Determining Cholesterol Binding to Membrane Proteins by Cholesterol ¹³C Labeling in Yeast and Dynamic Nuclear Polarization NMR

Abstract

We present a general strategy for determining the cholesterol-binding site of eukaryotic membrane proteins in native-like lipid membranes by NMR spectroscopy. The strategy combines yeast biosynthetic ¹³C enrichment of cholesterol with detection of protein-cholesterol ¹³C-¹³C cross peaks in 2D correlation NMR spectra under the dynamic nuclear polarization (DNP) condition. Low-temperature DNP not only allows high-sensitivity detection of weak protein-cholesterol cross peaks in 2D spectra but also immobilizes cholesterol and protein to enable intermolecular distance measurements. We demonstrate this approach on the influenza M2 protein, which utilizes cholesterol to conduct membrane scission in the last step of virus budding and release from the host cell plasma membrane. A ¹³C-¹³C double-quantum filter was employed to significantly simplify the 2D ¹³C-¹³C correlation spectra and facilitate the identification of protein-cholesterol cross peaks. A number of cross peaks between the M2 transmembrane residues’ sidechains and the cholesterol sterol group were detected, which complement recently measured protein contacts to the isooctyl tail of cholesterol to define an extended binding interface. These results provide atomic-level evidence of M2-cholesterol interaction to cause membrane curvature and scission, and the approach is generally applicable to other eukaryotic membrane proteins for understanding the influence of cholesterol on membrane protein function.

Graphical Abstract

Introduction

Cholesterol is an essential component of mammalian cell membranes ¹. It consists of four fused rings bonded to a polar hydroxyl group at one end and an isooctyl tail at the other end. In phospholipid bilayers, cholesterol is oriented with the long axis of the sterol rings approximately parallel to the bilayer normal; the polar OH moiety resides at the membrane-water interface whereas the aliphatic tail inserts into the bilayer center ^2,3. In mammalian membranes, cholesterol is present at 20–30 mol%, while in enveloped viruses cholesterol is enriched to ~45 mol% in the viral lipid envelope ^4,5. The rigid sterol plane of cholesterol renders phospholipid bilayers more viscous at high temperature but more dynamic at low temperature compared to cholesterol-free membranes, which has profound consequences to membrane physical properties such as viscosity, bilayer thickness, domain formation, and curvature stress ⁶.

Cholesterol is known to regulate a diverse range of membrane protein functions. For example, it modulates the conformational equilibria, stability, and ligand-binding affinities of G-protein coupled receptors to affect signal transduction ^7–11. It is implicated in Alzheimer’s disease and binds the amyloid precursor protein ^12,13. Cholesterol also changes the conformation and localization of viral fusion proteins ^14–16. Cholesterol can modulate membrane protein function by specific binding to proteins or by non-specific effects on the membrane physical properties. However, the precise mechanism of cholesterol’s influence on membrane proteins is so far poorly understood, and likely varies from protein to protein. The lack of information is largely due to the small size and complex dynamics of cholesterol in lipid bilayers, and the incompatibility of cholesterol with the detergent micelles and small isotropic bicelles that are commonly used for membrane protein reconstitution before solution NMR studies ¹⁷. In comparison, cholesterol can be advantageous for protein crystallization when used in concert with curvature lipids such as monoolein ¹⁸. Of the approximately dozen high-resolution (< 3 Å) cholesterol-bound membrane protein structures that have been deposited in the Protein Databank so far, most were solved using X-ray crystallography. However, in most these cases, the bound cholesterol in these structures results from co-crystallization from monoolein-cholesterol lipid cubic phases, and has no known function. The β₂ adrenergic receptor and the glutamate receptor structures are examples where the cholesterol-binding sites have partly understood functions ^7,19. Therefore, a strategy for determining functionally important protein-cholesterol binding sites in lipid bilayers is important for elucidating the biological roles of cholesterol in lipid membranes.

Magic-angle-spinning (MAS) solid-state NMR (SSNMR) is ideally suited to the investigation of protein-lipid interactions in lipid bilayers ^20–22. Solid-state NMR spectroscopy can detect angstrom-to-nanometer intermolecular contacts ^23,24 under a wide range of temperature, bilayer composition, and membrane viscosities, and thus can potentially provide detailed structural information about cholesterol-protein binding. A number of recent solution NMR studies have reported cholesterol interactions with membrane proteins by monitoring protein chemical shift perturbations ^12,25,26. However, it is recognized that chemical shift perturbations can occur through allosteric effects, so that residues far from the bound cholesterol may exhibit significant chemical shift changes. Therefore, to unambiguously determine the location of cholesterol relative to the protein, distance measurements are necessary.

Recently, we showed that aliphatic tail-fluorinated cholesterol provides an effective probe for determining the cholesterol-binding site to proteins using solid-state NMR of bilayer-bound proteins ²⁷. By combining ¹³C-labeled protein with aliphatic-heptafluorinated cholesterol, we determined the binding site of cholesterol to the influenza M2 protein. However, current commercially available ¹³C-labeled cholesterol compounds place ¹³C-enriched sites either at the sterol polar end or at the nonpolar aliphatic tail, whereas the vast central region of cholesterol is NMR-silent, thus restricting the detectable region of cholesterol for protein binding to the two ends of the molecule.

In this work we demonstrate biosynthetic ¹³C-labeling of cholesterol from yeast, together with sensitivity-enhancing dynamic nuclear polarization (DNP) NMR, for determining the cholesterol-binding site in membrane proteins with high sensitivity. Yeast normally produce the sterol ergosterol, which shares common biosynthetic steps with cholesterol until the final transformations, where three enzymes, Erg4p, Erg5p, and Erg6p catalyze methyltransferase and reductase activity at C24 and desaturase activity at C22 (Fig. 1a). A genetically engineered Saccharomyces cerevisiae strain, RH6829, has been developed in which the Erg5p and Erg6p enzymes were replaced with two dehydrocholesterol reductases for C24 and C7, thus producing cholesterol ^28,29. Using this metabolically engineered yeast strain, we show that cholesterol can be ¹³C skip-labeled at alternate carbons throughout the molecule ^28–30, which dramatically increases the detectable intermolecular interface with the influenza virus M2 protein.

