A Secondary Structural Transition in the C-helix Promotes Gating of Cyclic Nucleotide-regulated Ion Channels*

Background: Cyclic nucleotide-regulated channels are involved in sensory transduction and repetitive electrical activity. Results: Transition metal ion FRET and electrophysiology demonstrated a coil-to-helix transition within the ligand binding domain. Conclusion: Agonist binding stabilized the structure of the C-helix, triggering channel gating. Significance: This transition is important for ion channel activation and may be necessary for the activation of other cyclic nucleotide-regulated proteins. Cyclic nucleotide-regulated ion channels bind second messengers like cAMP to a C-terminal domain, consisting of a β-roll, followed by two α-helices (B- and C-helices). We monitored the cAMP-dependent changes in the structure of the C-helix of a C-terminal fragment of HCN2 channels using transition metal ion FRET between fluorophores on the C-helix and metal ions bound between histidine pairs on the same helix. cAMP induced a change in the dimensions of the C-helix and an increase in the metal binding affinity of the histidine pair. cAMP also caused an increase in the distance between a fluorophore on the C-helix and metal ions bound to the B-helix. Stabilizing the C-helix of intact CNGA1 channels by metal binding to a pair of histidines promoted channel opening. These data suggest that ordering of the C-helix is part of the gating conformational change in cyclic nucleotide-regulated channels.

Cyclic nucleotide-gated (CNG) 2 and hyperpolarization-activated cyclic nucleotide-modulated (HCN) channels are members of the superfamily of voltage-activated ion channels (1,2). CNG and HCN channels are unique in the voltage-gated channel superfamily in that their gating from closed to open is promoted by direct binding of the cyclic nucleotides cAMP and cGMP. CNG channels are responsible for the primary electrical signals in vision and olfaction (1). HCN channels contribute to pacemaking in the heart as well as repetitive firing, resting membrane potential, and dendritic integration in some neurons (1,3).
Cyclic nucleotides bind to a conserved domain in the cytoplasmic C-terminal region of each subunit (1). Several structures of the C-terminal region in complex with cyclic nucleotide have been solved for eukaryotic HCN channels (4 -8). The structures contain a novel helical domain called the C-linker, which is attached directly to the ion channel pore in intact channels and is the site of tetramerization of the cytoplasmic domain ( Fig. 1, A and B). Immediately following the C-linker is a cyclic nucleotide-binding domain (CNBD) that is very similar to the binding domains of the bacterial MloK1 potassium channel, the Escherichia coli transcription factor CAP, the guanine nucleotide exchange factor Epac, and the regulatory domains of the kinases PKA and PKG (9 -15). The core of the CNBD consists of an eight-stranded ␤-roll followed by two ␣-helices, the B-and C-helices (Fig. 1B). Extensive contacts are made between the cyclic phosphate and ribose moieties of the cyclic nucleotide and residues in the ␤-roll and a short helix between the sixth and seventh ␤-strands called the P-helix. The C-helix caps the ligand-binding site by forming interactions with the purine ring of cAMP or cGMP.
A mechanism for the activation of eukaryotic cyclic nucleotideregulated channels has been proposed based on electrophysiological and biochemical studies (1). In this model, initial contact is made between the ligand and the ␤-roll/P-helix. Mutations in this region of the channel have effects on cyclic nucleotide affinity, with little effect on the efficacy of the ligand to promote channel opening (4, 16 -19). Subsequent to ligand binding, the C-helix is thought to translate toward the binding pocket and form contacts with the ligand. Mutations in the C-helix have effects on the ability of ligand to promote channel activation, suggesting that C-helix movement is coupled to the opening conformational change of the channel pore (4, 16 -21).
Direct structural evidence for this hypothesis in eukaryotic channels is limited. The ligand-free (apo) structure of the mouse HCN2 C-terminal region has been solved (6). Overall, the apo-structure is very similar to the ligand-bound structure, but it differs in two key regions. The FЈ-helix of the C-linker adopts a nonhelical coil structure in the absence of agonist. Although the position of the proximal C-helix is quite similar between the apo-and ligand-bound HCN2 structures, the distal half of the C-helix is not resolved in the apo-structure. Furthermore, circular dichroism spectra reveal a marked increase in the helical content of the HCN2 C terminus in the presence of cAMP (22). These results raise the intriguing possibility that the C-helix becomes more ordered upon cyclic nucleotide binding and that this secondary structural transition may be coupled to the gating conformational change that opens the channel.
In this study, we used transition metal ion fluorescence resonance energy transfer (tmFRET) and electrophysiology to investigate the hypothesis that cyclic nucleotide binding stabilizes the secondary structure of the C-helix. tmFRET is a FRET technique that uses small, covalently attached fluorophores as donor molecules and colored transition metal ions (e.g. Ni 2ϩ and Co 2ϩ ) bound to pairs of histidine residues as acceptors (6,(23)(24)(25). This method can report very short distances (8 -20 Å), virtually eliminates the orientation dependence of classical FRET, and allows the experimenter to tune the dynamic range of distance measurements by changing the acceptor ion. Previously, a comparison of tmFRET measurements in the absence and presence of cAMP demonstrated that the C-helix undergoes a translation toward the ␤-roll in the presence of cyclic nucleotide (6). Furthermore, the transition metal ion Ni 2ϩ bound more tightly to pairs of histidines in the C-helix spaced four residues apart in the presence of cAMP than in its absence, suggesting cAMP causes a stabilization of the secondary structure of the C-helix (6). In this study, we used tmFRET within the C-helix and between the B-and C-helices of the purified HCN2 C-terminal domain to demonstrate further that the C-helix undergoes a secondary structural transition, becoming more ordered upon binding cAMP. Using patch clamp electrophysiology on intact, functioning ion channels, we show that this secondary structural transition is coupled to channel opening and is therefore part of the gating conformational change in cyclic nucleotide-regulated channels that translates ligand binding into channel opening.

