A novel NMDA receptor positive allosteric modulator that acts via the transmembrane domain

Ionotropic glutamate receptors (iGluRs) mediate fast excitatory neurotransmission and are key nervous system drug targets. While diverse pharmacological tools have yielded insight into iGluR extracellular domain function, less is known about molecular mechanisms underlying the ion conduction gating process within the transmembrane domain (TMD). We have discovered a novel NMDAR positive allosteric modulator (PAM), GNE-9278, with a unique binding site on the extracellular surface of the TMD. Mutation of a single residue near the Lurcher motif on GluN1 M3 can convert GNE-9278 modulation from positive to negative, and replacing three AMPAR pre-M1 residues with corresponding NMDAR residues can confer GNE-9278 sensitivity to AMPARs. Modulation by GNE-9278 is state-dependent and significantly alters extracellular domain pharmacology. The unique properties and structural determinants of GNE-9278 reveal new modulatory potential of the iGluR TMD.


Introduction
Glutamate (Glu) is the main excitatory neurotransmitter in the brain and rapid synaptic responses to Glu release are mediated by ionotropic glutamate receptors (iGluRs). The iGluR family consists of four homologous classes of tetrameric receptors: AMPA, NMDA, kainate, and delta receptors (Dingledine et al., 1999). All iGluRs share a modular structural organization with extracellular amino terminal domains (ATDs), ligand-binding domains (LBDs), intracellular domains, and a pore-forming transmembrane domain (TMD) consisting of three transmembrane helices (M1, M3, and M4) as well as a reentrant loop (M2) between M1 and M3 that serves as part of the selectivity filter. As dysfunction of NMDARs in particular has been implicated in many neurological diseases including schizophrenia, epilepsy, and Alzheimer's disease, investigation into the potential for pharmacological manipulation of NMDARs has been pursued with great interest Paoletti et al., 2013;Zhou and Sheng, 2013;Soto et al., 2014). While negative allosteric modulators (NAMs) have been examined for the purpose of normalizing pathological overactivation, positive allosteric modulators (PAMs) could be valuable for correcting hypofunction or enhancing normal physiological function of NMDARs (Coyle et al., 2003;Gonzalez-Burgos and Lewis, 2012).
Much of the insight into the mechanisms of PAM action on iGluRs comes from AMPARs, where multiple classes of PAMs have been described that target the inter-subunit interface of the LBDs (Sun et al., 2002;Jin et al., 2005) and constrain the LBD dimer movement so as to favor the agonist-bound conformation and thereby maintain tension on the linkers that connect to the TMD (Chen et al., 2014). Similarly for NMDARs, we have recently described a class of NMDAR PAM that bind the GluN1/GluN2A LBD interface at an analogous site to where AMPAR PAMs bind (Hackos et al., 2016).
In contrast to the LBD, there is relatively less structural information on the TMD, as there are currently no structures of iGluRs with a TMD in the open pore state, and the recent identification of a NAM binding site at the extracellular aspect of the TMD of AMPARs represents the first published TMD modulator co-crystal structure (Yelshanskaya et al., 2016). Additional insight into the role of this region in channel gating comes from the functional impact of mutations in the TMD. One key region is the highly conserved SYTANLAAF "Lurcher motif" of iGluRs located near the extracellular end of M3. This motif is found where the crossing transmembrane helixes form a physical constriction in the ion channel pore that prevents ion conduction when the channels are in closed or desensitized states (Karakas and Furukawa, 2014;Lee et al., 2014). This region was first identified as important to ion channel gating when it was found that neurodegeneration in the Lurcher mouse results from constitutive channel activity due to an A654T mutation within the SYTANLAAF motif (A8T) of the delta 2 glutamate receptor (GluD2) (Zuo et al., 1997). Subsequently, engineered mutations in this region have been shown to cause gain-of-function phenotypes in GluD2 (Kohda et al., 2000), AMPA (Klein and Howe, 2004;Schmid et al., 2007), and NMDA receptors (Hu and Zheng, 2005;Blanke and VanDongen, 2008;Chang and Kuo, 2008;Murthy et al., 2012), suggesting the importance of this region in iGluR gating. Importantly, study of the structural determinants of GluN2C/D subunit-containing NMDAR potentiation by the modulator CIQ has identified the M1 and pre-M1 regions as contributing to allosteric modulation (Ogden and Traynelis, 2013). Previously, CIQ has represented the only NMDAR PAM with well-characterized TMD structural determinants.
Here we describe the discovery of a new NMDAR PAM, GNE-9278, which has unique modulatory properties and a novel binding site in the TMD. Experiments with TMD mutations demonstrate the positive and negative modulatory potential of this binding site and the transferability of GNE-9278 potentiation to AMPARs. We then used GNE-9278 to examine the impacts of TMD modulation on extracellular domain function and pharmacology.

