A State-independent Interaction between Ligand and a Conserved Arginine Residue in Cyclic Nucleotide-gated Channels Reveals a Functional Polarity of the Cyclic Nucleotide Binding Site*

Activation of cyclic nucleotide-gated channels is thought to involve two distinct steps: a recognition event in which a ligand binds to the channel and a conformational change that both opens the channel and increases the affinity of the channel for an agonist. Sequence similarity with the cyclic nucleotide-binding sites of cAMP- and cGMP-dependent protein kinases and the bacterial catabolite activating protein (CAP) suggests that the channel ligand binding site consists of a b -roll and three a -helices. Recent evidence has demonstrated that the third (or C) a -helix moves relative to the agonist upon channel activation, forming additional favorable contacts with the purine ring. Here we ask if channel activation also involves structural changes in the b -roll by investigating the contribution of a conserved arginine residue that, in CAP and the kinases, forms an important ionic interaction with the cyclized phosphate of the bound ligand. Mutations that conserve, neutralize, or reverse the charge on this arginine decreased the apparent affinity for ligand over four orders of magnitude but had little effect on the ability of bound ligand to open the channel. These data indicate that the cyclized phosphate of the nucleotide approaches to within 2–4 Å of the arginine, forming a favorable ionic bond that is largely unaltered upon activation. Thus, the binding site appears to be polarized into two distinct structural and functional domains: the b -roll stabilizes the ligand in a state-independent manner, whereas the C-helix selectively stabilizes the ligand in the open state of the channel. It is likely that these distinct contributions of

Activation of cyclic nucleotide-gated channels is thought to involve two distinct steps: a recognition event in which a ligand binds to the channel and a conformational change that both opens the channel and increases the affinity of the channel for an agonist. Sequence similarity with the cyclic nucleotide-binding sites of cAMP-and cGMP-dependent protein kinases and the bacterial catabolite activating protein (CAP) suggests that the channel ligand binding site consists of a ␤-roll and three ␣-helices. Recent evidence has demonstrated that the third (or C) ␣-helix moves relative to the agonist upon channel activation, forming additional favorable contacts with the purine ring. Here we ask if channel activation also involves structural changes in the ␤-roll by investigating the contribution of a conserved arginine residue that, in CAP and the kinases, forms an important ionic interaction with the cyclized phosphate of the bound ligand. Mutations that conserve, neutralize, or reverse the charge on this arginine decreased the apparent affinity for ligand over four orders of magnitude but had little effect on the ability of bound ligand to open the channel. These data indicate that the cyclized phosphate of the nucleotide approaches to within 2-4 Å of the arginine, forming a favorable ionic bond that is largely unaltered upon activation. Thus, the binding site appears to be polarized into two distinct structural and functional domains: the ␤-roll stabilizes the ligand in a state-independent manner, whereas the C-helix selectively stabilizes the ligand in the open state of the channel. It is likely that these distinct contributions of the nucleotide/C-helix and nucleotide/␤-roll interactions may also be a general feature of the mechanism of activation of other cyclic nucleotide-binding proteins.
Cyclic nucleotides regulate the activity of a diverse family of proteins involved in cellular signaling. These include a transcription factor (the bacterial catabolite activating protein, CAP), the cAMP-(PKA) 1 and cGMP-dependent protein kinases (PKG) and the cyclic nucleotide-gated (CNG) ion channels in-volved in visual and olfactory signal transduction (1,2). Despite obvious divergence among the effector domains of these proteins, the cyclic nucleotide binding (CNB) sites appear to share a common architecture. Solution of the crystal structures of CAP (3) and a recombinant bovine PKA RI␣ subunit (4) has demonstrated that their CNB sites are formed from an ␣-helix (A helix), an 8-stranded ␤-roll, and two more ␣-helices (B and C), with the C-helix forming the back of the binding pocket. Six residues are invariant among all members of the CAP and kinase families: three glycines involved in turns between strands of the ␤-roll, an arginine and a glutamate, each of which contact the cyclic nucleotide, and an alanine whose function is uncertain (1) (see also Fig. 1). Strikingly, these six residues are conserved in the CNG channels. Thus, it has been suggested that the invariant residues play important and conserved roles in the folding/function of the CNB sites of these diverse proteins (1)(2)(3)(4)(5)(6). Interestingly, only three of these residues (two glycines and the arginine) appear to be conserved among the more distantly related voltage-gated channels that bear the CNB site motif and whose gating may be modulated by direct binding of cyclic nucleotide (KAT1 (7,8), AKT1 (9,10), and dEAG (11,12) see Fig. 1).