Molecular Biology
The gene encoding the cysteine-free C-terminal fragment (amino acids 443-640) of the mouse HCN2 ion channel (HCN2) was synthesized by Blue Heron Biotechnology (Bothell, WA) and cloned into the pMALc2T vector (New England Biolabs, Ipswich, MA) (6). The vector encodes an N-terminal maltose-binding protein tag that is separated from the gene product of interest by a thrombin-cleavable linker. CNGA1 Q was subcloned into the pGEMHE vector (26). This construct contains the full-length CNGA1 channel with all of the native cysteines mutated, with the exception of Cys-481 and the native Ni 2ϩ potentiation site mutated to a glutamine (H420Q) (16,27). All mutations were created using standard PCR techniques and verified using automated sequencing (Genewiz, Seattle, WA). To make RNA for Xenopus oocyte expression experiments, cDNAs encoding CNGA1 Q constructs were linearized with NheI (New England Biolabs) and transcribed in vitro using the mMessage mMachine Ultra kit with T7 polymerase (Ambion, Austin, TX).

Protein Purification
Soluble HCN2 constructs were prepared as described previously (28). Briefly, all constructs were expressed in BL21 (DE3) bacteria (New England Biolabs), grown in 2-liter cultures, and induced with 1 mM isopropyl thio-␤-D-galactopyranoside. Bacteria were lysed and centrifuged to remove insoluble cellular material. MBP-HCN2 constructs were purified on an amylose affinity column. The maltose-binding protein tag was removed by thrombin digest followed by ion exchange chromatography.

Protein Labeling
Labeling of cysteine residues inserted into the HCN2 construct via mutagenesis was carried out during purification after the thrombin digest but prior to ion exchange chromatography. Protein was eluted from the amylose affinity column into ϳ20 ml of buffer containing 150 mM KCl, 50 mM D(ϩ)-maltose monohydrate, 10% glycerol, and 30 mM HEPES, pH 7.2. Fluorophores (monobromobimane or dibromobimane) were added to the buffer at a final concentration of 100 M from a 100 mM stock in DMSO, which was stored at Ϫ80°C. At this concentration, the fluorophore was in molar excess of the cysteines intended for labeling. The solution was placed on an orbital shaker (60 -80 rpm) for 1 h at 20 -25°C. Unincorporated fluorophore was removed during the ion exchange step. Labeling efficiency was not quantitated for the subsequent experiments as any unlabeled protein would not be expected to contribute to the fluorescence signal.

UV-Visible Spectroscopy
Absorbance spectra of 1 mM Ni 2ϩ and Co 2ϩ bound to the peptide ACAAKAAAKHAAAHKA (3 mM) were acquired in a Beckman DU460 spectrophotometer.

Fluorometry
Fluorescence spectra were acquired using a Spex Fluorolog-3 spectrofluorometer outfitted with a MicroMax plate reader (HORIBA Jobin Yvon, Edison, NJ). The excitation wavelength was 390 nm, and emission was measured between 440 and 600 nm in 1-nm increments with a 0.1-s integration time and slit widths set between 5 and 14 nm on the excitation and emission monochromators. For FRET experiments, peak emission was averaged over a 7-nm window, usually between 477 and 483 nm. Bimane-labeled HCN2 constructs were diluted to a concentration between 30 nM and 6 M. Experiments were conducted in a buffer containing 150 -300 mM KCl, 10% glycerol, and 30 mM HEPES, pH 7.2, at 20°Ϫ25°C. Data were acquired from 250-l samples in black, flat-bottom 96-well plates (Corning; Lowell, MA) through fiber optics in the plate reader connected to the sample compartment of the fluorometer. To reduce the amount of scattered excitation light collected by the emission fiber optic, a 425-nm long pass filter was placed in line with the emitted light path. To each sample was added 7.5 l of water or cAMP (sodium salt, monohydrate) diluted to the desired concentration from a 10 mM aqueous stock. Unless otherwise noted, the final cAMP concentration was 300 M. 2.5 l of water or aqueous stocks of NiCl 2 ⅐6H 2 O or CoSO 4 ⅐7H 2 O were also added to each well.
Fluorescence anisotropy measurements were conducted using 200-l samples in a standard quartz cuvette. The Fluorolog-3 was outfitted with Glan-Thompson polarizers. Data were acquired in the L format with excitation set to 390 nm and emission recorded at 480 nm with 5-nm slit widths. Anisotropy was calculated as described elsewhere (29).
Electrophysiology cRNAs encoding CNGA1 Q constructs were expressed in Xenopus laevis oocytes prepared as described previously (30).
