Allosteric Changes in the NMDA Receptor Associated with Calcium-Dependent Inactivation

Abstract

N-methyl-D-aspartate (NMDA) receptors mediate synaptic excitatory signaling in the mammalian central nervous system by forming calcium-permeable transmembrane channels upon binding glutamate and coagonist glycine. Ca²⁺ influx through NMDA receptors leads to channel inactivation through a process mediated by resident calmodulin bound to the intracellular C-terminal segment of the GluN1 subunit of the receptor. Using single-molecule FRET investigations, we show that in the presence of calcium-calmodulin, the distance across the two GluN1 subunits at the entrance of the first transmembrane segment is shorter and the bilobed cleft of the glycine-binding domain in GluN1 is more closed when bound to glycine and glutamate relative to what is observed in the presence of barium-calmodulin. Consistent with these observations, the glycine deactivation rate is slower in the presence of calcium-calmodulin. Taken together, these results show that the binding of calcium-calmodulin to the C-terminus has long-range allosteric effects on the extracellular segments of the receptor that may contribute to the calcium-dependent inactivation.

Significance

Calcium-induced inactivation of NMDA receptors has been studied extensively using biochemical and functional studies. Calcium binding to the resident calmodulin at the intracellular C-terminal domain of the GluN1 subunit of the receptor has been shown to be responsible for this process. Here, we show that calcium-calmodulin binding to the receptor leads to changes at the extracellular entrance to the transmembrane segments and the glycine binding domain showing long-range allosteric changes that may play a role in the calcium-calmodulin-induced inactivation.

Introduction

N-methyl-D-aspartate (NMDA) receptors belong to the family of ionotropic glutamate receptors. They are obligate heterotetramers consisting of two glycine-binding (GluN1) and two glutamate-binding (GluN2) subunits (1, 2, 3, 4). Upon binding glutamate and coagonist glycine, they form cation-selective channels mediating excitatory neurotransmission at synapses (5,6). Given their role in synaptic signaling, they play a crucial role in physiological processes such as learning and memory (7,8). Dysregulation or overstimulation of NMDA receptors has been implicated in many neuropathological conditions, such as ischemic stroke, autism, depression, and excitotoxic neurodegeneration (5).

One of the unique features of NMDA receptors is their high calcium (Ca²⁺) permeability relative to other ionotropic glutamate receptors (5). The NMDA-receptor-mediated calcium influx plays a vital role in initiating various downstream signaling pathways mediated by kinases, such as protein kinase A and protein kinase C, and phosphatases, such as calcineurin (9,10). In addition to these slower processes, calcium influx also reduces NMDA receptor activation in a fast millisecond process known as calcium-dependent inactivation (CDI) (11, 12, 13). Using computational modeling along with rapid electrophysiological measurements, this process has been shown to be mediated by calcium binding to the resident calmodulin docked on the intracellular segments of the receptors (13). Electrophysiological studies using chimeras have narrowed down the calmodulin binding site mediating this calcium-dependent inactivation to the segment immediately adjacent to the last transmembrane segment of the GluN1 subunit of the NMDA receptor (14, 15, 16). Further biochemical and structural studies used the C-terminal peptide to identify two sites for calmodulin binding on GluN1, amino acids 838–863 (C0) and 875–898 (C1), and showed that apo-calmodulin has a higher affinity for the C1 segment and calcium-bound calmodulin has a higher affinity for C0 (17). Although these functional and biochemical measurements have clearly established that calcium-mediated inactivation occurs because of calcium binding to calmodulin docked on the intracellular domain of the GluN1 subunit, the conformational changes associated with this inhibitory process in the NMDA receptors are yet to be elucidated.

In particular, it is not known whether calcium-calmodulin binding to the receptor induces long-range allosteric changes similar to those observed for other allosteric inhibitors. The inhibitors that bind to the amino-terminal domain cause changes throughout the receptor, including alterations of agonist efficacy at the agonist binding domain, reduction of dynamics at the transmembrane segments, and preference for shorter distances across the GluN1 subunits leading to the closure of the channel (4,18,19). The extensive allosteric network of NMDA receptors is also seen in the case of negative cooperativity between the two agonists—glutamate that binds to the GluN2 subunit and glycine that binds to the GluN1 subunit—with significant differences in conformational dynamics at the second agonist site due to the binding of an agonist at the first site (20).

In this work, we aimed to answer the question of whether a similar long-range allosteric modulation across the transmembrane region could occur during calcium-mediated inhibition in NMDA receptors. We used single-molecule fluorescence resonance energy transfer (smFRET) to study changes at extracellular sites of the first transmembrane segments, as well as changes at the agonist binding domain, induced by calcium-calmodulin binding. Additionally, we used barium that does not bind to calmodulin to demonstrate that the conformational changes, as well as changes in dynamics, are specific to the calcium-bound state of calmodulin. Taken together, we provide an insight into conformational changes involved in the calcium-dependent inactivation of the NMDA receptors, as well as changes in agonist efficacy that have not been documented previously.