Figure 1.

Biosynthetic ¹³C labeling of cholesterol in yeast for investigating protein-cholesterol interactions. (a) Key steps in the Saccharomyces cerevisiae biosynthetic pathway for ergosterol. Replacement of Erg6p and Erg5p enzymes by DHCR7 and DHCR24 enzymes enables efficient production of cholesterol. The 1-¹³C and 2-¹³C glucose labeled carbons are colored in red and blue, respectively. (b) 1D and 2D ¹³C CP spectra of 1-¹³C CHOL in POPC/POPG (4 : 1) bilayers, measured at 275 K. The sterol ring naming convention is indicated on the chemical structure. Assignment of the cross peaks confirm the ¹³C labeled positions.

The influenza M2 protein is a tetrameric membrane protein ³¹ that conducts protons to initiate viral uncoating ^32–35 as well as mediating membrane scission of the budding virus ³⁶. The membrane scission function has been shown to require cholesterol ^37,38 and an amphipathic helix (AH) that is C-terminal to the proton-conducting transmembrane (TM) helix ^39,40. Direct interaction between cholesterol and M2 was first suggested based on biochemical assays ^41,42. We recently measured M2-cholesterol ¹³C-¹⁹F distances to show that cholesterol binds to two of the four TM helices from the vertices of the tetramer ²⁷, and binding does not require a putative cholesterol-recognition amino acid sequence ^43,44 but requires the L-shape formed by the TM helix and the AH. Based on the measured distances between the TM residues and cholesterol tail and the measured sterol plane orientation ²⁷, we proposed a structure of the bound complex. However, this previous study did not have any distance restraints for the extended sterol of cholesterol due to the lack of suitable ¹³C-labeled cholesterol. We now show 2D ¹³C-¹³C correlation spectra with multiple cross peaks between the protein and yeast-labeled cholesterol, which allow us to determine an extensive binding interface between M2 and cholesterol.

Materials and Methods

Biosynthesis of ¹³C-labeled cholesterol from yeast

Saccharomyces cerevisiae strain RH6829 ^28,29, was a gift from Professor Riezman (University of Geneva). The yeast was plated and inoculated into starter culture following established protocols ^45,46. The 500 mL production medium consisted of 3.34 g (0.67%) yeast nitrogen base with ammonium sulfate but without amino acids (VWR AMRESCO), 695 mg (0.14%) yeast dropout medium supplements without His, Leu, Trp, and uracil (Sigma), 190 mg Leu, 76 mg uracil, and 7.5 g (15%) of 1-¹³C glucose (Cambridge Isotope Laboratories). This medium was filter-sterilized before use. The production medium was split into two 250 mL portions, each inoculated with 25 μL (1 : 10,000) RH6829 starter culture, and incubated for 68 h at 30°C while shaking at 220 rpm. The cells were harvested and washed with trichloroacetic acid ⁴⁷. The cell pellets were lyophilized and transferred to 20 mL glass scintillation vials for sterol extraction.

The cells were resuspended in an extraction cocktail consisting of 3 mL 60% (w/v) KOH in water, 3 mL 0.5% (w/v) pyrogallol in methanol, and an additional 4.5 mL methanol; 10.5 mL was used per 250 mL culture. The cells were heated at 80°C for 2 h with occasional mixing, and then extracted three times with ~4 mL petroleum ether. The ether phase was pooled into 20 mL scintillation vials and dried by nitrogen gas. The dried material was redissolved in acetone and passed through a 0.22 μm syringe filter to prepare for HPLC purification. The concentration of the extracted material was estimated by comparing the extract solution against commercial cholesterol solutions spotted on a TLC plate and visualized with an acidic FeCl₃ stain (50% w/v FeCl₃ in water with 5% v/v glacial acetic acid and 5% v/v sulfuric acid, Fig. S1a).

The extracted fraction was purified by HPLC using a Vydac C18 reversed-phase column (4.6 mm i.d. × 100 mm, 5 μm particle size) under isocratic conditions of 30% solvent A (90% ethanol, 5% isopropanol, 4.6% methanol) and 70% solvent B (acetonitrile) and a 1.5 mL/min flow rate. The extract solution (~1.5 mg/mL) was purified over 14 injections of 300 μl; the main peak eluted at 9 minutes (Fig. S1b). Approximately 8 mL of eluant was collected per injection. The total mass of purified 1-¹³C CHOL from the 500 mL production culture was 9 mg.

Expression of recombinant M2(21–97) protein

An Udorn 1972/H3N2 M2 construct (residues 21–97) spanning the TM domain and the full cytoplasmic domain was expressed in E. coli as described previously ^48,49. The construct, termed M2T below, contains an N-terminal His₆ tag and a TEV protease cleavage site for purification. The protein was ¹³C, ¹⁵N-labeled at Ile, Phe, Gly, and Ala (IFGA-M2T) by adding 150 mg/L of each ¹³C, ¹⁵N-labeled amino acid to the M9 growth media (48 mM Na₂HPO₄, 22 mM KH₂PO₄, 8.55 mM NaCl, 1 g/L NH₄Cl, 2 g/L D-glucose, 2 mM MgSO₄, 0.1 mM CaCl_2, 100 mg/mL ampicillin). In addition, 50 mg/L unlabeled Ser was added to prevent Ser labeling due to scrambling from Gly ^50,51.

Cells were grown in 2 L of Luria broth media to an OD₆₀₀ of 0.8, then pelleted and resuspended in 1 L M9 medium. About 40 minutes after suspension, protein expression was induced by adding isopropyl-β-D-thiogalactopyranoside (IPTG) to a concentration of 0.4 mM. After 4 hours, cells were pelleted and resuspended in lysis buffer (50 mM Tris, 50 mM NaCl, pH 7.8) and sonicated on ice for 30 min in cycles of 30 s on, 30 s off. M2T was purified from both the lysate supernatant and inclusion bodies by Ni-NTA affinity chromatography, concentrated and washed, and incubated with ~1 mg TEV protease for 48 h at room temperature. Cleaved M2T was purified by HPLC on a Vydac C4 reversed-phase column (22 mm × 250 mm, 10 μm particle size) using a mobile phase of water/acetonitrile containing 0.1% trifluoroacetic acid. Purification was conducted with a gradient of 30–99% acetonitrile over 70 minutes at a flow rate of 10 mL/min. About 12 mg of M2T was purified from 1 L of M9 medium and was lyophilized to dryness before use.

