Structural and Molecular Determinants of Glutamate Transporter Allosteric Modulators

Abstract

Excitatory Amino Acid Transporters (EAATs) are critical regulators of synaptic glutamate levels in the central nervous system (CNS). Dysregulated CNS glutamatergic homeostasis is implicated in many neurological diseases, highlighting the key role of EAATs in neurological health. We previously identified a library of small compounds that function as either positive or negative allosteric modulators (PAMs or NAMs) of EAATs, with diverse selectivity for subtypes EAAT1, EAAT2, and EAAT3, including astrocytic EAAT1 and EAAT2, and neuronal EAAT3. In this work, we characterize compounds from our library using molecular modeling, mutagenesis and pharmacological approaches. We focused on three representative compounds: NA-014, an EAAT2-selective PAM, DA-038, an EAAT1-3 PAM; and NA-010, an EAAT2-selective NAM. Binding studies demonstrated that these compounds do not interact with the orthosteric glutamate-binding site, confirming an allosteric action. Docking studies suggested several potential binding poses of NA-014 between the scaffold and transport domains of EAAT2, which we then studied with mutagenesis approaches. We identified potential binding sites of representative compounds in transmembrane (TM) domains 1, 5, 8 and hairpin 2 (HP2) and demonstrated that these are necessary for their activity. Ten key amino acids residues within a subdomain of EAAT2 substituted into EAAT1 conferred EAAT2-selective PAM activity, demonstrating these residues are required and sufficient to enable selective PAM function. Collectively, these studies identified crucial subdomains and key amino acids linked to PAM activity, advancing our understanding of how to modulate EAAT activity. This knowledge can be integrated in future studies to develop EAAT allosteric modulators for neurological disorders.

Significance statement

We identified modulators of glutamate transporters, key regulators of CNS excitability and neuronal health. Using molecular modeling, mutagenesis, and pharmacology, we mapped their allosteric binding sites and identified ten residues that confer selective transport enhancement. This mechanism of transporter activation may guide development of therapies for disorders involving glutamatergic dysregulation, including stroke, neuropathic pain, and substance use disorders.

1. Introduction

Glutamate, the major excitatory amino acid neurotransmitter in the mammalian CNS, is imperative for many brain functions, including memory and learning, motor functions, and neural plasticity and development.¹ Glutamatergic signaling is terminated by reuptake of glutamate by sodium-dependent excitatory amino acid transporters (EAATs).^2,3 There are five subtypes of EAATs, with EAAT2/GLT-1 (human/rodent homologue) being the most abundant.⁴ It is mainly expressed in astrocytes throughout the brain and spinal cord, accounting for ~95% of CNS glutamate uptake,⁵ thus being crucial at maintaining extracellular balance and preventing excitotoxicity.^6,7 EAAT1/GLAST and EAAT3/EAAC1 are expressed in the cortex, with EAAT1 predominantly on astrocytes and EAAT3 predominantly on neurons.⁸

EAATs are transmembrane proteins formed by three protomers that independently transport glutamate.⁹ Each protomer has a scaffold (or trimerization) domain and a transport domain which moves almost as a rigid-body along the scaffold domain in a twisting “elevator-like” motion to transport glutamate and sodium ions into the cell.^10-14 Uptake of substrate and ions takes place in the outward-facing (OF)¹⁵ state of the protomers, and their release to the cytoplasm, in the inward-facing (IF) state. Substrate entry and exit in the respective OF and IF states are further regulated by pairs of residues serving as extracellular (EC) and intracellular (IC) gates, which may assume open and closed conformations in either state, thus enabling a tight control of neurotransmitter transport.^16,17 Dysregulation of this mechanism and ensuing disruption of glutamate homeostasis are associated with the development of various neuropsychiatric disorders, including ischemic stroke,¹⁸ substance use disorder,¹⁹ neuropathic pain,²⁰ and epilepsy.²¹

Modulation of the activity or expression of EAAT2 represents a novel therapeutic approach to treat these disorders/diseases. Our prior work identified positive allosteric modulators (PAMs), i.e., modulators that increase EAAT2 activity.^22-26 We identified an allosteric site at the interface between the scaffold and transport domains of the protomers, which led to the hypothesis that PAMs increase glutamate translocation by disrupting the interactions between these domains.^27-29 An in-silico screening for compounds that bind this allosteric region resulted in the discovery of GT949, an EAAT2-selective PAM.^22,23

We then generated GT949 analogs to improve drug-like properties (i.e., size, solubility, and polarity), which resulted in a library of ~50 new compounds.³⁰ Their characterization revealed different activities and selectivity towards EAAT subtypes,³⁰ despite their structural similarity, prompting further investigation into their pharmacological mechanisms.

In the present study, we focused on three compounds with different actions: NA-014 (EAAT2-selective PAM), DA-038 (EAAT1, 2 and 3 PAM, or pan-EAAT PAM), and NA-010 (EAAT2-selective negative allosteric modulator, NAM, Table 1). Binding experiments revealed that the compounds do not compete with the EAAT inhibitor ETB-TBOA, confirming that they act allosterically. Computational modeling and simulations revealed several potential binding poses and coordinating residues between the scaffold and transport domains, which provided the basis for experimental tests and validation. Site-directed mutagenesis and pharmacological experiments further helped identify key amino acids that define the allosteric sites, the mechanisms of pharmacological activity and EAAT selectivity.

Table 1: Structure and function of NA-014, DA-038 and NA-010. Chemical structures of three allosteric modulators and their mechanism and selectivity on EAATs.

Compound	NA-014 (C22H25FN4O)	DA-038 (C21H23N7)	NA-010 (C19H24N4O)
Structure
Mechanism	EAAT2-selective PAM	EAAT1, 2 and 3 PAM	EAAT2-selective NAM

2. Material and Methods

2.1. Compounds and reagents

Compounds NA-014, NA-010, and DA-038 were synthesized according to chemistry schemes reported,³⁰ dissolved in dimethyl sulfoxide (DMSO) to 100 mM stock solutions and were stored at −20° C. L-glutamate and (3S)-3-[[3-[[4-(Trifluoromethyl) benzoyl]amino]phenyl]methoxy]-L-aspartic acid (TFB-TBOA) were purchased from Tocris (Bristol, United Kingdom). L-[3,4-³H]-Glutamic acid, Specific Activity: 51.1 Ci/mmol) was purchased from Revvity (Waltham, MA, USA). [³H]-ETB-TBOA [(2S,3S)-2-amino-3-[[3-[[4-(1,2-ditritioethyl)benzoyl]amino]phenyl]methoxy]butanedioic acid] (Specific activity: 31.3 Ci/mmol) were synthesized by Moravek Biochemicals, Inc. (Brea, California, USA).