Materials and Methods

Molecular biology

Wild-type Rattus norvegicus GluN1 and GluN2A cDNAs were generously provided by Dr. S. Nakanishi (Osaka Bioscience Institute, Osaka, Japan). We first generated “cys-light” constructs (GluN1^∗ and GluN2A^∗) by mutating extracellular non-disulphide-bonded cysteines—C459 in GluN1 and C231, C399, and C460 in GluN2A, to serines. Twin streptavidin tag was introduced in both GluN1^∗ and GluN2A^∗ at the amino-terminus or C-terminus. This tag was used to selectively pull-down the protein of interest onto the streptavidin-coated microscope slide. For smFRET measurements, we introduced cysteines for labeling with the donor and acceptor fluorophores at desired sites. To probe the transmembrane domain of the GluN1 and GluN2A subunits, a cysteine was introduced at the 554 position in GluN1^∗TMD and at the 553 position in GluN2A^∗TMD. To monitor conformational changes at the agonist binding domain (ABD), cysteines were introduced at positions 701 and 507 in GluN1^∗ABD construct and at positions 503 and 701 in GluN2A^∗ABD construct. All the constructs were in pcDNA 3.1 vector. All mutations were checked by Sanger sequencing (Genewiz).

Sample preparation for smFRET

HEK-293T cells (ATCC, Manassas, VA) were maintained in Dulbecco’s modified Eagle’s medium (GenDepot, Barker, TX) supplemented with 10% fetal bovine serum (GenDepot) and penicillin/streptomycin (Sigma-Aldrich, St. Louis, MO) at 37°C and 5% CO₂. The day before an smFRET experiment, the cells were transiently transfected with corresponding GluN1 and GluN2A cDNA-containing pcDNA 3.1 vectors (10 μg total DNA per 10-cm dish) using jetPRIME transfection reagent (Polyplus, Berkeley, CA) according to the manufacturer’s protocol. The GluN1/GluN2 cDNA ratio was kept at 1:3. The transfection medium was replaced 4–5 h post-transfection with Dulbecco’s modified Eagle’s medium supplemented with 30 μM GluN1 antagonist 5,7-dichlorokynurenic acid (Abcam, Cambridge, MA) and 300 μM GluN2A antagonist (2R)-amino-5-phosphonopentanoate (Abcam) to prevent excitotoxicity.

Cells were harvested 24 h post-transfection and washed thrice with the extracellular buffer (135 mM NaCl, 3 mM KCl, 2 mM CaCl₂, 20 mM glucose, and 20 mM HEPES, pH 7.5). Cells were then labeled with premixed donor (Alexa 555-C2-maleimide, 600 nM) and acceptor (Alexa 647-C2-maleimide, 2.4 μM) fluorophores and incubated in the dark at room temperature for 30 min. After labeling, cells were washed thrice with the extracellular buffer and subsequently resuspended in 2 mL of solubilization buffer containing 1% lauryl maltose neopentyl glycol (Anatrace, Maumee, OH), 2 mM cholesteryl hydrogen succinate (MP Biomedicals, Solon, OH), and protease inhibitors (Pierce Protease Inhibitor Mini Tablets; ThermoFisher Scientific, Waltham, MA) in 1× phosphate-buffered saline and nutated at 4°C for 1 h. Solubilized cells were spun down for 1 h at 44,000 rpm at 4°C, and the supernatant was collected and kept on ice until preparation of slides for smFRET.

Preparation of slides for smFRET

Microscope glass slides were prepared as previously described (19, 20, 21, 22, 23, 24, 25). On the day of the experiment, the slides were washed with molecular-grade water and dried with nitrogen. The slides were then treated with 50 μL of short-chain PEG solution (25 mM short-chain 333-Da MS(PEG)4 Methyl-PEG-NHS-Ester Reagent (Thermo Scientific) in 0.1 M sodium bicarbonate) and incubated at room temperature. The slides were then prepared for loading of protein sample as described previously (19, 20, 21, 22, 23, 24, 25). The protein samples were loaded onto the slides and were incubated at 4°C for 20 min. Immediately before imaging and measurement, the slides were flushed twice with the reactive oxygen species scavenging solution (ROXS; 3.3% w/w glucose, 3 units/mL pyranose oxidase, 0.001% w/w catalase, 1 mM ascorbic acid, and 1 mM methyl viologen; all from Sigma-Aldrich, in 1× imaging buffer, pH 7.5) plus 1 mM glutamate and 1 mM glycine. Bovine calmodulin (1 μM; Sigma-Aldrich) was also added to ROXS buffer along with either 2 mM CaCl₂ or 2 mM BaCl₂.

smFRET data acquisition and analysis

The smFRET measurements were performed using a custom-built PicoQuant MicroTime 200 Fluorescence Lifetime Microscope (Picoquant, Berlin, Germany) with the pulsed interleaved excitation set at 80 MHz during the data collection. The sample slide was placed on a scanning x-y-z piezo stage (P-733.2CD; Physik Instrumente, Karlsruhe, Germany) and observed through a 100× objective (100× 1.4 NA; Olympus, Tokyo, Japan) immersed with oil. The donor and acceptor fluorophores were excited using 532 nm (LDH-D-TA-530) and 637 nm (LDH-D-C-640) lasers (Picoquant), respectively. The emitted photons from the sample were detected by two SPAD photodiodes (SPCM CD3516H; Excelitas technologies, Waltham, MA) after passing through the objective. A 550-nm (FF01-582/64; AHF, Tübingen-Pfrondorf, Germany/Semrock, Rochester, NY) emission filter and a 650-nm (2XH690/70; AHF) emission filter were used for donor and acceptor channels, respectively.

