Competing Lipid-Protein and Protein-Protein Interactions Determine Clustering and Gating Patterns in the Potassium Channel from Streptomyces lividans (KcsA)*

Background: Certain membrane proteins form supramolecular complexes that mediate cooperative cellular functions. Results: Clustering of KcsA results from competing lipid-protein and protein-protein interactions at nonannular sites and changes its gating. Conclusion: Clustering effects on gating occur via the Trp67–Glu71–Asp80 inactivation triad and its bearing on the selectivity filter. Significance: These interactions are potential targets for designing ion channel modulators. There is increasing evidence to support the notion that membrane proteins, instead of being isolated components floating in a fluid lipid environment, can be assembled into supramolecular complexes that take part in a variety of cooperative cellular functions. The interplay between lipid-protein and protein-protein interactions is expected to be a determinant factor in the assembly and dynamics of such membrane complexes. Here we report on a role of anionic phospholipids in determining the extent of clustering of KcsA, a model potassium channel. Assembly/disassembly of channel clusters occurs, at least partly, as a consequence of competing lipid-protein and protein-protein interactions at nonannular lipid binding sites on the channel surface and brings about profound changes in the gating properties of the channel. Our results suggest that these latter effects of anionic lipids are mediated via the Trp67–Glu71–Asp80 inactivation triad within the channel structure and its bearing on the selectivity filter.

There is increasing evidence to support the notion that membrane proteins, instead of being isolated components floating in a fluid lipid environment, can be assembled into supramolecular complexes that take part in a variety of cooperative cellular functions. The interplay between lipid-protein and protein-protein interactions is expected to be a determinant factor in the assembly and dynamics of such membrane complexes. Here we report on a role of anionic phospholipids in determining the extent of clustering of KcsA, a model potassium channel. Assembly/disassembly of channel clusters occurs, at least partly, as a consequence of competing lipid-protein and protein-protein interactions at nonannular lipid binding sites on the channel surface and brings about profound changes in the gating properties of the channel. Our results suggest that these latter effects of anionic lipids are mediated via the Trp 67 -Glu 71 -Asp 80 inactivation triad within the channel structure and its bearing on the selectivity filter.
Since the elucidation of the x-ray structure of KcsA (1-3), 4 this potassium channel from the Gram-positive soil bacterium Streptomyces lividans (4) has become an experimental system of choice in many ion channel and membrane protein studies. From the functional view point, the characterization of the KcsA single channel properties has been surrounded by certain controversy. Schrempf and co-workers (4 -6), discoverers of KcsA, reported a dependence of channel opening on acidic pH, multiple conductance states with fairly high opening probabilities, and unusual permeabilities to different cations in addition to K ϩ . In contrast, Miller and co-workers (7,8), using KcsA reconstituted into planar lipid bilayers, found a single conductance state with a much lower opening probability, as well as orthodox ion selectivity to validate KcsA as a bona fide K ϩ channel model. Such discrepancies were never fully explained, but still it became accepted that KcsA behaves as a moderately voltage-dependent, K ϩ -selective channel with low opening probability and the peculiar property of opening in response to intracellular acidic pH. Later on, however, different modes of KcsA activity have been reported, which differ greatly on their opening probability and channel kinetics (9). Moreover, it was found that KcsA may also open at neutral pH when subjected to certain experimental conditions (10). Also, it has been proposed that a more "physiological" version of KcsA might correspond to a supramolecular complex in which the channels would coassemble with polyhydroxybutyrate and inorganic polyphosphates (11), which are reservoir materials in prokaryotes. In our hands, purified KcsA reconstituted into giant liposomes made from asolectin phospholipids exhibits two major patterns of activity in patch-clamp recordings from excised, inside-out membrane patches: (i) a low opening probability (LOP) pattern, seen in ϳ55% of the recordings, in which channel openings are scarce and result primarily from single channel events or from coupled gating of a few channels, and (ii) a high opening probability (HOP) pattern observed in the remaining 45% of the patches, in which the channels are opened most of the time and exhibit positive cooperativity through coupled gating of a larger number of channels (12). Because abundant channel clusters were also detected in the giant liposomes, it was speculated that the different functional patterns could arise from "nonclustered" or "clustered" KcsA channels and that the assembly of such supramolecular entities would somehow determine changes in the integrated behavior of the channels involved. Interestingly, KcsA clusters have also been detected in vivo both in S. lividans and in Escherichia coli, suggesting that clusters may be the native form of KcsA in the bacterial membranes (13).
