Local constraints in either the GluN1 or GluN2 subunit equally impair NMDA receptor pore opening

Representative crystal structures of ligand-gated ion channels from multiple families are now available (Jiang et al., 2002; Hilf and Dutzler, 2008; Kawate et al., 2009; Sobolevsky et al., 2009). Nevertheless, the dynamics of their activation gating—the series of conformational changes that couple the binding of ligands to the opening of the ion channel pore—remain unresolved. Ionotropic glutamate receptors (iGluRs), including the N-methyl-d-aspartate (NMDA) receptor subtype, mediate fast excitatory synaptic transmission integral to central nervous system function. Kinetic studies of NMDA receptors have identified key energetic steps, including various intermediate states, in the activation gating process (Wyllie et al., 1996; Banke and Traynelis, 2003; Auerbach and Zhou, 2005; Schorge et al., 2005; Dravid et al., 2008; Kussius and Popescu, 2009). These kinetic mechanisms therefore provide a template to define the dynamics of gating in NMDA receptors, although the structural events underlying the kinetic steps are unknown.

A single iGluR subunit contains four modular domains: the extracellular amino-terminal domain and ligand-binding domain (LBD), a transmembrane domain (TMD), and an intracellular C-terminal domain (Oswald et al., 2007; Traynelis et al., 2010; Mayer, 2011). The core domains necessary and sufficient for activation gating are the LBD (composed of S1 and S2) and TMD (composed of M1–M4) (Fig. 1 A). These core domains are coupled by three short polypeptide linkers (S1–M1, M3–S2, and S2–M4) that mediate ligand-induced conformational changes in the LBD to the opening of the channel pore. In the tetrameric assembly of an iGluR, the four M3 transmembrane segments form the pore and constitute the main channel-gating element (Kohda et al., 2000; Jones et al., 2002; Qian and Johnson, 2002; Sobolevsky et al., 2002; Yuan et al., 2005; Blanke and VanDongen, 2008; Chang and Kuo, 2008). Homologous to the evolutionarily linked K⁺ channel (Wo and Oswald, 1995; Wood et al., 1995; Chen et al., 1999; Panchenko et al., 2001; Kuner et al., 2003; Sobolevsky et al., 2009), the opening of the channel pore in iGluRs involves, in some form, movements of the M3 segments away from the central axis of the pore (Fig. 1 B). The M3–S2 linker that connects the LBD to the M3 segment must therefore be a central element of the gating machinery coupling conformational dynamics at the LBD to pore opening.

In the only intact iGluR structure of a homomeric AMPA receptor, the LBD–TMD linkers mediate an unprecedented symmetry mismatch between the LBD (twofold symmetry relative to central axis of receptor) and the ion channel (fourfold symmetry) by adopting different conformations in adjacent subunits (Sobolevsky et al., 2009). If this symmetry mismatch also exists in the NMDA receptor, it is likely to have important gating consequences because the functional unit within the receptor is thought to be a heterodimer containing a glycine-binding GluN1 and a glutamate-binding GluN2 subunit (Furukawa et al., 2005). Because of the different linker arrangements, the energetic mechanisms coupling ligand binding to pore opening may differ within an NMDA receptor heterodimer. Still, in an intact NMDA receptor, all four subunits must bind their respective ligands for subsequent pore opening to occur (Benveniste and Mayer, 1991a; Clements and Westbrook, 1991; Schorge et al., 2005), although the mechanism and structural basis for this concerted gating process are unknown.

To investigate subunit-dependent mechanisms driving pore opening in NMDA receptors, we identified means to physically constrain the M3–S2 linker by disulfide cross-linking it with the S2–M4 linker within specific NMDA receptor subunits, either GluN1 or GluN2A. Using single-channel recordings and kinetic analysis to define energetic effects on gating, we find that constraining M3–S2 in either GluN1 or GluN2A strongly reduced channel open probability. The most dramatic effect was on the late gating steps mediating pore opening where a long-lived, energetically stable open state was nearly abolished. In these late gating steps, we found a tight coupling across all NMDA receptor subunits; full pore opening would not occur unless all four subunits undertook their own intrasubunit gating actions. Thus, concerted gating in NMDA receptors requires equivalent intrasubunit movements of M3–S2 relative to S2–M4 occurring together across both GluN1 and GluN2 subunits.

Cysteine substitutions in the rat GluN1a (NCBI Protein database accession no. P35439) and GluN2A (accession no. Q00959) subunits were generated using PCR-based methods (Sobolevsky et al., 2007). As a reference and background for mutagenesis, we used a GluN2A construct in which a reactive cysteine near the N terminus of S1 was mutated to an alanine (C399A). Xenopus laevis oocytes were prepared, injected with cRNA, and maintained as described previously (Sobolevsky et al., 2007). Recordings were made 2–5 d after injection. For mammalian cell expression, human embryonic kidney (HEK) 293 cells were cotransfected with cDNA for GluN1 and GluN2A subunits, as well as a vector for enhanced green fluorescent protein (pEGFP-Cl; Takara Bio Inc.), at a ratio of (in µg) 1:1:1, using Fugene 6 (Roche). Recordings were made 24–72 h after transfection.

