Structural Insights into GluK3-kainate Receptor Desensitization and Recovery

GluK3-kainate receptors are atypical members of the iGluR family that reside at both the pre- and postsynapse and play key role in regulation of synaptic transmission. For better understanding of structural changes that underlie receptor recovery from desensitized state, GluK3 receptors were trapped in desensitized and resting/closed states and structures analyzed using single particle cryo-electron microscopy. We show that receptor recovery from desensitization requires major rearrangements of the ligand binding domains (LBD) while the amino terminal (ATD) and transmembrane domains remain virtually unaltered. While, the desensitized GluK3 has domain organization as seen earlier for another kainate receptor-GluK2, antagonist bound GluK3 trapped a partially “recovered” state with only two LBD domains in dimeric arrangement necessary for receptor activation. Using these structures as guide, we show that the N-linked glycans at the interface of GluK3 ATD and LBD likely mediate inter-domain interactions and attune receptor-gating properties. Mutational analysis also identifies putative N-glycan interacting residues. These results provide a molecular framework for understanding gating properties unique to GluK3 and identify role of N-linked glycosylation in their modulation.


Main
Ionotropic glutamate receptors (iGluRs) mediate majority of fast excitatory neurotransmission at the chemical synapses in the central nervous system (CNS). Due to this pivotal role, their dysfunction is implicated in a wide range of neurological disorders such as schizophrenia, neuro-excitatory disorders, epilepsy, neurodegenerative, developmental disorders, neuropathic pain etc 1  KARs are abundantly expressed in hippocampus and cerebellum; display characteristic slow kinetics, are mainly involved in long-term memory formation and motor control 5,6 . They are divided into two families consisting of the "low-affinity" glutamate binding subunits (GluK1-GluK3) that form functional homomeric ion channels and "high-affinity" subunits (GluK4 and GluK5) that form functional receptors only on assembly with "low-affinity" subunits [7][8][9] .
Of all the KARs, GluK3 is one of the least studied subunit with respect to structurefunction analysis. However, investigations on GluK3 knockout mice have clearly shown their important role at presynaptic sites in the hippocampal mossy fiber synapse 10 . Electrophysiological assays have shown that unlike homomeric GluK2 or heteromeric GluK2/K5 receptors, the desensitization rates of partially bound homomeric GluK3 receptors are much faster than those of fully bound states 11 .
Further, GluK3 receptors are activated by higher glutamate concentrations compared to other KARs and are potentiated by zinc 12 [16][17][18] however the mechanism of this regulation is unclear. The polar, charged and complex sugar chains present in Nglycans may interact with other synaptic proteins or lay in close proximity to the functional domain of iGluRs 19 thereby affecting receptor functionality such as deactivation, desensitization and recovery from desensitization 20 . N-glycans have also been shown to be important for efficient assembly and trafficking of AMPA 21,22 NMDA 23,24 and KA receptors 20 .
In this study, we expressed and purified intact homotetrameric rat GluK3 receptors from baculovirus infected HEK293 GnTIcells, trapped them in agonist and antagonist bound states and determined their structures using single particle cryoelectron microscopy. We have also explored the involvement of N-glycosylation sites in the receptor function using structural information of GluK3 and electrophysiology.
These data shed light upon molecular mechanism of the transition between these two states during the gating cycle of receptor and the importance of N-glycosylation for GluK3 function.

