Introduction

Top Summary Introduction Materials and Methods Results Discussion References

Cyclic nucleotide-gated (CNG) channels play important roles in visual and olfactory signal transduction. They are activated by the direct binding of cAMP or cGMP to sites located at the intracellular side of channel protein (Fesenko et al., 1985; Zimmerman and Baylor, 1986; Nakamura and Gold, 1987). Because phosphorylation by cyclic nucleotide-dependent protein kinases is not involved, the activation and deactivation can take place at higher speed than in kinase-dependent pathways. The direct activation appears well suited for an important function in vertebrates: to mediate the detection of external stimuli by the peripheral sensory neurons of the visual and olfactory systems. In this regard the CNG channels resemble ion channels that are gated by extracellular ligands, such as the nicotinic acetylcholine receptors, which mediate fast communication between cells.

Molecular cloning data have revealed that CNG channels form a family of related proteins (reviewed in Yau, 1994; Finn et al., 1996). For each cloned channel subunit, the deduced amino acid sequence contains a "core" channel domain, followed by a carboxyl terminal cyclic nucleotide-binding domain. Similar to voltage-gated channels, the "core" has six putative transmembrane segments and a P region, which comprise part of the pore. In other important aspects, however, the CNG channels differ from a typical voltage-gated channel. First, unlike the voltage-dependent Na⁺ or K⁺ channels, the CNG channels are poorly selective among monovalent cations. Second, their activities depend only slightly on membrane potential; apparently the conformational changes involved in activation are energetically coupled more tightly to ligand binding than to changes in membrane potential.

Despite these differences from voltage-gated channels, it is generally believed that each CNG channel, similar to a voltage-gated channel, is formed by the association of four subunits, each possessing its own cyclic nucleotide-binding site. Thus the activation of CNG channels may involve, or even require, the binding of agonist molecules at all four binding sites. This would extend the similarity with voltage-gated channels, for which the conformational changes at all four "voltage sensors" contribute to the opening transition (see for example Hoshi et al., 1994; Stefani et al., 1994). The participation of multiple binding events during CNG channel activation is supported by the dose-response relations of the macroscopic currents: the Hill coefficient, an empirical measure of apparent cooperativity, is almost always greater than 1, and usually falls between 2 and 3 (Zimmerman and Baylor, 1986; Li et al., 1997).

An important goal of studies on CNG channels is to understand the mechanism of channel activation, especially how the binding events are coupled to the opening of the pore, and how the presumed four channel subunits interact to generate the apparent cooperativity. A companion paper reports structure-function studies on the CNG binding site (Li and Lester, 1999). To approach this goal it is also appropriate to examine the stochastic behavior of individual channels, because the macroscopic current, reflecting the collective activity of many channels, lacks the resolution to describe closely related models. It is also appropriate to examine the stochastic behavior of individual channels, because the macroscopic current, reflecting the collective activity of many channels, lacks the resolution to describe closely related models. In this regard, single-channel measurements provide a much richer set of data; in fact, they are among the few methods for monitoring the behavior of a single allosteric protein in real time. Furthermore, they allow us to recognize individual kinetic states and to specify the rate constants governing the transitions among them.

A number of laboratories have recorded single-channel currents of CNG channels. However, efforts toward a quantitative rather than merely descriptive kinetic analysis were previously hindered by at least three factors. 1) Channel openings are often flickery, i.e., the open-close transition are too brief for the recording apparatus to resolve (Matthews and Watanabe, 1988; Ildefonse et al., 1992; Sesti et al., 1994; Taylor and Baylor, 1995; Bucossi et al., 1997). 2) Many types of CNG channels open to multiple conductance levels (Taylor and Baylor, 1995; Liu et al., 1996; Bucossi et al., 1997; Ruiz and Karpen, 1997; Liu et al., 1998). When there is no clear correspondence between binding states and conductance classes, the multiplicity of conductance only complicates kinetic analysis. 3) Most studies used channels expressed in native tissues (Matthews and Watanabe, 1988; Sesti et al., 1994; Taylor and Baylor, 1995), in which the channel population is not homogeneous: the channel that is being recorded can be of any subunit composition out of the many different possibilities.

