A step towards probing the non-annular belt of membrane proteins

Abstract

Integral membrane proteins are embedded in the biological membrane where they carry out numerous biological processes. Although lipids present in the membrane are crucial for membrane protein function, it remains difficult to characterize many lipid binding events to membrane proteins, such as those that form the annular belt. Here we use native mass spectrometry along with the charge reducing properties of trimethylamine N-oxide (TMAO) to characterize a large number of lipid binding events to the bacterial ammonia channel (AmtB). In the absence of TMAO, significant peak overlap between neighboring charge states is observed, resulting in erroneous abundances for different molecular species. With the addition of TMAO, the weighted average charge state (Z_avg) was decreased. In addition, the increased spacing between nearby charge states enabled a higher number of lipid binding species to be observed while minimizing mass spectral peak overlap. These conditions helped us to determine the equilibrium binding constants (K_d) for up to 16 lipid binding events. The binding constants for the first few lipid binding events display the highest affinity, whereas the binding constants for higher lipid binding events converge to a similar value. These findings suggest a transition from non-annular to annular lipid binding to AmtB.

Graphical Abstract

Introduction

Membrane protein-lipid interactions are often classified into three groups: non-annular lipids, annular lipids, and bulk lipids.^{1, 2} Non-annular or ‘co-factor’ lipids are described as tightly associating with membrane proteins. In contrast, annular lipids form an annular shell of lipids around the protein that can exchange with bulk lipids, which are freely diffusing in the membrane. It has been challenging to study the different types of membrane protein-lipid interactions due to the difficulty in that membrane proteins can not only bind a number of lipids, but lipids themselves can self-associate. Traditional methods like surface plasmon resonance (SPR) and isothermal titration calorimetry (ITC) suffer from the requirement of chemical perturbation (e.g., immobilization to the sensor surface), the need for large amounts of sample, and the inability to distinguish protein-lipid from lipid-lipid interactions. ^3–5 Bioluminescence resonance energy transfer (BRET) has been applied to study protein-lipid interactions, in which the membrane protein is expressed as a fusion to nanoluciferase along with a lipid modified with a fluorophore. ⁶ The major drawbacks of BRET and other techniques such as electron paramagnetic resonance, ⁷ are the use of non-natural lipids (e.g., fluorescently labeled or brominated), and as ensemble measurements, it is difficult to dissect each binding event, which is necessary to measure equilibrium binding constants accurately.

Native mass spectrometry (MS) is a powerful biophysical technique for studying bio-macromolecular assemblies and their interactions with molecules. ^8–12 Under carefully tuned conditions, ¹³ native MS preserves non-covalent interactions in the gas phase, ^{14, 15} and provides insight into individual ligand binding events and binding stoichiometry. For this reason, native MS can be utilized to elucidate the interactions between biomolecules and small molecules, such as for proteins interacting with lipids, drugs, nucleotides, and peptides. ^16–21 Native MS has also advanced to determine binding constants and thermodynamics of protein-lipid interactions, ¹⁴ and allostery between protein-lipid interactions and small molecules, such as other lipids^{22, 23}, and drugs. ^{24, 25}

Native MS of membrane proteins is generally performed with the protein encapsulated in detergent micelles. ^{26, 27} Removing detergent molecules while keeping the protein complex structure intact has been challenging because the harsh conditions required to strip detergents can, in some cases, perturb protein structure. ²⁸ Non-charge reducing detergents like dodecylmaltoside (DDM) can give rise to highly charged ions that make it difficult to preserve non-covalent interactions. ²⁸ Charge reduced protein complexes have reduced columbic repulsion, thereby aiding the preservation of native-like structures and non-covalent interactions, such as lipid binding. ²⁹ The discovery of charge reducing detergents, such as n-dodecyl-N,N-dimethylamine-N-oxide (LDAO), and tetraethylene glycol monooctyl ether (C₈E₄), have yielded comparatively lower charged ions that are less susceptible to unfolding and dissociating non-covalent interactions. ²⁸ Another charge-reducing approach is the addition of small molecules such as glycerol carbonate and imidazole, ³⁰ which has been shown to improve mass spectra of nanodiscs collected in negative mode. ^30–32 For membrane proteins, trimethylamine N-oxide (TMAO), a natural osmolyte and protein stabilizer, and polyamines, such as spermidine, have been found to reduce the weighted average charge state (Z_avg) significantly. The reduction in charge state leads to increased spacing between nearby charge states allowing for more ligand-bound states to be resolved. ^{33, 34}