Receptor purification and structure determination
The wild type GluK3 receptor was optimized to improve its expression and stability in heterologous expression system as elucidated in Fig. 1a and Supplementary Fig.   1. The final construct referred to as GluK3 EM (Fig. 1a) henceforth was optimized using the fluorescence-detection size exclusion chromatography (FSEC) 25 and had a symmetrical profile corresponding to tetramer on size exclusion chromatography ( Fig. 1b). GluK3 EM exhibited gating profile similar to wild type (WT) GluK3 receptors as validated by whole-cell patch clamp recordings in HEK293T cells. It had similar rectification properties, rates of entry into desensitization and recovery from desensitized state (Fig. 1c-e, Table 1). Our electrophysiological recordings are also consistent with previous reports 11 .
In order to understand the structural basis of the receptor rearrangements during its gating cycle, we elucidated the GluK3 EM structure in desensitized and closed state using single particle cryo-EM. These states were captured in presence of agonist 2S, 4R-4-methylglutamate (SYM) 26 and antagonist UBP310 27 respectively ( Supplementary Fig. 2). Reference-free 2D classification obtained after cryo-EM data processing for both the complexes showed good distribution of different receptor orientations as well as identifiable external features resembling glutamate receptor (Supplementary Fig. 3a and b). Various steps of EM data processing and 3D reconstruction of the complexes reveals similar receptor architecture as previously observed for GluA2 28,29 and GluK2 30, 31 ( Fig. 2; Supplementary Fig. 4a and b) Fig. 5 and 6). Details of cryo-EM data collection and refinement are shown in Table   2. Our map shows well-defined features for the ATD and LBD domains while the S1-M1 and M3-S2 linker are poorly resolved (Supplementary Fig. 7). Co-ordinates of previously reported GluK3 ATD (PDB ID: 3OLZ) 14  LBD-LY complex (PDB ID: 5CMK) 31 . These individual models were rigid body fitted into EM maps of GluK3-SYM and GluK3-UBP310 to generate tetrameric models in UCSF-Chimera 32 and subsequently subjected to several rounds of real space refinement as implemented in Phenix software suite 33 .

Desensitized and closed state GluK3 receptor structure
The desensitized and closed-state structures have a similar three-layered assembly for the ATD, LBD and TM domains as reported earlier for other iGluRs. Receptor assembly is mainly mediated by ATD as dimer of dimer (AB and CD) and then undergoes domain swapping at the LBD layer (BC and AD) as observed for AMPA 28,34,29 and kainate 30,31 receptors.
The arrangement of ATD and TM domains in both GluK3-SYM and GluK3-UBP310 complex is similar conformation with an apparent 2-fold symmetry at dimer and dimer of dimers interface of ATD resulting into N-shaped arrangement; and 4-fold symmetry of the TM domains ( Fig. 2 and 3a-d; Supplementary Fig. 8). This infers that the dimer of dimer at the ATD level is intact in both resting and closed state. The recently reported O-shaped arrangement of the ATD layer was not observed in our study 35 . Consistent with this, the parallelogram formed by joining Cα atoms of His 3 from each subunit at top and Ile 385 at the base has similar dimension in desensitized and closed state ( Fig. 3b and c). This also suggests that there are no major structural changes at the ATD layer in order to undergo transition from desensitized to closed state. The resolution of our EM maps are adequate to (Supplementary Fig. 7 (Fig. 3i-l).
In contrast to ATD and TM domains, comparison of the SYM and UBP310 bound structures show major conformational changes at the LBD layer. LBD in GluK3-SYM bound state adopts an apparent 4-fold symmetry (Fig. 3e) and is similar to that reported for GluK2 receptors 30,31 (Supplementary Fig. 9). Thus, the progressively reducing parallelograms formed by joining Cα atoms of Leu 405, Lys 502 and Ser 668 at top, middle and at the base of LBD have dimensions indicative of a symmetric arrangement of both upper and lower LBD lobes ( Fig. 3e-f). In the GluK3-UBP310 complex however, we observe previously unseen asymmetric arrangement for kainate receptors trapped in a putative closed state. Interestingly, subunits B and C form the LBD dimer characteristic of an antagonist-bound form seen in both GluA2 and GluK2 receptors; while the subunits A and D are separated and exist in a desensitized-like state ( Fig. 3 h). Due to such arrangement, joining of Cα atoms of Leu 405, Lys 502 and Ser 668 at top, middle and at the base of LBD forms an asymmetric trapezium of reducing dimensions shown in Fig. 3g-h. It is interesting to note that all the four LBD domains are in an extended cleft conformation characteristic of an antagonistbound state in UBP310 complex; whereas SYM bound density map reveals LBD 'clamshell' closure ( Fig. 3d and Supplementary Fig. 10). It is possible that we have trapped an intermediate between desensitized and "fully-recovered" state. Since, our cryo-EM analysis does not reveal any 2D or 3D class similar to "fully-recovered" state observed in GluK2 and GluA2, where both LBDs are arranged as two dimers with 2-fold symmetry, the trapped structure likely represents a stable conformation. In addition, the LBD arrangement in GluK3-UPB310 complex is similar to that observed in GluA2 complex with partial agonist fluorowillardine 29 . Hence, it is likely that the resting state in GluK3 with both the LBD dimers recovered may not be as stable compared to other AMPA and kainate receptors of known structures. It is also important to note that we have imaged the receptors saturated with ~100 fold excess concentration of UBP310 than its EC 50 value calculated via electrophysiology 27 . It is likely that this partial recovery contributes to low glutamate sensitivity of GluK3.
In both desensitized and closed/resting state, the ligand binding is uncoupled from the transmembrane domain as strain on the linkers connecting LBD to TM domains are relaxed by extended conformation of antagonist bound LBD (even while they are in dimeric configuration). In desensitized state, the LBD-TM linker strain due to bound agonist is relaxed via rearrangements of LBD from a dimer of dimer scheme to a pesudo-4 fold symmetric arrangement, where the LBD dimers are disrupted (Fig. 2).
In concordance with this, the TM domains in both the structures adopt similar closed pore configuration.

