Structure of TRPV1 channel revealed by electron cryomicroscopy

To perform structural analysis of TRPV1, we used a yeast expression system shown to produce functional channels that allowed Ca²⁺ entry when activated by capsaicin and that bound the specific ligand, resiniferatoxin (RTX) (22). To improve the efficiency of purification, we used an epitope tag that has been successfully used to purify a variety of transmembrane proteins, the 1D4 epitope from the C terminus of the G protein-coupled receptor rhodopsin (23, 24). The nonapeptide epitope tag incorporated into the C terminus of TRPV1 allowed us to purify TRPV1 in a single step using an immobilized monoclonal antibody, 1-D4 (Fig. 1a). A typical yield from this procedure is 1.5 mg of purified protein per 15 liters of cultured cells. Two bands were detected in Coomassie-blue-stained SDS/PAGE of TRPV1 (Fig. 1a) purified from yeast by using an immobilized 1D4 column and peptide elution: a major band with an apparent mass of 106 kDa, and a minor band of 98 kDa that constitutes 5–10% of the entire protein. Immunoblots (Fig. 1a) revealed that both bands are due to recombinant TRPV1 protein with intact C-terminal epitope tags. Treatment with peptide-N-glycosidase F (PNGase F), which cleaves N-glycosidic bonds between the sugar portion and asparagine of glycoproteins, did not shift the upper band to the position of the lower one, and treatment of yeast cells with tunicamycin slowed growth and reduced protein expression but did not affect the relative levels of the two bands. Therefore, we conclude that the size heterogeneity is unlikely to be due to heterogeneous glycosylation and is likely due to proteolytic removal of an ≈10-kDa N-terminal peptide. The presence of 5–10% of cleaved protein is not likely to have a significant effect on the 3D structure determined by using a single-particle cryo-EM approach.

A tetrameric state of the purified channel protein was confirmed by using size-exclusion chromatography (Fig. 1b). The narrowness of the major tetrameric peak eluted from the gel-filtration column suggests that the assembled channel subunits in this peak are monodisperse, properly folded protein particles with an apparent molecular mass of 498 kDa, corresponding to the correct size of the tetrameric channel plus 30% additional mass due to carbohydrate modifications and bound detergent.

In previous work, the functionality of yeast-expressed TRPV1 was shown by capsaicin-induced calcium influx in yeast expressing the mammalian channel (22). To test the functionality of the purified protein, Ca²⁺-flux measurements were performed with TRPV1 channels reconstituted into unilamellar vesicles. The external Ca²⁺ from the vesicles solution was removed by sequential gel filtration and treatment with a “calcium sponge” chelating resin (25) to reduce contaminating metal ions to low nanomolar levels. To activate TRPV1, we used the naturally occurring high-affinity agonist, resiniferatoxin (RTX) (11). The Fura-2 excitation spectra in Fig. 2 a and b show that before agonist addition the suspension of TRPV1-containing vesicles and control protein-free vesicles had very low levels of residual external Ca²⁺. When 50 nM RTX was added to TRPV1-containing vesicles, Ca²⁺ was rapidly released, within the manual mixing time, immediately shifting the excitation spectrum of the dye toward that of the Ca²⁺-bound form. In contrast, RTX did not have a detectable effect on the excitation spectrum of the dye solution in which the control vesicles were suspended, ruling out Ca²⁺ contamination or a direct effect on bilayer Ca²⁺ permeability as the source of the agonist effect on Fura-2 fluorescence (Fig. 2b). The spectra shown are representative of six experiments in which, on average (±SEM), 16.8% (±3.2%) of total vesicular calcium was released by RTX from TRV1-containing vesicles, and 1.2% (±0.09%) was released from protein-free control vesicles (P < 0.001) [see supporting information (SI) Fig. S1].

Thus, the purified, reconstituted TRPV1 is able to bind RTX and conduct Ca²⁺, as expected for a fully functional channel. This assay provides a means to test functionality of wild-type or mutant TRPV1, and because it is fluorescence based, should be readily adaptable to high-throughput screens. Addition of the Ca²⁺ ionophore, ionomycin, to the vesicles (Fig. 2) revealed that a substantial portion of the entrapped Ca²⁺ was still sequestered within vesicles after TRPV1 activation. We repeated this experiment, carefully quantifying the protein:lipid ratio and mean vesicle area (see Materials and Methods) so that we could calculate the fraction of vesicles with and without inserted channels. In those experiments (data not shown), 14.8% of the total releasable Ca²⁺ was released by RTX, whereas 20.5% of the vesicles were estimated to contain TRPV1. Thus, at least 72% of the channels are active; this number is a lower limit because the TRPV1-containing vesicles are likely more leaky than those without protein, so these results are consistent with nearly all of the purified channels being functional.

