Identification of a Domain on the β-Subunit of the Rod cGMP-gated Cation Channel That Mediates Inhibition by Calcium-Calmodulin*

The cGMP-gated cation channel mediating phototransduction in retinal rods has recently been shown to be inhibited by calcium-calmodulin, through direct binding of the latter to the β-subunit of the heterotetrameric channel complex. Here, we report the characterization of this inhibition and the identification of a domain crucial for this modulation. Heterologous expression of the α- and β-subunits of the human rod channel in HEK 293 cells produced a cGMP-gated current that was highly sensitive to calcium-calmodulin, with half-maximal inhibition at approximately 4 nm. In biochemical and electrophysiological experiments on deletion mutants of the β-subunit, we have identified a region on its cytoplasmic N terminus that binds calmodulin and is necessary for the calmodulin-mediated inhibition of the channel. However, in gel shift assays and fluorescence emission experiments, peptides derived from this region indicated a low calmodulin affinity, with dissociation constants of approximately 3–10 μm. On the C terminus, a region was also found to bind calmodulin, but it was likewise of low affinity, and its deletion did not abolish the calmodulin-mediated inhibition. We suggest that although the identified region on the N terminus of the β-subunit is crucial for the calmodulin effect, other regions are likely to be involved as well. In this respect, the rod channel appears to differ from the olfactory cyclic nucleotide-gated channel, which is also modulated by calcium-calmodulin.

Visual transduction in retinal rods involves a light-triggered signaling cascade that leads to a decrease in cGMP concentration (for review, see Refs. [1][2][3]. In darkness, cytoplasmic cGMP binds to and opens cGMP-gated, nonselective cation channels on the plasma membrane (for review, see Refs. 4 -7). These open channels sustain an inward dark current, carried predominantly by Na ϩ and Ca 2ϩ , that keeps the cell depolarized. In the light, the hydrolysis of cGMP leads to the closure of these channels, resulting in a membrane hyperpolarization as the electrical response. When the cGMP-activated channels close, the Ca 2ϩ influx into the rod outer segment stops, but a Ca 2ϩ efflux through a Na ϩ /Ca 2ϩ , K ϩ exchanger continues, leading to a decrease in the cytoplasmic Ca 2ϩ concentration in the outer segment. This decrease in Ca 2ϩ triggers a negative feedback to produce light adaptation of the rod (for review, see Ref. 3). One mechanism underlying this Ca 2ϩ feedback involves a reduction of the apparent affinity of the channel for cGMP (8,9), through the action of one or more Ca 2ϩ -binding proteins, one of which is calmodulin (CaM) 1 (10 -14).
The rod channel belongs to a family of cyclic nucleotideactivated, nonselective cation channels now known to be important for both visual and olfactory transduction pathways (for review, see Refs. [5][6][7]. These channels are ligand-gated, being opened directly by cGMP and cAMP. They are composed of at least two subunit species (␣ and ␤, or 1 and 2) most probably forming heterotetrameric complexes (15,16). The ␣-subunit, but not the ␤-subunit, is capable of forming functional homomeric channels. Like the Shaker superfamily of potassium channels, both subunits have six transmembrane domains and a putative ␤-hairpin that forms part of the pore. The cyclic nucleotide-binding site is situated on the cytoplasmic C terminus and is homologous to the binding sites found on the cyclic nucleotide-activated kinases protein kinase C and protein kinase A, and the Escherichia coli catabolite gene activator protein, CAP (see . The reduction in affinity of the rod cGMP-gated channel for cGMP involves Ca 2ϩ -CaM binding to the ␤-subunit (Refs. 17 and 18; for review, see Refs. 19 and 20). A similar, but more potent, inhibition by Ca 2ϩ -CaM was found for the olfactory cyclic nucleotide-gated channel, although in this case the modulation involves Ca 2ϩ -CaM binding to the ␣-subunit of the channel (21,22). The mechanism by which the olfactory channel is modulated by Ca 2ϩ -CaM has been elucidated (22). A domain on the N terminus of the olfactory channel ␣-subunit influences gating by promoting the open state of the liganded channel. When Ca 2ϩ -CaM binds to this domain, the influence of the latter on channel gating is removed, leading to a decrease in the apparent affinity of the channel for cGMP due to the coupling of ligand binding and channel gating. Most recently, it has been reported that the N and C termini of the olfactory channel ␣-subunit interact directly with each other and that the domain on the N terminus that is important for this interaction coincides with the Ca 2ϩ -CaM-binding site (23). When Ca 2ϩ -CaM binds, the interaction between the two termini disappears (23). Presumably, this interaction influences the gating of the channel and accounts for the ability of Ca 2ϩ -CaM to modulate the channel.
In this paper, we address the question of how the rod channel is modulated by Ca 2ϩ -CaM. In an electrophysiological approach, we recorded cGMP-activated currents in the absence and presence of Ca 2ϩ -CaM from excised, inside-out membrane patches of HEK 293 cells expressing the human rod channel ␣-and ␤-subunits. Furthermore, site-directed mutagenesis and biochemical binding studies were carried out to identify the Ca 2ϩ -CaM-binding domain on the ␤-subunit important for this modulation. The results indicate similarities, but also differences, between the rod and olfactory channels with respect to the CaM inhibition.