Cell lines
Doxycycline-inducible HEK293 cell lines that express GluN1/ GluN2A, GluN1/GluN2B, GluN1/GluN2C, GluN1/GluN2D, or GluA2 were used for calcium imaging experiments and a CHO cell line expressing inducible GluN1/GluN2A was used for whole-cell electrophysiology, as previously described (Hackos et al., 2016). Receptor expression was validated by electrophysiology and general cell line authentication and mycoplasma testing were not performed.

Calcium influx assay
Compounds were tested for their ability to potentiate either NMDAR or AMPAR-expressing cell lines in calcium influx assays using a FDSS 7000 in a 384-well format as previously described (Hackos et al., 2016). For NMDAR assays, EC 30 Glu (concentration empirically determined each day the assay was run) and saturating Gly were used to activate the channels. For AMPAR assays, saturating (100 mM) Glu was used for channel activation.

Whole-cell electrophysiology
Doxycycline (5 mg/mL) was added to the CHO cell culture media in the presence of ketamine (1 mM) to induce GluN1/GluN2A NMDAR expression the night before recording. On the day of recording, cells were detached from the culture dish and kept at room temperature with 1 mM ketamine added to the media. Whole-cell patch clamp recordings from cell lines were obtained using a Molecular Devices Axopatch 200B patch clamp amplifier. Extracellular solution for whole-cell recording contained (in mM): 150 NaCl, 3 KCl, 1 CaCl 2 , 5 glucose, 10 HEPES, pH 7.4. Intracellular solution contains (in mM): 140 CsF, 10 NaCl, 1.5 MgCl 2 , 5 EGTA, 10 HEPES, pH 7.2. Rapid solution exchange was achieved using the 16channel Dynaflow Resolve system (Cellectricon).

Two-electrode voltage clamp
Expression of NMDAR and AMPAR channels in Xenopus oocytes was achieved by subcloning the human cDNAs for these channels into the pTNT vector (Clontech). To produce RNA for injection into oocytes, the constructs were linearized, purified, and used as the substrate for T7 RNA polymerase-mediated RNA synthesis (mMessage mMachine T7, Ambion). Mutagenesis was carried out using the Quikchange Lightning Multi kit (Agilent) as per manufacturer's instructions. Xenopus oocytes were purchased from Nasco (LM00935MX; Fort Atkinson, WI) and prepared using standard procedures. In brief, oocytes were digested by type 2 collagenase (720 U/ml final) in OR-2 solution (in mM: 82.5 NaCl, 2.4 KCl, 1 MgCl 2 , 5 HEPES, pH 7.4) for 2.5 h followed by several washes in ND-96 (in mM: 96 NaCl, 2 KCl, 1 MgCl 2 , 1.8 CaCl 2 , 5 HEPES, pH 7.4) after digestion. Oocytes injected with NMDAR mRNAs were incubated at 15 C for 1e3 days before recording. Oocytes injected with AMPAR mRNAs were incubated for about 1 week before recording. Two-electrode voltage clamp was carried out using standard procedure and the data were digitized at 2 kHz. Unless otherwise stated, the perfusion solutions were prepared using Ca 2þ and Mg 2þ free ND-96 plus 300 mM BaCl 2 and 50 mM Gly. The timing of solution exchange was controlled by custom software developed using MATLAB (MathWorks).

Data analysis
Data analysis was done using Clampfit (Molecular Devices) and/ or custom scripts written in MATLAB. The deactivation time constant of the NMDAR current was determined by fitting the decaying current with a single-exponential function. Outliers were removed by Grubbs' test and statistical significance was determined in a two-sided Wilcoxon rank sum test except where otherwise noted in the figure legends. Reported values and error bars in this article represent mean ± SEM.

GluN2A homology model
A GluN1/GluN2A homology model was constructed using the multiple-sequence alignment and homology modeling functions in the Molecular Operating Environment (MOE, 2015.10; Chemical Computing Group Inc., Montreal, Canada). The human GluN1/ GluN2A sequences were first aligned to the respective sequences in GluN1/GluN2B and then a structural model was constructed using the GluN1/GluN2B structure (PDB 4TLM) (Lee et al., 2014) as a template. The model was rendered using the molecular visualization program PyMol (The PyMOL Molecular Graphics System, Version 1.8 Schr€ odinger, LLC).
Whole-cell recordings from a cell line expressing GluN2Acontaining NMDARs using a rapid perfusion system confirmed that GNE-9278 acts as a PAM rather than as an agonist or co-agonist ( Fig. 1D). GNE-9278 increased current in response to pulses of saturating Glu (in the presence of saturating glycine) both by increasing the peak current amplitude and by slowing the deactivation process during Glu washout ( Fig. 1E). At 100 mM Glu and 50 mM glycine (Gly), the maximal GNE-9278-mediated potentiation of the peak amplitude was 2.43 ± 0.2 fold with an EC 50 of 3.06 ± 0.32 mM (Fig. 1E). Notably, the effect of enhanced peak amplitude occurred with higher potency than the effect of slowing of deactivation (Fig. 1E). In addition, outside-out patch recordings with small numbers of channels showed that GNE-9278 potentiated GluN1/GluN2A NMDARs without altering the single channel conductance (data not shown).