Recordings were carried out in the inside-out patch clamp configuration with an Axopatch 200A amplifier (Molecular Devices, Sunnyvale, CA) and digitized with an ITC-16 analog/digital converter (Instrutech/HEKA, Bellmore, NY). Data were low pass filtered at 2 kHz and sampled at 10 kHz. All data were acquired using Pulse software (HEKA). Both pipette and bath solutions contained 130 mM NaCl, 3 mM HEPES, pH 7.2, with N-methyl-D-glucamine. Bath solutions were supplemented with cGMP or inosine 3Ј,5Ј-cyclic monophosphate sodium salt (cIMP) and NiCl 2 ⅐6H 2 O as indicated. Pipettes were pulled from borosilicate glass and fire polished to a resistance of 0.2-0.6 megohms. Rapid solution exchange was achieved using an RSC-100 solution changer (Biologic, Claix, France). All recordings were performed at room temperature (20 -25°C). CNGA1 currents exhibit a time-dependent run-up following patch excision (31). To account for this phenomenon, the patches were repetitively pulsed in the presence of 200 M cGMP until the current magnitude stabilized, typically 15-20 min.

Data Analysis
Fluorescence-All data were blank-corrected by subtracting spectra recorded from solutions with the same components but no fluorescent protein. Because we used a plate reader for our fluorescence measurements of HCN2 constructs, it was impossible to determine the exact path length of the excitation and emission light and therefore not possible to mathematically correct for the inner filter effect of the absorbance of ions and fluorophore in our samples. Rather, we corrected for this experimentally by dividing normalized (to no metal Ϯ 300 M cAMP) fluorescence data from constructs containing bimane-conjugated cysteines and introduced di-histidine metal-binding sites by normalized fluorescence data from constructs containing the same fluorophore configuration but no metal-binding site. Not only did this partially account for the inner filter effect, but it eliminated the low affinity quenching component observed in proteins lacking an exogenously introduced metal-binding site, which is likely the result of quenching by ions in solution.
Corrected fluorescence data were plotted as a function of total metal (Ni 2ϩ or Co 2ϩ ) concentration and fit with a single binding site model as shown in Equation 1, where F/F no site is corrected fluorescence; [M 2ϩ ] is the total transition metal ion concentration; [protein] is the total concentration of the HCN2 protein being used; K d is the apparent metal binding affinity, and E FRET is the FRET efficiency, i.e. the amount of quenching at saturating metal concentrations.
The cAMP concentration dependence of tmFRET was fit with Equation 2, where [cAMP] is the total cAMP concentration; F 0 is the fluorescence in the absence of cAMP; F cAMP is the fluorescence in the presence of saturating cAMP, and all other terms are as defined above. Distances between FRET donors and acceptors were calculated using the Förster (Equation 3), where R is the distance between the donor and acceptor; R 0 is the distance at which FRET efficiency is 0.5, and E FRET is the FRET efficiency. R 0 for each FRET pair was calculated as described previously (6,29,32).
Electrophysiology-Current records were leak-and capacitance-corrected by subtracting off currents from the same patch in a solution with no cyclic nucleotide. Current magnitudes reported were the mean currents from steps to Ϫ60 mV for cyclic nucleotide concentration-response relationships and steps to Ϫ100 mV for Ni 2ϩ potentiation experiments. Currents in every patch were normalized to the current magnitude at 10 mM cGMP.
Cyclic nucleotide concentration-response relationships were fit with the Hill Equation 4, where I max, cGMP is the maximal current response in 10 mM cGMP; A cNMP is the fractional activation in cNMP; [cNMP] is the free cyclic nucleotide concentration; K d is the apparent affinity for cyclic nucleotide, and n is the Hill coefficient. Ni 2ϩ concentration-response relationships were determined in the presence of 16 mM cIMP. Ni 2ϩ concentration-response relationships for the CNGA1 Q construct, which exhibited block, but no potentiation, were fit with a single binding site model of Equation 5, where K b is the apparent affinity for Ni 2ϩ block. Ni 2ϩ concentration-response data for the CNGA1 Q E595H,Q599H construct were fit with Equation 6, which describes two independent components, Ni 2ϩ potentiation and Ni 2 block, where P is the fractional potentiation by Ni 2ϩ , and K p is the apparent affinity for Ni 2ϩ potentiation. Because the E595H, Q599H mutation should not affect the Ni 2ϩ block of the channel pore, K b was fixed to the value determined from the CNGA1 Q fit to Equation 5, reducing the fit of Equation 6 to only two free parameters.
To determine the ⌬⌬G for stabilization of the open state of CNGA1 Q E595H,Q599H by Ni 2ϩ binding to the C-helix, we first had to calculate L, the equilibrium constant for channel opening. Because the Ni 2ϩ potentiation experiments were conducted at saturating concentrations of the partial agonist cIMP, the gating scheme was simply represented as shown in Equation 7, In this scheme, L for each agonist is defined by Equation 8, where P open is the probability of channel opening, and I max is the current at maximal open probability. In wild type CNGA1 channels, maximal current is measured in saturating cGMP and 1 M Ni 2ϩ , where P open has been directly measured to be near unity (33). In the CNGA1 Q constructs, this determination was not possible because the native Ni 2ϩ potentiation site was mutated (H420Q). To determine I max , we assumed that the relative efficacy of cGMP and cIMP was preserved between wild type and our mutant constructs. This is a reasonable assumption as we did not mutate any of the amino acids that were expected to contact the ligand. We then used Equation 9 to calculate I max , where L cIMP /L cGMP was set equal to 0.06, based on the value reported in the literature (18). Equation 9 was rearranged to make Equation 10, Our experimentally determined values of I cGMP and I cIMP were then used to calculate I max and L for CNGA1 Q E595H,Q599H with and without Ni 2ϩ . ⌬⌬G was then calculated using Equation 11, where R is the gas constant, and T is temperature in Kelvin.

