Role of Conformational Dynamics in α-Amino-3-hydroxy-5-methylisoxazole-4-propionic Acid (AMPA) Receptor Partial Agonism*

Background: Agonist binds to an extracellular agonist-binding domain in AMPA receptors. Results: Willardiines induce a range of cleft closure states in the agonist-binding domain of AMPA receptors. Conclusion: The fraction of the agonist-binding domains in a closed cleft conformation correlates with the extent of activation. Significance: The dynamics and extent of cleft closure in the agonist-binding domain control activation of AMPA receptors. We have investigated the range of cleft closure conformational states that the agonist-binding domains of the α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid (AMPA) receptors occupy when bound to a series of willardiine derivatives using single-molecule FRET. These studies show that the agonist-binding domain exhibits varying degrees of dynamics when bound to the different willardiines with differing efficacies. The chlorowillardiine- and nitrowillardiine-bound form of the agonist-binding domain probes a narrower range of cleft closure states relative to the iodowillardiine bound form of the protein, with the antagonist (αS)-α-amino-3-[(4-carboxyphenyl)methyl]-3,4-dihydro-2,4-dioxo-1(2H)-pyrimidinepropanoic acid (UBP-282)-bound form exhibiting the widest range of cleft closure states. Additionally, the average cleft closure follows the order UBP-282 > iodowillardiine > chlorowillardiine > nitrowillardiine-bound forms of agonist-binding domain. These single-molecule FRET data, along with our previously reported data for the glutamate-bound forms of wild type and T686S mutant proteins, show that the mean currents under nondesensitizing conditions can be directly correlated to the fraction of the agonist-binding domains in the “closed” cleft conformation. These results indicate that channel opening in the AMPA receptors is controlled by both the ability of the agonist to induce cleft closure and the dynamics of the agonist-binding domain when bound to the agonist.

Glutamate receptors mediate excitatory neurotransmission by forming cation selective transmembrane channels upon binding to the neurotransmitter glutamate (1)(2)(3)(4)(5)(6)(7). They are involved in learning and memory and are implicated in neurodegenerative disorders such as Huntington, Parkinson, and Alzheimer diseases and in neurodegeneration associated with stroke and amyotrophic lateral sclerosis (8 -15). Glutamate receptors are classified into three subtypes: ␣-amino-5-methyl-3-hydroxy-4-isoxazole propionate (AMPA), 3 N-methyl-D-aspartate (NMDA), and kainate receptors, based on their sequence, functionality, and pharmacological profiles. The recent full-length structure of the antagonist-bound form of the AMPA receptor shows that the receptor is a dimer of dimers, with each subunit being made up of modular segments of the N-terminal domain, agonist-binding domain, transmembrane segments, and intracellular C-terminal domain (16). This structure, along with FRET investigations of the full-length receptor, establishes that the isolated agonist-binding domain is a good model of the domain in the full-length receptor, thus validating its use in detailed structure and dynamics investigations (17)(18)(19).
There are currently numerous structures available for the agonist-binding domain of the AMPA receptor determined in complex with antagonists, as well as agonists of varying efficacy (20 -25). Based on the x-ray structures, it was initially thought that the extent of cleft closure is the primary mechanism by which agonists mediate receptor activation, i.e., increased cleft closure leads to increased activation (23). However, there were several structures such as those of the glutamate-bound form of the T686A mutant and the structures of the AMPA bound form of the L650T mutant that do not follow this trend (22,(25)(26)(27). Ensemble FRET investigations with the wild type and L650T mutant were consistent with the x-ray structures, further validating these deviations from the cleft closure hypothesis (28,29).
Recent single-molecule FRET (smFRET) investigations of wild type and T686S mutant receptors provide additional insight into the complete structural landscape in terms of cleft closure (30). These studies showed that both the most probable state and the average state probed by the T686S mutant showed a closed cleft conformation consistent with the x-ray structures. However, the T686S mutant protein also probed a wider range of cleft closure states that covered more open states than the glutamate bound wild type receptor. Thus, the probability that the T686S mutant was in the closed state was lower than that of the wild type protein when bound to glutamate, and this decrease in probability could lead to lower activation (27,30). These smFRET investigations emphasized the need to study the overall cleft closure landscape and dynamics of the agonistbinding domain to obtain a better insight into the role of cleft closure in activation.
The smFRET results are also consistent with NMR investigations that show wide variation in the dynamics of the agonistbinding domain upon binding to different agonists. In particular, NMR studies investigating the methyl group dynamics indicate that at slower time scales, willardiines, which are partial agonists of the AMPA receptor, exhibit the largest flexibility in the agonist-binding domain relative to full agonists and antagonists (31,32). Here we have performed smFRET investigations with substituted willardiines bound to the agonistbinding domain and compared the spectrum of states that the protein probes to the previously reported NMR-based dynamics and additionally correlated these results with activation of the receptor.

