Structure of potassium channels

Potassium channels ubiquitously exist in nearly all kingdoms of life and perform diverse but important functions. Since the first atomic structure of a prokaryotic potassium channel (KcsA, a channel from Streptomyces lividans) was determined, tremendous progress has been made in understanding the mechanism of potassium channels and channels conducting other ions. In this review, we discuss the structure of various kinds of potassium channels, including the potassium channel with the pore-forming domain only (KcsA), voltage-gated, inwardly rectifying, tandem pore domain, and ligand-gated ones. The general properties shared by all potassium channels are introduced first, followed by specific features in each class. Our purpose is to help readers to grasp the basic concepts, to be familiar with the property of the different domains, and to understand the structure and function of the potassium channels better.


Introduction
Potassium (K ? ) channels locate in cell membranes and control transportation of K ? ions efflux from and influx into cells. They play crucial roles in both excitable and non-excitable cells and can be found in virtually all species, except for some parasites [1]. K ? channels have transmembrane helices (TMs) spanning the lipid bilayer. Based on the structure and function, the channels are categorized into three major classes: the voltagegated (Kv) (six TMs), inwardly rectifying (Kir) (two TMs), and tandem pore domain (K2P) (four TMs) channels [2]. Furthermore, the ligand-gated (Kligand) channels have either two or six TMs and are stimulated by various messengers. A K ? channel can, independent of which class it belongs to, be divided into two parts: the pore-forming domain and the regulatory domain. The pore-forming domain is responsible for transportation of K ? ions and its structure is similar in all types of K ? channels. The regulatory domain senses diverse stimuli and its structure differs among the classes.
In this review, the structure of K ? channels is discussed ( Table 1). The channel containing only the pore-forming domain is introduced first, followed by Kv, Kir, K2P, and Kligand channels. Since the pore-forming domain is shared by all of the channels, only specific features in each class are discussed. The function of one example in each class is given in more detail to demonstrate the importance of K ? channels for living cells.
In cells many K ? channels are regulated by their own auxiliary subunits, which profoundly affect the K ? channel physiological activities. Since these subunits are specific to each channel and the structural information of them is less well known, they are omitted in this review. Interested readers can refer to (Kv [3], Kir [4], K2P [5], and Kligand [3,6]) for more detailed information. ions move [7]. The general structure of the pore-forming domain can be described by the transmembrane part of KcsA, a two TMs K ? channel from Streptomyces lividans [8]. The active site of K ? channels is composed of four conserved signature sequences, TVGYG 75-79 functioning as a selectivity filter (SF) to conduct K ? ions. K ? ions are conducted very efficiently, at near diffusion-limited rates (10 7 ions channel -1 s -1 ) [9]. Simultaneously, K ? channels are highly selective and at least 10,000 times more permeant for K ? than sodium (Na ? ) ions [8]. Besides the feature with regard to selectivity and conductivity, K ? channels are tightly regulated. Another interesting property is that many K ? channels can be inactivated, meaning that they enter stable nonconductive states shortly after opening. One kind of inactivated state is closely coupled to a conformational change of the SF.

