The Selectivity Filter May Act as the Agonist-activated Gate in the G Protein-activated Kir3.1/Kir3.4 K+ Channel*

The Kir3.1/Kir3.4 channel is activated by Gβγ subunits released on binding of acetylcholine to the M2 muscarinic receptor. A mechanism of channel opening, similar to that for the KcsA and Shaker K+ channels, has been suggested that involves translocation of pore lining transmembrane helices and the opening of an intracellular gate at the “bundle crossing” region. However, in the present study, we show that an extracellular gate at the selectivity filter is critical for agonist activation of the Kir3.1/Kir3.4 channel. Increasing the flexibility of the selectivity filter, by disrupting a salt bridge that lies directly behind the filter, abolished both selectivity for K+ and agonist activation of the channel. Other mutations within the filter that altered selectivity also altered agonist activation. In contrast, mutations within the filter that did not affect selectivity had little if any effect on agonist activation. Interestingly, mutation of bulky side chain phenylalanine residues at the bundle crossing also altered both agonist activation and selectivity. These results demonstrate a significant correlation between agonist activation and selectivity, which is determined by the selectivity filter, and suggests, therefore, that the selectivity filter may act as the agonist-activated gate in the Kir3.1/Kir3.4 channel.

K ϩ channels open and close in response to a number of stimuli including voltage or intracellular ligands such as G proteins. The position of the gate or gates that control channel opening and closing remains a matter of interest. Studies on the proton-gated KcsA K ϩ channel (1,2), the voltage-gated Shaker K ϩ channel (3)(4)(5) and the G protein-activated Kir3.x (6,7) or ATP-sensitive Kir6.x (8,9) inward rectifier K ϩ channels suggest that an intracellular gate is formed by the bundle crossing region of the channel. These studies are consistent with spin labeling measurements made in the KcsA channel that suggest that during proton activation the second transmembrane (TM2) 1 domains rotate and tilt away from the axis of the pore about a pivot point in a scissoring-type motion (1,2). This work is supported by crystallization of the MthK bacterial K ϩ channel in the open state (10), which shows a pivot or hinge point at a highly conserved glycine residue at position 83 (equivalent to position 99 in the KcsA channel).
However, other parts of the channels may be associated with channel opening, because although access of large MTS re-agents (such as MTSET; radius ϳ2.9 Å) to the inner vestibule of a Ca 2ϩ -activated K ϩ channel (SK) is state-dependent, smaller MTS reagents (such as MTSEA; radius ϳ1.8 Å, compared with K ϩ radius of 1.33 Å) can access as far as the selectivity filter equally well in open and closed channel states (11). Similarly, access of large MTS reagents to a cyclic nucleotide-gated channel (CNG1) is gated, while Ag ϩ (radius 1.27 Å) has state-independent access (12). This suggests that, in some channels at least, the bundle crossing may not be sufficient to impede the passage of permeant ions through the pore. On channel activation, other than the large movements of the TM2 domains that are described above, small movements at the inner portion of the selectivity filter of the KcsA channel have been observed in spin labeling experiments (1). It is therefore possible that this region of the channel may also act as a gate. Other observations are consistent with this. For example, Ctype inactivation, which occurs in many voltage-gated K ϩ channels, closes the channel by constriction of the outer mouth of the selectivity filter (13)(14)(15). Also, mutations within the selectivity filter of the Kir2.1 (16,17) and Kir6.2 (18) inward rectifier K ϩ channels or exchanging the P-loop of the Kir2.1 channel with that of the Kir1.1b channel (19) alter the kinetics of single channel flickery openings. Furthermore, a naturally occurring mutation within the selectivity filter of the G proteinactivated Kir3.2 channel (Weaver mutation) alters channel sensitivity to agonist and G protein activation (20,21).
In this study, we show that mutations that alter the structure of the selectivity filter affect the response of the G proteinactivated Kir3.1/Kir3.4 channel to agonist. While some mutations within the selectivity filter had minimal effects, disruption of a salt bridge that maintains tension in the selectivity filter abolished agonist activation. We describe a correlation between selectivity, which is determined by the selectivity filter, and agonist activation that suggests that agonist activation of the Kir3.1/Kir3.4 channel may occur at the selectivity filter. We also address the role of a putative intracellular gate in inward rectifier K ϩ channels (as suggested by the KirBac1.1 crystal structure, Ref. 22), formed by a ring of bulky hydrophobic residues. We show that these residues also contribute to agonist activation of the channel in a way that may be coupled to the gate at the selectivity filter.

