The heteromeric cyclic nucleotide-gated channel adopts a 3A:1B stoichiometry

Distinct A subunits (or α-subunits) of vertebrate CNG channels, named CNGA1–4 according to the latest nomenclature⁵, mediate rod phototransduction (A1), cone phototransduction (A3) and olfactory transduction (A2 and A4). Two B subunits (orβ-subunits) are known, named CNGB1 and CNGB3 (CNGB2 does not exist⁵), with CNGB1 being part of native rod and olfactory channels and CNGB3 part of cone channels. When expressed heterologously, most A subunits (CNGA1–3) form functional homomeric channels^1–4; in contrast, B subunits do not, but they form functional heteromers with A subunits. Homologues of these A and B subunits have been identified in Drosophila⁹ and Caenorhabditis elegans^10,11.

While examining possible inter-domain interactions of the human (h)CNGA3 subunit, we found that its cytoplasmic C terminus, designated A3-C, interacted with itself. Thus, when co-expressed in HEK 293T cells, haemagglutinin (HA)-tagged A3-C could be co-immunoprecipitated with Myc-tagged A3-C by an anti-Myc antibody (Fig. 1a). This homotypic interaction was confirmed by an in vitro binding assay in which purified recombinant glutathione S-transferase (GST)-tagged and His-tagged A3-C were found to interact directly with each other (Fig. 1b). Furthermore, when 293T cells transfected with Myc-tagged A3-C were metabolically labelled and lysed, a single radioactive protein species with the expected molecular mass was immunoprecipitated by the anti-Myc antibody (Fig. 1c), confirming that no other proteins mediated the homotypic interaction. Finally, similar homotypic interactions were seen with C-terminal constructs of CNGA1 (bovine) and CNGA2 (rat) (Fig. 1d). The C terminus of CNGB1 (human) showed only very weak interaction (data not shown), and that of CNGB3 (mouse) showed no detectable interaction (Fig. 2b, right gel).

Sequence analysis identified a 22-residue leucine-zipper-like sequence (underlined in Fig. 2a) in the C terminus of hCNGA3, downstream of the cyclic-nucleotide-binding site. Deleting these residues abolished the homotypic interaction of a molecule comprising the last 114 residues of A3-C (Fig. 2b, compare A3C-28/A3C-28 and A3C-32/A3C-32 lanes). Further deletion and truncation analyses (Fig. 2b, see A3C-71/A3C-71; also data not shown) revealed that the full interaction domain covers the 22-amino-acid sequence plus 25 residues immediately downstream. This domain, which we name CLZ, is present in all examined CNG channel A subunits (Fig. 2a) but not in the B subunits. It consists of two leucine-zipper-like coiled-coils connected out of phase by a hydrophilic linker. At least several conserved leucine or hydrophobic residues (marked by asterisks in Fig. 2a) are important: when any of them was mutated into alanine, the homotypic interaction was undetectable in the co-immunoprecipitation assay (data not shown). Finally, the CLZ domain was able to confer homotypic interaction on the C terminus of CNGB3 when inserted after the cyclic nucleotide-binding site (Fig. 2b, right gel).

Despite the importance of the CLZ domain, its disruption only weakened, but did not abolish, the interaction of the complete A3-C, in that the construct A3C-ΔZ (Fig. 2b, see A3C-ΔZ/A3C-ΔZ lane) still showed a mild interaction. Further experiments suggested that a region upstream of the CLZ domain also interacts homotypically, but the CLZ-mediated interaction is much stronger (data not shown). From here on, we shall focus on the CLZ domain.

In a gel-filtration experiment, a C-terminal fragment (A3C-28; see Fig. 2b) of hCNGA3 having the CLZ domain as the only interacting motif was found to be fractionated into a peak with an apparent molecular mass (estimated from its Stoke's radius) 3.5-fold that of the monomer peak (Fig. 3a; see Methods). Because the heterotetrameric CNG channel contains at least one B subunit, which does not contain the CLZ domain, this result suggests that the fragment probably forms a trimer rather than a tetramer. Indeed, a similar experiment with a slightly larger C-terminal fragment of the channel (T565-End) gave a value of 3.1 (data not shown).