Synthesis of M2(22–61) peptides

A ¹³C, ¹⁵N-labeled M2 Weybridge peptide (residues 22–61) (NH₂-SSDPLVVAASII-GILHLILWILDRLFFKSIYRRLKYGLKR-amide) was synthesized using a custom-designed fast-flow peptide synthesizer ^27,52. The peptide was ¹³C, ¹⁵N-labeled at G34 and I39 (henceforth referred to as GI-M2W) to allow site-specific verification of the M2-cholesterol contacts. Amino acids were coupled using HATU (O-7-(azabenzotriazol)-1-yl-N,N,N’,N’-tetramethyluronium hexafluorophosphate) at 70°C under flow rates of 2.0 or 3.1 mL/min. Coupling times were 60 s for ¹³C, ¹⁵N-labeled residues and 80 s for unlabeled residues. Ten equivalents of unlabeled amino acids and four equivalents of labeled amino acids were used for each coupling. Resin cleavage and sidechain deprotection were conducted using 95% trifluoroacetic acid (TFA), 2.5% triisopropyl silane (TIPS), and 2.5% (H₂O) at room temperature for 2–3 h. The cleaved peptide was precipitated with cold diethyl ether. The crude peptide was purified by reverse-phase HPLC on a C4 column (22 mm × 250 mm) with an acetonitrile-water gradient of 30–70%. All mobile phases contained 0.1% TFA. Purified peptide masses were verified on an Agilent 6520 LC-MS equipped with a Zorbax 300SB-C3 column. Pure peptide fractions were pooled and lyophilized for later reconstitution into membrane bilayers. About 14 mg purified peptide was obtained from a 0.06 mmol synthesis scale.

Membrane sample preparation

IFGA-M2T and GI-M2W were reconstituted into lipid membranes containing 1-¹³C CHOL by organic-phase mixing. Two membrane mixtures were used: a POPC/POPG/CHOL membrane with a molar ratio of 8 : 2 : 2 or 9.8 : 3.3 : 2, and a virus-mimetic membrane (VMS) containing POPC, POPE, POPS, sphingomyelin, and CHOL at a molar ratio of 36 : 18 : 18 : 11 : 17 ⁵³. Thus, the cholesterol mole fractions in these membranes are 13–17 mol%. The protein-lipid molar ratios are 1 : 20 to 1 : 25, while the protein-cholesterol molar ratio was 1 : 4.5 for all samples. All lipids and cholesterol were dissolved in chloroform, with up to 10% methanol for POPS, while the M2 protein was dissolved in trifluoroethanol. The solutions were combined, dried with a stream of nitrogen gas, and placed under vacuum overnight to remove residual organic solvent. The samples were rehydrated in ~3 mL Tris buffer (20 mM, pH 7.5) and subjected to 6–7 freeze/thaw cycles between liquid nitrogen temperature and a 37°C water bath. The liposome suspension was pelleted by ultracentrifugation at 160,000 × g and 4°C for 4–16 h, and the supernatant was decanted to obtain the proteoliposomes pellet with ~70 μL associated buffer.

Membrane samples for DNP experiments were prepared by adding D₂O to the centrifuged membrane pellets to create a 3 : 1 D₂O/H₂O composition. This mixture was vortexed to obtain a uniform suspension, then the sample volume was reduced to ~100 μL by repeated lyophilization of 10–20 minutes at a time, with vortexing in between to maintain a uniform suspension and prevent sample losses along the walls of the vessel. An equivalent mass of “DNP juice”, which consists of 15 mM AMUPOL in 60% (v/v) ¹³C-depleted glycerol (Cambridge Isotope Laboratories), 30% D₂O, and 10% H₂O, was added to the membrane suspensions, and vortexed to mix the radicals and the membrane uniformly. The mixture was dried to ~50% hydration by brief lyophilization, and centrifuged (3000 × g) into 3.2 mm MAS rotors.

Solid-state NMR experiments

Non-DNP 1D and 2D ¹³C spectra at 293 – 255 K were measured on a 600 MHz (14.1 T) Bruker AVANCE III HD spectrometer using a 3.2 mm magic-angle-spinning (MAS) probe. Radiofrequency (rf) field strengths were 71–85 kHz for ¹H and 31–50 kHz for ¹³C. ¹³C chemical shifts were externally referenced to the CH₂ resonance of adamantane at 38.48 ppm on the tetramethylsilane scale. ¹³C cross-polarization (CP) spectra were measured under 9–13 kHz MAS from 293 K to 255 K, which spans both the liquid-crystalline phase and gel phase of the POPC/POPG membrane.

DNP sensitivity-enhanced 2D ¹³C-¹³C correlation spectra at ~110 K were measured on a 600 MHz NMR spectrometer equipped with a 7.2 T gyrotron operating at a Larmor frequency of 395 GHz (Bruker BioSpin, Billerica, MA). The samples were spun at 12 kHz or 13.33 kHz using a 3.2 mm MAS probe. Typical rf field strengths were 100 kHz for ¹H and 60–67 kHz for ¹³C. ¹H-¹³C CP used a tangent shape on H and contact times of 0.5 – 1.0 ms. Sample temperatures were ~98 K and ~110 K without and with microwave irradiation. The ¹³C sensitivity enhancement factors were 28–42, as evaluated from 1D ¹³C CP spectra measured with a 10 s recycle delay. 2D ¹³C-¹³C correlation spectra were measured using C CORD spin diffusion mixing times of 30 ms or 100–500 ms (Table S1) ⁵⁴. To simplify the ω₁ dimension of the 2D spectra, we applied a ¹³C-¹³C double-quantum filter (DQF) before the t₁ evolution period in many experiments, so that cholesterol ¹³C peaks that do not have directly bonded ¹³C sites are suppressed. This DQF strategy in 2D correlation experiments has been shown before in the absence of DNP for removing natural abundance ¹³C signals ⁵⁵. A 2D dipolar INADEQUATE spectrum of POPC/POPG bound IFGA-M2T was measured to assign the protein resonances. The 2D DQF-CC and INADEQUATE experiments used the SPC5 sequence ⁵⁶ for ¹³C dipolar recoupling with a combined DQ excitation and reconversion time of 0.6–0.9 ms ⁵⁷.