Surprisingly, this structural similarity of the CNB site does not appear to be reflected in the conformation of the bound agonist. Thus, the crystal structures reveal cAMP binds in an anti conformation to CAP (3) but in a syn conformation to PKA RI␣ (4), although this may not reflect the conformation of the ligand bound to the proteins in solution (1,2,19). While experiments with cyclic nucleotide analogs and modeling, based upon the CAP and PKA R1␣ structures, have been used to investigate the conformation adopted by agonists in other CNB sites, this issue is unresolved (1)(2)(3)(4)(5)(6). This uncertainty, coupled with the lack of a crystal structure for any of the CNB proteins, in either the absence of bound agonist or presence of antagonist, leaves an important question unresolved: what are the structural changes that take place within these binding sites that result in the activation of each of the CNB proteins?
By employing site-directed mutagenesis and patch clamp recording of CNG ion channels, it is possible to separate the coupled processes of ligand binding from activation, permitting a dissection of the molecular contributions of protein-ligand interactions to each of these events. Such studies have demonstrated that residues within the C-helix selectively contribute to channel activation (20,21). Indeed, an aspartic acid residue (Asp 604 ) in the bovine rod subunit 1 (bRET1 (16)) C-helix appears to interact with the purine ring of cGMP selectively when the channel is open (21). That is, the binding energy of this interaction predominantly serves to stabilize ligand binding in the active conformation of the binding site, thereby leading to stabilization of the active (open) state of the channel. However, the state dependence of interactions between the cyclic nucle- Here we ask whether regions other than the C-helix of the CNB site are likely to be altered upon channel activation and thereby contribute to the increased affinity of the open channel for agonist. Studies of cyclic nucleotide analogs bearing sulfur substitutions on one or another of the exocyclic oxygens of the cyclized phosphate raise the possibility that residues in the ␤-roll may also contribute to activation gating. These data show that, in the kinases, the equatorial sulfur-substituted derivative (Rp-cAMPS) is an antagonist, whereas the axial sulfur-substituted compound (Sp-cAMPS) is an agonist (22)(23)(24), a profile that appears to be reversed in CAP (4,25). CNG channels formed from the catfish olfactory neuron subunit 1 (fOLF1 (17)) show an identical pharmacological profile to that of the kinases (26). By contrast, in CNG channels formed from bRET1, both cGMP derivatives are agonists and both cAMP derivatives are antagonists (26). Since the exocyclic oxygen atoms interact with residues in the ␤-roll, these data raise the possibility that large and possibly divergent structural changes may take place in the ␤-roll of each of the CNB proteins upon activation (1)(2)(3)(4)(5)(6).
We have focused upon the conserved arginine residue in the ␤-roll (Arg 559 in bRET1, Arg 529 in fOLF1, see Fig. 1). The homologous residue forms an ionic bond with the cyclized phosphate of the nucleotide in both CAP and the RI␣ subunit of PKA (1)(2)(3)(4), which suggests that this residue is well placed to detect any significant rearrangement between the ligand and the ␤-roll upon activation. We have previously reported that substitution of this conserved arginine with the polar but uncharged glutamine residue leads to a 27-fold increase in the K1 ⁄2 (agonist concentration producing half-maximal activation) in a chimeric channel (ROON-S2, see "Experimental Procedures" and Fig. 1). Despite this reduction in sensitivity to ligand, there is no apparent change in the ability of bound ligand to activate the ROON-S2 channel, as determined from the maximum open probability (P max ) of the channel in the presence of a saturating concentration of ligand (27). These data suggest that either the conserved arginine contacts the bound agonist in a state-independent manner (that is, it interacts equally well with ligand in the open and closed states of the channel) or that the polar glutamine residue is able to substitute effectively for the arginine to maintain any state-dependent contacts. Here we explore further the role of Arg 559 by studying a wide range of mutations that conserve, neutralize, or reverse the charge of this residue. Such mutations are tolerated and cause a progressive decrease in the affinity for agonist with little or no detectable change in the ability of bound agonist to activate the channel. These data are consistent with the formation of a state-independent, electrostatic interaction between this arginine and the cyclized phosphate of the ligand, although they also reveal an unexpected steric influence of chain length.