Transitions from desensitized to resting/closed state require large conformational changes in LBD and the LBD-TM linkers
In the gating cycle, post-desensitization, receptor needs to recover to a closed/resting state, where the LBDs couple as 2-dimers for the next cycle of activation. Hence, in a "fully-recovered" receptor, the four LBDs should rearrange into two dimers as observed previously in GluK2 31 and GluA2 28, 29 closed-state structures to enable coupling of agonist binding with channel opening 36 . In order to understand this transition, we compared SYM and UBP310 bound structures. It was observed that there is very little change in the conformations of the ATD and TM domains. On the contrary, major changes take place in the LBD layer and in the linkers connecting S1-M1, M3-S2 and S2-M4. Surprisingly, ligand binding domains of only subunits B and C return to a dimeric state, while subunits A and D still remain in a desensitized-like state (Fig. 4). The LBD-distal subunit B swings clockwise by ∼105° in the horizontal plane, while the LBD-proximal subunit C rotates anticlockwise by ∼16° to achieve dimeric configuration ( Fig. 4a-b). In contrast to this, the LBD of subunit A and D undergo smaller degree of anti-clockwise rotation by ~7.7° and ~17.5° respectively and hence are unable to achieve a dimeric configuration ( Fig. 4a-b). In order to further analyze structural changes in detail, we aligned both receptors at TM domains and calculated center of masses (COM) for ATD and S1; S2 lobes of LBDs to measure their displacement in going from SYM to UBP310 bound state. We observe that receptor subunits A and D swing laterally by ~19.5 -21 Å at the ATD level and ~9.7 -12 Å at the LBD-S1 to accommodate this asymmetric arrangement of LBD in UBP310 bound state. While smaller movements of ~3.9 to 5.8 Å, are observed in S2 lobe ( Fig. 4c -e). On the other hand, subunits B/C that go from monomeric to dimeric LBD configuration swing laterally by ~20.9 -22.5 Å at the ATD level. Interestingly, the separation between COMs for the B and C ATD remains unchanged between SYM and UBP310 bound forms suggesting a rigid body movement of the entire ATD layer ( Fig. 4f -h). The movements are more pronounced and asymmetric in the LBD layer for B/C subunits where the LBD-S1 lobe in subunit B translates horizontally by 21.6 Å while the subunit C moves only by ~9.5 Å. Further, the S2 lobes move towards center by 23.6 Å and 2.7 Å, respectively. As a result of this, the separation between S1 lobes in B and C reduces from 51.3 Å to 35.2 Å and that for S2 lobes reduces from 37.2 Å to 25.8 Å ( Fig. 4f-h). The restricted movement of LBD for distal subunit C and larger movements of proximal subunit B leads to BC LBD dimer formation. We also compared the distal A/C and proximal B/D pairs ( Supplementary   Fig. 11) that show similar movements for ATD domains between 19.5 Å to 22.3 Å.
The LBD domains of distal subunits A/C move by ~ 9.7 Å while that for proximal B/D move by 21.6 Å for B and only 12.1 Å for subunit C, again highlighting the asymmetric arrangement of LBD in UBP310 bound state.
Owing to rearrangements at the LBD layer, there is also a substantial reorganization of LBD-TM linkers to go from desensitized to resting state. Consistent with 4-fold symmetric arrangement of LBDs in desensitized state, LBD-helix E separation in both distal A/C and proximal B/D subunit pairs is equal at ~21 Å ( Supplementary Fig. 12 a , b). Also helix-E is placed ~4 Å lower in proximal B/D subunits when compared to distal A/C protomers (Supplementary Fig. 12c). In contrast, helix-E in UBP310 bound receptor has a separation of ~22 Å for A/C pair similar to that in SYM complex but ~45Å for the B/D subunits. Further, due to formation of BC dimer at the LBD layer, the helix-E in subunit B moves lower in vertical plane increasing the distance between helix-E from subunits B and C to ~12 Å. Interestingly, the LBD helix-E from subunits A and D align in same plane in contrast to SYM complex where they are separated by ~4 Å (Supplementary Fig. 12). These conformational changes in helix E-M3 are necessitated because of the asymmetric architecture of UBP310 bound receptor.