Molecular docking with distance constraints

To model the M2-CHOL complex using the measured distance and orientation constraints, we used the docking program HADDOCK ^58,59. The input restraints include the ¹³C-¹³C distance contacts reported in this work, ¹³C-¹⁹F and ¹³C-¹³C distances previously reported on M2 peptides, and cholesterol orientation relative to the bilayer normal measured using ²H NMR ²⁷. HADDOCK allows designation of ambiguous and unambiguous interaction restraints to guide the docking: all unambiguous restraints were maintained throughout the docking process, while 50% of ambiguous restraints were discarded at the beginning of each run ⁶⁰. M2-CHOL contacts with specific residue and atom assignments, obtained from the site-specifically labeled peptide M2(22–61), were input as unambiguous restraints, while assignments obtained from M2T samples, which can correspond to multiple residues (e.g. I39 or I42), were input as ambiguous restraints (Table S3). Inter-vector angular restraints ⁶¹ were defined between the bilayer normal, which is equivalent to the channel axis, and a molecule-fixed cholesterol coordinate system ²⁷. Previously, we defined the channel axis as the average vector of the four TM helices where the TM Cα atoms define each helix axis, and the cholesterol coordinate system was defined with respect to the C3 moiety ^27,62. For HADDOCK, addition of artificial vectors interferes with docking, thus we adopted an alternative approach of selecting proxy vectors in the protein and cholesterol to represent the same coordinate axes. The I35 Cα–W41 Cα vector of helix D is nearly parallel to the channel axis; the Ch13–Ch3 vector of cholesterol defines the Z-axis of the cholesterol molecular frame, while the Ch10–Ch19 vector is approximately coincident with the X-axis of the cholesterol molecular frame (Table S4). All these vectors lie within 5° of the previously established bilayer normal and cholesterol coordinate axes, and were used for implementing orientation restraints in HADDOCK.

A number of changes were made to the default HADDOCK settings as recommended for protein-ligand binding. The mode for clustering analysis was changed to RMSD with a 2.5 Å cutoff and a minimum cluster size of 6. The weighting factors for scoring during rigid-body docking and solvent refinement were altered, starting temperatures for the second and third temperature-accelerated dynamics (TAD) steps were reduced to 500 and 300°C, respectively, and the number of MD steps for rigid high-temperature TAD and first rigid-body cooling were set to 0. Dimethylsulfoxide was used during the final explicit solvent refinement stage as a membrane mimic. These settings are modifiable with “guru” access to the HADDOCK webserver. The default settings of 1000 initial structures for rigid body docking and 200 structures each for semi-flexible refinement and explicit solvent refinement were used. The input M2 structure used for the docking was the first model of the SSNMR structure ensemble (PDB accession code: 2L0J) ⁴⁰. The residues were renumbered to avoid non-uniqueness to conform to HADDOCK guidelines. The input cholesterol structure was accessed from the ZINC ligand database ⁶³. The full input parameter file (except for the PDB coordinate file), is given in Appendix A.

To evaluate and compare different docked structures, we first used the HADDOCK cluster and ranking analysis based on the HADDOCK scoring function. However, the output score does not perform as well when ranking protein-ligand complexes ⁶⁴, thus we additionally analyzed the docking structures using the PRODIGY-LIGAND contact-based binding energy predictor ^64–67. In addition, we compared the output cholesterol β and γ orientation angles to the experimentally measured angles from ²H NMR spectra ²⁷. To quantify the docked cholesterol’s orientational deviations from the measured orientations, we calculated the residual sum-of-squares (RSS) for each structure using the equation:

$$RSS={\left({\beta }_i-{\beta }_o\right)}^2+{\left({\gamma }_i-{\gamma }_o\right)}^2,$$

where β_o and γ_o are the output angles of the docked structure while β_i and γ_i are the experimentally measured input angles of 11° and 10°, respectively. The top 10 structures by the RSS ranking are shown in Fig. S9 and the RSS, PRODIGY, and HADDOCK scores are summarized in Table S5. The docked structures were visualized in PyMOL and the orientation angles and RSS were calculated in Matlab.

Results

Yeast biosynthesis of 1-¹³C CHOL

Yeast-based ¹³C enrichment of sterols has recently been extended beyond ergosterol by metabolic engineering of S. cerevisiae ²⁹. The biosynthetic pathways of fungal ergosterol, vertebrate cholesterol, and plant campesterol, diverge near the end. This allowed the replacement of two ergosterol-synthesis enzymes by the corresponding cholesterol-synthesis enzymes, DHCR7 and DHCR24, to produce cholesterol (Fig. 1a). Supplying 1-¹³C labeled glucose labels 15 carbons in cholesterol, 12 of which do not have directly bonded carbons ^28,30. This skip labeling facilitates NMR structural studies by removing ¹³C-¹³C J-couplings and dipolar couplings, narrowing the linewidths and increasing spectral sensitivity. Compared to 2-¹³C glucose, 1-¹³C glucose labels all five methyl carbons in cholesterol - Ch19, Ch18, Ch21, Ch26 and Ch27 - which may have a higher propensity to form specific interactions with proteins. Moreover, the 1-¹³C labeled sites have larger chemical diversity and hence larger chemical shift dispersion (Fig. S2), making 1-¹³C labeling the more promising choice for studying M2-cholesterol interactions ⁶⁸.