EXPERIMENTAL PROCEDURES
Molecular Biology-Mutations in the Kir3.1 or Kir3.4 channel subunits were made using site-directed PCR mutagenesis and confirmed by sequencing. Plasmids (pTLNII or pGES) containing wild-type or mutant Kir3.1 or Kir3.4 channel subunits or the hD 2 (human dopamine) or M 2 (muscarinic acetylcholine (ACh)) receptor (required for agonist activation of the channel) were linearized using MluI or NotI (New England Biolabs, Beverly, MA) and transcribed in vitro using SP6 or T7 RNA polymerase (Riboprobe®; Promega, Madison, WI).
Electrophysiology-Xenopus oocytes were prepared as described previously (23). Oocytes were injected with 50 nl of cRNA encoding wildtype or mutant Kir3.1 (30 ng/l) and Kir3.4 (30 ng/l) as well as hD 2 (3.8 ng/l). In a few experiments, M 2 (3.8 ng/l) was injected instead of hD 2 ; * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  Currents were recorded using the two-electrode voltage clamp technique using a GeneClamp 500B amplifier (Axon Instruments, Union City, CA) filtering at 500 Hz and sampling at 2 kHz. Voltage protocols were generated using pClamp software (Axon Instruments) with a Digidata 1200 D/A converter (Axon Instruments). Electrodes were filled with 3 M KCl (tip resistance, 1-3 M⍀). Experiments were performed at 20 -25°C. Recordings were made in solution consisting of (mM): 90 KCl, 2 CaCl 2 , 5 HEPES, pH 7.4 (KOH). Oocytes were placed in a small bath (ϳ5 ϫ 5 ϫ 4 mm; ϳ100 l in volume) and were directly perfused at ϳ5 ml/min with a large bore manifold (ϳ1 mm in diameter) placed ϳ1 mm from the cell. To record current-voltage relationships, currents were recorded firstly in the absence of the agonist dopamine and then following 30 s perfusion of solution containing 10 M dopamine (along with 10 M ascorbic acid to prevent dopamine oxidation). Oocytes were held at 0 mV and 750 ms voltage pulses were applied from Ϫ130 to ϩ 60 mV in 10 mV increments. Currents were measured at the end of each voltage pulse. To study Rb ϩ permeation, the 90 mM KCl in the solution was replaced with 90 mM RbCl (pH titrated using RbOH). To measure the rate of channel activation on application of agonist and the effect of G␣ i3 , solution without agonist was rapidly exchanged for solution containing 10 M dopamine (to measure channel activation) or 10 M ACh (to measure the effect of G␣ i3 ) and current was recorded continuously at Ϫ80 mV. To test the ability of our perfusion system to rapidly change the solution around each oocyte, we measured the rate of current activation in the constitutively active Kir3.1[S170P]/Kir3.4[S176P] mutant channel on switching from ND96 to solution containing 90 mM K ϩ . Current activated in 0.7 Ϯ 0.1 s (n ϭ 5) and this is Ͼ10x faster than current activation on application of dopamine. In all experiments with mutant channels, data were also collected from the wild-type channel expressed in the same batch of oocytes. Data analysis was performed using Clampfit (Axon Instruments) and SigmaPlot (SPSS Science, Chicago, IL) software. Data are given as means Ϯ S.E. (n ϭ number of oocytes). Statistical analysis was performed using a Student's t test or ANOVA as appropriate. p values of Ͻ0.05 were considered to signify a significant difference.
Modeling-A comparative model of the Kir3.1/Kir3.4 tetrameric channel was constructed based on the crystal structure of KcsA as described previously (24,25).