EXPERIMENTAL PROCEDURES
Purification and Labeling of the Agonist-binding Domain of GluA2 Subunit of the AMPA Receptor-The plasmid expressing the agonist-binding domain of the GluA2 subunit of AMPA receptors was provided by Dr. Eric Gouaux (Oregon Health and Science University). Mutations T394C and S652C were made by standard mutation reactions using Pfu Turbo DNA polymerase (Agilent Technologies). Plasmid expressing this double mutant was transformed, and the protein was expressed in Escherichia coli origami DE3 cells (EMD Chemicals) in LB broth, Miller (Fisher Scientific) containing ampicillin (Sigma Aldrich), kanamycin (Fisher Scientific), and tetracycline (Shelton Scientific) at concentrations of 50, 15, and 12.5 g/ml, respectively. The protein expression was induced using isopropyl ␤-D-1-thiogalactopyranoside (Fisher Scientific) when cells reached an optical density of 0.8 -0.9. After isopropyl ␤-D-1thiogalactopyranoside addition, cells were grown at 20°C for 24 h. The cells were harvested and lysed in 20 mM Tris buffer containing 150 mM NaCl, 1 mM glutamate, 5 mM MgSO 4 , 0.5 mM PMSF, 50 g/ml lysozyme, 125 g/ml sodium deoxycholate, 25 g/ml DNaseI. The lysed cells were centrifuged at 35,000 rpm, 4°C for 45 min. The supernatant was then purified using a HiTrap nickel column (GE Healthcare). The purified protein was dialyzed in phosphate-buffered saline containing 1 mM glutamate. The protein was labeled with a 1:3 ratio of maleimide derivatives of Alexa 555 (donor) and Alexa 647 (acceptor) (Invitrogen), respectively. The ratio was confirmed using absorbance measurements. Protein was allowed to con-jugate with the fluorophores by incubating overnight at 4°C in dark. The protein sample was dialyzed to remove the excess unbound fluorophores in phosphate-buffered saline without glutamate, and the appropriate willardiine derivative (Abcam Biochemicals) was added to the protein. This sample was then treated with sulfolink resin for 30 min. The resin was removed by centrifugation, and supernatant was conjugated with biotinconjugated anti-His antibody (Rockland Immunochemicals) and used for smFRET investigations.
Sample Preparation for smFRET-Standard 22 ϫ 22-mm glass microscope slides were plasma cleaned to remove the organic residues and other impurities from the surface. The surface was then functionalized with aminosilane groups through the Vectabond procedure. In this procedure, the slides were first treated with Vectabond-acetone 1% (w/v) solution (Vector Laboratories) for 5 min, rinsed with molecular biology grade water (HyClone) for 30 s, and dried with an N 2 gas stream. The functionalized area was then exposed with 100:1 mixture of 5-kDa, methoxy-terminated, N-succinimidyl polyethylene glycol (Fluka; 33% w/w PEG in molecular biology grade water) and 5-kDa biotin-terminated PEG (NOF Corporation, 2.5% w/w in molecular biology grade water) in sodium bicarbonate (1% v/v, pH 8.0) buffer for ϳ3 h. A sample chamber was assembled by placing a custom hybriwell chamber (Grace Bio-labs) with the help of two silicon ports (press fit tubing connectors; Grace Bio-labs) on top of the biotin-PEG glass slide (33). The silicon ports provided the inlet and outlet for the flow system. The biotin-PEG chamber was filled with 40 l of 20% w/w streptavidin (Invitrogen) in PBS buffer (pH 7.4) and incubated in the dark for 10 min. The streptavidin acts as linker between the biotin-PEG slide and the biotin-conjugated antihistidine antibody bound to GluA2 subunit. A PBS solution containing ϳ250 nM protein tagged with biotin-conjugated anti-histidine monoclonal antibody was then added in ten 17-l increments into the chamber and incubated for 20 min. The excess protein was then rinsed with PBS buffer.