GNE-9278 slows deactivation and enhances the potency of both Glu and Gly
The slowing of deactivation following removal of Glu suggests that GNE-9278 could act in part by slowing the agonist off-rate and increasing agonist affinity. To further explore this, we performed experiments at a high GNE-9278 concentration and compared the agonists, D-Glu, L-Glu and L-CCG-IV, which have fast, medium and slow deactivation kinetics (in the absence of GNE-9278), respectively. With these agonists, GNE-9278 showed minimal, moderate, and strong slowing of deactivation, respectively, ( Fig. 2A and B). This is consistent with GNE-9278 enhancing the affinity of each agonist as reflected in a proportional slowing of off-rate (Fig. 2B). In the case of L-CCG-IV, a slow tail current was observed following GNE-9278 washout, which is expected since when GNE-9278 was removed very little L-CCG-IV dissociation had occurred in the presence of the PAM. Consistent with this, the measured exponential time constant of the slow tail (Tau_tail) once GNE-9278 was removed is the same as the deactivation time constant observed in the absence of GNE-9278 (Fig. 2C).
As our previous series of compounds that bind in the LBD selectively shift Glu but not Gly EC 50 of the NMDAR (Hackos et al., 2016), we next determined the effects of GNE-9278 on the EC 50 s of the co-agonists. These experiments showed a significant lowering of the EC 50 for both Glu and Gly in the presence of GNE-9278 ( Fig. 2D and E). The results are consistent with increased affinity of both agonists in the presence of GNE-9278.

GNE-9278 association is rapid and NMDAR activationdependent
Measurements of GNE-9278 effects indicated rapid association/ dissociation kinetics when applied in the presence of saturating Glu (Fig. 3A). However, when tested at lower Glu concentrations, the apparent rate of GNE-9278 association slowed, despite rapid dissociation (Fig. 3A). This Glu-dependence of the rate of GNE-9278 potentiation could be consistent with a requirement of channel activation prior to GNE-9278 binding, with the slow apparent association at low Glu concentrations reflecting the accumulation of activated receptors under low open probability conditions. To further explore this possibility, we pre-applied GNE-9278 and then tested different duration pulses of 1 mM Glu þ1 mM Gly application (in the continual presence of 50 mM GNE-9278; Fig. 3B). Subsaturating Glu/Gly concentrations were used to take advantage of slowed GNE-9278 effects, enabling easier measurement of activation-dependent changes in PAM occupancy over time. Because slowing of channel deactivation increases as a greater fraction of NMDARs are bound by GNE-9278 ( Fig. 1E), compound binding that is dependent on channel opening should be reflected in greater slowing of deactivation with increasing durations of agonist application. On the other hand, if compound binding is state-independent, there should be little relationship between the duration of agonist application and the degree of slowing of deactivation since pre-applied 50 mM GNE-9278 (near saturation) would have reached equilibrium at its binding site prior to Glu/Gly application. These experiments revealed an exponential timedependent slowing of deactivation kinetics with longer agonist pulses ( Fig. 3B and D), suggesting that GNE-9278 binding is significantly enhanced following channel activation/opening. Control experiments with agonist pulses in the absence of GNE-9278 demonstrate that the observed slowing of deactivation is dependent on GNE-9278 ( Fig. 3C and D) and therefore can serve as a measure of GNE-9278 occupancy. The observed enhanced binding of GNE-9278 following channel activation could be due to the binding site becoming exposed in a use/state-dependent manner, and/or the enhanced affinity of the binding site for GNE-9278 in agonist-activated channels.

GNE-9278 does not act at known extracellular domain modulatory sites
As a first step in identifying the binding site for GNE-9278, we examined the ability of GNE-9278 to potentiate GluN1/GluN2A NMDARs after disruption of known modulatory sites. The ATD region of NMDARs is a major site for allosteric modulation. Zinc, endogenous polyamines, and GluN2B antagonists including ifenprodil, all bind the GluN2 ATD or at the interface between GluN1 and GluN2 ATDs (Zhu and Paoletti, 2015). Therefore we tested for the ability of GNE-9278 to potentiate NMDARs lacking the GluN2 ATD, which should destroy known ATD binding sites. These experiments showed that NMDARs with the GluN2A ATD deletion could be potentiated similarly to wild-type NMDARs ( Fig. 4A and B), largely excluding the ATD as a potential GNE-9278 binding location.
We next examined the NMDAR PAM binding site at the LBD dimer interface by testing the effects of the GluN2A mutations T758A and V783F, which have previously been shown to disrupt PAM binding at this site (Hackos et al., 2016). As a side-by-side comparison, we also tested GNE-8016 (compound 29 from ), a GluN2A-selective PAM that acts at the LBD dimer interface. As expected from the LBD co-crystal structure (PDB 5I2N ,) showing GNE-8016 binding at this site ( Fig. 4C), GNE-8016 potentiation was eliminated by these mutations (Fig. 4D). On the other hand, GNE-9278 potentiation was not reduced (Fig. 4D), arguing against GNE-9278 binding at this LBD site. The absence of evidence for ATD or LBD binding sites suggested the possibility that GNE-9278 could be acting in the TMD region.