Macroscopic currents of Xenopus oocytes were recorded at room temperature (20–23°C) using two-microelectrode voltage clamp (TEV-200A; Dagan Corporation) with Cell Works software (npi electronic GmbH). Microelectrodes were filled with 3 M KCl and had resistances of 1–4 MΩ. The external solution consisted of (in mM): 115 NaCl, 2.5 KCl, 0.18 BaCl₂, 5 HEPES, and 100 µM EDTA, pH 7.2 (NaOH), unless otherwise noted. All reagents including glycine (20 µM) and glutamate (200 µM), competitive antagonists for GluN1, 5,7-dichlorokynurenic acid (DCKA; 10 µM), and for GluN2, DL-2-amino-5-phosphonopentanoic acid (APV; 100 µM), and the reducing agent dithiothreitol (DTT; 4 mM) were applied with the bath solution. All reagents were obtained from Roche or Sigma-Aldrich.

Steady-state reactions were quantified at a holding potential of −60 mV. Baseline glutamate-activated current amplitudes (I_pre) were established by three to five 15-s applications of glutamate and glycine. All agonist or any other reagent applications were separated by 30–120-s washes in agonist-free solution. 4 mM DTT was applied for 2 min in the absence of agonist but in the presence of the competitive antagonists DCKA (for GluN1) and APV (for GluN2). We found that DTT more effectively potentiated current amplitudes under these conditions than when applied in the presence of agonists. After DTT exposure, current amplitudes (I_post) were determined using three to five agonist applications. The change in glutamate-activated current amplitude, expressed as a percentage (percent change), was calculated as: = (I_post − I_pre)/I_pre × 100. In certain instances, we corrected for observed current rundown by fitting a single-exponential function to a minimum of three pre-DTT glutamate-activated current amplitudes.

MK801 is an irreversible (on the timescale of tens of minutes) open-channel blocker at hyperpolarized potentials (Huettner and Bean, 1988). MK801 inhibition was assessed with either 1 µM (Fig. 3, A–C) or 25 nM (Fig. 3 D–H) MK801 after agonist-induced current amplitudes had reached steady state. The change in glutamate-activated current amplitude, expressed as a percentage (percent change), was calculated as: = (I_post − I_pre)/I_pre × 100. For DTT and antagonist treatments, percent change was calculated relative to the current amplitudes preceding these treatments but after MK801 block. The kinetics of MK801 inhibition were fitted with either single- or biexponential functions. A higher-order exponential function was used only when it qualitatively minimized the residual currents (I_res) of these fits (see bottom graphs in Fig. 3, D and E).

Single-channel recordings were made at steady state using the cell-attached configuration on HEK 293 cells. Currents were acquired using an amplifier (Axopatch 200B; Molecular Devices), filtered at 10 kHz (four-pole Bessel filter), and digitized at 50 kHz (ITC-16 interfaced with Patchmaster; HEKA). The bath solution consisted of (in mM): 150 NaCl, 2.5 KCl, and10 HEPES, pH 7.2 (NaOH). Although gating in NMDA receptors can be modulated by external [H⁺] (Erreger et al., 2005a), we used pH 7.2 rather than a high pH to allow direct comparison between the HEK cell and oocyte recordings. Pipette electrodes were pulled from thick-walled borosilicate glass (Warner Instruments) and fire-polished immediately before use. The pipettes were filled with an external solution (pipette) consisting of the bath solution supplemented with 1 mM EDTA, 0.1 mM Gly, and 1 mM glutamate. Final resistances were 10–40 MΩ when filled with pipette solution and measured in the bath solution. A voltage of +100 mV was applied through the recording pipette to elicit inward sodium currents. In certain instances, patches were broken into the whole cell configuration to note the resting V_m, which was typically from −11 to −30 mV, after the recording period had ended.

For experiments assessing the effect of DTT on NMDA receptor single-channel activity, we paralleled the DTT application protocol used to efficiently reduce the disulfide bonds in our whole cell oocyte recordings (see Fig. 2 C). We pretreated the NMDA receptor–transfected HEK cells with 4 mM DTT, 10 µM DCKA, and 100 µM APV in our bath solution and then came in with the patch pipette (containing the agonists but no DTT) under positive pressure to form the cell-attached seal. For whole cell recordings in oocytes, current amplitudes of the double-cysteine–substituted receptors gradually decreased over time after washout of DTT, presumably because of disulfide bond reformation (not depicted). However, for the single-channel recordings, the DTT-induced high activity levels remained constant even in recordings lasting up to 50 min. We do not understand this apparent lack of disulfide bond reformation in the cell-attached mode, but it may indicate that the glass pipette does not provide an absolute oxidizing environment.