Oxygen Scavenging System-To minimize the photobleaching and blinking of fluorophores, all of the experiments were performed in the presence of an oxygen scavenger system, consisting of 33% w/w ␤-D-(ϩ)-glucose (Sigma-Aldrich), 1% w/w glucose oxidase, 0.1% v/v catalase (Sigma Aldrich), 1 mM methyl viologen (Sigma Aldrich), and 1 mM ascorbic acid (Sigma Aldrich) in molecular biology grade water saturated with phosphate buffer (34). In addition, 1 mM of the substituted willardiine was also added to the oxygen scavenging system, depending on the experimental conditions. Experimental Setup for smFRET-All single-molecule fluorescence measurements were performed using a custom built confocal microscope (35,36). A 532-nm diode-pumped solid state laser (Coherent, Compass 315M-100 SL) was used for sample excitation. The light was expanded to overfill the back aperture of a Fluar 100ϫ 1.3 NA oil immersion microscope objective lens (Carl Zeiss), which resulted in the expansion of the laser light in a 1/e 2 beam radius of ϳ 250 nm and height of ϳ 1 m, respectively. The sample chamber with the flow system for oxygen scavenger was placed on top of a closed-loop x-y-z piezo stage (P-517.3CL; Physik Instrumente) with 100 ϫ 100 ϫ 20-m travel range and 1-nm specificity (SPM 1000; RHK Technology, Maryville, TN). The power of the laser was controlled as necessary using neutral density filters. Fluorescence was collected and refocused by the same objective and was separated from the excited light by using a dichroic mirror (z532rdc; Chroma Technology). Fluorescence was collected and refocused by the same objective and the excited light was filtered via a notch filter (zet532nf, Chroma Technology). The refocused signal was further passed though a dichroic mirror (640dcxr, Chroma Technology) to split donor emission and acceptor emission by wavelength, and these fluorescence signals were collected by two avalanche photodiodes (SPCM-AQR-15; PerkinElmer Life Sciences). The signal to noise ratio was improved by the use of additional emission filters (NHPF-532.0, Kaiser Optical; and ET585, Chroma Technology) placed in front of the photodiode detectors.
Data Collection and Analysis-To obtain the smFRET trajectories for the individual protein molecules, a 10 ϫ 10-m area of the sample was scanned to spatially locate 20 -25 molecules. The fluorescence signals of the donor and the acceptor were collected until the fluorophores were photobleached. The emission intensity trajectories were collected at 1-ms resolution and later binned up to 10-ms time steps to improve the signal to noise ratio. All of the data were analyzed with programs written in-house using MATLAB (R2009b; Mathworks). The corrected fluorescence signal trajectories were used directly to calculate the apparent FRET efficiency (E A ) using the following equation, where I A and I D correspond to background corrected acceptor and donor fluorescence intensities, respectively. The distance between the two fluorophores was calculated with the following equation, where: r ϭ is the inter-dye distance, and R 0 ϭ is the Förster radius, which, for the Alexa 555-Alexa 647 pair, is ϳ5.1 nm (Molecular Probes). All the raw trajectories were analyzed by the above mentioned processing algorithm, and traces were automatically characterized into single step bleaching, multistep bleaching and high acceptor background. Traces were excluded if they met with the conditions of multistep bleaching and higher acceptor background. Details of the wavelet denoising technique have been described previously (30,37,38).
For experiments requiring multiple solution conditions on the same proteins, a secondary method of data collection and analysis was used. Single-molecule samples were prepared as described above along with two buffer solutions following the stated procedure for oxygen scavenging solutions and using glutamate (1 mM) and iodowillardiine (1 mM) as agonists. A dual syringe pump system initially delivered glutamate containing buffer with a flow rate of 0.001 ml/min followed by both the glutamate and iodowillardiine buffers and then finally iodowillardiine alone. A 30 ϫ 30-m area was repeatedly raster scanned using the same confocal setup as above, with a 30-s repetition rate between images. The images were analyzed using a custom-designed program that locates and tracks single molecules to correct for possible stage drift. Molecules were selected from the final frame of the acceptor channel images to avoid skewing of intensities by photobleaching and to ensure FRET occurred in all molecules during the imaging period.
