A Cyclic Nucleotide Modulated Prokaryotic K⁺ Channel

INTRODUCTION

Ion channels gated by cyclic nucleotides are widely distributed throughout eukaryotes (Kaupp and Seifert, 2002; Robinson and Siegelbaum, 2003). As the molecular hardware linking changes of intracellular cyclic nucleotide levels occurring in sensory transduction to cellular electrical responses, these proteins are especially important in the coherent functioning of electrically excitable tissues. They have been most widely studied in sensory neurons, where they carry out essential tasks in the detection of photons or odorants, in cardiac myocytes, where they underlie pacemaking activity, and in cerebellar neurons, where they contribute to fine tuning in motor-coordination circuits. All such channels belong to the S4 superfamily (Yellen, 2002; Yu and Catterall, 2003), which also includes voltage-dependent Ca²⁺, Na⁺, and K⁺ channels; these proteins are tetramers of identical or similar subunits, or of honorary subunits inhabiting the same polypeptide in tandem, with each subunit having six transmembrane sequences and a pore-forming sequence positioned between the fifth and sixth of these.

The cyclic nucleotide-modulated channels additionally carry a cyclic nucleotide-binding domain (CNBD) in the cytoplasmic COOH-terminal region. These channels fall operationally into two broad subclasses, denoted CNG (cyclic nucleotide gated) and HCN (hyperpolarization and cyclic nucleotide activated). Channels of the first class depend absolutely on cyclic nucleotides for opening, gate in a voltage-independent manner, and permit conduction only by monovalent cations, with little ionic selectivity among the group IA cations. In contrast, HCN channels are merely modulated by cyclic nucleotides in their primary mode of voltage-dependent gating, and, while also conductive to monovalent cations, favor K⁺ over Na⁺ by a factor of approximately four. At the molecular level, both CNG and HCN channels show the arginine-decorated fourth transmembrane sequence characteristic of the S4 family, and both carry similar CNBD sequences. In keeping with their K⁺ selectivity, HCN channels carry the “GYG” pore-forming selectivity sequence unique and common to K⁺ channels (Heginbotham et al., 1992), while CNG channels lack this sequence.

The recent high-resolution structures of several K⁺ channels (Doyle et al., 1998; Jiang et al., 2002, 2003) provide an outline of molecular architecture in this protein family but leave many questions specific to the cyclic nucleotide-binding subfamily unaddressed. Are these, like K⁺ channels, built as fourfold symmetric structures, as in the structure of an isolated CNBD of an HCN channel (Zagotta et al., 2003), or as dimers-of-dimers, as suggested by functional experiments on CNG and HCN channels (Liu et al., 1998; Ulens and Siegelbaum, 2003)? What accounts for the hyperpolarization-dependent opening of HCN channels, a voltage dependence with polarity opposite to that of conventional K_v-type K⁺ channels (Mannikko et al., 2002)? How does the “C-linker,” the ∼80-residue sequence connecting the CNBD to the sixth transmembrane helix, mediate between cyclic nucleotide binding and channel opening (Gordon and Zagotta, 1995)? Why, despite their K⁺ selectivity sequence, are HCN channels so much less K⁺ selective than other K⁺ channels, which typically show 100–1,000-fold selectivity for K⁺ over Na⁺?

These questions will ultimately require structural answers. The only productive avenue toward high-resolution ion channel structure is currently provided by bacterial homologues, which in some cases allow overexpression in Escherichia coli of properly assembled proteins amenable to purification and crystallization. For this reason, we searched prokaryotic databases for S4-type ion channels containing cyclic nucleotide-binding domains. We now describe the overexpression, purification, and functional reconstitution of an S4-superfamily homologue from Mesorhizobium loti, with a K⁺ selectivity sequence and a CNBD. The channel's chromatographic behavior in detergent micelles and its ionic flux properties in liposomes respond to micromolar levels of cyclic nucleotides, and its selectivity for K⁺ is higher than that of either CNG or HCN subtypes.