Data (in .dat format) were transferred to QuB (http://www.qub.buffalo.edu) for analysis. Each recording was visually inspected in its entirety for multiple simultaneous openings, signal-to-noise fluctuations, high frequency artifacts, and baseline drifts. For GluN1/GluN2A receptors without DTT and all receptors with DTT, the recordings consisted of long clusters of activity separated by seconds-long periods of zero activity, making it straightforward to detect more than one channel in the patch as simultaneous openings. In these cases, given the high P_o (0.5–0.98) of GluN2A-containing receptors and the minutes-long duration (with 10,000–320,000 events) of recordings without any apparent multiple openings, these recordings certainly contained only a single channel in the patch. For recordings of GluN1(R645C,S784C)/GluN2A and GluN1/GluN2A(Q642C,K785C) without DTT (P_o < 0.02), it was more challenging to detect single-channel patches. First, many patches were recorded but excluded from analysis because of obvious simultaneous openings of multiple channels. Of the remaining patches, only minutes-long recordings (with 2,500–52,000 events) without any apparent simultaneous openings were further analyzed. According to Colquhoun and Hawkes (1990), a two-channel patch containing channels with a P_o of 0.01 is expected to undergo at least one double opening every 200 single-opening events. Among our analyzed GluN1(R645C,S784C)/GluN2A and GluN1/GluN2A(Q642C,K785C) recordings, the least number of events recorded in a patch without any apparent double opening was 2,675. Therefore, we are reasonably confident that we analyzed patches containing only a single channel.

Further processing was done to eliminate occasional obvious brief noise spikes (e.g., high amplitude, stereotypical exponential-like decay, etc.) by matching them to the level of adjacent events using the “erase” function in QuB. Long periods of high noise were deleted with the remaining flanking segments separated as discontinuous segments. Baseline drift was corrected by resetting baseline to zero current levels. Processed data were idealized using the SKM algorithm after filtering to 12 kHz with a Gaussian digital filter. A conservative dead time of 0.15 ms was imposed across all recording files. The idealization protocol may have missed very fast events, and we did not correct for such missed events. Still, our goal was not to define an absolute kinetic mechanism of NMDA receptor gating, but rather to compare kinetic mechanisms across conditions. For our analysis, we assume that missed events were largely equal across different experimental conditions.

Kinetic analysis was performed using the maximum interval likelihood (MIL) algorithm in QuB. State models with increasing open and closed states were constructed and fitted to the recordings until log-likelihood (LL) values improved by <10 LL units/added state. We used a linear fully liganded state model containing three closed states, two desensitized states, and two to four open states (see Figs. 7 A and 8 A) of NMDA receptor gating (Kussius and Popescu, 2009). All recordings of GluN1/GluN2A and double-cysteine–substituted receptors were best fit by five closed and two to four open states (all additional open states branched out of the first one). The open-time components, comprising one common short-duration (O₁) and up to three long-duration (O_2-4) intervals, arise from modal gating of NMDA receptors (Popescu and Auerbach, 2003). The long open events O₃ and O₄ did not occur in every patch, either of wild-type (WT) or double-cysteine–substituted receptors. In addition, for the double-cysteine–substituted receptors, the overall number of modal shifts to these long open states was low, simply because of their low P_o of <0.03. Therefore, we could not perform any statistical analysis of shifts in modal gating. Rather, for ease of analysis and comparison, we aggregated the two to four open-time components into two: one short-duration (O₁) and one long-duration (O₂) component (Fig. 7 A). Time constants and the relative areas of each component, the transition rate constants, as well as mean closed time (MCT) and mean open time (MOT) were averaged for each receptor without and with DTT pretreatment and compared with each other.

2–3 d after injection of ~10 ng cRNA per oocyte, 10 healthy oocytes expressing WT GluN1/GluN2A or cysteine substituted GluN1/GluN2A receptors were selected for membrane purification. For a positive control, we used GluN1(N521C,L777C)/GluN2A(E516C,L780C) receptors that form disulfide bond–stabilized intersubunit dimers (Furukawa et al., 2005). Oocytes were washed with 1× PBS, lysed in 1 ml of lysis buffer (20 mM Tris and 0.5 mM N-ethylmaleimide [NEM]), and centrifuged (3,000 rpm for 3 min at 4°C) to separate out the yolk. Recovered supernatant was centrifuged (40,000 rpm for 10 min at 4°C). The pellet was washed with 1 ml PBS and recentrifuged (40,000 rpm for 10 min at 4°C). The resulting pellet was resuspended in 87 µl of solubilization buffer (20 mM Tris, 0.5 mM NEM, 50 mM NaCl, and 1/1,000 protease inhibitor cocktail) without detergent and bath sonicated in ice water four times (15 s on/15 s off). Detergents (1% Triton X-100 and 0.3% sodium deoxycholate [monohydrate]) were added to the solubilization buffer to a final volume of 100 µl and incubated with gentle agitation at 4°C for 1 h. Solubilized proteins were centrifuged (40,000 rpm for 20 min at 4°C), and the recovered supernatant contained the purified membrane fraction.

Membrane fractions were run on a 5% SDS-PAGE gel under nonreducing conditions. Proteins were transferred to 0.45-mm nitrocellulose membranes by semi-dry transfer (Bio-Rad Laboratories) using Bjerrum-Schafer-Nielsen buffer. Membranes were probed with either mouse anti-NMDAR1 (1:300; Millipore) or rabbit anti-NMDAR2A (1:300; Millipore). Blots were developed with Western Blotting Luminol Reagent (Santa Cruz Biotechnology, Inc.) and exposed to chemiluminescence film (Biomax; Kodak).