Despite the advantages of charge reduction, ^{26, 33, 34} it is not well understood whether the high concentration often used to achieve significant charge reduction is compatible with lipid binding studies. Does the charge reducing molecule enhance or compete with lipid binding? Can useful equilibrium binding constants be determined? Herein, lipid binding events to the ammonia channel (AmtB) from Escherichia coli is characterized in the presence of the charge-reducing molecule, TMAO. Similar to previous studies, ³³ in the presence of TMAO, up to 18 lipid binding events to AmtB can be observed without overlapping mass spectral peaks from neighboring charge states. In the absence of TMAO, up to 10 lipid binding events could be resolved. The ability to resolve a higher number of lipid binding species enables the determination of the equilibrium dissociation constants (K_d), providing insight into non-annular and annular lipid binding to AmtB.

Materials and Methods

Protein Expression and Purification.

The ammonia channel (AmtB) from Escherichia coli was expressed and purified as previously described with slight modifications. ¹⁵ In brief, AmtB was expressed with an N-terminal HRV3C protease cleavable 10x His-tag and maltose binding protein (MBP) from pCDFDuet-1 (Novagen) in E.coli C43 (DE3) (Lucigen). A single colony was used to inoculate LB Miller (IBI Scientific) and was grown overnight at 37°C. The overnight culture was used to inoculate terrific broth (TB, IBI Scientific) and allowed to grow at 37°C until it reached an OD₆₀₀ of 0.6. Protein expression was induced with the addition of IPTG to a final concentration of 0.5 mM and grown for 18 h at 20°C. Cells were harvested by centrifugation at 5,000 xg for 10 min, resuspended in lysis buffer (150 mM sodium chloride, 50 mM Tris, 5mM 2-mercaptoethanol (BME); pH 7.4 at room temperature), and lysed using a Microfluidics M-110P microfluidizer operating at 25,000 PSI. The lysate was clarified through centrifugation (25 min at 20,000 xg at 4°C). The supernatant was centrifuged (2 h at 100,000 xg at 4°C) to pellet the membrane. The membrane was resuspended in a buffer (100 mM sodium chloride, 20mM Tris, 20% glycerol; pH 7.4 at room temperature) and extracted overnight with 5% octyl glucoside (OG) at 4°C. The extracted protein mixture was centrifuged at 40,000 xg for 20 mins. The supernatant was filtered with a 0.45-micron syringe filter (Pall Corp.) and loaded onto a HisTrap HP 5 mL column (Cytiva) pre-equilibrated with NHA-DDM buffer (200mM sodium chloride, 20mM Tris, 20mM imidazole, 10% glycerol, 0.025% DDM; pH 7.4 at room temperature). After loading, the column was initially washed with 20 mL of NHA-DDM buffer supplemented with 2% OG followed by 40 mL of NHA-DDM until a steady baseline was reached. AmtB was eluted with NHB-DDM buffer (100mM sodium chloride, 20mM Tris, 500mM imidazole, 10% glycerol, 0.025% DDM; pH 7.4 at room temperature). The eluted protein was diluted with MBP-loading buffer (100mM sodium chloride, 20mM Tris, 10% glycerol, 0.025% DDM; pH 7.4 at room temperature) and loaded onto a 5 mL MBPTrap HP column (Cytiva) equilibrated with the MBP-loading buffer. The protein was eluted with MBP-elution buffer (100mM sodium chloride, 20mM Tris, 10 mM maltose, 10% glycerol, 0.025% DDM; pH 7.4 at room temperature). The protein was pooled, and HRV3C was added in a 100:1 protein-to-protease ratio, and the mixture was incubated overnight at 4°C. The cleaved protein was loaded onto a HisTrap HP column equilibrated in NHA-DDM buffer, and the flow-through containing the tag-less protein was collected and concentrated using a 100 kDa MWCO concentrator (MilliporeSigma). Concentrated protein was loaded onto a Superdex 200 Increase 10/300 GL column (GE Healthcare) equilibrated with GF buffer (100mM sodium chloride, 20mM Tris, 10% glycerol, 0.5% C₈E₄; pH 7.4 at room temperature). The peak fractions containing C₈E₄-solubilized AmtB were pooled, aliquoted into 50 μL shots, flash-frozen in liquid nitrogen, and stored at −80°C.