Identification of specific structural determinants of GNE-9278 potentiation in the TMD
We next performed a mutational scan of likely TMD regions that could be involved in GNE-9278 potentiation. Because mutations or chemical modification within the conserved SYTANLAAF motif can enhance gating of iGluRs, we focused on the residues broadly surrounding this region of M3 in GluN1 and GluN2A. In addition, given the proximity of the M1/pre-M1 helix to this region of M3 in NMDAR crystal structures (Karakas and Furukawa, 2014;Lee et al., 2014), we also examined the adjacent region of GluN1 that includes M1/pre-M1. For this set of experiments, we systematically mutated each residue to alanine, (or leucine in the case of residues that were already alanine), and examined the effects on potentiation of the response to 300 nM Glu (Fig. 5A).
Because mutations could alter basal channel gating and confound interpretation of the effects on GNE-9278 potentiation, we also characterized the baseline effects of each mutation by determining the Glu EC 50 in the absence of PAM application. In this analysis many mutations substantially lowered the Glu EC 50  5B). This increased sensitivity to Glu with some mutations could occlude potentiation and thereby confound identification of residues that are specifically important for GNE-9278 potentiation. Hence, as a control, we measured the effect of each mutation on potentiation by both 1 mM GNE-9278 and 10 mM GNE-8016 (which binds at the LBD interface). These PAM concentrations yield equivalent levels of potentiation in WT NMDARs. To identify residues that were uniquely important for GNE-9278 vs. GNE-8016 potentiation, we looked for mutations that resulted in GNE-9278 potentiation that was meaningfully smaller (at least 30% reduced) and significantly different (p < 0.05) from the level of GNE-8016 potentiation. Experiments testing 64 mutations using these screening criteria identified 5 residues in GluN1 pre-M1 (S549, T550, L551, D552, and F554), 4 residues in GluN1 M3 (Y647, N650, F654, and L655), and one residue in GluN2A M3 (F652) as ones which uniquely reduce GNE-9278 potentiation when mutated (Fig. 5A). We also identified a mutation in GluN2A M3 (A647) that dramatically increased the effects of GNE-9278. Notably, the residues found to uniquely reduce GNE-9278 potentiation are roughly clustered in regions where effects of mutation on Glu-EC 50 were also prominent. These regions also include other residues which when mutated diminished GNE-9278 potentiation, but couldn't be resolved as unique determinants because GNE-8016 potentiation was also similarly diminished (e.g. GluN1 S553). Importantly, some mutations did reduce GNE-9278 potentiation without affecting the Glu EC 50 or reducing GNE-8016 potentiation (most notably the GluN1 mutations T550A and D552A).
As a next step to validate structural determinants of GNE-9278, we measured the GNE-9278 dose-response for key mutations that were found to specifically reduce potentiation in the screen described above (Fig. 6AeD). These experiments were performed at saturating Glu concentrations in order to reduce the confounding effects of the left-shifted Glu EC 50 that was observed for some of the mutants (Fig. 5B). In these experiments GluN1 pre-M1 mutations T550A and D552A were found to significantly increase the EC 50 of GNE-9278, consistent with a direct contribution of these residues to GNE-9278 binding affinity (Fig. 6D). On GluN1 M3, F654A significantly reduced the efficacy of GNE-9278 ( Fig. 6C) without reducing potency (Fig. 6D) and the adjacent L655A mutant completely abolished GNE-9278 potentiation with a small inhibitory effect observed at all concentrations under these conditions (Fig. 6B, gray trace). At the same time, some of the residues identified in the screen caused only subtle reductions of GNE-9278 potentiation (GluN1 S549 and L551), or even showed greater potentiation than wild-type NMDARs (e.g. GluN1 F554 and Y647, Fig. 6C). This could reflect, 1) an increased window for observing potentiation with these mutations due to a lower basal open probability at the saturating Glu concentration and/or 2) false positives from the screen. In either case, as with the broader set of residues identified in the screen, the 4 key residues (GluN1 T550, D552, F654, L655) that confirmed as GNE-9278 structural determinants cluster in the vicinity of where GluN1 pre-M1 and GluN1 M3 are apposed to each other near the constriction of the channel pore (Fig. 6EeG).