Here we report a study that circumvents some of these problems. We found that the  subunit of the rat olfactory CNG channel, rOCNC1, when expressed in Xenopus oocytes, displayed a single conductance and opened to a relatively stable current level. We therefore examined the kinetic characteristics of these single, homogeneous CNG channels under steady-state conditions. We found that 1) the openings showed no conspicuous clustering, in contrast both to predictions of some common models and to properties of many other ligand-gated channels; 2) with more ligands bound the open state was further stabilized, whereas the closed state was further destabilized, just as predicted by allosteric models; and 3) the open- and closed-time distributions displayed only a limited number of distinguishable components; as a result, simple kinetic models invoking one or two binding events and one or two gating transitions can account for the observed kinetic properties.

To estimate kinetic parameters, we used the maximum interval-likelihood method (Qin et al., 1996, 1997), in which the probability of observing an actual sequence of events, according to a given scheme, is simply the joint probability of observing these events individually and in the observed sequence. The algorithm searches for the parameters that maximize this probability. The final likelihood score of a model is used to compare it with alternative models. The comparison employs objective, statistical criteria for penalizing over-parameterization (Horn, 1987).

During the analysis, special attention was given to the comparison between cAMP and cGMP, both of which can activate rOCNC1. Previous studies show that cGMP has a ~20-fold lower EC₅₀ than cAMP, although both produce the same maximal activation (Varnum et al., 1995; Gordon and Zagotta, 1995a,b). The present study verifies this finding and asks whether it arises 1) because of differences in the initial ligand-channel interaction, or 2) because of differences in the subsequent conformational transitions that open the channel (Li et al., 1997). The previous studies, based on macroscopic currents, assigned (2) as the likely explanation (Varnum et al., 1995; Gordon and Zagotta, 1995a,b). The single-channel kinetic study now suggests that explanation (1) holds: cGMP probably binds with much higher affinity, but the conformational changes are only modestly more likely than for cAMP.

	Materials and Methods

Top Summary Introduction Materials and Methods Results Discussion References

Expression. All experiments were carried out on the rat olfactory CNG channel  subunit (Dhallan et al., 1990), kindly provided by Dr. K. W. Yau. The cDNA was subcloned (at EcoRI and HinDIII) into the pGEMHE vector, originally constructed by E. R. Liman, containing the 5' and 3' untranslated sequences of the Xenopus laevis major -globin gene for enhanced expression in oocytes (Liman et al., 1992). Stage V and VI Xenopus laevis oocytes were injected with cRNA synthesized in vitro (Ambion T7 mMESSAGE mMACHINE Kit, Ambion Inc., Austin, TX) from plasmid linearized with PstI. To obtain patches that contain only one channel, we injected oocytes with 50 nl of each of three serial dilutions of cRNA, covering a concentration range of ~16-fold. The highest concentration of the three dilutions varied from 1 to 50 µg/ml, depending on the month-to-month variations in expression levels, and to allow recordings to be performed from 24 to 120 h after injection. To improve the viability of oocytes, horse serum (HyClone Laboratories, Logan, UT) was added at 5% to the incubation solution ND96 (Quick et al., 1992). ND96 contains: 96 mM NaCl, 2 mM KCl, 1 mM MgCl₂, 1.8 mM CaCl₂, and 5 mM HEPES (pH 7.4).

Recording and Signal Processing. The recording pipettes were fabricated from filamented, borosilicate glass tubing (Corning type 7740, A. 1.5 mm, i.d. 0.86 mm, Sutter Instrument Co. Novato, CA), using a Flaming/Brown Micropipette Puller (model P-87, Sutter Instrument Co.). The pipette tips were fire-polished (MF-83, Narishige Scientific Instrument Lab, Tokyo, Japan). The filled pipettes had resistances between 5 and 15 M in the bath solution.

Kinetic Modeling. The data were idealized in FETCHAN of pCLAMP 6 using a half-magnitude threshold-crossing criterion for detecting event transitions. Transitions were individually inspected and manually accepted or rejected. The resulting event-list files were converted into ASCII form, and were selected for further analysis by 1) discarding files that were apparently nonstationary, judged by inspecting the opening probability, the open times, or the closed times, and 2) discarding files with fewer than 800 events. For the selected files, open- and closed-time histograms were constructed in PSTAT of pCLAMP 6, and were fitted in PSTAT with sums of exponential functions using the Levenberg-Marquardt method with weighting by function. The histograms were binned with a logarithmic time axis and plotted with a square-root transformation of the vertical axis, so that the individual exponential components could be directly visualized as apparent peaks in the histograms (Sigworth and Sine, 1987). We verified that openings as brief as 90 µs could be detected at 50% of the original signal amplitude by passing pulses through the NeuroCorder-filter-computer combination. This value was used 1) to transform the event lists by deleting durations shorter than 90 µs and joining the adjacent events (Colquhoun and Sigworth, 1995), and 2) as the lower limit of the fitting range while we fit the histograms with exponential functions. In this study we used the time constant and relative fraction of the exponential functions for describing the main characteristics of data, whereas the kinetic rate constants were estimated using the maximum interval-likelihood method (see below).