From the structural view point, KcsA is arranged as a homotetramer with subunits of 160 amino acids, each comprising Nand C-terminal cytoplasmic domains and two transmembrane ␣-helices, TM1 and TM2, connected by a short "pore" helix and a P-loop containing the characteristic TVGYG sequence of the selectivity filter in potassium channels. A further examination of the KcsA crystal structure reveals that it contains noncovalently bound lipid (2,14) identified as phosphatidylglycerol (PG) (15). Tight binding prevents these lipids from dissociating appreciably from the protein by treatments such as detergent solubilization, and indeed, they cocrystallize with the protein and can be seen in the x-ray structure. The crystallographic evidence and other studies (16) conclude that the PG binding sites in KcsA have the features of the so called "nonannular" sites (17), which usually correspond to clefts or grooves between transmembrane helices or at protein-protein interfaces (18,19). In particular, nonannular sites in KcsA include a deep cleft on the protein surface, between the pore helix and TM2 of adjacent subunits, which bind in vitro other anionic phospholipids, in addition to PG, in a rather unspecific manner (16,20,21), and their occupancy seems important for a number of KcsA features. These features include its folding, tetramerization, structural stability, and channel function and therefore make it extremely interesting to study the interactions involved (22)(23)(24), which might even be potentially useful targets for the design of ion channel modulators.
Here we report on the role of anionic phospholipids in determining the extent of KcsA clustering, which seemingly occurs as a consequence of competing lipid-protein and protein-protein interactions at the nonannular sites of the channel and brings about changes in the gating properties. In regard to the latter, our results suggest that such changes are mediated via the Trp 67 -Glu 71 -Asp 80 inactivation triad within the channel structure and its bearing on the selectivity filter.
Cloning and Mutagenesis of KcsA-The pT7-837KcsA containing the R64A-KcsA gene mutant was kindly donated by Professor A. Killian (Utrecht University, Utrecht, Holland). The R89A-KcsA mutant was obtained through site-directed mutagenesis, using the wild-type gene inserted into the pQE30 (Qiagen) plasmid as a template. The S22C-KcsA mutant was prepared as described previously (12). All mutations were confirmed by dideoxynucleotide sequencing.
Overexpression and Purification of KcsA-Expression of the wild-type KcsA protein, S22C and R89A mutants were performed in E. coli M15 (pRep4) cells, whereas R64A mutant was expressed in E. coli strain BL21(DE3). Purification and determination of protein concentrations were performed as described (25). Yields ranging from 1 to 2 mg of purified, DDMsolubilized, tetrameric KcsA per liter of culture were obtained. The purified protein batches were also analyzed by SDS-PAGE (25).
Blue Native PAGE-Blue native (BN)-PAGE was performed in linear 4 -16% (w/v) polyacrylamide gradient gels (26 -28). Shortly before starting BN-PAGE, DDM-solubilized KcsA was incubated in the absence or in the presence of increasing concentrations of the corresponding phospholipids as mixed micelles containing 5 mM DDM. Just before their application onto the gel, sample aliquots containing 10 g of DDM-solubilized protein were mixed with 5% Coomassie Brilliant Blue G stock solution in 750 mM aminocaproic acid and 1 mM SDS to a final 4:1 (by weight) DDM to Coomassie ratio. Electrophoresis was initiated at 85 V for 30 min and then continued at 200 V for 2.5 h, at 4°C. After electrophoresis, the gels were stained overnight with colloidal Coomassie Blue G 250 (CCB) (29). All gels were scanned and analyzed with ImageQuant TL, version 2005 software (Molecular Dynamics).
Electrophysiology-Multilamellar giant liposomes (up to ϳ100 m in diameter) were prepared by submitting a mixture of the reconstituted vesicles from above (containing 50 g of wild-type KcsA or mutants at a lipid to tetrameric protein molar ratio of 500:1) and either asolectin, PC:PA:Chol, or PC:PG:Chol lipid vesicles (25 mg of total lipids), to a cycle of partial dehydration/rehydration (12). Cholesterol was included in the phospholipid mixtures to facilitate giant liposome formation and handling. The resulting giant liposomes have a final lipid:tetrameric protein molar ratio of 46,800:1. Standard, inside-out patch-clamp recordings (31) were carried out on excised patches from the giant liposomes, as reported previously (12). The pipette (extracellular) solution contained 10 mM Hepes buffer, pH 7, 100 mM KCl, and the bath (intracellular) solution contained 10 mM MES buffer, pH 4, 100 mM KCl. Voltage ramps or different holding potentials were routinely imposed on the membrane patches to record channel currents (see details in the figure legends). The extent of rectification of the I/V curves was assessed by determining the "rectification index," calculated as the quotient between the whole outward and inward currents recorded during the voltage-ramp protocol.