Conversion of GNE-9278 modulation from positive to negative by a mutation in GluN1 M3
Of the confirmed structural determinants of GNE-9278 potentiation, GluN1 L655 was particularly interesting as there was a complete loss of potentiation with the alanine mutation even with the highest tested concentration of GNE-9278 (100 mM), and because this residue is located near the Lurcher motif. We further examined this residue by exploring different sized side chain substitutions, including L655V, L655Y, and L655A. In contrast to the near complete loss of GNE-9278 potentiation in GluN2A-containing NMDARs with GluN1 L655A, L655V caused a partial loss of positive modulation and L655Y actually converted GNE-9278 effects into strong negative modulation (Fig. 7AeC). Notably, the GNE-9278 EC 50 for WT and L655V mutant NMDARs, and the IC 50 for the L655Y mutant, were all in the low micromolar range, suggesting this residue is key to the effects of GNE-9278, but not critical to the binding affinity. In contrast to GNE-9278, GNE-8016 still functioned as a positive modulator of GluN2A-containing NMDARs regardless of the mutations at this residue (Fig. 7D). Similarly, when paired with GluN2C, the three different GluN1 L655 mutations disrupted the positive modulation of GluN2C NMDARs by GNE 9278 (Fig. 7E) but not by the GluN2C/D-selective PAM CIQ (Fig. 7F). Together, these results demonstrate the specificity of these residues in controlling the nature of GNE-9278 modulation.

Three Pre-M1 residues are sufficient to confer GNE-9278 potentiation to AMPARs
The change in efficacy without decreasing the potency of GNE-9278 with mutations of GluN1 M3 residues F654 and L655 (Fig. 6) suggest residues on M3 may mediate the effects of GNE-9278 binding without directly contributing to binding affinity. In contrast, alanine mutations of pre-M1 residues T550 and D552 (which do not alter the Glu EC 50 or GNE-8016 mediated potentiation; Fig. 5) resulted in large increases in the GNE-9278 EC 50 (Fig. 6D), suggestive of a direct contribution to binding affinity. Therefore, we tested if we could confer GNE-9278 sensitivity to AMPARs by inserting the GluN1 pre-M1 residues T550-L551-D552 into the corresponding portion of pre-M1 in GluA2 (GVF / TLD) (Fig. 8A). Because AMPARs rapidly desensitize in response to Glu application, we measured Glu-induced currents from AMPARs in the presence of GNE-3419, a compound that binds the LBD of AMPARs and prevents desensitization (Hackos et al., 2016). In these experiments, as expected, GNE-9278 failed to potentiate Gluinduced currents in WT AMPARs (Fig. 8B upper panel and 8C). However, in the chimeric AMPARs containing the GluN1 TLD sequence, significant potentiation by GNE-9278 was observed (Fig. 8B lower panel and 8C). While these results are consistent with dose-response curves (right; n ¼ 12, 6, 10, 10 plates) are shown. The structure of GNE-9278 is shown inset. (B) In contrast to the non-selective iGluR PAM GNE-3419 (Hackos et al., 2016), 125 mM GNE-9278 does not potentiate GluA2 AMPARs containing either the flip or flop isoform (n ¼ 4 plates/compound/isoform). (C) Oocyte voltage-clamp recordings were used to measure GNE-9278 potentiation of currents elicited by 100 mM Glu in the presence of 50 mM Gly (n ¼ 10, 8, 7, and 8 oocytes). Fold potentiation is shown on the left and example recordings on the right. Note that for GluN2C and GluN2D there was a decrease in the current with 100 mM compared to 30 mM GNE-9278, potentially indicating contribution of an inhibitory component as well as the PAM effect at 100 mM. Therefore the 100 mM GNE-9278 data point was not used in PAM EC 50 calculations for GluN2C and GluN2D (open symbols). (D) A representative whole-cell recording from a GluN1/GluN2A-expressing CHO cell is shown. GNE-9278 application enhances the response to Glu þ Gly application but does not serve as a co-agonist when applied with either Glu or Gly alone. (E) 100 mM Glu-evoked whole-cell GluN1/GluN2A currents recorded in the constant presence of 50 mM Gly at different concentrations of GNE-9278 are shown. The maximal potentiation of the peak amplitude is 2.43 ± 0.2 fold with an EC 50 of 3.06 ± 0.32 mM (right panel, black, n ¼ 9 patches). The time constant of deactivation following agonist removal also increases in a dose-dependent manner (t ¼ 265 ± 12, 322 ± 17, 496 ± 30, 1277 ± 111 and 4649 ± 859 ms for 0, 0.5, 2.5, 10, and 50 mM GNE-9278, respectively), but this effect is less potent than the effect on peak current and shows no sign of saturation up to 50 mM GNE-9278 (right panel, red, EC 50 not determined). All data are shown as mean ± SEM. Glu EC 50 shifted from 1.32 ± 0.2 mM in control conditions to 0.11 ± 0.02 mM in the presence of GNE-9278 (n ¼ 9 patches, p ¼ 4.11 Â 10 À5 ). (E) The Gly EC 50 shifted from 1.31 ± 0.11 mM, in control conditions, to 0.12 ± 0.01 mM in the presence of GNE-9278 (n ¼ 10 patches, p ¼ 1.24 Â 10 À4 ). Representative traces with (red) or without (black) GNE-9278 are shown on top of the dose response curves. Two-sided Wilcoxon rank sum test was used and data represent mean ± SEM. *p < 0.05, **p < 0.01. these residues being sufficient for conferring binding to AMPARs, it is also possible that GNE-9278 binds wild-type AMPARs without effect, and these residues are necessary for conferring transduction of the allosteric effect of GNE-9278 binding. This striking ability of 3 pre-M1 residues to confer action of GNE-9278 onto AMPARs confirms understanding of the structural determinants of PAM action in the TMD, and suggests that the GNE-9278 binding site is in the vicinity of the GluN1 pre-M1 helix.