Selectivity and conductivity
The mechanism of selectivity and conductivity of K ? channels is well studied in KcsA [8,10,11]. The structure of KcsA at high K ? ion concentrations (protein data bank (PDB): 1K4C [11]) is demonstrated in Fig. 1a. The K ? ions usually go from the intracellular side (helical bundle), then enter the central water-filled cavity (Sc), followed by passing the SF (S4-S1), and finally reach the extracellular entryway (S0 and Sext), down the electrochemical gradient. The K ? ions are hydrated in the central cavity, dehydrated in the SF, and then rehydrated in the extracellular entryway. The noticeable feature is that there are four evenly spaced K ? binding sites (S1-S4), which are formed by the carbonyl oxygens of TVGYG and the side chain of threonine. Four K ? ions can bind to these sites, where each K ? ion sits in the middle of two oxygen layers (Fig. 1b, c). Since the arrangement of these protein oxygens can mimic the water oxygens surrounding a K ? ion in solution, the transfer energy of the K ? ions from the central cavity to the SF is low [7]. The net result is that the conduction of K ? ions occurs at rates near the diffusion limit. The proposal discussed above explains the mechanism of conduction perfectly and the selectivity could be explained by the fact that permeation of Na ? ions is energetically unfavorable, since the radius of a Na ? ion is smaller than a K ? ion [8,12]. In fact, it has been reported that Na ? ions and their bound protein oxygens are in the same layer, allowing shorter ligand-ion distance [13]. A similar structure is observed for Li ? ions [14]. However, if K ? ions bind the S1-S4 sites tightly due to their high selectivity, how can they be released from these sites and transported across the membrane at a very fast speed? The near diffusion-limited rate of conduction suggests that K ? ions have weak binding in the protein. After analyzing the ion occupancy in the SF by different methods, MacKinnon and his colleagues proposed that only two ions exist in the SF at a time and they adopt two configurations, in which the ions are either in S2 and S4 (2,4-configuration, Fig. 1b) or in S1 and S3 (1,3-configuration, Fig. 1c) with two water molecules occupying the two corresponding vacant binding sites [10,11,15,16]. The queue of ions and water molecules move in a concerted manner by either a concentration-independent path when the 2,4-configuration shifts to the 1,3-configuration or a concentration-dependent path when a third ion enters from one side of the SF and another ion exits from the opposite side [7,10,15]. The Fig. 1 The transmembrane part of KcsA. a The atomic structure of KcsA in the conductive state (PDB: 1K4C) viewed along the membrane plane. The pore-forming domain consists of the outer helix (magenta), loop regions (green), pore helix (blue), SF (yellow), and inner helix (orange). The conducted K ? ions are represented by purple balls with surrounding water molecules in red. EC is extracellular and IC is intracellular for short. The glycine hinge (Gly99) and the helical bundle are labeled. b, c The enlarged view of the boxed area in (a) containing the SF and the extracellular entryway. The K ? ions are in two configurations, either in S2 and S4 (b) or S1 and S3 (c) during conduction. The water molecules occupy the vacant ion positions in S1 and S3 (b) or in S2 and S4 (c). Other ions are located in the extracellular entryway (either S0 (b) or Sext (c)) and in the central cavity (Sc (a)). For clarity, only two monomers opposite to each other are shown. The amino acid sequence of the SF is labeled. All figures (Figs. 1,2,3,4,5,6,7,8) in this paper were made using Chimera [130] and GNU Image Manipulation Program (GIMP) Structure of potassium channels 3679 transfer energy cost between two configurations is expected to be low [10] and S2 favors binding of K ? ions, but not Na ? ions [17,18]. The repulsion between two ions in the doubly occupied SF facilitates the conduction as well [7,15]. K ? ions stabilize the conductive conformation of the SF, which in turn favors conduction of K ? ions [12,19]. On the other hand, Na ? ions may stabilize a nonconductive state, which has a distorted structure of the SF (structure in low K ? ion concentrations, PDB: 1K4D [11]) [20]. The SF in the conductive state can conduct Na ? ions as well, when K ? ions are absent [20]. However, under physiological conditions, two K ? ions residing in the SF prevent Na ? ion conduction [20]. Thus, the SF and K ? ions work together to keep the selectivity and conductivity. Furthermore, the amino acid residues surrounding the SF could contribute to maintaining the conductive state of the SF [8,12,[21][22][23]. In addition, other studies reveal that one or more of the following effects are involved to explain the selectivity and conductivity: different ion occupancies between two configurations [24], different binding sites favoring different ions [14,17,18,25], ion concentrations in solution [26], the number of binding sites [27], the effect of the already bound ions in the SF [28], water molecules in the central cavity [29,30], existence of the pore helix [8,29], and different conformations of the SF [21,22].
To summarize, the SF acts as an elegant machine to conduct K ? ions efficiently among other available ions. All factors that influence the integrity of the SF, even distant amino acid residues away from the SF, may change the selectivity and conductivity of the channel.
Gating K ? channels usually have three states: resting, activated, and inactivated. The channels are usually closed in the resting state, and opened after stimuli activation, followed by turning to the nonconductive states. The gating is the process to control closing and opening of the channel [31]. There are two kinds of gating mechanisms: the intracellular one is at a position where the inner helix bends and the extracellular one includes the SF [32]. These two gates are coupled, but the effects of coupling vary in different K ? channel classes. For example, the two gates in Kv channels (introduced in the ''Voltage-gated potassium channels'') are negatively linked to make them easily enter the inactivated states, whereas the two gates in the K2P channels (introduced in the ''Tandem pore domain potassium channels'') are positively coupled to facilitate their constitutive opening [33]. The coupling is involved in the arrangement of the K ? channel, mainly in the area at the lower part of the SF [32,[34][35][36][37].

The intracellular gate
The KcsA structure described above (PDB: 1K4C) is in a closed conformation, where the inner helices cross near the intracellular membrane interface (Fig. 1a). The open structure is represented by MthK, a calcium-gated K ? channel from Methanobacterium thermoautotrophicum, in which the inner helices are bent and splayed out following a glycine residue [38]. This glycine hinge is observed in other K ? channels, such as in KvAP, a voltage-gated K ? channel from Aeropyrum pernix [31]. Figure 2 shows the structural comparison of KcsA, KvAP, and MthK. The glycine hinges (G99 in KcsA, G220 in KvAP, and G83 in MthK) are located in a similar position in these proteins. In many eukaryotic Kv channels, this glycine hinge is replaced by P-X-P, where P is proline and X is any amino acid residue [39]. P-X-P allows the sixth helix (S6, corresponding to the inner helix in KcsA) to bend to interact with the linker between the pore-forming domain and the voltage sensor domain. In this way, the signal can be transferred from the voltage sensor domain to the poreforming domain [40].