Only the molecules showing a single acceptor and donor photobleaching step as well as anticorrelation between acceptor and donor were selected for smFRET analysis, which was performed according to the protocol reported previously (19, 20, 21, 22, 23, 24, 25). The raw fluorescence resonance energy transfer (FRET) efficiencies were calculated using the acceptor and donor intensities. The efficiency traces were further denoised by the wavelet decomposition method using MATLAB (The MathWorks, Natick, MA) and then combined and plotted as histograms depicting the relative occurrence of various FRET efficiencies. The number of molecules selected for generating the compiled histogram for each condition was as follows: GluN1^∗TMD – glu/gly/CaM/Ca²⁺ n = 50, glu/gly/CaM/Ba²⁺ n = 49; glu/gly/CaM/Ca²⁺/CaMIP n = 44, GluN2A^∗TMD- glu/gly/CaM/Ca²⁺ n = 51, glu/gly/CaM/Ba²⁺ n = 51; GluN1^∗ABD – glu/gly/CaM/Ca²⁺ n = 50, glu/gly/CaM/Ba²⁺ n = 46; and GluN2A^∗ABD – glu/gly/CaM/Ca²⁺ n = 55, glu/gly/CaM/Ba²⁺ n = 51.

Step transition and state identification was used to determine the number of discrete conformational states for each condition (26). To ensure unbiased determination of states, Gaussian curve fitting of raw FRET histograms was also performed using Origin software (OriginLAB).

Statistical analysis for FRET measurements

The mean (μ) FRET efficiency was calculated from the probability distribution of the denoised FRET efficiencies using the following equation:

$$\mu =\sum xP\left(x\right),$$ (1)

where x is the FRET efficiency, and P(x) is the probability of occurrence of that FRET efficiency. Standard deviation (σ) was determined using the statistical equation for probability distributions.

$$\sigma =\sqrt{\sum {\left(x-\mu \right)}^{2}P\left(x\right)}$$ (2)

Significance was determined using the p-values calculated based on mean, standard deviation, and number of molecules studied under a given condition, with a p-value of <0.05 considered significant and that of <0.001 considered highly significant.

Electrophysiology

Electrophysiological measurements were performed as previously described (20). Briefly, HEK-293T cells were transfected with indicated NMDA receptor cDNA constructs using Lipofectamine 2000 (Invitrogen) according to the manufacturer’s directions. To prevent excitotoxicity, (2R)-amino-5-phosphonopentanoate (300 μM) and 5,7-dichlorokynurenic acid (30 μM) were added to the transfection media. The whole-cell recordings were performed 24 h post-transfection. Cells were lifted and measured using borosilicate glass pipettes (Sutter Instruments) with internal solution (135 mM CsCl, 35 mM CsOH, 4 mM MgATP, 0.3 mM Na₂GTP, and 0.1 mM EGTA, adjusted to pH 7.4 with CsOH). Cells were exposed to external solution (150 mM NaCl, 5.4 mM KCl, 2 mM CaCl₂, and 10 mM HEPES, pH 7.4 with NaOH) with or without 1 mM glutamate and/or 1 mM glycine as indicated to create the appropriate experimental conditions. For conditions containing barium, 2 mM CaCl₂ was replaced with 2 mM BaCl₂ in the external solution. Traces were recorded at −60 mV hold potential using an Axopatch 200B amplifier (Molecular Devices). The data acquisition frequency was 10 kHz (pCLAMP10 software; Molecular Devices), and traces were then filtered online at 5 kHz frequency.

Outside-out patch recordings were performed 24–48 h post-transfection under conditions described for whole-cell recordings with the exception of the internal solution (135 mM CsF, 33 mM CsOH, 2 mM MgCl₂, 2 mM CaCl₂, 10 mM EGTA, 10 mM HEPES; adjusted to pH 7.4 with CsOH). Currents were recorded with outside-out patches from HEK-293T cells expressing wild-type NMDA (GluN1/GluN2A) receptors using a fast-piezo-driven triple jump method, with calcium (2 mM) and calmodulin (1 μM) in the recording pipette. The acquisition and filtering frequencies were also increased to 50 and 10 kHz, respectively. Rapid application was performed using a homemade device using the triple-barrel pipettes (Vitrocom) as described by Maclean (27). Translation of application pipettes was achieved using a LZ150M piezo translator (Burleigh Instruments) with voltage commands low-pass-filtered (eight-pole Bessel; Frequency Devices) at 100–200 Hz. The exchange time from open time currents ranged from 100 to 250 μs (10–90% rise time). The current decay data were fit to exponential decays. Fitting was performed on individual patches, and the results from each patch were averaged to obtain the reported time constants. Mean, standard deviation, and number of recordings were used to calculate the p-value, which was then used as a measure of statistical significance.