N-glycans at the ATD-LBD interface
iGluRs are post-translationally modified with multiple N-linked glycans that are studded on both the extracellular ATD and LBD domains 17,37 . Many of these glycans reside at the ATD-LBD interface and on the linkers connecting ATD and LBD (Supplementary Fig. 1) and have previously been shown to modulate gating properties 37,38 and trafficking of iGluRs 24,39 or interactions with other synaptic proteins 40 . In GluK3 desensitized EM map, we observed residual density at the ATD-LBD dimer interface, which was not satisfied by fitting of the ATD and LBD coordinates. Interestingly, the residual densities are in close proximity to potential Nlinked glycosylation sites ( Fig. 5a-b). In particular, the density near Asn 247 in ATD and Asn 402 in LBD appear more prominent in distal subunits as compared to proximal subunits (Fig. 5b). Similar extra density was also observed in the GluK2  is slightly higher than that of wild type GluK3 ( Fig. 5d and g) while that on LBD (Asn 402) slightly increased it. However, all the three-glycan knockouts either individually or in combination slowed down desensitization rates.

GluK3 N-glycan mutants recover faster from desensitized state
We next evaluated the recovery rates from desensitization by generating recovery currents via two applications of 30 mM glutamate at varying time intervals (two-pulse protocol) ranging from 50 ms to 5 s. We observe that all the GluK3 N-glycan mutants  Table 2). In particular the glycans at position Asn 247 and Asn 402 alone or in combination seem to play key role in receptor desensitization and recovery. We hypothesized that glycans at these positions might affect receptor properties by mediating glycan-glycan and/or glycan-protein interactions. Hence, we next focused our attention to the putative N-glycan interacting residue on the receptor surface.

Exploration of putative N-glycan interacting residues
It has been shown that the composition, content and length of N-glycans can be highly variable for iGluRs depending on the context of their expression. This in combination with flexibility of N-glycans makes it impossible to predict all the residues that might interact with them. Thus, we focused only on the residues that likely lay in close apposition to the N-acetyl glucosamine (NAG) residues of the oligo-mannose core at the ATD-LBD interface shown in Fig 5a -b. We mutated the potential interacting residues to their counterparts in GluA2 receptors. The core NAGs of Asn 402 Nglycan would likely lie in close proximity to ATD lower lobe (Fig. 5b). Interestingly, as in case of Asn 402 N-glycan knockout, single point mutants for positions Glu 239 and Tyr 243 to corresponding residues in GluA2 Leu (E239L) and Phe (Y243F) at the lower lobe of ATD slows down the receptor entry into desensitized state with a τ des of 2.0 ±0.2 ms (E239L) and τ des of 2.3 ±0.5 ms (Y243F) respectively (Fig. 6a). The mutant receptors also show faster recovery from desensitization than their wild-type counterparts with τ rec of 0.3 s and 0.6 s for E239L and Y243F respectively similar to phenotype seen in case of N-glycan knockouts at Asn 402 (Fig. 6b). Interestingly, introducing a negatively charged residue by mutating arginine at position 242 to glutamate (R242E) in Y243F background (double mutant R242E/Y243F) showed a significant decrease in the rate of desensitization (τ des of 3.3 ±0.5 ms) as in N-glycan knockouts but led to slower recovery from desensitized state (τ rec of 4.0 s) in contrast to faster recovery rates observed for knockouts and other putative interacting residue mutants ( Fig. 6c and d). The double mutant R242E/Y243F recovers ~6.3 fold slower than Y243F mutant receptors and ~4.2 fold slower than wild type GluK3. However, single mutation of R242E (corresponding residue in NMDA receptors) alters the measured receptor properties (τ des and τ rec ) τ des of 2.1 ±0.1 ms and τ rec of 0.8 s and is similar to wild type GluK3 (Fig. 6c-d). We don't fully understand the reason for this observation but it's likely due to modulation of protein-N-glycan interactions. Owing to 2-fold symmetry at ATD dimer and dimer of dimer interface, these mutations in a homotetrameric receptor in proximal subunits would lie close to the ATD dimer of dimer interface. However, these mutations are away from the central axis of receptor tetramer and are not likely to perturb tetrameric assembly. Hence, the affects seen are likely due to modulation of N-glycan-protein interactions on the distal subunits.
On evaluation of total surface expression of all the N-glycan and potential interacting residue mutants by surface biotinylation assay, it was found that all the mutants were able to reach the surface albeit in variable amounts. Surprisingly, N721A glycan knockout failed to reach cell surface ( Fig. 6e; Supplementary Fig. 13). This Nglycan site lies at the LBD-TMD interface and was shown to be glycosylated in LBD expressed in insect cells 42 . Interestingly, it is conserved in all the kainate receptor subunits (GluK1-GluK5) but absent in NMDA and AMPA receptors. This suggests that glycosylation at this site might be important for assembly and trafficking of kainate receptors. However, it needs to be explored further to fully elucidate its role.
Next, we investigated the potential interacting residues for Asn 247 N-glycan. The core NAGs of Asn 247 N-glycan likely lie in close proximity to S1 lobe of LBD. We checked double mutants Y744L/R745G and quadruple mutants Y744L/R745G/D746T/K747P for activity. However, in spite of reaching the cell surface ( Fig. 6e; Supplementary Fig. 13) we could not record measurable currents from these receptors. Despite this, our glycan knockout assays highlight the importance of Asn 247 glycan in modulation of receptor functions.
To summarize, our structural and functional data suggest potential inter-domain