EXPERIMENTAL PROCEDURES
Molecular Biology-Point mutations were made by a polymerase chain reaction/subcloning strategy, and the resulting cDNA was verified by dideoxy chain termination sequencing of the polymerase chain reaction fragment (17,20). The amino acids swapped in the construction of the chimeras are given in the legend to Fig. 1. The majority of these experiments were performed in the background of two chimeric channels for technical reasons. The chimera RO133 is comprised of bRET1 whose pore-forming P-region has been replaced with the corresponding amino acids from fOLF1 (28). Channels formed from this chimera have cyclic nucleotide-gating properties identical to those of bRET1 but have the large single channel conductance of fOLF1 channels, facilitating measurements of single channel currents and, hence, P max (28). The other construct we utilized was a double chimera, ROON-S2 (27), in which we replaced both the P-region and aminoterminal N-S2 domain of bRET1 with sequences from fOLF1. This construct has both a large single channel conductance and a very high sensitivity to cGMP (due to the presence of the fOLF1 N-S2 domain and the bRET1 CNB domain (20,27)). The high apparent affinity of this parent chimera permitted us to study ligand-dependent gating of CNBsite point mutants whose apparent affinities for cGMP were shifted by up to 4 orders of magnitude. In the parent bRET1 and fOLF1 backgrounds, such mutations shift the dose-response curve of the channel into a cGMP concentration range that is greater than 100 mM and thus unmeasurable. We have previously shown that the only effect of introducing the fOLF1 N-S2 domain in the bRET1 background is to increase the efficacy with which bound ligands activate this construct; the selectivity of the bRET1 CNB site for ligand is not compromised (20,27). Throughout the text, the invariant arginine in ␤7 is identified according to the numbering sequence of either bRET1 (Arg 559 ) for those constructs that contain the bRET1 CNB site (bRET1, RO133, and ROON-S2) or of fOLF1 (Arg 529 ). In constructs where this residue is mutated, the identity of amino acids substituted for the arginine is shown by a letter after a slash mark in the construct name (single letter amino acid code); constructs with no slash mark and letter contain an arginine.
Electrophysiological Recordings-Inside-out patches were obtained from Xenopus oocytes 1-7 days after injection with cRNA (Message Machine, Ambion). In most experiments, recordings were performed with symmetrical solutions (67 mM KCl, 30 mM NaCl, 10 mM EGTA, 1 mM EDTA, 10 mM HEPES, pH adjusted to 7.2 with KOH). Na-cGMP was included in the intracellular solution by iso-osmolar replacement of NaCl. In some experiments, we completely replaced the KCl and NaCl with 100 mM Na-cGMP. Data were acquired using an Axopatch 200A patch clamp amplifier (Axon Instruments) and then digitized (Macintosh Centris 650 personal computer; ITC-16 interface and PULSE software, Instrutech Corp.) following low pass filtering (8 pole Bessel filter, Frequency Devices 902). Single channel recordings were filtered at 4 kHz and digitized at 20 kHz. Macroscopic currents were filtered at 1 kHz and digitized at 2 kHz. All data were acquired at a holding potential of Ϫ80 mV.
Data Analysis-K1 ⁄2 was estimated from fits to the Hill equation, where K1 ⁄2 is the apparent affinity, [A] is the agonist concentration, h is the Hill coefficient, and P open is the observed open probability at a given concentration of cGMP. For all constructs except ROON-S2/D, this was determined from patches containing many channels and calculated according to P open ϭ (I cGMP / I max )P max , where I cGMP is the macroscopic current at a given concentration of cGMP and I max is the maximal current at a saturating concentration of cGMP, measured in the same patch. As such macroscopic recordings were never obtained for ROON-S2/D, all open probabilities were determined from single channel recordings, as described below. For the following constructs, P max was determined from single channel patches in the presence of a saturating concentration of cGMP S1-S6 transmembrane segments. The seventh open box, in the carboxyl terminus, corresponds to the C-helix of the CNB site. The CNB site extends from the hashed line in the carboxyl terminus to the end of the C-helix box. In RO133, only the P domain of bRET1 (Ala 344 -Ala 378 ) is replaced by that of fOLF1 (Ser 314 -Phe 348 ). The position of Arg 559 is indicated. B, comparison of the amino acid sequences of homologous cyclic nucleotide binding sites from CAP (residues 9 -133) (13,14), bPKA R1␣ (residues 141-258, A site; 259 -379 B site) (15), bRET1 (residues 483-609) (16), fOLF1 (residues 453-579) (17), dEAG (residues 577-702) (11), and KAT1 (residues 383-510) (7). We have aligned all presently deposited CAP, PKA, and PKG sequences and confirmed that the six invariant residues identified by Shabb and Corbin (1) (marked by asterisks) are retained (not shown). Bars above the sequence indicate the positions of the ␣-helices and ␤-strands identified in the CAP crystal structure (3). The invariant arginine in ␤7 is identified throughout the text