Effect of Mutation of the Bulky Hydrophobic Residues at the
Bundle Crossing-The comparative model of the Kir3.1/Kir3.4 channel (Fig. 1) shows a narrowing of the permeation pathway at the bundle crossing formed by the large side chains of the phenylalanine residues at position 181 in Kir3.1 and 187 in Kir3.4 (Fig. 1, A and C). The recently crystallized KirBac1.1 bacterial inward rectifier K ϩ channel also possesses phenylalanine residues at the equivalent position (F146 (22)). Because their large hydrophobic side chains are ideally positioned to prevent movement of water through the narrow bundle crossing, the four phenylalanine residues in KirBac1.1 were proposed to be involved in channel activation (22). We have investigated the role of these phenylalanine residues in agonist activation of the Kir3.1/Kir3.4 channel. Fig. 2 shows the effect of replacement of the phenylalanine residues with either methionine residues, the equivalent residue in the constitutively active Kir2.1 channel, or alanine residues, which have relatively small hydrophobic side chains. The left hand panel shows currents through the wild-type channel recorded in the presence of 90 mM K ϩ during 750 ms voltage pulses to Ϫ130 to ϩ60 mV from a holding potential of 0 mV. The top set of traces was recorded in the absence of agonist, the middle set was recorded in the presence of agonist (10 M dopamine) and the bottom set shows the agonist-activated current, the difference in current with and without agonist. Mean current-voltage relationships are also shown. In the absence of agonist, current was recorded through the wild-type channel (Fig. 2). Basal current is thought to be due to a high level of free G␤␥ and/or a low level of G␣ i within the Xenopus oocyte (26). Nevertheless, wild-type current was substantially increased, by 60.7 Ϯ 6.4%, on application of agonist (Fig. 2).
Replacement of both pairs of phenylalanine residues with methionine residues (Kir3. fect on the agonist dependence of the channel. This can be seen in Fig. 2, which shows current traces and mean current-voltage relationships for the mutant channel, and also in Fig. 3A (bar 4), which shows the agonist-activated current (at Ϫ130 mV) of various mutant channels normalized to that of the wild-type channel recorded from the same batch of oocytes. Although mutation to a methionine residue in all four subunits had no effect, mutation in either Kir3.1 or Kir3.4 alone did have a modest effect on agonist dependence ( In proline-scanning experiments of the TM2 domain (7), introduction of proline residues affected agonist activation with an ␣-helical periodicity (residues highlighted in blue in Fig. 1D). The double mutation, Kir3.1[S170P]/Kir3.4[S176P], had the greatest effect on agonist activation and produced this by introducing a kink into the helix of the TM2 domains (7). In the present study, the double mutation, Kir3.1[S170P]/ Kir3.4[S176P], reduced agonist activation by 80.5 Ϯ 4.9% ( Fig. 2; n ϭ 5, ANOVA, p Ͻ 0.05) compared with that of the wild-type channel ( Fig 3A, bar 6).
As a control, we investigated the effect of the mutation of an aspartate residue within the TM2 domain (D173 in Kir3.1; equivalent to D172 in Kir2.1) that is thought to be involved with polyamine block of inward rectifier K ϩ channels. The mutation, Kir3.1[D173Q], had no effect on agonist activation; agonist-activated current through the Kir3.1[D173Q]/Kir3.4 mutant channel was 93.1 Ϯ 1.0% that of the wild-type channel (n ϭ 2; ANOVA, not significant).
Mutations that Disrupt the Selectivity Filter Abolish Agonist Activation- Fig. 1B shows an enlarged view of the part of the P-loop that forms the selectivity filter of the Kir3.1/Kir3.4 channel. As in the Kir2.1 channel (27), a salt bridge is thought to exist behind the selectivity filter between the negatively charged glutamate residue at position 139 (Glu-139) and the positively charged arginine residue at position 149 (Arg-149) in the Kir3.1 subunit and between Glu-145 and Arg-155 in the Kir3.4 subunit (see also the alignment in Fig. 1D). and Kir3.4[E145R,R155E] mutations, had dramatic effects on agonist activation. Current through these mutant channels was not activated by agonist, and the channels were constitutively active (Fig. 4). Again, this can also be seen in Fig. 3A (bars 8 -10).
To investigate the effect of disrupting the salt bridge on channel activation further, we measured the sensitivity of the wild-type channel and the Kir3.1/Kir3.4[E145Q] mutant channel (taken as an example of a mutant channel that lacks agonist activation) to G␤␥ activation. Fig. 5A shows current through the wild-type channel and the Kir3.1/Kir3.4[E145Q] mutant channel recorded with and without injection of 3 ng/ oocyte of G␤␥. Injection of 3 ng/oocyte of G␤␥ activated the wild-type channel, but had no effect on the Kir3.1/ Kir3.4[E145Q] mutant channel. Panels B and C of Fig. 5 show mean current-voltage relationships and dose-response curves for a range of G␤␥ concentrations. At each concentration tested, G␤␥ activated the wild-type channel, but failed to activate the Kir3.1/Kir3.4[E145Q] mutant channel (Fig. 5, B and C). These data show that the Kir3.1/Kir3.4[E145Q] mutant channel, in which the salt bridge has been broken, is insensitive to G␤␥ activation as well as to agonist activation.
