Amantadine Inhibits NMDA Receptors by Accelerating Channel Closure during Channel Block

Introduction

Amantadine is a drug of broad clinical utility. It was first recognized as an antiviral agent (Davies et al., 1964) but soon afterward was also found to be useful in treating Parkinson's disease (Schwab et al., 1969). It continues to be a widely used and well tolerated drug in the treatment of parkinsonian movement disorders and may also slow the progression of Parkinson's disease (Blanchet et al., 2003), ameliorate chronic pain (Fisher et al., 2000), and improve recovery from traumatic brain injury (Meythaler et al., 2002).

The clinical utility of amantadine appears to result from its actions as a blocker of ion channels. Its antiviral action results from block of the M2 small viral membrane protein (Hay, 1992). Although the utility of amantadine in treating Parkinson's disease was initially assumed to result from effects on dopaminergic systems (Danysz et al., 1997), amantadine later was found to block the channel of NMDA receptors (Kornhuber et al., 1991; Lupp et al., 1992; Parsons et al., 1995; Blanpied et al., 1997). Increasing evidence suggests that the effectiveness of amantadine in treating nervous system disorders results predominantly from its inhibition of NMDA responses (Blanchet et al., 2003).

NMDA receptor channel blockers have diverse properties. Channel block by extracellular Mg²⁺ is of profound physiological significance because it is responsible for the powerful voltage dependence of postsynaptic Ca²⁺ influx at excitatory synapses (Dingledine et al., 1999). In contrast to clinically useful NMDA receptor channel blockers such as amantadine and its derivative memantine, many organic blockers, such as ketamine and phencyclidine, induce schizophrenic-like behavior in humans (Jentsch and Roth, 1999) and are neurotoxic (Olney et al., 1999).

Why do substances with an apparently similar pharmacological action, channel block of NMDA receptors, exhibit such divergent clinical properties? Previous data suggest that possible variations in either nonspecific drug actions or in NR2 subunit specificity do not provide adequate explanations (Javitt and Zukin, 1991; Bresink et al., 1996; Danysz et al., 1997; Monaghan and Larsen, 1997; Blanchet et al., 2003). Instead, the wide variation in clinical utility may arise from differences in the mechanisms of blocker interaction with NMDA receptors. Aspects of mechanism that can powerfully influence inhibitory properties of a channel blocker include the kinetics and affinity of blocker binding and the effects of bound blocker on transitions among channel closed states (Rogawski, 1993; Parsons et al., 1999b; Johnson and Qian, 2002). Despite the importance of these aspects of blocker action, we currently have limited understanding of the interaction between NMDA receptors and amantadine.

In this paper, we combined whole-cell and single-channel recording with quantitative modeling to investigate the mechanisms by which amantadine inhibits NMDA responses. We found that amantadine displays intermediate kinetics of block (Hille, 2001), which permitted us to measure blocking and unblocking rates and estimate rate of channel closure with amantadine bound. We show that, when amantadine is bound in the channel of NMDA receptors, it increases the rate of channel closure. As a result, the predominant inhibitory mechanism of amantadine is not blockade of current flow through open channels but rather increasing occupancy of channel closed states. The unusual properties of amantadine may play an important role in its clinical safety.

Materials and Methods

Cell cultures and solutions. Cerebral cortices were isolated from day 16 embryonic rats and used to prepare cultured neurons on glass coverslips as described previously (Antonov and Johnson, 1996). All procedures were approved by the Institutional Animal Care and Use Committee of the University of Pittsburgh. Neurons were used for experiments after 2-5 weeks in serum-containing medium. For recordings, coverslips were transferred to a recording chamber and bathed in an extracellular solution that contained the following (in mm): 140 NaCl, 2.8 KCl, 1 CaCl₂, and 10 HEPES. pH was adjusted to 7.3 with NaOH. The pipette solution contained the following (in mm): 120 CsF, 10 CsCl, 10 HEPES, and 10 BAPTA. pH was adjusted to 7.2 with CsOH.

Human embryonic kidney (HEK) 293T cells were maintained as described previously (Qian et al., 2005). For experiments, the cells were plated onto glass coverslips pretreated with poly-d-lysine and rat-tail collagen. Eighteen to 24 h after plating, the cells were transiently transfected with cDNAs encoding the NR1-1a and NR2B NMDA receptor subunits using a calcium phosphate precipitation procedure (Qian et al., 2005). cDNA of enhanced green fluorescent protein (eGFP) was cotransfected as a marker of successfully transfected cells. The amount of cDNA used per dish was as follows (in μg): 0.7 eGFP, 1.3 NR1-1a, and 2 NR2B. After a 6-8 h incubation in the transfection solution, the cells were washed with fresh culture medium that contained 200 μm APV. Experiments were performed 20-72 h after transfection.

