Phosphatidylinositol Interactions with the SARS-CoV-2 Envelope Protein Investigated by Lipid ¹³C Labeling and Solid-State NMR

Abstract

Membrane protein structure and function are intimately influenced by the surrounding lipids. Solid-state NMR spectroscopy is an important approach for investigating site-specific protein-lipid interactions under physiological conditions. To observe protein-lipid contacts with high spectral sensitivity and lipid selectivity, here we describe an efficient protocol for expressing and purifying ¹³C-labeled phospholipids from yeast. We focused on phosphatidylinositol (PI), an important lipid involved in cellular signaling and membrane trafficking. Using ¹³C-labeled PI, we investigated its interaction with the SARS-CoV-2 envelope protein E. ¹³C chemical shifts, T₁ relaxation times and cross-polarization buildup times indicate that the E transmembrane domain (ETM) rigidified the inositol headgroup and the acyl chains without causing significant chemical shift perturbations, indicating that ETM-PI interaction is weak. Despite this weak interaction, protein-lipid cross peaks are observed in two-dimensional ¹³C-¹³C correlation spectra, indicating that a sub-population of PI lipids has specific interactions with the protein. These protein-bound annular PI lipids interact with Thr, Asn and Ser residues at the N- and C-terminal ends of the transmembrane helix, likely through hydrogen bonding and other electrostatic interactions. These results provide direct evidence that the SARS E protein interact with anionic PI lipids, and this interaction may modulate cation conduction by this pathogenic viroporin.

Graphical Abstract

Introduction

The structure and function of membrane proteins are intimately dependent on the surrounding lipids^1–3. Numerous experimental and computational studies have demonstrated the functional importance of protein-lipid interactions. For example, nonlamellar lipids such as phosphatidylethanolamine interact with viral fusion proteins to induce membrane curvature^4–5. Cholesterol modulates the conformational equilibria of potassium channels⁶, facilitates membrane scission by the influenza M2 protein^7–9, and drives clustering of HIV gp41 trimers^10–11. Negatively charged lipids such as phosphatidylglycerol and phosphatidylserine induce domain formation by cationic membrane peptides in bacterial membranes^12–13, affect the structure of antibiotic peptides¹⁴, are essential for the function of blood-clotting proteins^15–16, and regulate the activities of potassium channels¹⁷ and virus ion channels¹⁸.

While many biophysical techniques have been used to study protein-lipid interactions, solid-state NMR spectroscopy provides one of the most site-specific methods for studying these interactions in hydrated lipid membranes near physiological condition¹⁹. Solid-state NMR experiments can observe protein conformational and dynamical changes induced by the lipid composition of the membrane²⁰ and lipid chain disorder through ²H quadrupolar couplings²¹. Solid-state NMR can also directly probe protein-lipid contacts through cross-peak intensities in multidimensional correlation spectra²² or quantitative distance experiments⁸. A simple two-dimensional (2D) NMR approach is to correlate lipid chain ¹H chemical shifts with protein ¹³C signals²³, which provides information about membrane protein insertion depths²⁴. However, the similarity of acyl chain ¹H chemical shifts among different lipids limits the ability of this technique to identify protein interactions with a specific lipid species in multi-component membranes. To overcome this limitation, ¹³C labeling of individual lipid species is a promising approach. By incorporating one ¹³C-labeled lipid component at a time into a complex membrane, one can conduct 2D ¹³C-¹³C correlation experiments to observe protein contacts with the ¹³C-labeled lipid of interest. This approach has been successfully demonstrated for cholesterol^{6, 9, 25} using a mutant strain of the yeast S. cerevisiae that produces cholesterol instead of ergosterol²⁶. More recently, ¹³C, ¹⁵N-labeling of phospholipids was also introduced²⁷. By extracting and purifying phosphatidylethanolamine (PE), phosphatidylglycerol (PG), and cardiolipin from bacteria and phosphatidylcholine from yeast, 2D NMR spectra correlating the signals of labeled lipids in a complex membrane were obtained.

In this study, we describe ¹³C labeling of phosphatidylinositol (PI) to investigate its interaction with a viroporin, the SARS-CoV envelope protein E^28–29. PI is a negatively charged phospholipid with a myo-inositol headgroup (Fig. 1a), which adopts a tilted orientation with respect to the membrane surface³⁰. PI is synthesized in the endoplasmic reticulum (ER) and distributed in many intracellular membranes³¹. Phosphatidylinositol phosphates (PIPs), also called phosphoinositides, play essential roles in receptor-mediated signaling, ion channel regulation, and membrane trafficking³². We recently discovered that the transmembrane peptide of SARS-CoV E (ETM), a pentameric cation-conducting viroporin, is clustered in PI-containing lipid membranes³³. ¹⁹F spin-counting solid-state NMR experiments^34–35 on singly fluorinated ETM showed that ETM assembles into pentameric channels, as evidenced by equilibrium intensities of ~0.20, consistent with previous biochemical data^36–37. This result was found for ETM bound to model DMPC:DMPG membranes as well as mixed POPC, POPE and POPS membranes. However, when PI is included in the membrane to better mimic the ER membrane in which E is localized²⁸, lower equilibrium values of 0.14–0.17 were measured, strongly suggesting that multiple ETM pentamers are closely packed in space in the presence of PI lipids³³.

Figure 1.

1D ¹³C spectra of ¹³C-labeled PI in POPC:POPE:PI (2:1:1) membranes without (black) and with (blue) ETM. The spectra were measured under 10.5 kHz MAS at 268 K. (a) PI chemical structure and nomenclature. (b) Mobile-selective ¹³C INEPT spectra. (c) Rigid-selective ¹³C CP spectra. (d) ¹³C DP spectra. For all spectra, peak intensities in the 80–50 ppm range were 4-fold amplified relative to the 40–10 ppm range for clarity. The inositol ¹³C linewidths are much narrower in the INEPT spectra than the CP spectra, indicating that two pools of PI exist in the membrane, one with highly mobile inositol headgroups and the other with more immobilized headgroups. ETM binding did not cause large chemical shift perturbations to PI.