Because the gel-filtration result could have been confounded by molecular shape, we did a cross-linking experiment on A3C-28 with dithiobis-succinimidyl propionate (DSP), a cross-linker cleavable by reducing reagents. Partial cross-linking produced three bands in SDS-PAGE under non-reducing conditions, but only one band in the presence of dithiothreitol (Fig. 3b). These bands presumably represented the monomer, dimer and trimer, respectively. More complete cross-linking shifted the product mostly to the highest (trimer) band. This cross-linking was CLZ-dependent because, when this domain was disrupted (A3C-32 in Fig. 3b), only the lowest (monomer) band was observed. Similar results were obtained for the corresponding C-terminal constructs of hCNGA1 and rat CNGA2 (Fig. 3c), confirming generality among A subunits.

To be definitive, a native-gel approach was used. A3C-28 migrated on a native gel primarily as multimers, as supported by the observation that a point mutant (A3C-87: L633A; Fig. 3d, upper gel) with a disrupted homotypic interaction migrated much more rapidly than the wild type. We constructed two other mutants of A3C-28, retaining the homotypic interaction but with extra negative (A3C-88: K596E, R603E) or positive (A3C-89: D613K, E615K) charges. As expected, A3C-88 had a much higher mobility than A3C-89 in native-gel electrophoresis (Fig. 3d, upper gel). When co-expressed, these two mutants gave four bands (Fig. 3d, upper gel), consistent with the combinatorial possibilities, P₃, P₂Q, PQ₂ and Q₃, from two proteins P and Q that form trimers. The four bands became particularly evident when the transfected complementary DNA ratio for A3C-88 and A3C-89 was manipulated (Fig, 3d, lower gel).

The final evidence came from analytical centrifugation, a method of determining the molecular mass independently of shape. A 48-residue peptide (Lys 624–Gly 671) corresponding to the CLZ domain was synthesized and analysed by equilibrium sedimentation. The resulting data were best fitted if the peptide formed a trimer, rather than a dimer or tetramer, in solution (Fig. 3e).

To examine whether the CLZ domain forms a trimer in the context of the complete channel protein, we replaced this domain in hCNGA3 with well-characterized leucine zippers¹² that form dimers, trimers and tetramers, respectively. All mutants readily formed homomeric channels when expressed alone (Fig. 4a). However, the mutants harbouring dimeric and tetrameric leucine zippers consistently differed, although moderately, from the wild-type channel in the K_1/2 value of the dose-response relation (Fig. 4b) and in the saturated cAMP-activated/saturated cGMP-activated current ratio¹³ (Fig. 4c). In contrast, the mutant harbouring a trimeric leucine zipper mimicked the wild type fairly well in both properties. Furthermore, like the wild type, the trimeric mutant gave comparable currents in the absence and presence of the B subunit, but not the dimeric and tetrameric mutants (Fig. 4d). Thus, the CLZ domain was best emulated by a generic trimeric leucine zipper in functional CNG channels.