Outside-out patch recordings. Pipettes were pulled from standard wall borosilicate glass with filament (Warner Instruments, Hamden, CT), coated with Sylgard (Dow Corning, Midland, MI), and fire polished. Pipette resistance ranged from 6 to 12 MΩ. Recordings were performed using an Axopatch 200A amplifier (Axon Instruments, Union City, CA). The patch was placed in front of the opening of one of a pair of perfusion tubes, and the solution was switched by moving the tubes. NMDA receptors were activated by solution containing 5 μm NMDA and 10 μm glycine alone or with the indicated concentration of amantadine. One to 4 min of channel openings at each drug concentration were low-pass filtered at 10 kHz and recorded to videotape for later analysis. The holding potential, after correction for the measured junction potential, was -67 mV.

Single-channel analysis. Data from videotape were filtered using an eight-pole Bessel filter with a cutoff of f_c = 4 kHz and digitized at 20 or 41 kHz. Segments containing unacceptable noise were removed using the Channel Analysis Program (R.C. Electronics, Santa Barbara, CA). Openings and closings were detected with a 50% threshold criterion (Colquhoun and Sigworth, 1995) using pClamp 6 software (Axon Instruments). The effective cutoff of the cascaded filters resulted in an estimated system dead time of 0.179/f_c = 0.048 ms, and events briefer than twice the system dead time (t_min) were deleted from all histograms and excluded from analysis. Histograms are presented as square root versus log time plots (Sigworth and Sine, 1987).

Binned dwell time distributions (eight bins per decade) were fit using the maximum likelihood method. Open-time distributions were well fit by a single-exponential component (Ascher and Nowak, 1988; Jahr and Stevens, 1990; Antonov and Johnson, 1996). Additional open-time components (Howe et al., 1988; Gibb and Colquhoun, 1992; Donnelly and Pallotta, 1995; Kleckner and Pallotta, 1995; Antonov et al., 1998; Antonov and Johnson, 1999) were not analyzed because they were either not visible or very small. The mean open time (τ_o) was corrected for missed brief closures, which otherwise could cause overestimation of τ_o, using the following equation (Ogden and Colquhoun, 1985; Marshall et al., 1990; Antonov and Johnson, 1996): τ_o = (t_rec - t_meas - t_miss)/(n_meas + n_miss + 1), where t_rec is total recording time of the data segment under analysis, n_meas is the measured number of closures with durations ≥t_min, t_meas is their summed closed duration, n_miss is the estimated number of closures with durations <t_min, and t_miss is their estimated summed closed duration. n_miss and t_miss were estimated by extrapolation of the fit closed duration distributions from t = t_min to 0. In control conditions, the value of τ_o calculated by this method was 4.19 ± 0.26 ms, 98 ± 3% of the unadjusted arithmetic mean open time calculated from open-time histograms, confirming the accuracy of the calculation when τ_o ≫ t_min. The correction had a greater effect when the unadjusted mean open time was briefer. In 100 μm amantadine, which increased the frequency of openings briefer than t_min, τ_o was 70 ± 2% of the unadjusted arithmetic mean open time.

Closed-time histograms in the absence and in the presence of amantadine were well fit in the large majority of cases by the sum of three exponential components. In the absence of amantadine, these components had time constants as follows: short duration (τ_C,S), 0.29 ± 0.04 ms; intermediate duration (τ_C,I), 7.22 ± 1.70 ms; long duration (τ_C,L), 494 ± 80 ms; with relative amplitudes of 0.43 ± 0.02, 0.16 ± 0.01, and 0.41 ± 0.02, respectively. As described in Results, the mean duration of amantadine blocking events (τ_C,B) was similar to τ_C,S. The accuracy of estimates of τ_C,B were optimized by establishing conservative criteria for acceptance of data to be used in the estimates. The criteria were used to minimize two potential errors. The first potential error resulted from the similarity of τ_C,S and τ_C,B. These two closed-time components were too similar to be fit by separate exponentials but in general should not have identical durations. Thus, short-duration channel closures could have interfered with measurement of τ_C,B. To minimize interference from these brief closures, we accepted τ_C,B measurements only if the short-duration component area was >70% of the total area of all closed-time components (see Fig. 2C). With this cutoff, approximately two-thirds or more of the short-duration component were attributable to amantadine block. The second possible error was overestimation of τ_C,B because of missed brief openings at high concentrations of amantadine. To minimize this error, we accepted τ_C,B measurements only when τ_O,B was >0.5 ms (Antonov and Johnson, 1996), limiting missed openings to a maximum of 18% of all openings. Because the number of missed openings increases with amantadine concentration, the observation that the estimated τ_C,B does not increase with amantadine concentration (see Fig. 2 B) supports the accuracy of our estimate. Only data in 10 μm (6 of 8 patches) and 30 μm amantadine (7 of 11 patches) satisfied the criteria described above. Data that met both these criteria are superimposed on all single-channel data in Figure 2, B and C.