The hypothesis that negatively charged PI may preferentially interact with ETM is consistent with single-channel conductance data that show that the SARS E channel activity is sensitive to the anionic lipid content of the membrane^38–39. Negatively charged phosphatidylserine (PS) increased the cation selectivity and unitary conductance of the protein. Despite these interesting observations, the interaction between PI and E has not been directly studied. Here we tailor a yeast-based lipid ¹³C labeling protocol to label PI and use the resulting ¹³C-PI to study the conformational dynamics and interaction of this lipid with SARS ETM using solid-state NMR. We show that ETM binding caused limited chemical shift perturbations to PI and slightly rigidified the lipid. Despite the weak interaction, we observed clear PI-ETM cross peaks in 2D ¹³C-¹³C correlation spectra, indicating that a sub-population of this anionic lipid preferentially interacts with the protein, especially with the polar residues in the N- and C-terminal regions of the transmembrane helix.

Materials and Methods

Yeast growth in ¹³C-labeled media

To produce ¹³C-labeled PI, we used Saccharomyces cerevisiae strain RH6829, which was engineered to synthesize cholesterol instead of ergosterol²⁶. Yeast cells were recovered from −80°C glycerol stocks, transferred to yeast peptone dextrose (YPD, Millipore) medium and incubated for three days, then plated onto 2% YPD agar (w/v) for two days. Isolated colonies were inoculated in YPD medium containing 50 μM ampicillin, 10 μM kanamycin, and 3 μM natamycin. 500 ml of precultures were grown at 30°C under 180 rpm shaking to an optical density at 600 nm (OD₆₀₀) of 4.2 over 20 hrs. Cells were harvested by centrifugation at 1,000 ×g for 30 min at 20°C and resuspended in 1.0 L of ¹³C-labeled media to an OD₆₀₀ of ~0.25 (Fig. S1a–c). The ¹³C-labeled media contained 7.0 g/L yeast nitrogen base (Thermo Fisher), antibiotics, 1.0 g/L of ¹³C-glucose or ¹³C-acetate, and 40 mg/L uracil, 100 mg/L leucine, 40 mg/L methionine, 40 mg/L histidine, 40 mg/L tryptophan, 40 mg/L phenylalanine, 200 mg/L threonine, 100 mg/L lysine and 40 mg/L adenine. The 1 L culture was grown at 30°C for 29 h for the glucose-based media and 39 h for acetate-based media to reach the stationary phase. Small-scale cultures (10 mL) were prepared in parallel to measure the OD₆₀₀ at various time intervals (Fig. S1b, c).

Extraction of ¹³C-labeled lipids

The ¹³C-labeled yeast culture was centrifuged at 4,000 ×g for 10 min to obtain cell pellets, which were resuspended in 10 mL of H₂O in a 1.0 L Erlenmeyer flask. MeOH and CHCl₃ were added to the suspension to a MeOH:CHCl₃:H₂O volume ratio of 3:2:1 to obtain a homogeneous mixture (Fig. S1a). The suspension was acidified with 20 mM acetic acid and sonicated for 10 min on ice at 40 kHz with a 50% duty cycle. The lysate was transferred to a separatory funnel, and 10 mL of H₂O was added to cause a ternary mixture: a clear lipid-containing organic phase at the bottom, a thin white emulsion in the middle, and a yellow aqueous phase at the top. The organic phase was collected, then chloroform was added to the funnel twice to further extract lipids. The emulsion containing precipitated proteins, cell debris and DNA were transferred to Pyrex glass vials and centrifuged at 1,000 ×g for 10 min. The resulting organic phase was collected while the solid white precipitate was washed again with CHCl₃ to extract residual lipids. All organic phases were pooled and rotor-evaporated at room temperature to concentrate the lipids.

Lipid purification by anion exchange and silica gel chromatography

To separate phospholipids from fatty acids, sterols and sphingolipids and to fractionate phospholipids, we used anion exchange chromatography⁴⁰ (Fig. S1a). A DEAE-cellulose resin in chloride form (BIOphoretics) was converted into acetate form to facilitate phosphate binding. About 20 g of resin was incubated in 100 mL of 0.5 M HCl for 30 min, rinsed with H₂O, then incubated in 100 mL of 0.5 M NaOH for another 30 min. The resin was rinsed again with H₂O and incubated in 100 mL of 1.0 M acetic acid (AcOH) for 2 hr. The resin was then washed with methanol and equilibrated in a 3:2:1 MeOH:CHCl₃:H₂O solution (solution A) before use.

Crude lipid extract was dissolved in solution A containing 10 mM ammonium bicarbonate to deprotonate the phospholipids. Here the bicarbonate concentration is specified with respect to the volume of water in solution A. The lipid-containing solution was loaded onto the equilibrated DEAE-cellulose column. Uncharged lipids were washed out with 3 column volumes (CVs) of fresh solution A, while phospholipids were eluted with 3 CVs of increasing concentrations (10–1000 mM) of bicarbonate in solution A.

The eluted lipid fractions were analyzed by thin layer chromatography (TLC) using glass plates coated with 60 Å-pore silica gels (Merck). The plates were developed in 43.5:43.5:10:3 MeOH:CHCl₃:H₂O:AcOH (solution B) at room temperature. The developed plates were incubated in an iodine chamber to identify sterols and fatty acids, then stained with molybdate to identify phospholipids. For the latter, the plates were sprayed with 1.3 % (m/v) ammonium molybdate tetrahydrate in 30% (v/v) sulfuric acid and heated at 80°C for 1–2 min until blue stains appeared. TLC data of phospholipid standards showed the relative positions of PI, PS, PC, PE and cholesterol bands (Fig. S1d). Among the fractions collected from the anion exchange column, the 50 mM fraction contained PI and cholesterol (Fig. S1e). This fraction was dried and washed with acetone to remove cholesterol, then the insoluble material containing mostly PI was dried and dissolved in solution B.

To further remove residual contaminants in the 50 mM fraction, we conducted silica gel chromatography using a silica gel column (60 Å, 230–450 mesh, 12 × 2.5 cm, Alfa Aesar) equilibrated in solution B (Fig. S1f). PI eluted at ~2.5 CVs. The fraction was pooled and rotor-evaporated at 20°C. The yield of purified ¹³C-labeled PI was ~2.0 mg from 1 L of yeast culture grown to an OD₆₀₀ of 1.0 in 1.0 g/L of ¹³C-labeled glucose media.

