Phosphatidylinositol 4,5-bisphosphate (PIP2) regulates KCNQ3 K+ channels by interacting with four cytoplasmic channel domains

Phosphatidylinositol 4,5-bisphosphate (PIP2) in the plasma membrane regulates the function of many ion channels, including M-type (potassium voltage-gated channel subfamily Q member (KCNQ), Kv7) K+ channels; however, the molecular mechanisms involved remain unclear. To this end, we here focused on the KCNQ3 subtype that has the highest apparent affinity for PIP2 and performed extensive mutagenesis in regions suggested to be involved in PIP2 interactions among the KCNQ family. Using perforated patch-clamp recordings of heterologously transfected tissue culture cells, total internal reflection fluorescence microscopy, and the zebrafish (Danio rerio) voltage-sensitive phosphatase to deplete PIP2 as a probe, we found that PIP2 regulates KCNQ3 channels through four different domains: 1) the A–B helix linker that we previously identified as important for both KCNQ2 and KCNQ3, 2) the junction between S6 and the A helix, 3) the S2–S3 linker, and 4) the S4–S5 linker. We also found that the apparent strength of PIP2 interactions within any of these domains was not coupled to the voltage dependence of channel activation. Extensive homology modeling and docking simulations with the WT or mutant KCNQ3 channels and PIP2 were consistent with the experimental data. Our results indicate that PIP2 modulates KCNQ3 channel function by interacting synergistically with a minimum of four cytoplasmic domains.

Lending support for a generalized structural interaction between PIP 2 and the region just distal to the final transmembrane helix of K ϩ channels is the crystal structure of PIP 2 bound to the K ir 2.2 channel (12), which shows a PIP 2 molecule interacting with residues not only in the proximal C terminus as it emerges from the lipid bilayer, but also residues at the distal end of the M2 helix. Thus, it behooved us to more systematically examine all of these regions of a KCNQ channel most amenable to study via a voltage-dependent phosphatase (VSP), which can dephosphorylate nearly all of the PIP 2 in the plasma membrane within about 500 ms (13). This method has been exploited to examine the PIP 2 sensitivity of KCNQ (14,15) and TRP (16) channels, among others. Most significantly, unlike reducing PIP 2 abundance by stimulating G q -and phospholipase C-coupled receptors, which could also produce inositol triphosphate, Ca 2ϩ rises, activate protein kinase C and induce other downstream signals, activation of VSP only dephosphorylates PIP 2 to PI(4)P, a singly phosphorylated lipid that does not allow activation of M channels (17,18).
The Hille group (14) studied KCNQ2/3 heteromers and found the time constant of dephosphorylation of available PIP 2 in the membrane of a tissue culture cell to be ϳ250 ms; in that work, while not quantifying the k on or k off of PIP 2 , they found a "dwell time" of ϳ10 ms to be consistent with the modeling of their data, most likely due to the low affinity of KCNQ2 subunits that determine whether KCNQ2/3 channels are open or closed due to PIP 2 interactions. Hence, mutations that decrease their apparent affinity of PIP 2 , resulting in "dwell times" necessarily shorter than 10 ms in KCNQ2-containing channels cannot possibly be meaningfully quantified during the decay of the current during the depolarization step to a very positive potential that activates Danio rerio VSP (Dr-VSP), because any shorter k off would be wholly confounded by the time required for PIP 2 dephosphorylation. In such a case, only an altered rate of recovery of the current, reflecting an altered k on , could be meaningful. Thus, such relatively low PIP 2 apparent affinity channels are unsuitable for this approach. For these reasons, we chose the KCNQ3 homomer as our test channel, due to its extremely high apparent affinity for PIP 2 , as manifested by its saturating open probability near unity at saturating voltages and its maximal depression by M 1 receptor stimulation of only ϳ40% (5,19) versus Ͻ0.3 and 90%, respectively, for all other KCNQ isoforms and compositions. Our assumption was that that this channel would be amenable to such analysis using the VSP approach and that high structural and mechanistic similarity with the other KCNQ subtypes should make our data generalizable among this K ϩ channel family. In some experiments, we used the alternative assay of quantifying the extent of depression of the current by stimulation of muscarinic M 1 acetylcholine receptors (M 1 Rs) co-expressed with the channels (see below).
In our patch-clamp experiments, we used the well-expressing KCNQ3-A315T (KCNQ3T) channel as a baseline, an innerpore mutant that increases whole-cell current amplitudes by Ͼ10-fold (20,21,73), without changing the open probability of the channels or their apparent PIP 2 affinity (19). We probed the effects of charge neutralizations in the S2-S3 linker, the S4 -S5 linker, the S6Jx domain, and the A-B helix linker on changes in the apparent PIP 2 affinity of the channels as well as their voltage dependence of activation. In addition, homology modeling and PIP 2 -docking simulations were performed to seek a structural framework for our experimental results. We find that all of the regions tested are involved, complementing the PIP 2 -binding "cationic cluster" described previously in the A-B helix linker of KCNQ2 and KCNQ3 (8). Whereas the four domains identified here for KCNQ3 as interacting with PIP 2 are conserved with KCNQ1, and likely KCNQ2, mutations that lowered the apparent affinity of the channels for PIP 2 were not correlated with alterations in voltage dependence.

Results
We chose Dr-VSP because it activates at ϩ40 mV, well positive to the saturating voltage for all KCNQ channels. Upon activation of Dr-VSP by depolarization to ϩ120 mV, which dephosphorylates PIP 2 into PI(4)P, quantification of the rate of decay of the current provides an estimate of changes in k off of PIP 2 from the channels due to mutations. We realize that this is an approximation, due to the confound of the known rate of Dr-VSP dephosphorylation of PIP 2 by Dr-VSP at that voltage ( ϳ250 ms). However, the deconvolution of those rates is beyond the scope of this paper; moreover, we would need information on the allosteric influence of the binding of one PIP 2 molecule to one subunit on its affinity with another and the precise number of PIP 2 molecules required for the opening of KCNQ3 homomers, and both sets of data are lacking at this time. Upon the step back to ϩ30 mV, changes in k on of PIP 2 due to mutations were estimated by the rate of recovery of the current. We again realize that this estimate is an approximation due to the confound of the known rate of PI(4)P-5 kinase ( ϳ10 s) (14). Again, an even more sophisticated deconvolution would be extremely difficult without more information, which is also not presently available.
Besides the measurements described above, we also compared the amplitude of tonic whole-cell currents between cells transfected with KCNQ3T and mutant KCNQ3T channels and the voltage dependence of activation. The first measurement is based on the correlation between the tonic open probability at the single-channel level, macroscopic current amplitudes, and PIP 2 apparent affinity observed for KCNQ2, KCNQ2/3, KCNQ3, and KCNQ4 channels (5,19) and other PIP 2 -regulated channels (e.g. GIRK channels) as well (6,22). The voltage dependence of activation is important because whether PIP 2 -mediated depression of KCNQ1-containing channels is accompanied by altered voltage dependence is still open to debate (3,10), and PIP 2 -mediated modulation of KCNQ2/3 channels does not change the voltage dependence of activation (23)(24)(25). Whereas the A-B helix linker "cationic" cluster domain identified in PIP 2 interactions with KCNQ2 and KCNQ3 in our previous work (19) is not conserved in KCNQ1, the S6Jx, S4 -S5 linker, and S2-S3 linker PIP 2 -interaction domains are conserved, which for KCNQ1 were suggested to form a network of PIP 2 -interacting domains that are involved in voltage sensor/gate coupling (10). We were therefore keen to investigate these issues for the case of KCNQ3, which is found primarily in neurons, as opposed to cardiomyocytes or epithelia. With the parameters and assumptions given, we can now present the data.