MATERIALS AND METHODS

RESULTS

In a BLAST search for S4-type channels containing CNBDs, we noticed, among four other hits, a putative K⁺ channel homologue in the plant symbiont M. loti, an organism commonly used for genetic studies of rhizobial symbiosis and nitrogen fixation. The gene (GenBank/EMBL/DDBJ accession no. NP104392) encodes a 355-residue polypeptide of 38 kD with six putative transmembrane sequences (S1–S6), where S4 bears three basic residues (two arginines, one lysine), a K⁺ selectivity sequence (TTGYGD) between S5 and S6, and a COOH-terminal CNBD (Fig. 1). The pore sequence for this protein, which we call MloK1, is homologous to the corresponding region of other K⁺ channels such as KcsA (60% identity) and Shaker (50%), and its CNBD displays convincing similarity (∼30%) to those of eukaryotic CNG and HCN subtypes. Unlike its eukaryotic counterparts, MloK1 does not contain the four-helix C-linker domain, which connects the end of S6 to the CNBD and is thought to transmit the cyclic nucleotide binding signal to the channel's gate (Gordon and Zagotta, 1995; Zong et al., 1998; Paoletti et al., 1999; Johnson and Zagotta, 2001; Wang et al., 2001). In MloK1, the CNBD sequence begins only 13 residues after S6 emerges from the membrane.

The channel protein, engineered with a COOH-terminal His6 tag and placed behind a tetracycline promoter, is efficiently expressed in E. coli (Fig. 2), with an induction-dependent band of ∼35 kD visible on SDS-PAGE even in the whole-cell lysate. After extraction of the cells in detergent, high purity is achieved in a single pass over a Ni²⁺-chelate column with a yield of ∼3 mg/L culture. The chromatographic behavior of the affinity-purified protein on gel filtration is cyclic nucleotide dependent (Fig. 3). In the absence of cyclic nucleotide, much of the protein elutes as an aggregate in the void volume, with some tetrameric material (∼150 kD) following, along with several other peaks. However, with 50 μM cAMP present, most of the material runs as a single tetrameric peak, as though the protein population is monodisperse under this condition. Although gel filtration chromatography does not rigorously determine particle size, we are reasonably confident of our measurement here, since the column was calibrated with six membrane proteins of known size, run under identical detergent conditions. We note that these experiments were performed on two separate preparations processed in parallel in which the protein was extracted, purified, and chromatographed in the presence or absence of cAMP. cGMP supports this chromatographic propriety as well (unpublished data). This result implies that occupancy of the CNBD by ligand promotes native folding and perhaps stabilizes the channel during the purification procedure; we have found (unpublished data) that the tetrameric peak purified without cAMP and maintained in its absence remains tetrameric for at least 7 d at 4°C.

Does MloK1 function as a cyclic nucleotide-gated ion channel? We reconstituted the protein into liposomes and examined its ability to catalyze transmembrane movement of ⁸⁶Rb⁺, a convenient radioactive K⁺ analogue. Influx was followed under conditions wherein liposomes loaded with high KCl are suspended in low KCl solution containing radioactive tracer, a circumstance that leads to concentrative accumulation of radioactivity. Under these conditions, significant tracer uptake can occur only in liposomes that are permeable to K⁺ but not to Cl⁻ (Heginbotham et al., 1998). Fig. 4 A illustrates a typical influx time course, in which ⁸⁶Rb⁺ becomes trapped within the liposomes with approximately exponential time dependence characterized by a half-time of ∼30 min. However, even on the long, 2-h time scale of this experiment, the influx does not reach completion, and so the equilibrium level of trapped ⁸⁶Rb⁺ remains uncertain. We measure the maximum capacity of ⁸⁶Rb⁺ uptake by adding valinomycin, an ionophore which instantly permeabilizes all liposomes to K⁺ and Rb⁺. Thus, in this experiment, MloK1 catalyzes ⁸⁶Rb⁺ uptake into ∼35% of the liposome space in 2 h, but because of the slowness of the time course, we cannot claim that this represents the fraction of liposomes carrying functional MloK1; it does place a lower limit on this fraction, however. Henceforth we normalize influx time courses to the maximum uptake measured by valinomycin to compare results obtained from different reconstitution conditions, as in Fig. 4 B. Here, the effect of cAMP on influx is shown. Over the ∼1-h time course, a small background uptake occurs in protein-free liposomes. Reconstitution of MloK1 increases the initial rate of uptake, and a maximal concentration of cAMP (200 μM) approximately doubles this. These results show that MloK1-mediated ion transport is not absolutely dependent on cAMP, but that the ligand enhances the protein's intrinsic activity.