We obtained 9.0 mg of purified and skip-labeled 1-¹³C CHOL from 500 mL of yeast medium containing 7.5 g of 1-¹³C glucose. One-dimensional ¹³C CP spectra of 1-¹³C CHOL reconstituted into POPC/POPG bilayers show linewidths of ~0.2 ppm in the liquid-crystalline phase of the membrane at 275 K. Two-dimensional ¹³C-¹³C correlation spectra measured with 300 ms mixing display up to four-bond correlations. These cross peaks are consistent with the spectra of cholesterol obtained with 2-¹³C sodium acetate, which produces the same labeling pattern as 1-¹³C glucose (Fig. 1b) ³⁰. The cholesterol ¹³C chemical shifts were assigned based on the connectivity patterns in the spectra and verified by comparison with literature chemical shifts ⁶⁹

Identifying protein-cholesterol contacts using 1-¹³C CHOL and ¹³C-labeled M2

Labeling with 1-¹³C glucose produced a large number of NMR reporters throughout cholesterol, which dramatically facilitated the search for the M2-CHOL interface and allowed a more detailed description of the binding site obtained from previous distance measurements, which involved only the cholesterol tail ²⁷. We reconstituted 1-¹³C CHOL with ¹³C-labeled M2T in POPC/POPG bilayers to measure cholesterol-protein ¹³C-¹³C cross peaks in 2D correlation spectra. To immobilize cholesterol in the membrane, we first conducted the 2D experiments at ~253 K, in the gel phase of the membrane. However, at this temperature the cholesterol ¹³C signals broaden significantly, to ~2 ppm, compared to the liquid-crystalline phase (Fig. S3), with the concomitant drop in sensitivity. Indeed, 2D ¹³C-¹³C correlation spectra measured at 250 K show relatively low sensitivity and no clear intramolecular long-range cross peaks for cholesterol are observed. For example, the cholesterol Ch3 cross section shows no cross peaks with any other carbons, even though the two-bond Ch3-Ch1 and Ch3-Ch5 correlations are fully expected (Fig. S4a). Under these conditions, M2 signals show up relatively well, and a potential Ile Cα-Ch18 cross peak was observed. However, no other intermolecular cross peaks are detected. Therefore, at this moderate low temperature regime, the dynamic properties of M2 and cholesterol are not conducive for detecting intermolecular cross peaks. We thus turned to DNP-based 2D experiments at ~110 K, where both the protein and cholesterol are well immobilized in the membrane. We routinely obtained sensitivity enhancement factors of 28–42. The cholesterol ring ¹³C signals do not broaden noticeably from 255 K to 110 K. facilitating spectral assignment. In addition, the cholesterol ¹³C chemical shifts are unperturbed between the high and low temperatures, thus allowing assignments made at 273 K to be transferred to the DNP spectra at 110 K (Fig. S3, S5a). The methyl ¹³C signals exhibit the expected line broadening due to the well known methyl rotation at this temperature ⁷⁰.

The 2D DNP ¹³C-¹³C correlation spectra are shown in Fig. 2–5 and Fig. S4b-S7. The 30 ms spectrum (Fig. 2) shows predominantly intramolecular correlations. In the 300 ms CHOL-only spectrum, Ch3-Ch17 and Ch3-Ch18 cross peaks are observed (Fig. 3a). These carbons reside on opposite ends of the sterol rings at distances of 8–9 Å ⁷¹, thus cross peaks for very long distances can be detected with DNP sensitivity enhancement. However, for the mixed M2T and CHOL sample, the 2D spectra shows significant congestion (Fig. 4a): for example, between 12 and 45 ppm, many CHOL resonances overlap with the ¹³C signals of the labeled Ile, Phe, Gly and Ala.

Figure 2.

30 ms 2D CC spectrum of IFGA-M2T and 1-¹³C CHOL in POPC/POPG membranes, showing full resonance assignment. The spectrum was measured at ~110 K under 13 kHz MAS with sensitivity enhancement by DNP. Significant spectral congestion is seen due to the large number of CHOL cross peaks in addition to the protein cross peaks. Cholesterol cross peaks are assigned in orange, protein cross peaks in black, and CHOL-M2 cross peaks in red.

Figure 5.

Regions of 2D ¹³C-¹³C correlation spectra of IFGA-M2T and 1-¹³C CHOL in POPC/POPG membranes, showing protein-CHOL cross peaks. All spectra were measured at ~110 K under DNP condition. Left column shows 30 ms spectra (purple) where only short-distance contacts are detected. Middle column (blue) shows 300 ms spectra where longer-range cross peaks are detected. Right column (black) shows 300 ms spectra of the protein-free control membrane with 1-¹³C CHOL. (a) Aromatic region of the 2D spectra, showing a Phe Cγ-Ch15 cross peak. (b) Ile regions of the 2D spectra, showing cross peaks between Ile sidechain carbons and CHOL Ch13, Ch17, Ch19 and Ch21.

Figure 3.

300 ms (a) 2D CC spectrum and (b) 2D DQF-CC spectrum of 1-¹³C CHOL in POPC/POPG membranes, showing full assignment of the intramolecular CHOL cross peaks. The spectra were measured at ~110 K with DNP.

Figure 4.

Simplification of 2D CC correlation spectra by DQF. All 2D spectra were measured at ~110 K with DNP. (a) 300 ms 2D CC spectrum of IFGA-M2T and 1-¹³C CHOL in POPC/POPG membranes measured without the DQF. Representative protein (blue) and CHOL (orange) ¹³C chemical shifts in the ω₁ dimension are shaded. (b) 300 ms 2D DQF-CC spectrum of protein-free POPC/POPG membranes with 1-¹³C CHOL. The DQF suppresses all ¹³C signals without a directly bonded ¹³C in the ω₁ dimension, leaving only Ch18, Ch13, and Ch17 signals. (c, d) 2D DQF-CC spectra of IFGA-M2T and 1-¹³C CHOL in POPC/POPG membranes measured with ¹³C spin diffusion mixing times of (c) 30 ms and (d) 300 ms. DQF significantly simplifies the ω₁ dimension of the spectra, facilitating assignment of intermolecular protein-CHOL cross peaks.