Interactions of PIP 2 with the S2-S3 and the S4 -S5 linkers in KCNQ3
Several recent studies have suggested a potential role of the S2-S3 and the S4 -S5 linkers in PIP 2 -KCNQ channel interactions (10,11,26,27). Because these sites are most novel in terms of PIP 2 interactions suggested for KCNQ2-5 channels, we begin here. For KCNQ1, the interactions with the S2-S3 linker involve Arg 190 and Arg 195 , and for the S4 -S5 linker, they involve Arg 243 , His 258 , and Arg 259 . The sequence alignment of the S2-S3 and the S4 -S5 linkers among KCNQ1-5 channels ( Fig. 1A and Fig. S3E) indicates the Arg 190 , Arg 195 , and His 258 residues to be conserved. The perforated patch variant of whole-cell recording was performed to maintain the intracellular milieu and prevent "run-down" of PIP 2 abundance. We tested the effect of charge-neutralizing mutations at the analogous positions, Arg 190 and Arg 195 , in the S2-S3 linker and His 257 (corresponding to His 258 in KCNQ1) in the S4 -S5 linker of KCNQ3T (Figs. 1 (A-C) and 2). In the S2-S3 linker, the R190Q mutation, but not R195Q, decreased current densities from 197 Ϯ 6 to 66 Ϯ 12 pA/pF ( Fig. 2 (A and  B) and Table 1). Using the Dr-VSP assay, we found the rate of current decay upon depolarization that turns on Dr-VSP ( decay ) for the R190Q mutant (0.35 Ϯ 0.11 s) to be much faster than for KCNQ3T (0.94 Ϯ 0.13 s) (Fig. 2 (D and E)); however, the rates of recovery of the current ( recovery ) were not significantly different (7.5 Ϯ 1.7 s versus 9.6 Ϯ 1.6 s for KCNQ3T and R190Q, respectively). The same result was obtained from the analogous mutant R190A ( decay ϭ 0.35 Ϯ 2 s, recovery ϭ 7.5 Ϯ 1.8 s). Neither response was altered by the R195Q mutation. Either Arg 190 influences K off for PIP 2 , but not k on , or our assay is not sensitive enough to detect changes in both rates accurately. As an alternative assay, we thus turned to the classic M 1 R-dependent depression of the current in cells co-expressing M 1 muscarinic receptors and KCNQ3T mutants. Because maximal M 1 R stimulation in tissue culture cells leads to about an 80% decrease in PIP 2 abundance, rather than to near zero when VSPs are activated (14), the maximal depression of KCNQ3 currents is only ϳ30 -40%, because enough PIP 2 molecules remain in the membrane to keep most KCNQ3 channels PIP 2 -bound (19,24). Thus, for such high PIP 2 apparent affinity channels, a change in that affinity is manifested most in the fractional suppression of the current, not a shift of the dose-response relation of [agonist] versus current suppression (Fig. S1F). In these experiments, we decided to use mutants in which the arginines at positions 190 and 195 were mutated to alanines instead of the highly hydrophilic and bulky glutamines, which can interact with PIP 2 via several types of H ϩ bonds, to avoid any such confounding effects. Consistent with previous work, we found the KCNQ3T current to be suppressed by a supramaximal concentration (10 M) of the receptor agonist, oxotremorine methiodide, by only 29.5 Ϯ 5.5%, and for cells expressing KCNQ3T-R195A, the maximal inhibition was only 24.9 Ϯ 5.8%. However, for KCNQ3T-R190A, the maximal inhibition was 63 Ϯ 13%, indicating that the R190A mutation reduces PIP 2 affinity, consistent with the Dr-VSP assay ( Fig. 2G and Fig. S1). Neither the R190Q/A nor the R195Q/A mutations affected the voltage dependence of activation (Fig. 2C), suggesting that the apparent affinity of PIP 2 for this site is unrelated to voltage dependence.
We found the H257N mutation in the S4 -S5 linker (Figs. 1 (B and C) and 2) to result in strongly reduced current densities, from 197 Ϯ 6 to 30 Ϯ 3 pA/pF. The rate of current decay after Dr-VSP activation was much faster than KCNQ3T (0.58 Ϯ 0.14 s), and the rate of recovery was slightly slower (11.0 Ϯ 1.4 s), although it was not suitable for analysis in most of the cells recorded, probably due to the astounding shift in the voltage dependence of activation from Ϫ34.0 Ϯ 1.9 mV for KCNQ3T to 2.5 Ϯ 2.8 mV for H257N. Thus, we turned again to quantifying the result of M 1 R stimulation. For cells co-transfected with M 1 Rs and the KCNQ3T-H257N mutant, the maximal inhibition was 81.6 Ϯ 7.9% ( Fig. 2G and Fig. S1). Together, these results indicate that the H257N mutation reduces the apparent PIP 2 affinity of the channels.
Because the R243H mutation in the S4 -S5 linker was shown to reduce the apparent affinity of KCNQ1 for PIP 2 (26) and Arg 243 is conserved in other KCNQ channels (Arg 242 in KCNQ3) (Fig. 1A), we also tested the effect of the R242A mutation in KCNQ3T channels (Fig. 2). This mutant resulted in reduced current densities (146 Ϯ 16 pA/pF) and slowed current recovery (14.2 Ϯ 1.7 s) in the VSP assay; however, the rate of decay upon turn-on of Dr-VSP was not significantly affected. M 1 R stimulation inhibited the current by 64.8 Ϯ 9.2%, 2-fold greater than for KCNQ3T ( Fig. 2G and Fig. S1). These results are consistent with a role of Arg 242 in PIP 2 interactions. This mutation resulted also in a pronounced shift of the voltage dependence of activation toward more positive potentials (V1 ⁄ 2 ϭ Ϫ4.0 Ϯ 3.2 mV) (Fig. 2C). The adjacent mutation R243A was also tested. This mutant displayed reduced current densities as well (56 Ϯ 16 pA/pF), a faster rate of current decay upon Dr-VSP turn-on (0.59 Ϯ 0.19 s), slowed recovery after Dr-VSP turn-off (13.1 Ϯ 1.8 s), and a significantly increased M 1 Rdependent inhibition of 42 Ϯ 5.5%. Surprisingly, the voltage dependence of activation for this mutant, which is adjacent to R242A, was not altered (V1 ⁄ 2 ϭ Ϫ31 Ϯ 4.7 mV). When both arginines were mutated to alanines, the whole-cell current densities were reduced (65 Ϯ 11 pA/pF), as for the R243A single mutant. The rate of current decay and recovery after Dr-VSP turn-on or turn-off were significantly affected (0.45 Ϯ 0.11 and 15 Ϯ 1.7 s) to a greater extent than for either of the single mutations. The voltage dependence of activation of the double mutant displayed the same positive shift as for the R242A single mutant (V1 ⁄ 2 ϭ Ϫ0.2 Ϯ 2.9 mV). Last, the M 1 R-mediated inhibition of the double mutant was very high (91.6 Ϯ 2.7%; Fig. 2G). These results are consistent with an interaction of the KCNQ3 S4 -S5 linker with PIP 2 , which again seems not to be coupled to the voltage dependence of activation of the channels. Clearly, however, the S4 -S5 linker of KCNQ3 itself is coupled to channel voltage dependence or to the coupling mechanism, just not in a way that involves PIP 2 . These data are summarized in Table 1.