Supported Lipid Bilayers (SLB) Containing Fluorescently Labeled KcsA-The sulfhydryl-containing mutant KcsA S22C was fluorescently labeled with a maleimide Alexa probe (Alexa Fluor 647 C2-maleimide; Molecular Probes), as reported (12). Planar bilayers were prepared as described by Chiantia et al. (32). Briefly, Alexa-labeled KcsA was reconstituted at 10,000:1, lipid to protein molar ratio in lipid vesicles of the indicated compositions. After a dilution to 0.7 mg/ml of lipid in 10 mM Hepes, pH 7, 100 mM KCl, the sample was bath-sonicated for ϳ15 min to obtain small unilamellar vesicles. 150 l of the sample were then placed on a freshly cleaved mica substrate glued to a glass coverslip within an open chamber. To induce vesicle fusion and SLB formation, 3 mM Ca 2ϩ was added, rapidly diluted with 400 l of buffer, and incubated for 15 min at 37°C. Then the chamber was rinsed thoroughly with prewarmed buffer to remove Ca 2ϩ and nonfused vesicles. Control experiments to check for the homogeneity of lipid distribution in the SLBs were done always before measuring KcsA distribution (not shown). To do that, the fluorescent probe 3,3Ј-dioctadecyloxacarbocyanine perchlorate was incorporated in the liposomes, by its addition to the lipid mixtures when in the chloroform:methanol (2:1) solution used in the initial stages of the liposome formation protocol (30).
Confocal microscopy was performed on an laser scanning microscope Meta 510 system (Carl Zeiss, Jena, Germany) using a 40 ϫ NA 1.2 UV-VIS-IR C Apochromat water immersion objective. SLB samples with Alexa-labeled KcsA were excited with the 633-nm line of a helium-neon laser, and the emitted light was filtered through a 650-nm long pass filter. In control experiments, DiO C18 (3) was excited with the 488-nm line of an argon laser, and the emitted light was filtered through a 505-550-nm band pass filter. Images were processed and analyzed with Scion Image (Scion Corp.).
Molecular Modeling-KcsA-DPPA complexes were built using the KcsA structure having cocrystallized lipids (2) (Protein Data Bank code 1K4C). The partial lipid structures seen in the crystal were used as a scaffold to build the entire DPPA lipid molecule. Lipid-protein contacts were optimized by energy minimization (see below).
Modeling of KcsA in the open conformation was done by homology, using the crystal structure of the open calciumgated potassium channel MthK as a template (33) (Protein Data Bank code 1LNQ). The sequence alignment was made with CLUSTALW (34) at the European Bioinformatics Institute site and manually supervised (final 24% identity and 40% similarity). The homology modeling was performed in the Swiss-Model protein modeling server (35) at the ExPASy Molecular Biology site under project mode. Structure preparation for modeling was made using DeepView, version 4.1 (36). The model was evaluated using PROCHECK to show the residues in the allowed regions of the Ramachandran plots (37). Model refinements were carried out as described by Molina et al. (38), followed by energy minimization. This latter process involved an initial short steepest descent minimization to remove bumps.
Next, we performed a simulated annealing minimization in which the simulation cell was slowly cooled toward 0 K by downscaling the atom velocities. The entire system was subject to an equilibration process before the molecular dynamics simulation. The equilibration consisted of an initial minimization of the fixed backbone atoms. Next, the restrained ␣-carbon atoms were minimized, and a short molecular dynamics (10 ps) minimization was performed to reduce the initial incorrect contacts and to fill the empty cavities. Finally, under periodic boundary conditions in the three coordinate directions, the full system was simulated at 310 K for 0.2 ns. All dynamic simulations were performed using Yasara (39) with the force field AMBER03 (40). The cutoff used for long range interactions was set at 10 Å.
The above open KcsA model was used to predict dimeric KcsA-KcsA interactions. The docking prediction was accomplished with GRAMM-X v.1.2.0 using default conditions. The solutions were checked and filtered manually, discarding those without biological significance, and then evaluated in terms of energy. The final model chosen was that having the highest interaction energy and the lowest van der Waals clashes. Finally, we perform molecular dynamic simulation on the resulting dimeric KcsA, following the protocol stated above.
As an alternative to build a dimer of open KcsA channels, we used the crystal structure of the dimeric potassium transporter TrkH (41) (Protein Data Bank code 3PJZ) as a scaffold to superimpose the three-dimensional structures of two open KcsA channels. The criteria for the superimposition were the correct alignment of the pore filter and the base of the pore helix.
The energy of the models and the docking complexes were tested using FoldX (42,43) on the CRG site. Molecules were edited and reconstructed with the general purpose molecular modeling software Yasara (39). The final molecular graphic representations were created using PyMOL, version 1.6.