EXPERIMENTAL PROCEDURES
Cloning of the Full-length Human Rod Channel ␤-Subunit-Poly(A) ϩ RNA was isolated from bovine retina with the oligo(dT)-selection method (Micro-FastTrack, Invitrogen, Carlsbad, CA). Oligo(dT)-primed cDNA was then synthesized using the SuperScript TM Choice system (Life Technologies, Inc.). Using this cDNA as template, PCR was performed with the primers AGGAAGAAGGCAAGTCCTG and AT-GGGCTTGATCTCCAAGG, corresponding to nucleotides Ϫ50 to Ϫ32 and 635-617 of the cDNA coding for bovine GARP (glutamic acid-rich protein; Ref. 24). The PCR product with the appropriate size was isolated, sequenced, and used as a probe to screen an adult human retinal cDNA library in gt10. Six positive clones were isolated. The clones fell into two groups according to their restriction patterns. One clone from each group was used for further characterization. The inserts were subcloned into pCIS and sequenced. These two clones, named hGARP1.6 and hGARP2.5, shared a common 5Ј-region of 0.9 kb but were divergent in their 3Ј-regions. The 3Ј-region of hGARP2.5 overlapped with the 5Ј-end of the hRCNC2b clone described in Chen et al. (25) (see "Results"). To combine hGARP2.5 and hRCNC2b, two of three Bsu36I sites in hGARP2.5 were deleted by silent mutations. A Bsu36I-BamHI fragment from hRCNC2b (spanning nucleotides 417-2849) was then ligated into hGARP2.5, generating a clone that appears to code for the full-length ␤-subunit of the human rod channel, by comparison to the bovine ortholog. In this paper, we refer to hRCNC2b as the trunc-␤-subunit and to the full-length ␤-subunit as the full-␤-subunit.
Channel Protein Expression and Electrophysiological Recordings-The cDNAs coding for the human rod cGMP-activated channel ␣-subunit (corresponding to hRCNC1 in Ref. 26 and renamed hRCNC␣ here) and the trunc-␤-or full-␤-subunit were subcloned in the pCIS expression vector and cotransfected into human embryonic kidney (HEK) 293 cells (American Type Culture Collection) using the calcium-phosphate method (27). For hRCNC␣, a Kozak consensus sequence (28) had been introduced into its 5Ј-region, which enhanced the level of protein expression by a modest degree. HEK 293 cells were cultured at 37°C in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum and penicillin/streptomycin in a humidified atmosphere containing 5% CO 2 . At 2-4 days after transfection, current recordings were made from voltage-clamped, excised, inside-out membrane patches using an EPC-9 patch-clamp amplifier and the PULSE/PULSE-FIT software package (both from HEKA Elektronik, Lamprecht, Germany, distributed by Instrutech, Great Neck, NY). The signals were low pass filtered at 2.9 kHz (4-pole Bessel filter). The patch pipettes were made from borosilicate glass capillaries, with a tip-lumen diameter of 1-1.5 m and a resistance of 2-4 M⍀. The relatively low signal-to-noise ratio in some of the traces had resulted from the small magnitudes of the currents and the lack of signal averaging.
For zero-Ca 2ϩ conditions, the pipette and bath solutions both contained 140 mM NaCl, 5 mM KCl, 2 mM EGTA, and 10 mM HEPES/ NaOH, pH 7.4. In experiments involving Ca 2ϩ -CaM, the bath was perfused with a solution containing CaM at a specified concentration and also 50 M buffered free Ca 2ϩ (achieved by substituting 2 mM nitrilotriacetic acid and 704 M CaCl 2 for the EGTA). cGMP was added to the bath solution as needed. In experiments involving the ␤-subunit, the cGMP-activated current was always tested for blockage at ϩ60 mV by 10 M L-cis-diltiazem applied to the bath solution to verify its functional expression (25). A solenoid-controlled rotary valve system (29) was used to change the bath solution, and the solution change around the membrane patch was complete within 1-2 s. All experiments were performed at room temperature.
Mutagenesis and Fusion Protein Construction-Deletion mutants were generated by performing site-directed mutagenesis on the cDNAs coding for the trunc-␤-and full-␤-subunits. For binding studies, fusion protein constructs containing the cytoplasmic N terminus (amino acid residues 1-313) or C terminus (residues 535-908) of the trunc-␤-subunit were made by PCR amplification using primers with flanking BamHI and EcoRI sites and subcloning the PCR fragments into pGEX-2T (Amersham Pharmacia Biotech). The resulting constructs were transformed into E. coli BL21 cells, and the fusion proteins were isolated and purified using the Bulk GST purification module from Amersham Pharmacia Biotech.
