Gas-Phase Protonation Thermodynamics of Biological Lipids: Experiment, Theory, and Implications

Abstract

Phospholipids are important to cellular function and are a vital structural component of the plasma membrane and organelle membranes. These membranes isolate the cell from its environment, allow regulation of the internal concentrations of ions and small molecules, and host diverse types of membrane proteins. It remains extremely challenging to identify specific membrane protein-lipid interactions and their relative strengths. Native mass spectrometry, an intrinsically gas-phase method, has recently been demonstrated as a promising tool for identifying endogenous protein-lipid interactions. However, to what extent the identified interactions reflect solution- versus gas-phase binding strengths is not known. Here, the “Extended” Kinetic Method and ab initio computations at three different levels of theory are used to experimentally and theoretically determine intrinsic gas-phase basicities (GB, −ΔG for deprotonation of the protonated base) and proton affinities (PA, −ΔH for deprotonation of the protonated base) of six lipids representing common phospholipid types. Gas-phase acidities (ΔG and ΔH for deprotonation) of neutral phospholipids are also evaluated computationally and ranked experimentally. Intriguingly, it is found that two of these phospholipids, sphingomyelin and phosphatidylcholine, have the highest GB of any small, monomeric biomolecules measured to date and are more basic than arginine. Phosphatidylethanolamine and phosphatidylserine are found to be similar in GB to basic amino acids lysine and histidine, and phosphatidic acid and phosphatidylglycerol are the least basic of the six lipid types studied, though still more basic than alanine. Kinetic Method experiments and theory show that the gas-phase acidities of these phospholipids are high, but less extreme than their GB values, with phosphatidylserine and phosphatidylglycerol being the most acidic. These results indicate that sphingomyelin and phosphatidylcholine lipids can act as charge-reducing agents when dissociated from native membrane protein-lipid complexes in the gas phase and provide a straightforward model to explain the results of several recent native mass spectrometry studies of protein-lipid complexes.

Graphical Abstract

Introduction

Phospholipids are a major component of physiological membranes, where they interact directly and indirectly with diverse membrane proteins. Native mass spectrometry^1,2 has recently been demonstrated as a promising tool for rapidly identifying native protein-lipid interactions^3–19. In many of these experiments, membrane proteins are embedded in a lipoprotein Nanodisc or lipid and detergent micelle environment in solution. These complexes, which initially contain many tens of lipid/detergent molecules, are then dissociated in the gas phase to determine which lipids remain associated with protein after extensive dissociation of detergent or other lipid molecules. The remaining lipids are typically inferred to be structural or annular lipids^4,14,16,20, and apparent binding constants determined from these gas-phase experiments can be remarkably similar to those measured using condensed phase methods^6,7,13. However, to what extent the observed relative abundances of specific lipid adducts identified using these experiments reflect gas-phase or solution-phase binding, or both, is not known. The relative lability of phospholipids in Nanodiscs in response to gas-phase activation has been attributed to their tendency to carry charge in the gas phase, and gas-phase basicity-based arguments have been used to explain effects of supercharging²¹ and charge-reducing reagents²² on charge states and stabilities of gas-phase native membrane protein-lipid assemblies. Gaining an understanding of how gas-phase lipids intrinsically behave from a physical chemical perspective is important for understanding protein-lipid interactions observed in native mass spectrometry studies and to what extent they may directly reflect physiological protein-lipid interactions in the condensed phase.

The gas-phase basicity (GB, negative ΔG of protonation at 298 K) and proton affinity (PA, negative ΔH of protonation at 298 K) of molecules and ions strongly affects the stability of ionic non-covalent hydrogen bonding interactions in the gas-phase and upon transfer of such species from solution to the gas phase.^23–25 The GB values of thousands of molecules have been measured and tabulated, and for biomolecules these range from < 346.3 kJ/mol for highly-elongated supercharged protein ions²⁶ to 1007 kJ/mol for arginine.²⁷ However, the GB of common phospholipids have not been measured to date. One of the most typical methods to experimentally determine GB is the Kinetic Method.^26,28,29 In this approach, a proton bound dimer ion of two bases is formed and transferred into a mass spectrometer (often by electrospray ionization), and the dimer is activated in the gas phase until it dissociates, typically resulting in one of the bases taking the proton and the other leaving as a neutral molecule. In the “Extended” Kinetic Method, proton-bound dimers of a base of unknown PA with other bases of known PA are dissociated as function of collision energy to determine the PA of the unknown base and investigate entropy effects.³⁰Seminal work by Johnson and coworkers²³ explains how the relative strength of gas-phase proton bonds can be attributed to how close together the GB of the constituent neutral monomers are; bases with highly disparate GB tend to form weaker proton-bound dimers than do bases with similar GB. The magnitude of differences in GB for dimers of bases with a shared-proton bond can also influence the site and geometry of the shared-proton bond and the stability of different isomers of the bases within the dimers.³¹

Using mass spectrometry and ab initio computations, we determine the GB and PA of six of the most common phospholipid headgroups (see Figure 1). Ab initio computations are used to rationalize experimental Extended and Conventional Kinetic Method results and are found to agree well with the experimentally determined relative and absolute basicities of the lipids. These studies indicate that sphingomyelin (SM) and phosphatidylcholine (PC) are the most basic known biomolecules in the gas-phase to date and form a benchmark for interpreting native protein-lipid interactions in gas-phase experiments on lipid-bound proteins and protein complexes.