Cell culture media included Dulbecco’s modified Eagle’s medium (DMEM) with glucose and glutamine, fetal bovine serum (FBS), penicillin/streptomycin, and phosphate-buffered saline (PBS) buffer were obtained from Corning Cellgro (Manassas, VA, USA). Mirus transfection reagent TransIT-LT1 was purchased from Mirus Bio LLC (Madison, WI, USA). Reagents for uptake assays including CaCl₂, MgCl₂, NaOH, and Sodium dodecyl sulfate (SDS) were obtained from Sigma-Aldrich (St. Louis, MO, USA). Optimem reduced serum media, scintillation fluid, EZ-Link^™ Sulfo-NHS-Biotin, NeutrAvidin Agarose Resins, protein measurement kit (Pierce BCA Protein Assay Kit), Halt protease inhibitor cocktail (100x), Glycine, Methanol, NuPAGE LDS Sample Buffer (4x), NuPAGE MOPS Running Buffer (20x), Restore Fluorescent Western Blot Stripping Buffer, and NuPAGE Novex 4-12% Bis-Tris Gels and were all purchased from Thermo Fisher Scientific (Waltham, MA, USA). Turbo-Blot Turbo Transfer System and TransBlot Turbo transfer buffer were obtained from Bio-Rad (Hercules, CA, USA). Odyssey Blocking Buffer was purchased from LI-COR Biosciences (Lincoln, NE, USA). Pellet Paint Co-precipitant was obtained from EMD Millipore (Billerica, MA, USA). Phusion High-Fidelity DNA polymerase, DpnI enzyme, Buffer 4, NEB 10-Beta Competent E. coli, and Phusion Buffer HF were all purchased from New England Biolabs (Boston, MA, USA). Deoxynucleotide (dNTP) solution mix was obtained from Promega (Madison, WI, USA). DMSO was purchased from Macron (Center Valley, PA, USA). Mini and Midiprep plasmid purification kits were purchased from QIAGEN (Valencia, CA, USA). Ethanol 200 Proof was obtained from Decon Labs (King of Prussia, PA, USA). For western blots, primary antibodies used were Anti-EAAT2 [(Rabbit, Kind gift from Susan Amara (NIMH), NIH #274, 1:10,000)], Anti-EAAT1 (Rabbit, Alomone Labs, Jerusalem, Israel, # AGC-021, 1:500, RRID: AB_2039885) and β-Actin (Mouse, Cell Signaling Technology, Danvers, MA, USA, # 8H10D10, 1:1000, RRID:AB_2242334); and secondary antibodies used were IgG (H+L) DyLight 800 4x PEG conjugate (Anti-Rabbit, Cell Signaling Technology, #5151S, 1:1000, RRID:AB_10697505) and IgG (H+L) DyLight 680 Conjugate (Anti-Mouse, Cell Signaling Technology, #5470S, 1:1000, RRID:AB_10696895).

2.2. Binding assays with radiolabeled ETB-TBOA

Cells were transfected as above with WT EAAT2 plasmid. Two days after transfections, membrane fractions were prepared according to previous methods,³¹ by rinsing cells with PBS, scraping them in 10 mL of PBS and pelleting by centrifugation at 1,000 g for 10 minutes at 4°C. Cell pellets were then resuspended in chilled 1 mL of 50 mM Tris HCl buffer (pH 7.4), triturated 5 times to gently hypotonically lyse cells, and then transferred to a fresh Eppendorf tube. Cell suspension was then centrifuged at 21,000 g for 20 mins at 4°C. Fresh membrane pellets were then resuspended in cold lysis buffer with 3 times the volume of the cell pellet and immediately analyzed by Bicinchoninic Acid (BCA) kit to determine protein concentration. The remaining membrane fraction was stored at −80°C until use. ³¹ On the assay day, 96-well plates (UniFilters GF/B, Whatman, Maidstone, UK) were soaked for 1 hour in 3% polyethylenimine (PEI) prior to use. Membranes were gently thawed on ice and diluted in binding buffer (50 mM Tris-HCl pH 7.4, 120 mM NaCl, 5 mM KCl, 5 mM MgCl₂, 25 mM HEPES). Saturation assays were performed with varying concentrations of [³H]-ETB-TBOA and 10 μM unlabeled TBOA for obtaining the non-specific binding. Displacement studies were performed by the addition of 10 nM −10 mM non-radiolabeled ligands (NA-014, NA-010, DA-038 or TFB-TBOA) and 20 nM [³H]-ETB-TBOA to membrane fractions. Assays were done in 96 well plates for 1 hour at 4 °C, then binding was terminated by six rapid washes with binding buffer using a FilterMate Harvester vacuum filtration system (PerkinElmer, Waltham, MA, USA) onto the 96-well PEI-soaked filter to trap radioligand that was bound to the membrane fraction (i.e., to EAAT2). Plates were dried overnight, then loaded with 50 μL of scintillation fluid and read on a 1450 MicroBeta Liquid Scintillation Counter (PerkinElmer, Waltham, MA, USA).

2.3. Docking simulations

Docking simulations were performed using as target EAAT2, in both OF and IF states using AutoDock Vina.³² Simulations were repeated for NA-014, NA-010, and DA-038. Prior to docking, the substrate and known inhibitor were removed to assess the binding sites and binding poses of the compounds to the substrate/ligand-free state of the transporter. Cubic grids of size 32 x 32 x 32 ų were used with the exhaustiveness parameter set to 50. A short-list of 20 bound conformations were generated as energetically most favorable poses for each compound and each state of the transporter (total of six sets of runs), which were subjected to further ranking and clustering to identify four most frequently recurring binding sites, Site 1A, 1B (or 4), 2 and 3. Binding affinities were calculated and ranked using AutoDock Vina.³² The specific binding affinities of each compound in the best binding pose at each of the four sites were calculated using PRODIGY-LIG.³³

We also investigated how the presence of sodium ions and the HP2 gate opening influence the docking results. We first generated a structural model for the OF state of human EAAT2 (hEAAT2) using the OF structure resolved for hEAAT1 (PDB: 5LLU)¹² as template in SWISS-MODEL. We then generated another structural model for hEAAT2 with HP2 in the open conformation, in the presence of two sodium ions, using as template the OF structure resolved for hEAAT3 (PDB: 8CV2).³⁴ To account for HP2 loop flexibility during docking, simulations were performed using the GNINA docking program.³⁵ While preserving the AutoDock Vina search settings, GNINA further considers side chain flexibility, and incorporates deep learning-based techniques to enhance the docking pose predictions and scoring. Blind docking simulations followed the same docking protocol, with the exception that side-chain flexibility was allowed for HP2 residues A435-V448.

2.4. In silico saturation mutagenesis analysis

The effect (pathogenic or neutral) of point substitution at each position (all 19 substitutions) at all sequence positions was computed using our AI-powered tool Rhapsody-2³⁶ for in silico saturation mutagenesis. We used the substrate-free IF EAAT2 structure (PDB: 7VR8)³⁷ as input. For each residue, Rhapsody-2 releases a pathogenicity score ranging from 0 (neutral) to 1 (deleterious), with intermediate scores scaling with the probability of pathogenicity. The output is an in silico saturation mutagenesis heatmap color-coded from blue (neutral) to red(pathogenic) by the probability of pathogenicity or tolerance to mutations (ordinate) at all positions (abscissa) for the investigated protein.