Results

Conformational changes at the transmembrane domain induced by calcium-calmodulin binding

To measure changes in the distance across the transmembrane domain, we used the cysteine-light GluN1 and GluN2 subunits with accessible cysteines—C459 in GluN1 (GluN1^∗) and C231, C399, and C460 in GluN2 (GluN2^∗)—mutated to serines. Cysteines to be labeled with donor and acceptor fluorophores in this cysteine-light background were introduced at F554 on GluN1 (GluN1^∗TMD) and F553 on GluN2 (GluN2^∗TMD) (Fig. 1, A and B). A streptavidin tag was introduced either at the C- or N-terminus to allow pull-down of the receptors onto the smFRET slides. The functionality of these constructs has been previously established (19,20). Electrophysiological measurements of these constructs showed CDI in the presence of calcium and calmodulin (Fig. 1 C; red trace, Fig. 1 D). This inactivation was lower when calcium was replaced with barium (Fig. 1 C; black trace, Fig. 1 D), and the recorded currents were similar to those obtained in the absence of calmodulin (19,20). Having established that these constructs exhibit expected CDI, we next performed smFRET measurements using GluN1^∗TMD/GluN2A^∗ labeled with donor and acceptor fluorophores to determine changes across the transmembrane domain (TMD) of the GluN1 subunit and GluN1^∗/GluN2A^∗TMD to measure changes across the GluN2 subunit. Only traces showing a single donor and single acceptor photobleaching step and anticorrelation between donor and acceptor intensity were used because these represent a single molecule with a donor and acceptor fluorophore (Fig. S1). The smFRET efficiency traces of individual molecules were obtained from intensity traces of donor and acceptor recorded with donor excitation; sample traces for GluN1^∗TMD/GluN2A^∗ are shown in Fig. S2, and those for GluN1^∗/GluN2A^∗-TMD are in Fig. S3. Normalized cumulative histograms were generated using efficiency traces from 45–50 molecules (Fig. 2). The normalized smFRET histograms obtained in the presence of calcium- calmodulin are shown in Fig. 2 A, and those obtained in the presence of barium-calmodulin are shown in Fig. 2 B. The smFRET states were identified using step transition analysis software, a t-test-based statistical method to determine transitions between FRET efficiency points using the wavelet-based denoised smFRET traces (26). These states were further confirmed by Gaussian fits of the observed data, and the FRET values for the states and the sum of Gaussian fits are shown in the histograms. The smFRET histograms of the glutamate-glycine-bound receptor obtained in the presence of barium-calmodulin (Fig. 2 B) represent NMDA receptor bound to apo-calmodulin due to a lack of barium binding to calmodulin. These histograms are similar to the previously reported histograms obtained in the absence of calmodulin (20). The similarity is also confirmed from the lack of significant differences between the mean FRET efficiency (listed in Table 1) for glutamate-glycine-bound receptor obtained in the absence of calmodulin (20) to the mean smFRET efficiency for glutamate-glycine-bound receptor in the presence of calmodulin and barium. This finding indicates that there are no significant changes across the transmembrane pre-M1 sites of the receptor at both GluN1 and GluN2 due to apo-calmodulin binding.

Figure 1.

Transmembrane sites for labeling and functional characterization of smFRET constructs. The Structure of the NMDA receptor showing sites being tagged by donor-acceptor fluorophores in GluN1^∗TMD/GluN2A^∗ (red) and GluN1^∗/GluN2A^∗TMD (blue): (A) side view, (B) top view. (C) Representative whole-cell current recording showing CDI in the presence of calcium-calmodulin (red) relative to barium-calmodulin (black). Calmodulin (1 μM) was added in the recording pipette. (D) Bar graph showing CDI as a ratio of steady-state current to peak current for GluN1^∗TMD/GluN2A^∗ and GluN1^∗/GluN2A^∗TMD. Error bars represent standard deviation. To see this figure in color, go online.

Figure 2.

smFRET efficiency histograms measuring changes at the transmembrane domain. Shown are smFRET efficiency histograms along with three representative FRET efficiency traces measuring (I) the distance across glycine binding GluN1 subunits at site 554 and (II) the distance across glutamate-binding GluN2 subunits at site 553 when bound to 1 mM glutamate and 1 mM glycine in the presence of (A) calcium-calmodulin and (B) barium-calmodulin. Histograms from denoised data are shown in blue or red, and observed data before denoising are shown in gray. The FRET efficiencies for states from Gaussian fits of the observed data are listed, and a summation of fits is shown as a gray line. Observed FRET efficiency traces are shown in gray, and denoised FRET efficiency traces are shown in blue or red. To see this figure in color, go online.

Table 1.