GNE-9278 interacts with modulators that act via extracellular domain sites
While the above experiments clearly indicate a TMD binding site for GNE-9278, we found evidence of a dominant interaction of GNE-9278 with extracellular domain modulators. Interestingly, we noticed that the LBD PAM GNE-8016 seemed to lose its ability to potentiate L655A NMDARs in the presence of GNE-9278 (Fig. 9A).
To follow up on this observation we performed dose-response experiments and found that the maximum effect of GNE-8016 was dramatically reduced in the presence of GNE-9278 (Fig. 9B). Because these experiments were performed on the L655A channels that lack GNE-9278 potentiation, this doesn't reflect occlusion of GNE-8016 potentiation by GNE-9278 (i.e. via a ceiling effect), but rather reflects silent allosteric modulation by GNE-9278, which prevents potentiation by the LBD-acting PAM.
Given the evidence that GNE-9278 modulation enhances agonist affinity and diminishes action of LBD PAMs, we asked if GNE-9278 could also reduce allosteric modulation via the ATD. To test this we examined inhibition of NMDAR current by ifenprodil, a GluN2B-selective negative allosteric modulator that acts by binding to the ATD. These experiments showed a dramatic reduction in inhibition by ifenprodil in the presence of GNE-9278 (Fig. 9C). This further demonstrates the ability of GNE-9278 modulation to alter the normal action of modulators that act via the extracellular domain.

Discussion
GNE-9278 is a novel PAM that potentiates NMDAR peak currents   shows the binding site at the LBD dimer interface between GluN1 and GluN2A. Colors used are consistent with panel A and the orientation has been rotated by 90 to view from above the receptor. (D) Compared to WT, GNE-8016 potentiation is reduced by the key binding site mutations GluN2A T758A (p ¼ 0.008, n ¼ 5, 5 oocytes) and GluN2A V783F (p ¼ 0.008, n ¼ 5, 5 oocytes). In contrast, GluN2A T758A caused a slight enhancement of GNE-9278 potentiation (p ¼ 0.008, n ¼ 5, 5 oocytes) and GluN2A V783F had no effect on GNE-9278 potentiation (p ¼ 1, n ¼ 5, 5 oocytes). Representative traces are shown in the bottom panel. Two-sided Wilcoxon rank sum test was used and bars represent mean ± SEM. **p < 0.01. Hits' from this screen were defined as having potentiation by 1 mM GNE-9278 that was at least 30% reduced and significantly different from potentiation by 10 mM 8016 (t-test p < 0.05, stars above the bars, n ¼ 3e4 oocytes for each mutant, n ¼ 41 oocytes for wild-type). The black dashed line represents no potentiation and the red dashed line indicates the WT level of GNE-9278 potentiation. Representative traces from WT (upper left) and GluN1 T550A (upper right), which affects GNE-9278 but not GNE-8016 potentiation, are shown (inset). (B) Glu EC 50 of each mutant is shown. The hits from panel A are indicated with an arrow. Example traces from WT and GluN1 S553A are shown (inset). All data are shown as mean ± SEM. and slows deactivation kinetics. GNE-9278 shows fast kinetics and state dependency, apparently potently associating with open but not closed channels. While a precise binding site cannot be fully established in the absence of a co-crystal structure, we took a rigorous approach to identify the site of GNE-9278 action. After excluding potential extracellular domain binding sites, a mutational scan identified key structural determinants of GNE-9278 potentiation near the Lurcher motif. Different side chain mutations of the GluN1 L655 residue were able to alter the polarity of modulation and three pre-M1 residues were identified as sufficient to install the binding site into AMPARs, which were normally not potentiated by GNE-9278. Interestingly, GNE-9278 potentiation is associated with significant alterations to NMDAR extracellular domain function and pharmacology.