Statistics and Fitting
Data were analyzed and fit using Origin 8.0 (OriginLab Corp., Northampton, MA) and Microsoft Excel (Microsoft Corp., Redmond, WA). Data were reported as the mean Ϯ S.E. For fluorescence anisotropy measurements, one-way analysis of variance was used to determine statistical significance with the threshold set to p Յ 0.05 after determining that the data were normally distributed (Fig. 3D). For comparison of distances in the presence and absence of cAMP, we used a paired samples t test with a significance threshold of p Յ 0.05 (Fig. 4B).

Chemicals and Reagents
All chemicals were purchased from Sigma except for monobromobimane and dibromobimane, which were obtained from Invitrogen.

RESULTS
tmFRET between Donors and Acceptors in the C-helix of Soluble HCN2-We previously suggested that the C-helix of the CNBD in HCN2 channels undergoes a coil-helix transition upon binding of cAMP (6). As a more stringent test of this hypothesis, we sought to measure distances between locations within the C-helix with and without cAMP. If there is a change in the secondary structure of the C-helix rather than just a rigid body movement induced by agonist binding, then there should be a change in dimensions of the helix. To this end, we engineered cysteines into the C-helix of an otherwise cysteine-free, soluble HCN2 C-terminal construct (which we will refer to here as HCN2), and we modified them with a cysteine-reactive fluorophore. We chose monobromobimane (mBBr) as our donor because of its small size and short linker ( Fig. 2A). mBBr reacts via nucleophilic substitution with cysteine, which forms a covalent bond between the methylene carbon on bimane and the sulfur of cysteine, with Br Ϫ as the leaving group. We inserted histidines, spaced at i and i ϩ 4 into the same helix to serve as metal-binding sites. The emission spectrum of bimane overlaps with the absorbance spectra of the transition metals Ni 2ϩ and Co 2ϩ when they are coordinated by histidine pairs in a helical peptide, making them suitable FRET acceptors (Fig. 2B). The calculated R 0 , the distance at which FRET efficiency is halfmaximal, is 9.2 Å for Ni 2ϩ with bimane and 11.7 Å for Co 2ϩ allowing us to measure short intrahelical distances over two different dynamic ranges (Fig. 2C).
We first sought to confirm that we were able to measure distances within an ␣-helix of a protein. We expressed and purified mutants of the soluble HCN2 C-terminal construct with a single cysteine on the N-terminal side of the C-helix labeled with bimane and metal-binding sites positioned one (A624C mB ,T627H,D631H) or two (A624C mB ,I630H,D634H) helical turns away from the donor fluorophore ( Fig. 2A). At the concentrations used in our experiments, these constructs are monomeric (8). As a control, we expressed a construct with the cysteine, but no exogenous metal-binding sites (A624C mB ). Fig.  2D (left panel) shows the emission spectra of the control construct with increasing concentrations of Co 2ϩ in the presence of 300 M cAMP. Some quenching of the bimane emission occurred at high Co 2ϩ concentrations (Ͼ ϳ1 mM). The apparent K d for quenching was 23 mM. This reduction in fluorescence was likely the result of quenching by ions in solution rather than specific binding. In contrast, the construct with a metal-binding site positioned one helical turn away from the fluorescent donor (A624C mB ,T627H,D631H) showed a much greater amount of quenching and quenched at a much lower Co 2ϩ concentration (Fig. 2D, right panel).
This quenching results from FRET between the bimane and the bound Co 2ϩ ion.
The concentration dependence for Ni 2ϩ and Co 2ϩ quenching of A624C mB ,T627H,D631H and A624C mB ,I630H,D634H in the presence of 300 M cAMP is shown in Fig. 2E. All data were normalized to quenching in the control construct (A624C mB ), which has no introduced metal-binding sites, and fit with a single binding site model (see under "Experimental Procedures"). The fits to the data produced two parameters, the FRET efficiency (the fraction of quenching at high metal concentration) and the apparent metal binding affinity. With both Ni 2ϩ and Co 2ϩ , there was a greater FRET efficiency (more quenching) for bimane and the closer metal-binding site (A624C mB ,T627H,D631H) than the farther site (A624C mB , I630H,D634H) as expected from the distance dependence of FRET (Figs. 2E and 3A). Also, there was a greater FRET efficiency for both constructs with Co 2ϩ than with Ni 2ϩ , consistent with the calculated R 0 values, although Ni 2ϩ bound with a higher apparent affinity (Fig. 2E). Distances determined from FRET measurements are the most accurate and precise when they are near R 0 , where the changes in FRET are largest per unit of distance. This underscores an important feature of tmFRET; the metal ion can be chosen to suit the desired distance range. The distances measured here are closer to the R 0 of Co 2ϩ than Ni 2ϩ . Therefore, although we have obtained similar results with Ni 2ϩ , we will use Co 2ϩ as the acceptor for the remainder of these tmFRET experiments.
The Förster equation was used to convert the measured FRET efficiencies into distances (Fig. 3A). Because bimane covalently bound to a cysteine has a number of possible rotamers, it is difficult to predict the absolute distances measured from tmFRET from the crystal structure of HCN2. Therefore, we calculated the difference between the distances to the two metal-binding sites (⌬ distance), making the assumption that bimane should adopt the same set of conformations for the constructs with the two different metal-binding sites (Fig. 3B). This distance should approximate the distance between the two sites because the bimane and metal sites are all predicted to lie on the same face of the ␣-helix. The ⌬ distance determined from the tmFRET measurements was 3.6 Ϯ 0.2 Å, within 1 Å of the 4.5 Å distance predicted based on an average spacing of 1.5 Å per residue of an ␣-helix.