Interactions of PIP 2 with the S6Jx domain in KCNQ3 channels
Three basic residues (Lys 354 , Arg 360 , and Lys 362 ) in the S6Jx of KCNQ1, which are conserved in KCNQ3 (Lys 358 , Arg 364 , and Lys 366 ) ( Fig. 1A and Fig. S3E), have been found to play a role in Structural determinants of PIP 2 regulation of KCNQ3 channels  (28). In addition, Telezhkin et al. (18) found the R325A mutation in KCNQ2, homologous to Arg 360 in KCNQ1 and Arg 364 in KCNQ3, to decrease the apparent affinity of the channel for DiC8-PIP 2 , and early work implicated a role of His 328 in KCNQ2, homologous to His 367 in KCNQ3 (4). Because the Lys 358 , Arg 364 , Lys 366 , and His 367 residues in the S6Jx domain are conserved among KCNQ channels (Fig. 1A), we asked whether PIP 2 interacts with the S6Jx domain in Figure 2. Effects of charge-neutralizing mutations located in the S2-S3 and S4 -S5 linkers on KCNQ3T channels. A, representative perforated patchclamp recordings from CHO cells transfected with KCNQ3T or the indicated mutant channels. B, bars show summarized current densities at 60 mV for the indicated channels (n ϭ 6 -19). C, voltage dependence of activation of the tail currents at Ϫ60 mV, plotted as a function of test potential (n ϭ 5-19). D, representative perforated patch-clamp recordings from CHO cells co-transfected with Dr-VSP and KCNQ3T or the indicated mutant channels. E, bars summarize time constant values from single exponential fits to current decay during Dr-VSP activation (n ϭ 5-10). F, bars summarize time constants of single exponential fits to current recovery after Dr-VSP turn-off (n ϭ 5-11). G, bars summarize fractional inhibition after M 1 R stimulation for the indicated mutant channels (n ϭ 3-7). *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001. Error bars, S.E.

Structural determinants of PIP 2 regulation of KCNQ3 channels
KCNQ3T channels (Fig. 3). We found that the R364A mutation significantly decreased current amplitudes (72 Ϯ 8 pA/pF versus 197 Ϯ 6 pA/pF for KCNQ3T), whereas the K358A and K366A mutations did not (195 Ϯ 7 and 187 Ϯ 12 pA/pF, respectively) (Fig. 3, A and B). As before, we measured the responses of each mutant to PIP 2 dephosphorylation by Dr-VSP and the rate of recovery upon Dr-VSP turn-off and found the R364A mutation to result in a much faster decay of the current (0.14 Ϯ 0.02 s) upon activation of Dr-VSP and a much slower recovery of the current (27.7 Ϯ 6.9 s) upon its turn off (  Table 1). Neither response was altered for K358A and K366A ( Fig. 3 (D-F) and Table 1); nor was the maximal inhibition by M 1 R stimulation (25.1 Ϯ 6.8%).
We also tested the effect of the K358A and K366A mutations in combination with R364A as the triple mutant KRK/AAA. The KRK/AAA mutant decreased the current amplitude similarly to that of R364A (79 Ϯ 11 pA/pF), and such channels displayed a similarly reduced apparent affinity for PIP 2 ( decay ϭ 0.29 Ϯ 0.04 s and recovery ϭ 17.9 Ϯ 2.6 s, n ϭ 6 -7, p Ͻ 0.001) using the Dr-VSP assay ( Fig. 3 (A, B, and D-F) and Table 1), echoing the results of the single point mutants. None of these single point mutations significantly affected channel voltage dependence (Fig. 3C). Strikingly, however, the KRK/AAA triple mutation uniquely in this domain resulted in channels with a voltage dependence of activation markedly shifted toward more positive potentials. For KCNQ3T and KCNQ3T-KRK/ AAA, the half-activation potentials were Ϫ34.0 Ϯ 1.9 and Ϫ6.3 Ϯ 2.5 mV, respectively. We also tested the effects of the H367C mutation on KCNQ3T, which is slightly downstream of Arg 364 in the S6Jx domain. This mutation only slightly reduced current densities (138 Ϯ 5 pA/pF) but significantly increased the rate of decay of the current (0.32 Ϯ 0.05 s) upon activation of Dr-VSP and slowed its recovery (36.7 Ϯ 6.9 s) upon Dr-VSP turn-off, indicating an interaction of this residue with PIP 2 , as shown for KCNQ2 (4). Such mutant channels displayed no significant shift in the voltage dependence of activation ( Fig. 3C and Table 1). Taken together, these results strongly implicate the S6Jx domain of KCNQ3 channels as an important site for PIP 2 interactions, as for KCNQ1 channels, and this altered apparent affinity for PIP 2 also seems not linked to an altered voltage dependence of activation. 2 We previously identified a cluster of basic residues (Lys 425 , Lys 432 , and Arg 434 ) within the linker between helices A and B (A-B linker) of both KCNQ2 and KCNQ3 to be critical for PIP 2 -mediated control of gating, with the effect of mutations of this cluster in KCNQ2 somewhat more potent than in KCNQ3 (8). However, a study that deleted the A-B helix domain of KCNQ2 did not find that this deleted domain reduced the PIP 2 apparent affinity for KCNQ2 channels (29). Thus, we tested the importance of this domain of KCNQ3 using the same assays as before. We found that the deletion of the A-B linker (⌬ linker) decreased whole-cell current amplitudes by about half (112 Ϯ 10 pA/pF; Fig. 4, A and B). In cells co-expressing KCNQ3T (⌬ linker) with Dr-VSP ( Fig. 4D), the rate of current decay upon Dr-VSP turn-on was ϳ3-fold faster (0.26 Ϯ 0.04 s), compared with KCNQ3T ( Fig. 4E and Table 1), and the rate of current recovery upon turn-off of Dr-VSP was significantly slower (13.5 Ϯ 2.2 s) ( Fig. 4F and Table 1). Such data reinforce a critical role of the helix A-B linker in PIP 2 interactions with KCNQ3 channels, correlating with changes in open probability found for the triple (K425E/K432E/R434E) KCNQ3 mutant within the A-B linker previously studied in excised single-channel patches (8). Lastly, as for the other PIP 2 -interacting domains, the KCNQ3T (⌬ linker) did not display a significant shift in channel voltage dependence, with V1 ⁄ 2 values for KCNQ3T and KCNQ3T (⌬ linker) currents of Ϫ34.0 Ϯ 1.9 and Ϫ32.5 Ϯ 1.5 mV, respectively ( Fig. 4C and Table 1).