To simplify the spectra and allow assignment, we took advantage of the uniform ¹³C-labeling of the selected protein residues, which contrasts with the mostly skip-labeled 1-¹³C CHOL, by adding a ¹³C-¹³C double-quantum filter (DQF) before t₁ evolution ⁵⁵. This 2D DQF-CC experiment removes most cholesterol peaks from the ω₁ dimension except for the directly bonded Ch17, Ch13, and Ch18 carbons. As a result, roughly 75% cholesterol cross peaks are removed in the 2D DQF-CC spectrum (Fig. 3b), while those that remain can be easily assigned from the CHOL-only spectra with or without DQF (Fig. S5). In the protein-containing sample, the 30 ms DQF-CC spectrum emphasized the protein signals, and enabled assignment of the protein ¹³C peaks (Fig. 4c). A 2D dipolar INADEQUATE spectrum (Fig. S6) was also measured to confirm the assignment. The protein-cholesterol cross peaks of interest can thus be identified in long-mixing-time 2D DQF spectra (Fig. 4d, Fig. 5a, Fig. S7b, S8).

The 2D DQF-CC spectra shows multiple Ile cross peaks with cholesterol. Ile Cγ1 cross peaks with Ch17 (Fig. 5a) and with Ch24 (Fig. 2) are already observed at 30 ms mixing. The cross peak with Ch17 remains in the 300 ms DQF-CC spectrum, while the cross peak with Ch24 is below detection. With DQF, a 36.9 ppm peak in the ω₁ dimension exhibits cross peaks with CHOL methyl carbons Ch19 (19.8 ppm) and Ch21 (20.6 ppm) and D ring carbons Ch13 and Ch17 (Fig. 5a). The 36.9 ppm chemical shift can in principle be assigned to either Ile Cβ or Phe Cβ; however, the dipolar INADQUATE spectrum (Fig. S6) shows cross peaks between Ile Cα (64.1 ppm) and Ch13 (43.1 ppm), thus supporting the assignment of these cross peaks in the DQF-CC spectra as Ile Cβ to cholesterol contacts.

For the aromatic region of the 2D spectra, the spectra without DQF already resolve several cross peaks between Phe sidechain and CHOL (Fig. 5b). Specifically, cross peaks between Phe Cδ/Cε and CHOL Ch3 and Ch9 are observed, thus establishing protein contacts with the sterol A and B rings. The Phe-C3 cross peak agrees with a previous report of Phe Cγ − Ch3 contact in DMPC bilayers by SSNMR at 262 K ⁷². These Phe-CHOL contacts were not observed in the DQF-CC spectra, due to the removal of Ch3 and Ch9 signals in the ω₁ dimension and the low overall sensitivity in the aromatic spectral region.

Protein-cholesterol contacts from site-specifically labeled M2 peptide

Since the M2T sample labels all Ile, Phe, Gly and Ala residues, sequence-specific assignment cannot be made. We thus sought to provide some residue-specific information about the cholesterol-protein cross peaks using a G34, I39-labeled M2W(22–61) peptide. The M2Wamino acid sequence was taken from the chicken/Weybridge H7N7 strain, which contains a putative cholesterol-recognition motif ⁴³. However, our recent study ²⁷ has shown that this motif is not required for cholesterol binding. Indeed, the Udorn M2 sequence used in the previous experiments ²⁷ and in the current IFGA-M2T sample both give significant protein-cholesterol ¹³C-¹⁹F dipolar couplings and ¹³C-¹³C cross peaks, despite the fact that Udorn M2 does not have a cationic residue thought to constitute part of the cholesterol-recognition motif. We reconstituted GI-M2W into a POPC/POPG membrane and a virus-mimetic (VMS) membrane, to investigate whether M2-cholesterol binding depends on the membrane environment.

Although only a single Ile was labeled in the peptide, the unfiltered 2D CC spectra remain sufficiently congested. We therefore conducted the same 2D DQF-CC experiments to measure I39-cholesterol cross peaks. Fig. 6 shows selected regions of the 2D spectra. The two membrane samples show very similar protein and cholesterol chemical shifts, with at most a 0.3 ppm deviation. I39 Cβ and Cγ2 cross peaks with cholesterol Ch13, Ch15, Ch17, and Ch21 are observed. These intermolecular contacts again involve the cholesterol D ring and the nearby methyl carbon, as was the case for the M2T sample (Fig. 5a). The similarity of the observed cholesterol cross peaks with the protein in both the cytoplasmic-containing M2 and the site-specifically labeled TM-AH peptide indicates that I39 is a major contributor of the Ile-cholesterol cross peaks observed in the M2T spectra, although it does not rule out the possibility that other Ile residues such as I42 may also interact with cholesterol. It also means that the M2-cholesterol binding is unaffected by the presence or absence of the cytoplasmic tail or the cholesterol-recognition sequence motif. Interestingly, the two lipid membranes do not give an identical set of M2-cholesterol cross peaks. For example, the I39 Cγ2 cross peak to Ch17 is detected only in the VMS membrane and not in the POPC/POPG membrane (Fig 4c, d). This difference suggests that the M2-cholesterol binding interactions may be slightly different between the POPC/POPG bilayer and the VMS membrane. This difference may be a result of changes in the TM helix tilt angle in the VMS membrane, whose rigidity and larger thickness can affect the TM helix orientation ^73–75.

Figure 6.

Protein-cholesterol correlations observed in DNP 2D DQF-CC spectra of GI-M2W(22–61) peptide in two lipid membranes. (a) Regions of the 400 ms 2D DQF-CC spectrum of GI-M2W in POPC/POPG membranes. (b) Regions of the 400 ms 2D DQF-CC spectrum of GI-M2W in the VMS membranes. I39 cross peaks with several CHOL carbons are detected. (c) 1D cross sections extracted from the I39 Cβ chemical shift, showing similar cross peaks in the two lipid membranes. (d) 1D cross sections extracted from the CHOL Ch17 chemical shift, showing a protein-cholesterol cross peak in the VMS membrane but not the POPC/POPG membrane.

