The first structure of a Kv channel, termed KvAP, from the archeabacterium Aeropyrum Pernix (4), was determined by crystallizing the channel as a complex with a monoclonal Fab fragment attached to its voltage sensors [Protein Data Bank (PDB) ID code 1ORQ] (5). In that structure the voltage sensors are in a non-native conformation, displaced toward the intracellular side of the transmembrane pore. Together with a second structure of the voltage sensor alone, termed the isolated voltage sensor (PDB ID code 1ORS), the following ideas about voltage-dependent gating were proposed (5, 6): that the voltage sensor is a very mobile structure, presumably because it must carry charged amino acids through the membrane electric field; that the charge bearing S4 forms a helix-turn-helix with the C-terminal half of S3 (S3b), and that the helix-turn-helix moves at the protein-lipid interface a large distance. Experiments using avidin capture of biotin linked to the voltage sensor indicated that four of the S4 arginine amino acids translate 15-20 Å across the membrane. These structural and functional studies formed the basis for a conceptual model in which the helix-turn-helix “paddle” element shifts its position at the protein-lipid interface, opening the pore while moving its charged amino acids (6).

Because of the non-native conformation of the voltage sensors in the crystal, the KvAP structure left two compelling questions unanswered. The first, which is important for understanding the mechanism of voltage-dependent gating, is what is the native membrane conformation of KvAP? The second question, related to the mechanism of gating but also relevant to the more general subject of membrane protein structure, is why does KvAP not maintain a native conformation in crystals? Much speculation surrounded this second question, including the wonder whether antibody fragments can “distort” membrane protein structures.

Two additional crystal forms were grown by using the detergent-like lipid 1,2-diheptanoyl-sn-glycero-3-phosphocholine. The most important difference between these crystals is that one was grown in the absence of antibody fragments and the other as a complex with monoclonal Fv fragments attached to the voltage sensors. Triclinic crystals of KvAP without antibody fragments diffracted to a resolution of 8 Å. Molecular replacement using the KvAP pore without voltage sensors (from PDB ID code 1ORQ) showed two channels in the unit cell (10). An electron density map shows clear helical segments for the pore and the voltage sensors, which were absent in the model phases (Fig. 1a). The helical organization of KvAP is discernable at this resolution. In contrast to the initial structure of KvAP (PDB ID code 1ORQ), the voltage sensors are located entirely within the “plane of the lipid membrane.” The crystal lattice is supported by contacts between the helix-turn-helix voltage sensor paddle elements, which make contacts with each other from adjacent channels (Fig. 1b).

Tetragonal crystals of KvAP in complex with Fv fragments diffracted X-rays to a resolution of 3.9 Å. Molecular-replacement phases using the KvAP pore were combined with heavy atom phases from Tl⁺-soaked crystals (10). After solvent flattening and averaging, a good-quality electron density map (Fig. 1c) allowed building of a model and refinement to R_free 0.39 (10-12). This crystal lattice is mediated by contacts between Fv fragments (Fig. 1d).

The α-carbon trace of KvAP in the 8-Å electron density map (Fig. 1a) is the KvAP-Fv complex structure after removal of the Fv fragments. It is evident that the channel structures in the two crystal forms are very similar even though one has Fv fragments bound to its voltage sensors and the other does not. The Fv fragments have not had a significant influence on the conformation of the voltage sensors. Because the KvAP-Fv complex structure is better defined it will be the focus of further consideration.

Comparison with Previous KvAP Structures. The structure of a single subunit from the KvAP tetramer (Fig. 2a) is shown from the side with the extracellular surface above. The essential features of this structure include voltage sensors (helices S1-S4) surrounding the centrally located pore, S1 and S2 helices wrapped around the pore, and helix-turn-helix S3b-S4 voltage sensor paddles positioned adjacent to the S2 helices. The major difference between this structure and the original KvAP crystal structure (Fig. 2b; PDB ID code 1ORQ) is the position of the paddle, which in this structure is drawn near to S2 at the level of the membrane rather than extended toward the cytoplasm.

In this structure the position of the paddle with respect to S2 is very similar to that observed in the crystal structure of the isolated voltage sensor, in which Arg-133 on S4 forms a salt bridge with Asp-62 on S2 (PDB ID code 1ORS) (Fig. 2c). The same salt bridge is present in the full-length channel structure. The presence of this salt bridge appears to be correlated with pH: crystals of the original KvAP structure (in which the paddle is pulled away from S2 and the salt bridge is broken) were grown at pH 5.0 (5), crystals of the isolated voltage sensor were grown at pH 7.5 (5), and crystals of the KvAP-Fv and KvAP (8 Å) crystals were grown at pH 8.5 and 9.0, respectively. Thus, conditions that favor salt-bridge formation (higher pH) support a conformation in which the paddle remains near S2, within the plane of the membrane. This conformation of the paddle with respect to S2 and salt-bridge formation involving Arg-133 is now observed in more than one crystal structure of KvAP. Shaker K⁺ channels do not function when the corresponding arginine residue (position 377) is mutated (2). For these reasons we suspect that the observed salt bridge and position of the voltage sensor paddle with respect to the S2 helix is functionally relevant.