Preparation of samples.

Trimethylamine N-oxide (TMAO) was purchased from Cayman Chemical Company. Protein was buffer exchanged into mass spectrometry buffer (200 mM ammonium acetate, 0.5% C₈E₄; pH 7.4) with a centrifugal desalting column (Micro Bio-Spin 6 Columns, Bio-Rad). Phospholipids were purchased from Avanti Polar lipids and were prepared as previously described. ^{15, 28, 35} In brief, dried lipid films were resuspended in deionized water to the desired concentration (2-5 mM) and stored at −20°C until further application. The concentrations were confirmed by phosphorus analysis. ³⁶ Lipids, TMAO, and buffer exchanged protein were prepared in mass spectrometry buffer and incubated for 2-5 minutes. Previous work¹⁴ on AmtB showed lipid binding reaches equilibrium in a few minutes and no difference in lipid binding profiles was observed after overnight incubation. The optimized final concentrations of AmtB and TMAO were 2 μM and 120 mM, respectively. For preparing samples without TMAO, the mixing is done in the same way, except the same volume of mass spectrometry buffer is used in place of TMAO.

Native Mass Spectrometry (MS) Analysis.

Protein, phospholipids, and TMAO were mixed and loaded into a gold-coated nano-electrospray ionization emitter as previously described. ²⁶ MS was performed on an Exactive Plus EMR Orbitrap Mass Spectrometer (Thermo Scientific) and samples analyzed at room temperature. The instrument settings with TMAO were as follows: Capillary temperature was set at 300°C and capillary voltage at 1.60 kV. In-source collision-induced dissociation (CID) and trap collision voltage (CE) were set at 30 V and 70 V. Trapping gas pressure was set to 6.0. Source DC offset was set to 38 V, injection flatpole DC to 8 V, inter flatpole lens to 5 V, bent flatpole DC to 16 V, and transfer multipole DC to 3 V. The same without TMAO were as follows: Capillary temperature 300°C; capillary voltage 1.50 kV; In-source collision-induced dissociation 65 V; trap collision voltage 100 V; Trapping gas pressure 6.0; Source DC offset 40 V; injection flatpole DC 8 V; inter flatpole lens 4 V; bent flatpole DC 3 V; and transfer multipole DC 3 V.

Native MS data analysis.

Acquired native MS spectra were processed and deconvoluted using UniDec with the settings: m/z range 6500 to 11000 without TMAO and 7800 to 16000 with TMAO, charge range 8–24, mass sampling every 1 Da, and peak FWHM 0.85. ³⁷ An in-house Python script was used to streamline the species assignment and to calculate the equilibrium dissociation constants (K_d) for each individual protein-ligand interaction, ¹⁴ which can be found at https://github.com/LaganowskyLab/Laganowsky_Lab_Code. The weighted average charge states (Z_avg) were obtained using UniDec.