¹³C-¹³C distances constrain cholesterol docking onto the M2 protein

The many intermolecular ¹³C-¹³C cross peaks observed in these 2D spectra allow us to establish distance restraints between M2 and the sterol rings of cholesterol, which were previously inaccessible in commercially available terminally labeled cholesterol. These ¹³C-¹³C distance restraints were input into the HADDOCK program for molecular docking ^58,59. In addition, the previously measured ¹³C-¹⁹F distances and cholesterol orientation angles ²⁷ were also supplied as restraints. HADDOCK uses both ambiguous and unambiguous distance restraints (Table S3), of which the unambiguous restraints are always applied, while half of the ambiguous restraints were used for each of the 1000 initial runs. The ambiguous restraint list consisted of contacts obtained from the M2T sample for which sequence-specific assignment cannot be made. The three lipid-facing Ile residues of the M2 TM domain are I35, I39, and I42, where I39 and I42 reside at a similar depth while I35 lies in the middle of the bilayer. Thus, many of the observed Ile-cholesterol cross peaks were assigned to either I39 or I42. Of the four Phe residues of the amphipathic helix, only F47 and F54 face lipids. Based on our previous docking model, we had assigned the Phe–Ch3 contact observed by Ekanayake et al. to F47 ^27,72; nevertheless, we submitted this restraint at first as ambiguous. Preliminary docking simulations ruled out F48 and F54 as possible residues to contact cholesterol, since structures with these contacts strongly violated the submitted distance or angular restraints for cholesterol (Tables S2 and S3). Thus, we re-categorized the F47- cholesterol cross peaks as unambiguous, and the Phe/Ile Cβ cross peaks to cholesterol were assigned to I39 or I42 based on closer relative distances compared to F47. Additional unambiguous restraints included ¹³C-¹³C contacts determined using the G34, I39-M2W peptide, and distances from ¹³C-¹⁹F experiments and ¹³C-¹³C correlations with previously reported M2W and M2 Udorn peptides ²⁷.

We used a combination of metrics to critique the 200 HADDOCK output structures, including the HADDOCK score, PRODIGY score, and residual sum of squares (RSS) of the cholesterol β and γ orientation angles. These metrics and the rankings for the top 10 structures according to RSS are listed in Table S5. The HADDOCK score is applied both to whole clusters and to individual structures. However, in a recent protein-ligand binding challenge, ranking by HADDOCK score alone produced less than 20% good matches ⁶⁴. Consistently, we observed that similar M2-cholesterol docked structures can show large discrepancies in the HADDOCK scores. In comparison, the PRODIGY-LIGAND method identifies and categorizes atomic contacts to the ligand within 10.5 Å and assigns an energy value based on the atom pairs, resulting in more accurate structures and binding affinities ⁶⁴. The PRODIGY scores for the top 100 PRODIGY-ranked structures differ only by ~0.3 kcal/mol. Thus, we used RSS as the primary metric for ranking the docking results and at the same time monitored the PRODIGY scores for red flags when comparing structures. Of the top 10 structures ranked by RSS (with the lowest RSS values), the two poorest ranking structures by PRODIGY show cholesterol ring orientations that differ substantially from the rest of the ensemble (Fig. S9). We consider these structures as not accurately representing M2-bound cholesterol, despite the good RSS rankings. The remaining eight structures show consistent docking poses for the bound cholesterol. For the subsequent discussion, we examine the top-ranking structure by RSS that also ranked favorably by PRODIGY.

The HADDOCK docked M2-cholesterol structural model is shown in Fig. 7, while the measured distances and orientations are given in Table S3 and S4. Two cholesterol molecules are docked onto the four-helix bundle (Fig. 7a), in accordance with the previous finding that cholesterol binds M2 with a stoichiometry of 2 per tetramer. However only one cholesterol molecule was used during the simulation ²⁷. In the docked model, cholesterol is positioned on the outer face of the TM helix, with the sterol rings aligned with a groove formed by the lipid-facing sidechains from I39 to F47. The rough, methyl-studded β-face of cholesterol is flush with the TM helix, allowing the Ch18 and Ch19 methyl groups to form van der Waals contacts with the protein. Overall, Ile sidechains interact with the aliphatic tail and the D ring. Residue F47 is in contact with the A ring and likely forms CH-π interactions with the cholesterol hydrogens ⁷⁶. Meanwhile L43 and L46, which were not labeled in this study, constitute the lipid-facing residues in the final turn of the TM helix and are located at the same depth as the cholesterol A and B rings. The position of L43 suggests the presence of CH-π interaction with the Ch5-Ch6 double bond as another means of stabilizing the bound complex. Other stabilizing factors may stem from interactions between the distal cholesterol tail and more deeply embedded sidechains such as I35, but the principal forces stabilizing the bound complex likely originates from the interactions between M2 and the sterol ring system.

Figure 7.

Cholesterol docking onto M2 (PDB code: 2L0J) using HADDOCK. Distance restraints obtained from the 2D CC spectra were used to constrain the binding site. Cholesterol carbons that exhibit cross peaks with the M2 protein are shown as spheres. Key Ile and Phe residues at the binding interface are indicated, along with the distances to the cholesterol carbons.

Discussion

Implication of the M2-cholesterol complex structure for membrane scission

The current data demonstrate the feasibility and utility of using biosynthetic ¹³C-labeled cholesterol to determine protein-cholesterol binding structures in lipid bilayers. Out of a total of 27 cholesterol carbons, 15 were ¹³C-enriched by 1-¹³C glucose, and among these 9 carbons, distributed on all four sterol rings as well as in the aliphatic tail, showed correlations peaks with the M2 protein (Table S3). Importantly, detecting these protein-cholesterol correlation peaks required a careful choice of the experimental temperature at which both the protein and cholesterol are immobilized in the lipid membrane. In the liquid-crystalline phase of the membrane, cholesterol is highly dynamic ^30,77, giving well-resolved spectra but reducing the efficiency of ¹³C spin diffusion for measuring long-range correlation peaks. Cooling the membrane to moderate low temperatures of about 253 K improves ¹³C spin diffusion, but the spectra show heterogeneously broadened lines for the protein and cholesterol (Fig. S3), which reduce both the resolution and the sensitivity of the spectra. Use of DNP NMR experiments at cryogenic temperatures circumvent these problems and allow high-sensitivity detection of intermolecular correlation peaks that are inherently much weaker than one- or two-bond cross peaks.