To extend our distance measurements, we repeated the above experiment using the bBBr, which has two reactive moieties and can react with neighboring cysteines or cysteines placed three residues apart on an ␣-helix (28). By using bifunctional bimane, we could lock the fluorophore in two different orientations relative to the helix and greatly decrease the mobility of the fluorophore (Fig. 3C). We expressed and purified HCN2 C-terminal constructs containing the same metal-binding sites as before, but with pairs of cysteines reacted with bifunctional bimane, either at adjacent positions (R623C bB A624C) or three residues apart (M621C bB A624C) on the C-helix (Fig. 3C). The emission spectrum of bifunctional bimane conjugated to cysteine pairs in these constructs was indistinguishable from that of bimane attached at A624C. In all of these constructs, the closest point of attachment of the bifunctional bimane relative to the metal ion-binding site was the same as for bimane (A624C).
To show that the bifunctional bimane reacted with both cysteines and decreased its mobility, we measured fluorescence anisotropy (Fig. 3D). As expected, attaching bimane to a single cysteine on the C-helix of HCN2 (A624C mB ,T627H,D631H) greatly slowed its tumbling rate, resulting in a much higher anisotropy compared with mBBr in water (a change from 0.008 Ϯ 0.001 to 0.117 Ϯ 0.001). Covalently attaching bimane on both sides of the molecule by reacting bifunctional bimane with cysteine pairs further reduced its mobility. The anisotropy of bifunctional bimane attached to cysteines spaced three residues apart (M621C bB A624C,T627H,D631H) was 0.174 Ϯ 0.001, and the anisotropy of bifunctional bimane attached to

. tmFRET between bimane and bound transition metal ions in the C-helix of the HCN2 C-terminal domain.
A, schematics of the C-helix with a cysteine at position 624 conjugated to bimane and two di-histidine metal-binding sites at T627H,D631H and I630H,D634H. B, emission spectrum of bimane conjugated to HCN2 A624C (gray) overlaid with the absorbance spectra of Ni 2ϩ (green) and Co 2ϩ (pink) bound to a helical peptide with the sequence ACAAKAAAKHAAAHKA. C, calculated plots of the distance dependence of FRET between cysteine-conjugated bimane and Ni 2ϩ and Co 2ϩ coordinated by histidines four residues apart. The R 0 value (distance between donor and acceptor at which FRET efficiency is 0.5) is 9.2 Å for Ni 2ϩ and 11.7 Å for Co 2ϩ . Plots were calculated from the Fö rster equation. D, fluorescence emission spectra (in arbitrary units, A.U.) of HCN2 A624C mB (left panel) and HCN2 A624C mB ,T627H,D631H (right panel) in 300 M cAMP and increasing concentrations of Co 2ϩ . E, plots of the metal concentration dependence of tmFRET for Ni 2ϩ (left panel) and Co 2ϩ (right panel) for HCN2 A624C mB ,T627H,D631H (circles) and HCN2 A624C mB ,I630H,D634H (squares) in the presence of 300 M cAMP. Data were fit with a single binding site model (see "Experimental Procedures"). n ϭ 6 for all measurements.
neighboring cysteines (R623C bB A624C,T627H,D631H) was further increased to 0.205 Ϯ .001, a value approaching that for mBBr in glycerol (0.228 Ϯ 0.002). Both values for bifunctional bimane were significantly greater than for bimane with one reactive functional group. These data indicate that bifunctional bimane reacted with both cysteines on the C-helix, reducing its rotational degrees of freedom.
Despite the difference in orientation and mobility of bimane when it is covalently attached to two residues, the distances measured using tmFRET were very similar. We measured the concentration-dependent quenching of bifunctional bimane in two different orientations by Co 2ϩ bound to two different pairs of histidines (T627H,D631H and I630H,D634H) in the presence of 300 M cAMP. The calculated distances were similar for bimane and bifunctional bimane (Fig. 3A). The slight differences in our measurements (1-3 Å) were expected because bimane, which can freely rotate, should be on average slightly closer to the metal-binding sites than either bifunctional bimane construct. Bifunctional bimane attached between M621C and A624C appeared slightly closer than bifunctional bimane attached between R623C and A624C. Because we do not know the precise position of the fluorophores in these constructs, we again calculated the distance between the two metal-binding sites, which should factor out the fluorophore position (Fig. 3B). The distance between the sites was similar, regardless of the way in which bimane was attached to the C-helix, indicating that the flexibility of the short linker and the orientation of the bimane do not have large effects on distance measurements when using tmFRET. This minimal orientation dependence might result, in part, from the fact that the transition metal acceptors in tmFRET have multiple absorption dipoles and therefore no single, defined orientation.
cAMP-dependent Conformational Changes in the C-helix-We measured tmFRET between Co 2ϩ bound at the T627H, D631H site and bimane attached in three different arrangements, A624C mB , M621C bB A624C, and R623C bB A624C in the presence and absence of cAMP (Fig. 4A). For every combination tested, the metal binding affinity increased in the presence of cAMP. An increased affinity for metal binding to two histidines in helical register suggests a stabilization of the ␣-helical structure, i.e. an ordering of the C-helix in the presence of cAMP (6, 34 -36). Fig. 4C shows the increase in Co 2ϩ -binding energy (⌬⌬G) for the T627H,D631H site as a result of agonist binding. Co 2ϩ binding was stabilized between 0.8 and 1.6 kcal/ mol. We observed a greater cAMP-dependent stabilization of transition metal binding in the construct with bimane attached to a single cysteine (A624C mB ,T627H,D631H) than in either construct in which bifunctional bimane was reacted with two cysteines within the C-helix. This may be the result of reduced flexibility of the helix when the conformational space that can be sampled is restricted by cross-linking two residues.