Modeling Lipid-Protein versus Protein-Protein Interactions-
Specific interactions of anionic phospholipids with nonannular binding sites at intersubunit crevices in the KcsA homotetramer were first evidenced crystallographically (2) and more recently modeled (21). Fig. 1A focuses on the details of such interaction at the level of the extracellular membrane-water interphase and shows that Arg 89 at one of the KcsA subunits and Arg 64 at the contiguous subunit forming the intersubunit crevice interact with the phospholipid, "clogging" the crevice's entrance and forcing the side chain of Trp 67 toward the channel core, where it interacts with Glu 71 and Asp 80 . Such Trp 67 -Glu 71 -Asp 80 triad has been reported to play a critical role on the dynamics and conformational stability of the channel selectivity filter and to determine inactivation in KcsA (44), an analogous process to C-type inactivation in eukaryotic K v channels. The involvement of Arg 89 and mainly Arg 64 residues in the interaction with anionic lipids at these nonannular channel sites has been confirmed by NMR experiments (23) and MD simulations (45).
On the other hand, KcsA is known to self-associate into clusters, both in vivo and in vitro (12,13,46,47). Our hypothesis is that such clusters are responsible for the observed HOP pat-terns of KcsA activity, where the channels are opened most of the time and exhibit coupled gating (12). Therefore, in an attempt to explain how the individual channels could be assembled into clusters to give rise to channel activity with a high opening probability, we performed molecular docking between two adjacent KcsA molecules, using an open channel model of KcsA based on the x-ray open channel structure of MthK as a template (33). The reason for choosing MthK for these studies is that it still remains as the only crystal structure of a potassium channel having open both the inner and outer channel gates, and therefore, it provides the best possible available structure to model the almost permanently open HOP patterns in KcsA. The presence of lipid at the nonannular binding sites prevents docking between the two channels, and therefore, lipids were eliminated from the model. In the absence of the competing lipids (Fig. 1B), the intersubunit crevice defining the nonannular sites in each channel becomes now wider (the distance between the ␣-carbons of Pro 63 in the pore helix and Trp 87 in M2 segment increase 1-2 Å, as predicted by the model), being occupied by the N-terminal end of transmembrane M2 from the other channel, which contributes Trp 87 to the interaction network. As a consequence of docking, the Trp 67 side chain rotamer pointing away from the channel central core becomes now favored, thus disrupting the interactions within the Trp 67 -Glu 71 -Asp 80 triad. Moreover, Arg 64 and mainly Arg 89 interact now with Glu 71 and Asp 80 , further contributing to the disassembly of the Trp 67 -Glu 71 -Asp 80 triad responsible for channel inactivation. In other words, the model predicts that a clustered KcsA channel should have different gating and inactivation properties than those seen when anionic phospholipids occupy the nonannular sites of the channel.
As a second approach to test the above hypothesis, we took advantage of the reported x-ray structure of the potassium transporter TrkH (41), which already crystallizes as a dimer and therefore does not require any docking procedure to be applied. These circumstances should make the dimeric TrkH an adequate structure to model potassium channel clusters. In this case, it is the N-terminal end of the pore helix of one channel (instead of the M2, as mentioned above) that goes closer to the also wider intersubunit crevice in the adjacent channel (Fig.  1C). Nonetheless, the consequences of such interaction on the Trp 67 -Glu 71 -Asp 80 triad are practically identical to those seen in the docking above, because the positioning of the Trp 67 , Arg 64 , and Arg 89 side chains in the dimer structure contributes similarly to disrupt the interactions within the inactivation triad.
From the above observations, two related predictions can be made: (i) binding of anionic lipids to the nonannular sites should compete with KcsA clustering and (ii) assuming that clustering is indeed a determinant factor of gating behavior, anionic phospholipids should be expected to influence the frequency of appearance of LOP or HOP activity patterns. These two predictions are tested below.
