2 and 40 kJ mol -1 nm -2 , respectively, in the x, y and z directions. The temperature reached its target value (300 K) within the first 10 ps and remained stable for the rest of the equilibration. The system contained 189 sodium ions and 185 chloride ions (150 mM NaCl), and 64,809 TIP3P water molecules. The total number of atoms was 201,129 in a periodic box size of 12.9 nm x 12.9 nm x 12.9 nm. Simulation systems were built using CHARMM-GUI 92 . In all simulations the pressure was maintained using a Parrinello-Rahman barostat 93 and the temperature was maintained using a Nose-Hoover thermostat 94 . The temperature of the systems was 300 K and the timestep was 2 fs.

Simulated Annealing
Simulated annealing calculations were carried out in XPLOR-NIH 95 . Crosslinks were treated as distance restraints with a flat-well energy potential using noePot. A rigid-body calculation (100 calculations in total) was performed, where each domain was treated as a rigid body and residues in the linker regions were given torsion angle degrees of freedom. Pseudopotential energy terms describing covalent geometry restraints were applied to restrict deviation from bond lengths, angles and improper torsion angles. The first step in the structure calculation consisted of 10000 steps of energy minimization, followed by simulated annealing dynamics with all the potential terms active, where the temperature is slowly decreased (3000-25 K) over 4 fs and a final energy minimization in torsion angle space.
During the hot phase (T = 3000 K) the crosslink terms were underweighted to allow the domain to sample a large conformational space and they were geometrically increased during the cooling phase. For each calculation the coordinates of P1 and P2 were randomized by applying a random translation within 20 Å and a random rotation within 90 o of their initial positions. The linkers were re-built using torsionDB 96 to enforce correct geometry before the first step in the structure calculation protocol. Given this simulated annealing approach will drive the structure to compact states, more extended states of SurA that smFRET data show are populated in solution, will not be captured by this method. The structures of the 10 lowest energy conformations of SurA are available at the University of Leeds data repository (https://doi.org/10.5518/701).

Labelling of Cys-OmpX with Alexa Fluor 488
where S obs is the observed signal, S U is the signal from unbound OmpX, S B is the signal from bound OmpX, and n is the Hill coefficient. Data fitting was carried out using IgorPro 6.3.4.1 (Wavemetrics, Oregon, USA).
Photocrosslinking was performed for 30 sec using a UV LED irradiation platform 46

Analysis of electrostatic surface potential and residue conservation of SurA
Calculation of the surface electrostatic potential of SurA was performed using the APBS plugin for PyMOL 100 . Amino acid conservation analysis was carried out using the ConSurf webserver using default parameters 101 .  Fig. 1a-c). Regions are coloured as in (a).       Whether SurA-bound OMP is in a collapsed globule (represented here as a sphere) or a more extended state remains unclear. However, the XL data are consistent with the presence of multiple, specific OMP recognition sites on SurA, suggesting a dynamic ensemble of bound structures. Note that the images presented are schematic and aim to portray the dynamics of the P1, P2 and core domains relative to each other, rather than atomic-level detail.