Fluorescence Measurements-Fluorescence experiments to measure the binding of peptides to Ca 2ϩ CaM were performed using the Perkin-Elmer Luminescence Spectrometer LS50B. Peptides, representing putative Ca 2ϩ -CaM-binding sites on the trunc-␤-subunit, were synthesized in the Howard Hughes Medical Institute Biopolymer Facility at Johns Hopkins University School of Medicine. Dansyl-CaM (Sigma) was incubated at a given concentration with increasing peptide concentration in a buffer containing 50 mM Tris-HCl, pH 7.3, 150 mM NaCl, and 0.5 mM CaCl 2 or 2 mM EGTA. The emission spectrum at 400 -600 nm was recorded using an excitation wavelength of 340 nm, the bandwidth being 10 -15 nm for both excitation and emission. The increase in fluorescence at 480 nm was used to assay for the concentration of dansyl-CaM bound to peptide. Assuming a 1:1 stoichiometry of binding between Ca 2ϩ -CaM and peptide, the fraction of peptide, f b , bound to CaM is given by the equation , where I f is the dansyl-CaM fluorescence with no peptide present, I b is the fluorescence when all dansyl-CaM is bound to peptide, and I m is the fluorescence of intermediate mixtures (see Ref. 22). The dissociation constant, K d , between Ca 2ϩ -CaM and the peptide was derived from the relationship between the fractional increase of fluorescence and the calculated concentration of free peptide. According to vendor specifications, the mole ratio between the dansyl moiety and CaM in dansyl-CaM is about 0.6. However, this factor does not influence the results, provided that CaM and dansyl-CaM behave identically in the binding experiments.
Antibody Generation-A polyclonal antibody, Ab1859, was raised in rabbit (HRP Inc., Denver, PA) against the resin-coupled peptide MLG-WVQRVLPQPPGTPRKTK, which corresponds to amino acids 1-20 in the full-␤-subunit (see Fig. 1B). For immunoprecipitation experiments, the antibody was purified by protein A-Sepharose chromatography. Ab1859 cross-reacts with the bovine homolog of the protein.
CaM Overlay Experiments and Western Blotting-Overlay experiments were performed on purified fusion proteins (see above), retinal lysates, and the heterologously expressed full-␤-subunit. Retinal lysates were prepared from human retinal tissue that had been isolated and frozen in liquid nitrogen within 5 h post-mortem. The low salt lysis buffer contained 10 mM Tris-HCl, pH 7.5, 2 mM EDTA, 1 mM dithiothreitol, 10 g/ml leupeptin, 2 g/ml aprotinin, and 1% Triton X-100. For the heterologously expressed full-␤-subunit, HEK 293 cells were harvested 3-4 days after transfection and lysed in the above low salt buffer. The ␤-subunit was immunoprecipitated by incubating the cell lysate with 10 g of the purified Ab1859 antibody and a suspension of protein A-Sepharose (Sigma) for 2 h at 4°C. Alternatively, the antibody was covalently coupled to CNBr-activated Sepharose 4B and used instead.
The fusion proteins, retinal lysates, and immunoprecipitates were loaded on SDS gels and transferred to nitrocellulose (TransBlot, Bio-Rad) or polyvinylidene difluoride membranes (Immobilon, Millipore, Bedford, MA) in 10 mM CAPS, pH 10.8, or Towbin buffer containing 2-10% methanol (30). After transfer, the blots were probed with CaMcoupled alkaline phosphatase or biotinylated CaM in the presence of 0.1-1 mM Ca 2ϩ or 5 mM EGTA. The synthesis of CaM-coupled alkaline phosphatase and the detection procedure were both according to Walker et al. (31). In assays with biotinylated CaM, the membranes were blocked in a buffer containing 150 mM NaCl, 10 mM Tris-HCl, pH 7.5, 1 mM CaCl 2 or 5 mM EGTA, 0.1% antifoam A, and 5% nonfat dry milk for 30 min. Biotinylated CaM (Biomedical Technologies, Stoughton, MA) was added to give a final concentration of 1 g/ml, followed by an incubation for 1-2 h at room temperature. After extensive washing in the same buffer without additives, the membrane was incubated with avidin and horseradish peroxidase (ABC system, Vector Laboratories, Burlingame, CA) and developed using the ECL system (Amersham Pharmacia Biotech).
For Western blotting, the membranes were blocked in 2% nonfat dry milk in Tris-buffered saline (140 mM NaCl, 10 mM Tris-HCl, pH 7.5) and incubated with the antibodies in the same buffer for 1 h at room temperature or overnight at 4°C. The antibody Ab1859 was used at 1:2000 dilution. Fusion proteins were detected using the commercially available anti-glutathione S-transferase (GST) antibody (Amrad, Melbourne, Victoria, Australia) at a dilution of 1:5000. The bands were visualized using an horseradish peroxidase-coupled secondary antibody (Amersham Pharmacia Biotech) or the ABC system (Vector Laboratories). For Western blotting after the CaM overlay experiment, blots were stripped using 1% SDS, 1 mM EDTA in Tris-buffered saline.
Nondenaturing Polyacrylamide Gel Shift Assays-CaM (375 pmol) was incubated with different molar amounts of peptide in a buffer containing 10 mM HEPES/NaOH (pH 7.2) and 2 mM CaCl 2 or 5 mM EGTA, respectively, for 30 min at room temperature. The CaM-peptide complexes were then resolved by nondenaturing gel electrophoresis on 15% gels according to standard procedures for SDS-polyacrylamide gel electrophoresis, but omitting SDS and adding 2 mM CaCl 2 or 5 mM EGTA, respectively. Bands were visualized by Coomassie Blue staining.