The extracellular gate
Since the SF controls conduction of K ? ions, it can serve as a gate at the extracellular side as well. Indeed, the SF adopts different structures in the resting, activated, and inactivated states [32]. Understanding the structure of inactivation is of considerable clinical and pharmaceutical interest. In Kv11.1, this process is critical for normal cardiac repolarization and disturbance is related to unintended side effects of arrhythmia and sudden death [41].
There are two mechanisms for inactivation of the K ? channels: N-and C-types. The former is a fast autoinhibitory process, existing in some Kv channels, where the N-terminal part interacts with the open K ? channel and occludes it [42,43]. In the proposed N-type inactivation model [42], the first three amino acid residues at the N-terminus (inactivation ball) bind to the central cavity, the following eight hydrophobic amino acid residues extend from the cavity to the intracellular entryway, and the subsequent nine hydrophilic amino acid residues interact with the aqueous protein surfaces of the cytosolic domain (T1-S1 linker region). A mutagenesis study supports the view that electrostatic interactions between the inactivation ball and the T1-S1 linker region facilitate N-type inactivation [44]. C-type inactivation is usually a slow process and results from conformational changes of the SF, together with elimination of K ? ions and water molecules [32,45]. The nonconductive structure obtained at low K ? ion concentrations (PDB: 1K4D) may represent an inactivated conformation [46]. In fact, the SF of this structure (PDB: 1K4D) resembles the one from an open-inactivated structure (PDB: 3F7V [45]), although the intracellular gate is closed in the former and open in the latter.
The majority of K ? channels, including both prokaryotic and eukaryotic ones, undergo C-type inactivation [36]. A molecular mechanism is proposed based on a series of KcsA structures with different degree of opening of the intracellular and extracellular gates [45]. When both gates are opened, the channel conducts K ? ions. As the intracellular gate opens to a certain degree (the distance between two T112 residues on diagonally positioned inner helices [17 Å ), the SF starts to change its conformation from the conductive (activated) state to the nonconductive (inactivated) state (Fig. 3a). The first step may be pinching of the G77 backbone carbonyls together. The rearrangement of oxygens destabilizes the ion in the S2 site, which might further destabilize the ion in the S3 site and narrow V76 as well. As a result, ions only occupy the S1 and S4 sites and two amino acid residues in the SF (G77 and V76) are reorientated. This prevents conduction of K ? ions in the nonconductive state. Another noticeable SF structure, in the so-called flipped conformation (Fig. 3b) is reported from the E71A mutant of KcsA [21,22]. Whether this conformation represents a conductive [22] or a nonconductive state [21] has not received agreement yet. The presence of negatively charged lipids has been reported to be important for ion conduction in KcsA [47,48] and binding of these lipids may decrease C-type inactivation [48].

Voltage-gated potassium channels
The ability to learn, memorize, and perceive depends on the exchange of signals among neurons [49]. The action potential, the electrical signal generated by nerve cells involves several types of voltage-gated ion channels [50]. Our understanding of the action potential is based on the analysis of the squid axon [51], where voltage-gated Na ? (Nav) channels open for a short period followed by a rapid inactivation; after a short while, Kv channels are activated b Comparison of conductive (PDB: 1K4C, black) and flipped (PDB: 2ATK [21] and 3OGC [22] are similar and 2ATK is shown in gray) structures. V76 and Y78 are reorientated in the flipped conformation. The S2 and S4 binding sites are labeled and remain open for a longer period [49,50]. The net result is that when the action potential travels on depolarization, Na ? ions influx to the cell and K ? ions efflux to the extracellular environment. Action potentials have different functions in the neuron cell bodies and in the axons; in addition, various types of neurons have their own patterns of action potentials [50].
The majority of Kv channels open when the membrane is depolarized (less negative inside) and close when the membrane is hyperpolarized (more negative inside) [52]. The Kv channel has six TMs (Fig. 4a) and the first four helices (S1-S4) form the voltage sensor domain (VSD) [31,40,53,54]. The last two helices (S5-S6, corresponding to the outer and inner helices in KcsA, respectively) form the pore-forming domain. The VSD senses the membrane potential alteration, followed by a conformational change that is coupled to gate the poreforming domain.