	Results

Top Summary Introduction Materials and Methods Results Discussion References

In this study we focused on the rOCNC1 homomeric channel. The rOCNC1/rOCNC2 heteromeric channels resemble the native channels more closely, but they are likely to coassemble with varied and undefined stoichiometry, each probably having a distinct set of functional characteristics. It has been known that presence of rOCNC2 raises the apparent sensitivity to cAMP (Bradley et al., 1994; Liman and Buck, 1994). Furthermore in the oocyte expression system, rOCNC2 causes desensitization, which is absent in the rOCNC1 homomeric channel (Liman and Buck, 1994). It is certainly true that the native olfactory channels are likely to be heteromeric. However, in this study, seeking a better understanding of the activation mechanism, we exploited the greater simplicity of homomeric channels.

We used the oocyte expression system. In the past, the HEK293 cell system has served well for our macroscopic studies. The expression level was high and was highly reproducible. HEK293 cells have the major drawback, however, that it is nearly impossible to isolate single channels: the channels tend to form clusters, even at the lowest levels of expression that we tested. The expression level in Xenopus oocytes, in contrast, seems to depend more linearly on the amount of mRNA injected, and the channels are distributed more evenly in the plasma membrane.

CAMP and cGMP Produce Similar Maximal Macroscopic Responses. The ligand selectivity of the olfactory CNG channels is very different from that of photoreceptor channels: although cGMP is a much more potent agonist than cAMP for the rod channels (Fesenko et al. 1985), both agonists can fully activate the olfactory channels (Nakamura and Gold, 1987; Zufall et al., 1994). It has been reported that the cloned rat olfactory channel rOCNC1 can be activated to roughly equal maximal levels by cAMP and cGMP (Dhallan et al., 1990; Gordon and Zagotta, 1995b). We confirmed this finding in our macroscopic recordings. For the patch shown in Fig. 1, cAMP activated 96% of the current activated by saturating cGMP. Two other patches yielded maximal current ratios between cAMP and cGMP of 99% and 87%, respectively. In contrast, for the bovine rod channels, cAMP activates less than 1% of the current activated by saturating concentrations of cGMP (see for example, Gordon and Zagotta, 1995b).

View larger version (14K): [in this window] [in a new window]   Fig. 1.   Dose-response relation for macroscopic currents activated by cAMP and cGMP in a representative excised patch at Vm = 80 mV. The smooth lines are fits to Hill equation: I = Imax/(1 + (EC50/cNMP)n), where Imax is maximal current, EC50 is concentration that activates 50% of maximal current, cNMP is concentration of cyclic nucleotides, and n is the Hill coefficient. Imax, EC50 and n were free parameters. For patch shown, Imax for cAMP is 96% of Imax for cGMP (1743 pA versus 1804 pA). Other parameters are: EC50 for cAMP, 78 µM; for cGMP, 2.7 µM; n for cAMP, 1.8; and for cGMP, 1.9.

Basic Characteristics of Single Channels. Figure 2 presents consecutive current traces showing openings of a homomeric rOCNC1 channel, recorded at +60 mV and 60 mV, respectively, exposed to 50 µM cAMP at the cytoplasmic face. At +60 mV the channel opened to a single, relatively stable conductance level. The lack of subconductance and lack of flicker stand in clear contrast to the native rat olfactory CNG channels (Frings et al., 1992), the expressed homomeric bovine photoreceptor CNG channel (Ruiz and Karpen, 1997), and expressed heteromeric channels (Bradley et al., 1994; Liman and Buck, 1994; Liu et al., 1996, 1998). At 60 mV the open-channel noise became noticeably larger, indicating unresolved brief closings that interrupt opening. The baseline noise, however, was not increased at 60 mV, suggesting that there are few unresolved brief openings.

View larger version (27K): [in this window] [in a new window]   Fig. 2.   Consecutive current traces from a single channel in response to 50 µM cAMP. A, +60 mV; B. 60 mV. One section in A and one in B are redisplayed on a 10× expanded time scale. C and O indicate open and closed current level. Note increased open-channel noise at 60 mV.