Statistical Analysis for smFRET Measurements: Summary of Mean FRET Efficiencies and Standard Deviation under Different Conditions

Construct	Condition	Mean ± S.D.	p-Value
GluN1∗TMD-GluN2A∗	Glu + Gly	0.75 ± 0.12	NS
	Glu + Gly + Ba2+ + CaM	0.76 ± 0.13	–
	Glu + Gly + Ca2+ + CaM	0.91 ± 0.08	<0.0001
	Glu + Gly + Ca2+ + CaM + CaMIP	0.78 ± 0.13	NS
GluN1∗-GluN2A∗TMD	Glu + Gly	0.81 ± 0.15	NS
	Glu + Gly + Ba2+ + CaM	0.83 ± 0.13	–
	Glu + Gly + Ca2+ + CaM	0.84 ± 0.11	NS
GluN1∗ABD-GluN2A∗	Glu + Gly	0.86 ± 0.06	NS
	Glu + Gly + Ba2+ + CaM	0.86 ± 0.05	–
	Glu + Gly + Ca2+ + CaM	0.89 ± 0.06	0.0075
GluN1∗-GluN2A∗ABD	Glu + Gly	0.85 ± 0.07	NS
	Glu + Gly + Ba2+ + CaM	0.88 ± 0.07	–
	Glu + Gly + Ca2+ + CaM	0.86 ± 0.08	NS

p < 0.05 = significant. p-Values are calculated compared to the (Glu + Gly + Ba²⁺ + CaM) condition for that particular measurement. NS, not significant.

Comparison of the smFRET histogram of glutamate-glycine-bound receptor in the presence of calcium-calmodulin (Fig. 2 A-I) to that in the presence of barium-calmodulin (Fig. 2 B-I) shows that the occupancy is shifted to high FRET states with no occupancy of the low FRET states across GluN1 subunits in the presence of calcium-calmodulin. This indicates that the receptor populates states in which the distances are shorter and possibly a tighter packing across the GluN1 transmembrane segments in the presence of calcium-calmodulin. The low FRET states, corresponding to the longer distances across the GluN1 transmembrane segments, were previously assigned to open-channel states (19,20). Thus, the absence of the low FRET states in the calcium-calmodulin-bound receptor is consistent with the receptor being nearly completely inactive when bound to calcium-calmodulin, as shown in Fig. 1. Consistent with the lack of lower FRET efficiency states, the mean FRET efficiency in the presence of calcium-calmodulin is significantly higher relative to the mean FRET efficiency obtained in the presence of calmodulin and barium (listed in Table 1). Further, to determine whether the observed changes across GluN1 transmembrane domain in the presence of calcium-calmodulin are specific to calmodulin binding to the receptor, we performed control experiments by adding calmodulin inhibitory peptide that has been previously established to have a picomolar affinity to calmodulin (28) and hence is expected to compete with the C-terminus of GluN1 binding to calmodulin. The smFRET histogram for GluN1 transmembrane domain obtained after addition of the peptide removes the effect seen with calcium-calmodulin, and the histogram shows a similar distribution of states (Fig. S4), as seen in the absence of calmodulin and as seen in the presence of barium-calmodulin. This is also reflected in the mean smFRET values being similar to that observed in the absence of calmodulin (Table 1), thus establishing that changes observed are specific to calmodulin binding to the receptor.

The smFRET histograms representing changes across the GluN2 subunit of the glutamate-glycine-bound receptor show only small changes in distribution, and no significant differences are seen with the mean smFRET values (Table 1) between calcium-calmodulin-bound form (Fig. 2 A-II) and in the presence of barium-calmodulin (Fig. 2 B-II). These results suggest that the changes induced by calcium-calmodulin binding are larger across the GluN1 transmembrane segments than across the GluN2 transmembrane segments.

Conformational dynamics across the transmembrane domain induced by calcium-calmodulin binding

To determine whether there are any significant differences in the dynamics of the conformational transitions across the GluN1 and GluN2 transmembrane subunits, we calculated the transition probabilities between the states in the calcium-calmodulin-bound state (Fig. 3 A-I and II, respectively) to that in the presence of barium-calmodulin (Fig. 3 B-I and II, respectively). To be able to compare between different conditions for a given site being investigated, we normalized all transitions to that showing the highest number of transitions, which for FRET studies using GluN1^∗TMD/GluN2A^∗ is the transition from 0.45 to 0.60 FRET efficiency state in the barium-calmodulin condition and for FRET studies using GluN1^∗/GluN2^∗TMD, is the transition from 0.80 to 0.93 FRET efficiency state in the calcium-calmodulin condition. For the GluN1 subunits, in which changes in the conformational landscape were observed, there was a significant reduction in the number of transitions across different conformations in the presence of calcium-calmodulin (Fig. 3 A-I) when compared with the presence of barium-calmodulin (Fig. 3 B-I). This suggests that the binding of calcium-calmodulin lowers the dynamics of the transitions of the GluN1 subunit across the conformations, lowering the probability of the transmembrane segments exploring conformations with longer distances across the transmembrane segments, leading to the channel being primarily in the closed state. The conformational dynamics across the GluN2 subunit at the pre-M1 segment are different in the presence of calcium-calmodulin (Fig. 3 A-II) when compared with those observed in the presence of barium-calmodulin (Fig. 3 B-II), with more transitions between the higher FRET states in the presence of calcium-calmodulin and more transitions between the lower FRET states in the presence of barium-calmodulin. These trends would translate to a higher barrier to transition into the lower FRET states in the presence of calcium-calmodulin relative to that in the presence of apo-calmodulin, consistent with the receptor favoring a closed configuration in the presence of calcium-calmodulin.