Cloning of the Full-length ␤-Subunit of the Human Rod
Channel-Previously, we cloned the cDNA for a ␤-subunit of the human rod cGMP-gated channel (hRCNC2b; see Ref. 25). Subsequently, this clone appeared to be a truncated form of the cDNA for the full-length ␤-subunit, with the coded protein missing a segment of the N terminus (17,18). Nonetheless, when expressed, the truncated human protein exhibited all of the hallmark properties of the bovine full-length protein that was subsequently cloned by Körschen et al. (18). One of these common properties is the modulation by Ca 2ϩ -CaM. In this study, we obtained the human full-length channel cDNA to use for experiments. We generated a probe by performing PCR on cDNA synthesized from bovine retinal poly(A) ϩ -RNA, using primers based on the 5Ј-segment (the GARP region) of the bovine full-length ␤-subunit (Ref. 24; see also Ref. 18). This probe was used to screen an adult human retinal cDNA library (see "Experimental Procedures"). Two different types of clones were obtained. One clone, named hGARP1.6 (see Fig. 1), is identical to a human cDNA clone previously published by Ardell et al. (32) and to the human t-GARP clone more recently reported by Colville and Molday (33), except for the absence of six amino acid residues (GAASDP) at position 188 in our hGARP1.6 (indicated by insertion in Fig. 1; see also Refs. 32 and 33). The reason for this difference is unclear, because it appeared that the same cDNA library was used by all three groups. The hGARP1.6 clone shares a common 5Ј region with the other clone, hGARP2.5, but has a unique 3Ј region that can also be found in the genomic DNA for the ␤-subunit (data not shown), suggesting that it is a differentially spliced variant. hGARP1.6 was not pursued further. As for hGARP2.5, its 3Јend shows regions of identity with the 5Ј-end of the translated region of hRCNC2b, except that it has an extra insertion of 264 bases and a unique tail (Fig. 1); furthermore, there is a stop codon 147 bases into the insertion. In the translated protein from this clone, the C terminus overlaps with the N terminus of hRCNC2b for a stretch of 117 amino acids, followed by a unique stretch of 49 residues. The overall protein coded by the hGARP2.5 clone has 502 amino acids and a calculated molecular mass of 55,322 Da. However, in SDS-gels, the expressed protein showed an apparent molecular mass of around 110 kDa ( Fig. 2A, hGARP2.5), probably because of its high glutamic acid content. hGARP2.5 is presumably the human equivalent of the bovine f-GARP described by Colville and Molday (33).
We fused hGARP2.5 and hRCNC2b using Bsu36I and BamHI restriction sites to produce a single 3.8-kb insert coding for the complete ␤-subunit of the human rod channel (see Fig. 1), by comparison to the bovine ortholog (18). The translated protein has 1245 amino acid residues and a calculated molecular mass of 139,160 Da. The protein, obtained either directly from lysate of transfected HEK 293 cells or by immunoprecipitation from the lysate with an antibody (Ab1859) generated against its GARP part (see "Experimental Procedures"), migrated on an SDS-gel with an apparent molecular mass of approximately 220 kDa (Fig.  2, A and B). This full-length clone is identical to that recently identified by Colville and Molday (33), except again for the missing six residues in our protein described above. Co-transfection of HEK 293 cells with this full-length ␤-subunit cDNA and that for hRCNC␣ (renamed from hRCNC1 of Ref. 26) produced cGMPactivated channels with properties very similar to those observed in co-transfections involving the hRCNC␣ and hRCNC2b cDNAs. For instance, both types of heteromeric channels were blocked by L-cis-diltiazem (see Fig. 2C for channels containing the full-length ␤-subunit). Also, Ca 2ϩ -CaM inhibited both types of channels to about the same extent. This is consistent with the results obtained by Körschen et al. (18) with the bovine protein.
We refer to hRCNC2b as the trunc-␤-subunit and to the full- length ␤-subunit as the full-␤-subunit.
Inhibition of the ␣/␤-Heteromeric Channel by Ca 2ϩ -CaM-We first characterized the inhibition of the ␣/␤-heteromeric rod channel by Ca 2ϩ -CaM in greater detail than before (17). The cDNAs for the ␣and trunc-␤-subunits were transfected into HEK 293 cells and the macroscopic cGMP-activated current was recorded from excised, inside-out membrane patches. At Ϫ60 mV, 60 M cGMP evoked a current less than half-maximum, and 250 nM CaM reversibly inhibited this current by 60 -80% in the presence of 50 M Ca 2ϩ (Fig. 3A). In the absence of Ca 2ϩ , CaM was unable to elicit the inhibitory effect (data not shown). Also, the recovery of the current from the Ca 2ϩ -CaM inhibition required the removal of both CaM and Ca 2ϩ (Fig. 3A). Fig. 3B shows the dose-response relationships, averaged from two patches, between current activation and cGMP concentration in the absence and presence of Ca 2ϩ -CaM, respectively. In the presence of CaM, the dose-response relationship was shifted by about 2-fold to higher cGMP concentrations. This extent of shift is consistent with our previous finding (17).
To investigate the relationship between the extent of channel inhibition and Ca 2ϩ -CaM concentration, we evoked a current from a patch at Ϫ60 mV with 60 M cGMP and measured the extent of inhibition of this current at different CaM concentrations in the presence of 50 M Ca 2ϩ (Fig. 3C). One or two low CaM concentrations were tested on each patch, and the inhibition of channel activity was normalized with respect to that obtained with 250 nM CaM on the same patch. Averaged results from 27 experiments indicated that half-maximal inhibition occurred at a concentration of approximately 4 nM CaM, with a Hill coefficient of 0.93 (Fig. 3D). Because practically all CaM should be Ca 2ϩ -bound in the presence of 50 M free Ca 2ϩ , Fig.  3D essentially describes the dependence of channel inhibition on the concentration of Ca 2ϩ -CaM. The value we obtained for half-maximal inhibition of the channel by Ca 2ϩ -CaM is fairly close to that (1-2 nM) previously measured with the native rod channel, either reconstituted in lipid vesicles or directly from rod outer segment membranes, using Ca 2ϩ flux as an assay for channel opening (11,13).