The value of 1/τ_C,B provides an estimate of the sum of rates for leaving the blocked state (Colquhoun and Hawkes, 1995), which according to model 1, is α′+ k_-. Our use of 1/τ_C,B as an estimate of k_- alone is based on the assumption that k_- ≫ α′. Analysis of burst durations indicated that α′ is 251 s^-1 (see Results). Because this is much smaller than the estimated value of 1/τ_C,B (4480 s^-1), the approximation k_- = 1/τ_C,B should be in error by <6%.

n × P_open (number of channels in a patch multiplied by mean open probability per channel) was calculated as total channel open time divided by recording time using pClamp 9 (Axon Instruments). The patch recording durations used here ranged from 71 to 335 s (mean of 163 s).

Burst analysis was performed to estimate the rate of channel closure in the absence and presence of amantadine. Burst durations were measured as the duration of openings with closures briefer than a critical duration (t_crit) counted as part of open time (Colquhoun and Sigworth, 1995). Only patches in which <5% of all openings were multilevel openings were used. In the absence of amantadine, channel openings of NMDA receptors are often interrupted by short-duration closures (time constant τ_C,S). These short-duration closures constitute a trivial fraction of total closed time. This state therefore was ignored by choosing a t_crit between τ_C,S and τ_C,I (Dilmore and Johnson, 1998) and was not represented in the model we used for data analysis (see Fig. 3A). As a result, channel closures were defined as transitions to closed states of intermediate or longer duration. To minimize the unavoidable error of misclassification of closures, t_crit was calculated so that there were equal numbers of short-duration closures counted as between burst and intermediate closures counted as within burst (Magleby and Pallotta, 1983). Values for t_crit in control conditions were 1.23 ± 0.33 ms, in good agreement with values obtained in previous studies of forebrain neurons (Howe et al., 1988; Jahr and Stevens, 1990; Gibb and Colquhoun, 1992; Kleckner and Pallotta, 1995; Antonov and Johnson, 1996). The average separation of τ_C,S and τ_C,I by a factor of 27 is somewhat less than preferred for burst analysis (Colquhoun and Sigworth, 1995). However, burst duration showed little sensitivity to small variations in t_crit: doubling t_crit increased burst-duration estimates by only 9.0 ± 1.8%, whereas halving t_crit decreased burst-duration estimates by only 10.5 ± 1.3% (n = 6 patches). These results suggest that burst-duration estimates were reliable. In 1 of 32 patch data segments used for burst analysis, the middle closed-time component was <1 ms and was included within bursts.

Burst analysis in the presence of amantadine was designed to count amantadine blocking events as within burst. Because the value of τ_C,B was very close to the value of τ_C,S, the procedures used were identical to those used in the absence of amantadine. The mean values of t_crit in amantadine (1.02 ± 0.11 ms in 10 μm, 1.19 ± 0.13 ms in 30 μm, and 1.45 ± 0.06 ms in 100 μm amantadine) were similar to the control value (1.23 ± 0.33 ms).

Whole-cell recording. Whole-cell recordings from cultured neurons or transfected HEK 293T cells were performed as described previously (Blanpied et al., 1997), using the same solutions as for outside-out patch recordings. Rapid applications of the indicated NMDA concentration plus 10 μm glycine with or without amantadine were accomplished by moving a set of gravity-fed flow pipes under computer control. Solution exchanges were at least 98% complete within 120 ms (Blanpied et al., 1997).

All experiments were performed at room temperature.

Curve fitting and modeling. Concentration-inhibition curves were constructed by fitting data with the following equation: (1)

where Response is either normalized n × P_open (see Fig. 2 D) or I_Aman/I_Control (see Fig. 4 B), [Aman] is the amantadine concentration, IC₅₀ is the [Aman] at which the response is 50% inhibited, and n_H is the Hill coefficient, which reflects the cooperativity of drug action.