The S1 and S2 helices in the structures have the same horizontal disposition as seen previously, although in membranes we know they must have a more vertical orientation (9, 23) (Fig. 2 a and b). In this regard the crystal structures in this study, like the original structure, do not represent a truly native conformation of the voltage sensor. The horizontal positioning of S1 and S2 cannot be attributed to the antibody fragments bound to the voltage sensor because we observe the same conformation in three structures: with Fab fragments (5), with Fv fragments, and without antibody fragments. If not antibody fragments, what accounts for the horizontal positioning of these helices?

Role of the Lipid Bilayer in Voltage Sensor Orientation. Several features of the crystal structures in this study provide a likely explanation for the horizontal orientation of the voltage sensor in detergent micelles. The rather elaborate turn connecting S2 to S3 contains five tyrosine amino acids that project out from its perimeter (Fig. 3a). In the crystal this turn is located at a level corresponding to the middle of the membrane (Fig. 2a), which is unexpected because tyrosine side chains are energetically favored to reside at the interface between the hydrophobic and head group layers (e.g., near the edge of the membrane) (14-16). The detergent-like lipid 1,2-diheptanoyl-sn-glycero-3-phosphocholine used to crystallize the channel does not form planar lipid membranes, only micelles, so a planar membrane constraint is absent in the crystal. In the context of a lipid membrane with planar boundaries we should expect the tyrosine residues to be driven toward the interface layer. A rigid body rotation of a voltage sensor domain (or of helices S2-S4) around its connection to S5, as shown (Fig. 3b), would place the tyrosine residues at the interface. The same rotation would also bring the S1-S2 loop (which has near it several tyrosine residues) to the extracellular lipid-water interface. From these energetic considerations we suggest that a planar lipid membrane constrains the voltage sensor domains with respect to the pore, and when the channel is removed from the membrane, the unconstrained domains can easily adopt a non-native orientation. This behavior is interesting because it would be expected to occur only if the domains are loosely attached to the pore as fairly self-contained structural units.

Several lines of evidence support the idea that the voltage sensors do indeed function as nearly self-contained structural domains. First, expression of the isolated voltage sensor of KvAP (in the absence of a pore) enabled us to determine its structure, and upon expression the domain is targeted to and stable in the cell membrane (5). Second, voltage dependence is transferable by splicing a voltage sensor onto a nonvoltage-dependent K⁺ channel (17). And third, a voltage-dependent phosphatase with an S1-S4 sequence attached to a soluble enzyme shows that the voltage sensor can operate on a target other than an ion channel (18).

Defining the Position of the KvAP Voltage Sensor by Analogy to Kv1.2. The structure of a related Kv channel, Kv1.2 from the rat, helps to define the native position of the KvAP voltage sensor with respect to the pore (Fig. 4a) (PDB ID code 2A79) (8). Potential distortions of the Kv1.2 voltage sensor upon extraction from the membrane are restricted by the connection of the S1 helix to a cytoplasmic domain (not shown in Fig. 4a), which does not exist in KvAP. The KvAP voltage sensor (Fig. 2a) can be adjusted to have the appearance of the sensor in Kv1.2 by rotating the sensor vertically (as was done in Fig. 3b), straightening the S4-S5 linker helix, and folding the S1 helix back onto the body of the sensor, similar to its position in the KvAP isolated voltage sensor crystal structure (PDB ID code 1ORS) (Fig. 4b). In the resulting KvAP model, the S4-S5 linker is like that in Kv1.2: it forms an amphipathic helix with a hydrophobic surface (to face the membrane) and a hydrophilic surface (to face the cytoplasm) (Fig. 5a). Furthermore, in KvAP the sharp turn connecting the S4 transmembrane helix to the S4-S5 linker interfacial helix occurs at the position of a glycine amino, which is frequently found at such turns. The straightforward transformation from the crystal structures of KvAP to the model in Fig. 4b, with preservation of chemical features that correspond well to those in Kv1.2, provides good evidence that the native structure of KvAP is similar to Kv1.2.