Channel Protein Binding Studies-We examined the binding of CaM to the native ␤-subunit in human retinal lysate using either biotinylated-CaM or CaM-coupled alkaline phosphatase as a probe. For comparison, lysates from bovine retina were also examined in parallel. For the bovine retinal lysate, the CaM probe recognized a band at a molecular mass of about 240 kDa in the presence of 1 mM Ca 2ϩ , which is expected for the bovine channel ␤-subunit (data not shown). For the human retinal lysate, the CaM probe likewise recognized a band at a molecular mass of approximately 240 kDa, which is slightly higher than the 220 kDa expected from the Western blot of the expressed protein (data not shown; see Fig. 2). It is likely that this signal resulted from the binding of CaM to a protein other than the ␤-subunit. We have not carried out CaM binding experiments with the immunopurified channel protein from human retinal lysate because of the scarcity of tissue. However, we performed similar experiments with the Ab1859-immunoprecipitated full-␤-subunit expressed in HEK 293 cells (see Fig.  2B) and failed to detect a binding signal. This could imply that the CaM affinity of the expressed ␤-subunit is very weak and therefore not detectable, or that under the gel overlay conditions, the human protein had not renatured enough to permit significant CaM binding. The intense band in Fig. 2B between 106 and 205 kDa is probably due to proteolytic cleavage of the full-length ␤-subunit.
We have repeated the same experiments using GST fusion proteins of the N and C termini of the ␤-subunit. For the N-terminal fusion protein, the N terminus of the trunc-␤-sub- unit was used instead of the full-␤-subunit, simply because it was shorter. Because the trunc-␤-subunit confers the same CaM effect as the full-␤-subunit, this point is immaterial. Both the N-and C-terminal fusion proteins were found to bind CaM when probed with biotinylated-CaM or CaM-coupled alkaline phosphatase in gel overlay assays ( Fig. 4; see also Figs. 5D and 8B). However, their affinities for CaM were much weaker than that observed for the GST-fusion protein of the N terminus of the ␣-subunit of the olfactory cyclic nucleotide-gated channel studied in parallel (Fig. 4, N-OCNC␣), which previous work has shown to have a high affinity for CaM (22). On the other hand, the much weaker CaM affinity of the C-terminal fusion protein compared with the N-terminal fusion protein (Fig. 4) could have resulted from a lower blotting efficiency, as suggested by Western blots (data not shown). We found that the N-and C-terminal fusion proteins of the trunc-␤-subunit retained the binding to CaM even in the absence of Ca 2ϩ (Fig. 4, right  panel). This finding could have suggested the presence of Ca 2ϩindependent binding sites for CaM on the N and C termini. However, even for the GST-fusion protein of the N terminus of the olfactory channel ␣-subunit, some CaM binding was retained in the absence of Ca 2ϩ . Furthermore, the experiments described below with peptides corresponding to the putative binding sites on the two termini indicated that these peptides required Ca 2ϩ to bind CaM. These observations, together with the Ca 2ϩ requirement for the functional modulation of the ␣/␤-heteromeric channel complex described earlier, indicate that the apparent Ca 2ϩ -independent CaM binding of the fusion proteins was possibly an artifact of the experimental conditions. Contradictory observations on the Ca 2ϩ dependence of CaM binding under different conditions has been reported for the Ras-like GTPase RIC, which required Ca 2ϩ to bind CaM in a gel overlay but did not require Ca 2ϩ to bind to CaM-agarose beads in solution (34). We have not pursued this point further. In any case, no strong binding of CaM to the ␤-subunit could be found for the heterologously expressed protein, either in its entirety or as fusion proteins, although this may be due to incomplete renaturation.