Expression and purification of isotopically labeled ETM

Isotopically labelled ETM was expressed and purified as reported before⁴¹. Briefly, a plasmid containing the ETM sequence (residues 8–38) with an N-terminal His₆-SUMO tag was transformed in E. coli BL21(DE3) competent cells (New England Biolabs). Cells were grown at 37°C in M9 media supplemented with 3.0 g/L carbon source and 0.5 g/L ¹⁵N-labeled ammonium chloride (NH₄Cl). The carbon sources were 2-¹³C-labeled glycerol for the 2-¹³C-ETM sample and unlabeled glucose and ¹⁵N-labeled NH₄Cl for the ¹⁵N-ETM sample. For the ¹³C, ²H, ¹⁵N (CDN)-ETM sample, M9 media was prepared in 99% D₂O and contained U-¹³C, ²H-labeled glucose and ¹⁵N-labeled NH₄Cl. Protein expression was induced with 0.5 mM IPTG and proceeded under shaking at 18°C for 16 hours.

The cells were harvested and resuspended in a 50 mM pH 8.0 Tris buffer containing 100 mM NaCl, 1.0 mM MgSO4, 0.5 mg/mL of lysozyme, 1% Triton X-100, and 500 U of Benzonase nuclease (Millipore). Cells were lysed by sonication on ice, followed by centrifugation at 17,000 ×g to remove cell debris. The supernatant was incubated with 2 mL Ni²⁺ Profinity IMAC resin (Bio-Rad). Contaminants were removed with 30 mM imidazole and His₆-SUMO-ETM was eluted with 250 mM imidazole in 50 mM Tris buffer at pH 8.0 containing 100 mM NaCl and 0.1% n-dodecyl-β-d-maltoside. The fusion tag was cleaved by adding Cth SUMO protease at 0.1 mg/mL⁴² for 2 h at 20°C, then ETM was purified by reverse-phase high-performance liquid chromatography using an Agilent C3 column and a linear gradient of 5–99% solvent B (9:1 acetonitrile/isopropanol) while solvent A was water. ETM eluted at 95% solvent B, as verified by MALDI-TOF mass spectrometry.

Preparation of membrane samples for solid-state NMR

All membrane samples used in this study contain POPC:POPE:PI at a molar ratio of 2:1:1. All ¹³C natural abundance lipids, including POPC, POPE and bovine liver PI, were purchased from Avanti Polar Lipids. For samples in 3.2 mm rotors for ¹³C magic-angle-spinning (MAS) NMR experiments, unlabeled POPC and POPE were mixed with yeast ¹³C-labeled PI. For samples in 1.3 mm rotors for ¹H-detected experiments under 55 kHz MAS, d₃₁-POPC and d₃₁-POPE were mixed with ¹³C-labeled PI to reduce the proton density of the membrane. In total, seven membrane samples were prepared for this study (Table S1). Samples 1–3 contain ¹³C-labeled PI and no ETM (sample 1), ¹⁵N-labeled ETM (sample 2), or 2-¹³C labeled ETM (sample 3). Sample 4 contains 2-¹³C labeled ETM and unlabeled phospholipids. Three samples in 1.3 mm rotors contain d₃₁-POPC, d₃₁-POPE and ¹³C-labeled PI: sample 5 has no protein, sample 6 contains 30 mM Ca²⁺ ions, and sample 7 contains ¹³C, ¹⁵N and ²H-labeled ETM.

Lipid stocks were prepared in 1:1 CHCl₃ : MeOH solution and mixed at the desired ratios. ETM was solubilized in TFE and added to the lipid solution to a protein monomer / total lipid molar ratio (P/L) of 1:15. Thus, PI is in 3.75-fold molar excess to the protein. Solvents were evaporated by nitrogen gas, then the protein-lipid film was lyophilized. The dry film was suspended in 30 mM Tris pH 7.5 buffer containing 30 mM NaCl and 1 mM EDTA or 30 mM CaCl₂ and 10 mM NaCl. The liposome suspension was subject to ten freeze-thaw cycles, then centrifuged at 180,000 ×g overnight to obtain a membrane pellet, which was slowly dried to a water content of ~40 wt% relative to the total mass of the pellet before packing into MAS rotors.

Solid-state and solution NMR experiments

Solid-state NMR experiments were conducted on a Bruker 600 MHz (14.1 T) spectrometer using a 3.2 mm ¹H/¹³C/¹⁵N probe for experiments under 10.5 kHz MAS and a 1.3 mm ¹H/¹³C/¹⁵N probe for experiments under 55 kHz MAS. Typical radiofrequency (rf) field strengths were 83.3 or 71.4 kHz for ¹H and 62.5 kHz for ¹³C. ¹³C chemical shifts were externally referenced to the adamantane CH₂ signal at 38.48 ppm on the TMS scale. ¹H chemical shifts were referenced internally to the POPC Hγ chemical shift at 3.264 ppm on the TMS scale. All temperatures reported in this study refer to sample temperatures, which were estimated from the measured water ¹H chemical shift⁴³. For experiments under 55 kHz MAS, the bearing temperature was 20–30°C lower than the sample temperature to account for the significant sample heating under this fast MAS rate.

¹³C NMR spectra were measured at sample temperatures of 258–303 K. ¹H-¹³C cross-polarization (CP) spectra were measured using CP contact times of 1.0–4.0 ms. 2D ¹³C-¹³C correlation spectra were measured using either ¹H-¹³C CP or direct polarization (DP), and ¹³C spin diffusion mixing times were 100–300 ms. ¹³C T₁ relaxation times were measured in two ways. The ¹³C T₁’s of all lipids were measured approximately quantitatively using an inversion-recovery pulse sequence after ¹³C direct polarization with a recycle delay of 3.0 s. For comparison, the ¹³C T₁’s of partially immobilized lipids were measured using a ¹³C CP experiment followed by a z-filter⁴⁴.

¹H chemical shifts of ¹³C-labeled PI were measured using ¹H-detected 2D ¹³C-¹H correlation experiments under 55 kHz MAS. A 2D refocused INEPT experiment probed dynamic PI lipids whereas a dipolar 2D CP-hCH experiment preferentially detected immobilized PI. Finally, a 2D ¹³C-¹H heteronuclear single-quantum correlation (HSQC) spectrum was measured for ¹³C-labeled PI in CDCl₃:d₈-2-propanol solution on an 800 MHz AVANCE III solution NMR spectrometer equipped with a TXI probe.