The rate of ⁸⁶Rb⁺ uptake increases with the density of channels in the reconstituted membranes (Fig. 5). At low protein density (<2 μg protein/mg lipid), uptake increases linearly, as liposomes in excess are titrated with channel protein. Eventually, uptake levels off at high protein density >5 μg/mg (Fig. 5 B), as channel protein increases in excess of the liposomes. Unfortunately, this effect cannot be quantified, as has been done by Poisson statistics in other liposome systems (Goldberg and Miller, 1991; Heginbotham et al., 1998; Maduke et al., 1999), since the final steady level of ⁸⁶Rb⁺ uptake cannot be accurately determined here; nonetheless, the value of the half-saturation density (∼1 μg/mg) indicates that a large fraction of the protein is functionally active, since it is at this density that the number of liposomes is roughly equal to the number of protein molecules present in the reconstitution mix (Heginbotham et al., 1998).

The appearance of a K⁺ selectivity sequence (TTGYGD) between S5 and S6 suggests strongly that MloK1 should show a familiar K⁺ selectivity profile, permitting permeation by K⁺ and Rb⁺ and effectively barring Na⁺ and Li⁺. To test this expectation, we adapted the concentrative uptake assay to assess ion selectivity, comparing ⁸⁶Rb⁺ uptake into liposomes loaded with different test cations in the presence of cAMP (Fig. 6). Uptake is similar in K⁺ and Rb⁺, while neither Na⁺ nor Li⁺ supports significant uptake. These experiments do not allow precise estimates of ionic permeability ratios, but it is clear that Rb⁺ influx rate is close to the K⁺ rate, and Na⁺ and Li⁺ permeate, if at all, at much lower rates.

Cyclic nucleotides bias the open–closed equilibrium of eukaryotic CNG and HCN channels toward open conformations. This behavior is mirrored in MloK1, although the maximum stimulatory effect of cAMP is only twofold (Fig. 4). To examine the ligand dependence of this effect, we reconstituted the channel into liposomes and used the uptake assay to determine dose–response curves for cAMP and cGMP. ⁸⁶Rb⁺ flux time courses in the presence of varying cyclic nucleotide concentrations are shown in Fig. 7 (A and B). Both cAMP and cGMP stimulate the uptake, and both give similar approximately twofold maximal stimulation. Dose–response curves were derived from these time courses by normalizing the uptake values measured at 15 or 45 min to the corresponding isochronal uptake at maximal ligand concentration. These follow a simple Langmuir-type concentration dependence (Hill coefficient of unity), with cAMP being ∼10-fold more effective than cGMP (K_1/2 = 60 and 600 nM, respectively).