Structural determinants of PIP 2 regulation of KCNQ3 channels
sity was dramatically decreased, from 197 Ϯ 6 to 16 Ϯ 2 pA/pF ( Fig. 4B and Table 1). PIP 2 depletion induced by Dr-VSP rapidly and nearly completely abolished currents from the RH-AC/⌬ linker mutant, with a much faster rate of decay upon Dr-VSP turn-on (0.53 Ϯ 0.1 s), and a much slower rate of recovery upon turn-off of Dr-VSP (45.8 Ϯ 5.2 s), than for KCNQ3T channels ( Fig. 4 (D-F) and Table 1). The small amplitude of the currents from such severely mutated channels tested here precludes any significant meaning from comparing data from those channels and those from the RH-AC or the ⌬ linker mutant alone. They do reinforce the presence of two major PIP 2 -interaction sites within the C terminus of KCNQ3 channels, one in the A-B linker, as previously reported (8), and the other within the S6Jx domain.
Recently, the first two residues of a three-lysine cluster located at the end of the B-helix of KCNQ1 (Lys 526 -Lys 527 -Lys 528 ) have been identified as a critical site where CaM competes with PIP 2 to stabilize the open state of KCNQ1-contain- . C, voltage dependence of activation of the tail currents at Ϫ60 mV, plotted as a function of test potential (n ϭ 6 -19). D, representative perforated patch-clamp recordings from CHO cells co-transfected with Dr-VSP and KCNQ3T or mutant KCNQ3T channels. E, bars summarize time constants from single-exponential fits to current decay during Dr-VSP activation (n ϭ 5-10). F, bars summarize time constants from single-exponential fits to recovery after Dr-VSP turn-off (n ϭ 5-11). The current traces from KCNQ3T in A and D are from the same cell as in Fig. 2 (A and D), and the summarized data for KCNQ3T in B-F are the same as in Fig. 2, as these data serve as the baseline for all of the sets of mutants shown in Figs. 2-4. KCNQ3T and all mutants were tested contemporaneously. **, p Ͻ 0.01; ***, p Ͻ 0.001. Error bars, S.E.

Structural determinants of PIP 2 regulation of KCNQ3 channels
ing channels (30,31). Because this site is conserved in KCNQ3 (Lys 531 -Lys 532 -Lys 533 ), we independently mutated the three lysines to asparagines and tested them for interaction with PIP 2 using our VSP approach. Neither the current decay nor recovery was altered for any of the three mutations (Table 1), indi-cating that this basic cluster is not involved in PIP 2 interactions with KCNQ3. Whether this site plays a role in CaM modulation of KCNQ3 channels remains to be determined. It is likely that the involvement of this domain differs between KCNQ1 and KCNQ3.  8 -19). C, shown are the amplitude of tail currents at Ϫ60 mV, plotted as a function of test potential from KCNQ3T and KCNQ3T (⌬ linker) channels (n ϭ [11][12][13][14][15][16][17][18][19]. D, representative perforated patch-clamp recordings from CHO cells co-transfected with Dr-VSP and KCNQ3T or KCNQ3T (⌬ linker) or the RH-AC/⌬ linker mutants. E, bars summarize time constants from single-exponential fits to current decay during Dr-VSP activation (n ϭ 6 -10). F, bars summarize time constants from single-exponential fits to recovery after Dr-VSP turn-off (n ϭ 6 -11). The current traces from KCNQ3T in A and D are from the same cell as in Fig. 2, A and D, and the summarized data for KCNQ3T in B-F are the same as in Fig. 2, as these data serve as the baseline for all of the sets of mutants shown in Figs. 2-4. KCNQ3T and all mutants were tested contemporaneously. *, p Ͻ 0.05; ***, p Ͻ 0.001. Error bars, S.E.

Structural determinants of PIP 2 regulation of KCNQ3 channels Differences in plasma membrane expression of KCNQ3T mutant channels do not explain altered current amplitudes
Because we use whole-cell current amplitudes as one measure of PIP 2 sensitivity in this study, it was incumbent upon us that we rule out the possibility of differential membrane expression of the mutants suggested to have altered apparent affinity for PIP 2 , because this would confound our results. We and others have found visualization of membrane proteins tagged with fluorescent proteins under total internal reflection fluorescence (TIRF, evanescent wave) microscopy, which isolates emission from fluorophores within 300 nm of the membrane (32), to be by far the most reliable measure of such membrane expression (20,33). Under TIRF illumination, we measured the emission from enhanced yellow fluorescent protein (EYFP)tagged WT and mutant KCNQ3T channels expressed in Chinese hamster ovary (CHO) cells (Fig. 5). These data indicate that the decrease of the whole-cell current density is not due to divergent expression of mutant KCNQ3T channels in the plasma membrane. In fact, the EYFP emission from KCNQ3T ϩ H257N was even higher than that of KCNQ3T, suggesting that the H257N mutation increases the number of channels at the plasma membrane. Thus, differential membrane abundance of channel proteins does not underlie the differences in macroscopic current amplitudes reported in this study.

PIP 2 is predicted to interact with the S4 -S5 linker/S6Jx interface of KCNQ3 channels
Our electrophysiological data are consistent with localization of KCNQ3-PIP 2 interactions to four distinct cytoplasmic locations: the A-B helix linker, the S6Jx domain, the S2-S3 linker, and the S4 -S5 linker (Fig. S3E). In an attempt to construct a framework of these four sites into a coherent structural model of PIP 2 interactions with the channels, we performed homology modeling and PIP 2 docking simulations for all of the mutants studied in this work. Our overall hypothesis emerging from the experimental data supposes a network of interactions between basic residues located in the S2-S3 linker, the S4 -S5 linker, and the S6Jx that, together with the A-B helix linker, governs PIP 2 -mediated regulation of KCNQ3 channel gating. As above, we divide the channel into three basic modules: the voltage sensor domain (VSD), comprising S2-S4, the pore domain (PD), from S5 to S6, and the C terminus, of which the proximal half (up to the end of the B helix) is the site of several regulatory molecules, and so we call it the regulatory domain. We show models of the VSD, PD, and S6Jx based on the coordinates of the Kv1.2 channel solved in the activated/open conformation (34). Arg 190 and Arg 195 lie within the S2-S3 linker, which is part of the VSD; Arg 242 and His 257 lie within the PD; and Lys 358 and Arg 364 are within the S6Jx, which our model predicts also to be in continuous interface with the PD (Fig. 1 (B and C); Lys 366 and His 367 are not displayed). We did not construct a model of PIP 2 binding to the A-B helix linker, due to the lack of a suitable template.
To model the putative network of interactions of PIP 2 with KCNQ3 channels, we first built structural models of WT and mutant KCNQ3 channels and performed PIP 2 docking simulations to the most energetically favorable WT (Fig. 1, D and E) and mutant KCNQ3 models (Fig. 6). It is widely thought that positively charged amino acids are mostly responsible for interactions with PIP 2 . Thus, we first simulated the interaction of PIP 2 in the presence of all available positive charges on the protein in the open conformation of WT KCNQ3. In the preferred location for PIP 2 binding in the WT KCNQ3 model (Fig. 1E), the phosphate headgroup of PIP 2 is predicted to be directed toward Arg 242 and Arg 243 in the S4 -S5 linker and Lys 358 and Lys 366 in the S6Jx and also predicted to form hydrogen-bond interactions with the nearby residues within the same subunit in both the S4 -S5 linker and S6Jx (Fig. 1E, residues in blue in Sub-D). Of note, the acyl tail of PIP 2 is predicted to be directed toward residues in the inner face of S5 (His 257 ) and S6 (Phe 343 , Phe 344 , Leu 346 , and Pro 347 ) in the neighboring subunit (Fig. 1E, residues in orange in Sub-C). Thus, PIP 2 appears to be cross-linking neighboring subunits, in analogy with a role for PIP 2 reported for GIRK2 channels (35). Taken together, our simulations find that PIP 2 is predicted to interact with the S4 -S5 linker/S6Jx interface (Fig. 1E), suggesting a mechanistic basis for the effect of mutations in these regions on the favorability for activation (i.e. PIP 2 interactions with the S4 -S5 linker/S6Jx interface to stabilize and promote opening).