The equilibrium predictions of model 1 (see Fig. 3A) shown in Figures 3, B and C, 5, and 6 D were made with the equations derived in the supplemental material (available at www.jneurosci.org). The equilibrium constants among unblocked channel states (K_a = 11.0 μm; K_g = 39.0; K_ds = 0.2) were fixed at values chosen as described previously (Dilmore and Johnson, 1998) based on previous measurements (Benveniste and Mayer, 1991; Lester et al., 1993; Rosenmund et al., 1995). Because estimates of maximal P_open (which is established by K_g) for NMDA receptors have varied greatly, a value for K_g could not be set with confidence. However, the predictions based on model 1 shown in Figures 3C, 5, and 6 D varied only slightly when K_g was varied from 39 [corresponding to a maximal P_open of 0.025 (Rosenmund et al., 1995)] to 2.33 [corresponding to a maximal P_open of 0.3 (Jahr, 1992)]. Similarly, our general conclusions are not affected by changes in the values of K_a or K_ds within a plausibly physiological range. The K_d of amantadine was fixed at 110 μm, which equals the ratio of rate constants (k_-/k₊) determined here from single-channel measurements. Equilibrium constants among blocked channel states were set to equal the corresponding equilibrium constants for unblocked channels, except when otherwise indicated. Burst-duration fitting and predictions (see Fig. 6C) were made with Equation 2, which is derived in the supplemental material (available at www.jneurosci.org).

Predictions from and fitting of equations were performed in Origin 7 (OriginLab, Northampton, MA) or SigmaPlot 8 (SPSS, Chicago, IL). Errors in text and error bars in figures indicate SEM.

Results

Discussion

A crucial feature of clinically useful drugs is an affinity for their target molecule(s) that is high enough to prevent undesirable nonspecific effects. The affinity of a drug can be increased through increasing its binding rate or decreasing its unbinding rates. Because binding rates are limited, most fundamentally by diffusion, clinically useful drugs tend to unbind relatively slowly. The data presented here demonstrate that amantadine binds to its principal target, the open channel of NMDA receptors, with a binding rate typical of channel blockers, ∼40 μm^-1s^-1 at resting potential. The corresponding unbinding rate of amantadine, however, is remarkably fast at over 4000 s^-1. How does a drug with such a fast unbinding rate nevertheless exhibit inhibitory effects on its target in a useful concentration range? The answer shown here is that, after blocking the open channel, amantadine encourages NMDA receptor channels to occupy closed conformations. This unusual ability of amantadine increases its affinity, despite its fast unbinding from receptors with an open channel.

The principal mechanism by which amantadine inhibits NMDA responses at clinically useful concentrations is atypical of characterized channel blockers, which by definition plug open channels. Inhibition of NMDA responses by amantadine at concentrations below 100 μm, in contrast, depends mostly on stabilization of channel closed states. Thus, the principal mode of action of amantadine is that of a gating antagonist rather than channel blocker. Burst analysis (Fig. 6C) revealed that the main mechanism by which amantadine stabilizes channel closed states is through acceleration of channel closure by a factor of ∼2. Observation of a slow component of amantadine unblock after a depolarizing voltage jump (Fig. 7) supported the idea that amantadine binding accelerates channel closure. Additional support for this idea can be found in our previous study of inhibition by memantine and amantadine of whole-cell NMDA-activated currents (Blanpied et al., 1997). The best fit of a simplified version of model 1 to whole-cell responses during agonist and amantadine concentration jumps yielded channel closing rates that were 2.9 times faster when amantadine was bound. Thus, although amantadine binding and unbinding rates were unknown in the previous study, its modeling results were consistent with the results obtained here using burst analysis.

Acceleration of channel closure cannot fully explain the low IC₅₀ of amantadine, which requires that block by amantadine stabilize the closed state by a factor of 2.8 (Fig. 3C). It is possible that amantadine slows channel opening rate as well as accelerating channel closing rate. We cannot exclude the possibility, however, that amantadine has other inhibitory effects on NMDA receptors, although any additional inhibitory actions are unlikely to be at the NMDA binding site (Fig. 5).

Although only a single closed-liganded state (A₂R) is shown in the absence of blocker in model 1, NMDA receptors are known to have access to multiple closed-liganded states (Gibb and Colquhoun, 1992). The shortest-duration closed state that we observed, which was not included in model 1, was briefer than t_crit and so could not have terminated bursts. Shortening of burst duration by amantadine therefore must have resulted from increased occupancy of longer-lived closed states. It is possible that a new longer-lived closed state appears only when amantadine is bound. However, a simpler explanation would appear to be that entry into a state or states that also exist in the absence of amantadine is accelerated. Although understanding of connections among NMDA receptor closed states is limited, an explicit model to explain three components of closed-time histograms was proposed in an impressive recent study (Banke and Traynelis, 2003). In this model, open channels can close only to the shortest-lived closed state; from this state, either NR1 or NR2 subunits can make transitions to longer-lived closed conformations. The closed durations reported here differ from those recorded under different conditions by Banke and Traynelis (2003). Nevertheless, our burst data could be plausibly explained within the framework of their model if channel block by amantadine increases the rate of any of the three proposed types of closing transitions. Because amantadine binds in the channel, the most straightforward idea may be that amantadine increases the reverse rate of the final, rapid, concerted, pore-opening transition. This would mean that amantadine increases the rate into the shortest-lived state, the duration of which would be within burst. However, in this model, the shortest-lived state is a gateway state to all other closed states. Thus, increasing occupancy of the shortest-lived state would result in an increase in the rate of entry into longerlived states and would lead to shorter bursts.