A representative section of an electron micrograph of ice-embedded TRPV1 channel is shown Fig. 3a. Images show individual molecules in what appears to be a random distribution on the grid: depending on their orientation in the layer of vitreous ice, the TRPV1 channel particles appeared as either square-like projections with a central depression or approximately trapezoidal-shaped projections with variable density distribution. This observation was further confirmed through image processing (Figs. 3b and 4). The fourfold symmetry was clearly revealed (see Fig. S2) in class-averages generated during the initial steps of image analysis without imposing symmetry (26).

The 3D structure of TRPV1 (Fig. 3c) is 150 Å tall and consists of two major regions with nearly square cross-sections connected via four bridging densities. The corners of the squares and the points of contact with the bridging densities are rotated ≈35° with respect to each other (the hand of the rotation is assigned arbitrarily). The small region is compact and comprises 30% of the total volume. It has dimensions of 60 Å × 60 Å and a height of 40 Å. The large region comprising 70% of the total mass is basket-like, containing a large open volume in its center. It measures 100 Å on a side and has a height of 110 Å. The central floor of the basket is only 18 Å thick. The large region has four radial protuberances that are twisted with respect to those of the smaller region.

The relative volumes estimated for large and small regions by using the isosurface shown in Fig. 3c are in good agreement with those expected from the sequence prediction for the cytoplasmic and transmembrane regions (70% and 30%, respectively) of the channel protein. We therefore propose that the larger basket-like region is cytoplasmic, containing the N and C termini of the channel protein, including the N-terminal ankyrin repeats and the “tetramerization domain” of the C termini; the smaller region contains the membrane-spanning parts of the four TRPV1 channel subunits as indicated by the proposed position of the lipid bilayer in Fig. 5. This topological assignment of two major channel regions with respect to the membrane is consistent with visualization of continuous densities within the smaller compact region that are sufficient to span the lipid bilayer and are large enough to encompass six transmembrane segments from each protein monomer, as shown in the cutaway view in Fig. 5. In contrast, the basket-like region exhibits a quite hollow appearance and lacks a central portion thick enough to form the ion conduction pathway across the phospholipid bilayer.

A similar two-domain arrangement referred to as a “hanging gondola” has been observed in 3D structures of K channels (17–19) and the cyclic nucleotide-gated channels (20, 21), in which the cytoplasmic region “hangs” underneath the transmembrane domain of the channel.

Fig. 6 compares the architecture of the TRPV1 channel with that of the one eukaryotic tetrameric channel protein for which a high-resolution structure is available, Kv 1.2 (27). The predicted topology of TRP channels has been proposed to resemble that of voltage-gated potassium channels (3, 8), based on biochemical and mutagenesis data. These channels include intracellular N- and C-terminal regions of variable length and six membrane-spanning helices with the “P-loop” pore-forming segment consisting of a short helix and loop between transmembrane helices 5 and 6. In addition to a similar predicted secondary structure and topology in the transmembrane domain, TRPV1 is a tetramer (based on the size and symmetry we observe) as observed for potassium channels.

Other eukaryotic homotetrameric ion channels have a two-layer architecture with distinct cytoplasmic and membrane-embedded domains, including potassium channels, the InsP₃ receptor, and the ryanodine receptor (28–33).

Because of their similar topologies and symmetries, we asked how well the crystal structure of the Kv1.2 potassium channel (27) fits into the map for TRPV1 (Fig. 6). Kv1.2 has six segments per transmembrane subunit and forms a tetramer. The size and shape of the transmembrane region of Kv1.2 manually fit well with the portion of the TRPV1 map assigned as the likely transmembrane domain, and the two layer architectures are qualitatively similar, supporting the assignment of the N- and C-terminal domains to the hanging basket.

Recently, structures of TRPV1 and TRPV2 ankyrin repeat domains were solved (34–36). The N-terminal cytoplasmic portion contains ankyrin repeats that are thought to be important for protein function, and it was suggested that the flexible structures of ankyrin domains can provide excellent scaffolds for protein–protein interaction and that they bind the channel modulators, ATP and calmodulin (36). We fitted the structures of the ankyrin repeats that have been solved for TRPV1 (36) into the 3D map of TRPV1 (Fig. 6) and found that they fit well into densities in the proposed cytoplasmic domains. Although further evidence is needed to confirm this location for the ankyrin repeats of TRPV1, it is consistent with current results and would provide an accessible location for interactions with other proteins at the interface between the cytoplasm and membrane surface.