Voltage sensor domain
Considerable evidence supports the idea that the VSD functions as an independent domain. This domain is portable among K ? channels [55][56][57] and also exists in Nav and voltage-gated calcium (Cav) channels [58], the voltage-gated proton channel 1 (Hv1, also known as VSOP) [59,60], and the voltage-sensor-containing phosphatase [61]. In addition, the solved structures (PDB: 2A79 [54], 2R9R [53], 3BEH [62], and 3LUT [63]) show that the VSDs are located at the periphery of the channel and are weakly attached to the pore-forming domain. Furthermore, although the structure of the VSD in KvAP (PDB: 1ORQ [31]) resembles the ones in other K ? channels (PDB: 2A79, 2R9R, and 3LUT), its position is distorted as compared to others [64]. The structural comparison of published VSDs including those in Nav and transient receptor potential channels shows that the VSDs adopt different orientations Fig. 4 The VSDs in channels. a Alignment of monomers of different channels, viewed from the extracellular side. When pore-forming domains are aligned, the VSDs adopt various orientations. The VSD (in an ellipse) is composed of the first four helices (S1-S4). The poreforming domain (in a box) consists of S5 (corresponding to the outer helix in KcsA in Fig. 1) and S6 (corresponding to the inner helix in KcsA in Fig. 1). The pore helix is labeled as PH. b Alignment of published VSDs structures, viewed along the membrane plane. b Shows an enlarged side view, rotated 90°from (a). Different VSDs are compared: Kv1.2 (a Kv channel from Rattus norvegicus, PDB: 3LUT, light magenta), MlotiK1 (a non-voltage-gated K ? channel from Mesorhizobium loti, PDB: 3BEH, light brown), NavAb (a Nav channel from Arcobacter butzleri, PDB: 3RVY [131], light green), NavRh (a Nav channel from Rickettsiales sp. HIMB114, PDB: 4DXW [132], light orange), and TRPV1 (a transient receptor potential channel from Rattus norvegicus, PDB: 3J5P [133], light blue). Although the VSDs adopt different orientations in the channels (a), they show a substantial overlap when only these domains are compared (b). The VSDs in PDB: 3LUT, 3RVY, and 4DWX are aligned best. Two loop regions between S1-S2 and S3-S4 in PDB: 3LUT are omitted. The resembling structures (PDB: 2A79, 2R9R, 3RVZ [131], 3RW0 [131], 4EKW [134], 3J5R [37], and 3J5Q [37]) are not depicted. The VSD of KvAP is not shown either, since the solved structure is either distorted (PDB: 1ORQ) or resembles (PDB: 1ORS [31] and 2KYH [70]) the one in Kv1.2 (PDB: 3LUT) (Fig. 4a), although the structures themselves in these channels show a substantial overlap (Fig. 4b).
In the native conformation (PDB: 2A79, 2R9R, 3BEH, and 3LUT), the sensor domain from one subunit loosely contacts with the pore-forming domain from the adjacent subunit. These two domains interact in two coevolved interfaces (Fig. 5a) [65]. The first part is where the S4-S5 linker and S6 interact. The second part is where S1 in the VSD and the pore helix in the pore-forming domain interact. The first one is on the intracellular side and the second one is on the extracellular side. The large empty space between the VSD and the pore-forming domain is occupied by lipids (PDB: 2R9R), which play important functional and structural roles for Kv channels [52,53,64]. The VSD of Kv1.2 channels in the open conformation (PDB: 2A79, 2R9R, and 3LUT) contains a water-filled crevice at the extracellular side, followed by a hydrophobic region (Fig. 5b) [52,53,63].

The positive charges and the paddle motif in the voltage sensor domain
Positively charged amino acid residues (arginine and lysine) in S4 make Kv channels electrically sensitive. In Shaker, a well-studied Kv channel from Drosophila melanogaster, approximately 13 positive charges are displaced across the membrane during activation [39,66]. Up to eight positively charged residues have been found in each S4 and this number varies among different channels [1]. Shaker has seven such residues, denoted as R1-R4, K5, R6, and K7, of which R1-R4 contribute to most charge movement during activation [52,66]. The positively charged residues are separated by hydrophobic residues and this triplet residue pattern (one positively charged residue and two hydrophobic residues) is evolutionarily conserved [1] and necessary for voltage sensing. The triple residue pattern, rather than the specific sequence, accounts for charge translocation [57]. The positive charges are counterbalanced by several negatively charged residues in other helices (E183, E226, E154, E236, and D259 in the Kv1.2-Kv2.1 chimera channel, based on PDB: 2R9R). The electrostatic interactions are believed to assist the movement of S4 in the membrane bilayer.
A conserved helix-turn-helix motif composed of the second helix in S3, S3b, and the N-terminal half of S4 is proposed to move together during activation [31,67]. This motif, which contains R1-R4, was termed as the voltage sensor paddle. Similar to the entire VSD, the paddle motif is portable [53,56,57], suggesting that this motif resides in a relatively unconstrained environment. Although S3b and S4 may couple together as proposed in KvAP, some accessibility experiments suggest that S3 and S4 are movable relative to each other [68,69]. One explanation for this contradiction could arise from the properties of VSD in different Kv channels. For instance, S3 breaks into S3a and S3b in KvAP (PDB: 1ORS [31] and 2KYH [70]), but is a continuous helix in Kv1.2 (PDB: 2A79, 2R9R, and 3LUT), whereas S3b is almost absent in KvLM (a Kv channel from Listeria monocytogenes) [71]. Other reasons are discussed in [39].