View larger version (29K): [in this window] [in a new window]   Fig. 3.   Consecutive current traces of a channel recorded at 60 mV. A, 5 µM cGMP; B, 100 µM cAMP. C and O indicate open and closed current level. Note basic similarities in records.

View larger version (12K): [in this window] [in a new window]   Fig. 4.   An I-V plot of single-channel currents. Current levels recorded at a series of membrane potentials, from 10 mV to 150 V, are plotted against potentials.

View larger version (21K): [in this window] [in a new window]   Fig. 5.   Open- (left) and closed-time (right) histograms at three different cAMP concentrations. Smooth curves are sum of exponential functions used to fit observed distributions. Time constant and fractional area of major component in each histogram are given in Table 1.

Fluctuations in P_open. Even in the presence of a constant concentration of agonist, the P_opens often showed strong fluctuations. Figure 6A plots the P_open of a channel during continuous activation by 5 µM cGMP. The 158-s period can be roughly divided into three segments, between which there are marked differences in channel activity. The division between segments I and II was supported by the observation that in the first ~20-s period the major closed intervals were longer than for the rest of the record and there was a discernible lack of brief closed intervals. The channel displayed relatively stable kinetics in segment II; and this portion of the data was selected for kinetic analysis. Figure 6B describes another channel, activated by 250 µM cAMP. The four recording periods in Fig. 6B were interrupted during the actual experiment by other operations, including switches to a different concentration or to another agonist. The channel entered segments II and IV with activity patterns different from the ones at the end of the previous segment. Also, the channel underwent spontaneous mode changes within segments II and IV, in a fashion similar to that shown in Fig. 6A. Segments I and III were relatively stable and comparable in P_open; they were regarded as representing a single channel "mode", and joined as one continuous record in further kinetic analysis.

View larger version (22K): [in this window] [in a new window]   Fig. 6.   Fluctuations of Popen. A and B, Popen for two different channels activated by 5 cGMP and 250 cAMP, respectively. Three segments in A are contiguous recording periods demonstrating direct switching between different "modes" of activation, whereas four segments in B are unconnected recording periods, between which the channel was exposed to other cAMP concentrations or to a different agonist. Popen values were calculated within 2-s time windows.

Open-Channel Noise. As noted above, most of our analyses were performed at 60 mV to maintain stable seals and to avoid stretch-activated channels. At this membrane potential, the open channels display noticeable excess noise (Figs. 2B and 3). Previous studies suggest that noise during CNG channel openings arises primarily from open-channel block by Ca²⁺ or protons (Root and MacKinnon, 1994). Therefore the open-channel noise would not be expected to depend on the agonist; but we sought to test this point directly.

View larger version (23K): [in this window] [in a new window]   Fig. 7.   Two cyclic nucleotides produce indistinguishable open-channel fluctuations. Power spectra of single channel current activated by 250 µM cAMP (A) and 5 µM cGMP (B). Data were calculated by subtracting spectra in the absence of cyclic nucleotides from spectra in the presence of cyclic nucleotides. Smooth lines represent fits to single Lorentzian functions. Corner frequencies and zero frequency power levels are indicated. Frequency-independent power offsets were, for A and B respectively, 5.7 × 1010 pA2/Hz and 2.3 × 109 pA2/Hz.

Kinetic Properties: Comparisons between Agonists. Figures 8 and 9 present our kinetic analyses in three separate recordings, each on a single channel that was tested with both cGMP and cAMP. Although the details of the most likely model vary among experiments, certain unifying characteristics appear.

View larger version (23K): [in this window] [in a new window]   Fig. 8.   A comparison between cAMP and cGMP. A and C, activation by 5 µM cGMP; B and D, activation by 250 µM cAMP. A and B, upper panels, Popen plots (data are plotted for successive 2- and 4-s intervals, respectively); middle and lower panels, closed- and open-time histogram, respectively, with fitting by sums of exponential functions. For both cAMP and cGMP, open times are fitted by a single component, shown by superimposed curves, whereas closed times require two components. C and D, upper panel, simplest kinetic models that can account for observations. Each horizontal arrow represents binding (toward left) or dissociation (toward right) of a single agonist molecule, whereas each vertical arrow represents opening (downward) or closing (upward) of channel. Middle and lower panels, same histograms as in A and B, with superimposed theoretical pdfs calculated from most likely kinetic parameters. These parameters are shown in kinetic models in lower panels. Units are µM1 s1 or s where appropriate. Ratios of rate constants for each transition, i.e., equilibrium binding and gating constants, are documented in Table 2 together with the exponential fitting parameters and other information about data set.