2.5. Site-directed mutagenesis experiments

Twenty-six point mutants were generated in EAATs, using the QuikChange site-directed mutagenesis kit (Agilent Technologies, Wilmington, DE) as previously described.²³ First, we generated mutations from wild-type (WT) pCMV5-EAAT2³⁸ to either alanine or to sequentially alignment residues in EAAT1 or EAAT3, resulting in EAAT2 mutants: V75K, M76A, A79V, A79S, F80A, K90Q, A302I, I303A, F348A, F352L, F352A, W355L, T361A, S441G, S444A, I464V, V468I, and R476K. Mutations were verified by DNA sequencing (Azenta LifeSciences, Burlington, MA, USA). The following previously published mutations were also used: H71S, M86V, L290A, L295A, G298A, P443A, S465L, and W472I.^23,27 Forward and reverse primers are listed in Supplemental Table 3. Additionally, we generated an EAAT1 construct with 10 mutations: EAAT1-S74H, K79M, S82A, Q93K, V291L, E303A, M304I, I349F, Q445S, and I469V, this construct was named “EAAT1-EAAT2 hybrid”. All mutants were evaluated for functionality (kinetic parameters of V_max and K_m) and expression (total and surface, via cell surface biotinylation). Mutants that were not functional or not expressed on the surface were excluded from the study.

2.6. Cell culture and transfection

Glutamate uptake assays were performed as previously reported.^23,39 Briefly, COS-7 cells (ATCC, Manassas, AV, RRID: CVCL_0224) were maintained in Dulbecco's Modified Eagle Medium (DMEM) containing 10% FBS (fetal bovine serum), 100 units/mL of penicillin, and 100 μg/mL streptomycin in a humidified incubator with 5% CO₂ at 37 °C. For glutamate transporter assays, COS-7 cells at ~70% confluency were transiently transfected with 0.5 μg of plasmid DNA (WT EAAT or mutant) per well using TransIT-LT1 transfection reagent and plated at a density of 50,000 cells/well and uptake experiments were performed two days after plating. Transfection with empty vector pCMV5 was used to control the level of endogenous uptake of radiolabeled glutamate in each experimental condition. For binding assays, COS-7 cells at ~90% confluency were transfected as described above with pCMV5 or WT EAAT2 and plated in 15-cm dishes.

2.7. Cell surface biotinylation

Cell surface biotinylation was performed as previously described.²⁸ Briefly, COS-7 cells were transfected and plated into 6-well plates at 250,000 cells/well. After two days, cells were incubated with 1 mg/mL EZ-Link Sulfo-NHS-Biotin for 30 minutes, then any remaining biotin was quenched using 100 mM glycine. Cells were lysed in TNE Buffer (10 mM Tris, 150 mM NaCl, and 1 mM EDTA, pH 7.5 and 1% Triton X-100 and Halt Protease Inhibitors) at 4°C for 90 minutes, then centrifuged at 4°C for 10 minutes at 12,000 g. Supernatants were collected and divided into separate fractions for lysate input, protein measurement and incubation with NeutrAvidin Agarose Resins. After centrifugation at 2,460 g at 4°C, supernatants were removed and beads were washed three times with 1 mL TNE lysis buffer, followed by a final wash with 1 mL PBS-CM. Beads were then diluted with water, LDS-NuPAGE (4x) and DTT (10x) and frozen until further analysis.

Equal amounts of protein sample were loaded into a 4-12% Bis-Tris gel and separated for 60 min at 200 V, followed by a dry transfer into a Bio-Rad apparatus per the manufacturer’s protocol. After transferring, PVDF membranes were blocked, rinsed and incubated with primary antibodies EAAT2 (1:10,000) or EAAT1 (1:500), and β-actin (loading control, 1:1000) prepared in PBS-T overnight. After rinse with PBS-T, secondaries antibodies were applied for 60 minutes at room temperature, washed and imaged on LI-COR Image Studio Lite 4.0.21 (LI-COR Biosciences; Lincoln, NE, USA).

2.8. Glutamate transporter-based assays

2.8.1. Kinetic assays:

To compare the functionality (V_max and K_m) of EAAT1 and EAAT2 mutants to their WT counterparts, kinetics assays were completed as described before.^28,40 Briefly, two days after transfection, cells were washed with room temperature PBS-CM buffer (2.7 mM KCl; 1.2 mM KH₂PO4, 138 mM NaCl; 8.1 mM Na₂HPO₄, added 0.1 mM CaCl₂ and 1 mM MgCl₂, pH 7.4). Uptake reactions were initiated by the addition of unlabeled L-glutamate and [³H]-L-glutamate (1–1000 μM, final concentration, 99% unlabeled and 1% labeled). After 10 min, uptake was terminated by two washes with PBS-CM and addition of lysis buffer (1% sodium dodecyl sulfate /0.1 M NaOH). Lysates were added to scintillation vials with 3 mL scintillation fluid (ScintiVerse, Fisher Scientific, Pittsburgh, PA), and radioactivity was measured using an LS 6500 counter (Beckman Coulter, Brea, CA).

2.8.2. Dose-response assays:

Dose response assays were performed in presence of several concentrations (0.01–100 μM) of either NA-014, NA-010 or DA-038 or vehicle, in cells transfected with WT EAAT, mutants or pCMV5 (empty vector), two days after transfections. Compounds were pre-incubated for 10 min at 37 °C, uptake reactions were initiated by the addition of [³H]-L-glutamate at a final concentration of 50 nM and terminated after 10 min as above.

2.9. Statistical Analysis

All experimental data were analyzed using GraphPad Prism version 10 for Windows (GraphPad Software, La Jolla, CA, USA).

For binding studies, specific signals were obtained by subtracting background signal obtained from treatment with non-radiolabeled 10 μM TFB-TBOA. K_d and B_max values of [³H]-EFB-TBOA binding to EAAT2 were calculated using non-linear regression and One site sigmoidal equation from two separate experiments performed in triplicate. Dose-response curves were fitted by nonlinear regression to obtain IC₅₀ values and are presented as geometric mean [95% confidence interval (CI)] from three experiments. Statistical significance of the effect of compounds was assessed using one-way analysis of variance (ANOVA) followed by Dunnett’s multiple-comparisons post hoc test with vehicle as control (*p < 0.05).

For kinetic analysis, the Michaelis–Menten equation was used for calculating V_max and K_m. Background signal was obtained from pCMV5-transfected cells and subtracted before analysis. V_max of individual mutants were then normalized to WT EAAT2 to compare across experiments. For comparisons between V_max and K_m of WT EAAT and mutants, data were analyzed using a mixed-effects model (REML), with mutant treated as a fixed effect and experimental day as a random effect to account for pairing within experiments. Post hoc comparisons versus WT were performed using Dunnett’s multiple-comparisons test to control the family-wise error rate. K_m values were log-transformed prior to statistical analysis, and comparisons were performed on log(K_m) using the same method (*p < 0.05).

For western blots analysis, band intensities of both surface (biotinylated) and total EAAT2 monomers (~62 kDa) were measured using LI-COR Image Studio Light software. Surface density values were normalized to total density values. Then, these values were normalized again to be expressed as percentage of EAAT2 WT densities. Comparisons between WT EAAT and mutants transport expression (% of WT) were conducted using a mixed-effects model (REML) and post hoc comparisons as described above (*p < 0.05). We also obtained a ratio of difference between mutants and WT EAAT2 by dividing normalized V_max by normalized biotinylation values (shown as activity/expression ratio over WT).