The uptake experiments above demonstrate that MloK1 mediates membrane transport for Rb⁺ and K⁺ but do not distinguish between electrically conductive ion movement, as would be expected for a channel homologue, and an electroneutral mechanism. The preferred way of demonstrating a conductive mechanism is to reconstitute the protein into planar lipid bilayers and record electrical currents; however, we have been so far unable to do this. We can nevertheless address this question by an alternative method using liposomes. The method employs proton uptake to report on the electrical potential set up across the liposome membrane by a K⁺ gradient. The principle is illustrated in the cartoon of Fig. 8 A; liposomes loaded with high K⁺ are suspended in low K⁺ solution, and the pH of the suspension is recorded. A K⁺-selective conductance present in the membrane produces a negative-inside electrical potential that attracts protons into the intraliposomal space. The proton-impermeable liposome membrane does not allow protons to “follow” this potential, but treatment with a proton ionophore removes this kinetic constraint and allows protons to enter, as may be observed by a pH increase in the external solution. We performed this experiment first with protein-free liposomes (Fig. 8 B, top). Addition of the weak-acid uncoupler FCCP, which permeabilizes the membranes specifically to protons, has no effect on the pH of the suspension by itself. Addition of the K⁺-specific ionophore valinomycin polarizes the membrane and produces an immediate rise in pH, as protons are taken up against a pH gradient but down their electrical gradient. When this experiment is repeated on liposomes reconstituted with MloK1, however, FCCP causes prompt H⁺ influx without addition of valinomycin (Fig. 8 B, bottom). Thus, MloK1 substitutes functionally for valinomycin in setting up a negative internal membrane potential, and is therefore conductive, not electroneutral. The experiment further shows that MloK1 is selective for K⁺ over both Na⁺ and H⁺. Subsequent addition of valinomycin causes only a small additional pH increase, as though <25% of the liposomes in the population are free of functional MloK1 protein. These pH responses are completely abolished in the absence of a K⁺ gradient (unpublished data).

DISCUSSION

All structures of ion channels determined so far have been prokaryotic in origin, mainly because of the empirical and tragic fact that overexpression of eukaryotic integral membrane proteins in E. coli has been so reliably unsuccessful. For this reason, if we wish to know what a eukaryotic channel protein looks like, we are currently forced into the unsatisfactory but practical necessity of seeking bacterial or archeal homologues of the channel of interest. In such a search we found five putative S4-type ion channels of the cyclic nucleotide-binding subclass. In striking contrast to the abundance of other channel families among bacterial genomes, such as K_v and K_ir K⁺ channels and CLC Cl⁻ channels, these CNBD-containing homologues are rare, so far found only in five bacterial genomes, three of which are rhizobial symbionts. In no case is the biological function of any of these prokaryotic channels known.

We chose to focus on MloK1, which expresses well in E. coli and may be readily purified in quantity. To maintain the protein in a homogenous tetrameric state, it is necessary to keep cyclic nucleotide present throughout all steps of the purification procedure. The purified protein reconstituted into liposomes catalyzes ⁸⁶Rb⁺ uptake in a cyclic nucleotide-stimulated fashion. These two phenomena, along with the low concentration range in which the ligands act, imply that the COOH-terminal CNBD is properly folded and functionally competent. Furthermore, the tetrameric nature and K⁺-like ionic selectivity of conductive fluxes are as expected for a K⁺ channel. For these reasons, we consider that the purified MloK1 protein, in fulfilling the expectations of its sequence, is correctly folded and functional.

To which subtype of cyclic nucleotide binding channel does MloK1 belong: CNG or HCN? It is certainly not of the CNG subtype, since it is K⁺ selective and shows a K⁺ signature sequence. We are tempted to assign it to the HCN subfamily, on the basis of this K⁺ selectivity, as well as the substantial basal activity in the absence of cyclic nucleotides. But in fact we cannot definitively assert that the channel is of HCN subtype. Since we have not been able to record electrical currents from MloK1, we cannot gauge its voltage dependence, a fundamental defining characteristic of HCN channels. Moreover, the high selectivity of MloK1-mediated flux for K⁺ over Na⁺ would not be expected for a conventional HCN channel, as these are only weakly selective for K⁺. We are therefore unwilling to assign this channel firmly to either of the categories defined in eukaryotic channels.

