activating only between 10% and 35% of the maximal currents induced by cGMP. Unlike ligand-gated neurotransmitter receptors, CNG channels do not desensitize in the continuous presence of ligand. This feature allows CNG channel activity to faithfully track the cGMP concentration in photoreceptors. Closer inspection of the dose-response relation suggests that CNG channels bind multiple cGMP molecules in a cooperative manner. Specifically, double-logarithmic plots of the activation (log I/Imax vs. log [cGMP]) display a limiting slope of up to about 3.5. This suggests that channel opening requires the binding of 3–4 cGMP molecules [27]. The highly cooperative activation maximizes the sensitivity of CNG channels, and therefore the sensitivity of the photoreceptor dark current, to small changes in the free cGMP concentration.

TRANSMEMBRANE TOPOLOGY AND FUNCTIONAL DOMAINS

The transmembrane topology of CNG channel A subunits is derived from immunogold labeling and electron microscopy of rod photoreceptors [28, 29], and it is supported by the results of a gene fusion approach using enzyme reporters [30]. According to these studies, both the N- and the C-terminus of A subunits are cytoplasmic, and the segment connecting S5 to the p region is extracellular. Based on sequence similarities, it is assumed that CNG channel A and B subunits adopt a similar transmembrane topology. Several key properties of CNG channels are attributed to specific domains of the channel proteins. These domains deserve a closer look.

The Cyclic-Nucleotide-Binding Domain

The C-terminal cytoplasmic region of all CNG channel subunits harbors a cNMPbinding site comprised of 80–100 amino acid residues. The three-dimensional structure of the catabolite activator protein (CAP) of Escherichia coli has been used for molecular modeling of the cNMP-binding site in CNG channels (for a detailed discussion, see [31]). The cAMP-binding site of CAP is comprised of three α-helices (A, B, and C) and eight β-strands (β1–β8). The β-strands form a flattened β-roll consisting of two antiparallel β-sheets, each with four strands. The A helix connects to the β-roll, followed by the B and C helix. The ribose and cyclic phosphate interact exclusively with the β-roll, while the purine ring interacts with residues in the β-roll and the C helix.

The molecular modeling of the cNMP-binding site in CNG channels predicts ten different interactions between cGMP and the binding pocket of CNGA1 or CNGA3, accomplished by eight different amino acids (Fig. 2B,C). It is thought that interactions with the common ribofuranose moiety are similar for all nucleotides, whereas different interactions with different purine rings control ligand selectivity. In particular, the residues T560 and D604 (in bovine CNGA1) are postulated to be key residues for ligand discrimination. Residue T560, located in β7, is expected to form a hydrogen bond with the amino group of cGMP, probably mediated by a water molecule. In contrast, no interaction is expected to occur between T560 and the purine moiety of cAMP, thus lowering the affinity for cAMP. Consistent with this prediction, the substitution of threonine for alanine dramatically decreases the cGMP sensitivity but has little effect on the cAMP sensitivity. The carboxylate side chain of D604, located in the C helix, presumably shares a single hydrogen bond with either N1 or the amino group in cGMP, while repulsive

forces are expected with the free electron pair at N1 of cAMP. In agreement with this prediction, the ability of cAMP to activate CNGA1 is enhanced if D604 is replaced with a neutral amino acid (D604Q or D604M), at low pH when D604 is protonated, or when the channel includes CNGB1, which carries a neutral asparagine at the position equivalent to D604. According to the model, T560 and D604 are unlikely to interact simultaneously with the purine ring of cGMP since the two interactions require different conformations of the cyclic nucleotide [32]. It is been speculated that cGMP initially binds in the syn conformation to interact with T560; a switch to the anti conformation then allows the interaction with D604. Moreover, it is also thought that the interaction with D604 initiates a movement of the C helix toward the β-barrel, which is essential for channel gating [33].