The original rat TRPV1 cDNA was a generous gift from David Julius (University of California, San Francisco). The cDNA clone was modified by PCR to include restriction enzyme sites, SpeI at the 5′ end and MluI at the 3′ end. A 1D4 immunoaffinity tag (TETSQVAPA) was added to the C terminus to enable purification. This construct was subcloned into the yeast expression plasmid YEpHIS (41) with a strong constitutive PMA1 promoter. The protease-deficient S. cerevisiae yeast strain BJ5457 was used for overexpression. Procedures for overexpression and plasma membrane isolation were performed as described in ref. 22. Plasma membranes (2 mg/ml) were solubilized at 4°C, using 1% n-decyl-β-d-maltoside (DM) for 2 h in the presence of 20 mM sodium phosphate buffer (pH 8.0), 500 mM NaCl, and 1 μM DDT. Insoluble membranes were removed by centrifugation at 100,000 × g for 1 h, using the Type 45 Ti rotor (Beckman Coulter). The supernatant containing TRPV1 was added to an affinity column of CNBr-activated Sepharose 4B coupled to 1D4 antibody (2 ml, 3–4 mg of Ab/ml) (42), and after washing with 20 mM sodium phosphate buffer, 500 mM NaCl, 1 μM DDT, 10% glycerol, and 0.1% DM, protein was eluted by adding 0.1 mg/ml of 1D4 peptide into the washing buffer. Purified protein was resolved by electrophoresis on a 4–20% linear gradient SDS-polyacrylamide gel (Invitrogen) and transferred to a supported nitrocellulose membrane. The presence of TRPV1 was assayed by using 1D4 antibody and detected by chemiluminescence (Pierce). Polyacrylamide gels were Coomassie blue-stained (Pierce) to verify the presence of purified proteins. Protein concentration was determined by using the BCA method (Pierce). Size-exclusion chromatography analysis was performed by using a Superdex 200 HiLoad 16/60 (GE Healthcare).

TRPV1 purified protein was dialyzed in the presence of phospholipids and 5 mM Ca²⁺ to prepare unilamellar vesicles, Ca²⁺ external to the vesicles was rapidly removed by gel filtration, and then the vesicle suspension was treated with a “calcium sponge” resin to reduce contaminating metal ions to low nanomolar levels (25). A fluorescent Ca²⁺-indicator dye, Fura-2 (pentasodium salt; Invitrogen), and ligands were prepared in metal-free form by using the same resin. Vesicles that contain TRPV1 protein and 5 mM internal Ca²⁺ were transferred into 4-ml cuvettes that contained 2–4 μM metal-free Fura-2 and a small stir bar. The cuvette was placed in the chamber of a spectrofluorimeter fitted with a magnetic stirrer. The emission at 510 nm was continuously recorded while agonist RTX was added directly to the cuvette, and then the excitation spectrum was recorded. The decrease in fluorescence at 380 nm and the increase at 340 nm were used to detect calcium released from the vesicles. After RTX (50 nM) addition, ionomycin (2 μM) was added to determine the total vesicular Ca²⁺. Phospholipids were quantified as inorganic phosphate after hydrolysis to give a ratio of 2.26 × 10⁵ mol lipid per mol TRPV1 tetramer. The average vesicle surface area was determined from the root-mean-square diameter of vesicles imaged in ice (78.7 nm, n = 66). Using an average lipid surface area of 75 Ų, and assuming equal areas of both leaflets, there were 51,930 phospholipids per vesicle and 0.230 channels per vesicle. For a Poisson distribution, the fraction of vesicles with at least one channel is 1 − exp(−0.230) = 0.205.

The purified TRPV1 channel was embedded in a thin layer of ice on a holey carbon grid by using standard procedures (43). Commercially supplied Quantifoil grids coated with a “holey film” were used. They were coated with a thin continuous carbon film. The sample was quick-frozen by using a robotic instrument designed for this purpose, the Vitrobot (FEI). Frozen-hydrated TRPV1 channel specimens were imaged by using a JEM2010F electron microscope operated at 200 kV and −176°C. Images were recorded at effective magnification of 82,873 on a Gatan 4k × 4k CCD camera with an electron dose of ≈15–20 e⁻/Ų per image frame. Image processing and 3D reconstruction were performed by using the EMAN software package (26). The reconstruction from boxed particles (13,592) extracted from CCD images, spanning a range of defocus values from 1.6 to 4.9 μm, was computed by using EMAN with appropriate contrast transfer function corrections.

Particles were selected manually and boxed by using the BOXER routine (26). The particles were initially aligned by using a reference-free procedure in EMAN (26), and initial classes were sorted based on statistical criteria and averaged. A Fourier common lines procedure was used to determine the relative orientations of the class-average images, and an initial 3D model was generated. This initial model was then used to begin an iterative process of model-based alignment, classification, averaging, reconstruction of a new model, realignment using the new model, etc. In each cycle of this procedure, the current model was used to generate projections over a full range of Euler angles at user-defined angular resolution (7° steps). This cycle was repeated until the model converged, i.e., did not change significantly from one round to the next, as monitored by the Fourier shell correlation, which fell to a criterion value of 0.5 at a resolution of 19 Å after convergence for correlations between both successive iterations and arbitrarily divided halves of the data (see Fig. S3). To evaluate model bias, the structure was re-refined from randomly generated starting models, with and without imposed symmetries (from C1 to C4), and the results were compared. Projections of the models under C2 and C4 symmetry agreed well with each other and with the class averages (99 classes for C4 and 199 classes for C4; see Figs. S2 and S4), lending support for the validity of the fourfold symmetric model.