Thus far, we have described mutant channels that do not respond to agonist, such as Kir3.1/Kir3.4[E145Q], as being constitutively active. However, it is possible that these channels are instead highly sensitive to G␤␥. Mutant channels may be maximally activated by endogenous G␤␥ and so do not respond to agonist. To distinguish between these two possibilities, we co-expressed G␣ i3 , which sequesters endogenous G␤␥ and abolishes basal current through the wild-type channel (26). Fig. 6A shows continuous current recordings at Ϫ80 mV from oocytes injected with wild-type or Kir3.1/Kir3.4[E145Q] mutant channels in the absence or presence of 3 ng/oocyte of G␣ i3 . These experiments were performed with the M 2 muscarinic ACh receptor. In the absence of G␣ i3 , substantial basal current was recorded through the wild-type channel and on application of agonist (10 M ACh), current was increased (Fig. 6A, left as observed previously e.g. Fig. 2). Co-expression of G␣ i3 abolished basal current, but not the agonist-activated current, through the wild-type channel (Fig. 6A, left). In contrast, basal current through the Kir3.1/Kir3.4[E145Q] mutant channel was not affected by G␣ i3 (Fig. 6A, right). Furthermore, in the absence and presence of G␣ i3 , agonist application still failed to activate the channel (Fig. 6A, right). These results are confirmed by the mean data in Fig. 6B, which shows mean current amplitudes in the absence and presence of ACh for the wild-type and mutant channels with and without G␣ i3 . These data suggest that the Kir3.1/Kir3.4[E145Q] mutant channel is indeed constitutively active and not highly sensitive to G␤␥.
Also highlighted in Fig. 1B are the alanine residue at position 142 in Kir3.1 and the threonine residue that is in the equivalent position in Kir3.4 (Thr-148). In Kir2.1, this site has been shown to be important in high affinity block of the channel by Cs ϩ , Rb ϩ , and Ba 2ϩ (28,29).  Fig. 3A (bars 11 and 12) shows that the agonist-activated , had no effect on agonist activation (Fig. 3A, bar 13; Fig. 7; n ϭ 5; not significant).
Correlation between Agonist Activation and Selectivity-The salt bridge behind the selectivity filter is likely to be important for the correct conformation of the selectivity filter (30). The dramatic effect of disruption of the salt bridge in Kir3.4 could, therefore, be the result of the disruption of the selectivity filter, rather than the salt bridge per se. It is possible that the effects of the mutation of Ala-142 and Thr-148 in Kir3.1 and Kir3.4 may again be the result of disruption of the selectivity filter. To test this, the effect of the various mutations on channel selectivity as well as agonist activation was investigated. Fig. 3B shows a measure of channel selectivity. In these experiments, the 90 mM K ϩ in the extracellular solution was replaced with 90 mM Rb ϩ . Rb ϩ permeated the wild-type channel to a certain extent; at Ϫ130 mV, Rb ϩ current was 0.35 Ϯ 0.01 (n ϭ 5) the size of the K ϩ current. However, in the case of the mutant channels in which the salt bridge in Kir3. 4  currents were 0.29 Ϯ 0.02 and 0.32 Ϯ 0.01 the size of the K ϩ current, respectively (n ϭ 2-5; ANOVA, not significant). Fig.  3C shows that agonist dependence and channel selectivity are significantly correlated (r 2 ϭ 0.47, p Ͻ 0.02). These data suggest that the selectivity filter of the Kir3.1/Kir3.4 channel may be the agonist-activated gate.
Rate of Agonist Activation-In Fig. 3C , had no effect on the size of the agonist-activated current, it did reduce selectivity; Rb ϩ current was 0.81 Ϯ 0.06 the size of the K ϩ current (n ϭ 5; ANOVA, p Ͻ 0.05). However, although the size of the agonist-activated current was not altered, the rate of agonist activation of the Kir3.1[E139Q]/Kir3.4 mutant channel was. Fig. 8 shows the rate of agonist activation in the wild-type channel and those mutant channels (including the Kir3.  1[A142T]) had no effect on the rate of agonist activation (n ϭ 6 -8; not significant).