Figure 1:

Chemical structures of the six lipid headgroups studied, shown in their common physiological total charge and protonation states. PhA = phosphatidic acid, PC = phosphatidylcholine, PE = phosphatidylethanolamine, PG = phosphatidylglycerol, PS = phosphatidylserine, SM = sphingomyelin. PhA, PG and PS are shown here in their deprotonated forms, and all other lipids are shown in their net neutral forms. Two diastereomers of PG were evaluated due to a racemic mixture of both isomers being present in natural PG samples from the non-stereoselectivity of the biosynthesis of the glycerol headgroup. R₁ and R₂ represent acyl chains. For the calculations here, methyl groups were used for both R₁ and R₂ to limit computational time, whereas in Extended and Conventional Kinetic Method experiments, R₁ and R₂ were longer acyl chains more typical of physiological membranes (see Methods section).

Methods

Materials.

L-arginine and L-alanine were purchased from Sigma-Aldrich (St Louis, MO, USA); L-lysine from Sigma-Aldrich (Buchs, Switzerland); L-histidine and L-proline from Sigma-Aldrich (Tokyo, Japan); L-tryptophan from Alfa Aesar (Lancashire, United Kingdom); 34:2-L-α-phosphatidylcholine (egg yolk, Type XVI-E), 36:2-L-α-phosphatidylethanolamine (egg yolk, Type III), 36:1–3-sn-phosphatidic acid (egg yolk) from Sigma-Aldrich (St Louis, MO, USA); 18:0-sphingomyelin (egg yolk) from Sigma-Aldrich (Ontario, Canada); L-α-36:1-phosphatidylserine (bovine brain) and 36:1-L-α-phosphatidylglycerol (egg yolk) from Avanti (Alabaster, AL, USA). Methanol was obtained from Honeywell (Seoul, South Korea). Chloroform was obtained from ACI Labscan (Bangkok, Thailand). Deionized water (18 MΩ/cm resistivity) was obtained using a water purification system (MilliQ, Merck, Darmstadt, Germany). All chemicals were used without further purification. Stock solutions of amino acids were prepared in deionized water and lipids in chloroform. To form protonated dimer complexes, solutions contained 60 μm and 40 μm of the respective amino acid and lipid of interest in 99:1 vol:vol methanol:acetic acid. For protonated heterodimer lipid complexes, solutions for nanoelectrospray ionization (nanoESI) were prepared containing 20 μm of each lipid in 99:1 vol:vol methanol:acetic acid. For deprotonated heterodimer complexes, nanoESI solutions contained 20 μm of each lipid in 99:1 vol:vol methanol:ammonium hydroxide.

Mass Spectrometry.

Conventional Kinetic Method mass spectrometry experiments were performed using a hybrid linear quadrupole ion trap and Orbitrap mass spectrometer (LTQ Orbitrap XL; Thermo Fisher Scientific), equipped with a nanoelectrospray ionization (nanoESI) source. NanoESI emitter tips were fabricated by pulling borosilicate capillaries (1.2 mm OD, 0.69 mm ID, Harvard Apparatus Limited) to an inner diameter of ~1.3 μm using a microcapillary puller (Model P-97, Sutter Instruments).^32,33 The emitters were coated with a mixture of Au and Pd for 20 s using a Scancoat Six (Edwards; Au/Pd alloy targets). NanoESI emitters were positioned ~2 mm on axis from the heated capillary entrance to the mass spectrometer (200 °C). For nanoESI, a voltage of +1.0 to 1.2 kV was applied to the emitters relative to the capillary entrance of the mass spectrometer. Collision-induced dissociation was conducted in an octopole collision cell (referred to as higher-energy CID by the instrument supplier) using an isolation width of 5 m/z centered on the ion of interest. The normalized collision energies (NCE) that were used resulted in ~50% depletion of the precursor ions are given in the Supporting Information (Supplementary Table S1). For the Extended Kinetic Method, experiments were conducted on an Orbitrap QExactive Plus mass spectrometer (Thermo Fisher Scientific). Samples were injected at a flow rate of 5 μL/min using electrospray ionization (ESI) through a stainless-steel capillary tip with inner diameter ~76 μm. A spray voltage of +4 kV was applied relative to the capillary entrance of the mass spectrometer. For higher-energy CID, an isolation width of 5 m/z centered on the ion of interest and collision voltages from +10 to +30 V were used. In higher-energy CID, ion fragmentation is performed in an external octupole collision cell, which eliminates the low mass cutoff effect of fragmenting isolated ions inside of quadrupole ion traps.³⁴For both CID experiments, N₂ (~1–5 mTorr) was used as the target gas in the collision cell.

Computations.

Ab initio computations were performed for phosphoric acid and model phosphatidic acid (PhA), PC, phosphatidylethanolamine (PE), phosphatidylglycerol (PG), phosphatidylserine (PS), and SM in the 1- (except for PC), 0, and 1+ charge states in order to estimate the GB and PA of each lipid headgroup. In these model lipid headgroups, methyl groups were used as the acyl chains to reduce computational time as compared to much longer physiological chains, and the resulting structures provide more realistic models than those previously reported, which were truncated at the glycerol backbone with a methyl group.³⁵ Because the lipids all contain highly basic sites that interact more strongly with other polar groups than nonpolar aliphatic groups, it is expected that this choice of acyl chains should only slightly affect computed GB/PA in comparison to longer aliphatic chains.