Discussion
Kainate receptors are comparatively less studied than AMPARs and NMDA receptors but the current advancement in KAR knowledge indicates that they are involved in multifunctional neuronal activity and have a profound role in health and diseases 2, 43-45 . Using a multi-pronged approach, combining cryo-EM, X-Ray crystallography, Also, the desensitized state in kainate receptors is ~100 fold more stable than their AMPA counterparts 46 Fig. 1). Similarly, for the receptors containing the "high-affinity" GluK4 and GluK5 subunits the residues corresponding to E239 and Y243 are aspartate and glutamate/aspartate respectively for the two sites in GluK4 and GluK5 subunits and may still interact with the N-glycan corresponding to Asn 402 on heteromeric assembly with GluK1-GluK3 subunits. Further, the interactions mediated by Nglycans would also affect the functional modulation by auxiliary proteins like Neto1 and Neto2 and has been shown earlier for GluK2 20 but need more exploration in case of GluK3 receptors.

Construct design
We initially tried expression of the full-length rat GluK3 subunit containing point mutants R591Q (Q/R site) but this construct had very low expression and poor profile on size exclusion chromatography (SEC) and hence was not suitable for structural analysis. In order to achieve better surface expression, native signal peptide of GluK3 was replaced with that of GluK2 and was sub-cloned into the pEGBacMam vector 48 for baculovirus based expression in mammalian cells using standard molecular biology techniques. To screen constructs via fluorescence detection 25 and for affinity purification, a thrombin recognition site along with linker sequence (GLVPRGSAAAA) was inserted between GluK3 and the coding sequence for the A207K non-dimerizing EGFP mutant, with a C-terminal octa-histidine (His8) affinity tag. To improve solubility and stability, the GluK3 construct was truncated at C-ter ∆826 and cysteine residues at positions 86, 305 were mutated to threonine while cysteine 547 was replaced with valine. This construct had good expression and tetrameric receptor profile as screened via FSEC and hence was chosen for large-scale expression and purification. We refer to this construct as GluK3 EM .

Whole-cell voltage-clamp Recordings
In order to test the functionality of GluK3 wild type, GluK3 EM construct, and various In order to trap the purified receptors in different states, GluK3 receptor was incubated with ligands directly before imaging. 2S, 4R-4-methylglutamate (SYM) at a final concentration of 2mM was added to protein to trap desensitized state whereas, 100 µM UBP310 was used to stabilize the receptor in resting state. Screening of the concentration and effect of these agonist and antagonist onto stability of the receptor was checked by FSEC analysis (data not shown).