If agonist binding is accompanied by a secondary structural change in the C-helix, it might result in a change in the distance between the donor and acceptor sites on the helix. In addition to the stabilization of metal binding gained by the addition of cAMP, we also observed a change in the distance between the donor fluorophore and the acceptor metal ion in the C-helix. Again, regardless of the configuration of bimane used (A624C mB , M621C bB A624C or R623C bB A624C), we measured a cAMPdependent change in the FRET efficiency such that there was always significantly greater FRET in the presence of agonist than in its absence (Fig. 4, A and B). From the FRET efficiencies, we calculated distances with and without cAMP present and expressed these values as a change in distance upon addition of agonist (Fig. 4D). With bimane attached at position 624 (A624C), the fluorophore moved 0.52 Ϯ 0.08 Å closer to the di-histidinebinding site (T627H,D631H) upon addition of cAMP. A similar value was obtained with bifunctional bimane attached to cysteines spaced three residues apart (M621C bB A624C) where the donor and acceptor moved 0.52 Ϯ 0.26 Å closer together. Although small, these significant distance changes occurred within a single turn of the helix and likely reflect some disorder of the helix at this location in the absence of agonist.
Surprisingly, the agonist-dependent change in distance between the donor and acceptor site was much greater when measured with bifunctional bimane reacted with neighboring cysteine residues (R623C bB A624C,T627H,D631H; Fig. 4D).
These donor and acceptor sites moved 2.72 Ϯ 0.14 Å closer in the presence of cAMP. This increase in the agonist-induced distance change arose largely from the greater distance separation between the donor and acceptor site in this construct in the absence of agonist (Fig. 4, A and B). Most likely, tethering adjacent cysteines together distorted the structure of the C-helix with no agonist present, and this distortion was overcome by the ordering of the helix in the presence of cAMP. Whatever the case, the dramatic FRET changes observed for R623C bB A624C,T627H,D631H further support the hypothesis that cAMP induces a conformational change within the C-helix.
Dependence of FRET Changes on cAMP Concentration-The changes in tmFRET within the C-helix due to cAMP binding could be used to measure the apparent affinity for cAMP. Fig. 5 shows the quenching of R623C bB A624C,T627H,D631H by a nearly saturating Co 2ϩ concentration (1 mM), as a function of the total cAMP concentration. The apparent K d values for cAMP binding derived from fits to the data were 2.8 Ϯ 1.5 M. This value is in the concentration range reported from fluorescence and calorimetry studies of monomeric HCN2 proteins  ). B, cAMP-dependent changes in FRET efficiency and distances between bimane in three different configurations on the C-helix and Co 2ϩ bound at the T627H,D631H site. * denotes p Յ 0.05, paired samples t test. C, change in the free energy of Co 2ϩ binding to the T627H,D631H di-histidine site resulting from the binding of cAMP expressed as ⌬⌬G (⌬G cAMP Ϫ ⌬G no cAMP ). D, distance changes with cAMP between bimane covalently attached in three different configurations to the C-helix HCN2 constructs and Co 2ϩ bound to the T627H,D631H di-histidine site. Distances were calculated as distance no cAMP Ϫ distance cAMP . n ϭ 6 for all measurements.
(130 nM to 13.2 M), indicating that the mutations and modifications of the C-helix did not dramatically alter the cAMP binding (22,37,38). In the absence of metal, 300 M cAMP had no significant effect on the fluorescence emission of R623C bB A624C (data not shown).
cAMP Increases the Distance between the B-helix and the Distal C-helix-If the C-helix secondary structure was altered by cAMP binding, we also expect to observe changes in distance between residues on the C-helix and its nearest neighbor, the B-helix. We measured FRET between bimane at position 635 on the distal C-helix and Co 2ϩ bound to a pair of histidines on the B-helix (R635C mB ,N612H,E616H, see Fig. 6A). Co 2ϩ bound to the B-helix quenched bimane in a concentration-dependent fashion in both the absence and presence of 300 M cAMP (Fig.  6B). Unlike Co 2ϩ binding to histidines on the C-helix, there was no change in the free energy of Co 2ϩ binding to the B-helix upon addition of cAMP (Fig. 6C). There was, however, a large increase in the distance between the distal C-helix and the B-helix upon addition of cAMP. In the absence of cAMP, bimane at position 635 was 13.3 Ϯ 0.1 Å away from Co 2ϩ bound to the B-helix. In the presence of cAMP, this distance was increased to 15.4 Ϯ 0.2 Å. Such a distance change would not be possible if the C-helix moved as a rigid body. The most likely explanation for this observation is that in the absence of cAMP, the C-helix adopted a flexible coil conformation, allowing bimane tethered to the helix to more closely approach a Co 2ϩ ion bound to the B-helix.
Ordering of the C-helix Promotes Gating in Intact Channels-We have shown that there is a change in the secondary structure of the C-helix of the C-terminal fragment of HCN2 induced by agonist binding. Does this conformational change occur in intact functioning channels? If there is a transition of the C-helix from a less ordered state to a more stable helical conformation, is this structural change simply a consequence of agonist binding, or is it coupled to opening of the ion channel pore, i.e. is it a transition associated with agonist binding or channel gating?