Models for voltage sensing
VSDs exist in both depolarization-and hyperpolarizationactivated Kv channels. It has been suggested that S4 has the same outward movement during activation, whereas the coupling between the VSD and pore-forming domain is opposite in these two kinds of channels [72,73]. How does the VSD translocate the charges across the membrane when the membrane potential changes? The transporter, helical screw, and paddle motif models have been previously proposed to answer this question (reviewed in [39]). In all models, several explanations are shared: (1) translocation of the charges is carried out by the movement of the positively charged amino acid residues in S4; (2) S4 can rotate, translate, and adopt different helical conformations; and (3) the positive residues in S4 are counterbalanced by the negative residues in other helices in the VSD. The main differences are how far S4 has been moved vertically in the membrane and what environment S4 faces. With the atomic structures of Kv channels (PDB: 2A79, 2R9R, and 3LUT), as well as new biophysiological data (such as the observed omega current [74,75] and proton leakage [76,77]), a consensus model is appearing [39,53,63,78]. In this focused electric field model (named in [63]), a hydrophobic region of approximately 10 Å thickness separates the external and internal solutions (Fig. 5b). Therefore, the charges only need to travel through the focused field, rather than across the entire membrane. Two structurally conserved negative residues in S2, which interact with a pair of positive residues in S4, are located at both sides of this hydrophobic region (Fig. 5b). E226 is on the extracellular side and E236 on the intracellular side (E226 and E236 are in the Kv1.2-Kv2.1 chimera channel, based on PDB: 2R9R).
In the resting state, R1 interacts with E226 and R4 with E236 [78]. On depolarization, S4 moves upwards to the extracellular side. A pair of positive residues (R1-R2, R2-R3, and R3-R4) pass through the hydrophobic region sequentially and each time interact with E226 (and maybe E236 as well) [78]. At least two positively charged residues are required to maintain the voltage-gated function [57]. The activated conformation is evident from the solved Kv1.2 structures (PDB: 2R9R and 3LUT), in which R1-R2 expose to the extracellular crevice and interact with the lipid; R3-R4 stay in the hydrophobic region and interact with E226; and K5-R6 expose to the intracellular crevice and interact with E236 (Fig. 5b) [53]. The intracellular crevice is deduced to be significant in the resting state, but may vanish in the activated state [53,78]. The helical segment of S4 adopts a 3 10 -helical conformation when passing through the hydrophobic region [53]. This special helical conformation extends the helical length of S4 so that the paired positive residues can have the same orientation and interact with E226. The stability of this conformation may be affected by the protein environment from other helices in the VSD as well [79]. The S4 segment outside the hydrophobic region at both sides retains its ahelical conformation and the length of the 3 10 -helix may vary in different states [80]. F233 is proposed to have a structural role for facilitating formation of the electrostatic interactions between the positively and negatively charged residues [53,81,82]. The electrostatic interactions between the charged residues in Kv channels may be replaced by the hydrogen bonds in the non-voltage-gated K ? channels [83].
The structure of a Kv channel in the resting state has not been determined experimentally yet. The models are generated by computer simulation with the restriction from the available biochemical and biophysical data (e.g., [78,80,82]). Although these models generally agree with the current interpretation of the experimental data, the resting structure is still in question due to several uncertainties, such as the multiple sub-states of the channel and the network interactions between the charged residues. Thus, the focused electric field model, proposed interactions, and the computationally simulated structures may still be finetuned in the future.
Lipids are closely associated with both the VSDs and the pore-forming domains of Kv1.2-Kv2.1 chimera channel (Fig. 5a). The phospholipid head groups interact with the positive charges in S4 in the activated structure (PDB: 2R9R) [53]. The functional studies showed that lipid composition causes the VSDs to switch conformations in KvAP: phospholipids are required for it to reach the activated state, and non-phospholipids could stabilize its VSDs in the resting state [84]. Thus, gating of Kv channels is both lipid-and voltage-dependent [53,84].

Inwardly rectifying potassium channels
Kir channels have diverse physiological functions in the cell, depending on their type and location, and are modulated by various mediators, such as ions, phospholipids, and binding proteins [4]. Kir channels can be divided into seven subfamilies (Kir1.x-Kir7.x, where x is the number of each member) based on their mediators and the properties of ion conduction [4]. In vivo, they can be either homo-or hetero-tetramers [4].
Kir6.x channels are sensitive to nucleotides and are involved in glucose homeostasis. In pancreatic b-cells, Kir6.x channels and their partners (sulfonylurea receptor (SUR) subunits) work together to control insulin secretion [4]. Mutations in the Kir6.x or SUR gene result in a range of diseases and in fact, drugs targeted to SUR are routinely used to treat type 2 diabetes [85]. Kir6.x channels together with the SUR subunits have also been found in cardiac, smooth muscle, and brain nerve cells [4].
The unique feature of Kir channels is that they conduct K ? ions on hyperpolarization, rather than on depolarization as in other K ? channels. The inward rectification occurs because they are blocked by intracellular magnesium ions and polyamines on depolarization, whereas these blockers are released on hyperpolarization to allow K ? ions to influx into the cell [4]. The Kir channel contains a pore-forming domain and a cytosolic domain, where the pore-forming domain is responsible for ion conduction while the cytosolic domain regulates the gating of the channel.