Biochemical Information on KvAP in Lipid Membranes. Perozo and colleagues (9) have used EPR spectroscopy to measure lipid exposure of protein surfaces of the KvAP voltage sensor. These data agree well with the position of the voltage sensor in Kv1.2 (8) (Fig. 4a) with one small discrepancy. In the Kv1.2 crystal structure, one edge of the S4 helix leans against the S5 helix of a neighboring subunit; however, the corresponding amino acids in KvAP exhibit high lipid exposure according to the EPR data (Fig. 5 b and c, residues 124 and 127). In addition, a face of S1 in Kv1.2 that would be expected to exhibit high lipid exposure (because it is on the lipid-facing surface) appears in KvAP to be in a protein environment (i.e., neither lipid nor water) according to the EPR data (Fig. 5 b and c, residues 31, 35, and 39). These differences between expected lipid exposure on the basis of the Kv1.2 crystal structure and observed lipid exposure of corresponding amino acids in KvAP suggest that the voltage sensor in KvAP might be rotated slightly with respect to the pore, compared with the sensor in Kv1.2. The rotation of the voltage sensor that would be required to account for the EPR data is very modest and in fact has already been made in the model shown in Fig. 4b, relative to the Kv1.2 voltage sensor in Fig. 4a. That small variations in the orientation of the voltage sensors might occur between KvAP and Kv1.2 is not surprising, especially because the constraint imposed by the linker to the T1 domain in Shaker family Kv channels (e.g., Kv1.2) does not exist in KvAP.

We carried out disulfide cross-bridge studies to further examine the relationship between KvAP and Kv1.2. Several studies have shown that a cysteine residue introduced at the position of the first arginine residue on S4 (the extracellular-most arginine, corresponding to position 117 in KvAP) in the Shaker K⁺ channel (which is very similar to Kv1.2) readily forms a cross-bridge with a cysteine introduced near the extracellular end of S5 of a neighboring subunit (19). We observe a qualitatively similar result with KvAP channels in membranes (Fig. 5d). In KvAP, however, we do not observe the same quantitative degree of complete covalent tetramer formation that is observed in the Shaker K⁺ channel (19). The data on KvAP are thus consistent with the voltage sensor being in a similar location as in Shaker (or Kv1.2), but the lower efficiency of tetramer formation is consistent with the possibility that S4 may be slightly further away from S5, as the EPR data suggest.

For proteins like KvAP the detergent micelle may be an imperfect mimic of the lipid bilayer, but detergent-mediated crystallization is still a valuable approach for understanding such proteins. A non-native protein structure simply means that additional information is required to help interpret what we see. In the present analysis we have combined crystal structures of the KvAP channel with biochemical cross-bridge studies in membranes, EPR data from the literature, and comparisons with a related Kv channel structure to arrive at a plausible answer to the first question raised in the Introduction: what is the native structure of KvAP in lipid membranes? The model is shown in Fig. 6. Coordinates for the model shown in Fig. 6 are available on request from R.M.

This proposed KvAP structure is quite similar to the Kv1.2 channel structure, which was used to help guide its construction (8). The pore conformation corresponds to that observed in two KvAP crystal structures (PDB ID codes 1ORQ and 2A0L) with the S6 inner helices adjusted slightly to accommodate their interaction with the S4-S5 linker helices. The S6 inner helices near the intracellular pore entryway have a wide diameter (≈12 Å), suggesting that the pore is open. The conformation of the voltage sensors is very similar to the sensor in the crystal structure of KvAP (PDB ID code 2A0L) but with S1 moved near to its position in the isolated voltage sensor structure (PDB ID code 1ORS) (5). The voltage sensor domains in KvAP, compared with those in Kv1.2, are rotated slightly about an axis perpendicular to the membrane plane relative to the pore in such a way that S1 is more buried in protein and S4 is more exposed to lipid to account for lipid exposed surfaces mapped by EPR analysis (9). The conformation of the voltage sensor paddle (the helix-turn-helix formed by S3b and S4) is near to an open conformation because the four conserved arginine residues on S4 are near the extracellular surface.

The turn connecting S2 to S3, which is not resolved in the structure of Kv1.2, is defined by several crystal structures of KvAP. On the basis of amino acid sequence analysis we think that this turn will be conserved in voltage-dependent K⁺ and Na⁺ channels. In particular, there is a salt-bridge pair that is always conserved, with a nearly fixed number of amino acids between them (see Fig. 7, which is published as supporting information on the PNAS web site). In KvAP, and probably other voltage-dependent channels, we propose that one of the functions of this turn is to serve as a membrane interface anchor. The various crystal structures of KvAP suggest that this turn also serves a second function, to allow the voltage sensor paddle to undergo large movements. We base this statement on the observation that the turn is different in its conformation from one crystal structure to the next, indicating that it is intrinsically flexible (PDB ID codes 1ORS, 1ORQ, and 2A0L).