The C Terminus of the ␤-Subunit Does Not Contain a CaMbinding Site Crucial for the CaM-mediated Inhibition-We scanned the N-and C-terminal sequences of the trunc-␤-subunit and identified potential binding sites for CaM in both locations, based on the presence of conserved hydrophobic residues within short stretches of amino acid sequence (35,36). The sites on the C terminus are more prominent, so we focused on them first. On the C terminus, there are two potential regions, one upstream and the other downstream of the cyclic nucleotide-binding domain. The first region (see Fig. 5, arrowhead), corresponding to residues Ile 629 to Leu 646 and containing two consensus motifs, was ignored because this region bears strong homology to the corresponding region in the ␣-subunit of the rod channel (26), a protein known not to bind or be inhibited by CaM (10,11,17,22). For the second region, we synthesized a peptide, KY19 (Ala 781 -Lys 804 ; see Fig. 5A), and performed gel shift experiments. The peptide and CaM were loaded in different mole ratios onto a nondenaturing gel to examine the ability of the peptide to bind CaM and retard its migration. In the absence of Ca 2ϩ , KY19 did not retard the migration of CaM (data not shown). In the presence of 2 mM Ca 2ϩ , a mixture of KY19 and CaM at a ratio of 50:1 showed a limited ability to retard the migration of CaM (Fig. 5B). To measure the affinity between KY19 and CaM, we employed the fluorescent CaM derivative, dansyl-CaM. When a peptide is bound to dansyl-CaM, the fluorescence increases and shifts to shorter wavelengths (22,37). The relationship between the fraction of bound dansyl-CaM and the concentration of free peptide measured in these fluorescence experiments can be approximately described by a binding isotherm corresponding to the Hill equation with a coefficient of unity and a dissociation constant, K d , of approximately 15 M (Fig. 5C). To be certain that this low affinity did not result from an incomplete CaM-binding site on the peptide, we examined three other peptides, KY13, KY14, and KY15, spanning adjacent regions also rich in hydrophobic and positively charged amino acid residues (see Fig. 5A). KY13 and KY15 had a somewhat higher ability to retard CaM migration, but the effect was still weak (Fig. 5B). In fluorescence experiments with dansyl-CaM, the measured K d was approximately 3 M for KY13, 0.4 M for KY14, and 0.3 M for KY15 (Fig. 5C). As in the gel shift experiments, these peptides did not appear to interact with dansyl-CaM in the absence of Ca 2ϩ .
We examined deletion mutants of the C-terminal fusion protein for their ability to bind CaM in gel overlay experiments. When a mutant lacking the entire region downstream of the cyclic nucleotide-binding site was expressed in BL21 cells, the protein was hardly expressed. Another mutant (MG8) lacking the region (residues 749 -805) that includes the combined KY14, KY15, and KY19 lost the ability to bind CaM (Fig. 5D).
Thus, it appears that there is a specific Ca 2ϩ -CaM-binding site on the C terminus covered by these peptides, but the affinity of this site for CaM is weak.
We tested the functional importance of this CaM-binding site by generating deletion mutants of the trunc-␤-subunit lacking the region corresponding to KY13, KY19, or the combined re-gion of KY14, KY15, and KY19. When co-expressed with hRCNC␣, the mutant lacking KY13 did not seem to express well, as suggested by the lack of blockage of the cGMP-activated current by L-cis-diltiazem (data not shown). However, when the mutants lacking KY19 (MG2) or KY14, KY15, and KY19 together (MG8) were co-expressed with hRCNC␣, the resulting cGMP-activated current was still inhibited by CaM to a similar extent as the wild-type (Fig. 6). Thus, the CaMbinding site identified in the region spanning KY14, KY15, and KY19 on the C terminus does not appear to be necessary for the modulation of the heteromeric channel by CaM, but the possibility cannot be excluded that it contributes to the high sensitivity of the rod channel heteromeric channel to CaM, as observed in the electrophysiological experiments.
The N Terminus of the ␤-Subunit Contains a Region Crucial for the CaM-mediated Inhibition-Because the experiments on the C terminus indicated that not every CaM-binding site is necessarily involved in the modulation of the rod channel, we adopted a functional assay to study the N terminus. A number of deletion mutants were generated spanning consecutive regions of the N terminus of the trunc-␤-subunit (Fig. 7A) and were co-expressed individually with hRCNC␣. To test for a loss of CaM-modulation of the heteromeric channel expressed in HEK 293 cells, we performed electrophysiological recordings. Among these, we found that only mutants MG4 and BA104, which lacked a region close to the first transmembrane domain, had lost the inhibitory effect (Fig. 7B). This region is also rich in hydrophobic and basic residues (Fig. 8A). Experiments on the full-␤-subunit lacking the region corresponding to MG4 gave the same result. Sequence scanning suggests that the region deleted in MG17 (see Fig. 7A) may also contain a CaMbinding site, but this deletion mutant retained the CaM-mediated inhibition (Fig. 7B).
A fusion protein of the trunc-␤-subunit N terminus with a deletion corresponding to MG4 likewise lost its binding to CaM (Fig. 8B). Two peptides, MEG1 and MEG2, covering the deleted region were synthesized and tested for their ability to bind CaM. In a gel shift assay, both peptides appeared to bind CaM, but again very weakly (Fig. 8C). As controls, we examined several other peptides in the same experiment. One peptide, KY9, corresponds to the CaM-binding site on the N terminus of the ␣-subunit of the olfactory cyclic nucleotide-gated channel and binds Ca 2ϩ -CaM with nanomolar affinity (22). It retarded the migration of CaM completely at a peptide:CaM ratio of 2:1. Another peptide, RH106, corresponds to one of the CaM-binding sites on a N-methyl-D-aspartate receptor channel subunit (NR1) and has previously been shown to bind CaM with a K d of lower than 100 nM (38). RH106, likewise, shifted CaM according to the published results. In experiments with dansyl-CaM, the measured K d for both MEG1 and MEG2 was in the range of 3-10 M, compared with a K d of approximately 30 nM for RH106, which is consistent with the published value. A third control peptide, covering a region in the N terminus of the olfactory channel ␣-subunit and not able to bind CaM (KY8; see Ref. 22), neither shifted the CaM band in a gel shift assay nor showed any binding to dansyl-CaM at up to micromolar concentrations (data not shown). Thus, the CaM binding ability of MEG1 and MEG2 seems genuine, albeit weak. This CaM binding was Ca 2ϩ -dependent.
