Assembly and Channel Opening in a Bacterial Drug Efflux Machine

Summary Drugs and certain proteins are transported across the membranes of Gram-negative bacteria by energy-activated pumps. The outer membrane component of these pumps is a channel that opens from a sealed resting state during the transport process. We describe two crystal structures of the Escherichia coli outer membrane protein TolC in its partially open state. Opening is accompanied by the exposure of three shallow intraprotomer grooves in the TolC trimer, where our mutagenesis data identify a contact point with the periplasmic component of a drug efflux pump, AcrA. We suggest that the assembly of multidrug efflux pumps is accompanied by induced fit of TolC driven mainly by accommodation of the periplasmic component.


Figure S2
Asymmetry of the TolC open state.
Opening and asymmetry of the periplasmic end of the TolC channel as observed in the available crystal structures. The subunit non-equivalence in the crystal structure is most clearly seen in the inter-protomer distances. The outer rim of the TolC channel as defined by G365 is opening up in both novel crystal forms as compared with the closed state. The displacement of this residue also indicates clearly the asymmetric opening of the channel. 3

Figure S3
A docking model of the interaction of the AcrA hairpn (yellow) with TolC C2 structure.
The residues highlighted in red include K383 and R390 from TolC and D149 from AcrA. The numbering of AcrA is according to the full-length sequence of the protein as represented in the 2F1M PDB entry.

Protein purification
C41(DE3) and C43(DE3) cells were grown in 2xYT media at 37°C, induced at OD 600 of roughly 0.6, and then were grown for 12 hours at 24 o C. Cells were harvested in lysis buffer (20 mM Tris-Cl pH 8.0, 150 mM NaCl, 20 mM MgCl 2 , complete EDTA free proteinase inhibitor cocktail (Roche), supplemented with DNase I and lysozyme), lysed using an Emulsiflex C5 cell homogenizer (Avestin, Canada) and cell pellet removed by centrifugation at 10,000 x g. The membrane fraction from the supernatant was pelleted by centrifugation at 100,000 x g for 3 hours. The membranes were solubilized in 20 mM Tris-Cl pH 8.0, 150 mM NaCl, 20 mM MgCl 2 , 2.5 mM CaCl 2 , 2.5 mM KCl, 2% v/v Triton X100 on a rotary shaker at 4 o C overnight, centrifuged at 100,000 x g for 2 h and supernatant diluted 4 fold in column equilibration buffer (20 mM Tris-Cl pH 8.0, 150 mM NaCl, 0.15% v/v Triton X100 and 5 mM imidazole) and applied to a Ni-Chelating column (HiTrap, Amersham Biosciences). Protein was eluted with a linear gradient to 400 mM imidazole. TolC enriched fractions were applied on to anion exchange column (HiTrapQ, Amersham Biosciences), over which the detergent was exchanged to 0.05% (w/v) βDDM. Protein was eluted with a salt gradient and applied to an S200 size-exclusion column (Pharmacia) to transfer to storage buffer (20 mM Tris-Cl pH 8.0, 100 mM NaCl, 0.05% w/v βDDM).

Crystallographic model building
In the regions of structural movement, sections of the model were manually fitted into the electron density as rigid bodies. The poorly fitting sections as identified by realspace scoring function as implemented in RAPPERtk (Gore et al., 2007) and regions of poor geometry as identified by MolProbity (Davis et al., 2007), were rebuilt using RAPPERtk following the protocol used by RAPPER for low resolution model building (Furnham et al., 2006). The resulting models were refined using CNS (Brunger et al., 1998) and manually built in COOT (Emsley and Cowtan, 2004). In the final stages, REFMAC 5.3 was used (Murshudov et al., 1997). Residues after 428 are disordered in the structure. Model and refinement statistics are presented in Table   2. Figures were made using PyMol (http://pymol.sourceforge.net/). Secondary structures were assigned using DSSP (Kabsch and Sander, 1983). Ramachandran statistics for the refined structures were determined using PROCHECK (Laskowski et al., 1993) and RAMPAGE (Lovell et al., 2003)

6
Docking models for TolC/AcrB interaction were refined using ad-hoc protocols. For the two possible orientations we obtained, different conformations were generated by randomly sampling (within a maximum of 5 degrees) the mutual rotation of both molecules TolC and AcrB, together with random sampling (within 5 Å) of the intermolecular distance, while keeping the trimeric symmetry axis. These docking orientations were later refined by interface side-chain sampling with SCWRL, (Canutescu et al., 2003) and the one with the best pyDock (Man-Kuang Cheng et al., 2007) energy was selected.

Optimal Docking Area calculations
Optimal Docking Areas (ODA) of unbound TolC, AcrA, and AcrB, were calculated following a variation of a previously described method (Fernandez-Recio et al., 2005). For each surface residue, we computed the desolvation energy based on accessible solvent area for a surface patch formed by itself and the surrounding residues. Several distance values d (d = 1, 2, …, 20 Å) were used to find the surface patches with the best desolvation energy. Residues with patch desolvation energy < -10.0 kcal/mol define regions on the protein surface that are likely to be involved in protein-protein interactions. D795,G796 D795 a Refined docking models for TolC:AcrB interaction. Orientation 1 is the same as in Figure 3B. Orientation 2 is an alternative low-energy rotation of AcrB and TolC along the three-fold symmetry axis.
b Binding energy of the best refined model for each orientation is evaluated with pyDock (van der Waals + electrostatics + desolvation) c Satisfied restraints in the model are defined by those residue pairs whose Cα atoms would be at < 10Å, as expected from cross-linking experiments (Tamura et al., 2005).