on the basis of the numbering sequence of bRET1 (Arg 559 ) or fOLF1 (Arg 529 ) (see "Experimental Procedures"). As these histograms included all open and closed events, the area of the closed peak represents the closed probability (P closed ) and P max is equal to 1 Ϫ P closed . However, for ROON-S2/L, ROON-S2/E, ROON-S2/D, and RO133/Q, 30 mM cGMP was not saturating. Higher concentrations of cGMP caused the maximal current to decrease, possibly due to desensitization. Accordingly for these four constructs, we first normalized the dose-response data by the open probability directly measured with 30 mM cGMP. P max was then obtained by fitting the Hill equation to the normalized data. This introduced only a minor correction for ROON-S2/L, ROON-S2/E, and RO133/Q. The correction was larger for ROON-S2/D, which had the most displaced dose-response curve. For ROON-S2/L and ROON-S2/D, this procedure can lead to P max values that are slightly larger than 1, reflecting the error inherent in this procedure given that the observed open probabilities are so close to 1 originally. Where appropriate, the values for P open (with 30 mM cGMP) are reported in the legends to Figs. 4 and 6, in addition to the estimated value of P max . This small error will not significantly affect our estimates of K1 ⁄2 . Throughout the manuscript, data are given as mean Ϯ S.E. or mean Ϯ range for those cases in which n ϭ 2. Fits are weighted to the reciprocal of the standard deviation of the mean data.
Determination of Electrostatic Distance-Our goal in these experiments is to dissect out the contribution of the conserved arginine in ␤7 to ligand binding and channel activation. Although K1 ⁄2 values depend, in general, on both ligand affinity and the coupled gating reaction, for those mutations that do not alter channel gating (P max ), changes in K1 ⁄2 must reflect a selective change in ligand affinity. Since the Arg 559 mutations studied here do not alter P max , we have used the observed changes in K1 ⁄2 with the various Arg 559 point mutants to calculate the change in free energy of the actual binding reactions. Thus, the change in free energy of binding, ⌬(⌬G), upon changing charge at Arg 559 is given by, where K 1 and K 2 are the K1 ⁄2 values for the wild-type and mutant channels, respectively. Assuming that the change in free energy reflects a simple coulombic interaction between the residue at position 559 and the cyclized phosphate of cGMP, it follows that, and from Equations 1 and 2, we obtain the electrical distance r, where R is the gas constant, T is the temperature, ⌬q 1 is the change in charge at position 559 between wild-type and mutant channels, q 2 is the charge on the cyclized phosphate of the ligand, N is Avogadro's number, and ⑀ and ⑀ 0 are the dielectric constant of the binding site environment and the vacuum permittivity, respectively. As this is a solvent-accessible part of the binding site, we assume that the charged groups are fully ionized and that the dielectric constant equals that of water. Fits of the Monod-Wyman-Changeux Gating Model-We have previously shown (20,27) that the simplest kinetic scheme that describes the equilibrium gating properties of CNG channels is the cyclical allosteric model of Monod, Wyman, and Changeux (18) (Fig. 1D). According to the model, the channel undergoes an allosteric transition between the In these fits, the only free parameter was K O . P max was constrained to the value determined from single channel recording (see above), and the number of ligand-binding sites was assumed to be four (29). K C was determined from the relation: for RO133 (7999) and fOLF1 (443) were constrained to the values previously determined from the unliganded open probability, P sp , measured for each of these channels (27). The values of L0 for RO133/Q and fOLF1/Q were assumed to be equal to those of the parent channel. This assumption seems reasonable since we have previously shown that mutation of Arg 559 to a glutamine had no effect on the value of L0 in the ROON-S2 background (27).  1 (top traces). These data thus suggest that neutralization and reversal of the charge at position 559 leads to a progressive decrease in the sensitivity of the channel to cGMP. A concern in all mutagenesis experiments is that the observed effects are due to a global disruption of the structure and function of the protein. As the ion conducting pore of the CNG channels is largely formed from the loop between the 5th and 6th transmembrane domains (28), with no detectable contribution from the carboxyl terminus, we determined the single channel conductance properties of each of these mutants. Despite the large change in ligand sensitivity, the representative single channel traces (Fig. 2) reveal that the current flow through the open channel for the two point mutants is indistinguishable from that of ROON-S2. The open states of all three channels are characterized by pronounced open channel noise, which is readily seen by comparison with the base-line noise when the channels are closed (lower traces of each pair). This excess noise is due to the rapid, partial block and unblock of the open channel by external protons (28,30).