The cytosolic domain
The C-terminal cytosolic domain is rich in b-sheets and is located below the pore-forming domain, extending the ion conduction pathway. The cytosolic domain also forms a binding site to interact with diverse intracellular regulatory mediators [4]. Multiple ion binding sites (D173, E225, D256, and E300 in Kir2.2, Fig. 6b, c) in this domain are conserved and critical to inward rectification [86,88,95,96]. In addition to the intracellular and extracellular gates in the transmembrane part, the Kir channel has a third gate (called G-loop), which is located at the apex of the cytosolic domain and forms a girdle around the central fourfold axis (Fig. 6b, c) [95]. This gate is intrinsically flexible and indeed, some solved Kir channel structures have these gates closed whereas others have them opened [86, 88-92, 95, 96].
Since the cytosolic domain is relatively independent from the pore-forming domain, it may adopt various conformations in the resting, activated, and inactivated states. An interesting paper illustrates a series of structures of KirBac3.1 (a Kir channel from Magnetospirillum  view, rotated 90°from (a). The pore-forming domain locates above the cytosolic domain. The G-loop (residues from 301 to 311, magenta), and conserved multiple ion binding sites (gray) are labeled. The PIP 2 lipids (yellow) are located at the interface between two domains. c The extended ion conduction pathway in the cytosolic domain, an enlarged view of (b) magnetotacticum), which adopts twist, non-twist, latched, unlatched, and semi-latched conformations [97]. Reorientation and rotational movement of the cytosolic domain correlate with the ion configuration in the SF as well as binding of polyamine [97]. Growing evidence supports the view that rotation of this domain facilitates gating of the channel, although the rotational angles deviate in prokaryotic [90,92] and eukaryotic [91] channels.

Lipid regulation
Activation of the Kir channels depends on the signaling lipid phosphatidylinositol 4,5-bisphosphate (PIP 2 ) [98]. As shown in Fig. 6b, PIP 2 binds at the interface between the pore-forming domain and the cytosolic domain in Kir2.2. PIP 2 induces a 6 Å translational movement of the cytosolic domain towards the membrane layer, concomitant with local conformational changes [93]. Interestingly, these changes are induced by binding of the head group of PIP 2 to the cytosolic domain [93]. Besides Kir2.2, it has been observed that the cytosolic domain of Kir3.1 (a Kir channel from Mus musculus)-KirBac1.3 (a Kir channel from Burkholderia xenovorans) chimera in the open state gets closer to the pore-forming domain as compared to the closed state [87]. Although PIP 2 usually activates eukaryotic Kir channels, it has an opposite effect on prokaryotic channels [99].
Kir3.2 (a Kir channel from Mus musculus, GIRK2) is activated by PIP 2 , as well as G proteins or Na ? ions [89]. Based on the structures from wild-type Kir3.2 with (PDB: 3SYA [89]) and without (PDB: 3SYO) PIP 2 , and R201A mutants with (PDB: 3SYQ [89]) or without (PDB: 3SYP [89]) PIP 2 , it was concluded that binding of PIP 2 alone (in a similar position as in Kir2.2 shown in Fig. 6) does not open the gates in the pore-forming domain or the G-loop in the cytosolic domain, whereas G proteins alone open the G-loop but not the intracellular gate. When both PIP 2 and G proteins are present, the channel opens [89]. Furthermore, it was proposed that the interaction between the Na ? ion and D228 could promote a similar conformational change as with G proteins [89,100].
Tandem pore domain potassium channels K2P channels are abundant in both excitable and non-excitable cells, where they play diverse functions. The K2P channels are regulated by a variety of mediators, e.g., ions, pH, lipids, and regulatory proteins [2]. They set resting membrane potential and are targets of volatile anesthetics [5].
The TRAAK channel (TWIK-related arachidonic acid activated K ? channel, KCNK4, K2P4.1) exists exclusively in brain, spinal cord, and retina in mouse [101]. It is partially inhibited by barium ions at high concentrations, but is insensitive to other classical K ? channel inhibitors [101]. TRAAK can be stimulated by unsaturated fatty acids [5,101]. Considering that these lipids exert both anti-ischemia and anti-convulsant effects, TRAAK channel was suggested to be a candidate in neuroprotection [5,101,102]. Indeed, riluzole, a neuroprotective drug is found to stimulate expression of the TRAAK channel [101]. Apart from the unsaturated lipids, the TRAAK channel can be activated by mechanical force, elevation of temperature, and alkalization from the intracellular side [5,103,104].
The K2P channels have several unique features. Functionally, the K2P channels are usually constitutively open [105], whereas the other K ? channels are tightly regulated in their closed and open states. Thus, the principle gating site might be the extracellular gate (see also C-type inactivation discussed in ''The extracellular gate'') [104,106]. Structurally, each mammalian K2P channel has four TMs and two pore-forming domains. Thus, the biological assembly of a K2P channel is a dimer [103,104,[106][107][108][109].