Data analysis was done using Igor Pro (WaveMetrics), QuB, and Excel (Microsoft). For analysis and illustration, leak currents were subtracted from total currents. Results are presented as mean ± SEM. An ANOVA was used to define statistical differences. The Tukey or Dunnett tests were used for multiple comparisons of means. Significance was defined at P < 0.05.

In an iGluR subunit, the M3–S2 linker connects the LBD to the M3 transmembrane segments, the main channel-gating element (Fig. 1). To dissect out subunit-dependent mechanisms of pore opening in NMDA receptors, we sought to constrain the gating movements of M3–S2. Therefore, we substituted cysteines in M3–S2 as well as S2–M4 (connecting the LBD to M4) to generate intrasubunit disulfide bonds (Fig. 2). Our experimental design used spontaneously formed disulfide bonds rather than oxidizing agent–induced formation of disulfide bonds because the latter can more readily trap receptors in rarely visited conformations. We tested multiple pairs of M3–S2 and S2–M4 cysteine-substituted positions in GluN1 (Fig. 2 A) and GluN2A (Fig. 2 B), anticipating that spontaneously formed disulfide bonds could be broken by the reducing agent DTT and that, if these constraints affected gating, treatment with DTT would yield significant changes in current amplitudes.

For each subunit, we identified a pair of positions displaying the anticipated phenotype: for GluN1 R645 in M3–S2 and S784 in S2–M4, and for GluN2A Q642 in M3–S2 and K785 in S2–M4. Exposure of the double-cysteine–substituted receptors GluN1(R645C,S784C)/GluN2A or GluN1/GluN2A(Q642C,K785C) (hereafter referred to as GluN1(C,C) and GluN2A(C,C), respectively) to extracellularly applied DTT significantly potentiated whole cell current amplitudes (percent potentiation; mean ± SEM; 151 ± 28%, n = 11, and 102 ± 14%, n = 8, respectively) compared with GluN1/GluN2A (5.6 ± 3.5%, n = 5) (Fig. 2, C and D). (Note that DTT most effectively potentiated current amplitudes of the double-cysteine–substituted receptors when it was applied in the presence of competitive antagonists; see Materials and methods.) In contrast, NMDA receptors containing a corresponding single-cysteine substitution showed no significant DTT-induced current potentiation (Fig. 2 D). Therefore, we conclude that DTT-induced current potentiation in the double-cysteine–substituted receptors results from reduction of spontaneously formed disulfide bonds between substituted cysteines. The observed disulfide cross-linking between M3–S2 and S2–M4 was within a single subunit (intrasubunit) rather than between like subunits (e.g., substituted cysteine in M3–S2 of one GluN1 cross-linking with substituted cysteine in S2–M4 of the other GluN1 subunit), because, in contrast to the positive control, no dimers were detected for either GluN1(C,C)/GluN2A or GluN1/GluN2A(C,C) (Fig. 2 E).

As an initial assessment of whether constraining the M3–S2 linkers affects gating, we used MK801, an irreversible open-channel pore blocker (Huettner and Bean, 1988; Jahr, 1992). The treatment of GluN1/GluN2A, GluN1(C,C)/GluN2A, or GluN1/GluN2A(C,C) with a high concentration (1 µM) of MK801 strongly inhibited current amplitudes (percent inhibition: 95 ± 0.2%, n = 4; 94 ± 1.4%, n = 4; and 92 ± 0.9%, n = 9, respectively) (Fig. 3, A–C). Surprisingly, however, the remaining MK801-resistant current for GluN1(C,C)/GluN2A and GluN1/GluN2A(C,C) was significantly potentiated by DTT applied in the presence of competitive antagonists (percent potentiation: 704 ± 88%, n = 4, and 453 ± 35%, n = 8, respectively), compared with that in GluN1/GluN2A (86 ± 15%, n = 4) (Fig. 3, A–C). This observed effect was specific to DTT, as the application of antagonists alone on MK801-treated GluN1(C,C)/GluN2A or GluN1/GluN2A(C,C) receptors did not produce significant current potentiation (Fig. 3 C). Thus, a population of double-cysteine–substituted receptors is resistant to persistent pore block by MK801 but sensitive to DTT-induced current potentiation. In conjunction with the later single-channel results, we conclude that this MK801-resistant component of double-cysteine–substituted receptors is the portion of cell surface receptors containing intact DTT-sensitive cross-links. In contrast, the portion of cell surface receptors that undergo pore block by MK801 likely contains un–cross-linked disulfides. The application of the strong oxidizing reagent copper(II):phenanthroline did not affect the pre-DTT current amplitudes (not depicted), suggesting that the un–cross-linked disulfides exist as superoxide species, such as −SOH, −SO₂H, or SO₃H forms (Cline et al., 2004).