The similarity of the open channel current properties among the constructs is confirmed from the all-points current amplitude histograms (Fig. 2, right panels). These To interpret the effect of these mutations quantitatively, we measured P open over a broad range of cGMP concentrations and fit the dose-response relationships by the Hill equation. As is seen in Fig. 3, the effect of these mutations was to cause essentially parallel shifts in the dose-response curves toward greater concentrations of ligand. Thus, the slope of the relationships and the P max values were largely unaltered while the K1 ⁄2 for activation of ROON-S2, ROON-S2/Q, and ROON-S2/E by cGMP increased from 1.8 Ϯ 0.3 M (n ϭ 10) to 50 Ϯ 8 M (n ϭ 8) and 3379 Ϯ 1005 M (n ϭ 5), respectively. That is, neutralization of Arg 559 resulted in a 28-fold increase in the K1 ⁄2 value, whereas charge reversal increased further the K1 ⁄2 value by 68-fold.
We next asked if the chemical identity of the residue at position 559 was important or if the altered activation of the mutant channels was simply a consequence of the change in charge on the side chain. To investigate this, we constructed a more extensive series of mutations in the ROON-S2 background, generating channels with basic (arginine or lysine), neutral (glutamine, asparagine, or leucine), or acidic (glutamate or aspartate) residues at position 559. The gating properties of each construct were then determined. Fig. 4A shows that P max for all of the constructs was Ն 0.98, indistinguishable from the parent chimera ROON-S2. Together, the data in Figs. 2, 3, and 4A show that neither the charge nor chemical identity of the side chain of residue 559 has a detectable influence upon the ability of bound ligand to open the channel.
In contrast, a plot of K1 ⁄2 versus charge on the side chain of residue 559 reveals that there are both electrostatic and steric effects of side chain substituents upon the apparent affinity for ligand (Fig. 4B). Thus, introducing the charge-conserving lysine (ROON-S2/K) residue resulted in a decrease in apparent affinity. Surprisingly, the 70-fold increase in K1 ⁄2 was larger than the 28-fold increase seen upon neutralization with glutamine. Lysine has two important differences when compared with arginine. First, it is the equivalent of one methylene bridge shorter, and second, it has a point charge on a primary amine, whereas arginine has the charge delocalized over the guanidinium group. As there are no other amino acids with basic side chains, it is not possible to distinguish between the steric effect of shortening the side chain from an effect of alteration in local field strength.
Mutation of Arg 559 to neutral and acidic amino acids does permit us to address this question further. Replacement of Arg 559 with an asparagine (which is one methylene bridge shorter than glutamine, but otherwise identical), to generate the ROON-S2/N mutant, gives rise to a far more pronounced increase in K1 ⁄2 (392-fold) than does replacement with glutamine (ROON-S2/Q). This result suggests that chain length or the exact location of the polar groups, in addition to charge, is an important determinant of ligand affinity. The importance of side chain polarity is demonstrated upon introduction of the non-polar residue leucine, which increased the K1 ⁄2 by 882-fold, a more pronounced modification than that seen with either of the polar substitutions or with lysine. Leucine is effectively an asparagine in which the carbonyl oxygen and amino group of the side chain have been replaced by methyl groups and which has a volume intermediate between that of glutamine and arginine.
The importance of charge at position 559 was further ex- plored by reversing the sign of the charge by introduction of either glutamic or aspartic acid. The K1 ⁄2 values of these two mutants was increased by 1877 and 5489 fold, respectively. The magnitudes of these increases in K1 ⁄2 are consistent with the generation of a repulsive interaction between the acidic side chain of the amino acid and the cyclized phosphate of the cyclic nucleotide. However, here again we see that amino acid residue with shorter side chain produced a more pronounced increase in K1 ⁄2 .
A linear regression through the plot of log(K1 ⁄2 ) values versus charge at position 559 (solid line in Fig. 4B) yields a slope corresponding to a 19.5-fold increase in K1 ⁄2 for an elementary change in charge (the mean value of K 2 /K 1 , Equations 1 and 3 under "Experimental Procedures"). Assuming a coulombic interaction between the residue at position 559 and a single negative charge on cGMP, this relationship yields an approximate distance of 2.4 Å between the ligand and the charge at position 559 (determined from Equation 3, under "Experimen-tal Procedures"). Approximate upper and lower bounds for this value are obtained from the largest and smallest changes in K1 ⁄2 observed upon reversal of charge. The 5489-fold increase in K1 ⁄2 upon replacing arginine by aspartate is equivalent to a distance of 1.7 Å, whereas the 27-fold increase in K1 ⁄2 upon replacing lysine by glutamate indicates a slightly longer distance of 4.4 Å. The electrostatic nature of this interaction is supported by the roughly similar fold increase in K1 ⁄2 seen upon changing the residue at position 559 either from a glutamine to a glutamate or from an asparagine to an aspartate. In each case, chain length is held essentially constant while a negative charge is introduced by conversion of the amide to the acid (Fig. 4B). Taken together, the data in Fig. 4 are consistent with the formation of a state-independent ionic bond between the side chain of residue 559 and the cyclized phosphate of the nucleotide.