Spermine Can Permeate the Channel in the Absence of Agonist-We have previously shown that polyamines, such as spermine, can permeate the Kir3.1/Kir3.4 channel (32-34). Fig.   9 shows that agonist is not required for spermine to permeate the wild-type channel. Fig. 9A (left) shows typical wild-type currents recorded in the presence of 90 mM K ϩ and in the absence and presence of agonist. Agonist-activated current and mean current-voltage relationships are also shown. The data are similar to those in Fig. 2. Fig. 9A (right) shows currents recorded from the same cell when the 90 mM K ϩ was replaced with 90 mM spermine. With spermine as the charge carrier, substantial current was recorded showing that spermine can permeate the wild-type channel (32)(33)(34). A substantial spermine current was present in the absence of agonist and this was not increased by agonist application (Fig. 9A, right). The spermine current through the wild-type channel was small, but a large spermine current was recorded through channels consisting of Kir3.1 subunits alone (Fig. 9B). In these experiments, the mutation, Kir3.1[F137S], was made to restore K ϩ conductance (35). In the Kir3.1[F137S] channel, as in the wild-type channel, whereas K ϩ current was agonist-dependent, the spermine current did not depend on agonist (Fig. 9B).

The Bundle Crossing May Not be the Agonist-Activated
Gate-Our comparative model of the Kir3.1/Kir3.4 channel predicts a narrowing of the permeation pathway at the bundle crossing created by bulky side-chain phenylalanine residues (Fig. 1C) just as in the KirBac1.1 crystal structure (22). It is proposed by Kuo et al. (22) that the presence of large hydrophobic residues at this position may act as a barrier to ion permeation. Consistent with this idea, replacement of the phenylalanine residues in Kir3.1/Kir3.4 with small side chain alanine residues, but not with bulky side chain methionine residues, produced channels that were constitutively active (Figs. 2 and 3A). However, these residues may not form the gate that opens and closes in response to agonist activation, because replacement with methionine residues, as in Kir2.1, had only minimal effects on agonist activation (Figs. 2 and 3A) and did not render the channel constitutively active, like the Kir2.1 channel. Furthermore, replacement with alanine residues reduced ion selectivity at the same time as agonist activation (Fig. 3B), suggesting that the effect on agonist activation may be due to disruption of a gate at the selectivity filter, not the bundle crossing. The state-independent access of small MTS reagents (11) and Ag ϩ (12) to the inner vestibule of some K ϩ channels also suggests that the bundle crossing does not form a significant barrier to ion permeation.
The Selectivity Filter May Act as the Agonist-Activated Gate-Neutralization of a glutamate residue in the P-loop (Kir3.4[E145Q]) of the Kir3.1/Kir3.4 channel abolishes K ϩ selectivity (31) and reduces Ba 2ϩ block of the channel (23). The residue E145 in Kir3.4 is thought to form a salt bridge behind the selectivity filter (27). It may provide the selectivity filter with its tension and hence hold it in its correct conformation (30). We refer to this salt bridge as a bowstring that gives the bow (the selectivity filter) its tension. In the present study, we found that the Kir3. 4 Fig. 6). Furthermore, under these conditions, agonist still failed to activate the channel. This suggests, there- fore, that the mutant channels are constitutively active rather than hypersensitive to G␤␥. Each of the mutations in Kir3.4 that abolished selectivity for K ϩ also abolished agonist activation (Fig. 3), suggesting that disruption of the normal conformation of the selectivity filter to such an extent that it cannot discriminate between monovalent ions renders the channel constitutively active and, therefore, that the selectivity filter itself acts as the agonist-activated gate.
Interestingly, disruption of the salt bridge in Kir3.1 had comparatively little effect on channel properties: the mutation Kir3.1[E139Q] had no effect on the amplitude of the agonistactivated current (Fig. 4), although it did slow the rate of agonist activation (Fig. 8). It also increased Rb ϩ current to some degree (Fig. 3B). However, using a wider spectrum of monovalent ions, we have previously shown that the effect of the Kir3.1[E139Q] mutation on selectivity is considerably less than that of the equivalent mutation, Kir3.4[E145Q] (31). This suggests that disruption of the salt bridge in Kir3.1 has a more subtle effect on the structure of the selectivity filter. This may be explained by asymmetry of the Kir3.1/Kir3.4 channel pore: the contribution of the two Kir3.1 and two Kir3.4 subunits to the structure of the selectivity filter may not be equal. There is already evidence for this: for example, the tyrosine residue of the Kir3.4 GYG motif, but not of the Kir3.1 GYG motif, is required for selectivity (37). Also, the mutation Kir3.4 [E145Q] alters Ba 2ϩ block and inward rectification whereas the equivalent mutation, Kir3.1[E139Q], does not (31).
