Dimyristoylphosphatidylcholine

For the lipid tails hexadecane
was chosen as a model compound for computing the charges. It is well-known
that partial atomic charges are conformation dependent⁵² but previous FFs have been parametrized from optimized
geometries. In order to address this issue, we performed a 10 ns long
MD simulation with pure hexadecane with FF parameters earlier derived
by our group.^{31 (link)} After the simulation, 54
random conformations were extracted and used for computing the charges
which were then averaged over all conformations in order to obtain
a final set. In this way, we obtained Boltzmann-averaged charges over
an ensemble of conformations in a procedure equivalent to the one
used by Sonne et al.^{30 (link)} We hope that by
averaging over an ensemble of conformations the effects of the conformational
dependence of partial charges are minimized. In computation of atomic
charges, a dielectric constant of 2.04 was used to mimic the dielectric
environment of the membrane’s hydrophobic part.
Atomic
charges for the lipid head group were obtained in a similar fashion
where 26 random conformations were chosen from a 20 ns long simulation
of an equilibrated bilayer (DMPC) with the same FF parameters used
in the initial simulation of hexadecane. A large part of the hydrophobic
parts of the lipids were then cut off in order to save CPU time and
the cropped lipids were placed in dielectric continuum with ε
= 78.4 in order to mimic the aqueous environment. Inclusion of solvent
effects results in a FF with implicitly polarized charges optimized
for condensed phase simulations. This has been proven to give reliable
results without any performance loss.^{53 (link)}For each molecular conformation, the charges were computed
using
the restricted electrostatic potential approach⁵⁴ (RESP) with the DFT method using the B3LYP exchange-correlation
functional^{55 −58} and the cc-pVTZ basis set.⁵⁹ The electrostatic
potential was sampled with the Merz–Singh–Kollman scheme⁶⁰ by single-point calculations and fitted during
the two-stage procedure developed by Cornell et al.⁶¹ All solvent effects were modeled by placing the molecule
in a polarizable continuum with different dielectric constant (see
above) with the IEFPCM continuum solvent model.^{62 ,63} The quantum mechanical calculations were performed with the Gaussian09
software package,⁶⁴ and the RESP calculations
were performed with the Red software.^{65 (link)} In subsequent molecular dynamics simulations, Coulombic 1–4
interactions were scaled by a factor of 0.8333.
The way the
atomic charges have been calculated and used in MD
simulations makes them compatible with the AMBER03 FF^{53 (link)} and since the charges in all AMBER FFs are derived from
the RESP the lipid FF presented here is compatible with most members
of the AMBER FF family. This is of importance since there is a growing
interest in simulating membrane proteins in their native environment^{19 (link),66 (link)} and also peptide partitioning in biological membranes.^{67 (link),68 (link)} Ongoing work aims to clarify which AMBER biomolecular FFs that work
sufficiently well together with the current parameters. A preliminary
test of the compatibility of the lipid parameters and the AMBER03
FF is presented further down.
Boltzmann averaging over charges
introduces temperature dependency
on the charges and in order to see the impact of temperature, simulations
with different temperatures (298, 303, 310, 318, and 325 K) were performed
with hexadecane using the methodology described above. No explicit
temperature dependence could be found over this range of temperatures,
making the charges reliable and robust with respect to temperature,
at least within the interval tested here (data not shown).

EmrE was expressed and purified using a 6× His-tag, which was then removed with thrombin. Purification was performed in decylmaltoside (DM) or dodecylmaltoside (DDM). EmrE was reconstituted into dimyristoylphosphatidylcholine (DMPC) or dilauroylphosphatidylcholine (DLPC) liposomes using standard methods and then formed into isotropic bicelles by addition of dihexanoylphosphatidylcholine (DHPC) and several freeze-thaw cycles. Protein concentration was determined using absorbance at 280 nm, with an extinction coefficient determined by amino acid analysis. ITC was performed by titrating 54 μM TPP⁺ stock solutions into 10 μM EmrE, with matching concentrations of detergent or lipid in both solutions.
Bulk FRET labeling was performed in liposomes to separately label residues on either side of the membrane. An “antiparallel” sample was labeled with donor and acceptor on opposite sides and a “parallel” sample was labeled with donor and acceptor on the same side. Donor-only and acceptor-only controls were labeled with dye only on the exterior of the liposome, reconstituted into bicelles and then mixed. Single-molecule FRET samples were labeled in micelles and experiments were performed using a wide-field total internal reflection fluorescence microscope (TIRF) setup^{43 (link)}.
NMR experiments were performed using a 700 MHz Varian NMR spectrometer or 800 MHz Bruker spectrometer equipped with a cryoprobe. All NMR samples contained 0.5–1.0 mM ²H/¹⁵N-EmrE in buffer conditions of 2 mM TPP⁺, 20 mM NaCl, 20 mM potassium phosphate, 2 mM TCEP, pH 7.0, 45°C. The membrane mimetic in each sample (DDM micelles or isotropic bicelles) is as listed. The TROSY-selected ZZ-exchange experiment^{35 (link)} was modified to include a lipid “flipback” pulse. Data were processed and analyzed with NMRPipe^{46 (link)}, NMRView^{47 (link)}, Sparky⁴⁸ , and IgorPro (Wavemetrics). All EmrE structure figures were created in PyMOL using PDB 3B5D with the backbone rebuilt to render the cartoons. Full page versions of the spectra in the main figures are included in the supplementary information.