The Amino Terminal Domain and Modulation by Calmodulin

Deletion studies demonstrated that an unconventional calmodulin-binding site in the N-terminal domain of the β′ part of CNGB1 is essential for the Ca2+/calmodulin sensitivity of rod CNG channels [3, 34]. The detailed mechanism of how binding of Ca2+/calmodulin to this site decreases ligand sensitivity remains unknown. Nevertheless, it appears that disruption of an interaction between CNGB1 and the C-terminal region of CNGA1 is a critical element [34]. Both subunits of cone CNG channels contain conserved calmodulinbinding motifs in their N-termini, and CNGB3 contains an additional calmodulin-binding motif in the C-terminal region (e.g., [35]). Heteromeric cone CNG channels remain sensitive to modulation by Ca2+/calmodulin when either one or the other of the two binding motifs in CNGB3 is deleted but lose sensitivity on deletion of both motifs.

The P Region

The p region of CNG channels shows high similarity to the p region of voltage-gated K+-selective channels, although CNG channels discriminate poorly between monovalent cations [36, 37]. Therefore, not surprisingly, the sequence similarity between the p regions in both channel types ends at the selectivity filter. The crystal structure of KcsA, a bacterial K+-selective channel, provided detailed insight into how selectivity is achieved. In KcsA, the p region consists of an α-helix of about 15 amino acids, followed by a loop of 6 amino acids that forms the selectivity filter in the narrowest part of the pore [22]. It is thought that the gross structure of the pore is conserved in CNG channels, but the structural detail around the selectivity filter is expected to be different. While the selectivity filter in K+ channels comprises only neutral amino acids, CNGA1 and CNGA3 carry one negatively charged glutamate residue in the pore loop. Apparently, pairs of pore glutamates from different subunits interact by sharing a single proton rather than repel each other [38].

Insight into the physiological role of the pore glutamates comes from studies of Ca2+ permeation. Increasing extracellular Ca2+ concentrations progressively impedes the current of monovalent ions through CNG channels (Fig. 3A). At physiological Ca2+ concentrations, CNG channel currents are reduced to a few percent of the maximal value found under Ca2+-free conditions. For example, the single-channel conductance of rod CNG channels drops from about 25 pS in a Ca2+-free saline to less than 1 pS in the presence of 1 mM extracellular Ca2+ [39, 40].

Fig. 3. Blockage of cyclic nucleotide-gated (CNG) channels by extracellular calcium. A Increasing the extracellular Ca2+ concentration successively impedes currents through homomeric CNGA1 (black dashed line) and CNGA3 (gray dashed line) channels as well as currents through heteromeric rod (CNGA1/CNGB1; black solid line) and cone (CNGA3/CNGB3; gray solid line) channels. Data were obtained on heterologous expression of CNG channel subunits at a membrane voltage of −60 mV [13, 26]. B Each subunit in rod (left) and cone (right) CNG channels contains an α-helix followed by a loop in the pore-forming region. The pore loop in CNG channel A subunits contains a glutamate that participates in the formation of the ion-binding site within the pore (E). B subunits carry a glycine (G) instead of glutamate in the pore loop.

The effect of Ca2+, and the similar effect of Mg2+, on CNG channel currents can be explained by a model in which permeating monovalent and divalent ions compete for a common cation-binding site within the channel pore. Since divalent ions are bound more strongly by the binding site, they occlude the permeation pathway for monovalent ions. The fact the Ca2+ blockage can be described by a simple binding isotherm demonstrates the presence of a single Ca2+-binding site within the pore [41]. The Ca2+-binding site is apparently formed by the pore glutamates of CNGA1 and CNGA3 since the replacement of glutamate by the neutral residue glutamine dramatically reduces the Ca2+ blockage [42].

Homomeric channels composed only of CNGA1 generally display a higher sensitivity for Ca2+ blockage than those composed only of CNGA3. An extensive mutagenesis study showed that all aspects of Ca2+ blockage can be transferred from one channel to another by swapping the p region and the two adjacent transmembrane segments (S5–p region–S6) of the channel-forming subunits. Therefore, the S5–p region–S6 segment is considered the basic pore module that governs both blockage and ion permeation in CNG channels [41].