Effects of Lipids on the Assembly/Disassembly of KcsA Clusters-Mixed micelles containing the purified KcsA channel protein solubilized in DDM and different concentrations of specific lipids have been used here to evaluate the extent of KcsA clustering by BN-PAGE, a method of choice to study the FIGURE 1. Lipid-protein and protein-protein interactions at nonannular lipid binding sites. A shows a top view (normal to the membrane plane) of an extracellular portion comprising amino acid residues 62-101 (roughly indicated by the box on the small diagram to the right of A) in the crystallographic structure of KcsA (Protein Data Bank code 1K4C) defining one of the four intersubunit crevices acting as nonannular lipid binding sites. A DPPA molecule (atoms represented as spheres) has been drawn bound to such a site, using as a scaffold the partial lipid structure appearing in the protein crystal (see "Materials and Methods"). Adjacent protein subunits defining the nonannular site are colored in light brown and green. The one-letter codes and numbering in the KcsA sequence for the main amino acid residues involved in the interactions have been included. Hydrogen bonds are indicated by red dashed lines. It should be noted that the Trp 67 -Glu 71 -Asp 80 inactivation triad is fully configured and that the lipid polar head interacts primarily with Arg 64 and Arg 89 . B shows the results from computer docking two KcsA channels in an attempt to model a cluster. In this case, the KcsA sequence was previously modeled on the basis of homology onto the open structure of MthK (see "Materials and Methods"). The nonannular binding sites, flanked by the pore helix and M2 in each channel, interact now with the M2 from the adjacent channel (channel 2; subunits colored in light blue and pink), whereas the inactivation triad has been disrupted. C shows the results of modeling the two KcsA channel cluster based on the x-ray structure of the potassium transporter TrkH, which already crystallizes as a dimer (see "Materials and Methods"). Notice that in this case, the nonannular binding site interacts with the pore helix from the adjacent channel, but the effect of such interaction on the disruption of the inactivation triad is identical. organization of protein complexes (27, 48 -50) including KcsA (26). In BN-PAGE, samples are first solubilized with a mild detergent, usually digitonin or DDM, but it is the negative charge from the Coomassie Brilliant Blue bound to the hydrophobic protein surfaces what determines the electrophoretic mobility. Fig. 2 shows that in the absence of added lipids, the DDM-solubilized protein (left-hand lane in all gels) exhibits several bands able to enter the gel, which correspond to different KcsA supramolecular assemblies, which remain as such despite detergent presence (26). The fastest migrating band in the gels is the well known KcsA homotetramer (labeled as T species in the figure), whereas the other gel bands correspond to supramolecular clusters comprising 2, 3, 4, and 5 tetrameric channels (labeled 2T, 3T, 4T, and 5T in the figure). Fig. 2 also shows that increasing the amount of zwitterionic phospholipids (PC or PE in upper panel, respectively) in the mixed micelles, regardless of the phospholipid to protein ratio, have no significant effects on the relative abundances of the KcsA species detected by BN-PAGE. Conversely, when anionic phospholipids (PG or PA in lower panel) are used in the experiments, disassembly of the larger clusters occurs progressively as the phospholipid to protein ratio is increased, so that eventually the gels show mostly T and 2T KcsA species instead of the more complex starting population of clustered species. Quantitative analysis of cluster disassembly caused by the anionic phospholipids shows no apparent differences between the effects of either PG or PA, suggesting that the concentration-dependent process requires just a negatively charged anionic phospholipid to take place.
In addition to the mixed micelles from above, SLBs have also been used as an experimental system to assess the concentration-dependent effects of anionic phospholipids on KcsA clusters. SLBs are likely to model native membranes better than mixed micelles and also lack detergents, thus providing data complementary to those from above. Fig. 3 shows that fluorescently labeled KcsA in the SLBs indeed forms clusters of variable sizes, up to 1-2 microns in diameter. The abundance of such clusters decreases dramatically, and their size decreases slightly when the presence of anionic lipids in the supported bilayer is increased, thus qualitatively confirming the observation from the mixed micelles above on the ability of high levels of anionic phospholipids to interfere on the assembly of KcsA into clusters.
Effects of Lipids on KcsA Function-We reconstituted KcsA into giant liposomes made from mixtures of different phospholipids to assess their effects on channel function by patch-clamp techniques. Fig. 4 (A and B) shows that under these conditions the reconstituted KcsA still exhibits both LOP and HOP patterns of channel activity, comparable in all aspects to those described in detail when using asolectin phospholipids for reconstitution of wild-type KcsA (12). However, the frequencies of appearance of such activity patterns are now found to be dependent on the relative proportion of anionic to zwitterionic phospholipids used in the experiments (Fig. 4C). Thus, giant liposomes containing only 5% of either PA or PG in a PC/cholesterol matrix have a high percentage of patches exhibiting HOP patterns of activity (ϳ70% of the recordings, taking together the two populations of PA-and PG-containing giant liposomes; n ϭ 84). Conversely, giant liposomes containing 25% of either PA or PG in the lipid matrix show an increased presence of LOP patterns, i.e. a marked decrease in the frequency of appearance of HOP patterns (which account for only ϳ30% of the recordings; n ϭ 91; p Ͻ 0.01 when compared against the group of 5% anionic lipids, z test). Interestingly, the frequency of appearance of HOP patterns reported in asolectin lipids (12), which is between those found in the lipid media containing 5 or 25% anionic phospholipids from above (Fig.  4C), seems consistent with the report that asolectin lipids contain 10 -13% anionic phospholipids (51). These observations indicate that the ability of KcsA to exhibit HOP patterns of channel activity is inversely related to the concentration of ani-  OCTOBER 16, 2015 • VOLUME 290 • NUMBER 42 onic phospholipids within the reconstituted membrane bilayer.

Modulation of Clustering and Gating in KcsA
In any case, it should be noted that at comparable concentrations, PA was always more efficient than PG in decreasing the percentage of HOP patterns.