Dose–response curves evaluating the effects of the compounds on glutamate uptake were obtained from independent biological experiments (n indicated in the corresponding tables), each performed in technical triplicate. Curves were analyzed by nonlinear regression using a three-parameter sigmoidal equation, from which EC₅₀ values were derived. For PAMs, the top (maximum effect) was unconstrained, and the bottom (baseline) constrained to 100%, whereas for NAMs, the top (baseline) was constrained to 100% and the bottom (maximum effect) left unconstrained. For each individual experiment, one-way ANOVA followed by Dunnett’s multiple-comparisons test (vehicle as control) was used to assess whether compound treatment produced a statistically significant effect across tested concentrations within each experiment. Dose-response parameters (EC₅₀ and efficacy) were derived from nonlinear regression and used for quantitative comparisons between conditions.

Efficacy values (derived from the fitted top and bottom parameters of the nonlinear regression) are presented as mean [95% CI]. Comparisons between EAAT mutants and WT EAAT2 were performed using a mixed-effects model (REML) with post hoc testing as described above. EC₅₀ values were log-transformed to pEC₅₀ prior to statistical analysis, and all comparisons were performed on log-transformed values; results are presented as geometric mean [95% CI].

For comparisons between two independent groups, an unpaired two-tailed Welch’s t-test was used, which does not assume equal variances (p < 0.05).

Based on dose–response glutamate uptake analyses, the effects of compounds on mutants were categorized as follows: (i) mutation abolishes the compound effect (colored in purple); (ii) mutation does not abolish effect of compound, but increases EC₅₀ (i.e., it decreases compound potency, colored in green); or (iii) mutation does not abolish compound effect (colored in orange).

3. Results

3.1. NA-014, NA-010 and DA-038 do not compete with the orthosteric ligand ETB-TBOA

In our glutamate transporter cell-based assays,³⁰ we found that compounds NA-014, NA-010 and DA-038 do not affect K_m but change V_max, suggesting that they have an allosteric action. To further validate this conjecture, we performed displacement assays against a known orthosteric inhibitor of EAATs, [³H]-ETB-TBOA. We first performed saturation experiments and observed a dose-dependent binding effect of [³H]-ETB-TBOA in EAAT2-expressing COS-7 membranes. In two independent experiments, we observed B_max values of 492 and 562 pmol/mg and K_d values of 36 and 21 μM, yielding a geometric mean of 27.5 μM. Background binding of [³H]-ETB-TBOA of less than 10 % was obtained in the presence of non-radiolabeled TFB-TBOA [(3S)-3-((4-(Trifluoromethyl)benzoyl)amino)phenyl) methoxy)-L-aspartic acid] (Figure 1a), confirming appropriate assay conditions. We then tested NA-014, NA-010, DA-038, and non-radiolabeled TFB-TBOA in displacement assays. TFB-TBOA displaced the binding of [³H]-ETB-TBOA with an IC₅₀ = 1.4 μM (geometric mean [0.9, 2.2], n = 3) (Figure 1b). Conversely, our compounds did not compete with [³H]-ETB-TBOA binding to EAAT2, suggesting they do not bind the same orthosteric glutamate site, but target an allosteric site.

Figure 1: The three allosteric modulators do not bind to the orthosteric site that TBOA binds.

a, A representative saturation plot of binding of [³H]-ETB-TBOA to WT EAAT2-transfected COS-7 cell membranes, with incubation performed at RT for 1 hour. Open black squares represent the total binding; open blue circles represent EAAT2- mediated binding and open red triangles represent nonspecific binding in the presence of 10 μM TFB-TBOA. EAAT2-expressing membranes had a B_max of 492 pmol/mg and K_d of 21 μM for ³[H]-ETB-TBOA binding. Data are shown as mean ± SD. b, Representative displacement experiment with 20 nM [³H]-ETB-TBOA in WT EAAT2-expressing COS-7 cell membranes. Specific binding is normalized vehicle (control). TFB-TBOA displaced binding with an IC₅₀ of 1.4 μM (geometric mean [0.9, 2.2], n = 3). Curves were fit by non-linear regression to calculate IC₅₀ values. One-way ANOVA followed by Dunnett’s multiple-comparisons test (vehicle as control) was used was used to confirm the presence of a concentration-dependent effect (*p < 0.05, and **p < 0.01). Data are shown as mean ± SD.

3.2. Computational Modeling and Simulations Revealed the EAAT Sites Preferentially Selected by Allosteric Modulators

Our previous experiments determined compound NA-014 to be an EAAT2-selective PAM; DA-038 is EAAT1, 2 and 3 (pan EAAT) PAM; and NA-010 is an EAAT2-selective NAM.³⁰ To explore their potential binding sites and poses by molecular modeling and simulations, we first generated a structural model for the OF state of human EAAT2 (hEAAT2) using the OF structure resolved for hEAAT1 (PDB: 5LLU)¹² as template in SWISS-MODEL.⁴¹ For the hEAAT2 IF state, we used a glutamate- and a selective inhibitor-bound (PDB: 7XR6)⁴² and a substrate-free (PDB: 7VR8)³⁷ structure, after removing the substrate and inhibitor. Docking simulations were repeated using both OF and IF hEAAT2 as targets.

Three distinct sites were identified for NA-014 binding to OF EAAT2: Sites 1A, 1B and 3, shown in Figure 2a-b; and two additional sites were detected for IF EAAT2, Sites 2 and 4 (Supplemental Figure 1). Site 1A was previously identified and evaluated by our group using prior generations of EAAT2 PAMs.^23,27 As described above, the transport domain undergoes an “elevator-like” motion enabling alternating access to the EC and IC environments in the respective OF and IF states.^{12,34,37,42,43} Due to this sliding motion, Site 4 partially aligns with Site 1B in the OF conformation, hence the identification of four distinct sites: 1A, 1B/4, 2, and 3. We find that M76, A79, G82, D83, M86, K90, D238, G239, L295, K299, G360-A362, S363, I442, P443, and R476 are key residues lining Site 1A of the OF EAAT2 to help coordinate NA-014. The residues in boldface are not conserved across hEAAT subtypes. Closeup views (Figure 2b) show some of these residues that interact with the compound. P443 is unique as it interacts with different ligands/poses and is conserved across EAAT1-3 (Supplemental Figure 2). A440-S444 line both Sites 1A and 1B. Site 3 harbors several aromatic residues including F345, F348, F352, and W355 in close contact with NA-014 in both the IF and OF states. Site 2 is selected in the IF state only, and V468 in Site 2 is unique to hEAAT2, which may serve as a determinant of selectivity for binding the IF state exclusively.

Figure 2: Simulations of NA-014 (PAM) and NA-010 (NAM) docking onto OF EAAT revealed residues that may play a role in coordinating PAMs and NAMs.

a, NA-014 binding sites obtained by docking simulations onto the OF EAAT2 structural model. The sites are shown in stick and mesh representations colored cyan, dark blue, and orange respectively for Sites 1A, 1B and 3. Left and right panels represent two different views. Site 1A offers a large pocket accommodating alternative binding poses, two of which are shown. b, Closeup view of the interactions of NA-014 at its three binding sites. Residues making close contacts with the modulators and specific to EAAT subtypes are labeled and shown in stick representation. c-d, Similar results for NA-010. See also the results in Supplemental Figures 1, 3 and 4 for the results for all compounds for IF states of the transporter.