To investigate further these redox-heterogenous populations of cell surface receptors, we quantified the rate of MK801 block of WT and GluN1(C,C)/GluN2A receptors. A low concentration of MK801 (25 nM) inhibited whole cell current amplitudes of WT receptors, the rate of which was well described by a single-exponential function (time constant τ = 6.8 ± 0.6 s, n = 5) (Fig. 3, D and G). In contrast, MK801-mediated (25 nM) decay of whole cell current amplitudes of GluN1(C,C)/GluN2A was best fitted (see Materials and methods) with a biexponential function (τ_f = 14.8 ± 1.4 s and τ_s = 70.3 ± 5.2, n = 4) (Fig. 3, E and G). DTT treatment of GluN1(C,C)/GluN2A did not affect the kinetics of the biexponential components of MK801 inhibition (τ_f = 12.5 ± 0.9 s and τ_s = 71.9 ± 11, n = 5) (Fig. 3, F and G) but reversed their fractional contributions (post-DTT: τ_f = 37.7 ± 1.9% and τ_s = 62.3 ± 6.4% vs. pre-DTT: τ_f = 82.1 ± 1.9% and τ_s = 17.3 ± 1.9%) (Fig. 3 H), resulting in an increase in the overall rate of MK801 block (Fig. 3 F).

We interpret these data as follows. For GluN1(C,C)/GluN2A cell surface receptors, τ_f, which approaches the faster single-exponential function in WT receptors, underlies MK801 block of the un–cross-linked double-cysteine–substituted receptors. Conversely, τ_s represents MK801 block of the cross-linked receptors. Before DTT, the un–cross-linked and cross-linked receptors account for ~40 and ~60% of the macroscopic current amplitudes. DTT treatment breaks the disulfide bonds in the cross-linked receptors, resulting in ~80% of the macroscopic currents now being carried by un–cross-linked receptors. This biophysical assessment of the properties of MK801 block reveals that the heterogenous population of cell surface double-cysteine–substituted receptors significantly complicates the macroscopic current profile of these receptors. To more cleanly study the gating effects of the intrasubunit cross-links, we therefore used single-channel recordings

Fig. 4 (left) shows representative steady-state single-channel recordings of GluN1/GluN2A (A), GluN1(C,C)/GluN2A (B), or GluN1/GluN2A(C,C) (C) receptors acquired in the cell-attached mode on transiently transfected HEK cells. For each receptor, a parallel set of recordings was made from separate HEK cells pretreated with DTT (Fig. 4, right; see Materials and methods). Currents were recorded in the cell-attached mode at a pipette potential of +100 mV with saturating concentrations of glycine (0.1 mM) and glutamate (1 mM) in the pipette. The recordings shown as well as all others used for subsequent analysis were performed on patches that contained a single channel (see Materials and methods).

Qualitatively, the double-cysteine–substituted receptors, compared with WT GluN1/GluN2A, displayed considerably reduced single-channel activity, with only flickery short-duration openings apparent (Fig. 4, B and C, left). Strikingly, the predominant long-duration openings typical in WT receptors were essentially absent in double-cysteine–substituted receptors (Fig. 4, left, bottom traces). Nevertheless, there were similarities between the single-channel activity profiles; like GluN1/GluN2A, the double-cysteine–substituted receptors showed: (a) bursts of activity separated by long periods of no activity, (b) no obvious subconductance levels, and (c) comparable unitary current amplitudes. These similarities suggest that basic NMDA receptor gating behavior was intact in the double-cysteine–substituted receptors. Importantly, the observed gating differences in the double-cysteine–substituted NMDA receptors were mainly, if not exclusively, a result of the intrasubunit disulfide cross-links because with DTT treatment, the double-cysteine–substituted receptors, like WT, showed comparable high frequency and long-duration channel openings (Fig. 4, right).

Table I summarizes the quantified single-channel properties for WT and cross-linked NMDA receptors. Importantly, cross-linking M3–S2 and S2–M4 in either GluN1 or GluN2A did not affect unitary current amplitudes. Rather, the constraints drastically reduced equilibrium channel open probability (eq. P_o; ~30- and 75-fold, respectively) compared with GluN1/GluN2A. This reduction in equilibrium channel open probability resulted mainly from a decrease in MOT with a corresponding increase in MCT. Interestingly, although MOT was similarly affected, MCT was differentially affected in GluN1(C,C)/GluN2A and GluN1/GluN2A(C,C), hinting at subunit-specific contributions to gating transitions restricted to channel closed periods.

After exposure to DTT, GluN1/GluN2A, GluN1(C,C)/GluN2A, and GluN1/GluN2A(C,C) showed similar equilibrium channel open probability, MCT, and MOT (Table I). This result indicates that the low levels of gating activity in untreated double-cysteine–substituted receptors was because of the physical constraint of the disulfide cross-links between M3–S2 and S2–M4, an effect that can be fully reversed by breaking the disulfide bonds with DTT. Thus, constraining relative movements of M3–S2 and S2–M4 in either GluN1 or GluN2A deters NMDA receptor activation gating. DTT treatment increased single-channel activity, specifically MOT, in WT receptors (Table I). Because activity in GluN1/GluN2A, GluN1(C,C)/GluN2A, and GluN1/GluN2A(C,C) was comparable after DTT treatment, DTT must be having this same effect on the double-cysteine–substituted receptors. Nevertheless, the focus of the present study is on the gating effects of intrasubunit cross-link rather than any additional effects of DTT on WT or double-cysteine–substituted receptors. Hence, we restricted our analysis of the single-channel results to within the DTT-untreated and separately within the DTT-treated sets of recordings, not between them.