Figure 3.

State transition maps obtained from smFRET data at the transmembrane segments. State transitions for FRET across the GluN1 subunit at site 554 (I) and across GluN2 subunits measured at site 553 (II) in the presence of 1 mM glutamate and 1 mM glycine and in the presence of (A) calcium-calmodulin and (B) barium-calmodulin. Normalized number of occurrences represents the transitions occurring from one FRET efficiency state to another FRET efficiency state normalized to the maximal probability observed at a given site. All transitions for GluN1-TMD were normalized relative to the 0.45–0.60 transition under the barium-calmodulin condition, and all transitions for GluN2-TMD were normalized to the 0.80–0.93 transition under the calcium-calmodulin condition. To see this figure in color, go online.

Conformational changes and dynamics at the agonist binding domain induced by calcium-calmodulin binding

Considering significant changes observed at the pre-M1 transmembrane site of the GluN1 subunit and the extensive allosteric network in NMDA receptors, we next performed smFRET measurements across the agonist binding cleft of the GluN1 and GluN2 subunits to determine whether there are any long-range allosteric changes at the agonist binding site induced by calcium-calmodulin binding. For these experiments, we used the GluN1^∗ABD and GluN2^∗ABD constructs in which cysteines for the attachment of fluorophores were introduced across the cleft: at T701-S507 in GluN1^∗ and Q503-M701 in GluN2^∗ (Fig. 4, A and B). GluN1^∗ABD/GluN2A^∗ was used to study changes across GluN1 cleft segment, and GluN1^∗/GluN2A^∗ABD was used to study changes across GluN2 cleft segment (see (20) for detailed strategy). The functionality of these smFRET constructs has been shown previously (20), and here we confirmed that they exhibit expected CDI in the presence of calcium-calmodulin but not in the presence of barium-calmodulin (Fig. 4, C and D).

Figure 4.

Sites for labeling at the agonist binding domain and functional characterization of smFRET constructs. Structure of NMDA receptor showing sites being tagged by donor-acceptor fluorophores in GluN1^∗ABD/GluN2A^∗ (red) and GluN1^∗/GluN2A^∗ABD (blue): (A) side view, (B) individual agonist binding domains. (C) Representative whole-cell current showing CDI in the presence of calcium-calmodulin (red) relative to barium-calmodulin (black). Calmodulin (1 μM) was added in the recording pipette. (D) Bar graph showing CDI as a ratio of steady-state current to peak current for GluN1^∗ABD/GluN2A^∗ and GluN1^∗/GluN2A^∗ABD. Error bars represent standard deviation. To see this figure in color, go online.

Normalized cumulative smFRET histograms generated from 45–50 glutamate-glycine-bound receptor molecules in the presence of calcium-calmodulin and in the presence of barium-calmodulin are shown in Fig. 5, A and B, respectively (representative single-molecule traces showing transitions are shown in Fig. S5 for GluN1^∗ABD and Fig. S6 for GluN2A^∗ABD). The smFRET histograms of both GluN1 as well as GluN2 cleft segments in the presence of barium-calmodulin (Fig. 5 B-I and II, respectively) are similar to those previously observed in the absence of calmodulin (20), suggesting that apo-calmodulin does not induce any significant changes at the agonist-binding domain. Consistent with this observation, the mean smFRET for the two conditions (listed in Table 1) shows no significant statistical difference. This observation is consistent with the measurements at the pre-M1 sites described above (Fig. 2 B), confirming that apo-calmodulin has no transmembrane or long-range effects on the receptor. The smFRET histograms measuring the changes across the cleft at the GluN2 agonist binding domain in the presence of glutamate-glycine and calcium-calmodulin (Fig. 5 A-II) show a small increase in the 0.84 FRET efficiency fraction from those obtained in the presence of barium-calmodulin (Fig. 5 B-II); however, as seen in the mean FRET values, this is not significant. This finding indicates that calcium-calmodulin binding is consistent with observations at the pre-M1 site of GluN2 reported above, which shows no significant changes in the overall landscape (Fig. 2, A-II and B-II). The smFRET histogram of the GluN1 agonist binding domain in the presence of calcium-calmodulin, on the other hand, shows a shift to the higher-efficiency population, representing shorter distances (Fig. 5 A-I) when compared with the populations observed in the presence of barium-calmodulin (Fig. 5 B-I); this shift is also observed in the mean smFRET values and is significant (Table 1). This suggests that upon binding of calcium-calmodulin, the GluN1 agonist binding domain favors a more closed-cleft state.