To answer these questions, we turned to intact CNG channels. CNG channels are highly homologous to HCN channels (31.7% identical in the C-terminal region, see Fig. 1C), and available data suggest that their C-terminal regions are structurally very similar (39,40). However, cyclic nucleotides have a  . FRET between the B-helix and C-helix. A, schematic illustrating bimane conjugated to R635C on the C-helix and the di-histidine mutant N612H,E616H in the B-helix bound to a Co 2ϩ ion. B, metal concentration dependence of FRET between bimane attached to R635C and Co 2ϩ bound to N612H,E616H in the presence and absence of 300 M cAMP. n ϭ 6 for all measurements. C, free energy of Co 2ϩ binding to R635C mB ,N612H,E616H in the presence and absence of cAMP. n ϭ 6 for no cAMP and 5 for cAMP. D, distance between bimane attached to R635C and Co 2ϩ bound to N612H,E616H in the presence and absence of 300 M cAMP. * denotes p Յ 0.05, paired samples t test. n ϭ 6 for no cAMP and 5 for cAMP. much larger energetic effect on the gating of CNG channels than they do on HCN channels (1). Furthermore, although cGMP is a full agonist of CNG channels, cIMP is a partial agonist of CNGA1 channels, i.e. it does not fully activate channels at saturating concentrations. This can be useful for differentiating between binding effects and gating effects. Finally, the effects of metal binding to CNGA1 channels have been well characterized, and the endogenous metal-binding site can be removed (27,(41)(42)(43).
To test the hypothesis that ordering of the C-helix is coupled to gating of cyclic nucleotide-regulated channels, we inserted a di-histidine metal-binding site into the C-helix of a mutant CNGA1 channel with its native Ni 2ϩ potentiation site removed (CNGA1 Q , see under "Experimental Procedures"). The di-histidine mutation, CNGA1 Q E595H,Q599H, was placed in an analogous position to the T627H,D631H site in HCN2 based on sequence alignment and homology modeling (Fig. 7A). We reasoned that if the conformational change induced by cyclic nucleotide-stabilized metal binding to a di-histidine site in the C-helix, then binding metal to the di-histidine site in the C-helix should promote the conformational change induced by agonist binding. To separate the effects of Ni 2ϩ on initial cyclic nucleotide binding from the effects on gating the channel, we used a saturating concentration of the partial agonist cIMP. If metal binding to the di-histidine site caused an increase in the current response to cIMP relative to cGMP when all the cyclic nucleotide-binding sites were occupied, then the increase in current must have come from an increase in the favorability of the opening transition, i.e. from promotion of the gating conformational change and not a change in the initial binding affinity for agonist. Fig. 7B shows CNGA1 Q E595H,Q599H currents in response to a voltage pulse from a holding potential of 0 to Ϫ60 and ϩ60 mV in the presence of a saturating concentration of the full agonist cGMP (10 mM) and the partial agonist cIMP (16 mM). Similar data were compiled to generate a concentration-response for cGMP and cIMP on CNGA1 Q E595H,Q599H (Fig.  7C). At a saturating concentration of cIMP, the partial activation at Ϫ60 mV was 0.38 Ϯ 0.06, relative to saturating cGMP, consistent with the value reported for wild type CNGA1 channels (18).
We then examined the effect of Ni 2ϩ on the current at saturating concentrations of cIMP. Fig. 7D shows CNGA1 Q control currents resulting from steps to Ϫ100 and Ϫ60 mV from a holding potential of 0 mV in 16 mM cIMP in the absence and presence of 100 M Ni 2ϩ . In the absence of an introduced metal-binding site, the current magnitude did not change in response to this concentration of Ni 2ϩ . We tested multiple concentrations of Ni 2ϩ and found no effect on current magnitude in CNGA1 Q , except at 1 mM, where some degree of Ni 2ϩ block was evident, as described previously (Fig. 7E) (43). In contrast, the current magnitude in CNGA1 Q E595H,Q599H channels di-histidine site, which is homologous to the T627H,D631H site in HCN2. B, CNGA1 Q E595H,Q599H currents in response to voltage steps to Ϫ60 and ϩ60 mV from a holding potential of 0 mV in the presence of 10 mM cGMP (black) or 16 mM cIMP (blue). C, normalized (to 10 mM cGMP) CNGA1 Q E595H,Q599H currents in response to steps to Ϫ60 mV as a function of cGMP (black) or cIMP (blue) concentration. Data were fit with the Hill equation (see "Experimental Procedures"). K d, cGMP ϭ 76.7 M, n cGMP ϭ 1.9, K d, cIMP ϭ 2.0 mM, n cIMP ϭ 1.7. n ϭ 6 for all measurements. D, CNGA1 Q (top) and CNGA1 Q E595H,Q599H (bottom) currents in response to voltage steps to Ϫ100 and Ϫ60 mV from a holding potential of 0 mV in the presence of 16 mM cIMP (blue) or 16 mM cIMP plus 100 M Ni 2ϩ (green). E, normalized (to 16 mM cIMP) CNGA1 Q (gray) and CNGA1 Q E595H,Q599H (black) currents in response to steps to Ϫ100 mV as a function of Ni 2ϩ concentration. Data were fit as described under "Experimental Procedures." The solid gray line represents a fit to the CNGA1 Q data in which only block by Ni 2ϩ was observed. was significantly enhanced upon binding of 100 M Ni 2ϩ (Fig.  7D). We observed an increase in cIMP-induced currents as the Ni 2ϩ concentration was increased (Fig. 7E). Again, there was a decrease in current at 1 mM Ni 2ϩ . We fit these data with a model that takes into account both the potentiation and block by Ni 2ϩ to determine the maximal effect of Ni 2ϩ on channel gating (see under "Experimental Procedures"). Ni 2ϩ bound to CNGA1 Q E595H,Q599H with an apparent affinity of 5.0 M, similar to the binding affinity of the homologous site in the C-helix of HCN2 (Fig. 2E). From our fits, we determined that stabilizing the C-helix by Ni 2ϩ binding to the E595H,Q599H site potentiated the current in saturating cIMP by a factor of 1.6. This corresponded to a stabilization of the opening gating transition by 0.47 kcal/mol, on the same order of magnitude as the values we measured for stabilization of Co 2ϩ binding to the T627H,D631H site in the C-helix of HCN2 in response to cAMP (Fig. 4C). These results indicate that ordering of the C-helix is tightly coupled to the opening conformational change in these cyclic nucleotide-regulated channels.