In summary, the physical constraint produced by the intrasubunit disulfide cross-links in either GluN1 or GluN2A dramatically impaired pore opening. Thus, the conformational freedom of the M3–S2 and S2–M4 linkers is critical to the energetics of the gating process, coupling ligand binding to ion channel pore opening. Further, when pore opening does occur in receptors with gating constraints in either subunit, they are of a single conductance level indistinguishable from WT receptors, suggesting that even though gating is constrained in only two of the four subunits, the two unconstrained subunits cannot gate independently.

To examine in more detail the effects of the subunit-specific gating constraints, we characterized the durations and relative areas of the individual components within the channel closed and open dwell-time histograms (Fig. 5). The kinetics of NMDA receptor activation contain at least four closed- and two open-time components (Gibb and Colquhoun, 1992; Banke and Traynelis, 2003; Popescu and Auerbach, 2003; Auerbach and Zhou, 2005). The analyzed single-channel records of GluN1/GluN2A (Fig. 5 A), GluN1(C,C)/GluN2A (Fig. 5 B), or GluN1/GluN2A(C,C) (Fig. 5 C) were best fit with five closed- and two to four open-time components (see Materials and methods), suggesting that the basic kinetic mechanism of activation gating is unaffected by the intrasubunit disulfide cross-links. Although two to four open-time components, comprising one common short-duration and up to three long-duration intervals arising from modal gating of NMDA receptors (Popescu and Auerbach, 2003), could be fitted to our single-channel dataset for each receptor, we collapsed them into two components, one short and one long duration, for ease of analysis and comparison.

For WT, the short-duration openings (τ_O1: 0.12 ± 0.01 ms) covered only a small fraction of the channel open time (a_O1: 6 ± 1.1%), whereas the long-duration openings represented the predominant open-time component (τ_O2 and a_O2: 8.7 ± 0.9 ms and 94 ± 1.1%) (Table II). Either GluN1 or GluN2A intrasubunit cross-links most profoundly affected gating by severely shortening the duration of the longer open-time component (τ_O2) (N1(C,C)/N2A, 1.6 ± 0.2 ms; and N1/N2C(C,C), 1.9 ± 0.9 ms) (Fig. 6 A, left). Although the shorter open-time component (τ_O1) was intact, its mean duration (0.25 ± 0.05 ms and 0.45 ± 0.09 ms, respectively) was increased by the intrasubunit cross-links (Fig. 6 A, left).

Of the five closed-time components, the GluN1 and GluN2A intrasubunit cross-links increased the same two long-duration components (τ₃ and τ₄; τ₃: N1/N2A, 4.2 ± 0.4 ms; N1(C,C)/N2A, 38.1 ± 8.6 ms; and N1/N2A(C,C), 55.5 ± 9 ms; and τ₄: 26.9 ± 5.6 ms; 208 ± 45 ms; and 154 ± 27 ms, respectively) (Figs. 6 B, top). The relative area shifted away from the shorter duration (a₁ and a₂) and toward the longer duration (a₄ and a₅) closed-time components (Fig. 6 B, bottom). Intriguingly, the shortest duration closed-time component (τ₁) was prolonged, albeit modestly, exclusively by the GluN1 intrasubunit cross-links (0.14 ± 0.005 ms, 0.25 ± 0.04 ms, and 0.11 ± 0.01 ms, respectively), with its relative area (a₁) reduced to significantly different degrees by the cross-links in the different subunits (20.4 ± 1.5%, 12.6 ± 1.4%, and 3.7 ± 0.3%, respectively) (Fig. 6 B). In summary, constraining relative movements of M3–S2 and S2–M4 in either GluN1 or GluN2A decreases long channel openings and consequently increases receptor dwell times in long-duration channel closed states.

To define subunit-specific contributions to pore-opening steps, we fitted a previously described linear model of NMDA receptor activation (Fig. 7 A) (Popescu and Auerbach, 2003; Auerbach and Zhou, 2005; Kussius and Popescu, 2009) to the idealized sequence of single-channel closed and open times using the MIL method (see Materials and methods). Because all single-channel recordings were done under saturating agonist concentrations, full occupancy of the agonist-binding sites was assumed, and thus explicit agonist-binding steps were excluded from the model. The linear scheme C₃–C₂–C₁–O₁–O₂, comprising the three shortest duration closed-time components (τ₃, τ₂, and τ₁) and the two aggregate open-time components (τ_O1 and τ_O2), represents the central activation pathway. C₅, comprising the longest duration closed-time component (τ₅), represents the main microscopic desensitized state (Dravid et al., 2008). C₄, the other long-duration closed-time component (τ₄), is perhaps also a desensitization-related state (Dravid et al., 2008). This sequential activation gating scheme, with the indicated connectivity of the off-pathway desensitization-related states (Fig. 7 A), fitted best the idealized data from each single-channel record as determined by the highest log (likelihood) value, with rate constants for GluN1/GluN2A receptors within the range of previously published values using comparable recording conditions and configurations (Kussius and Popescu, 2009).