Chemical computations were carried out using the University of Oregon’s high-performance computing cluster, Talapas, and the Molecular Graphics and Computation Facility at the University of California, Berkeley in a manner similar to that previously described.^21,22 In short, initial structures of the lipids and phosphoric acid in each charge state were constructed in the molecular modeling software Maestro v. 2019.2 (Schrödinger, Inc.) and candidate structures were generated using Monte Carlo Multiple Minimum (MCMM) conformational searching in MacroModel v. 2019.2 (Schrödinger, Inc.) with a potential defined by the MMFF94 forcefield. All candidate structures were then geometry optimized using this forcefield with a steepest descent algorithm. The three conformers of each lipid in each charge state with the lowest predicted energy were selected to be further geometry optimized using quantum mechanical computations using Gaussian 09 (Gaussian, Inc.) first at the B3LYP/6–31G* level of theory. Conformers were then further geometry optimized and harmonic vibrational frequencies were calculated using the B3LYP/6–31++G** level of theory. Subsequent geometry optimizations and harmonic frequencies computations were performed using the B3LYP/6–311++G** level of theory. Both B3LYP/6–31++G** and B3LYP/6–311++G** levels of theory were selected for these computations for their previously documented accuracy in predicting experimental thermodynamic trends of biomolecules containing organic phosphate groups.^35,36 Zero-point corrected enthalpies and Gibbs free energies at 298 K were calculated without rescaling of the vibrational frequencies. Proton affinities (PA) and gas-phase basicity (GB) values of a given molecule were calculated by subtracting the 298 K enthalpy or Gibbs free energy, respectively, of the protonated structure from that of the neutral structure and then correcting for the standard 298 K enthalpy or Gibbs free energy of formation for a proton gas.³⁷When this method is applied to water, computed values for PA and GB are found to be 687.7 kJ/mol and 656.6 kJ/mol respectively, in excellent agreement with accepted values from Hunter and Lias²⁷ (691.0/ 660.0 PA/GB, in kJ/mol). Gas-phase enthalpies of deprotonation of the neutral form were calculated analogously using the neutral and anionic form of the lipids studied. Single-point energy computations were also carried out using the MP2/6–31++G** and the M06–2X/6–31++G** levels of theory on the optimized geometries found using the B3LYP/6–31++G** level of theory. GBs and PAs were computed for these levels of theory using these single-point energies and the Gibbs free energy and enthalpy corrections from the B3LYP/6–31++G** frequency computations.

Results

Basicity Ranking Based on Collision-Induced Dissociation Bracketing.

GB and PA values for many amino acids have been measured by both the Kinetic Method,²⁸ion-molecule reactions in ion trapping instruments,^26,38–40 and high-pressure mass spectrometry⁴¹ experiments as well as computational methods.^21,22,24,42 Based on computational results (see below), all of the net neutral lipids investigated were expected to have relatively high GB/PA, near the range spanned by alanine and arginine (which have GB of 867.7 ± 8.0 and 1006.6 ± 8.0 kJ/mol, respectively²⁷). Conventional Kinetic Method experiments using lipids bound via a shared proton bond with these amino acids were performed to determine a range of possible GB/PA values for each lipid. Upon collisional activation of the gas-phase proton-bound dimers, the species preferentially retaining the proton after dissociation was determined to be the more basic molecule, and formation of their respective protonated product ions in similar relative abundances was taken to indicate similar GB for the two bases. Intriguingly, 18:0-SM and 36:2-PC were both found to be more basic than arginine (the most basic of the 20 common amino acids) in the gas phase²⁷ (Table 1 and Figures S1 and S2, respectively). PE was found to be less basic than arginine but more basic than lysine, which has a GB of 951.0 ± 8.0 kJ/mol²⁷(Figures S3a and S3b, respectively). PS was found to be less basic than lysine and histidine and more basic than tryptophan, but closer in basicity to histidine than to tryptophan, with a GB of 915.0 ± 8.0 kJ/mol²⁷ (Figures S4a and S4b, respectively). Based on the close GB/PA of lysine and histidine, PE and PS are therefore close together in GB. PhA and PG were determined to have the lowest GB among the lipids studied, being less basic than proline (GB of 886.0 ± 8.0 kJ/mol²⁷) but more basic than alanine (GB of 867.7 ± 8.0 kJ/mol²⁷) (Figures S5a, S5b, and S6).

Table 1:

Relative GB and quantitative PA values of lipid headgroups as determined by Conventional and Extended Kinetic Method experiments compared to select amino acids.

Analyte	GB (kJ/mol)	PA (kJ/mol)
SM	>1006.6	N/A
PC	>1006.6	N/A
argininea	1006.6	1051.0
PE	951.0 to 1006.6	996.6 (± 5.5)b
lysinea	951.0	996.0
histidinea	950.2	988.0
PS	915.0 to 950.2	989.3 (± 9.5)b
tryptophana	915.0	948.9
methioninea	901.5	935.4
asparaginea	891.5	929.0
prolinea	886.0	920.5
PG	867.7 to 886.0	909.1 (± 14.0)b
PhA	867.7 to 886.0	895.6 (± 6.3)b
alaninea	867.7	901.6

^aExperimental GB values of amino acids from Hunter and Lias,²⁷ PA values from this source are reported to have an approximate uncertainty of ±8.0 kJ/mol (RMSD) and GB values are assumed to have at the least the same magnitude of uncertainty as PA values

^bReported here based on Extended Kinetic Method measurements. PAs for SM and PC are not available (N/A) because as sufficiently basic reference molecule is unavailable. Uncertainties for the measured PA values are based on 95% confidence intervals (see SI for full details) and do not include the error in the GB values for each reference amino acid (ca. ±8.0 kJ/mol).