The presence of anionic phospholipids in the lipid matrix has an additional effect on the observed HOP currents: although the samples containing 5% anionic lipids, in which HOP patterns are predominant, show a clear inward rectification at negative potentials (rectification index: 0.63 Ϯ 0.03, n ϭ 59; see   containing 25% anionic lipids show HOP currents with a lesser inward rectification (rectification index: 0.79 Ϯ 0.04, n ϭ 27; p Ͻ 0.01; see Fig. 4, lower traces (HOP) B and the gray column in D). Such different features were previously observed as variants of HOP patterns upon reconstitution of KcsA in asolectin lipids (12).
Analysis of KcsA Single Mutants-In the previous paragraphs, we hypothesized that the different patterns of channel clustering and gating observed in the presence of increasing concentrations of anionic phospholipids could be explained on the basis of differential interactions involving key amino acid residues, namely Arg 64 , Arg 89 , and Trp 67 , that bear on the integrity of the Trp 67 -Glu 71 -Asp 80 inactivation triad and thus on the properties of the selectivity filter. Therefore, we conducted experiments with several mutants of KcsA to test our hypothesis.
The electrostatic interaction between anionic phospholipids and positively charged amino acid residues in the KcsA sequence is believed to constitute the basis for lipid binding to the nonannular sites (16,20). Therefore, substitution of the Arg 64 or Arg 89 residues by uncharged amino acids in KcsA mutants should disfavor the stabilization of anionic phospholipids at such sites, and according to our working hypothesis, such mutants should be expected to display an altered pattern of channel clustering and gating. Fig. 5 shows that compared with the wild-type channel in Fig. 2, the R64A KcsA mutant is indeed relatively insensitive to the effect of anionic phospholipids on the disassembly of channel clusters observed by BN-PAGE. Conversely, the R89A mutant still shows a partial response of cluster disassembly by anionic phospholipids, more markedly to PA than to PG, suggesting that lipid interaction and the subsequent cluster disassembly relates mainly to the presence of the Arg 64 residue in the channel sequence.
As to the functional responses, the two arginine mutant channels have in common a marked loss of their ability to inactivate. Fig. 6 (A and B) shows that the predominant pattern seen in the R64A mutant is very similar to the HOP pattern exhibited by the wild-type channel (68%; 17 of 25 patches), including the characteristic coupled gating observed in channel openings and closings within the HOP recordings. Single channel recordings taken from this mutant channel (Fig. 6C) show that the channel conductance is also similar to that reported for the wild-type channel (12), but the channel opening probability is greatly increased, as expected, approaching that reported by others for this and other mutant channels such as the E71A mutant (52), considered an archetypical "noninactivating" channel. As different from the above, the R89A channel mutant shows unusual HOP patterns (73%; 30 of 41 patches), with lower overall current levels throughout the voltage ramps, particularly at negative potentials (Fig. 6D), and a marked loss of the coupled gating feature (Fig. 6E), as if the ensemble of channels within the clusters were now opening and closing in a mostly independent manner. We termed this behavior of the R89A mutant as an "uncoupled" HOP pattern, which is similar to that reported by Perozo and co-workers (52) for several KcsA mutant channels. Nonetheless, the R89A mutant has in common with the R64A mutant a similar conductance and channel opening probability in single channel recordings (Fig. 6F).
Despite having been reported by others (44,52), W67F and E71A KcsA mutants were also prepared in our lab to compare their behavior under our experimental conditions with the arginine mutants from above. In agreement with such reports, we found that, even more markedly than in the R89A mutant, HOP recordings in both the E71A and W67F mutants have an even lower extent of channel coupling, i.e. were also uncoupled HOPs, and when recorded as single channels, exhibited also a high P o , as expected from these two amino acid residues being part of the Trp 67 -Glu 71 -Asp 80 inactivation triad (data no shown). Finally, it should be noted that the uncoupled HOPs seen in the R89A mutant, as well as those in the W67A and E71A mutants (not shown), showed I/V curves with outward rectification (rectification index for R89A mutant: 1.54 Ϯ 0.12, n ϭ 30; see Fig. 6G), instead of the inward rectification or almost ohmic behavior seen in conventional HOP patterns from either the wild-type KcsA or the R64A mutant (rectification index: 0.92 Ϯ 0.06, n ϭ 25, and 1.01 Ϯ 0.08, n ϭ 17, for wild-type and R64A mutant channels, respectively; see Fig.  6G). Outward rectification is a hallmark feature of KcsA single channels (5, 8, 12), thus further supporting that the uncoupled HOP recordings arise from channels that, despite forming clusters, act independently.