Human K2P structure
The overall structure of human TRAAK (PDB: 3UM7 [108], 4I9 W [109], 4RUE [104], 4RUF [104], 4WFE [103], 4WFF [103]), TREK-2 (KCNK10, K2P10.1, PDB: 4BW5 [106], 4XDJ [106], 4XDK [106], 4XDL [106]), and TWIK-1 (KCNK1, K2P1.1, PDB: 3UKM [107]) resembles each other, regardless of the up and down movements of the inner helices in different states (see below). Each of them shows a rhomboid-shaped helical cap (Fig. 7a), which is not observed in other solved ion channel structures [108]. This cap is located on top of the transmembrane part of each K2P channel and is formed by the extracellular region within the first pore-forming domain, but is absent in the second pore-forming domain. In contrast to other K ? channels, K2P channels may exhibit domain swap where the units of the outer helix 1 (red M1 in Fig. 7b)-cap helix 1 (not shown in Fig. 7b) exchange between two subunits [103,104,109]. More importantly, the domain-swapped K2P channels are concluded to exist in the cell membranes [109]. The cap makes K ? ions accessible only from the bifurcated extracellular pathway [107,108]. The coordination of ions in the SF in most structures resembles those from KcsA (PDB: 1K4C), except for those in the down state of TREK-2 (PDB: 4XDJ, 4XDK, and 4XDL) where only three ions were found in the SF.

Up and down states of K2P
A series of K2P structures illustrates the gating/activation mechanism of the channel (PDB: 3UM7, 4I9W, 4RUE, 4RUF, 4WFE, 4WFF, 4BW5, 4XDJ, 4XDK, 4XDL, and 3UKM). The essence is the movement of the inner helix 2 (blue M2 in Fig. 7b). In the down state (PDB: 3UM7, 4WFF, 4XDJ, 4XDK, 4XDL, and 3UKM), the intracellular side of M2 is straight; whereas in the up state (PDB: 4I9W, 4RUE, 4RUF, 4WFE, and 4BW5), M2 is kinked approximately halfway through the membrane around the hinge glycine (G268 in TRAAK, G312 in TREK-2, and G256 in TWIK-1). The consequence of the up movement of M2 is that the lateral openings, which provide binding sites for regulatory lipids or for hydrophobic inhibitors to interact, are closed [103,104,106,109]. Therefore, no lipid binds to the channel in the up state. On the other hand, lipids have been modeled in a similar position (Fig. 7b) in the down state of TRAAK (PDB: 4WFF), TREK-2 (PDB: 4XDJ), and TWIK-1 (PDB: 3UKM). The down state is believed to represent a nonconductive state, since the lipid occupies the central cavity thus blocking conduction of K ? ions [103,106]. In accordance with the elevation movement of M2, other helices change their conformations [103,104,106]. These movements work together to gate the channel.

Ligand-gated potassium channels
Most Kligand channels have either two or six TMs, and a cytosolic domain usually at the C-terminus, although some of them have additional TMs such as in BKca (a large conductance K ? channel from Homo sapiens, both voltage and calcium gated). The transmembrane part of a two TMs Kligand channel is similar to KcsA, and that of a six TMs resembles a Kv channel. The cytosolic domain in the Kligand channel acts as a receptor domain for binding of various messengers, including cAMP [62], calcium [110][111][112][113][114][115], and NADP [116,117].
BKca participates in many biological processes including generation of action potentials, modulation of the tone of blood vessels, and release of hormones and neurotransmitters [118,119]. It has been suggested that BKca can coassemble with multiple types of Cav channels to form macromolecular complexes in the central nervous system [50,119]. Thus, complex formation with distinct Cav channels can tune the BKca activity and further control the diverse processes in which BKca is involved [119].
Although ligand-binding domains in diverse Kligand channels have their own structures and properties for interacting with various messengers, the general gating mechanisms among the channels are similar. The ligandbinding domains in BKca and its homologs are discussed in the following text as an example for other Kligand channels.
The fourfold symmetric RCK octameric gating rings BKca and MthK have large cytosolic domains, called RCK (the regulator of the conductance of K ? ion) [120]. The RCK domains modulate the function of some prokaryotic [110][111][112] and eukaryotic [113][114][115] K ? channels. Its homolog, KTN (K ? ion transport and nucleotide binding) domain also exists in prokaryotes for uptake of K ? ions [116,117]. A prokaryotic genome analysis shows that a large number of two and six TMs K ? channels carry an RCK or a KTN domain, which illustrates a general theme utilized by this group of channels [1]. The RCK domain can be expressed as an entire soluble protein through an internal methionine in the gene which encodes the full-length channel as well [110,[120][121][122][123]. Two soluble RCK proteins or one soluble RCK protein and one RCK domain from the full-length protein form an RCK dimer due to the strong interactions at the dimer interface (corresponding to the flexible interface in Fig. 8, this RCK dimer is also called a hinge dimer [117]) [120]. Four such dimers in BKca [113][114][115], MthK [110][111][112], and KtrA (a cytosolic subunit of a K ? transporter from Bacillus subtilis, with nucleotides as its ligands) [116] build up the octameric gating ring in an alternative arrangement (Fig. 8). The effect of ligand binding has been well documented in BKca and MthK, where different gating ring states have been determined [110][111][112][113][114][115]. In both proteins, the outer rim of the gating ring expands after binding of ligands. The free energy of ligand binding is transferred through a presumed rigid helical linker to the pore-forming domain and results in a conformational change of the SF to conduct K ? ions [112]. In BKca, only the layer facing the membrane undergoes a substantial conformational change and the other layer remains static [114]. On the other hand, both layers alter their arrangements in MthK after ligand binding.
A dimer of dimer assembly of RCK Growing references suggest that the RCK assembly is flexible and the fourfold symmetry observed in the octameric gating ring of BKca, MthK, and KtrA may not be valid in other RCK-containing proteins.