The fact that the trimeric CLZ domain is present only in the A subunits but not the B subunits seems to leave only one natural stoichiometry, 3A:1B, for the hetero-tetrameric CNG channel. To examine this hypothesis directly, we purified the native CNG channel from bovine rod photoreceptor membranes by affinity chromatography and determined the relative amounts of A and B subunits in the channel complex. The native rod channel consists of a truncated A subunit (CNGA1) and a large B subunit (CNGB1) that migrate on SDS gels as polypeptides of apparent molecular masses 68,000 and 240,000 (M_r 68K and 240K), respectively^14,15 (Fig. 5a). We quantified the relative tryptophan contents in these bands from the fluorescence arising from ultraviolet-induced trichloroethanol-modified residues¹⁶ (Fig. 5b). The ratio of tryptophan signals for the A:B subunits was 1.5 ± 0.4 (mean ± s.d.) (n = 10), in close agreement with the predicted ratio of 1.3 for a 3A:1B stoichiometry but not with the ratio of 0.4 for a 2A:2B stoichiometry (the truncated A subunit has 9 tryptophan residues; the B subunit has 21). We also determined the relative cysteine contents of the A and B subunits labelled by the fluorescent Oregon Green maleimide (Fig. 5c). The observed modified cysteine ratio of 1.8 ± 0.5 (n = 3) for A:B bands was also closer to the predicted ratio of 1.2 for a 3A:1B stoichiometry than the ratio of 0.4 for a 2A:2B stoichiometry (the A subunit has 6 cysteine residues; the B subunit has 15). The higher than expected value observed in our experiments might arise from the limited accessibility of some cysteine residues in the GARP (glutamic-acid-rich protein) part¹⁵ of the B subunit to the bulky Oregon Green maleimide compound. Finally, we examined whether the channel subunits were degraded during purification. Only intact A and B subunits were detected on western blots labelled with a variety of monoclonal antibodies against different epitopes on these subunits (see Fig. 5d for an example). The absence of degraded fragments indicates that no significant proteolysis of the channel had occurred. Together, our data support a stoichiometry of 3A:1B for the native rod CNG channel.

Our results refute the current belief that the CNG channel adopts a 2A:2B composition, which was largely supported by two studies on the rod channel^6,7. One study⁶ relied on a Ni²⁺ potentiation of the opening of homomeric bovine (b)CNGA1 channels by coordinating corresponding histidine residues (His 420) of two adjacent subunits¹⁷; the corresponding residue in bCNGB1 is asparagine (Asn 546). These authors found that Ni²⁺ potentiated the heteromeric channel whether formed by wild-type bCNGA1 and bCNGB1 or by reciprocally mutated bCNGA1 (H420Q) and bCNGB1 (N546H), concluding that there was an AABB stoichiometry/ordering. However, their implicit assumption was that Ni²⁺ acted only on the histidine residue in question, whereas others⁷ have reported that a mild potentiation by Ni²⁺ persisted in the heteromeric channel formed by wild-type bCNGB1 and mutant bCNGA1 (H420Q). The second study⁷ arrived at an ABAB stoichiometry/ordering based on expressions and co-expressions of the tandem dimers A–A, A–B, B–A and B–B. This conclusion assumed that only the leading subunit, not the trailing subunit, of a tandem dimer could incorporate singly into a functional channel. Otherwise, the data would be consistent with a 3A:1B stoichiometry.

In summary, our results suggest that the preferred stoichiometry of heteromeric CNG channels is 3A:1B, although it is impossible to rule out a small percentage of native CNG channels with a different stoichiometry. The olfactory CNG channel is composed of two A subunit species (CNGA2 and CNGA4) and one B subunit species (CNGB1b)^5,18–21. Because these A subunits both contain the CLZ domain, the native channel is expected to contain one B subunit together with two A2 and one A4, or one A2 and two A4, subunits (previous work on the stoichiometry of channels formed by co-expressed A2 and A4 is unfortunately not informative because the B subunit was absent²²). Our findings also provide perhaps the first example of a breakdown of symmetry in tetrameric channels. Previous studies based on homomeric CNG channels have suggested that the four CNG subunits in a channel complex are organized into two functional dimers^23,24. A three-dimensional structure of purified rod CNG channel at 35Å resolution²⁵ also suggests that the intracellular domains are arranged as a pair of dimers. Incorporating these findings, the linker between the cyclic-nucleotide-binding site and the CLZ domain might be flexible enough that, despite its asymmetric composition, the other parts of the CNG channel can still configure and function as a `dimer of dimers'.