Results

To study how the spacing between neighboring charge states could impact the accuracy of deconvolution, we first examined cardiolipin binding to AmtB in the absence of TMAO to determine the number of lipids that could be resolved prior to observing mass spectral peak overlap. The weighted average charge state (Z_avg) for AmtB in the C₈E₄ detergent is 16.2 ± 0.1, which is lower compared to AmtB solubilized in dodecylmaltoside (DDM) (Z_avg of 21.9). ³⁸ AmtB was then titrated with 18:1 cardiolipin (TOCDL, 1,1´,2,2´-tetraoleoyl-cardiolipin) in the range of 1-50 μM (Figure 1). At a lower lipid concentration (15 μM), up to four TOCDL binding events and no overlap of neighboring charge states were observed (Figure 1A). In contrast, at concentrations greater than 30 μM, significant peak overlap between different charge states becomes evident (Figures 1B and C). For example, the peak corresponding to the 9 lipids bound species from the 16⁺ charge state overlaps with the 3 lipids bound species from the 15⁺ charge state (Figure 1D). Mass spectral peak overlap presents challenges for the accurate deconvolution of mass spectra, which could lead to erroneous results when calculating binding parameters for protein-ligand interactions. More specifically, the abundance for the same molecular species varies depending on whether all charge states or only a subset of charge states were used for deconvolution. This prompted us to compare the abundances from deconvolution of all charges or those clearly separated using UniDec. ³⁷ This phenomenon is more pronounced when processing with only the higher charge states where the spectral overlap is more frequent (Figure S1). The differences in the abundance of molecular species makes it difficult to confidently determine binding parameters.

Figure 1. Mass spectral peak overlap for different charge states of AmtB bound to TOCDL.

A-C) Mass spectra of 2 μM AmtB in C₈E₄ titrated with A) 15 μM, B) 35 μM, C) 50 μM TOCDL. D) Zoom of the 16⁺ and 15⁺ charge states from the mass spectrum containing 50 μM TOCDL. The peak for the nine-lipid bound species from the 16⁺ charge state and the 3-lipid bound species from the 15⁺ charge state overlap.

We next investigated whether lipid binding to AmtB would be affected by the addition of TMAO. The binding of AmtB in the presence of varying concentrations of TMAO was investigated. The native mass spectrum of AmtB mixed with 40 μM TOCDL showed up to seven binding events with a Z_avg of 16.2 (Figures 2A and D). The addition of either 60 mM or 120 mM TMAO reduced Z_avg to 14.1 and 12.8, respectively (Figure 2B and C). The concentration-dependent charge reduction by TMAO is consistent with our previous report. ³³ The charge-reduced mass spectra reduced congestion and the appearance of charge states for AmtB bound to TOCDL with reduced or no peak overlap. Notably, the presence of TMAO did not change the total number and abundance of TOCDL bound to AmtB (Figure 2D). Mass spectra of a AmtB and TOCDL mixture collected under different instrument settings show the mole fraction of TOCDL bound the channel are similar (Figures 2E). At higher trap energies we observe AmtB monomers (Figure S2). Similar measurements performed In the presence of TMAO show the mole fraction of TOCDL bound to AmtB is consistent and no monomers are observed, even at the highest energy (Figure 2F, S3 and S4).

Figure 2. The effect of varied TMAO concentration on TOCDL binding to AmtB.

A-C) Mass spectra of 2 μM AmtB mixed with 40 μM TOCDL in the presence of A) 0 mM, B) 60 mM, and C) 120 mM TMAO. D) Mole fractions determined from deconvolution of mass spectra shown in A and C. E-F) Plot of mole fractions of of AmtB(TOCDL)_0-7 recorded under different trap energies and in the presence E) 0 mM, F) 120 mM TMAO. The final concentration of AmtB and TOCDL was 2 μM and 10 μM, respectively. Reported are the mean and standard deviation (n=3).