Discussion
Channel activity in KcsA results from a concerted equilibrium between two defined ion gates within the protein structure (53). The inner gate is formed by the intracellular transmembrane helix bundle crossing, which is destabilized at acidic pH, shifting the equilibrium toward its open conformation. On the other hand, the outer gate at the selectivity filter defines primarily the channel opening probability and may shift its conformation equilibrium from the permeation state toward a nonpermeation state, causing channel inactivation. Both the conformational state and functional responses of KcsA seem strongly modulated by its surrounding membrane milieu (43, 54 -56), probably through an alteration of one or both gating equilibria. So far, different mechanisms have been proposed to explain such effects. On one hand, although still controversial, nonspecific electrostatic interactions between anionic lipids and the N-terminal portion of the protein are proposed to favor the open conformation of the inner gate of the channel (57), resulting in increased activity. On the other hand, specific interactions of anionic lipids with the nonannular binding sites of KcsA have been proposed to modulate ion channel function through its influence on the outer gate (45,56), although there is not a clear explanation for such an effect. In this work, we propose that a competition between lipid-protein and proteinprotein interactions at the nonannular binding sites constitutes a central element in such KcsA modulation.
Both in vivo (13) and in vitro (12,25,46,47), KcsA assembles into supramolecular clusters, which could even be the predominant forms in which this channel is present in bacterial membranes (12,25). Here we report that both in a mixed lipid/detergent micellar system and in supported bilayers, the presence of anionic phospholipids modulates assembly/disassembly of such KcsA clusters. Therefore, low proportions of anionic phospholipids allow KcsA to remain associated into clusters, whereas higher concentrations lead to cluster disassembly. Moreover, such conditions of high or low proportion of anionic phospholipids in reconstituted giant liposomes containing KcsA increase the frequency of appearance of either LOP or the more complex HOP patterns of channel activity, respectively, in patch clamp recordings. Thus, the effects of the anionic phospholipids on the assembly/disassembly of clusters and on the appearance of HOP/LOP patterns of activity are correlative and suggest that the more complex HOP recordings indeed arise from large channel clusters, whereas LOP recordings originate mostly from nonclustered, phospholipid-bound KcsA channels. It should be noted here that experimental variables such as the presence of detergent, lipids, or the lipid to protein ratios used in reconstitution greatly influence KcsA assembly, and therefore, the comparison of results from the different experimental approaches used here should only be done in qualitative terms.
In an attempt to find a cause and effect relationship for the above phenomena, we modeled the consequences of the interaction of either an anionic phospholipid or another tetrameric channel at the nonannular site of KcsA. Fortunately, these  Fig. 4. Notice that the R64A mutant shows HOP patterns of activity very similar to those seen in wild-type KcsA (12) and exhibits coupled gating, as illustrated in the continuous recording taken at ϩ150 mV shown in B. None of these features are clearly present in the R89A mutant (D and E), which exhibits mainly uncoupled HOPs (see "Results"). C and F show single channel recordings taken also at ϩ150 mV of the two mutant channels to illustrate that despite the above differences; the channel conductance and opening probability when analyzed at the single channel level are similar in both cases. G shows a column histogram of the rectification index of the HOP pattern of the wild-type KcsA, as well as those of the R64A and R89A mutant channels.
interactions are partly documented from x-ray crystallographic data of both the anionic phospholipid-KcsA complex (2) and that of the dimeric potassium transporter TrkH (40), thus providing a valuable experimental support to our models. Opposite to the channel-channel interaction, binding of the phospholipid to the nonannular site on the channel allows Trp 67 to form the Trp 67 -Glu 71 -Asp 80 triad, which should leave the phospholipid-bound channel prone to inactivate. Because KcsA is a homotetramer, there are four identical nonannular sites, each of them acting on just one of the four rows of Thr 75 -Val 76 -Gly 77 -Tyr 78 -Gly 79 residues from each subunit, which contribute their carbonyl groups to configure the selectivity filter. It is to be expected that anionic lipids disrupt channel-channel interactions at more than one of such sites to affect sufficiently the selectivity filter conformation as to make inactivation a truly effective process, whose occurrence should be dependent on the phospholipid concentration. Obviously, a readily inactivating channel should produce recordings characterized by a low channel opening probability, i.e. a LOP pattern of activity, as has been found experimentally.
On the contrary, because the inactivation triad in clustered KcsA is predictably disrupted because channel-channel interaction causes that Trp 67 swings away from its Glu 71 and Asp 80 partners, these clustered channels would not be expected to inactivate as readily as the phospholipid-bound, nonclustered KcsA channel does. This is indeed the case in the observed HOP patterns of channel activity.