CNGB1 and CNGB3 do not carry a negatively charged pore loop glutamate but instead have a neutral glycine. Thus, native rod channels have a total of three pore glutamates, and native cone channels presumably a total of two pore glutamates (Fig. 3B). The fact that the pore affinity for Ca2+ is reduced in channels with incorporated B subunits is likely to be due to the reduction in the pore’s charge density [31]. The high affinity of the CNG channel pore for Ca2+ has two important consequences for the physiology of photoreceptors. First, it ensures that the pore of CNG channels is preferentially occupied with Ca2+ rather than monovalent ions. This allows a relative Ca2+ flux of 10–25%, even though Ca2+ constitutes less than 2% of the extracellular cations. Second, Ca2+ blockage

reduces the contribution of each single channel to the total dark current, making it less sensitive to single-channel current fluctuations and thereby increasing the signal-to- noise ratio of photoreceptor currents. This aspect is of special importance for rod photoreceptors since they act as high-sensitivity photon detectors.

Interestingly, amino acid substitutions in or near the pore glutamate not only change the permeation properties but also the gating behavior of CNG channels (e.g., [26]). This observation indicates that conformational rearrangements of the pore around the selectivity filter are also involved in channel gating.

The GARP Domain of CNGB1

CNGB1 is unique among CNG channel subunits because it contains a large N-terminal cytoplasmic domain that is almost identical to two soluble rod-specific GARPs. The GARPs are alternatively spliced forms of a single gene different from that encoding CNGB1. The most conserved structural elements in GARPs, four short proline-rich repeats of about 15 amino acids in their N-terminus, were used as bait to test for interactions of CNGB1 with other rod proteins using peptide affinity chromatography [43]. This study found that the proline-rich repeats from GARPs interact with PDE, GC, and the retina-specific ATP-bind- ing cassette receptor (ABCR) transporter. The ABCR transporter is also known as the rim protein as it is distributed along the rim region of rod discs.

The interaction between GARPs and ABCR and GC has been called into question by the results of a second approach to identify GARP-interacting proteins [44]. In immunoprecipitation experiments, neither ABCR nor GC was pulled down with GARP-specific antibodies, even when GARP-binding partners were covalently attached using cross-linking reagents. Instead, peripherin, another protein located at the disk rim, was shown to interact with the shorter GARP splice variant and the GARP domain of CNGB1 (Fig. 4). The hydrodynamic properties of GARPs and little secondary and tertiary structure are consistent with an elongated, unfolded GARP domain of CNGB1 that is able to span the 10-nm gap between the rod plasma membrane and the disk rim [45]. The interaction between CNGB1 and proteins at the disk rim would align the CNG channels in stacked circles along the outer segment. Indeed, a nonuniform distribution of CNG channels in the outer segment of rod photoreceptors has been reported [46]. This suggests that the interaction between CNGB1 and disk rim proteins supports the flat appearance and the arrangement of membrane disks in the outer segment; however, mice lacking CNGB1 form rods of normal morphology [47].

Due to the high density of glutamate residues in the GARP domain of CNGB1 (137 of 571 residues in bovine CNGB1), this domain functions as a low-affinity Ca2+ buffer. It has been speculated that the GARP domain of CNGB1 guides Ca2+ ions from the intracellular CNG channel pore to the disk surface, the locale of the Ca2+-binding proteins that regulate the sensitivity and the kinetics of the photoresponse [45].

CNG CHANNELS ARE COMPONENTS OF LARGER

PROTEIN COMPLEXES

The interaction of peripherin and the GARP domain of CNGB1 argue against a model in which CNG channels diffuse freely and are isolated in the photoreceptor membrane. In fact, further biochemical and pharmacological data suggest that CNG channels of