TrkA, a dimer of dimer assembly in two layers
The TrkA gene contains two tandem RCK domains. Four soluble TrkA proteins are assembled with their membrane proteins to form a main K ? uptake system in bacteria. In contrast to BKca that also contains two tandem RCK domains within one single gene, the recently reported TrkA gating ring adopts a dimer of dimer assembly (PDB: 4J9U [124]). In BKca, four RCK domains within the same layer of the octameric gating ring are identical, but different between two layers (See also the KtrA gating ring shown in Fig. 8. The blue and red RCK monomers are identical in KtrA, but homologous to each other in BKca). However, homologous RCK monomers exist in the same layer in TrkA, rather than in two layers as in BKca. Furthermore, as compared to the twisted structure from the isolated TrkA gating rings (PDB: 4J9V [124] and a comparably similar TrkA structure from another species, PDB: 4G65), the authors proposed that four assembly interfaces are not identical [124]. Two of them are static (fixed), and another two are flexible (mobile). Thus, different proteins may have their own properties of the assembly interfaces.

KefC-CTD, a dimer of dimer assembly in one layer
KefC is a K ? efflux system from Escherichia coli and bears a C-terminal KTN domain (KefC-CTD). Although the KTN domains have been observed to arrange into an octameric gating ring structure in KtrA [116], only onelayer assembly composed of four domains is proposed to  [116]). a Is viewed down the fourfold axis from the extracellular side. The RCK monomers in the top layer (blue) and bottom layer (red) are connected through the alternative flexible and assembly interfaces, but the monomers in the same layer do not interact with each other. (b) Is the side view, rotated 90°from (a). One RCK monomer (red) interacts with two adjacent monomers in another layer (blue). The adenosine triphosphate ligands are shown in yellow be formed in KefC-CTD (PDB: 3EYW [125] and PDB: 3L9W [126]). Furthermore, instead of a fourfold symmetric layer as in BKca [113][114][115], MthK [110][111][112], and KtrA [116], two KefC-CTD hinge dimers are assembled together to form a dimer of dimer complex. As compared to the octameric gating ring structure, the assembly interfaces are disturbed due to existence of the helices 7 and 8 in KefC-CTD. This helix-turn-helix occupies the space in the assembly interface thus blocking formation of a higher assembly [125]. More interestingly, after binding of various ligands helices 7 and 8 are repositioned, which may alter the accessibility of the assembly interfaces [126]. The authors proposed that this one-layer assembly exists in the cell and the unidentified regulators or auxiliary proteins prevent formation of the octameric gating ring [125].

Summary of the RCK assembly
The results of the RCK assembly have not received an agreement yet. Some show two-layer arrangements and some have one-layer assemblies. Several factors may explain the discrepancy. First of all, it is due to the structure of the RCK domains. Although all of the RCK domains share a similar Rossmann fold [120], each domain varies at its C-terminus, which affects the assembly as discussed in KefC-CTD. In addition, some RCK domains also have extra sub-domains, such as the C-terminal subdomains in BKca, MthK, and TrkA. The effect of these extra domains remains to be elucidated [127]. Ligand binding changes the angle of the hinge dimer [112], as well as the order of the assembly [126]. The regulators or auxiliary proteins in the cell may further stabilize or prevent a certain assembly. Another important factor to be considered is the lipid environment for membrane proteins. Recent electron crystallographic studies of Kch (a six TMs K ? channel from Escherichia coli) suggested that the RCK domains expose to the solvent and do not interact with each other as observed in other octameric gating rings [128,129]. Since the majority of studies concern only the soluble RCK domains, the resulting structures may be due to the artifacts where the restriction of the transmembrane partner is lacking. Besides the effect of absence of the transmembrane partner, the observed conformations may be affected by the crystal contacts as well [124]. Furthermore, the angle of the hinge dimer and its higher assembly may vary in different states (resting, activated, and inactivated) [125].
To summarize, both the hinge dimer and the higher arrangement between the dimers are flexible. The published results may be influenced by the experimental conditions. Other unidentified partners in the cell may affect the assembly as well.

Perspective
Diverse biochemical and biophysical methods have been applied to understand the structure and function of potassium channels, besides X-ray crystallography that is still used as the main method to determine protein structures. Although considerable progress has been made, more studies are needed to explain the discrepancy in different reports, answer unclear questions, and aid in drug design. One important topic, the effect of the lipid environment may be worth investigating. Lipids not only provide a suitable environment for channels to fold, but also can participate in their activation. Another interesting topic is to study protein complexes, since membrane proteins often interact with other proteins to perform specific functions in the cell (such as the BKca/Cav complex mentioned above).
In the future, more information and deeper understanding of channels will be obtained with developing techniques.