Despite the pronounced effect of these point mutations on cGMP sensitivity, the conductance properties of the mutant channels are essentially identical to the parent chimera, ROON-S2. This is evident in Fig. 5, a two-dimensional plot of conductance versus fractional occupancy of the three open channel conductance states (unprotonated, singly and doubly protonated, see Fig. 2). The variability in the amplitude of the largest conductance state among the different mutants is not correlated with ionic charge at position 559. Rather, it is likely to reflect a technical difficulty in fitting this infrequently occupied conductance level next to the two dominant conductance levels, which represent the partially and fully protonated states. Taken together, the lack of an effect of the mutations upon either the single channel conductance or P max indicate that the mutations of Arg 559 result in a discrete disruption of the binding site that selectively lowers the apparent affinity for ligand.
As these experiments were performed in the background of ROON-S2, a chimeric channel with unusually high apparent affinity, we were concerned that the introduction of either the fOLF1 P-region or N-S2 domains may have altered the normal interaction between Arg 559 and the ligand. To address these concerns, we tested the effect of one of the Arg 559 point mutations in the backgrounds of both bRET1 and the chimera RO133 (bRET1 with the fOLF1 P-region). Fig. 6A shows that the mutation R559Q in the R0133 background increased the K1 ⁄2 of the resulting construct (RO133/Q) 42-fold with no change in P max . A similar decrease in the apparent affinity was observed upon introduction of the R559Q mutation in bRET1 (52 Ϯ 6 M, n ϭ 9 to 2793 Ϯ 368 M, n ϭ 2; data not shown). The qualitative and quantitative similarity of the R559Q mutation in bRET1, RO133, and ROON-S2 demonstrate that the selective change in apparent affinity upon mutation of Arg 559 is an intrinsic property of the bRET1 CNB site.
What is the basis for this selective increase in K1 ⁄2 ? To address this, we have utilized the cyclic allosteric model of Monod, Wyman, and Changeux (18) (Fig. 1D), which we have previously demonstrated to be the simplest kinetic scheme that adequately describes CNG channel activation (20,27). Based upon this model, an increase in K1 ⁄2 can be produced either by a reduction in affinity for ligand (increase in K O or K C ) or from an increase in the allosteric equilibrium constant, L0, between the open and closed state of the channel (L0 ϭ [C]/[O]). However, we have previously shown, using measurements of agonist-free openings, that the arginine to glutamine mutation does not alter L0 in ROON-S2 (27). Moreover, a change in L0 in any of the mutants would not only increase K1 ⁄2 but would also significantly reduce P max , which was not observed (Figs. 2A, 4A, and 6A). Rather, these data suggest that mutation of Arg 559 lowers the apparent affinity by specifically decreasing the absolute affinity of the CNB site for ligand.
To investigate the effect of the arginine to glutamine muta-tion quantitatively, we fit cGMP dose-response curves with the MWC model to determine K O and K C for RO133 and RO133/Q (Fig. 6A), constructs that allow us to determine P max with a high degree of accuracy. Plotting these equilibrium constants as free energy terms (⌬G ϭ ϪRT ln(1/K)) shows that the mutation reduces the absolute binding affinities of both the open (K O ) and closed (K C ) states of the channel (Fig. 7A). Indeed, the change in free energy of cGMP binding between the wild-type channel and the glutamine mutant (⌬(⌬G), filled circles in each plot) shows that K O and K C are destabilized equivalently, by 2.21 and 2.27 kcal mol Ϫ1 , respectively. These amounts are consistent with the disruption of an ionic bond. Is this state-independent interaction between the conserved arginine in ␤7 and the cyclic nucleotide a common characteristic of all CNB domains or do marked structural rearrangements occur around the homologous arginine in the ␤-roll of other binding domains? To address this, we have investigated the effects of mutation of the homologous arginine to a glutamine in fOLF1. This comparison is particularly interesting given the differential handling of the Rp-and Sp-cyclic nucleotide analogs by bRET1 and fOLF1 (26), which suggests that the ␤-roll portion of the binding site of fOLF1 may differ significantly from that of bRET1. Fig. 6B compares dose-response curves for fOLF1 and fOLF1/Q. Although there is a shift in the fOLF1/Q dose-response curve to higher concentrations, this effect is less marked (7.6-fold) than is seen in the bRET1 CNB site (28 -54-fold, depending upon the channel background). Surprisingly, the fOLF1/Q mutant shows a higher P max compared with wild-type fOLF1, despite the decrease in cGMP sensitivity.