Replacement of the threonine residue at position 148 in Kir3.4 also significantly reduced the response of the channel to agonist (Figs. 3A, 7, and 8). This mutation also altered selectivity for K ϩ (Fig. 3B) suggesting that it probably also alters the conformation of the selectivity filter. Thr-148 probably lines the selectivity filter, because in the Kir2.1 channel mutation at this site (T141A) reduces both Cs ϩ (28) and Ba 2ϩ (29) 3A and 7) and no effect on selectivity (Fig. 3B). The experiments shown in Fig. 9, where spermine was used as a charge carrier, suggest that spermine acts as a foot-in-thedoor at the selectivity filter. We have previously shown that spermine can permeate the Kir3.1/Kir3.4 channel (34) suggesting that it can access the selectivity filter. The results in Fig. 9 show that spermine can access the selectivity filter and permeate the channel not only in the presence, but also in the absence, of agonist. This suggests that spermine binding to the selectivity filter acts as a foot-in-the-door: it opens the selectivity filter and this allows permeation.
Proton-activation of the KcsA channel is associated with relatively large movement of the TM2 helices that open the aperture of the bundle crossing (1). Subsequent work on this and other K ϩ channels suggests that the TM2 helices tilt away from the central pore axis about a pivot point so that an intracellular gate at the bundle crossing opens (see Introduction). However, small movement was also observed at the KcsA selectivity filter on proton-activation (1). It is possible that these relatively small movements alter the conformation of the selectivity filter and effectively gate the access of K ϩ . The idea of a flexible selectivity filter is supported by observations that the filter adopts one conformation in the presence of low extracellular K ϩ and another in the presence of high extracellular K ϩ (38), and also it was recently shown that different distortions of the filter could produce channels with differing conductance (39). The idea that the selectivity filter might move during channel opening has been suggested previously. The affinity of the Kir3.1/Kir3.4 channel to block by Ba 2ϩ and Cs ϩ is reduced on binding of G␤␥ (40) (41). Several other studies suggest that selectivity may be coupled to gating. Replacement of the second glycine in the GYG motif of the Kir3.2 G proteinactivated channel with a serine residue (as occurs in the Weaver mutation) alters both selectivity and agonist activation (20, 21). Also, the selectivity of cyclic nucleotide-gated channels for Ca 2ϩ over Na ϩ increases with open probability (42) and the selectivity spectrum of a mutant Shaker channel changes as it passes through subconductance states on its way to the fully open state (43).
Working Hypothesis of Agonist Activation-How does the selectivity filter of the Kir3.1/Kir3.4 channel act as a gate when the activating signal, the binding of G␤␥, occurs within the cytoplasmic domain of the channel? Given the wealth of evidence supporting the idea of an intracellular gate and the evidence reported in this study, it is possible that two gates are present in the Kir3.1/Kir3.4 channel: one at the bundle crossing and one at the selectivity filter. A similar suggestion was recently proposed for the Kir1.1 channel in which the intracellular gate is regulated by internal pH and an extracellular gate at the selectivity filter is regulated by external K ϩ (44). Whether the two gates of the Kir3.1/Kir3.4 channel are coupled to one another or act independently would require further investigation. Another possibility is that a lateral force exerted by G␤␥ binding at the cytoplasmic domain could be transduced not only to the TM2 domains, causing them to rotate and translate (7,22), but also across the glycine hinge to induce subtle conformational changes within the selectivity filter. Many of the observations reported in this study support this latter hypothesis; both agonist activation and selectivity were altered by the replacement of phenylalanine residues at the bundle crossing, suggesting that changes at the bundle crossing can indeed alter the conformation of the selectivity filter. Also, disruption of the salt bridge that maintains the tension of the selectivity filter abolished agonist activation and selectivity, suggesting that relaxation of the filter renders the signal transduced from the cytoplasmic domains on binding of G␤␥ unnecessary. Introduction of a proline kink within the TM2 domain (by the double mutation Kir3.1[S170P]/Kir3.4[S176P]) abolished agonist activation, but did not affect selectivity; this double mutation perhaps had a more subtle effect on the selectivity filter. However, of course, G␤␥ binding presumably also opens the channel without causing a major change in selectivity. These data sug- gest, therefore, that transduction of the activating stimulus to the selectivity filter may explain why mutations at both intracellular and extracellular sides of the channel affect agonist activation and, also, the correlation between agonist activation and selectivity for K ϩ .