Structural determinants of GNE-9278 effects
An important aspect of our mutational scan was our use of GNE-8016, which binds at the LBD, as a "control" PAM. This allowed discrimination of specific structural determinants of GNE-9278 potentiation from mutations that simply altered channel gating and thus occlude potentiation in general. Interestingly, many of the mutations in the TMD did alter basal Glu sensitivity of NMDARs. This observation builds on previous work showing effects of various TMD mutations on channel gating (Kashiwagi et al., 2002;Hu and Zheng, 2005;Sobolevsky et al., 2007;Blanke and VanDongen, 2008;Chang and Kuo, 2008;Murthy et al., 2012;Ogden and Traynelis, 2013), and emphasizes the role of both GluN1 pre-M1/ M1 and M3 in channel gating.
Of the 4 confirmed GluN1 structural determinants, the pre-M1 residues (T550 and D552) were particularly important for GNE-9278 potency and thus likely contribute directly to GNE-9278 binding. This is convincingly supported by conferral of GNE-9278 potentiation to AMPARs by transferring the 3-amino acid sequence containing these GluN1 residues into GluA2 (Fig. 8). In contrast to the pre-M1 residues, the two critical M3 residues (F654 and L655) are key to the efficacy of GNE-9278. This is seen with the reduced maximal potentiation by F654A, despite somewhat increased potency. Similarly, while the L655A mutation eliminated potentiation, L655V reduced potentiation without dramatically affecting potency and L655Y caused GNE-9278 to act as a NAM with a similar potency as the PAM effects on WT channels (Fig. 7). These observations suggest that L655 determines the nature of GNE-9278 modulation without affecting binding. Interestingly, GNE-8016 can potentiate L655A NMDARs, but this potentiation is greatly diminished in the presence of GNE-9278, indicating that the apparent loss of GNE-9278 modulation with L655A actually represents neutral/silent allosteric modulation that dominates over GNE-8016 potentiation ( Fig. 9A and B). Overall our results support a model where, upon channel opening, GNE-9278 binds to the TMD with residues on pre-M1 contributing directly to binding affinity. The effects of GNE-9278 binding then propagate through residues located near the extracellular end of the Lurcher motif in M3.

Comparison to other NMDAR modulators that act via the TMD
Interestingly, structural determinants of endogenous neurosteroid inhibition of GluN2B NMDARs include residues within the SYTANLAAF motif, and a binding site created by channel opening has been proposed (Vyklicky et al., 2015). However, the proposed neurosteroid inhibitory binding site is an interior funnel formed by the M3 helices, while the structural determinants on pre-M1 support GNE-9278 interaction with an exterior facing aspect of M3 and the extracellular end of the SYTANLAAF motif. At the same time, L655 is among the residues important for blocking NMDARs by MK-801 and TB-3-4 (Kashiwagi et al., 2002), emphasizing the key importance of this residue in channel function and pharmacology.
Other NMDAR PAMs with structural determinants in the TMD include the NMDAR modulator pregnenolone sulfate (Jang et al., 2004) and the GluN2C/2D-selective compound CIQ (Ogden and Traynelis, 2013). Interestingly, while the structural determinants of pregnenolone potentiation include pre-M4/M4 of GluN2, the structural determinants of CIQ and GNE-9278 include pre-M1/M1 of GluN2, and GluN1, respectively. This raises the possibility that CIQ and GNE-9278 bind distinct but related sites involving GluN2 for CIQ (which exhibits GluN2 subunit selectivity) and GluN1 for GNE-9278 (which is GluN2 non-selective). The distinct properties of GNE-9278 and CIQ, however, suggest very different modes of action. In contrast to GNE-9278, CIQ does not prolong channel opening after agonist removal and has minimal effects on Glu and Gly EC 50 s (Mullasseril et al., 2010). These distinct modes of potentiation by GNE-9278 and CIQ are consistent with their distinct structural determinants.