These data raise two questions. First, to what extent do these differences in gating properties between bRET1 and fOLF1 result from a fundamental difference in the mechanistic behavior of their binding sites? Second, how can a mutation destabi- lize binding but increase efficacy? To investigate these questions, we fit the dose-response data for fOLF1 and fOLF1/Q with the MWC model and determined K O and K C for these two channels. This analysis reveals that the impact of the R to Q mutation on ligand binding in fOLF1 is, in fact, very similar to the effect observed in the bRET1 CNB site. Thus, the destabilization of K O (1.57 kcal mol Ϫ1 ) and K C (1.90 kcal mol Ϫ1 ) in fOLF1/Q are similar in sign and magnitude to the changes seen in the bRET1 background. The less marked shift in the doseresponse curve and the increase in P max seen with the arginine to glutamine mutation in fOLF1 arise from small quantitative differences in the magnitude of the effect of the mutation upon binding of agonist to the open and the closed states of the channel, not from a qualitatively different utilization of the binding energy. DISCUSSION Here we have investigated which regions of CNB sites contribute to activation and, in particular, whether there is likely to be a significant change in the interaction between the ␤-roll of the CNB site and the ligand upon activation. Our studies focused on an arginine residue in the ␤-roll that is conserved among diverse CNB proteins and that makes an ionic interaction with the cyclized phosphate of cyclic nucleotides in both the bacterial CNB protein CAP as well as in the regulatory subunit of cAMP-dependent protein kinase (1-4) (see also Fig. 1).
Mutations of this conserved arginine, in the background of the chimeric CNG channel ROON-S2, to a series of residues that conserve, neutralize, or reverse its charge, caused a progressive decrease in apparent affinity of the channel for ligand. Although an unexplained steric effect of chain length contributed to this decrease, the clear dependence of the K1 ⁄2 values on charge at position 559 strongly supports the formation of an ion pair between Arg 559 and the cyclized phosphate. This result is consistent with the x-ray crystallographic structural studies of PKA RI␣ and CAP (1)(2)(3)(4). Indeed, the estimate of the electrostatic distance between Arg 559 and the cyclized phosphate from these experiments (1.7-4.4 Å) is close to that predicted from the crystal structures of CAP (3.1-3.5 Å, (3)) and PKA RI␣ (Ͻ3.3 Å, (4)). Despite the large changes in ligand sensitivity with the Arg 559 mutants (spanning nearly four orders of magnitude), the ability of the bound ligand to activate the channel (as determined from P max ) was virtually unaltered. These data suggest that Arg 559 plays an important role in stabilizing cyclic nucleotide and that these interactions do not contribute to channel activation. The absence of an effect of the mutations on the single channel conductance or on P max shows that these mutations are unlikely to cause a global disruption of the protein.
This surprising result, that the Arg 559 point mutants have large effects on ligand sensitivity but little effect on activation gating, can be readily explained within the context of the MWC allosteric reaction scheme (18). According to this scheme, a concerted allosteric conformational change in the channel both opens the channel pore and alters the binding site, causing the ligand affinity of the open state to be considerably higher than the ligand affinity of the closed state (dissociation constants K O and K C , respectively). By measuring ligand-independent openings, we previously determined the allosteric equilibrium constant between closed and open channels in the absence of agonist, L0, for both bRET1 and fOLF1 (27). We found that a 20-fold difference in L0 between bRET1 and fOLF1, which contributes to physiologically important differences in ligand gating (20,27,31), was localized to the amino-terminal N-S2 domain (27). Since this region of the channel interacts with the carboxyl terminus (32,33) and is involved in subunit assembly in the homologous voltage-gated K channels (34 -38), we have postulated that channel activation involves a change in quaternary structure.