Structural determinants of PIP 2 regulation of KCNQ3 channels Multiple sites of PIP 2 interactions at the VSD-PD interface of mutant KCNQ3 channels
In line with previous studies on KCNQ1 and KCNQ2 channels (11,28), positively charged residues of the S4 -S5 linker (Arg 242 and Arg 243 ) and S6Jx (Lys 358 and Lys 366 ) in the same subunit (Fig. 1E, residues colored in blue) and S5 of the neighboring subunit (His 257 ) (Fig. 1E, residues colored in orange) are predicted to be involved in the interactions of PIP 2 with WT KCNQ3. However, our experimental data demonstrate that mainly Arg 190 , Arg 242 , Arg 243 , His 257 , Arg 364 , and His 367 are the determinants of PIP 2 interactions, whereas Lys 358 and Lys 366 did not seem important. Therefore, we used our model to ask whether these sites are predicted to alter PIP 2 interactions. We analyzed PIP 2 docking simulations for the following mutants: R190Q, R242A, H257N, R364A, KRK/AAA, H367A, K358A, and K366A (Fig. 6). Unlike WT KCNQ3, PIP 2 docking simulations of R190Q (Fig. 6A), H257N (Fig. 6C), R364A (Fig. 6D), H367A (Fig. 6F), and K366A (Fig. 6H) predict a network of interactions mainly with two positively charged residues of the S4 -S5 linker (Arg 242 and Arg 243 ) and one in S6Jx (Lys 358 ) of the same subunit. Simulations of KRK/AAA (Fig. 6E) and K358A (Fig. 6G) mutants predict that PIP 2 interacts similarly with Arg 242 and Arg 243 of the S4 -S5 linker but in those cases stabilizes the network of interactions with His 257 in S5 of the neighboring subunit. Noteworthy for all these mutants, Arg 242 is predicted as a common residue in the network of interactions of PIP 2 . Moreover, PIP 2 docking simulations of R242A (Fig. 6B) suggest a network of interactions with Arg 243 of the S4 -S5 linker and two positively charged residues in S6Jx (His 363 and Lys 366 ). Moreover, the R242A, H257N, and KRK/AAA mutations are predicted to cause major structural rearrangements in the S4 -S5 linker, S5, S6, and S6Jx (Fig. S2). Again, we realize that the experimental data reported little functional effects of charge neutralization of the Lys 358 and Lys 366 residues that might have been predicted to stabilize the interactions of PIP 2 with the channels. However, the simulations of PIP 2 with K358A and K366A (Fig. 6, G and H) predict that whereas the orientation of PIP 2 in the inner face of S6Jx is opposite of that predicted for WT channels, the predicted interactions at residues Arg 242 and His 257 are predicted to preserve coupling to channel gating by maintaining coupling between the S4 -S5 linker and the S6Jx. Alternatively, as stated above, our model may not have such single-residue precision that corresponds to a transmembrane ion channel in situ.

Additional sites of PIP 2 interactions at the S2-S3 interface with KCNQ3 channels
Given the lack of correlation between PIP 2 interactions and modification of the voltage dependence of activation observed in our data (Fig. S3E), we generated additional structural models of KCNQ3 in the closed state using as a template the coordinates of the Kv1.2 channel solved in the resting/closed state (34). For the modeled closed KCNQ3 channels, the inositol ring of PIP 2 is predicted to be oriented toward Lys 103 in S1, Arg 188 in the S2-S3 linker, and Arg 227 and Arg 230 in S4, whereas the acyl tail of PIP 2 is predicted to form hydrogen bonds with residues in S2 and S4 within the same subunit (Fig. 7, A-C). To correlate these predictions with function, we performed additional patch-clamp experiments, evaluating the effect of charge-neutralizing mutations on current density and on the apparent PIP 2 affinity, again using the Dr-VSP approach. We found that substitution of positively charged residues with an alanine significantly reduced the current density of K103A (19.9 Ϯ 2.7 pA/pF) and R188A (20.3 Ϯ 6.5 pA/pF), compared with KCNQ3T (36.5 Ϯ 2.7 pA/pF) but had no effect on R227A and R230A (Table 2). However, all of the mutants displayed an accelerated rate of decay of the current upon turn-on of Dr-VSP, compared with KCNQ3T. For KCNQ3T, KCNQ3T-K103A, KCNQ3T-R188A, KCNQ3T-R227A, and KCNQ3T-R230A, the rates of decay were 0.84 Ϯ 0.13, 0.29 Ϯ 0.05, 0.48 Ϯ 0.11, 0.18 Ϯ 0.03, and 0.20 Ϯ 0.03 s, respectively (Fig. 7D). All of the point mutants displayed a slower rate of recovery compared with KCNQ3T. We then wondered whether the combined K103A/ R188A or R227A/R230A double mutations would result in a synergistically greater reduction in current density and in apparent PIP 2 affinity than either mutation alone. We found the current density of KCNQ3T-K103A/R188A (21.5 Ϯ 2.9 pA/pF) to be similar to that of the single-point mutations, but the rate of current decay upon turn-on of Dr-VSP was 2-fold faster (0.39 Ϯ 0.05 s) than that of KCNQ3T channels and intermediate between the K103A (0.29 Ϯ 0.05 s) and R188A (0.48 Ϯ 0.11 s) mutants. In contrast, the current density of the R227A/ R230A double mutant was significantly lower (23 Ϯ 3 pA/pF) than that of KCNQ3T. Moreover, the rate of current decay upon turn-on of the DR-VSP was 3-fold faster (0.27 Ϯ 0.04 s) than that of KCNQ3T but slower than those from single R227A and R230A mutants (Table 2). Finally, both double mutants displayed a slower current recovery after turn-off of Dr-VSP (25.4 Ϯ 3.4 and 28.6 Ϯ 5.0 s) than KCNQ3T, quite similar to those of single mutants ( Table 2). These data suggest that Lys 103 in S1, Arg 188 in the S2-S3 linker, and Arg 227 and Arg 230 in S4 play roles in PIP 2 interactions with KCNQ3, but that they do not act synergistically. These data are also consistent with the predictions of our modeling/docking simulations, giving us further confidence in the fidelity of our modeling. Interestingly, Arg 188 is conserved in KCNQ2 but not in other KCNQ channels, suggesting that this residue may also interact with PIP 2 in KCNQ2. Unlike Arg 188 , Arg 227 is conserved in all KCNQ channels and may also be involved in PIP 2 interactions with KCNQ1-5 channels.
While this manuscript was being prepared, the structure of most of the frog Xenopus oocyte analog of mammalian KCNQ channels, ("KCNQXem") bound to CaM (PDB entry 5VMS) was solved by cryo-EM (36). We here used as a template the mammalian shaker Kv1.2 K ϩ channel (37), which has well-validated data to build new structural models based on the highly conserved structural organization of the voltage sensor domain and the pore domain and moreover is not bound by CaM (37)(38)(39)(40). Sun and MacKinnon (36) suggested their KCNQXem-CaM complex to be in a "decoupled" state (PIP 2 -free state) or in a transitory conformational state between an open PIP 2 -bound activated state and a closed PIP 2 -bound deactivated state. We analyzed the alignments between Kv1.2 and KCNQXem-CaM structures, along with our KCNQ3 structural model, and found striking differences between the structures that suggested that  36. D, top, representative perforated patch-clamp recordings from CHO cells co-transfected with Dr-VSP and KCNQ3T or the indicated mutants. Cells were held at Ϫ60 mV, current decay was measured at 100 mV, and recovery of the current was measured at 0 mV after the depolarization to 100 mV. Note the larger amplitude of the recovery current in these experiments after turn-off of Dr-VSP, due to the voltage used (0 mV), at which the "leak" current is expected to be minimal, compared with ϩ30 mV. Bottom, bars summarize the data from these experiments (n ϭ 5-11). *, p Ͻ 0.05; **, p Ͻ 0.01. Error bars, S.E.