Electrophysiology-The GluA2-flip plasmid was donated by Dr. Seeburg (Max Planck Institute, Heidelberg, Germany). Endogenous accessible cysteine residues were removed, and the T394C and S652C mutations were introduced into wild type GluA2 as described above. This construct was co-transfected with enhanced GFP into human embryonic kidney 293 tSA201 (HEK 293) cells (ATCC CRL 11268) using the calcium phosphate technique with 1-2 g of GluA2/ml for 10 -12 h. Electrophysiological experiments were performed 48 -72 h later. Alexa 555 (100 nM) was added to the recording dish 10 -20 min prior to recording. Whole cell recordings were obtained using thick walled borosilicate pipettes with resistances of 2-4 M⍀ and filled with solution containing 135 mM CsF, 33 mM CsOH, 2 mM MgCl 2 , 1 mM CaCl 2 , 11 mM EGTA, and 10 mM HEPES (pH 7.4). The extracellular bath solution consisted of (in mM) 150 NaCl, 2.8 KCl, 1.8 CaCl 2 , 1 MgCl 2 , 5 mM glucose, and 10 mM HEPES (pH 7.4). All of the ligand solutions were prepared in extracellular buffer and kept at pH 7.4. Cyclothiazide at a concentration of 100 M was used for all experiments. Cells expressing mutant GluA2 were voltageclamped at Ϫ60 mV, and solutions were locally applied using computer controlled valve switcher (VC-6; Warner Instruments) and homemade application pipette. All of the recordings were performed using an Axon 200B amplifier (Molecular Devices), with data acquired at 10 kHz, low pass filtered at 3 kHz (8-pole Bessel, Ϫ3 dB) and under the control of pCLAMP 10.1 software (Molecular Devices). Two-tailed repeated measures t test was performed for statistical analysis of differences in the responses evoked by the different ligands.  Fig. 1. The data shown in the bar graph are consistent with the numerous previous more detailed investigations and show that the willardiines are partial agonists with iodowillardiine have a lower response relative to chloro-and nitro-willardiines (31,32,39). A two-tailed repeated measures t test was performed for statistical analysis of differences in the measured currents, and a p value of Յ0.05 was considered significant in our experiments. The p values for chlorowillardiine, nitrowillardiine, and iodowillardiine were calculated to be less than 0.0005, 0.0025, and 0.0005, respec-tively, in comparison with mean currents with glutamate. As expected, the antagonist UBP-282 produced no response beyond the base-line noise. These willardiines, along with glutamate, thus provide a spectrum of ligands that allow us to draw correlations between the smFRET-based states and receptor activation.

Functional Characterization of Willardiines-Whole
Conformational Changes in the Agonist-binding Domain-A representative smFRET trace of a single GluA2 subunit agonist-binding domain tagged with donor and acceptor fluorophores is shown in Fig. 2. The sites tagged with the donor and acceptor fluorophores are T394C and S652C, similar to those used in our previous smFRET investigations (30). The smFRET data from a number of such traces (170 -210 traces) for each willardiine were then processed for background and cross-talk correction (35,36) and denoised using wavelet decomposition as described previously (37). The data were then plotted as histograms of fraction of occurrence versus the FRET efficiency to determine the spread of states that the protein explores (Fig. 3). The denoised histograms for the chlorowillardiine, nitrowillardiine, iodowillardiine, and the antagonist UBP-282-bound forms are shown in Fig. 3. The distances calculated for the average state for the glutamate-, UBP-282-, and willardiine-bound forms, based on the FRET efficiencies are listed in Table 1. The average distances follow the rank order Glu Ͻ nitrowillardiine Ͻ chlorowilladiine Ͻ iodowillardiine Ͻ UBP-282. Although the correlation of cleft closure with activation is present for the extreme cases, this does not hold true for nitrowillardiine and cholorwillardiine. Nitrowillardiine and cholorwillardiine have similar electrophysiological responses; however, the average smFRET distances indicate that the nitrowillardiine-bound state is on average slightly more closed relative to chlorowillardiine bound state. A similar breakdown was observed in the crystal structures where no significant changes were observed in the distances at the positions labeled for the FRET measurements between these willardiines (shown in Table 1) (39).
The smFRET studies provide insight into possible reasons for this discrepancy because they are able to provide the spread of closed cleft states that the protein occupies (Fig. 3). Because the time resolution of the smFRET measurements is in milliseconds, the breadth of the histograms reflect the dynamics in the millisecond time scale. The histograms show that the chlorowillardiine-bound form probes a narrower range of cleft closures relative to the nitrowillardiine, whereas the iodowillardiine-bound form exhibits a wider range of cleft closure states (Fig. 3). The willardiine-based antagonist, UBP-282, showed the broadest distribution of closed cleft conformations (Fig. 3d). These data suggest that the cleft is more stabilized when bound to chlorowillar-  diine relative to nitrowillardiine, with the iodowillardiine-and UBP-282-bound forms being the most destabilized, exploring a wide range of cleft closures and consequently spending little time in conformations able to activate the receptor.