Similar computations and analyses performed for NA-010 and DA-038 led to the results presented in Figure 2c-d and Supplemental Figures 1, 3 and 4. The pan EAAT1-3 PAM, DA-038 showed binding patterns similar to those of NA-014: mainly Sites 1A, 1B and 3 were occupied in the OF state; and Sites 2-4 in IF. This may be expected as both compounds act as PAMs. NAM, NA-010, on the other hand, was distinguished by lack of binding to Site 2 in the OF state of EAAT2, or to Site 3 of IF EAAT2, pointing to its high selectivity.

Supplemental Table 1 lists all residues making < 4 Å atom-atom contacts with the allosteric modulators in different sites, along with their binding affinities. Supplemental Table 2 lists those shared or distinctive between the three EAAT subtypes at each site. These data provided the basis for mutagenesis analyses presented below.

We also investigated how the presence of sodium ions and the HP2 gate opening influences the docking results. When considering the OF hEAAT2 model with the open HP2 gate and bound sodium ions, we observed that Site 1A and Site 3 were preferentially selected by modulators. Supplemental Figure 5a-c illustrates examples of binding poses distinguished by their high binding scores for all three compounds. Notably, no binding was observed to the neighborhood of HP2, despite the substantial opening of the pocket created by HP2-loop opening. These results indicate that NA-014, DA-038, and NA-010 are predicted to bind EAATs irrespective of HP2 loop conformation or ion occupancy.

3.3. Role of Site 1A for mediating allostery

Site 1A is located at the interface between the scaffold and the transport domains, and this binding site aligns with molecular docking results previously generated for compound GT949 binding to EAAT2.³⁰ The EAAT2 specific PAM, NA-014, shows the highest binding affinity to Site 1A in both the IF and OF states of EAAT2 (Table 1); DA-038, and NA-010 bind to Site 1A in the OF state of EAAT2. Thus, Site 1A may be important to allosterically regulate the transport function of EAATs. In particular, the residues that coordinate both PAMs at Site 1A (Supplemental Table 1, first row) such as M76, A79, K90, L295, and R476, are of interest for experimental validation. Furthermore, specific residues were found to be involved in binding one of the PAMs, and not the other: M86, T361 and P443 bind NA-014 but not DA-038; while V75, F80, G298, and A302 bind DA-038 but not NA-014. The role of these 12 residues (or hot spots) will be investigated by site-directed mutagenesis coupled to functional assays.

Finally, we assessed whether the mutations might give rise to loss of structure or function, using our AI-powered tool for saturation mutagenesis in silico, Rhapsody-2.³⁶ Rhapsody-2 yields a pathogenicity score, s, varying from 0 (neutral) to 1 (strongly pathogenic/destabilizing) for any residue substitution at each position. The results for selected sites are presented in a color-coded heatmap, as shown in Supplemental Figure 6a; red regions refer to deleterious substitutions, and dark blue to neutral mutations. Substitutions predicted by Rhapsody-2 to retain the transporter structure and function (with s < 0.5) were selected for experimental tests (see below), including seven (H71S, F80A, M86V, L290A, I303A, F352A, and W355L) anticipated to have a marginal effect (0.2 < s < 0.5) (Supplemental Figure 6b).

3.4. Glutamate uptake assays with EAAT2 mutants confirmed the critical role of Site 1A in mediating NA-014 PAM activity

Combining data from docking simulations, sequence alignment, and in silico saturation mutagenesis, we shortlisted 26 single-point mutations. These contain the 12 residues noted above to participate in Site 1A, which are specific to EAAT subtypes and selectively bind to different compounds. Additionally, we considered three residues at Site 3 (on helical hairpin 1, HP1), two at Site 2 (transmembrane helix 8, TM8), and three at Site 1B (HP2 and TM8), prioritizing those not conserved between EAAT subtypes. Finally, we also considered the Site 1A mutations H71S and L290A, and Site 1B mutation S465L, generated earlier, ^23,27 along with I303A on Site 1A, adjacent to the hotspot A302, leading to a total of 15 residues at Site 1A, and 9 at Sites 1B, 2 and 3, listed in the Supplemental Table 3.

We evaluated the kinetics of substrate transport by these mutants. Table 2 shows a summary of the kinetic parameters, cell surface expression, and activity/expression ratios. Effect sizes are reported in Tables 2-5 as mean or geometric mean with 95% confidence intervals and are referenced in the text where relevant.

Table 2: Results from kinetics study and expression analysis of the EAAT2 mutants. Summary of the kinetic data for glutamate uptake (V_max and K_m), transport expression, and activity/expression ratios in COS-7 cells transfected with WT EAAT2 and mutants.

Transporter	Vmax (% of WT): Mean [95% CI] (n)a	Km (μM): Geometric mean [95% CI] (n) b	Transport expression (% of WT): Mean [95% CI] (n)c	Ratio activity/ expression over WT d
WT EAAT2	100 [84, 116] (27)	96 [82, 112] (27)	100 [91, 109] (20)	1.0
H71S	107 [61, 154] (5)	96 [51, 180] (5)	102 [95, 110] (3)	1.0
V75K	65 [35, 96] (6)	102 [63, 166] (6)	107 [−50, 265] (3)	0.6
M76A	121 [103, 138] (8)	139 [70, 277] (8)	156 [24, 287] (3)	0.8
A79S	22 [5, 40] (5) **	90 [14, 571] (5)	49 [−14, 112] (5)	0.6
A79V	80 [21, 140] (5)	80 [50, 127] (5)	41 [−49, 131] (3)	2.0
F80A	135 [14, 255] (4)	260 [120, 567] (4)	138 [−131, 408] (3)	1.0
M86V	111 [47, 174] (3)	48 [13, 172] (3)	68 [−15, 150] (3)	1.6
K90Q	107 [39, 177] (4)	47 [7, 313] (4)	59 [−6, 124] (4)	1.8
L290A	82 [38, 126] (4)	146 [65, 325] (4)	139 [−3, 280] (3)	1.4
L295A	119 [37, 201] (3)	91 [11, 766] (3)	85 [−4, 174] (3)	1.4
G298A	289 [250, 327] (5) ***	49 [25, 95] (5)	133 [82, 185] (4)	2.2
A302I	86 [54, 115] (5)	56 [10, 352] (5)	82 [−63, 399] (4)	1.0
I303A	36 [13, 53] (7) **	84 [40, 175] (7)	168 [−63, 399] (4)	0.2
F348A	100 [58, 143] (6)	85 [55, 134] (6)	86 [−120, 292] (3)	1.2
F352A	4 [−2, 10] (5) ****	1159 [340, 3945] (5) ***	−22 [−225, 182] (3) *	−0.2
F352L	51 [−16, 118] (3)	143 [78, 261] (3)	92 [43, 141] (3)	0.6
W355L	79 [4, 153] (5)	63 [26, 153] (5)	99 [−16, 215] (4)	0.8
T361A	71 [40, 102] (6)	55 [52, 109] (6)	92 [−7, 190] (4)	0.8
S441G	32 [19, 44] (4) *	17 [7, 38] (4) ****	94 [−8, 196] (4)	0.4
P443A	32 [22, 41] (3) **	84 [78, 90] (3)	73 [−58, 203] (3)	0.4
S444A	88 [51, 126] (6)	76 [54, 109] (6)	102 [35, 170] (3)	0.9
I464V	93 [63, 124] (4)	84 [49, 144] (4)	87 [15, 160] (3)	1.1
S465L	72 [16, 127] (3)	83 [65, 106] (3)	65 [14, 115] (3)	1.1
V468I	132 [71, 193] (5)	99 [88, 110] (5)	166 [−2, 335] (3)	0.8
W472I	43 [7, 78] (3)	99 [24, 398] (3)	49 [−10, 109] (3)	0.9
R476K	66 [31, 101] (5)	86 [42, 172] (5)	124 [1, 247] (4)	0.5

CI, confidence interval.