Structural determinants of PIP 2 regulation of KCNQ3 channels
the CaM-free Kv1.2-based KCNQ3 model was superior. The details of those structural comparisons are shown in Fig. S3  (A-D).

Discussion
In the present work, we investigate the molecular determinants involved in the regulation of KCNQ3 channels by PIP 2 . Many studies have investigated the sites of action of PIP 2 on ion channels, including voltage-dependent K ϩ channels (K v ). However, the location of these sites remains controversial. For KCNQ2 and KCNQ3 channels, we have previously highlighted critical PIP 2 -interaction domains in the A-B helix linker (8).
Others have identified the S6Jx domain as important for KCNQ1-3 (10,18,41), and our results here are in accord with those reports. Recent work studying KCNQ1-containing channels has illuminated important PIP 2 -interaction domains in the S2-S3 and S4 -S5 linkers that play a role in coupling to gating (10,42,43). This study is in accord with those findings as well for KCNQ3, in terms of there being additional domains of PIP 2 interactions. Another recent study suggested that the voltage dependence of KCNQ2 channels is regulated via PIP 2 interactions with the S2-S3 and S4 -S5 linkers (11). We do not find similar results for KCNQ3. Finally, another group recently suggested that deletion of the A-B linker does not affect the apparent affinity of KCNQ2 for PIP 2 (29); however, in retrospect, we wonder if the VSP method is suitably applicable for such low-PIP 2 -affinity channels, given the extremely brief "dwell time" that PIP 2 must manifest for them and a correspondingly high k off rate, especially compared with the rate of PIP 2 dephosphorylation by Dr-VSP. Finally, the current work here, studying KCNQ3, is consistent with our earlier studies implicating the importance of the A-B linker domain (8).

Comparison of the regions of KCNQ1-3 channels contributing to PIP 2 interactions
The present work, reporting that Arg 364 and His 367 mutations of KCNQ3T, corresponding to R325A and H328C in KCNQ2, are also highly involved in PIP 2 interactions, is in accord with previous work on KCNQ2 (11,18). For the family of PIP 2 -regulated inward rectifier K ϩ (K ir ) channels, the JxS6 domain of KCNQ channels is analogous to the C-terminal domain just after M2, which has long been identified as a hot spot for PIP 2 interactions by mutagenesis studies (44) and confirmed by the solved crystal structure of PIP 2 bound to GIRK2 channels (35). Remarkably, our simulation studies predict that PIP 2 is stabilized between neighboring subunits in the S6Jx, which is similar to that reported for GIRK2 channels in the analogous domain. Hence, we suppose this structural mechanism to be likely conserved among PIP 2 -regulated channels in general. We speculate that the dual A and B helices, both containing calmodulin-binding domains, possessed by KCNQ, but not K ir , channels, endow the A-B linker of KCNQ channels as a more unique site of PIP 2 interactions, for reasons that will likely require more structural studies of these proteins.
Although our results here also show PIP 2 interactions with the S2-S3 and S4 -S5 linkers in the VSD of KCNQ3, as for KCNQ1, and that small, yet definite PIP 2 -sensitive and voltagegated currents are still produced by KCNQ3T channels mutated to lack interactions with both domains in the C terminus, we do not find the interactions with the S2-S3 and S4 -S5 linkers to be coupled to modifications of voltage dependence of the currents. Because the work on KCNQ1 channels showed that such linkage to PIP 2 was not via alterations in the sensitivity of the voltage sensor but rather due to the efficiency of coupling between the voltage sensor and the gating machinery (10), we hypothesize that the role of PIP 2 interactions in such coupling is probably similar in nature between KCNQ1 and KCNQ3, and likely KCNQ2 as well. Interestingly, a striking difference between KCNQ1-containing channels and KCNQ2-4 is that whereas currents from the latter are depressed by Ca 2ϩ /calmodulin, those of the former are enhanced (45)(46)(47)(48)(49)(50)(51). Given that both critical PIP 2 -interaction domains in the C terminus of KCNQ1-3 channels are very likely to be surrounded by Ca 2ϩ /calmodulin, we are very interested to learn the relationship between calmodulin and PIP 2 interactions and voltage-dependent coupling and the perhaps subtle yet important differences that confer opposite effects of Ca 2ϩ loading of calmodulin on the function of KCNQ1-containing channels versus KCNQ2-4.
The basic residues of both S2-S3 and S4 -S5 linkers are highly conserved among KCNQ channels. In our experiments, K103A, R188A, R190Q, R227A, and R230A, but not the R195Q or R195A mutations, in S1, the S2-S3 linker, and S4 induced a decrease of the apparent affinity for PIP 2 . Lys 162 in the S2-S3 linker of KCNQ2 has been implicated in PIP 2 -channel interactions in the closed state, supported by molecular dynamics simulations (11). Our PIP 2 -docking simulations of KCNQ3 channels also suggest that PIP 2 interacts with S1 (Lys 103 ), the S2-S3 linker (Arg 188 and Arg 190 ), and S4 (Arg 227 and Arg 230 ) of closed

apparent affinity of mutations predicted to interact with KCNQ3 channels in the closed state
Values represent mean Ϯ S.E. * and **, p Ͻ 0.05 and p Ͻ 0.01 (one-way analysis of variance with Dunnett's multiple-comparison test) statistically different from WT. ND, not determined.