1

1

1

1

1

1

View larger version (32K): [in this window] [in a new window]   Fig. 9.   Two other comparisons between cAMP and cGMP. A, a channel tested with activation by 5 µM cGMP (a) and with 100 µM cAMP (b). B, a channel tested with activation by 5 µM cGMP (a) and with 250 µM cAMP (b). Upper panels, Popen plots; middle and lower panels, closed- and open-time histograms, respectively. Lines represent theoretical pdfs calculated from most likely kinetic parameters. These parameters are shown in kinetic models in lower panels. Units are µM1 s1 or s1 where appropriate. Ratios of rate constants for each transition, i.e., equilibrium binding and gating constants, are documented in Table 2 together with exponential fitting parameters and other information about data set.

Kinetic Properties: Concentration Dependence. We also performed simultaneous fits of data recorded at several different concentrations for a given channel. This was done by scaling the ligand association rate constants with the ligand concentrations for individual data sets, and optimizing the total likelihood of the combined data. We found that such global fittings generally provided less satisfactory results than fitting the data sets separately, i.e., without the constraint of holding the association rate constants in proportion to concentration. For instance, if a model that adequately fits two concentrations separately was used to fit the two data sets simultaneously, the predicted dwell-time probability density functions (pdfs) usually had noticeably poorer "goodness-of-fit" to the observed histograms than fitting separately.

1

View larger version (19K): [in this window] [in a new window]   Fig. 10.   Simultaneous fitting of data obtained at two cAMP concentrations. A, 100 µM; B, 250 µM. C, A model that provides an adequate fit to both records simultaneously. Theoretical pdfs calculated from kinetic parameters shown in C are superimposed as smooth lines in A and B.

	Discussion

Top Summary Introduction Materials and Methods Results Discussion References

Our experiments use modern single-channel kinetic analysis to deduce the mechanism of cyclic nucleotide channel activation. We studied homomultimeric channels formed by the rat olfactory cyclic nucleotide-gated channel  subunit. We observed that this channel has a single, full-conductance level even at very low cyclic nucleotide concentrations, allowing for detailed kinetic analysis. In other recent studies, other CNG channels have had several open conductance values (Ruiz and Karpen, 1997), and these variations have been used to deduce facts about stoichiometry and order of the subunits (Liu et al., 1996; Liu et al., 1998).

cAMP and cGMP Differ Primarily in Binding Affinities. One significant finding in this study is the similarity between cAMP and cGMP in the parameters that govern the open-closed transitions for liganded channels. This result was unexpected, because previous studies on macroscopic responses were interpreted to suggest that cGMP binds with similar affinity to cAMP, but then is more likely to open the channel (Varnum et al., 1995; Gordon and Zagotta, 1995a,b). Our conclusion of dissimilar binding but similar subsequent conformational changes is of course quite consistent with the findings that cAMP and cGMP induce equal maximal macroscopic conductances (Fig. 1; see also Gordon and Zagotta, 1995a). A greater gating equilibrium constant for cGMP (Ls in Table 2), would be associated either with more stabilized open states, or more destabilized closed states, or both. None of these differences was observed in our recordings: when the comparison was made at similar open probabilities, the open and closed times were comparable between cAMP and cGMP. Our experiments resolve events as fast as ~90 µs, and follow the channel activity for 10 to 20 min. Within this time frame, the lack of significant difference between cAMP and cGMP is true for most of the channels recorded, despite the differences in the detailed model that accounted for the data among recordings. The only noticeable exception is that some channels showed one more closed-time component with cGMP than with cAMP. Noise analysis on open channels would be expected to revealor at least to suggestmajor differences in conductance substates or gating kinetics not clearly resolved by the single-channel recordings; but no such differences were found (Fig. 7), again arguing against differences other than at the binding steps.

Cooperativity Arises Partially Because Agonist Binds To Open State. A second significant finding in this study is that cyclic nucleotide can bind to the open state, and that the open state is open longer with additional bound agonist molecules. This result could only be obtained with a kinetic approach like that used here. In the example of Fig. 10, the second binding would lengthen the open state by a factor of ~2 at a concentration of 55 µM cAMP. Because the EC₅₀ for cAMP is 78 µM, this lengthening of the open state also contributes importantly to the sigmoid start of the dose-response relation.