Both cAMP and cGMP bind to this channel, as indicated by chromatographic behavior and ion fluxes. In the flux assay, cAMP shows ∼10-fold higher potency than cGMP, with the two ligands giving similar maximal effects. Several ligand-selectivity determinants have been identified in CNBDs from eukaryotic channels, from both electrophysiological analysis of CNG channels (Altenhofen et al., 1991; Varnum et al., 1995) and direct structure determination of an isolated CNBD from an HCN channel (Zagotta et al., 2003). All eukaryotic CNBDs, regardless of ligand preference, have a conserved arginine residue that coordinates the cyclic nucleotide phosphate (Zagotta et al., 2003), and this residue is also present in MloK1 (R307; Fig. 1). A threonine is always found adjacent to this arginine in eukaryotic channels that bind both cAMP and cGMP; in the structure of HCN2 CNBD, the hydroxyl group of this threonine hydrogen bonds to the N2 atom of the guanine ring, while cAMP-specific channel isoforms, such as the sea urchin HCN channel (Gauss et al., 1998), lack this hydroxyl group. In MloK1, this position uses a serine (S308), the hydroxyl of which may play an equivalent H-bonding role, permitting cGMP to bind. Finally, CNG channels that are highly selective for cGMP have an acidic residue closer to the COOH terminus (Fig. 1, asterisk), and neutralizing this residue removes the ligand selectivity (Varnum et al., 1995); MloK1 has an alanine residue here, in harmony with responsiveness to both cAMP and cGMP. On several grounds, therefore, the ligand dependence of MloK1 is consistent with expectations based on molecular determinants in its CNBD.

One feature of MloK1, however, is unexpected from precedent: the absence of a C-linker sequence, a motif found in all eukaryotic CNG and HCN channels. The C-linker domain has been proposed to transmit the ligand-binding signal to the channel's gate in S6. This proposal arose from mutagenesis studies in CNG and HCN channels (Gordon and Zagotta, 1995; Zong et al., 1998; Paoletti et al., 1999; Johnson and Zagotta, 2001; Wang et al., 2001) and was strengthened by high-resolution structures of an isolated CNBD from an HCN isoform (Zagotta et al., 2003), where most of the intersubunit contacts in the tetramer are formed by the C-linker, which is composed of four α-helices. This structure suggested that cyclic nucleotide binding drives a rearrangement of the C-linker domains, which opens the channel's ion permeation pathway. Our results refute this proposal, at least as a necessary, general mechanism of channel modulation by cyclic nucleotides, and they suggest that a minimal linking sequence is able to substitute for this function in the prokaryotic homologue.

Finally, we should ask whether the results here prove that MloK1 is a K⁺ channel. In fact they do not. To be sure, the gross functional characteristics of the reconstituted protein are consistent with a channel mechanism: electrically conductive transport with K⁺-like ionic selectivity. But without a direct measurement of unitary transport rate, it is impossible to rigorously distinguish channel from transporter mechanisms, and the liposome system on which we rely does not permit an estimate of this. We note that MloK1-mediated ⁸⁶Rb⁺ uptake is 10–100-fold slower than the rate expected for an always-open channel with a “typical” unitary conductance in the range 10–100 pS, or than that observed under similar conditions for KcsA (Heginbotham et al., 1998). This low uptake rate might reflect a small unitary conductance or a low open probability, even when maximally activated by cyclic nucleotide, or a combination of both. In the absence of direct single-channel recording, this point must remain unresolved, but it is worth noting that eukaryotic HCN channels have extremely low single-channel currents (DiFrancesco and Mangoni, 1994). In light of its tetrameric architecture, K⁺-selective cyclic nucleotide-modulated ion fluxes, and unambiguous membership in the S4 superfamily of ion channels, all accessible evidence implicates MloK1 as a cyclic nucleotide-modulated K⁺ channel.