Many organic NMDA receptor channel blockers affect channel gating (for review, see Johnson and Qian, 2002; Sobolevsky, 2003). However, occupation of the channel of NMDA receptors by organic channel blockers has been shown by previous studies to stabilize the channel open state. Amantadine appears to be the first organic NMDA receptor channel blocker to be shown to stabilize closed states. In contrast to most organic blockers, physiological block of NMDA receptor channels by Mg²⁺_o was proposed recently to accelerate channel closure (Kampa et al., 2004; Vargas-Caballero and Robinson, 2004). This proposal was supported by earlier measurements of an Mg²⁺_o-induced decrease in burst duration (Ascher and Nowak, 1988). We observed that amantadine, like Mg²⁺_o, exhibits a slow component of unblock after a voltage jump to a positive potential (Fig. 7, Table 1). Thus, the blocking actions of amantadine and of Mg²⁺_o may exhibit two notable similarities: unusually fast unblocking kinetics and the ability to accelerate channel closure while blocking. However, amantadine exhibits a much slower component of unblock that we did not observed with Mg²⁺_o (Fig. 7A, Table 1) (but see Kampa et al., 2004). Also, in contrast to amantadine, the IC₅₀ and K_d of Mg²⁺_o are nearly identical (Qian et al., 2002). This observation and others (Sobolevsky and Yelshansky, 2000) argue against the idea (Kampa et al., 2004; Vargas-Caballero and Robinson, 2004) that block by Mg²⁺_o significantly favors occupation of channel closed states. Thus, there must be yet uncharacterized differences in how amantadine and Mg²⁺_o influence gating while blocking the channel of NMDA receptors.

It is perhaps surprising that occupation of an ion channel by a relatively large molecule such as amantadine would accelerate rather than slow channel closure. Stabilization of closed states of a mutant Shaker K⁺ channel by an organic channel blocker has been observed previously (Holmgren et al., 1997). Tetraethylammonium (TEA) and its analog C₁₀ strongly stabilize the open state of wild-type Shaker channels. However, when Shaker channels with a point mutation (I470C) near the middle of the S6 region were examined, C₁₀ was found to weakly stabilize the channel open state and TEA to weakly stabilize channel closed states. These results indicate that the precise relationship between blocker structure and the shape of the channel cavity in which blockers are likely to bind (Jiang and MacKinnon, 2000) can determine the effects of blocker binding on channel gating. Structural and functional similarities between K⁺ channels and glutamate receptors (Panchenko et al., 2001) suggest that similar ideas may apply to block of NMDA receptors. Thus, systematic molecular modification of amantadine and other blockers (Antonov and Johnson, 1996; Bolshakov et al., 2003) can be used to help define channel shape while advancing the search for more useful therapeutic drugs.

Amantadine and its close analog memantine are both NMDA channel blockers that exhibit high clinical utility. Their spectra of utility, however, differ substantially (Danysz et al., 1997); for example, amantadine appears to be the more effective antiparkinsonian drug, whereas memantine currently is the only drug approved by the FDA for use in the treatment of moderate to severe Alzheimer's disease. Correspondingly, each drug exhibits distinct antagonistic properties that influence their affinity, kinetics, ability to be trapped, and use dependence. The properties of amantadine described here permit its action to resemble in important respects block by Mg²⁺_o, although amantadine block is less voltage dependent because of the lower valence of the drug (+1 at physiological pH). Thus, the remarkably fast unblocking kinetics of amantadine allow it, like Mg²⁺_o, to at least partly unblock during the brief depolarization of an action potential, a characteristic that may decrease deleterious clinical effects. The weaker voltage dependence of amantadine should permit it to inhibit NMDA responses more effectively than Mg²⁺_o during the prolonged depolarizations that accompany neurological insults, enhancing its neuroprotective potential. Together, the unique combination of kinetics, effects on channel gating, and voltage dependence of amantadine may enhance both its clinical safety and effectiveness.