^aV_max values were normalized to WT (%) and are presented as mean [95% CI]. Values were derived from independent experiments performed in technical triplicate; the number of biological replicates (n) is indicated in parentheses. Comparisons were performed using a mixed-effects model (REML), with mutants treated as a fixed effect and experimental day as a random effect to account for pairing within experiments. Post hoc comparisons versus WT were conducted using Dunnett’s multiple-comparisons test (* p < 0.05 *** p < 0.001 **** p < 0.0001).

^bK_m is shown in μM as geometric mean [95% CI]. Biological replicates (n) are derived from the same data set shown in V_max. K_m analyses were performed on log(K_m) using the same method as described for V_max.

^cPercentage of transport expression was calculated as follows: band intensities of surface (biotinylated) EAAT mutants were normalized to total values, then normalized to WT EAAT2. Comparisons with WT were performed

^dRatios of activity/expression were obtained by dividing normalized V_max by normalized surface expression values.

Table 5: Identification of key residues underlying the stimulation by the PAM DA-038. Dose-response data for glutamate transport assays mediated by WT EAAT2 and mutants, in the presence of DA-038, listing efficacy and EC₅₀ values.

Membrane Domain	Transporter a	Efficacy (%): Mean [95% CI] b	EC50 (nM): Geometric Mean [95% CI] c
	WT EAAT2 (n=7)	139 [118, 161]	9 [2, 48]
Scaffold	H71S (n=3)	No effect
	V75K (n=4)
	A79V (n=3)
	F80A (n=3)
	M86V (n=3)
	K90Q (n=4)
	L290A (n=3)
	L295A (n=3)
Transport	W355L (n=4)	129 [89, 170]	3 [0.04, 188]
	P443A (n=4)	No effect
	I464V (n=3)
	V468I (n=3)
	W472I (n=3)
Scaffold	M76A (n=3)
	A302I (n=3)	115 [79, 152]	6 [0.03, 128]
	I303A (n=3)	No effect
Transport	F348A (n=3)
	S444A (n=3)	143 [80, 206]	16 [2, 104]
Scaffold	G298A (n=3)	No effect
Transport	F352L (n=3)
	T361A (n=3)	136 [54, 218]	4 [0.04, 391]
	S441G (n=3)	No effect
	S465L (n=3)
	R476K (n=3)

CI, confidence interval.

^aWT EAAT2 and mutants were transiently transfected in COS-7 cells, and a dose response of DA-038 was performed in glutamate transport assays. Efficacy, and EC₅₀ values were derived from independent experiments performed in technical triplicate; the number of biological replicates (n) is indicated in parentheses. The effect of DA-038 is indicated by the colors: purple indicates that residue abolished the effect of the compound, and orange indicates that residue did not abolish the effect of the compound.

^bEfficacy values were normalized to WT (%) and are presented as mean [95% CI].

^cEC₅₀ values are shown as geometric mean [95% CI].

^b-cComparisons of efficacies and EC₅₀ between WT EAAT2 and mutant were performed using a mixed-effects model (REML), with mutants treated as a fixed effect and experimental day as a random effect to account for pairing within experiments. EC₅₀ analyses were performed on log (EC₅₀).

V_max values differed across mutants (p < 0.0001). Post hoc comparisons using Dunnett’s test showed that A79S, I303A, F352A, S441G, and P443A exhibited marked reductions in V_max (to <40% of WT levels, adjusted p = 0.0104, 0.0022, < 0.0001, 0.0141, and 0.0018, respectively), whereas G298A had a higher V_max compared to WT (2.9 fold increase, adjusted p < 0.0001), and the remaining mutants were comparable to WT (adjusted p > 0.05).

Regarding K_m analysis, values differed across mutants (p < 0.0001). F352A exhibited an increased K_m relative to WT (adjusted p < 0.0001), whereas S441G showed a decreased K_m compared to WT (adjusted p < 0.0001). The remaining mutants did not differ appreciably from WT (adjusted p > 0.05).

Most mutations showed expression levels comparable to WT EAAT2 (p = 0.0584), except for F352A, which exhibited reduced expression (p = 0.0381). Of note, F80A, I303A and F352A were among the seven mutations predicted by Rhapsody-2 to potentially have a marginally damaging effect (Supplemental Figure 6a).

We also calculated the ratios of activity to expression over WT EAAT2, by dividing normalized V_max by normalized biotinylation values, which further indicated that the mutants I303A, S441G and P443A had decreased activity, but similar expression compared to WT EAAT2 (Table 2). We note that negative lower bounds in some confidence intervals for expression values reflect statistical variability and limited sample size rather than biologically meaningful values; these estimates should therefore be interpreted with caution.

Figure 3a shows representative kinetic curves for WT EAAT2 and mutant V468I, illustrating that V468I displays similar V_max and K_m to WT EAAT2. Figure 3b displays representative western blots for total and surface sample fractions of pCMV5, WT EAAT2, and F352A and V468I mutants. For the totals WT EAAT2 and V468I, a band at ~62 kDa is observed, consistent with the size of EAAT2, whereas the expression of F352A is decreased and there is no expression by empty vector pCMV5. WT EAAT2 and V468I showed similar surface expression, whereas no surface expression was detected for F352A.

Figure 3: Representative kinetic and expression analysis of EAAT2 mutants.

a, Representative Michaelis-Menten curves of glutamate uptake by WT EAAT2 and mutant V468I. V_max values were 324 and 347 pmol/well/min, and K_m values were 85 and 110 μM for WT EAAT2 and V468I, respectively. Data are shown as mean ± SD. b, Representative immunoblot of COS-7 cells samples transfected with pCVM5, WT EAAT2, F352A and V468I, subjected to cell surface biotinylation followed by immunoblotting. Total and surface/membrane EAAT2 bands (~62 kDa) are shown in green; actin is shown in red.

Since mutant A79S also displayed decreased surface expression and V_max, it was excluded from further study, as well as F352A. Instead, we made the substitutions A79V and F352L, which displayed non-significant changes in surface expression and V_max, compared to EAAT2 (Table 2).

Next, we performed dose-dependent glutamate uptake assays with NA-014 for WT EAAT2 and mutants. Figure 4a shows the topological diagram of the transporter including its eight transmembrane helices (TM1-TM8) and two helical hairpins (HP1 and HP2). The location of the mutation sites are indicated in both panels a and b, and color-coded by three types of effects they exerted: (1) abolished the PAM effect of NA-014 (purple), (2) did not abolish the effect but increased the EC₅₀ i.e., decreased NA-014 potency (green), or (3) did not abolish the effect (orange).