Figure 5.

smFRET efficiency histograms measuring changes across the cleft at the agonist binding domain. smFRET efficiency histograms along with three representative FRET efficiency traces: (I) distance across glycine binding GluN1 subunit between sites T701 and S507 and (II) distance across glutamate-binding GluN2 subunit between sites Q503 and M701 in the presence of 1 mM glutamate and 1 mM glycine and (A) calcium-calmodulin or (B) barium-calmodulin. Histograms from denoised data are shown in blue and red, and observed data before denoising are in gray. The FRET efficiencies for states from Gaussian fits of the observed data are listed, and a summation of fits is shown as a gray line. Observed FRET efficiency traces are shown in gray, and denoised FRET efficiency traces are shown in blue or red. To see this figure in color, go online.

Dissociation of glycine or glutamate in the presence of calcium-calmodulin

Taking into account that calcium-calmodulin induces a more closed-cleft state in the glycine binding GluN1 subunit, we next explored the effect of calcium-calmodulin on the glycine dissociation rates. To study the deactivation of glycine channels, patches were preincubated with 100 μM glycine and then jumped into 1 mM glycine and 1 mM glutamate solution for 4 ms to activate the receptors before the final jump into 1 mM glutamate with 10 mM 5-methyl-indole 2-carboxylic acid (MeICA) (Fig. 6 A). As previously reported (20), MeICA is a low-affinity glycine site antagonist that prevents the rebinding of any contaminating glycine. The deactivation/dissociation of glycine decay curves could be fit to a primary component with a time constant of 230 ± 20 ms (n = 3) (Fig. 6 C). This time constant is significantly higher than 119 ± 7 ms (Fig. 6 G, +Glu), which was previously reported in the absence of calcium-calmodulin (20). The reduction in dissociation rate for glycine in the presence of calcium-calmodulin is consistent with the smFRET investigations above that showed a propensity for a more closed-cleft state at the GluN1 agonist binding domain in the presence of calcium-calmodulin. We also studied the dissociation rate for glycine in the presence of calcium-calmodulin but in the absence of glutamate. For this experiment, the outside-out patch containing the receptors was pre-equilibrated with glycine and then jumped into MeICA solution for variable times before a test pulse of saturating 1 mM glutamate to determine the fraction of channels still bound to glycine (Fig. 6 B). The rate of dissociation in the absence of glutamate obtained by fitting the decay was 560 ± 40 ms (n = 3) (Fig. 6 C) when the receptors are bound to calcium-calmodulin, which is significantly slower than the 400 ± 50 ms reported in the absence of calcium-calmodulin (Fig. 6 G, −Glu; (20)). This indicates that the higher affinity of glycine in the presence of calcium-calmodulin is independent of the second agonist glutamate being bound.

Figure 6.

Dissociation rates of glycine or glutamate. Current recordings from outside-out patches of HEK-293T cells expressing wild-type GluN1/GluN2A receptor to estimate the rate of glycine and glutamate dissociation and deactivation with calcium-calmodulin in the patch pipette. Shown are the (A) deactivation of channel upon glycine removal in the presence of glutamate; (B) glycine dissociation in the absence of glutamate obtained by jumping into solution with MeICA for variable times, followed by test pulses in the presence of glutamate; and (C) summary across the patches from all experiments in (A) and (B). Shown are the (D) deactivation of channel upon glutamate removal in the presence of glycine, error bars represent standard deviation; (E) glutamate dissociation in the absence of glycine obtained by jumping into solution with MeICA for variable time followed by test pulses in the presence of glycine; and (F) summary across the patches from all experiments in (D) and (E), error bars represent standard deviation. (G) A summary of deactivation time constants for glycine for all the experiments with and without glutamate is shown. (H) A summary of deactivation time constants for glutamate for all the experiments with and without glycine is shown. ^∗p < 0.01. ^∗∗∗p < 0.001. Data obtained in the absence of calmodulin were reported previously (20). To see this figure in color, go online.

Similar investigations were performed to study the rate of glutamate dissociation in the presence and absence of glycine using outside patches with calcium-calmodulin in the pipette (Fig. 6, D and E). The decays were fitted with primary time constants of 40 ± 3 (n = 4) and 106 ± 4 (n = 4) in the presence and absence of glycine, respectively (Fig. 6 F). Contrary to the glycine dissociation rate in the presence of glutamate, which decreased in the presence of calcium-calmodulin relative to the values in the absence of calcium-calmodulin, there were no significant changes in the glutamate dissociation rate in the presence of glycine between the two conditions (Fig. 6 H, +Gly; (20)), suggesting that calcium-calmodulin binding does not significantly alter the rate of dissociation of glutamate when the coagonist is present. This observation is again in agreement with the data above showing a lack of any significant changes at the GluN2 agonist binding domain in the presence of calcium-calmodulin. On the other hand, the glutamate dissociation rate in the absence of glycine was slower in the presence of calcium-calmodulin (Fig. 6 H, −Gly). However, the observed difference was not as high as that observed for glycine.