To obtain proton affinity values from CID measurements and investigate entropy effects, the Extended Kinetic Method can be used.^30,43–45 Full details of this approach are given in the SI. Here, CID spectra of each respective proton bound dimer of PE, PS, PhA, or PG with two reference bases were fragmented as a function of collision energy and the abundances of the protonated lipid (I_L) and the protonated reference base (I_i) were measured (Figure S15). By plotting ln(I_L/I_i) against the average PA (PA_avg) of the reference amino acids,⁴⁵the linear regression best fit line has a slope (m₁) corresponding to –1/RT_eff and an y-axis intercept (y01) corresponding to ([PA_L-PA_avg]/RT_eff – ΔS/R), in which PAL is the proton affinity of the lipid of interest, R is the gas constant, T_eff is the effective temperature, and ΔS corresponds to the average change in entropy for proton exchange between the reference bases and the lipid. By plotting y01 vs 1/RT_eff (negative of m₁) for each collision voltage that resulted in a product ion yield that is ≤ 95%, the slope of this second plot (m₂) can be used to obtain the PA of the lipid and the intercept (y02) is –ΔS/R. Thus, the PA and ΔS can be obtained for each lipid that was of a sufficiently low basicity to be bracketed by an amino acid reference base using the Extended Kinetic Method (Table 1 and Figure S15). The relative PA values obtained from the Extended Kinetic Method experiments are in excellent agreement with the basicity rankings obtained using the more conventional bracketing method. For example, the PA of PE is bracketed by that for arginine and lysine, and the PAs of PG and PhA are both between the PAs of proline and alanine (Table 1). (PE, PS, PG, and PhA were bracketed by arginine and lysine, histidine and tryptophan, proline and alanine, and proline and alanine, respectively.) The PA of PS is measured to be slightly higher than histidine but within measurement uncertainty (Table 1). These results suggest that the entropy effects (ΔS ranging from approximately −36 to +13 J mol⁻¹ K⁻¹) for the dissociation of the proton bound dimers (Figure S15) are not sufficient to significantly reorder the relative gas-phase basicity rankings of the six phospholipids within measurement uncertainty.

Based on these results and the principles described by Johnson and co-workers,²³it is expected that SM and PC lipids typically form relatively weak shared proton bonds with peptides and proteins in the gas-phase, due to the large difference in GB (calculated to be ≥ 39 kJ/mol; see below) between these two lipids and even the most basic amino acid (arginine). Under gas-phase slow-heating conditions, as in collision-induced dissociation (CID), these lipids should preferentially leave with the shared proton upon dissociation from the protein or peptide surface, analogous to charge-reducing reagents^.22 By contrast, PE and PS lipids should be able to form stronger shared proton bonds with many amino acids in gas-phase peptides and proteins owing to the relatively small difference in GB values (estimated as < 10 kJ/mol, see below) between these lipids and protonated basic amino acids. In positive ion mode, many of the protons on positively charged, protonated proteins are expected to reside on arginine side chains and, to a lesser extent, lysine side chains, due to their especially high GB/PA and because the number of these amino acids on the surface of most protein ions in positive ion mode typically exceeds their net charge under native-like conditions. Thus, PE and PS should often be able to form relatively strong shared proton bonds with arginine and lysine side chains, which may become more accessible upon activation and migration of the lipids. This, in return, should make these phospholipid types require higher energy ion activation to dissociate from the peptide or protein surface by gas-phase slow-heating methods such as CID. In positive mode, all four of these lipids (PE, PS, PhA, and PG) should have a stronger tendency to leave as neutral lipids in CID and other slow-heating experiments than should SM and PC, particularly for complex ions that are multiply charged owing to Coulomb repulsion between such charge sites lowering the GB of peptides and proteins. the measured PA values are based on 95% confidence intervals (see SI for full details) and do not include the error in the GB values for each reference amino acid (ca. ±8.0 kJ/mol).

Collision Induced Dissociation of Lipid Heterodimers.

Activation and dissociation of proton-bound lipid heterodimers allows for direct comparison of the GB of different lipids relative to one other (see Figures S7–S11). Consistent with the high GB of SM and PC determined above, dissociation of the proton-bound dimer complex between these two lipids resulted in a rich fragment ion spectrum (Figure S7). Fragment ions included neutral losses from both protonated SM (730.6 Da) and PC (757.6 Da) (relative fragment ion abundance of ~65%), in addition to neutral losses directly from the proton bound dimer (~28%). Specifically, high-energy collision induced dissociation of the proton bound dimer of SM and PC resulted in formation of abundant peaks at m/z 703.6 and 184.1. The former ion either corresponds to the direct loss of ethene from protonated SM or results from reaction that substitutes a C18 acyl chain for a C16 acyl chain from PC (~35%). The ion at m/z 184 is assigned to the loss of the lipid phosphocholine headgroup from protonated SM or PC or both (~30%), respectively. In addition, a minor fragment ion is assigned to protonated PC (~7%). Importantly, CID of the proton-bound PC and SM dimer can result in the cleavage of covalent bonds in preference to cleaving non-covalent, i.e., shared-proton bonds. Upon cleavage of covalent bonds due to the high dimerization energy that these species possess, alternative CID pathway are likely to be observed. Many examples of these alternative dissociation pathways have been discussed thoroughly by Blanksby and coworkers,³⁵but in short, it has been observed in many cases that the phosphate group on a phospholipid is able to readily transfer between dimerized bases in a substitution-like reaction. Other pathways observed in this article were intramolecular reactions that result in small, neutral losses for gas-phase monomeric phospholipids. For example, fragment ions at m/z 786.6 and 969.7 correspond to the neutral losses of 702.6 Da (~13%) from the SM lipid backbone, and 519.5 Da (~4%) from the PC lipid backbone, respectively. These results indicate that ionic hydrogen bonding and other intermolecular forces between these lipids are unusually high. Overall, these data indicate that SM has approximately the same basicity as or slightly higher basicity than that of PC, making it the most basic phospholipid among those studied and the most basic monomeric gas-phase biomolecule identified to date.