Structural determinants of PIP 2 regulation of KCNQ3 channels
KCNQ3 channels. In the simulations of KCNQ3 (R188A and R190Q), PIP 2 was predicted to interact with the S2-S3 linker and to lose intersubunit contacts, which might favor channel deactivation. As opposed to previous observations in Shaker and Kv1.2 channels in which the S2-S3 linker has been suggested to interact with PIP 2 preferentially in the closed state (52,53), our experimental results suggest that disruption of PIP 2 interactions with the S2-S3 linker hinder opening. The modeling/docking simulations are consistent with the opening of KCNQ3 channels involving PIP 2 interactions at the VSD-PD interface, consistent with PIP 2 /KCNQ channel interactions involving a complex network of basic residues along the VSD-PD interface and the C terminus that cooperatively favor opening. They also suggest that a structural mechanism of channel opening involves PIP 2 -mediated intersubunit interactions. Interestingly, such PIP 2 -channel interactions have also been described in the crystal structures of K ir 2.2 and GIRK2 (K ir 3.2) channels, corresponding to the S4 -S5 linker, pore domain, and C terminus in KCNQ channels (12,35). Although we do not here find the involvement of PIP 2 interactions with the S4 -S5 linker per se to be coupled to voltage dependence of activation, our electrophysiological data and our homology modeling are fully in accord with S4 -S5 linker and S6 being critical in the coupling between the VSD and the pore domain, as is generally widely seen for voltage-dependent K ϩ channels (10, 38, 54 -56).
Because only charge-neutralizing mutations in the S4 -S5 linker (R242A and H257N) and the S6Jx (K358A/R364A/ K366A), reduced PIP 2 apparent affinity and shifted the voltage dependence of KCNQ3 toward more depolarized potentials, we hypothesize that 1) cooperation between the S4 -S5 linker and the S6Jx stabilizes opening of KCNQ3 and 2) PIP 2 likely plays a role in this coupling, a hypothesis consistent with the Kv1.2-2.1 crystal structure in which anionic lipids are bound at the VSD-PD interface of the channel (57). However, one central question remains unclear as to generality among K ϩ channels: Does PIP 2 affect the voltage-sensor movement and, by that mechanism, the voltage dependence of K v channels, or do any effects of PIP 2 on channel voltage dependence generally arise from changes in coupling between the VSD and the PD? In Kv1.2, replacement of an arginine with a glutamine (R322Q) in the S4 -S5 linker, which is involved in VSD-PD coupling, affected the channel voltage dependence of activation when PIP 2 was depleted. Moreover, gating current experiments showed that PIP 2 affects the VSD movement of Shaker channels through interactions with the S4 -S5 linker (53). However, unlike for Shaker, depletion of PIP 2 does not affect VSD movement of KCNQ1 homomers (10). Different laboratories have come to divergent conclusions about whether PIP 2 -dependent modulation of KCNQ1-containing channels shifts the voltage dependence of activation, with one group positing that it does (3,58) but another group concluding that it does not (10,59). Our data here are in accord with the latter conclusion for the case of KCNQ3 channels, consistent with the conclusions for KCNQ2/3 heteromers (23)(24)(25). The presence or absence of KCNE1 subunits is unlikely to alter such conclusions for KCNQ1, because KCNE1 was shown to have no direct impact on VSD activation or pore opening, but rather to affect VSD-PD coupling (60). Consistent with this, a point mutation (F351A) at the VSD-PD interface had similar effects on KCNQ1 as did inclusion of KCNE1 in the channel. In that work, both KCNE1 and the F351A mutation abolished the "intermediate-open state" of KCNQ1-containing channels, promoting the activated-open states of KCNQ1 by increasing its PIP 2 affinity (59 -61), besides the suppression of inactivation (62). We tentatively conclude that PIP 2 does not contribute generally to the voltage dependence of all KCNQ channels, including KCNQ1, as we found for KCNQ3, but is much more likely to be involved in the efficiency of VSD-PD coupling. We suspect, but cannot at this point provide evidence, that the underlying reason is the display of two distinct open states of all KCNQ channels (42,63), leading to state transitions, and PIP 2 actions on voltage dependence, differing from those of other K v channels.
Although we now are in accord with four distinct regions of KCNQ1-3 channels interacting with PIP 2 , we cannot rule out yet more PIP 2 -binding sites. The distal C terminus contains basic residues that are conserved in all KCNQ channels, which may also contribute to PIP 2 . Our experiments show that the triplet of lysines (Lys 531 , Lys 532 , and Lys 533 ) located at the end of the B-helix of KCNQ3 do not interact with PIP 2 . However, Arg 539 and Arg 555 located in the distal C terminus of KCNQ1 (within the C-helix) were reported to decrease the affinity of the channel to DiC8-PIP 2 (26), and Lys 526 , Lys 527 , and Lys 528 have been identified as a critical fifth site where CaM competes with PIP 2 to stabilize the open state of KCNQ1-containing channels (30,31). The possibility of other PIP 2 -interacting sites at the end of the regulatory domain is intriguing, given the location of the site of phosphorylation of KCNQ3 channels by protein kinase C (64), because such phosphorylation would add a counteracting negative charge at that locus. This could be a "hot spot" of PIP 2 /protein kinase C cross-talk, both of which are affected by stimulation of G q -coupled receptors. Such a highly intriguing possibility needs to be carefully examined for all KCNQ2-4 channels as well as KCNQ2/3 heteromers that underlie most M-type K ϩ currents in the nervous system.

Cell culture and transfection
CHO cells were grown in 100-mm tissue culture dishes (Falcon, Franklin Lakes, NJ) in Dulbecco's modified Eagle's medium with 10% heat-inactivated fetal bovine serum plus 0.1% penicillin/streptomycin in a humidified incubator at 37°C (5% CO 2 ) and passaged every 4 days. Cells were discarded after ϳ30 passages. For patch-clamp and TIRF experiments, CHO cells were first passaged onto 35-mm plastic tissue culture dishes and transfected 24 h later with FuGENE HD reagent (Promega), according to the manufacturer's instructions. The next day, cells were plated onto coverglass chips, and experiments were performed over the following 1-2 days.