Specimen vitrification and Cryo-electron microscopy
For both GluK3-SYM and UBP310 complex, Quantifoil R1.2/1.3 Au 300 mesh grids were glow discharged for 90 secs at 15mA. 2.5µl of protein at 1.7 mg/ml was applied to the grid thrice followed by blotting for 5 s at 100% humidity, 4°C and vitrified by

Electron Microscopy Image Processing and model building
All the images (719 images for GluK3-SYM and 1693 for GluK3-UBP310) were subjected to beam induced drift correction using UCSF MotionCor2 49 followed by CTF estimation via Gctf 50 . Manual curation was carried out to remove micrographs with large ice contamination or poor CTF fits. RELION 2.1 51 was used to pick ~1000 particles manually and subjected to reference-free 2D classification. Selected 2D classes from this step were used as reference for automated particle picking from entire datasets. Autopicked particle stacks were then exported to cryoSPARC 52 for GluK3-SYM and cryoSPARC v2 for GluK3-UBP310 to carryout 2D classification.
Data were cleaned up via iterative rounds of 2D classification and subsequently removal of classes with unclear features, ice contamination or carbon. Initially, 77276 particles selected for GluK3-SYM and 138369 particles for GluK3-UBP310.
After successive rounds, 14648 and 54001 particles were selected for the GluK3-SYM and GluK3-UBP310 respectively. These were used to generate reference free abinitio 3D reconstruction. Additional round of 2D classification and removal of poorly defined particles selected 9730 particles that were utilized for 3D classification and refinement for SYM bound receptor in C1 symmetry. Homogenous refinement as implemented in cryoSPARC resulted into ~7.4Å resolution map according to gold standard Fourier Shell Correlation (FSC) for GluK3-SYM complex. For GluK3-UBP310, 3D reconstruction and homogeneous refinement by imposing C1 symmetry lead to ~8 Å resolution map. Sorted stack of 38853 particles was subjected to particle local motion correction followed with 2D classification and abinitio reconstruction into two 3D classes in cryoSPARC V2. The best 3D class (27194 particles) was further subjected to homogenous refinement resulting into a map with resolution of ~8.3 Å. This was followed with non-uniform and local refinement that improved the map resolution to ~ 8.1 Å and ~7.7 Å (0.143 FSC) respectively. Final maps for both the complexes were sharpened and ResMap 53 and Local Resolution module as implemented in cryoSPARC workflow was used for estimation of local resolution.

Crystallization and Structure Determination of GluK3 LBD with SYM
GluK3 LBD S1S2 domain from residues N402-K515 of S1 domain linked by GT  Fig. 14, and 15).
Structure of GluK3-SYM LBD is very similar to GluK2-SYM structure reported earlier with r.m.s.d of 0.78 for superimposition of Cα atoms (Supplementary Fig.   16).

Model building of the full-length receptor
Model for the GluK3-SYM tetramer was built by rigid body fitting in UCSF Chimera

Site Directed Mutagenesis
All the mutations were made in wild type GluK3 receptors following standard protocol for site directed mutagenesis and confirmed by sequencing of the entire and Y243F were made to test potential N402 glycan interacting residues.

Surface Biotinylation Assay
To Anti-GluR6/7 monoclonal Antibody (Sigma) to identify surface expressed wild type and mutant GluK3 receptors. Blots were analyzed via ImageJ 60 to quantitate total and surface expressed protein ( Fig. 6d; Supplementary Fig. 16).

Statistics
No statistical methods were used to predetermine sample size. The experiments were not randomised, and the investigators were not blinded to allocation during experiments and outcome assessment.

Data availability
The cryo-EM density reconstruction and final model were deposited with the Electron Microscopy Data Base (accession code EMD-XXXX) and with the Protein Data Bank (accession code XXX). All other relevant data supporting the key findings of this study are available within the article and its Supplementary Information files or from the corresponding author upon request.

Author Contributions
Jyoti Kumari optimized GluK3 construct, expressed and purified protein, carried out molecular biology, biochemical experiments and processed EM data with assistance from J.K., R.V. carried out the electrophysiology experiments. Janesh Kumar supervised the overall project design and its execution. All authors contributed to analysis and preparation of manuscript and approve the final draft.   Amplitude of the second glutamate application in a two-pulse experiment is reported as a normalized percentage of the first glutamate application and is plotted against interpulse intervals. Recovery rates (τ rec ) were calculated with single exponential association fits.