Similarly, for the assigned fragment ions of the proton bound dimer of PhA (702.5 Da) and PG (776.6 Da), the major peaks at m/z 605.6 Da and 735.5 Da correspond to neutral losses from protonated PhA or PG (relative fragment ion abundance of ~36%) and the PG lipid backbone of the proton bound dimer (~48%), respectively (Figure S11). Thus, PG has basicity similar to or slightly higher than PhA. In contrast, for the proton bound dimer of PE (743.6 Da) and PS (789.6 Da), the two most abundant peaks at m/z 744.6 Da and 790.5 Da correspond to protonated PE (relative fragment ion abundance of ~55%) and PS (~31%), respectively (Figure S9). Thus, PE is slightly more basic than PS. Overall, these results also indicate that CID of the proton-bound PhA and PG dimer leads to preferential covalent bond cleavage.

The relative gas-phase acidities of PhA, PG, and PS were also determined by CID of the deprotonated lipid heterodimers (Figures S12 to S14). For the PS-PG and PS-PhA heterodimers, the most abundant peak was deprotonated PS (relative ion abundance of ~77% and 87%, respectively), and for the PG-PhA heterodimer it was deprotonated PG (~65%). Thus, these results indicate that non-covalent bond cleavage is preferred for these deprotonated dimers. In addition, the order of gas-phase acidity was determined to be PS > PG > PhA, i.e., PS is the most easily deprotonated of these three net-neutral lipids. (Heterodimer complexes containing deprotonated PE, PC, and SM were not readily generated owing to the low stability of their anionic forms, so relative gas-phase acidity values for these phospholipids were not determined experimentally.) These results agree with a previous report from Blanksby and co-workers,³⁵ which found the same acidity ranking for lipids with these headgroups and two palmitoyl acyl chains.

Computational Results.

Ab initio computations were used to theoretically determine ΔG and ΔH values of protonation of each lipid from a charge state of 1- to 0 and from 0 to 1+ using the lowest-energy structures identified for each charge state. In order to properly evaluate the conformational space of each structure studied, Monte Carlo torsional sampling was employed to ensure that true ground state structure for each lipid/amino acid was found. Conformer searching was used to rank each structure in terms of potential energy computed using the MMFF94 forcefield, and the three lowest-energy structures from this search were subjected to subsequent ab initio computations detailed in the Methods section. In general, these three structures were found to be within 5–10 kJ/mol of each other and differ with respect to the network of intramolecular hydrogen bonding. For example, Figure 2 shows the different hydrogen bonding motifs for the three different structures of PG 1 in the 1+ charge state. These structures were all computed to be within 4 kJ/mol of one another, indicating that these structures may all be present in the gas phase and suggesting that many competitive hydrogen-bonding patterns may be present for the dimer ions investigated. Good agreement was found between the set of low energy structures found for amino acids here and previously reported structures^46–49. The computed thermodynamics for protonation and deprotonation here use the energies for the lowest-energy forms of each species, thus, if multiple structures are present at the temperature of the experiment, the ensemble-averaged thermodynamic values, including entropic contributions, could be slightly different from those reported here.

Figure 2:

Three lowest energy structures of PG 1 in the 1+ charge state found using Monte Carlo conformational searching and optimized at B3LYP/6–311++G**. The global minimum in terms of ΔG/ΔH of formation is conformer a. Conformer b is higher in energy by 0.7/3.8 kJ/mol (ΔG/ΔH, respectively) and conformer c was found to be higher in energy by 1.5/3.5 kJ/mol (ΔG/ΔH, respectively).

In general, low-energy structures for 1- and 1+ lipids were found to be either salt-bridge (SB) structures, with three sites carrying formal charges, or charge-solvated (CS), with only one site bearing formal charge, and low-energy structures for net neutral lipids were either SB (i.e., zwitterionic) or non-zwitterionic (NZ). The lowest-energy structures at 298 K identified for each lipid in each charge state are summarized in Table 2. A summary of all computations carried out using the B3LYP/6–311++G** level of theory that lists zero-point energies, enthalpies of formation and Gibbs free energies of formation along with geometries (tabulated and pictured) can be found in the Supporting Information (Supplementary Table S2, Tables S6–S40 and Figures S16–S19, respectively). For the 1- charge state, all of the lowest energy structures identified were deprotonated at the phosphate group (Figure S16). Only PS and SM 1- ions were found to have SB structures, which are preferentially stabilized by extensive self-solvation of the positive charge site at the ammonium group for PS and by the presence of the quaternary ammonium in SM. In their net neutral charge state, PC and SM were found to be SB structures, due to the presence of the quaternary ammonium group, but all other lipids were found to adopt CS structures for this charge state (Figure S17). Intriguingly, in the 1+ charge state, all lipids investigated were found to adopt CS structures, representing an asymmetry with the 1- charge state that arises in part due to the different intrinsic energies of protonation and deprotonation of phosphoric acid in the gas phase (Figure S18). That is, neutral phosphoric acid and the phosphoester groups on these lipids are much more basic in the gas phase than they are acidic. Furthermore, for several of the lipids, “head-to-tail” chains of hydrogen bonds lend stability to the CS 1+ structures that are not possible for the corresponding 1- lipids, which have two fewer hydrogen-bondable protons.

Table 2:

Computationally determined GB/PA for relevant neutral gas-phase amino acid brackets, lipid models and phosphoric acid. The ΔG/ΔH of deprotonation was only computed for neutral lipid models and phosphoric acid, with all values computed at the B3LYP/6-311++G** level of theory^a.

