A Mass‐Spectrometry‐Based Approach to Distinguish Annular and Specific Lipid Binding to Membrane Proteins

Abstract Membrane proteins engage in a variety of contacts with their surrounding lipids, but distinguishing between specifically bound lipids, and non‐specific, annular interactions is a challenging problem. Applying native mass spectrometry to three membrane protein complexes with different lipid‐binding properties, we explore the ability of detergents to compete with lipids bound in different environments. We show that lipids in annular positions on the presenilin homologue protease are subject to constant exchange with detergent. By contrast, detergent‐resistant lipids bound at the dimer interface in the leucine transporter show decreased koff rates in molecular dynamics simulations. Turning to the lipid flippase MurJ, we find that addition of the natural substrate lipid‐II results in the formation of a 1:1 protein–lipid complex, where the lipid cannot be displaced by detergent from the highly protected active site. In summary, we distinguish annular from non‐annular lipids based on their exchange rates in solution.

Phospholipids were purchased from Avanti Polar Lipids Inc (USA) and 400 μM stocks were prepared as described, [4] with the exception of E. coli polar lipids which were solubilized at a concentration of 20 mg/mL in H 2 O supplemented with 10% DDM.
For detergent competition experiments, 5 μL lipid stocks were added to 10 μL protein in ammonium acetate to make 15 μL protein/lipid stock solutions. 1 μL of detergent in H 2 O at concentrations between 0 and 4 % (NG) or 0 and 0.4% (DDM) were then added to 3 μL of the protein/lipid stock to produce protein solutions with fixed lipid concentrations of 50 μM (E. coli polar lipids) or 40 μM 1-Palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE 16:0/18:1(9z)) and final detergent concentrations between 0 and 1% (NG) or 0 and 0.1 % (DDM). Mass spectra were recorded immediately after addition of lipids and secondary detergents (see Table S1), with incubation times ranging from 1 to 10 minutes, during which samples were stored on ice. Delipidation of LeuT was carried out as described. [3] Briefly, the protein was exchanged from 1% OG into 2% NG using size exclusion chromatography, followed by 16h incubation at 4 C to remove all co-purified lipids.Samples were introduced into the mass spectrometer using gold-coated borosilicate capillaries produced in-house. Mass spectra were acquired on a Waters LCT time-of-flight 1 mass spectrometer (MS Vision) modified for analysis of intact protein complexes [5] , and on a Q Exactive EMR or an UHMR Orbitrap mass spectrometer (Thermo) [6,7] . Settings on the Q Exactive EMR were: capillary voltage 1.2 kV, S-lens RF 100%, collisional activation in the HCD cell 100-200 V, argon UHV pressure 1.12 × 10 −9 mbar, temperature 60 °C, Xcalibur (Thermo) software packages. Protein structures were visualized using Chimera 1.13.1 [8] .

MD Simulations
To model the kinetics of LeuT-CDL, we built the crystal dimer of PDB 2A65 (as per reference [9] ) into a membrane composed of 95% POPE and 5% cardiolipin (CDL). The protein was described using the Martini open beta 3.0.b.3.2. [10,11] Additional bonds of 500 kJ mol -1 nm -2 were applied between protein backbone beads within 1 nm. For modelling undecaprenyl pyrophosphate (UDP) interactions, for each of the MurJ PDB entries 5T77, 6NC6, 6NC7, 6NC8 and 6NC9, the protein was described using Martini 2.2 [10,11] with additional stability imposed using EINeDyn. [12] Four copies of each protein were built into a single POPE membrane (ca. 23 x 23 x 10 nm) with 24 copies of the UDP molecule in a -2 charge state (parameters to be published separately). Positional restraints on a single backbone bead kept each monomer from interacting with each other. Both systems were built using insane [13] and solvated with Martini waters and ions to a neutral charge.
Systems were minimized using the steepest descents methods, then run for 1 ns with 5 fs timesteps and 100 ns with 20 fs timesteps, both in the NPT ensemble, with velocityrescale temperature coupling at 323 K [14] and semi-isotropic Berendsen pressure coupling [15] at 1 bar. Electrostatics were described using the reaction field method, with a cut-off of 1.1 nm using a potential shift modifier, and van der Waals interactions were shifted from 0.9-1.1 nm. Following this, the systems were simulated over 4 repeats of 15-18 µs (LeuT) or 1 repeat of 7-8 µs (MurJ), using velocity-rescale temperature coupling at 323 K and a semi-isotropic Parrinello-Rahman barostat. [16] All simulations were run using Gromacs 2019. [17] Images were produced using VMD [18] and data were plotted in Prism 7 (GraphPad).
For modelling the LeuT kinetics, we followed an approach described previously. [19] Minimum distance analyses were run between each of the residues shown in Figure 2c and each CDL residue in the system. These data could then be plotted to follow individual residue-CDL interactions (for example, see Figure S2a). Sites were defined on the surface of LeuT based on a previously-generated free energy landscape for LeuT-CDL interaction. [20] This landscape suggested two small CDL binding sites per LeuT protomer, one on and one distant from the dimer interface. We used these two sites, as well as a region where no significant CDL binding was observed, for the analyses here. For simplicity, each site was probed based on interaction with two key residues on the flanking edge of the site.
LeuT-CDL binding events were determined based on continuous time of a CDL molecule within a 0.8 nm cutoff of either of these two residues. Note that if a CDL molecule unbound and rebound within 1 ns it was considered as a continuous interaction. Cumulative CDL residencies were binned and plotted (see Figure 2c and Figure S2b), with binding events even shorter than 25 ns considered non-specific and disregarded. The binned data were fitted to a single exponential ( Figure S2b; red line) to provide a k off of the CDL interaction.
For the MurJ-UDP binding analyses, contact occupancies (as a fraction of total simulation time spent within 0.8 nm) were calculated between the UDP molecule and the specified residue, for each of the 4 proteins in the system. These residues were highlighted in the original structural studies as important for UDP binding.   notable preference for a specific lipid type. It should be noted that peak splitting due to a 58 Da adduct was consistently observed for PSH, which did however not appear to affect lipid binding. (b) Native MS of PSH in LDAO reveals that increasing the detergent concentration from 0.025% to 0.2% reduces signal intensity and results in the appearance of lower and higher charge states of PSH, indicating unfolding. It should be noted that in addition to the appearance of higher charge states, we also observe strong charge reduction. This is due to the increased concentration of LDAO concentration, which charge-reduces proteins in a concentration-dependent manner. [21] Table S1.