Discussion

The smFRET measurements of the GluN1 and GluN2A TMDs showed that the pre-M1 GluN1 transmembrane segments undergo a significant conformational shift toward higher FRET/shorter distances/closed states of the receptor in the presence of calcium-calmodulin. However, no significant difference in the conformational landscape of the pre-M1 GluN2A transmembrane segments was observed between the calcium-calmodulin and apo-calmodulin (barium) bound states. The pre-M1 helix is a short α-helix located between the agonist-binding domain and the first transmembrane helix in the primary sequence of the protein and is oriented parallel with the surface of the lipid bilayer. The arrangement of the four pre-M1 helices around the ion channel pore forms a sort of “collar” around the transmembrane segments on the extracellular side of the receptor. The pre-M1 helix engages in interactions with the transmembrane helices of its own subunit, including the M1, M3, and M4 helices (29,30). Thus, the pre-M1 helix forms a coordinated network of residues involving the different transmembrane helices, which can link the conformational changes of one helix to that of the other helices and enable the transduction of allosteric signals among the transmembrane helices.

The calmodulin binding sites on the NMDA receptor are located on the C-terminal domain of the GluN1 subunit (14), after the M4 helix in the primary sequence, putting them on the intracellular side of the plasma membrane. Based on the presented data, we propose that the binding of calcium-calmodulin to these intracellular sites on the GluN1 C-terminal domain causes conformational changes in the nearby M4 transmembrane helices of the GluN1 subunits. These conformational changes in the M4 helices of the GluN1 subunits are then transduced across the plasma membrane and lead to conformational changes in the pre-M1 helices of the GluN1 subunits that interact with the extracellular ends of the M4 helices. The rearrangement of the pre-M1 helices in response to the calcium-calmodulin-induced movement of the M4 helices then leads to the reduction in the distance between the pre-M1 helices of the GluN1 subunits, manifested as the loss of the lower FRET states relative to the apo-calmodulin bound form of the NMDA receptor. The lack of any significant conformational changes at the GluN2A pre-M1 helices suggests that because GluN2A subunits do not bind calcium-calmodulin directly, they have no calmodulin driving the conformational changes that are transduced through the M4 helix to the pre-M1 helix.

The idea that small perturbations in the pre-M1 helix could lead to significant changes in the activity and properties of the NMDA receptor is supported by several facts. The GluN2A pre-M1 helix has been shown to be the site of several disease-associated mutations (29,31), indicating that a single amino acid change in the GluN2A pre-M1 helix could be sufficient to alter channel properties significantly enough to cause neurological issues. Indeed, mutational analyses of the pre-M1 helix of the GluN2A subunit revealed alterations of receptor properties such as glycine and glutamate EC₅₀, rate of deactivation, peak amplitude, open probability, and open time (29). Mutations in the GluN2A pre-M1 helix can also alter the desensitization profile of the receptor. When the F553 in GluN2A is replaced with leucine, a residue located at the corresponding site in the closely related AMPA receptors, the desensitization profile of mutated NMDA resembles that of an AMPA receptor (32). Additionally, mutations in the pre-M1 helix of the AMPA receptor have been shown to alter deactivation and desensitization time constants (30). Those studies demonstrate that the pre-M1 helix plays a significant role in determining channel activity and properties not only in the NMDA receptor but also in other ionotropic glutamate receptors. Given the significant changes that can be induced in single-channel receptor activity through relatively small changes in the pre-M1 helix of the GluN2A subunit or the AMPA receptor, it seems reasonable to extrapolate that the pre-M1 helix of GluN1 could have the ability to modulate the activity of the NMDA receptor in response to M4-mediated effects induced by calcium-calmodulin binding to the GluN1 C-terminal domain.

To reveal the allosteric conformational changes induced in the extracellular domains of the receptor by calcium-calmodulin binding to the intracellular C-terminal domain, we evaluated the effect of calcium-calmodulin binding on the GluN1 and GluN2A agonist-binding domains. One fluorophore attachment site was placed on each side of the clamshell-shaped agonist-binding domain to measure the degree of closure of the agonist-binding cleft that is found between the two lobes of this domain. Although binding of apo-calmodulin to the receptor did not cause significant conformational changes in either the GluN1 or the GluN2A agonist-binding domain, the binding of calcium-calmodulin did induce conformational changes in the GluN1 agonist-binding domain but not in the GluN2A agonist-binding domain. These observations are consistent with our transmembrane domain observations in which calcium-calmodulin altered only the conformational landscape of GluN1 and apo-calmodulin had no effect on either GluN1 or GluN2A. We had also shown that the higher smFRET/shorter distances/closed-cleft states of GluN1 translate to lower glycine dissociation rates. Taken together, the presented study further emphasizes the intricate allosteric network in the NMDA receptors and provides evidence for communication across the membrane in response to calcium-calmodulin binding leading to the increased glycine affinity.

Author Contributions

N.K.B. did the molecular biology, acquired and analyzed smFRET data, and wrote the article. E.C. acquired and analyzed the electrophysiology experiments. R.J.D. assisted with smFRET experiments and wrote and edited the article. V.B. assisted with acquisition and analysis of smFRET experiments. V.J. designed the experiments, analyzed smFRET data, and wrote and edited the article.

Editor: Meyer Jackson.