Modelled phospholipid (most stable 1−, neutral, and 1+ structure types)b	GB/PA for the neutral charge state (kJ/mol)	ΔG/ΔH of deprotonation for the neutral charge state (kJ/mol)
SM (SB, SB, CS)	1055.2/1089.6	1414.5/1439.8
PC (N/A, SB, CS)	1044.7/1068.2	N/Ac
arginine	1017.0/1051.0	-
PE (CS, NZ, CS)	946.4/978.9	1320.0/1356.0
lysine	951.1/993.9	-
histidine	949.7/982.3	-
PS (SB, NZ, CS)	945.7/992.9	1283.7/1310.6
tryptophan	914.9/949.8	-
proline	907.6/939.2	-
PG 2 (CS, NZ, CS)	913.6/941.5	1274.1/1307.5
PG 1 (CS, NZ, CS)	880.0/923.6	1280.4/1314.4
PhA (CS, NZ, CS)	867.5/901.9	1316.2/1346.6
alanine	869.1/901.9	-
phosphoric acid	795.7/828.5	1332.1/1361.4 (1351/1383d)

^aThe uncertainty of these computations at this level of theory is estimated to be ±19.2/±15.7 (GB/PA respectively, 95% confidence interval; see main text). Single-point energies for all structures calculated at the B3LYP/6-31++G**, MP2/6-31++G**, and M06-2X/6-31++G** levels and thermal corrections computed at the B3LYP/6-31++G** level can be found in the Supporting Information (Supplementary Table S3).

^bSalt bridge (SB; i.e. zwitterionic), charge solvated (CS; one site bearing formal charge), or NZ (non-zwitterionic).

^cEnergies of deprotonation for PC were not computed due to the lack of deprotonatable groups present on the structure of PC in the neutral charge state.

^dExperimental values.³⁵

The protonation and deprotonation energies determined from these computational results at the B3LYP/6–31++G** and B3LYP/6–311++G** level of theory are largely consistent with the Kinetic Method experiments described above (see Figure 3 for a comparison of experimental, literature, and computed GB values).

Figure 3:

Comparison of experimental Kinetic Method bracketing, literature, and computed GB values for amino acids and model lipids. The arrows for PC and SM brackets indicate that only lower bracket values were found due to the extremely high basicity of these lipids.

A minor exception is that these computations place PE and PS slightly below the experimental GB of lysine, but only by ~5 kJ/mol, which is likely within the uncertainty of the calculations. The RMSD between our computed values for the GB/PA of the amino acids studied and experimentally determined values listed in Table 2 indicates an approximate uncertainty of ±19.2/±15.7 kJ/mol (GB/PA respectively, 95% confidence interval) for these computations for the 7 lipid structures evaluated. The computed GBs for each lipid model at the MP2/6–31++G** and M06–2X/6–31++G** levels of theory, which are given in the Supporting Information (Supplementary Tables S4 and S5) yielded a slightly different ordering, with PS more basic than arginine, in contrast with experiment. All computed deprotonation and protonation thermodynamics values obtained using the 6–31++G** basis set are tabulated and compared graphically in the Supporting Information (Supplementary Tables S3, S4 and S5 along with Supplementary Figures S20 and S21 respectively).

Notably, the computations confirm that PE and PS, while falling into different brackets defined by known amino acid GB/PA values in experiments, have GB values that are very close together. Similarly, the computations indicate that PhA and PG indeed have similar GB/PA, consistent with the Kinetic Method results (Figure S14). Finally, although we were unable to determine experimental gas-phase acidity values owing to the dearth of well-established gas-phase acidity values for anions in the literature that are also suitable for readily forming anionic proton-bound dimers using nanoESI, computed gas-phase acidity values for PhA, PG and PS were computed to have the same acidity ranking as that found experimentally, and PE and SM were calculated to be much less acidic than these three lipids. (The especially high ΔG/ΔH of deprotonation for SM indicates that an exceptionally high amount of energy is required to deprotonate the neutral molecule and is attributed to its lowest-energy deprotonation site being the amide NH group).

Discussion

Each lipid studied here is quite basic in the gas phase compared to many amino acids due to the high basicity of the phosphoester group, which is the lowest-energy site for protonation for each net neutral lipid. Intriguingly, computations indicate that the phosphoester group is so basic in the gas phase that it causes net neutral conformers of PE and PS to have less stable zwitterion structures (the form they adopt in biological bilayers) than nonzwitterionic structures (where the phosphoester group and amine groups are both neutral). This result echoes many studies of positively and negatively charged gas-phase amino acids and small peptides^{24,25,50–52}, which indicate that nonzwitterionic forms of these ions are often stable in gas phase complexes, despite the fact that all common amino acids are zwitterionic in aqueous solution at neutral pH. Furthermore, the relative ordering of aqueous solution phase pK_a values of the phosphoester groups on dimyristoyl (DM-) phospholipids is DMPC (pK_a < 1) > DMPE (1.7) > DMPS (2.6) > DMPG (2.9) > DMPhA (3.5), in order from most acidic to least acidic.⁵³Because DMPC has a quaternary ammonium group, the pK_a of DMPC here corresponds to the equilibrium between its positively charged forms and net neutral forms, whereas all of the other pK_a values correspond to equilibria between net neutral and negatively charged forms. The experimental and computational results presented here for the analogous gas-phase charge states indicate a somewhat different ordering from most acidic to least acidic (PC > PG ~ PS > PhA > PE). Likewise, in solution, the amine group of DMPS (pK_a 11.55) is slightly more basic than that of DMPE (11.25),⁵³whereas PE is found here to be more basic than PS in the gas phase. These different orderings are attributed to the different extents of intramolecular hydrogen bonding and self-solvation of charge possible for the gas-phase lipids studied here, owing to their disparate numbers and types of polar and hydrogen-bonding groups. Thus, relative gas-phase protonation energetics of phospholipids can be very different from those in aqueous solution.