In comparing the rate constants, the GluN1 and GluN2A intrasubunit cross-links had widespread and largely overlapping effects on the NMDA receptor activation mechanism (Fig. 7). All the central activation transitions, C₃→C₂→C₁→O₁→O₂, as well as C₄→C₂, were similarly slowed in either GluN1(C,C)/GluN2A or GluN1/GluN2A(C,C) from those in GluN1/GluN2A (Fig. 6, A and B). The deactivation transitions O₂→O₁ and C₂→C₃ were similarly accelerated in GluN1(C,C)/GluN2A and GluN1/GluN2A(C,C) as compared with those in GluN1/GluN2A (Fig. 7, A and B). Interestingly, the reverse transition, C₁→C₂, was affected in a subunit-specific manner, with only GluN1/GluN2A(C,C) accelerating it (Fig. 7, A and B). Transitions to and from the main desensitization state C₅ were unaffected by the intrasubunit cross-links (Fig. 7, A and B).

Free energy plots of the central activation pathway (excluding the desensitization-related states) showed that the GluN1- or GluN2-specific intrasubunit cross-links generally raised the energy barriers governing the kinetic steps of activation, shifting occupancy away from the open states and toward the longer duration closed states (Fig. 7 C). Notably though, C₂–C₁ exclusively underwent GluN2A-specific alteration, wherein restricting M3–S2 and S2–M4 separation in GluN2A resulted in decreased occupancy and therefore energetic destabilization of C₁ (Fig. 7). In contrast, equivalent gating constraints in GluN1 did not affect C₁ in such a manner.

DTT treatment of GluN1/GluN2A, GluN1(C,C)/GluN2A, and GluN1/GluN2A(C,C) restores similar kinetic parameters of activation gating (Fig. 8). Now the energetics of gating was overwhelmingly shifted toward channel opening (Fig. 8 C), especially for the double-cysteine–substituted ones as compared with their untreated counterparts (Fig. 7 C). Although there appears to be some effects of DTT beyond the reduction of disulfide cross-links, for example, faster rate of C₂→C₁ transition for all three receptor types after DTT treatment (compare Figs. 7 A and 8 A), presently we do not analyze them further. Nevertheless, these results further emphasize the impairment of pore opening generated by the intrasubunit disulfide cross-links and its reversibility through DTT-mediated breakage of such cross-links.

In summary, constraining separation of M3–S2 and S2–M4 in either GluN1 or GluN2A symmetrically slowed the late gating steps (C₁–O₁–O₂), with entry rates into the long-lived open state being most dramatically reduced. Thus, GluN1 and GluN2 undergo tightly coupled conformational changes leading to pore opening. A single pre-open gating step (C₂–C₁), however, appears to be associated exclusively with the GluN2A subunit, suggesting some degree of subunit-independent early gating transitions before converging on concerted pore-opening movements across all subunits.

For NMDA receptors, defining kinetic mechanisms of activation gating has identified key energetic steps, including intermediate states between ligand binding and full channel opening (e.g., Banke and Traynelis, 2003; Auerbach and Zhou, 2005; Schorge et al., 2005; Kussius and Popescu, 2009). These states probably do not represent a single-protein conformation but rather a continuum of transiently stable conformations with similar energy profiles. To characterize the dynamics of activation gating in NMDA receptors, we took the novel approach of constraining the M3–S2 linker through intrasubunit cross-linking with S2–M4. The advantage of this approach is that it places a physical constraint at a specific location in the receptor, in these instances at a presumed pivotal juncture between the LBD and the ion channel. Restricting the relative intrasubunit movements of M3–S2 and S2–M4 greatly affected the energetics of gating, reducing channel open probability by 30–75-fold (Table I), while leaving intact the basic kinetic mechanism (e.g., a similar number of closed-/open-time components) (Figs. 4 and 5, and Table II). Through kinetic analysis, we addressed the contributions of this localized constraint in either the GluN1 or GluN2A subunit to the energetics of the gating process in NMDA receptors.

An open-state iGluR structure does not exist. However, based on the closed-state AMPA receptor structure and ligand-bound LBDs and an open K⁺ channel, a general model of conformational changes occurring during gating has been proposed (Sobolevsky et al., 2009). One feature of this model is that during pore opening, the M3–S2 and S2–M4 linkers within a subunit undergo a rotational and/or linear separation of ~9 Å (Fig. 9 A). Although there are limitations in comparing the state-dependent positioning of these linkers in AMPA and NMDA receptor subunits because of significant differences in their sequences and sizes (e.g., Fig. 9 B), we assume a similar general conformational rearrangements in the NMDA receptor linkers. Hence, by disulfide cross-linking M3–S2 to S2–M4 in an individual NMDA receptor subunit, we covalently constrain their relative separation required for movement of M3 away from the central axis of the pore. Consequently, the most robust effects of the intrasubunit gating constraints was impairment of pore opening by greatly shortening the lifetime of the main long-lived channel open state, leading chiefly to brief or “flickery” openings (Figs. 4–6 and Table II).

Activation gating in fully liganded NMDA receptors can generally be divided into early channel closed/pre-open gating steps (C₃→C₂→C₁) and late gating steps mediating pore opening to its most stable state (C₁→O₁→O₂) (Banke and Traynelis, 2003; Auerbach and Zhou, 2005; Kussius and Popescu, 2009). Gating constraints in either the glycine-binding GluN1 or glutamate-binding GluN2A subunits symmetrically affected the late gating steps mediating full pore opening (C₁–O₁ and O₁–O₂) to largely the same extent (Fig. 7). In addition, although open durations were greatly reduced in NMDA receptors with constrained subunits, these openings had the same conductance level as those found in WT receptors (Table I), rather than displaying prominent subconductance levels. Thus, if pore-opening movements of the M3–S2/M3 linker/transmembrane segment are restricted in one subunit (either GluN1 or GluN2A), pore-opening movements in the other nonconstrained subunits are also blocked. These results demonstrate that the intrasubunit pore-opening movements at the level of the linkers are tightly coupled across all four NMDA receptor subunits.