Analogous experiments were conducted for 1-palmitoyl-2-oleyl phosphatidic acid (POPA, 16:0-18:1) and POPE (1-palmitoyl-2-oleyl phosphatidylethanolamine, 16:0-18:1) binding to AmtB. In the presence of 160 μM POPA, up to 15 lipid binding events were observed without the addition of TMAO (Figure 3 and S5). In some cases, different POPA bound states overlapped, such as for AmtB with 120 mM POPA and 120 μM TMAO (Figure S6). The addition of 60 mM or 120 mM TMAO resulted in a charge reduction comparable to that observed for TOCDL (Figures S5B, C, E, and F). The total number of POPA bound to AmtB and the abundance of higher order POPA-bound states was enhanced, and the reduced charge states by TMAO resulted in lipid-bound states being completely resolved. A comparison of the mole fraction data showed the values were statistically indistinguishable (Figures S6B and C), which is in contrast to data collected in the absence of TMAO (Figure S1). Mass spectra of a AmtB and POPE mixture recorded under different instrument settings show dissociation of the bound lipid beginning at trap energy of 110 eV (Figures 3C and S7). However, in the presence of TMAO the mole fraction of POPE bound to AmtB doesn’t significantly change (Figure 3D and S8–S9). This result highlights the benefit of charge-reduction to preserve non-covalent interactions.

Figure 3. Charge-reduction preserves AmtB-POPE interactions.

A-B) Mass spectra of 2 μM AmtB mixed with 15 μM POPE in the presence of A) 0 mM and B) 120 mM TMAO. C-D) Plot of mole fractions of of AmtB(POPE)_0-2 recorded under different trap energies and in the presence E) 0 mM, F) 120 mM TMAO. Reported are the mean and standard deviation (n=3). Reported are the mean and standard deviation (n=3).

To investigate whether there is a transition from non-annular to annular lipid binding to AmtB, lipid titrations were performed for four different lipids that are found in E.coli: POPA, TOCDL, POPE and POPG (1-palmitoyl-2-oleyl phosphatidylglycerol, 16:0-18:1). In the presence of 120 mM TMAO, the average charge state (Z_avg) for all spectra ranged from 11 to 12.5 (Figure 4A–C). For TOCDL, up to 11 binding events could be fully resolved compared to those in the absence of TMAO, where only up to 10 binding events could be resolved (Figure 1C). Similar to TOCDL, an enhancement in the total number of lipid binding events that could be resolved was also observed with POPA (18 bound), POPE (15 bound), and POPG (14 bound) (Figure S10–13). The equilibrium dissociation constants (K_d) for each individual lipid binding species were computed (Figure 4D–E, S14, and Table S1). Comparison of the K_d values from deconvoluting all or only lower charge states gave statistically indistinguishable results (Figure S15). For all lipids, a stepwise increase of K_d values between each subsequent binding event to AmtB was observed (Figure 4D–E and Table S1). Among the four lipids examined, TOCDL had the shallowest K_d increase for all the binding events (Figure S16D). For the other lipids, a steeper K_d increase was observed among the first 9 binding events, followed by a tapering off of the K_d values for higher lipid bound states (Figure S16A–C). In all cases, the binding constants plateaued for higher lipid binding events.

Figure 4. Equilibrium binding constants for AmtB-lipid interactions in the presence of TMAO.

A-C) Native Mass spectra of 2 μM AmtB with 120 mM TMAO and A) 10 μM B) 30 μM, C) 50 μM TOCDL. D) Plot of mole fraction data (dots) for AmtB(TOCDL)_0-11 determined from a titration series of TOCDL and subsequent fit of a sequential lipid binding model (lines). E) Equilibrium dissociation constants for AmtB binding POPA, POPE, POPG, and TOCDL. Reported are the mean and standard deviation (n=3).

We next compared the K_d values determined in the presence and absence of TMAO (ΔK_d) (Figure 5). K_ds for the first three lipid-binding events for all four lipids were reported in our previous study. ¹⁴ For the most part, there are similarities between the two datasets, with some notable exceptions. Binding constants for TOCDL_1-3 to AmtB are statistically indistinguishable. POPE and POPG have similar values for binding the first two, with the third binding event being weaker in the presence of TMAO. For POPA, the first binding event is weaker, whereas the second and third are enhanced in the presence of TMAO. In short, in the presence of high concentrations of TMAO, lipid binding to AmtB is largely similar and, in some cases, promotes lipid binding.