The measured ¹³C-¹³C restraints support our previous finding that M2 binds cholesterol at an annular binding site on the lipid-facing face of the TM helix, as opposed to a buried or internal binding site. The cholesterol-binding interface consists mostly of Ile and Leu sidechains, with F47 at the polar head of cholesterol. The observed contacts between the Ile sidechains and cholesterol methyl groups suggest that the methyl-studded β face of cholesterol is in direct contact with M2, as is supported by the HADDOCK model (Fig. 7b). Sidechain-methyl interactions provide key stabilization of the complex in a manner similar to models of α-synuclein ⁷⁸, the human type 3 somatostatin receptor ⁷⁹, and the human β₂-adrenergic receptor ¹⁹. The M2-cholesterol interface shows both methyl-methyl interactions and CH-π interactions, includes at least five TM residues, and spans a region that is larger than the 9 Å length of the cholesterol rings. These results indicate an extended TM binding interface with cholesterol. Interestingly, previous biochemical data suggest that perturbation of some of these TM residues may not abolish M2-cholesterol interaction. For example, single alanine mutation at L46 did not affect the formation of Udorn viral filaments ³⁷, and triple alanine mutations of L46, F47, and F48 showed only a minor effect on viral infectivity ⁸⁰. In comparison, a penta-Ala mutation of five aromatic residues in the AH abolished cholesterol-dependent membrane scission ^36,81,82. This is consistent with our observation that the TM-only peptide, lacking the AH, does not bind cholesterol ²⁷. Moreover, TM and AH conformation and dynamics of the penta-Ala mutant was shown to be insensitive to the presence or absence of cholesterol in the membrane ⁸². Therefore, our M2-cholesterol complex structure and the mutagenesis data together indicate that the amphipathic helix is necessary for keeping the TM helix in the correct conformation to interact with cholesterol. Furthermore, cholesterol binding to the neck of the TM and AH may control the insertion depth of the AH in the bilayer, as indicated by recent EPR studies of spin-labeled M2 in the presence and absence of cholesterol ^38,82. This effect represents a mechanism for regulating AH-mediated membrane curvature induction through cholesterol binding to M2.

Binding cholesterol at two of the four vertices of the four-helix TM bundle provides a means to target M2 to the site of virus budding. This is consistent with M2’s preferential localization to the boundary of the raft and non-raft domains ^{4,36,42,83,84}. A previously suggested binding site in the groove between two TM helices ⁷² does not agree with our measured restraints. In addition to cholesterol, other factors that have been proposed to drive M2 to the neck of the budding virus including M2’s preference for high curvature regions of the membrane ^36,83 and for negative-curvature phosphatidylethanolamine ⁸⁵, S-palmitoylation at Cys50 ⁸⁶, and M2’s interaction with other influenza matrix proteins ^87,88. Cholesterol is also known to promote the formation of lipid stalks in membrane fusion ¹⁴, thus the M2-cholesterol interaction represents a confluence of several physiological mechanisms to form the high-curvature neck of the budding virus to induce the eventual membrane scission.

High-sensitivity studies of cholesterol-protein interactions by biosynthetic ¹³C labeling of cholesterol

In this study, we used 1-¹³C glucose to preferentially label methyl carbons in cholesterol, since we hypothesized that they are an important element for stabilizing M2-cholesterol binding. However, the similar ¹³C chemical shifts between the cholesterol and protein methyl signals and the line broadening at the ~110 K temperature used for the DNP experiments complicates resonance assignment. Use of 2-¹³C glucose labeled cholesterol, termed 2-¹³C CHOL, is a promising alternative ²⁸. 2-¹³C CHOL contains no ¹³C chemical shifts upfield of 20 ppm, and only one ¹³C resonance between 20 and 25 ppm (Fig. S2b), which is a region densely populated by Leu, Val and Ile methyl ¹³C signals. Thus, 2-¹³C CHOL will be especially useful for measuring protein methyl-cholesterol cross peaks. Although the ¹³C chemical shift dispersion of 2-¹³C CHOL is smaller than that of 1-¹³C CHOL, only 12 carbons are ¹³C-enriched, thus resonance overlap should be reduced. 2-¹³C CHOL can be prepared both from 2-¹³C glucose and 1-¹³C acetate. Likewise, it has been shown that 2-¹³C-acetate, through the mevalonate pathway, produces the same 1-¹³C CHOL labeling pattern as 1-¹³C glucose, but with higher yields per mass of ¹³C starting material and greater than 80% ¹³C incorporation ³⁰. Thus, acetate labeling has the benefits of higher sensitivity, increased statistical probability for protein-cholesterol cross peaks, and increased efficiency for spectral editing techniques such as the DQF experiment used here. The higher ¹³C labeling level also gives more freedom in preparing membrane samples with appropriate compositions. For the membrane samples used in this study, the mass ratio of 1-¹³C CHOL is 1 mg per 20 mg total dry mass, including lipids and protein. Given the 30 μL volume of the 3.2 mm MAS rotors, it is impractical to add more sample for boosting sensitivity. The membrane lipid molar ratio is dictated by the need for sufficient lipids to solvate the protein and the restriction to less than ~17 mol% cholesterol for studying curvature-inducing M2-cholesterol interactions ³⁶. Thus, with the limited mass of cholesterol, doubling the ¹³C enrichment level will facilitate NMR studies of protein-cholesterol binding.

This work focuses on the use of biosynthetic ¹³C-cholesterol in conjunction with NMR to determine the M2-cholesterol binding interface structure. The ¹³C-labeled cholesterol involves natural isotopes without the potential for structural perturbation, which makes it a superior approach to fluorescent labels, spin-labeled cholesterol derivatives, and even fluorinated cholesterol. While these non-natural molecules can be used to detect much longer distances to map protein-cholesterol binding sites, they have the potential to behave non-natively and perturb protein structure ⁸⁹. ¹³C-labeled cholesterol can also be used to study cholesterol dynamics in the membrane and when associated with proteins ^3,30,77. Beyond NMR, biosynthetically labeled cholesterol finds various other uses. For example, it has been exploited for mass spectrometric studies of membrane domain formation and lateral organization ⁶⁸, and perdeuterated cholesterol has been used for Raman spectroscopy and neutron scattering studies of membrane physical properties and intracellular cholesterol storage ^90–92. Therefore, biosynthetic cholesterol labeling opens many avenues for investigating biologically important interaction of this ubiquitous component of mammalian membranes with proteins and lipids.