Figure 4: Identification of key residues underlying the activation of EAAT2 by the PAM NA-014.

a-b, Topological diagram of EAAT2 (a), and the structure of one protomer from two different orientations (b) are shown. Mutation sites are labeled (a-b), shown in space filling (b), and color-coded by three types of responses to the PAM NA-014 as listed Table 3. The scaffold domain is shown in the light gray in both panels, and transport domain in beige (a) or green (b). The sites 1A, 1B, 2, and 3 predicted by docking simulations are shown by cyan, blue, magenta, yellow-orange meshes in b.

Table 3 summarizes the results of the dose-response experiments that led to the identification of key residues underlying the activation of EAAT2 by the PAM NA-014, and Supplemental Figure 7a-l illustrates representative graphs of dose-response experiments of all mutants that showed different behaviors, in comparison to WT EAAT2. The effect of NA-014 on glutamate uptake was abolished in mutants H71S, V75K, A79V, F80A, M86V, K90Q, L290A, L295A, W355L, P443A, I464V, V468I, and W472I, resulting in flat dose–response curves. A representative comparison between WT EAAT2 and H71S is shown in Figure 7a; the remaining mutants with the same response patterns are not shown for simplicity. Representative dose-response curves on NA-014 on mutants M76A, A302I, I303A, F348A, S444A, G298A, F352L, T361A, S441G, S465L and R476K are shown in Figure 7b-l.

Table 3: Identification of key residues underlying the enhanced activation of EAAT2 by the PAM NA-014. Dose-response data for glutamate transport assays mediated by WT EAAT2 and mutants, in the presence of NA-014, listing efficacy and EC₅₀ values.

Membrane Domain	Transporter a	Efficacy (%): Mean [95% CI] b	EC50 (nM): Geometric Mean [95% CI] c
	WT EAAT2 (n=19)	158 [144, 172]	1 [1, 2]
Scaffold	H71S (n=5)	No Effect
	V75K (n=6)
	A79V (n=4)
	F80A (n=4)
	M86V (n=5)
	K90Q (n=4)
	L290A (n=4)
	L295A (n=4)
Transport	W355L (n=4)
	P443A (n=4)
	I464V (n=4)
	V468I (n=4)
	W472I (n=4)
Scaffold	M76A (n=3)	161 [78, 245]	97 [42, 233] *
	A302I (n=5)	132 [107, 156]	28 [14, 55] *
	I303A (n=4)	123 [100, 145]	31 [1, 813] *
Transport	F348A (n=3)	133 [97, 169]	46 [18, 119] *
	S444A (n=4)	126 [99, 152]	39 [5, 299] **
Scaffold	G298A (n=3)	125 [91, 159]	3 [1, 11]
Transport	F352L (n=3)	145 [124, 167]	4 [0.08, 166]
	T361A (n=3)	174 [67, 281]	1 [0.0004, 701]
	S441G (n=4)	148 [122, 175]	1 [0.15, 11]
	S465L (n=3)	160 [140, 180]	2 [1, 4]
	R476K (n=3)	169 [74, 265]	7 [1, 38]

CI, confidence interval.

^aWT EAAT2 and mutants were transiently transfected in COS-7 cells, and a dose response of NA-014 was performed in glutamate transport assays. Efficacy and EC₅₀ values were derived from independent experiments performed in technical triplicate; the number of biological replicates (n) is indicated in parentheses. The effect of NA-014 is indicated by the colors: purple indicates that residue abolished the effect of the compound, green indicates that residue did not abolish the effect of the compound but increased the EC_50, orange indicates that residue did not abolish the effect of the compound.

^bEfficacy values were normalized to WT (%) and are presented as mean [95% CI].

^cEC₅₀ values are shown as geometric mean [95% CI].

^b-cComparisons of efficacies and EC₅₀ between WT EAAT2 and mutant were performed using a mixed-effects model (REML), with mutants treated as a fixed effect and experimental day as a random effect to account for pairing within experiments. EC₅₀ analyses were performed on log(EC₅₀). Post hoc comparisons versus WT were conducted using Dunnett’s multiple-comparisons test (*p < 0.05 and **p < 0.01).

Efficacy values for mutants M76A, A302I, I303A, F348A, S444A, G298A, F352L, T361A, S441G, S465L and R476K were comparable to WT (p = 0.2459). However, these mutations affected EC₅₀ values (p = 0.0027). A post hoc comparison showed that mutations M76A, A302I, I303A, F348A and S444A exhibited increased EC₅₀ values relative to WT (adjusted p = 0.0422, 0.0149, 0.0277, 0.0499 and 0.0061, respectively), whereas mutations G289A, F352L, T361A, S441G, S465L and R476K exhibited EC₅₀ and efficacy values comparable to WT EAAT2 (adjusted p< 0.9999).

The results show that: (i) all the 12 Site 1A mutations on TM2 and TM5, except for G298A, either abolish the PAM action, or reduce its potency, confirming the substantial role of Site 1A (and helices TM2 and TM5). Interestingly, G298A was previously shown to increase V_max, without affecting surface expression, when compared to WT EAAT2²⁸ (Supplemental Figure 7g). P443A on HP2, also abolishes the PAM effect, which is understandable due to the critical role of HP2 acting as a gate⁴⁴, while T361A (HP1) and R476K (TM8), retain the PAM activity. This result supports the critical involvement of Site 1A in mediating the enhancement of activity driven by the PAM NA-014. (ii) Similarly, mutants V468I and W472I both at Site 2 abolish the effect of NA-014, suggesting that these spots play a potential role in binding or mediating the effect of the PAM; (iii) In contrast, Site 1B S441G and S465L (Supplemental Figure 7k), did not abolish the effect of NA-014, and S444A only reduced the potency, suggesting that NA-014 likely does not interact directly with these residues, or does not bind to Site 1B. Yet, Site 1B I464V abolished the PAM effect of NA-014. The latter could be explained by a cooperative effect with V468I and W472I, on successive turns of TM8, with all abolishing the transport-enhancing effect of NA-014; (iv) Finally, Site 3 F348A, F352L and W355L, all on HP1, show a mixed behavior. However, some EC₅₀ estimates exhibited wide confidence intervals (e.g., T361A), indicating that these values are poorly constrained by the data and should be interpreted with caution.

3.5. Glutamate uptake assays with NAM NA-010 and pan-PAM DA-038 further indicate critical residues for modulatory actions

We expanded our studies to evaluate the effects of NA-010 and DA-038. The results in Table 4 and Supplemental Figure 8a-f illustrate the effect of NA-010 on W355L, I303A, L295A, G298A, F352L, and R476K, in comparison to WT EAAT2. Table 5 and Supplemental Figure 9a-e show the effect of DA-038 on P433A, A302I, W355L, S444A and T361A.

Table 4: Identification of key residues underlying the modulation of EAAT2 by the NAM NA-010. Dose-response data for glutamate transport assays mediated by WT EAAT2 and mutants, in the presence of NA-010, listing efficacy, IC50 and EC₅₀ values.