Whereas the difference in the allosteric transition between fOLF1 and bRET1 is localized to the amino terminus of the channel, the postulated increase in ligand affinity of the open state of these channels is mediated, at least in part, by interactions of the cyclic nucleotide with the C-helix of the carboxyl terminus CNB domain (20,21). In particular, an aspartate residue in the C-helix of bRET1, Asp 604 , has been shown to make important contacts with cGMP in the open state, but not closed state, of the channel (21). These results suggested a model of channel gating in which the allosteric transition that opens the channel is coupled to a change in the orientation of the C-helix relative to the ␤ roll, leading to an enhancement of C-helix/ligand contacts. According to this model, the ␤-roll would provide a relatively stable structure that is involved in the initial binding of ligand, which orients the nucleotide within the binding pocket. The lack of effect of mutation of Arg 559 on ligand-dependent gating is consistent with this hypothesis.
A quantitative analysis of the effect of mutating arginine 559 to glutamine was performed by fitting the MWC model to the cGMP dose-response data. This was done in the background of a chimeric channel, RO133 (bRET1 with the fOLF1 P region), because the gating properties and large single channel conductance of this construct facilitated accurate determination of P max , and hence, the channel activation parameters (28). This analysis shows that the R559Q mutation decreases the affinity of the open (K O ) and closed (K C ) state of the channel for ligand by an identical amount. From these data we can conclude that there is no significant structural rearrangements between this deep part of the ␤-roll and the ligand upon channel activation. Conversely, we can also conclude that all bonds between the protein and the ligand that are made more favorable when the channel goes from the closed to the open state, and stabilize the latter, are unaffected by the electrostatic and steric effects of substitutions at position 559.
Does this analysis of the interaction between Arg 559 in bRET1 and the cyclic phosphate hold true for other cyclic nucleotide binding pockets? Although mutation of the conserved arginine in CAP (to lysine, histidine, glutamine, or leucine) and PKA (to either lysine or tryptophan) has been shown to interfere with ligand-dependent activation, it has been difficult in these molecules to separate out effects of binding from activation (1, 25, 39 -43). To address this question, we therefore constructed the homologous mutation in the fOLF1 CNB domain. This is particularly interesting given the different actions of Rp-and Sp-substituted ligands in fOLF1 and bRET1 (26). In the background of the olfactory channel, mutation of the homologous arginine (Arg 529 ) actually enhanced P max despite a decrease in ligand sensitivity. This result suggested that there might be a qualitatively different interaction between cGMP and the ␤-roll of the fOLF1 binding site compared with the cGMP/bRET1 ␤-roll interaction. However, a fit of the MWC model showed that these differences can be explained by relatively small quantitative changes, amounting to only an ϳ0.3 kcal mol Ϫ1 difference between the effects of the R529Q substitution on K O and K C , in which closed state binding is decreased to a slightly greater extent than open state binding. Such small changes (equivalent to a fraction of a hydrogen bond) may readily be explained by indirect effects of the R529Q mutation on the orientation of the bound ligand rather than a large scale change in the structure of the fOLF1 ␤-roll during channel activation.
The state-independent interaction with the conserved argi-nine in ␤7 in the CNG channels is in contrast to results suggesting that the neighboring residue, Thr 560 in bRET1, may contribute to activation gating (21,44). Thus, the mutation T560A produces a somewhat greater decrease in binding to the open state compared with the closed state, resulting in a 6 -7fold decrease in P max . Although this suggests that there might be a state-dependent interaction between Thr 560 and ligand, this effect on gating could also be due to an indirect effect of the mutation, either by altering the conformation of the binding pocket or the orientation of the ligand in the binding site. For example, the T560A mutation might slightly decrease the ability of bound ligand to form optimal contacts with the C-helix in the open state. Although there are many possible interpretations for mutations that alter gating, only one interpretation is consistent with the profound state-independent changes in ligand binding seen with the wide range of Arg 559 mutations, that this region of the channel does not alter its contacts with ligand during gating. The data presented here, taken together with previous results, suggest that the CNB site of both fOLF1 and bRET1 CNG channels comprises two distinct structural and functional domains. The ␤-roll forms state-independent contacts with ligand that are important for stabilizing the ligand in the binding pocket, whereas the C-helix makes state-dependent contacts that increase ligand affinity upon channel activation and stabilize the channel in its open state (1-4, 20, 21). Based on the qualitatively similar effects of the mutation in bRET1 (R559Q) and fOLF1 (R529Q), our data suggest that these proteins undergo a common structural change upon activation despite their different patterns of activation with Rp-and Spcyclic nucleotide analogs (26). The distinct pharmacology of these two proteins probably reflects relatively small variations in the conformation of the binding pockets or the bound ligand rather than qualitatively different mechanisms of activation. Given the sequence similarity among CAP, the kinases, and the CNG channels and the similar effects seen in each upon mutation of the conserved arginine residue, we expect that the CNB sites of these diverse proteins share a similar functional organization that underlies the mechanism of ligand activation.