To confirm that tethered and tagged ligand-binding domains retained their function during imaging, we performed a displacement experiment. A collection of single ligand-binding domains were imaged in the presence of 1 mM glutamate alone for ϳ20 min. Subsequently, the perfusing buffer was switched to a glutamate and iodowillardiine containing solution for 20 min and finally 1 mM iodowillardiine alone for the final 20 min. As seen in Fig. 4, the average FRET intensity from this experiment progressively decreases from 0.81 in glutamate alone to 0.76 in iodowillardiine. This demonstrates that the labeled and tethered single ligand-binding domains retain their ability to bind and unbind ligands. Single-molecule histograms from this experiment also exhibit a progressive shift from a narrow,   focused distribution in glutamate to a broader distribution in iodowillardiine (Fig. 4) similar to the trends seen in the detailed histograms shown in Fig. 3.

Fraction in Closed Cleft
Conformations versus Activation-Based on the histograms, the fraction of the protein exhibiting efficiencies higher than 0.76 was determined for each of the willardiine bound state. The fraction for the glutamate-bound forms of GluA2 and T686S mutant protein, as well as the apo state of the protein, was also determined from the previous report (30). The FRET efficiency of 0.76 was chosen because it corresponds to a FRET distance of 42 Å, which is less than the most probable (46 Å), as well as the average state (43 Å) of the apo state, and thus is expected to be the least distance at which the cleft closure could trigger channel opening. This fraction of the protein exhibiting efficiencies higher than 0.76 shows a strong correlation with the currents mediated by the agonist, as shown in Fig. 5. The activation shown for the glutamate-bound form of T686S mutant is relative to that obtained with quisqualate on the same mutant (22,25,26). Quisqualate and glutamate mediate similar currents in wild type receptors (22,25,26).
The linear dependence both for the willardiines and the T686S mutant indicates that the fraction of agonist-binding domains in the closed cleft conformation, which takes into account the ability of the agonist to induce cleft closure, as well as the range of states that the protein probes, is a good determinant of the ability of an agonist drive activation rather than the most probable state or the average state that the proteinagonist complex occupies.

DISCUSSION
The modular nature of the AMPA receptors has allowed for the study of the isolated agonist-binding domain by crystallography, NMR, FRET, and FTIR spectroscopy (4, 6, 20 -25, 28, 31, 32, 39 -49). The initial crystal structures indicated a correlation between cleft closure and activation (50). However, later investigations showed a number of partial agonists and mutant proteins that did not follow this trend.
NMR and more recently smFRET experiments, along with functional studies, showed that the stability of closed cleft states and the spread of conformations that the protein explores play a key role in translating cleft closure to activation, thus accounting for some of these discrepancies (30,31,49). Indeed, as seen in Fig. 5, the smFRET data allow us to explicitly determine the fraction of agonist-binding domains that exist in a "productive" closed cleft conformation, which correlates with the extent of activation even in previously "discrepant" cases such as the T686S mutation (30). The willardiines were initially thought to be classical examples for showing a graded cleft closure consistent with their activation (50). However, further studies using crystal structures (39) and NMR showed that the correlation is not as straightforward as first thought (31,49). NMR studies investigating the exchange at the side chain methyl groups showed differences among the various willardiines and indicated that the dynamics must also play a role in the activation. Specifically, these studies show that the number of residues exhibiting exchange was higher for chlorowillardiine relative to iodowillardiine. The smFRET investigations reported here are consistent with this observation, with the chlorowillardiinebound form exhibiting a narrower range of states than the protein probes relative to iodowillardiine-bound form (Fig. 4) and add to these studies by showing the whole range of conformations probed. Additionally, the crystal structures of GluA2 and GluA3 agonist-binding domain in complex with chlorowillardiine show that despite the small size of the chloro substituent, the chlorowillardiine-bound form is on average more open than the glutamate-bound form (39). The nitrowillardiinebound form, despite nitro being a larger substituent, is more closed in six of the nine structures studied relative to the fluorowillardiine-bound form. The larger nitro group was accommodated by the change in the rotamer state of the ␤-carbon of M712 as well as elimination of a water group. Furthermore, the structures showed that the nitrowillardiine-bound form exhibited a wider range of lobe openings relative to the glutamatebound form of the protein. The smFRET data presented here are in agreement with these structures and show that the chlorowillardiine-bound form is on average slightly more closed than the nitrowillardiine-bound form (Table 1) and that the nitrowillardiine-bound form has a broader range of cleft closure states relative to the chlorowillardiine-bound form.
The smFRET data are also consistent with recent crystal structure "lobe-locking" experiments, which found that ligand-binding domains could be trapped in closed conformations when bound by partial agonists or even antagonists (51). Taken together, these data provide strong support for recent proposals that agonist efficacy at AMPA receptors (27), and perhaps kainate receptors (52), is governed not by the extent of closure in a single state of the binding domain but by the relative stability of a range of variously productive closed cleft conformations.