Modeling of ETM in lipid membranes

A molecular representation of ETM in PI bilayers was generated using CHARMM-GUI’s Membrane Builder assistant⁴⁵. The closed-state ETM structure (PDB: 7K3G) was inserted into a bilayer composed of POPI lipids randomly positioned around the protein. The bilayer consisted of 60 and 65 lipids in the top and bottom leaflets, respectively. Given the purpose for depicting the relative position of ETM across the membrane, no further energetic equilibration was undertaken.

Results

¹³C-labeling and purification of PI lipids from yeast

We chose S. cerevisiae to produce ¹³C-labeled PI because this yeast produces elevated quantities of PI that represent 10–30% of its total phospholipid pool⁴⁶. In addition, S. cerevisiae has simple nutrient requirements and fast growth rates⁴⁷, making it an attractive organism for cost-effective ¹³C labeling. We used the genetically engineered strain RH6829, which produces cholesterol instead of ergosterol by replacing ERG5 and ERG7 genes with the cholesterol-synthesizing DHCR7 and DHCR24 genes^{26, 47}. Thus, this yeast strain produces both ¹³C-labeled phospholipids and ¹³C-labeled cholesterol, the latter allowing future studies of protein-cholesterol interactions.

S. cerevisiae cells can utilize both glucose and acetate as the sole carbon source. We thus compared the growth yields using these two carbon sources. Yeast cultures grown on 1.0 g/L of glucose plateaued at an OD₆₀₀ of ~1.0 after ~35 hours. Doubling the glucose concentration increased the maximum OD₆₀₀ 1.5-fold, representing a lower cell density per gram of glucose. Yeast grown on acetate required 3 days to reach the stationary phase, during which bacterial and fungal contaminations occasionally occurred. We added 30 μM ampicillin and 10 μM kanamycin to prevent bacterial contamination and 3 μM of the ergosterol-targeting antifungal natamycin⁴⁸ to prevent fungal infection. With these precautions, 1.0 g/L ¹³C-acetate led to an OD₆₀₀ of ~0.6 after 40 hours; higher acetate concentrations decreased cell densities, indicating toxicity to cells⁴⁹ (Fig. S1b, c). In comparison, 1.0 g/L ¹³C-glucose led to an OD₆₀₀ of 1.0 after ~29 hours. Therefore, glucose leads to faster growth as well as higher biomass per gram of carbon source, making it the more efficient carbon source than acetate for lipid ¹³C labeling^{25–26, 50}.

We extracted ¹³C-labeled lipids using a methanol, chloroform and water solution, then separated them using anion-exchange chromatography with a DEAE-cellulose matrix (Fig. S1d). ¹³C-labeled PI and cholesterol were found in the 50 mM bicarbonate fraction (Fig. S1e). Cholesterol was extracted with acetone and other residual lipids were removed using silica-gel chromatography (Fig. S1f).

¹³C and ¹H NMR of PI dynamics in lipid membranes

We measured ¹³C NMR spectra of ¹³C-labeled PI in a POPC:POPE:PI (2:1:1) membrane to obtain PI chemical shifts and investigate its conformational dynamics in the membrane. One-dimensional (1D) ¹³C spectra show multiple ¹³C signals between 71 and 78 ppm, characteristic of sugar ¹³C chemical shifts (Fig. 1). These signals are well resolved from glycerol G3 and G1 ¹³C chemical shifts of 63–65 ppm and acyl chain ¹³C chemical shifts of 10–40 ppm⁵¹. To investigate PI dynamics, we measured ¹³C spectra using J-coupling-based ¹H-¹³C INEPT polarization transfer to preferentially detect highly mobile segments and dipolar-coupling-based cross-polarization (CP) to preferentially detect immobilized groups⁵². At 268 K, both INEPT and DP spectra show a distribution of linewidths, with the narrowest lines observed for ω, ω−1, and inositol carbons, consistent with the large-amplitude fast dynamics of these functional groups^{51, 53}. Two CH₂ peaks are observed at 30.7 ppm and 31.4 ppm in all three types of spectra, which persist at ambient temperature (vide infra). The small chemical shift difference is inconsistent with their assignment to all-trans and trans-gauche chains⁵⁴. Instead, because the ternary membrane used here contains natural abundance POPC and POPE in addition to ¹³C-labeled PI, and POPC and POPE have different order due to their distinct headgroup structures⁵⁵, we attribute these two CH₂ signals to slightly different conformations of PI lipids that are in contact with POPC versus POPE. Future experiments on binary membrane mixtures will be required to test this hypothesis.

Between INEPT and CP spectra, the glycerol backbone and middle of the chain carbons exhibit similar linewidths (Fig. 1b, c), indicating that these segments have homogeneous dynamics in the membrane. In contrast, the inositol and chain-end ω and ω−1 signals show much narrower linewidths in the INEPT spectrum than the CP spectrum: the former resolves ¹³C-¹³C J-splittings of ~40 Hz whereas the latter shows larger linewidths of 60–80 Hz. These differences indicate that two PI populations coexist in the membrane, with highly dynamic versus partially immobilized inositol headgroups and chain ends. A ¹³C direct-polarization (DP) spectrum that detects all PI lipids exhibits a superposition of narrow and broad peaks (Fig. 1d), confirming the coexistence of PI lipids with dynamic and partly immobilized headgroups and chain ends.

In addition to PI signals, we observed natural abundance POPC Cα, Cβ and Cγ signals. The intensity of the highly dynamic PC headgroup Cγ is much lower than the intensity of the similarly dynamic acyl chain ω, consistent with ¹³C enrichment of PI. Based on the relative intensities of these two peaks, we estimate a PI ¹³C labeling level of ~50%. Since the ¹³C-glucose yeast culture reached OD₆₀₀ ~1.0 from an unlabeled preculture at OD₆₀₀ 0.25, ~25% of cells are unlabeled in the culture, contributing to this partial ¹³C labeling of PI.