DISCUSSION
These data led us to a revised model for the agonist-induced conformational changes in eukaryotic cyclic nucleotide-regulated channels (Fig. 8). In the absence of cyclic nucleotide, the C-helix does not form a tight helical structure, and the channel pore is closed (unbound/closed). Agonist initially binds to residues in the P-helix and ␤-roll, with the ion channel pore still closed (bound/closed). Subsequent to agonist binding, there is an ordering of the C-helix as well as the FЈ-helix. Additionally, the C-helix moves closer to the agonist-binding site. This ordering of the C-helix, and possibly the translational movement as well, is coupled to the opening conformational change in the pore (bound/open).
Previous work in HCN channels has suggested that there is an ordering of the C-helix induced by agonist binding (6). In the apo state crystal structure of HCN2, there is no density resolved for at least half of the C-helix. This may indicate that the helix is disordered or adopts multiple conformations. However, crystal packing interactions at the C-helix may have distorted the structure. There is also the possibility that Br Ϫ ions in the agonist-binding pocket acted as pseudo ligands. We also previously demonstrated that Ni 2ϩ bound more tightly to several histidine pairs in helical register on the C-helix when cAMP was present. In this study, we extended this observation by directly examining distance changes between locations within the C-helix with tmFRET. The cAMP-dependent increases in FRET efficiency observed between bimane and bifunctional bimane attached to the N-terminal end of the C-helix and transition metal ions bound to a site roughly one helical turn away are best interpreted as a change in conformation within the helix itself. We again observed an increase in metal affinity for a di-histidinebinding site in the C-helix in the presence of cAMP. We also examined the distance between the distal end of the C-helix and the neighboring B-helix, which was significantly shorter in the absence of cAMP than in its presence. The simplest explanation for these observations is that the C-helix adopts a flexible, coiled conformation in the absence of agonist. Finally, working in intact CNG channels, we demonstrated that stabilizing the helical conformation of the C-helix promoted channel opening in a saturating concentration of the partial agonist cIMP. This result suggests that the transition in the C-helix from a less ordered structure to a stable helix is directly coupled to channel gating.
Agonist-induced secondary structural transitions in the C-helix have been reported for other CNBD-containing proteins. NMR studies comparing the apo-and cyclic nucleotidebound forms of the E. coli transcription factor CAP clearly demonstrate a coil-to-helix transition in this region subsequent to agonist binding (44). Crystal structures of the apo state of the prokaryotic, cyclic nucleotide-activated K ϩ channel MloK1 show a variety of conformations of the C-helix, including some structures in which density of part of the C-helix is absent (9). However, a recent NMR structure of MloK1 suggests that although both the B-and C-helices move closer to the binding site following agonist binding, they move as rigid bodies, with no accompanying change in secondary structure (45). Although the details vary, large conformational changes of this sort have also been reported for the kinases PKA and PKG as well as for Epac (46). Therefore, although we believe there is a rearrangement of the C-helix associated with cyclic nucleotide binding in almost all CNBD-containing proteins, the details of the conformational change may differ for different proteins.
This study demonstrates further the utility of tmFRET to measure very short distances and small distance changes in proteins. Using the small fluorophore bimane and transition metal ion acceptors, we were able to reproducibly measure changes in FRET efficiencies between adjacent rungs of an ␣-helix. This measurement would not have been possible using FIGURE 8. Model for the gating conformational change in cyclic nucleotide-regulated ion channels. cAMP binds to the ␤-roll of the closed channel. Subsequent to agonist binding, it forms contacts with the C-helix, stabilizing its structure. This stabilization is coupled to opening of the channel pore.
classical FRET methods where long R 0 values of commonly used donor/acceptor pairs would make it necessary to place the donor and acceptor on the order of 30 -50 Å apart to be in the proper range for distance measurements from FRET. To suit our particular distance range, we chose Co 2ϩ among multiple transition metals to improve the accuracy and precision of our measurements. Finally, we used bifunctional bimane to reduce the flexibility of the fluorophore and demonstrated that tmFRET measurements are not strongly influenced by reduced flexibility or fixed orientation of the fluorescent donor. This orientation independence likely resulted, at least in part, from the multiple absorption dipoles of the acceptor ion. tmFRET has allowed for increased insight into the conformational transitions in the CNBD of cyclic nucleotide-regulated channels and offers hope for measurements of the structural dynamics in many proteins.