Mechanism of PAM action
Due to the lack of open pore iGluR structures, attempting to accurately define a binding pose for GNE-9278, which acts on activated NMDARs, is difficult. Nonetheless, based on the similarity of the iGluR pore to an inverted potassium channel pore , we can assume the TMD undergoes a rearrangement and expansion during channel opening, and that the binding site is somehow created or becomes accessible to GNE-9278 on the exterior aspect of the channel pore. It is striking that the equivalent region in AMPARs to the pre-M1/linker region of GluN1, where the critical residues for GNE-9278 potency were found, is very close to the binding site for the non-competitive AMPAR inhibitors GYKI-53655, CP-465022, and perampanel, which has recently been demonstrated by x-ray crystallography (Yelshanskaya et al., 2016). These AMPAR non-competitive inhibitors exhibit higher affinity for agonist-unbound vs. activated receptors (Balannik et al., 2005), and the proposed mechanism of inhibition involves stabilization of a closed state of the AMPAR by compound binding (Balannik et al., 2005;Yelshanskaya et al., 2016). Conversely, GNE-9278 appears to preferentially bind to agonist-bound NMDARs (Fig. 3) and would appear to function by stabilizing an activated state of the NMDAR.
The ability of GNE-9278 to act in the TMD and alter agonist affinity and slow deactivation is reminiscent of the effects of certain NMDAR blockers like 9-amino-acridine (9-AA), which has been shown to trap glutamate and glycine at their binding sites and slow deactivation while inhibiting channels via an open channel block mechanism (Benveniste and Mayer, 1995). GNE-9278 appears to share this property of increasing agonist affinity while stabilizing an open state of NMDARs via action at the TMD, while lacking the blocking effects of 9-AA. This stabilization of a specific state of the NMDAR is reflected in the insensitivity to LBD and ATD allosteric modulators during GNE-9278 modulation (Fig. 9). In addition, we speculate that negative allosteric modulation of GluN1 L655Y NMDARs and silent allosteric modulation of GluN1 L655A NMDARs by GNE-9278 could possibly reflect an open channel block mechanism that is dominant over the PAM effects in the case of L655Y, and offsets the PAM effects in the case of L655A. The GNE-9278 dose-response profile seen with wild-type NMDARs (Fig. 1C) could also reflect a subtle contribution of less potent channel blocking effects superimposed on the predominant PAM effects. In the case of GluN2C and GluN2D receptors in particular, a contribution of a channel blocking effect could be responsible for the diminished PAM effect observed at the highest GNE-9278 concentration.
An interesting aspect of GNE-9278 potentiation is that both type I (increased maximum current) and type II (increased agonist potency and slowing of deactivation) PAM effects are observed (Hackos and Hanson, 2017). Interestingly, the effect on maximum current is more potent (i.e. lower EC 50 ) than the effect on slowing of  Overall effects of treatment type were determined by one-way ANOVA on the ranks, and individual comparisons were made with the two-sided Wilcoxon rank sum test. Data represent mean ± SEM. **p < 0.01, ***p < 0.001. For experiments in panels C and D, n ¼ 8 WT, 5 L655A, 6 L655V and 5 L655Y oocytes, and in panels E and F, n ¼ 11 WT, 7 L655A, 5 L655V and 6 L655Y oocytes. deactivation (Fig. 1E). One hypothetical cause for this phenomenon could be that binding of one PAM molecule is sufficient to increase peak potentiation while binding to two binding sites is required to slow deactivation. On the other hand, this phenomenon can also be explained simply by the observed fast GNE-9278 dissociation kinetics. The enhanced peak current observed in the presence of saturating agonists is expected to simply be proportional to receptor occupancy by GNE-9278. However, the fact that GNE-9278 is a PAM means that in the absence of agonists, GNE-9278 can't potentiate the channels once they close. Consequently, the effects on deactivation will be greatly diminished at low GNE-9278 concentrations since the channel will close rapidly (in the absence of Glu) once the PAM dissociates unless GNE-9278 is able to re-bind prior to channel closure, the probability of which is dependent on the PAM concentration. Accordingly, a simple kinetic model of this process suggests that the extent of slowing of NMDAR deactivation kinetics in the presence of a PAM with fast dissociation kinetics should be inversely proportional to the fraction of unoccupied receptors (Supplemental Fig. 1), which is sufficient to explain the less potent slowing of deactivation compared to the potentiation of peak current. Regardless of the mechanism, this mixture of different types of PAM effects, with slowing of deactivation only occurring at higher concentrations, could represent an interesting therapeutic profile if drug-like compounds with similar properties were to be developed. Overall, the discovery of a novel NMDAR PAM with a TMD 2.06 ± 0.24 to 1.15 ± 0.05 (p ¼ 4 Â 10 À4 , n ¼ 9, 6), while the EC 50 wasn't significantly changed (5.58 ± 1.48 vs. 5.31 ± 1.89 mM). (C) Quantification of ifenprodil inhibition (100 mM Glu, 50 mM Gly) in oocytes expressing GluN1/GluN2B receptors in the absence or presence of GNE-9278 is shown. GNE-9278 resulted in a significant reduction in the maximal inhibition by ifenprodil from 80 ± 2% to 23 ± 4% (p ¼ 6.7 Â 10 À4 , n ¼ 8, 6) and an increase in the IC 50 from 123 ± 9.8 nM to 961 ± 379 nM (p ¼ 1.7 Â 10 À4 ). All data represent mean ± SEM and two-sided Wilcoxon rank sum tests were used to determine the significance.
binding site that can be recreated in AMPARs or mutated to cause NAM effects broadens our knowledge of the modulatory potential of iGluRs. These results combined with the previous studies of AMPAR NAMs that bind the TMD (Balannik et al., 2005;Yelshanskaya et al., 2016), predict that state-dependent binding in this region should allow discovery and development of both positive and negative modulators of both NMDA and AMPA receptors. In addition to being useful for further studies of iGluR structure and function, compounds such as GNE-9278 that act via the TMD will help unlock the full potential of iGluR modulation as a therapeutic approach.

Disclosure/conflict of interest
All authors are current or former Genentech employees.