As can be seen in Table 2, the computed and experimental values for the ΔG/ΔH of deprotonation of phosphoric acid differ somewhat, but excellent agreement is found between computed and experimental values for the GB/PA values of single amino acid molecules at the same level of theory. A similar discrepancy has been reported on previously by Moser and coworkers,⁵⁴who found that density functional theory methods tend to slightly underestimate GB/PA values of biomolecules, with the largest deviations from experiment occurring with phosphorus-containing compounds. Thus, computed results for the GB/PA and the ΔG/ΔH of deprotonation for the phospholipids reported here may be systematically lower than the true values by approximately 10 kJ/mol, though differences between them may be much more accurate. It also appears that computed ground state geometry of lysine is level of theory/basis set dependent

While experimentally determining relative gas-phase acidities of some of these phospholipids proved difficult, relative values for ΔH of deprotonation that we report are consistent with an earlier study by Blanksby and coworkers.³⁵In that publication, the relative ordering of gas-phase acidities of PS, PG, PhA, and PE is reported to be PS > PG > PhA > PE, which not only aligns with our computational results, but also our experimentally determined ordering of PS > PG > PhA. Although computed relative orderings in this publication are identical to our results, it should be noted that this earlier publication reports absolute ΔH of deprotonation for each lipid that are higher than our results by 13–41 kJ/mol. Here, the same level of theory and basis set were used as in this publication, but we modelled the entire phospholipid headgroup along with the glycerol backbone and attached methyl ester groups to include possible charge-solvation and hydrogen-bonding effects from the fatty acid sidechains (Supplementary Figures S16, S17, S18). These additional ester groups provide polar sites that contribute to charge stabilization in many of the structures studied here computationally, and likely account for much of this 13–41 kJ/mol discrepancy.

The experimental and computational results presented here explain to a large extent why some lipids readily dissociate from gas-phase phospholipid Nanodisc and detergent micelle complex ions and also which phospholipids tend to remove charge from these complexes upon dissociation (Figure 4). Due to their especially high GB, SM and PC lipids should often bear positive charge in positively-charged protein-lipid complexes that have been strongly activated, and they should furthermore carry positive charge away upon dissociation, akin to charge-reduction reagents. Moreover, in ions with densely packed lipids, such as Nanodiscs, these headgroups should facilitate dissociation by loss of charged lipids due to Coulomb repulsion. By contrast, all of the other lipid headgroups studied are less likely than PC and SM to bear positive charge in activated protein-lipid complexes in positive ion mode and to carry positive charge away upon dissociation. Lower-basicity lipids should also lend stability to Nanodiscs relative to SM and PC lipids due to lower Coulomb repulsion within the lipid bilayer. In the absence of lipid clustering effects, in negative ion mode, PS and PG are the most likely among the lipids studied to bear negative charge in intact ions and to dissociate as negative ions, whereas PC and SM should preferentially adopt a neutral form.

Figure 4:

Schematic illustration of lipoprotein Nanodisc CID experiments under moderate activation conditions under which positively charged Nanodiscs lose lipids, but negatively charged Nanodiscs do not²¹. Due to their especially high GB, lipids in positive ion mode are more likely to bear charge, repel other charge sites, and dissociate than are lipids in negative ion mode under similar activation conditions.

These results straightforwardly explain observations in several literature studies of native-like Nanodisc ions formed by nanoESI and dissociated using CID. Recent work by Hoi, Robinson, and Marty⁵ reveals that, in positive ion mode, single-lipid palmitoyloleoyl (PO-) phospholipids POPG and POPS Nanodiscs resist dissociation upon moderate activation, whereas POPC Nanodiscs more readily dissociate under the same conditions, with prominent loss of positively-charge POPC lipids. Similarly, in negative ion mode, POPC Nanodiscs were found to be more stable to dissociation than are POPG and POPS Nanodisc complexes⁵. Intuition based solely on solution-phase equilibrium data (i.e., pKa values) would incorrectly predict that POPS should be the most likely of these lipids to bear net positive charge in the gas phase and to destabilize gas-phase lipid bilayers in positively-charged Nanodiscs. Marty and co-workers also found that negative ion mode nanoESI tends to produce globally lower charge states for native-like Nanodiscs than does positive ion mode.²¹ This contrasts with expectations based on the Charge Residue Model^55,56 for native protein charging in electrospray ionization, for which the same charge states are expected in positive and negative ion mode, and indicates that factors other than ion size and shape, such as gas-phase basicity and acidity, are at play in determining charge states for these ions. A separate study from Marty and coworkers²¹ reveals that negative ion mode is more stabilizing for both POPC and POPG Nanodisc complexes than is positive ion mode. These results are consistent with lower Coulomb repulsion and lipid charging in the lipid bilayer in negative ion mode relative to positive ion mode, with the observed total charge on native-like Nanodiscs resulting from a combination of their size and shape and the gas-phase basicity/acidity of titratable groups in the lipids and scaffold proteins.

The high GBs of SM and PC compared to amino acids indicate that these lipids are likely to remove protons from protonated protein ions upon activation and dissociation. Mechanistically, this is similar the behavior of charge-stripping reagents, such as imidazole derivatives, as described in a recent report²². The other lipids studied have lower intrinsic GBs than arginine, thus they should typically bind more strongly to protonated groups on protein ions than SM and PC and have a greater tendency to dissociate as neutral molecules upon activation. This mechanism is consistent with recent work from Laganowsky and coworkers⁶. In this publication, POPhA, POPG, POPS, and POPE were each separately titrated in solution into C8E4 detergent micelles containing transmembrane AmtB protein complexes, and the resulting micelle-bound complexes were studied with native MS and CID. All four of these lipids were found to dissociate as neutral species upon activation, consistent with the GBs reported here.