Two-dimensional (2D) ¹³C-¹³C correlation spectra allowed us to assign and verify inositol ¹³C chemical shifts based on connectivity patterns and prior literature⁵⁶ (Fig. 2a). Among the six inositol carbons, the phosphodiester i1 exhibits the most downfield (largest) ¹³C chemical shift (77.03 ppm at 293 K) whereas i6 has the most upfield (smallest) chemical shift (72.28 ppm) (Table S2). In addition to PI signals, we observed correlation signals of PS Cβ, Cα and Cγ (Fig. 2a), indicating that a small amount of ¹³C-labeled PS was co-purified from yeast.

Figure 2.

2D ¹³C-¹³C and ¹³C-¹H correlation spectra of ¹³C-labeled PI in the POPC:POPE:PI membrane. (a) 2D DP ¹³C-¹³C correlation spectrum measured with 100 ms ¹³C mixing under 55 kHz MAS. Chemical shift assignments are shown in red for ¹³C-PI and black for unlabeled PC and co-purified ¹³C-labeled PS. The spectrum was measured at a sample temperature of 268 K. (b) ¹H-detected 2D CP-hCH spectrum of ¹³C-labeled PI in the mixed membrane measured under 55 kHz MAS (black contours) at 276 K. This is overlaid with the ¹³C-¹H HSQC spectrum of PI in CDCl₃ : IPA solution (cyan contours).

In addition to ¹³C chemical shifts, we measured ¹H chemical shifts of PI using 2D ¹H-¹³C correlation experiments. The inositol ¹H signals resonate at 3.3–4.2 ppm, in good agreement with values measured in solution (Fig. 2b)⁵⁶. i4 and i6 exhibit small ¹H chemical shift differences between the membrane and solution, indicating that these two sites are sensitive to the solvent. To obtain further information about PI dynamics in the membrane, we compared mobile-selective 2D ¹H-¹³C refocused INEPT spectra and rigid-selective CP-hCH spectra at 275 K and 303 K (Fig. 3). At both temperatures, inositol i2 exhibits a peak in the INEPT spectrum but not in the CP-hCH spectrum, whereas the other inositol sites are observed in both spectra, indicating that i2 is more dynamic than the other inositol groups. Among the six inositol sites, i2 is the only one with an axial OH group whereas all other sites have equatorial OH groups, suggesting that this axial OH might facilitate local dynamics. At 303 K, we observed a second set of weak inositol signals in both CP-hCH and INEPT spectra (Fig. 3c, d), suggesting that a second PI conformation or alternative inositol isomers such as scyllo, epi- or allo-inositol become visible in the fluid membrane at high temperature⁶⁹.

Figure 3.

2D ¹³C-¹H correlation spectra of ¹³C-PI in d₃₁-POPC:d₃₁-POPE:PI membranes measured under 55 kHz MAS. (a) CP-hCH spectrum at 275 K. (b) INEPT spectrum at 275 K. (c) CP-hCH spectrum at 303 K. (d) INEPT spectrum at 303 K. The INEPT spectra suppressed most signals of the rigid glycerol backbone at both temperature, but exhibit the inositol i2 signal, which is absent in the CP-hCH spectrum, indicating that i2 is highly dynamic. A second set of PI peaks are observed at 303 K in both CP-hCH and INEPT spectra (green arrows and asterisks).

Because the PI headgroup is negatively charged and contains multiple hydroxyl groups, we hypothesized that PI’s conformational dynamics in the membrane should be sensitive to Ca²⁺ ions. To test this hypothesis, we measured variable-temperature ¹³C DP spectra with and without Ca²⁺. In the absence of Ca²⁺ and protein, the lipid chain carbons, particularly ω, ω−1, CH₂, and C3, exhibit downfield shifted peaks (larger chemical shifts) with decreasing temperature, consistent with increased populations of the trans conformation, which increases ¹³C chemical shifts through the trans-gauche effect^{54, 57}. Below 268 K, the DP spectra broaden severely (Fig. 4a) while residual INEPT intensities remain (Fig. S2), indicating that the membrane transitioned to an ordered phase that retains residual segmental dynamics of the acyl chains. The glycerol backbone and headgroup peaks are more severely broadened than the chain carbons, indicating that the headgroup and glycerol backbone are more immobilized. Upon addition of Ca²⁺, the inositol line broadening and the ω−1 chemical shift increase are observed at high temperature (275 K) (Fig. 4b), indicating that Ca²⁺ interaction with the negatively charged PI headgroup stabilized the liquid-ordered phase.

Figure 4.

Variable-temperature ¹³C DP spectra of ¹³C-labeled PI in the POPC:POPE:PI membrane. The glycerol backbone and inositol headgroup intensities in 60–80 ppm are scaled up 8-fold relative to the acyl chain intensities in 10–40 ppm. Sample temperature were estimated based on water ¹H chemical shifts. (a) ¹³C spectra of PI at 258–298 K, measured under 10.5 kHz MAS. Severe line broadening was observed at 258 K, indicating membrane transition to an ordered phase. (b) ¹³C spectra of Ca²⁺-containing membrane, measured under 55 kHz MAS at 275 and 303 K. Line broadening was observed already at 275 K, indicating that Ca²⁺ stabilized the ordered phase. (c) ¹³C spectra of ETM-containing membrane from 258 to 298 K, measured under 10.5 kHz MAS. Decreasing temperature shifted the ω−1 peak downfield in all three samples, but ETM and Ca²⁺ stabilized this downfield peak at higher temperature compared to the protein- and Ca²⁺-free membrane, indicating that ETM and Ca²⁺ ions slightly rigidified the PI acyl chain.

PI interactions with SARS ETM

To investigate PI interaction with ETM in the POPC:POPE:PI membrane, we first compared PI chemical shifts in the absence and presence of the protein. ¹³C DP spectra (Fig. 1) and 2D CP-hCH spectra (Fig. S3) indicate that ETM binding did not cause large chemical shift changes to PI. However, small (0.1–0.2 ppm) downfield shifts are observed for some of the lipid chain ¹³C signals, including C2, CH₂, ω−2 and ω−1, indicating that ETM slightly rigidified PI (Fig. 1c, d). Variable-temperature ¹³C spectra show that decreasing temperature increased the chemical shifts of the lipid chains (Fig. 4a, c). This effect is observed both without and with the protein. However, ω−1 exhibits a downfield-shifted peak at 268 K, which is absent from the protein-free spectrum at the same temperature, indicating that ETM stabilized the ordered conformation of the acyl chains for a sub-population of PI⁵⁷.