Figure 5. Comparison of the first three lipid equilibrium binding constants to AmtB in the presence and absence of TMAO.

The equilibrium dissociation constants (K_d) collected in the absence of TMAO in C₈E₄ are from a previous study.¹⁴ The calculated K_d differences (ΔK_d) between the two conditions (no TMAO minus with TMAO).

Discussion

TMAO is a well-known protein stabilizer used to counteract the protein destabilizing effects of pressure and urea. ³⁹ Recently, TMAO has also been found to reduce the charge on membrane proteins, which is beneficial for preserving non-covalent interactions and native-like structure. ³³ The addition of TMAO aids in producing lower charged ions which leads to better separation of lipid-bound states. This result will be beneficial for cases where significant overlap of different charge states of lipid bound states becomes problematic. For example, deconvolution of data collected in the absence of TMAO can skew abundances of species that are dependent on the selection of charge states (Figure S1). The presence of TMAO mitigates peak overlap in part by increasing the spacing between charge states. Another contributing factor is the increased number of charge states that minimizes the impact of overlapping peaks in the deconvolution process by producing more signals that do not overlap. Therefore, TMAO can play an important role in resolving higher lipid-bound states of membrane proteins.

As noted above, membrane protein-lipid interactions are often classified into different groups. The range of K_d values determined here provide insight into protein-lipid interactions. The first few binding events display high affinity toward AmtB, implying these lipids bind specifically. However, after binding 8-9 lipids, the K_d values for subsequent lipid binding events follow a different trendline. This may suggest a transition from non-annular to annular lipid binding to AmtB. It would be anticipated that annular lipid binding events would display weaker affinity than non-annular but higher than bulk lipids. Although the affinity is weak, it is still sufficient for lipids to interact with AmtB to form a seal or impermeable bilayer. However, it is difficult to rule out that the higher lipid binding events correspond to bulk or a mixture of bulk and annular lipids. It is estimated from molecular dynamics simulations of AmtB in POPC bilayers²⁸ that ~90 lipids makeup the annular lipid belt. This implies that binding up to 18 lipids likely represents those of non-annular and annular lipids. Moreover, we have previously shown that lipid binding to AmtB can allosterically modulate the binding of other lipids. ¹⁵ While we focus here on binding one lipid type, TMAO and other charge reducing molecules may prove useful for investigating a higher number of heterogenous lipids binding events to AmtB.

Another approach to access higher lipid bound states of membrane proteins is the use of membrane proteins embedded nanodiscs. The majority of studies have largely been limited to one lipid type but have shown the ability to eject protein complexes that retain different lipid bound states. ^{40, 41} It remains unclear if the collisional activation to eject membrane protein-lipid complexes disturbs or alters lipid binding. Mass spectra of membrane proteins in nanodiscs are often complicated and, due to the presence of a distribution of lipids (ranging from ~75-225) per disc, suffer from overlap of mass spectral peaks of different charge states and number of lipids. ⁴² Charge modulating molecules, such as glycerol carbonate, have led to improved mass spectra of protein-nanodiscs. ³¹ Peak overlap can also complicate studying the allosteric effect of lipids binding to membrane proteins. ⁴⁰ A recent study has looked into nanodiscs containing mixed lipids but has been limited to those that are similar in mass and up to ~125 different lipids. ⁴³ An alternative approach has been developed where lipid exchange is monitored between empty and membrane protein-loaded nanodiscs. ⁴⁴ Results from these studies have agreed with studies of the same protein in detergent.

In closing, a better understanding of protein-lipid interactions is crucial to a better understanding of the lipid belt that is composed of a mixture of annular and non-annular lipids. The characterization of up to 18 lipids binding to AmtB represents a step forward in our basic understanding of protein-lipid interactions. However, there remains open challenges regarding characterizing even larger numbers of lipid binding events, to the point of forming a complete annulus, and exploring the impact of acyl chain composition. This challenge is augmented when considering mixtures of lipid types and how these may alter the formation of the lipid belt.