This tight coupling at the level of the linkers must be a manifestation of concerted gating in NMDA receptors wherein all four ligands—two glycine and two glutamate molecules—have to bind for subsequent channel pore opening to occur (Benveniste and Mayer, 1991a; Clements and Westbrook, 1991; Schorge et al., 2005). Here, we demonstrate that the conformational dynamics of specific structural elements, the M3–S2 and S2–M4 linkers and presumably their associated transmembrane segments M3 and M4, act in concert across all four subunits to open the pore in NMDA receptors. The gating transduction mechanism in NMDA receptors is an integrated process stretching from the LBD to the ion channel, with the linkers’ gating dynamics firmly incorporated across all kinetically detectable states, C₃→C₂→C₁→O₁→O₂. Ligand interactions with and initial lobe closure of the LBD occurs before the receptor reaches the C₃ conformational state (Kussius and Popescu, 2010).

In contrast to concerted pore opening in NMDA receptors, gating in the related AMPA receptor apparently occurs in a subunit-independent manner manifested as openings to multiple conductance levels (Rosenmund et al., 1998; Smith and Howe, 2000; Poon et al., 2010; Prieto and Wollmuth, 2010). Mechanisms underlying pore opening in AMPA receptors are unknown but may include subunit-independent movements of the M3 segments, the reentrant M2 pore loop, and/or other gating elements more proximal to the LBD such as the pre-M1 helix. Using homologous positions at the linkers of the AMPA receptor, for which a structure already exists (Sobolevsky et al., 2009), the cross-linking approach we used here could be a means to test for the mechanisms of subunit-independent gating in AMPA receptors.

Immediately after ligand binding, the NMDA receptor undergoes at least two kinetically distinct channel closed/pre-open gating steps. In the sequential model of NMDA receptor activation, the slower pre-open gating step (C₃→C₂) must occur before the faster one (C₂→C₁) (Fig. 7) (Kussius and Popescu, 2009). Although the gating constraints in either the GluN1 or GluN2A subunit commonly affect overlapping rate constants in these two early gating steps, we do detect significant subunit-specific components. The reverse rate C₁→C₂ was exclusively affected by the GluN2A-gating constraints, being approximately twofold accelerated and resulting in C₁ attaining a more unstable energetic state (Fig. 7). Although we do not presently explore this further, these results suggest some degree of subunit-independent isomerization events in pre-open gating steps of the NMDA receptor.

Such subunit-independent gating steps in the NMDA receptor have been suggested previously using GluN1- and GluN2-specific partial agonists (Banke and Traynelis, 2003; Erreger et al., 2005b), as well as noting GluN2A- versus GluN2B-specific differences in gating kinetics (Erreger et al., 2005a). Furthermore, a crystal structure of an iGluR has revealed key structural asymmetry between subunits in the extracellular domains, especially at the LBD–TMD linkers, transitioning to fourfold symmetry at the TMD (Sobolevsky et al., 2009). The subunit-independent pre-open gating steps, initiated in the LBD subsequent to clam-shell closure, may reflect structural differences in the LBDs where the functional pair is the GluN1/GluN2 heterodimer or the asymmetry between M3–S2s within a dimer pair (Sobolevsky et al., 2009). Hence, the asymmetry between subunits in the pre-open steps may reflect subunit-independent differences between the S2–M4s or relative movements of M3–S2 to S2–M4 in the different subunits. Future kinetic studies of NMDA receptors with constrained gating elements encompassing other LBD–TMD linkers will be needed to fully resolve the structural mechanisms coupling the energetics of ligand binding to channel pore opening.

Our results with the NMDA receptor are consistent both with previous functional experiments and structural features of iGluRs. In NMDA receptors, allosteric interactions occur between the two subunits in the LBDs (Benveniste and Mayer, 1991a,b; Lester et al., 1993; Regalado et al., 2001), presumably initiating concerted gating. However, from the LBD to the TMD, there appears to be some degree of subunit independence ultimately (re)converging on concerted pore-opening actions undertaken by all four subunits. This concerted pore opening in NMDA receptors may have functional significance in Ca²⁺ permeation, pore block, and receptor kinetics. Nevertheless, its biological significance as well as its structural basis remain unclear.

We thank Catherine Salussolia, George Zanazzi, William Borschel, and Drs. Gabriela Popescu and Stephen F. Traynelis for helpful discussions and/or comments on the manuscript. We also thank Janet Allopena for technical assistance with molecular biology.

This work was supported by a National Institutes of Health RO1 grant from National Institute of Mental Health (MH066892 to L.P. Wollmuth), travel support from SUNY REACH (to L.P. Wollmuth and I. Talukder), and an American Heart Association Predoctoral Fellowship (to I. Talukder).

Kenton J. Swartz served as editor.