These results indicate that a region on the N terminus of the ␤-subunit is crucial for the CaM-mediated inhibition of the human rod channel. When this region is removed, the ability of the N-terminal fusion protein to bind CaM disappears, as does the CaM-modulation of the heteromeric channel. However, as determined with fusion proteins and peptides, the Ca 2ϩ -CaMaffinity for this binding site is surprisingly low, being almost 1000-fold lower than the apparent affinity measured in the electrophysiological experiments.

DISCUSSION
Based on Ca 2ϩ -flux measurements, others have shown that in the presence of Ca 2ϩ , CaM inhibits the native rod cGMPgated channel by shifting the dependence of its activation on cGMP to higher cGMP concentrations; the half-maximal inhibition of the current at low cGMP concentrations occurs at 1-2 nM CaM (10,11,13). Our electrophysiological recordings described here, on excised, inside-out patches of HEK 293 cells transfected with the cloned human cGMP-activated channel ␣and ␤-subunits, have led to a similar conclusion, giving a halfmaximal inhibition at approximately 4 nM CaM. These values suggest a high apparent affinity of Ca 2ϩ -CaM for the rod chan- nel. By contrast, our CaM overlay experiments on human retinal lysates and the heterologously expressed human ␤-subunit, which is known to confer the CaM effect (Ref. 17; see also Ref. 18), have failed to detect any obvious binding to CaM. The lack of CaM binding to human retinal lysates could have been due to the limited amount of tissue in the experiments or to the degradation of the tissue, which was obtained several hours post-mortem. However, these explanations cannot account for the lack of CaM binding to the heterologously expressed ␤-subunit, which we were able to harvest from transfected cells in higher quantity and with presumably little degradation. Previously, iodinated ( 125 I) CaM was used as a probe in gel overlay experiments for detecting CaM binding to the native bovine rod channel ␤-subunit (10,11) and to the heterologously expressed rat olfactory channel ␣-subunit (22), whereas biotinylated CaM and CaM-coupled alkaline phosphatase were used in our experiments. Because the latter two reagents have been shown to be as sensitive as, or more sensitive than, iodinated CaM in overlay assays on hair bundles of cochlear hair cells (31), it seems unlikely that our negative binding results have resulted from insensitive probes. More probably, insufficient renaturation of the human ␤-subunit under the gel overlay conditions led to undetectable CaM binding.
In experiments with the GST-fusion proteins of the trunc-␤subunit N and C termini, we did detect CaM binding, but the signals were very weak when compared with that for the Nterminal fusion protein of the olfactory cyclic nucleotide-gated channel ␣-subunit, which has a high affinity site for CaM (22). These weak affinities with fusion proteins were consistent with the results from gel shift and fluorescence experiments with synthetic peptides corresponding to the binding sites on the two termini, with the fluorescence experiments giving dissociation constants in the micromolar range. The low affinity of the identified C-terminal CaM-binding site may be genuine, especially considering that deletion mutants of the ␤-subunit lacking this region still conferred CaM-modulation to the ␣/␤-het- The experiments with 100 nM dansyl-CaM had the drawback of giving weaker fluorescence signals, but they gave more reliable K d measurements for the RH106 peptide, which had a higher CaM affinity. No difference between using 100 or 300 nM dansyl-CaM was observed for MEG1 and MEG2. eromeric channel complex. The low affinity of the N-terminal CaM-binding site is more surprising, however, because from deletion studies this is the site that is crucial for the CaMmodulation of the channel. Its affinity for CaM (K d of 3-10 M) as measured with synthetic peptides is approximately 1000fold lower than the half-maximal inhibition constant of 4 nM CaM derived from the electrophysiological experiments. From previous work on the olfactory channel, it is clear that CaM affects the gating of the channel (22). Assuming that CaM acts on the rod channel with a similar mechanism (see below), the apparent affinity of CaM for the intact channel protein can, in principle, be enhanced by the coupling between the ligandbinding and channel-gating steps. However, because of the relatively weak modulation of the rod channel by CaM (approximately 3-fold current reduction at low cyclic nucleotide concentrations; see "Results") compared with that for the olfactory channel (over 100-fold current reduction by CaM; see Refs. 21 and 22), it seems unlikely that this binding-gating coupling can lead to a 1000-fold increase in the apparent affinity for CaM. Moreover, despite the strong modulation of the olfactory channel by CaM, the CaM affinity for the binding site on this channel as measured with peptide-binding assays is not any lower than that measured with the electrophysiological experiments on the current inhibition (see below). There are several possible explanations. One is that the N-terminal peptides were structurally unable to reproduce the binding properties of the whole protein. Even though the peptides we synthesized are over 20 amino acid residues in length, they may still be not long enough to reproduce the native binding site. For the adenylyl cyclase in the bacterium Bordetella pertussis, for example, it was found that an increase in the length of a peptide spanning the CaM-binding site beyond 43 residues continues to increase its affinity for CaM (39,40). Another possibility is that more than one ␤-subunit is present in the ␣/␤-heteromeric channel complex and that CaM binds to them with positive cooperativity. Based on Ca 2ϩ -flux measurements, Bauer (13) suggested that two CaM molecules bind to a native rod channel complex. In our experiments on the heterologously expressed channel, on the other hand, the relationship between current inhibition and CaM concentration had a Hill coefficient near 1 (see Fig. 3), which suggests that perhaps only one CaM molecule binds to the channel complex. Because the two studies employed different preparations and different measurements, a strict comparison between them may not be meaningful. Nonetheless, considering that there is no known precedence for cooperative binding of CaM to a target protein, such a scenario is probably unlikely in our situation. A third, and more realistic, possibility is that even though the identified domain on the N terminus is a bona fide CaM-binding site, other regions on the channel protein facilitate or stabilize this binding through allosteric, electrostatic, or hydrophobic interactions. In the skeletal muscle phosphorylase kinase, for example, it has been suggested that two physically separate domains interact simultaneously with a single CaM molecule (41,42). This situation can be considered to be a more extreme variation of the first possibility mentioned above. In principle, other domains on the rod channel ␤-subunit participating in the interaction with CaM can be somewhere on the N terminus, as far away as in the C terminus, or even located in other subunits of the oligomeric channel. In this respect, the rod channel differs from the olfactory channel. Half-maximal inhibition of the current through the olfactory channel at low ligand concentrations occurs nominally at 4 -5 nM (this being calculated as the square root of the value of 21 nM CaM in Fig. 3 of Ref. 21, based on a Hill coefficient of 2 for the activation of the channel by cGMP), which matches quite well the K d of 3-4 nM for the peptide corresponding to the CaM-binding site on the ␣-subunit of the channel (22). Thus, for the olfactory channel, other regions besides the binding site may be minimally involved in interacting with CaM.