Based on the experimental and computational results reported here, the GB for several common phospholipid headgroups are similar to those of amino acids that are much less basic than arginine. For example, the GB for PG is within 2 kJ/mol of that for tryptophan.²⁷Previous work by Johnson and coworkers²³ demonstrates that the relative strength of a shared proton bond increases when the bases have comparable GB. Many proteins and protein complexes have a number of arginine residues exposed at their surface that exceeds their experimentally observed native charge state in nanoESI-MS experiments. Thus, without taking into account charge stabilization by adducts or salt bridges, one might assume that the majority of protons on the surface of most protonated, native-like protein ions reside at arginine residues, based on the very high intrinsic GB of this residue. By contrast, the above results indicate that, for example, PG can strongly adduct to exposed, protonated tryptophan (Trp) residues located on the outside of a gas-phase protein. Due to the strength of the resulting [PG·H+·Trp] shared-proton bond, interaction of PG with a Trp residue from a protein in the gas phase may in many cases stabilize charge site isomers of the protein that would not typically be stable in the absence of PG. We expect that the basicity and acidity values reported here, as well as the method that is used to obtain them, will be useful for modeling the relative energies of charge site isomers for proteins with lipid adducts, as has previously been done for protonated proteins using the concept of “apparent proton affinity” that combines Coulomb repulsion between charge sites and intrinsic proton affinities of the individual protonation sites.^29,57,58

It should be emphasized that predictions of ion behavior based solely on the gas-phase phospholipid thermodynamics described here may not accurately reflect the outcomes of some mass spectrometry experiments in which protein-lipid complexes are formed in solution are kinetically trapped using nanoESI. For example, in membrane protein analysis using phospholipid-doped detergents, such as those pioneered by Robinson and Laganowsky,^6,7,11,13,18 if the number of phospholipids in the micelles is kept lower than the number of lipid binding sites at the membrane protein surface, solution-phase stoichiometries and relative lipid abundances—and solution-phase binding thermodynamics derived from them^7,13—may be accurately determined from native MS experiments if the lipids are not activated to the point of rearrangement and/or dissociation from the complex. This can explain the apparent discrepancy between gas-phase phospholipid binding propensities discussed here and results from Laganowksy and coworkers,⁷who found that AmtB embedded in C8E4 micelles binds to POPS, POPG, POPE, tetraoleoyl-cardiolipin (TOCDL), and POPhA, with (solution-phase) ΔG within uncertainty of one another. As a corollary, it is recommended based on our results that the number of lipids doped into detergent micelles in experiments such as these be kept below the number of lipid binding sites at the membrane protein surface to prevent the possibility of perturbing apparent affinities after lipid activation, rearrangement to more favourable gas-phase binding sites, and subsequent dissociation (Figure 5).

Figure 5:

Gas-phase dissociation of a phospholipid-detergent micelle ion with an embedded transmembrane peptide/protein. Extensive activation of the micelle complex may lead to rearrangement of the phospholipids around the peptide/protein to a site that is more energetically favourable in the gas phase and is stabilized by the high GB of the phospholipid headgroup. Possibly even some phospholipids not initially associated directly with the peptide/protein surface in the condensed phase (blue headgroups) may bind to it after activation.

Conclusions

In this study, relative and/or absolute GB, PA, and ΔG/ΔH of deprotonation for six common physiological phospholipid types were determined by experiments using the (Conventional and Extended) Kinetic Method and were found to be in excellent agreement with values computed using ab initio methods. Based on these results, phospholipids with PhA, PC, PE, PG, PS, and SM headgroups are highly basic in the gas phase and have different chemical properties from lipids in the condensed phase. Notably, SM and PC have the highest reported intrinsic GB/PA values of any small biomolecule in the literature. Thus, the behavior of SM and PC in native lipoprotein MS experiments under gas-phase activation and dissociation can be similar to charge-reducing reagents in positive ion mode while the other headgroups studied have a greater tendency to dissociate as neutral molecules. Indeed, the phosphocholine headgroup common to both PC and SM, which is readily soluble in much higher concentrations than full PC and SM lipids, may prove a highly effective, new, charge-reducing reagent in the native MS arsenal. Computational results show that, in the gas phase, non-zwitterionic isomers of PE and PS are preferentially stabilized relative to condensed-phase-like zwitterionic isomers, and that the GB/PA of all the lipids studied here in the gas phase do not strictly correlate with solution pK_a values.

These results also help explain literature observations on the relative stabilities of lipid Nanodisc complexes in the gas phase when ionized by electrospray in either positive or negative ion modes as well as the relatively low charge states adopted in negative ion mode. Thus, these results provide an interesting exception to the Charge Residue Model in native mass spectrometry, illustrating that gas-phase acidity and basicity, in addition to ion size and shape, can play a role in the total charge adopted by biomolecular complex ions, especially those containing molecules other than protein. From the (de)protonation thermodynamics values determined here, we also conclude that it is advisable to keep the total number of phospholipids lower than the number of lipid binding sites at membrane protein surfaces in gas-phase activation- and dissociation-based experiments aimed at inferring condensed-phase lipid binding stoichiometries and thermodynamics. Alternatively, if one aims to characterize the tendency of a membrane protein complex to recruit various phospholipids beyond the first lipid annulus, it is advisable to infer this information from the entire content of the lipid-doped micelle or other membrane model. If, instead, the membrane protein-phospholipid complex is activated to the point where lipid rearrangement to more favorable gas-phase binding sites is likely, accurate relative condensed-phase lipid binding preferences could be challenging to infer from the resulting mass spectra. The results presented here thus help establish a framework for reliable native MS protocols aimed at characterizing lipid-protein interactions.