Perforated patch electrophysiology
Pipettes were pulled from borosilicate glass capillaries (1B150F-4, World Precision Instruments) using a Flaming/ Brown micropipette puller P-97 (Sutter Instruments) and had resistances of 2-4 megaohms when filled with internal solution Structural determinants of PIP 2 regulation of KCNQ3 channels and measured in standard bath solution. Membrane current was measured with pipette and membrane capacitance cancellation, sampled at 5 ms, and filtered at 500 Hz by means of an EPC9 amplifier and PULSE software (HEKA/Instrutech). In all experiments, the perforated patch method of recording was used with amphotericin B (600 ng/ml) in the pipette (65). Amphotericin was prepared as a stock solution as 60 mg/ml in DMSO. In these experiments, the access resistance was typically 7-10 megaohms 5-10 min after seal formation. Cells were placed in a 500-l perfusion chamber through which solution flowed at 1-2 ml/min. Inflow to the chamber was by gravity from several reservoirs, selectable by activation of solenoid valves (Warner Scientific). Bath solution exchange was essentially complete by Ͻ30 s. Experiments were performed at room temperature.
Currents were studied by holding the membrane potential at Ϫ80 mV and applying 800-ms depolarizing pulses from 60 to Ϫ80 mV, every 3 s. Basal KCNQ current amplitudes were measured at 60 mV. To estimate voltage dependence, tail current amplitudes were measured ϳ10 -20 ms after the repolarization at Ϫ60 mV, normalized, and plotted as a function of test potential. The data were fit with Boltzmann relations of the form, I/I max ϭ I max /(1 ϩ exp((V1 ⁄ 2 Ϫ V)/k)), where I max is the maximum tail current, V1 ⁄ 2 is the voltage that produces half-maximal activation of the conductance, and k is the slope factor. Cell populations were compared using a two-tailed t test. To evaluate the apparent affinity of WT and mutant KCNQ3T channels for PIP 2 , we used Dr-VSP cDNA cloned into the pIRES-EGFP bicistronic vector, so that transfected cells would express similar copies of Dr-VSP and EGFP. The cells patched were chosen based on their visible EGFP fluorescence as described previously. Current decay was measured at 120 or 100 mV, normalized, and plotted as a function of time. Recovery of the current was quantified at 30 or 0 mV (which is negative to activation of Dr-VSP) after depolarization to 120 or 100 mV. The rate of current recovery was quantified with a single-exponential fit as described previously, which we realize is an approximation due to the confound of the known rate of PI(4)P-5 kinase ( ϳ 10 s at room temperature) (14), and the rate of current decay was quantified ϳ30 ms after the activation of Dr-VSP at 120 mV with single exponential fits. Finally, the steady-state inhibition of the current by Dr-VSP was quantified by comparing current at 30 mV or 0 mV before and after activation of Dr-VSP. Data are given as the mean Ϯ S.E.

TIRF microscopy
Fluorescence emission from EYFP-tagged KCNQ3T and KCNQ3T mutants (R190Q, R242A, H257N, R364A, KRK/ AAA, H367C, ⌬ linker, and RH-AC/⌬ linker) were collected at room temperature using TIRF (also called evanescent field) microscopy. Total internal reflection fluorescence generates evanescent field illumination normal to the interface between two media of differing refractive indices, the coverglass and water in this case, that declines exponentially with distance, illuminating only a thin section (300 nm) of the cell very near the coverglass, including the plasma membrane (32). All TIRF experiments were performed on a Nikon TE2000 microscope mated to a Prairie Technologies laser launch delivery system, as described previously (20). Images were not binned or filtered, with a pixel size corresponding to a square of 122 ϫ 122 nm. The reader should know that this system has now been very significantly upgraded.

Structural homology, simulation, and docking models
The human KCNQ3 channel sequence in FASTA format (Uniprot ID O43525) was loaded into Swiss-PdbViewer version 4.10 (66) for template searching against the ExPDB database (ExPASy). Then the structural model for the full length of the Rattus norvegicus voltage-gated K ϩ channel subfamily A member 2 (Kv1.2; PDB entry 3LUT) (54) was identified as the best template. The initial sequence alignments between the KCNQ3 channel and Kv1.2 were generated with full-length pairwise alignments using ClustalW (67). Sequence alignments were inspected manually to assure accuracy among structural domains solved from the template. Because the turret domain of the KCNQ3 subunit was absent in the solved Kv1.2 structure, residues 287-296 were excluded from the modeling. The A315T pore mutation was also omitted from the template, because it does not change the apparent PIP 2 affinity of the channel (19). Full-length multiple alignments were submitted for automated comparative protein modeling implemented in the program suite incorporated in SWISS-MODEL (http:// swissmodel.expasy.org). 5 Before energy minimization using GROMOS96 (68), the resulting structural models of KCNQ3 subunits were manually inspected, the structural alignments were confirmed and evaluated for proper hydrogen bonds, and the presence of clashes and missing atoms was estimated using Molegro Molecular Viewer. Further structural models were generated by rearrangement of four KCNQ3 subunit models as a tetramer. Coordinates of the Kv1.2 channel in the resting/closed and activated/open states (34) were used to model the KCNQ3 channel in both forms. The calculated energies for the corresponding KCNQ3 open and closed stated structural models were highly favorable (Ϫ35,580 and Ϫ27,656 kJ/mol, respectively). Neighborhood structural conformational changes caused by the introduction of single point mutations of the KCNQ3 structure were simulated using Rosetta version 3.1 (69) and implemented in the program suite incorporated in Rosetta Backrub. As Rosetta 3.1 does not allow cysteine substitutions, we modeled KCNQ3 subunits (WT or mutant) with cysteines exchanged for alanines. Simulations for single point mutations were carried out for dimers, for which identical mutations were presented in neighboring subunits, excluding distal residues of the C terminus (residues 404 -557).
Up to 20 of the best-scoring structures were generated at each time by choosing parameters recommended by the appli- 5 Please note that the JBC is not responsible for the long-term archiving and maintenance of this site or any other third party hosted site.

Structural determinants of PIP 2 regulation of KCNQ3 channels
cation. The root-mean square (r.m.s.) deviation was calculated between the WT structures and superimposed on the simulated mutant structures. For each mutation, the r.m.s. average over 10 low-energy structures was computed, and conformational changes were displayed among neighboring structural domains considered significant for values of r.m.s. Ͼ 0.5 Å. PatchDock (70), a molecular docking algorithm based on shape complementarity principles, was used to dock PIP 2 with proposed interacting domains at the interfaces of dimer homology models based upon the Kv1.2 structure. One PIP 2 ligand was simulated docked per subunit, with the structure of PIP 2 used as in the solved PIP 2 -bound structure of Kv2.2 (12). PatchDock was implemented using an algorithm applied for protein-small ligand docking with a default clustering of 1.5 Å of the r.m.s. as recommended. Before the simulation, a list of residues for three predicted binding sites for PIP 2 in the docking site was derived, as indicated by functional studies, which included domains within the S2-S3 and S4 -S5 linkers and the proximal C terminus. Twenty solutions for the first and the fifth best-scoring simulated mutant were ranked according to the geometric shape complementarity score and the atomic contact energy (Ϫ171 and Ϫ243 kcal/mol for the open and closed states, respectively) (71) and inspected manually to assure accuracy among representative orientations of bound PIP 2 . The energy electrostatic interactions for a given docking pose (ligandprotein complex) were analyzed using the ligand energy inspector implemented through the Molegro Molecular viewer. The short-range electrostatic interactions (r Ͻ 6 Å) between the PIP 2 and residues in WT or mutant were computed, and the lowest solutions among those with the highest geometric score and the right orientation are represented here. We prepared the modeling figures using Chimera version 1.7 (72).