Since temperature-dependent chemical shifts indicate a slight rigidification of PI by ETM, we next investigated PI dynamics in the absence and presence of ETM in more detail using variable-contact ¹³C CP experiments (Fig. 5). ¹H-¹³C CP is mediated by dipolar couplings, thus mobile segments with motionally averaged dipolar couplings require longer CP contact times to reach maximum intensities than rigid segments. In the absence of ETM, most inositol carbons exhibit maximum CP intensities at contact times of 3–4 ms. In the acyl chains, the carbons adjacent to the C=C double bond (Cdb±1) also reached maximum CP intensity at 3.0 ms. ETM binding to the membrane accelerated the CP buildup of most PI segments: for example, the maximum-intensity contact times decreased to 1.0 ms for i4 and 2.0 ms for Cbd±1. Therefore, ETM immobilized the PI headgroup as well as acyl chains, giving rise to larger dipolar couplings that speed up the CP intensity buildup.

Figure 5.

¹³C CP buildup of PI in the POPC:POPE:PI membrane at 268 K in the absence and presence of ETM. (a) CP spectra of protein-free membrane as a function of contact time. (b) CP spectra of ETM-containing membrane as a function of contact time. (c, d) ¹³C CP buildup curves of PI in the protein-free membrane (c) and ETM-containing membrane (d). Protein binding accelerated the CP buildup, particularly for i4 and Cdb±1.

We next investigated nanosecond motions of PI in the absence and presence of ETM by measuring ¹³C T₁ relaxation times. We measured the average T₁’s of all PI lipids (Table S3) using an inversion recovery experiment following ¹³C direct polarization (DP) and the T₁’s of partially immobilized PI using a z-filter following ¹³C CP (Fig. 6, Fig. S4). Most inositol headgroup, glycerol backbone and acyl chain carbons exhibit longer T₁’s upon ETM binding. Temperature-dependent ¹³C T₁ measurements indicate that T₁ relaxation at 268 K is in the slow limit (Fig. S5a), thus the protein-induced increase of T₁ relaxation times indicate that ETM slowed down the motional rates of PI. The T₁ increase is found for all segments in the DP spectra and most segments in the CP spectra, but the extent of T₁ increase is larger for the inositol headgroup and top of the acyl chains than the chain ends (Fig. S5b, c), indicating that the N- and C-termini of ETM are chiefly responsible for slowing down the PI segmental motion. The acyl chain C2 shows a more pronounced T₁ increase in the CP-detected spectra than the DP spectra, indicating that within the partially immobilized population of PI, C2 is particularly slowed down by ETM. Interestingly, the ¹³C CP-detected inositol i4 signal is unique among all sites in exhibiting a shorter ¹³C T₁ upon protein binding, whereas DP-detected i4 displays a T₁ increase similar to other carbons. The i4 also exhibits the largest acceleration of CP buildup rate by ETM among all sites (Fig. 5c, d), indicating that i4 in the inositol ring is more rigidified than other sites by the protein. Since T₁ relaxation times are inversely proportional to the square of dipolar coupling, the reduction of the T₁ relaxation time of i4 upon protein binding can be attributed to the significantly increased order parameter of this site despite its slower motional rates. Thus, the partially immobilized PI lipids respond to protein binding in subtly different ways from average PI lipids in the membrane.

Figure 6.

¹³C T₁ relaxation times of PI in POPC:POPE:PI membranes without (black symbols and curves) and with ETM (red symbols and curves). All data were measured at 268 K under 10.5 kHz MAS. (a) 1D ¹³C DP spectrum of ETM-containing membrane with peak assignment. (b) ¹³C DP T₁ inversion recovery curves, reflecting the dynamics of all PI lipids. (c) ¹³C CP T₁ relaxation curves, preferentially detecting partially immobilized PI. ETM increased the T₁ values of all PI lipids observed in the DP spectra but caused more varied T₁ changes in the CP-detected partially immobilized PI.

The accelerated CP buildup, increased ¹³C T₁ relaxation times, and larger ¹³C chemical shifts at higher temperatures, all point to ordering of PI lipids by ETM. However, this structural ordering can in principle result from non-specific effects of the protein to the membrane rather than specific PI binding to ETM. To directly investigate whether PI is in molecular contact with ETM, we measured 2D ¹³C-¹³C correlation spectra using spin diffusion mixing times of 200–300 ms (Fig. 7). With ¹³C-¹³C dipolar coupling driving magnetization transfer, cross peak intensities provide information about the spatial proximity of two carbons. If PI is more than ~8 Å from the protein⁵⁸, or if the residence time of the lipid at a binding site is extremely short, then no intermolecular cross peaks would be observed. To reduce resonance overlap and unambiguously assign protein-lipid correlations, we combined skip-labeled 2-¹³C ETM with ¹³C-PI and additionally measured two control spectra, one on a sample containing ¹³C-PI but ¹³C natural abundance ETM and the other on a sample containing ¹³C-labeled ETM but unlabeled PI. Protein-lipid correlations should be present only in the doubly ¹³C-labeled spectrum but not in the two control spectra.

Figure 7.

2D ¹³C-¹³C correlation spectra indicate ETM-PI contact. (a) Spectrum of ¹³C-labeled PI and 2-¹³C-labeled ETM sample (red contours) measured with a combination of CP and DP. This is overlaid with the spectrum of ¹³C-labeled PI and ¹⁵N-labeled ETM sample (blue contours) measured using CP. PI-ETM cross peaks are observed in the former and are absent in the latter. (b) Spectrum of ¹³C-labeled PI and 2-¹³C-labeled ETM sample (red contours) overlaid with the spectrum of unlabeled PI and 2-¹³C labeled ETM sample measured with CP (black contours). Identified PI-ETM cross peaks are absent in the latter. ETM-PI correlations are observed from inositol carbons to Thr and Asn residues and from lipid chain carbons to Phe sidechains and C-terminal residues of ETM. All 2D spectra were measured under 10.5 kHz MAS at 268 K using ¹³C spin diffusion mixing times of 200 and 300 ms. (c) Model of lipid packing around the closed state of ETM (PDB: 7K3G), built in CHARMM-GUI without molecular dynamics simulations. This schematic should be taken only to mark the approximate depths of protein residues and lipid carbons. The measured protein-lipid correlations are indicated with dashed lines from the protein to one of the neighboring lipid molecules.