Despite the apparent difference in CaM-binding characteristics between the rod and olfactory channels, it is interesting to note that the functionally important CaM-binding site is situated at about the same location on both channels, both being a short distance N-terminal of the beginning of the first putative transmembrane domain (Fig. 9). Nonetheless, the binding site for the rod channel is situated on the ␤-subunit, but on the ␣-subunit for the olfactory channel. In comparison, the ␣-subunit of the rod channel neither binds CaM (Ref. 10; see also Ref. 22) nor is modulated by CaM (17). As for the ␤-subunit of the olfactory channel, it also does not confer any CaM modulation to the heteromeric channel complex when co-expressed, for example, with the rod channel ␣-subunit (43), although whether it binds CaM has not yet been studied. Finally, it should be mentioned that the ␣-subunit of the cone channel also has a high affinity CaM-binding site in a homologous position on its N terminus (44); 2 surprisingly, however, there is no CaM modulation at least of the homomeric channel formed by this subunit (Ref. 44; see also Ref. 45). 2 Taken together, these observations suggest an evolutionary path in which the ancestral channel that gave rise to the various ␣and ␤-subunits of the cyclic nucleotide-gated channels might already have a CaM-binding site-like domain on its N terminus. As the various subunit species arose over time, the CaM-binding capacity of the domain evolved as well, becoming either stronger or weaker in the different subunits. In parallel, other regions of the subunits evolved independently, such that a functional modulation by CaM occurs only when a CaM-binding site and other interacting regions on the channel are simultaneously present (as is the case with the olfactory channel ␣-subunit; see below). A sequence alignment of the CaM-binding sites on the olfactory channel ␣-subunit and the rod channel ␤-subunit is shown in Fig. 9. The CaM-binding site on the olfactory channel ␣-subunit conforms well to both the 1-8-14 and 1-5-10 motifs characteristic of many CaM-binding sites, where the numbers indicate the positions of key aromatic or long-chain aliphatic residues separated by other residues, including some positively charged ones (36). At the same time, the binding site shows a basic amphiphilic structure in a Kyte-Doolittle plot, with a net charge of ϩ2 (46,47). Both the hydrophobic and positively charged residues are thought to interact with CaM (48). The CaM-binding site on the rod channel ␤-subunit, on the other hand, shows only partial resemblance to these motifs. example, although the residues in positions 1, 5 and 8 are aromatic or hydrophobic in nature, residue 14 is not (although residue 15 is). At the same time, the sequence has a net charge of only ϩ1. Finally, a Kyte-Doolittle plot of the sequence does not show obvious amphiphilicity. These latter features may account for the relatively weak CaM affinity found for the peptides and fusion proteins.
The detailed mechanism by which CaM modulates the rod channel remains unclear. For the olfactory channel, CaM binds to a domain on the N terminus of the ␣-subunit that, in the absence of CaM, promotes the open state of the channel (Ref. 22; see also Refs. 15 and 49). When this domain is deleted, or when CaM binds to it, the influence of this domain in promoting the open state of the channel is lost, consequently leading to an inhibition of the current (22). Most recently, co-immunoprecipitation experiments have indicated that the N and C termini indeed directly interact with each other, and this interaction is disrupted by the binding of CaM to the N terminus of the channel (Ref. 23; see also Ref. 50). Thus, it appears that this N-terminal-C-terminal interaction promotes the opening of the channel. As for the rod channel, because the modulation by CaM is relatively weak (see above), we have not, unfortunately, been able to conclusively demonstrate whether or not a deletion mutant of its ␤-subunit lacking the CaM-binding site (MG4) behaves like the wild-type with CaM bound, that is, with a shift of the cGMP dose-response relationship to higher cGMP concentrations (data not shown). By analogy to the olfactory channel, however, the mechanism may still be similar.