Finally, site-directed mutagenesis at specific residues involved in the interaction with the anionic phospholipids was carried out to provide additional evidence to support or to discard our working hypothesis. In this regard, Arg 64 and Arg 89 residues are claimed to be responsible for electrostatic interaction with the anionic phospholipids at the nonannular protein sites and their substitution by alanine should be expected to minimize such interactions, partly preventing cluster disassembly and channel inactivation. This seems clearly the case in the R64A mutant, which does not disassemble significantly when in presence of anionic phospholipids, as shown by the BN-PAGE assay, and exhibits an essentially noninactivating pattern of channel activity, very similar to the HOP patterns in wild-type KcsA, in which coupled-gated currents are observed throughout. These results further evidence of a major role for Arg 64 in lipid interaction.
Conversely, the R89A mutant shows clusters that partly disassemble in response to the presence of anionic phospholipids, more markedly to PA than to PG, and exhibits lower current levels and less coupled gating than the wild-type KcsA. The larger effect of PA over PG on cluster disassembly could be due to the fact that, in the appropriate environment, PA could become a dianionic species, which may further favor the interaction with the positively charged groups in the protein.
Indeed, PA has been reported to have a higher affinity than PG to bind to the nonannular sites of KcsA (21). As to the decrease in the current level observed in the R89A mutant, we found that both the single channel conductance and the opening probability in these two arginine mutants were similar, and therefore, the lower current levels observed in the R89A mutant should be mainly attributed to a diminished capacity to recruit individual channels into the coupled-gated clusters. Therefore, the behavior of the R89A mutant channel suggests that the Arg 89 residue is important to maintain gating features characteristic of HOP patterns in wild-type KcsA, because its absence results in uncoupled HOP patterns with an outward rectification. This is also similarly observed in the W67F and E71A mutants, and therefore, it seems that the residues involved in determining channel inactivation, including Arg 89 , are also involved in conferring channel coupling and current rectification properties through a still undefined mechanism.
The above experimental observations, as well as the predictions from the models, are depicted in a simplified manner in Scheme 1. Interaction with the anionic phospholipid, in which Arg 64 is heavily involved, impedes cluster assembly and leaves Trp 67 , Glu 71 , and Asp 80 residues free to form the inactivation triad, resulting in individual, lipid-bound channels (LOP patterns of channel activity). On the contrary, cluster assembly displaces Trp 67 from the inactivation triad. Such a displacement seems difficult to reverse because the remaining inactivation triad members (Glu 71 and Asp 80 ) are now predictably stabilized by interaction with Arg 89 , which swings into the channel core as a consequence of channel-channel interaction. This results in clustered channels quite unable to inactivate, which may exhibit either (i) coupled gating, as in the conventional HOP patterns characteristic of wild-type KcsA, or (ii) uncoupled HOPs, when in the absence of strategic amino acid residues, as seen in some of the mutants above. Interestingly, altering interactions among residues equivalent to those configuring the inactivation triad in KcsA have been found to change selectivity, block, and current rectification properties in a Kir channel (58).
KcsA clusters have been detected in vivo (13), and indeed, they might account for the complex patterns of electrical activity reported earlier for this channel (4,5). Nonetheless, the assembly of clusters in vitro upon membrane reconstitution from purified channel preparations is a protein concentrationdependent phenomenon (12), and therefore, the experimental conditions used in reconstitution by the different authors, particularly the lipid to protein ratios used, which ranges within more than an order of magnitude, should be expected to greatly influence the occurrence of clustering and its functional consequences. For instance, clustering might go unnoticed under conditions such as those used to obtain recordings in planar lipid membranes, which use a very low concentration of KcsA, sometimes in the presence of detergents or organic solvents, which may also compete for binding to the KcsA nonannular sites. Indeed, the low opening probability mode of KcsA is believed to be predominant in planar bilayer experiments (9). In any case, considering that the molar ratio of KcsA to phospholipids in our giant liposome samples ranged approximately from 1:10,000 to 1:46,800, it becomes obvious that the affinity of KcsA to associate with other KcsA channels to form clusters SCHEME 1 OCTOBER 16, 2015 • VOLUME 290 • NUMBER 42 above a certain threshold of protein concentration is much higher than that to bind the anionic phospholipids contained in the bilayer, and therefore, cluster formation should be favored under our experimental conditions. All the evidence from above seemingly completes an scenario in agreement with the starting hypothesis, in which competing lipid-protein or protein-protein interactions at the nonannular sites of KcsA determine both the occurrence of clustering and changes in gating behavior. The latter changes seemingly result from interfering with the assembly of the Trp 67 -Glu 71 -Asp 80 inactivation triad, which in turn bears on the conformation of the selectivity filter. In other words, clustered KcsA channels have different dynamics and conformational stability at the selectivity filter than nonclustered, phospholipid-bound KcsA. Both clustering-and lipid-related effects in gating behavior have also been reported for other channels (59 -64), and although diverse mechanistic explanations have been given, it may be possible that phenomena similar to that reported here might also underline some of those observations.