In the inositol headgroup and glycerol backbone region (60–75 ppm) of the 2D spectrum, we observed lipid cross peaks with T9 and/or T35, L12, N15 and/or I33, and S16 (Fig. 7a, 7b). In the lipid chain region (10–40 ppm) of the spectrum, we observed ω and ω−1 correlations with Phe, Val and T30 in the middle of the TM helix, and C2 and C3 correlations with I33, A36, L37. These cross peaks are absent from both control spectra, confirming that they result from protein-lipid contacts. The ω/ω−1 cross peaks are consistent with the positions of Phe, Val and T30 residues in the middle of the TM peptide and hence the hydrophobic interior of the lipid bilayer, whereas the C2/C3 cross peaks are consistent with the shallower insertion depths of I33, A36 and L37 (Fig. 7c). This depth consistency between the lipid segments and protein residues indicates that ETM-bound PI acyl chains are relatively ordered, without signs of chain upturns⁵⁹ that could move ω and ω−1 outside the middle of the bilayer. Among the PI-ETM cross peaks, the correlations of i4 (73.1 ppm) with T9 or T35 Cα (61.8 ppm) and Cβ (67.4 ppm) are especially clear. The interaction of PI with T9 is consistent with recent data that T9I mutation and PI removal from the membrane perturbed Glu8 sidechain Cδ chemical shifts similarly⁶⁰. At shorter ¹³C mixing times of 100 ms, these PI-ETM correlations are no longer observed (data not shown), indicating that the PI-ETM interaction is weak. We also measured a 1D NHHC spectrum using ¹³C-labeled PI and ¹⁵N-labeled ETM but did not detect any correlation signals (Fig. S6), further indicating that the PI-ETM interaction is weak.

Discussion

It is now well appreciated that lipids are key regulators of membrane protein structure and function¹. But atomic-level experimental information about protein–lipid interactions remain difficult to obtain because of the small size and dynamic nature of lipids under physiological conditions. Solid-state NMR spectroscopy is well suited for site-specific studies of protein-lipid interactions. To probe these interactions with sufficient spectral sensitivity and selectivity, lipid ¹³C labeling is a powerful approach and is increasingly adopted in recent years^{9–10, 25, 27, 54, 61}. The current study describes a cost-effective method for producing milligram quantities of ¹³C-labeled phosphatidylinositol from yeast, to investigate how PI interacts with the SARS-CoV-2 E protein in complex lipid membranes.

In the POPC:POPE:PI membrane, PI ¹³C and ¹H chemical shifts are similar to the values found in solution. ¹³C linewidths indicate that the inositol headgroup and acyl chain ends are dynamically heterogeneous in the membrane. Ca²⁺ binding rigidified PI (Fig. 4b, Fig. S3a), consistent with the favorable electrostatic attraction between the negatively charged PI headgroup and Ca²⁺ ions. However, these changes are small, indicating that Ca²⁺-PI interactions are weak. Importantly, ¹³C labeling allowed us to probe PI interaction with ETM. PI exhibited limited ¹³C and ¹H chemical shift changes upon ETM binding, indicating that the protein did not cause significant changes to the average conformation of the lipid. More noticeable changes were observed in PI dynamics, as shown by faster CP buildup rates (Fig. 5) and longer ¹³C T₁ relaxation times (Fig. 6). The former indicates increased ¹H-¹³C order parameters and reduced motional amplitudes whereas the latter indicates slower motional rates. These changes indicate that ETM rigidified PI, especially at the inositol headgroup and beginning of the acyl chains. Most importantly, 2D ¹³C-¹³C correlation spectra showed clear lipid-protein cross peaks (Fig. 7), proving that the PI headgroup interacts with polar residues in ETM, including Thr, Asn and Ser, at the N- and C-termini of the TM helix. The inositol-Thr interaction likely occurs via hydrogen bonding of the hydroxyl groups, while the cationic R38 sidechain at the C-terminus may interact with the anionic phosphate group of PI. The absence of NHHC correlations between ¹³C-PI and ¹⁵N-ETM (Fig. S6) indicates that PI mainly interacts with protein sidechains rather than the backbone. Integrating the ¹³C T₁ relaxation and 2D correlation data, we find that ETM affected the same segments in PI, including inositol i4, glycerol backbone, and top of the acyl chains (Fig. 8a, b).

Figure 8.

Schematic model of PI-ETM interactions obtained from the solid-state NMR data. (a) Sites in PI that show correlations with ETM. (b) Sites in PI that show the largest ¹³C T₁ changes upon ETM binding. (c) The observed PI-ETM correlations despite small chemical shift and dynamics changes of PI indicate that protein-bound, annular, PI lipids coexist with unbound PI. (d) Schematic model of PI interactions with polar residues in the N- and C-terminal regions of ETM to modulate pore hydration and ion conduction.

It is relevant to consider the reason for the lack of large chemical shift perturbations (CSPs) of either PI or ETM despite the presence of ETM-PI correlation peaks in 2D spectra. The lack of lipid CSPs can be explained by conformational exchange between bound and unbound PI. The membrane samples used here have molar ratios of 5 : 37.5 : 18.8 : 18.8 for the ETM monomer, POPC, POPE and PI. For pentameric helical bundles, the predicted number of first-shell lipids is about 20^{54, 62}. Thus, about a quarter of the 75 lipids lies in the first shell of the protein. Although we do not know the exact partitioning of PI versus POPC and POPE in the first shell, it is unlikely that all ~19 PI lipids constitutes the first shell at the exclusion of POPC and POPE. This is supported by the fact that dynamic PI signals in ¹³C INEPT spectra coexist with more rigid PI signals in ¹³C CP spectra, indicating that bound and unbound PI coexist in the membrane (Fig. 8c). The unbound PI lipids are expected to exchange with protein-bound PI through lateral diffusion. Assuming standard translational diffusion coefficients, the estimated exchange rate should exceed ¹³C chemical shift differences (~1 ppm, or 150 Hz) between bound and unbound PI, thus averaging the chemical shifts of most carbons in PI.

In contrast to chemical shifts, the 2D ¹³C-¹³C correlation experiment specifically detects protein-bound PI through distance-dependent ¹³C-¹³C dipolar couplings. At mixing times of 200–300 ms, the protein-lipid contact is insensitive to lipid exchange as long as a binding site is occupied for a sufficient amount of time. Taken together, the presence of ETM-PI correlation peaks in the 2D spectra, the coexistence of dynamic and rigid PI in ETM-containing membranes, and the limited lipid CSPs between membranes with and without the protein, support a model that ETM-bound, annular, PI lipids coexist and exchange with unbound PI (Fig. 8c). These annular lipids interact with ETM through hydrogen-bonding and other electrostatic interactions with polar residues at the N- and C-terminal regions of the protein (Fig. 8d). Because these polar segments are critical for channel hydration and ion conduction^{41, 60, 63}, the PI-ETM contacts observed here are likely involved in modulating the conformational equilibria of ETM for function.

The lack of measurable ETM CSPs by PI has a separate origin. Previous NMR studies of protein-lipid interactions similarly did not report any protein CSPs attributable to specific lipid binding, despite the fact that quantitative protein-lipid distances and qualitative cross peaks have been measured for a variety of systems, including Arg-phosphate complexation in antimicrobial peptides^64–65, cholesterol binding to influenza M2⁹ and HIV gp41¹⁰, lipid II binding to antibiotic peptides^{14, 61}, and anionic phospholipid binding to potassium channels^{17, 54}. We attribute the lack of protein CSPs attributable to specific lipid binding to the fact that removal of essential lipids change the membrane fluidity, charge and other physical properties, thus causing protein chemical shift changes that cannot be distinguished from CSPs due to specific lipid binding. A case in point is cholesterol, whose removal from the membrane increases the membrane fluidity and hence conformational dynamics of the embedded protein, so that any protein CSPs between cholesterol-free and cholesterol-containing membranes cannot be uniquely attributed to specific cholesterol binding. The antibiotic peptide nisin showed clear ¹H and ¹⁵N chemical shift changes in membranes without and with lipid II¹⁴; however, the peptide secondary structure changed significantly in lipid-II containing membranes, thus making it impossible to separately measure CSPs due to specific lipid binding. Given the pronounced influence of the membrane environment on the conformational dynamics of membrane proteins, protein chemical shifts are not convenient reporters of specific lipid binding. Instead, protein-lipid dipolar couplings are a reliable indicator of specific lipid binding, as a binding site needs to be sufficiently occupied on the NMR timescale to give measurable dipolar couplings.

One of our motivations for investigating ETM-PI interactions is to understand the mechanism for clustering of ETM pentamers in PI-containing lipid membranes. This clustering was detected in ¹⁹F spin diffusion NMR data, which showed that ETM formed isolated pentamers in POPC:POPE:POPS membranes but clustered pentamers in the POPC:POPE:POPS:PI membrane³³. Our current data indicate that a subset of PI binds ETM as annular lipids while the average interaction of PI with ETM is weak. It is unclear whether the amount of protein-bound PI lipids are sufficient to cause clustering of the pentameric helical bundles in the membrane. The molar concentrations of PS and PI in our previous study were 8% and 15%, respectively, whereas the current samples contain 25 mol% PI with a small amount of copurified PS lipids. Thus, the total molar concentrations of negatively charged phospholipids are similar between these studies. One experimental condition that differs is that the current study employs yeast PI, which are predominantly POPI and DOPI, whereas the previous study used bovine liver PI, which contains an sn-1 palmitoyl chain and an sn-2 arachidonic chain with four C=C double bonds. The lipid chain dynamics of PI thus likely differs between these two membranes. It remains to be seen whether the polyunsaturated sn-2 chain of natural PI increases the propensity of interaction with ETM to promote clustering.

Our data show that Ca²⁺ interacts with PI weakly, causing a small chemical shift perturbation of i1 (Fig. S3a) and stabilizing the ordered phase of the membrane (Fig. 4b)⁵². This weak interaction is consistent with electron paramagnetic resonance data and coarse-grain simulations of PC/PI membranes⁶⁶, which found that Ca²⁺ moderately ordered PI without inducing phase separation of the two lipids⁶⁷. In comparison, Ca²⁺ induces significant phase separation of PC/PS membranes and strongly coordinates PIPs⁶⁸. Therefore, Ca²⁺ interactions with anionic lipids are sensitive to the nature of the functional groups and lipid headgroup charge.

Conclusions

These data demonstrate an economical approach for purifying ¹³C-labeled phosphatidylinositol from S. cerevisiae. Building on previous approaches for labeling the more abundant PC, PE and PG lipids²⁷, we purified the scarce PI by anion exchange chromatography. Using ¹³C-labeled PI, we investigated the conformational dynamics of this lipid and its interactions with Ca²⁺ and the SARS E protein in the POPC:POPE:PI membrane. We found that Ca²⁺ and ETM caused small chemical shift perturbations to PI and slightly rigidified the lipid. Despite the small chemical shift and dynamics changes, 2D ¹³C-¹³C correlation spectra show unambiguous cross peaks between ETM and PI, indicating that a sub-population of PI interacts specifically with ETM. These interactions between the inositol headgroup and polar residues at the C- and N-terminal segments of the protein and between the lipid acyl chains and Phe and Val residues in the middle of the TM peptide. These results indicate that protein-bound annular PI lipids coexist with unbound PI in the membrane. These results provide an example of how unphosphorylated PI interacts with viral ion channels. Phospholipid and cholesterol ¹³C labeling has emerged as a powerful approach for studying lipid interactions with membrane proteins¹⁷. Increased future efforts in this direction should provide much needed experimental information and biophysical insight into how lipids regulate membrane protein structure and function.

Supplementary Material

Additional yeast lipid purification protocol and analytical data, additional NMR spectra, and tables of membrane samples, chemical shifts, and ¹³C T₁ relaxation times. This material is available free of charge via the Internet at http://pubs.acs.org.