Membrane Domain	Transporter a	Efficacy (%): Mean [95% CI] b	IC50 or EC50 (nM): Geometric Mean [95% CI] c
	WT EAAT2 (n=8)	ND d	>100,000 nM
Scaffold	H71S (n=3)	No effect
	V75K (n=4)
	A79V (n=3)
	F80A (n=3)
	M86V (n=5)
	K90Q (n=5)
	L290A (n=3)
	L295A (n=3)	127 [115, 140]	8 [1, 76]
Transport	W355L (n=3)	No effect
	P443A (n=3)
	I464V (n=3)
	V468I (n=3)
	W472I (n=3)
Scaffold	M76A (n=4)
	A302I (n=3)
	I303A (n=3)	ND d	>100,000 nM
Transport	F348A (n=3)	No effect
	S444A (n=4)
Scaffold	G298A (n=3)	129 [118, 140]	21 [2, 234]
Transport	F352L (n=4)	131 [117, 145]	5 [1,42]
	T361A (n=3)	No effect
	S441G (n=3)
	S465L (n=3)
	R476K (n=3)	186 [2, 370]	5 [1, 22]

CI, confidence interval; ND, not determined.

^aWT EAAT2 and mutants were transiently transfected in COS-7 cells, and a dose response of NA-010 was performed in glutamate transport assays. Efficacy, IC₅₀ and EC₅₀ values were derived from independent experiments performed in technical triplicate; the number of biological replicates (n) is indicated in parentheses. The effect of NA-010 is indicated by the colors: purple indicates that residue abolished the effect of the compound, orange indicates that residue did not abolish the effect of the compound, and cyan indicates that the NA-010 effect changed from a NAM to a PAM.

^bEfficacy values were normalized to WT (%) and are presented as mean [95% CI].

^cEC₅₀ values are shown as geometric mean [95% CI].

^b-cComparisons of efficacies and EC₅₀ between mutants L295, G298A, F352L and R476K were performed using a mixed-effects model (REML), with mutants treated as a fixed effect and experimental day as a random effect to account for pairing within experiments. EC₅₀ analyses were performed on log(EC₅₀).

^dND: Efficacy was not determined because full inhibition was not achieved under the experimental conditions.

For NA-010, almost all mutants evaluated in this study abolished their negative allosteric effect except for one, I303A (Table 4 and Supplemental Figure 8b). Furthermore, mutants L295A, G298A, F352L and R476K inverted its effect from a NAM (inhibition) to a PAM (activation), blue in Table 4 and Supplemental Figure 8c-f), with efficacy and EC₅₀ values comparable across these mutants (p = 0.1626 and p = 0.2616, respectively).

Additionally, W355L and F348A (Site 3) were predicted to be residues that bind NA-010 in the OF state of EAAT2, and the replacement of the aromatic rings by smaller hydrophobic residues may have impacted the allosteric action and/or binding affinity of NA-010 on Site 3. S444A, I464V and W472I are mutations at the Site 2 residues which were predicted by docking simulations to bind NA-010 in the IF state of EAAT2, and these substitutions may have also impaired binding or allosteric action.

For DA-038, most mutations also abolished the effect of the compound, except for A302I, W355L, S444A and T361A (Table 5 and Supplemental Figure 9). For these mutants, efficacy and EC₅₀ values were comparable to WT (p = 0.5521 and p = 0.7872, respectively).

While most mutants responded similarly to both PAMs, a few are distinguished by different responses (Table 6 and Figure 5). To gain a deeper understanding of the mechanistic origin of the differential responses of these mutants to the two PAMs, we focused on local interactions near these mutation sites and examined how these mutations could elicit local switches of allosteric responses resulting in the observed behaviors. Figure 5a illustrates how a triad of aromatic residues on HP1 (F348, F352, and W355) can mediate either PAM bound to Site 3, but in different ways, such that their mutations would be differentially affecting them. Specifically, F348 abolishes DA-038 effect, decreases potency of NA-014, F352 does not abolishes NA-014 effect, but abolishes DA-038 effect, and W355 abolishes NA-014 effect but not DA-038, suggesting that they may function as a molecular switch distinguishing EAAT2-specific PAM NA-014 from pan EAAT1-3 PAM DA-038. In Figure 5a, NA-014 binds near to W355 and F348 and far from F352. On the other hand, DA-038 binds closer to F352 and F348, and far from W355. These distinct interaction patterns suggest that F348, F352, and W355 may act as a molecular switch distinguishing between the two PAMs upon side chain isomerization and resulting in the experimentally observed distinctive response to the two PAMs. Additionally, Figure 5b shows the important role of a hydrophobic pocket near L295 at Site 1A (preferential site for both PAMs) and the perturbation of this tight network of interactions, also involving HP2 P443 (mutation that abolishes both compounds’ activities), possibly disrupt the allosteric effect of these two PAMs. R476 at TM8 coordinates both PAMs. As observed in our functional assays (summarized in Table 6), mutating L295 or R476 affected both activity and selectivity of the compounds, possibly through the disruption of this network.

Table 6: Summary of the results for the effects of three compounds (NA-014, NA-010 and DA-038) on all mutations.

Membrane domain	EAAT2 Mutant a	NA-014 effect b	NA-010 effect b	DA-038 effect b
Scaffold	H71S	A	A	A
	V75K	A	A	A
	A79V	A	A	A
	F80A	A	A	A
	M86V	A	A	A
	K90Q	A	A	A
	L290A	A	A	A
	L295A	A	S	A
Transport	W355L	A	A	N
	P443A	A	A	A
	I464V	A	A	A
	V468I	A	A	A
	W472I	A	A	A
Scaffold	M76A	P	A	A
	A302I	P	A	N
	I303A	P	N	A
Transport	F348A	P	A	A
	S444A	P	A	N
Scaffold	G298A	N	S	A
Transport	F352L	N	S	A
	T361A	N	A	N
	S441G	N	A	A
	S465L	N	A	A
	R476K	N	S	A

^aGlutamate transport assays were performed in COS-7 cells transiently transfected with the indicated EAAT2 mutants. Complete datasets are provided in Tables 3-5.

^bThe effects of compounds are indicated by both color and symbol: purple (A), abolished effect; green (P), reduced potency (increased EC₅₀); orange (N), no significant change; and cyan (S), switch from NAM to PAM.

Figure 5: Selected residues act as specific molecular switch distinguishing between EAAT2-specific PAM NA-014 from pan EAAT1-3 PAM DA-038.

a, F348, F352, and W355 may function as a molecular switch distinguishing EAAT2-specific PAM NA-014 from pan EAAT1-3 PAM DA-038. OF EAAT2 structure is shown in surface representation, with the scaffold domain in white and the transport domain in green. Representative docked poses of NA-014 and DA-038 at Site 3 are shown in yellow-orange and light pink mesh, respectively. NA-014 binds near to W355 and F348 and far from F352. On the other hand, DA-038 binds closer to F352 and F348, and far from W355. b, Representative docking poses of NA-014 and DA-038 at Site 1A, displayed in cyan and blue sticks, respectively. Site 1A comprises subsites, as shown in Figure 2a. Here, we displayed the most frequently sampled subsite. L295, highlighted in magenta, at the interface of scaffold and transport domains, interacts with hydrophobic residues shown in orange and green space-filling. TM2 and TM5 are rendered in transparent orange and marine cartoons, respectively. Critical gating domain HP2 is shown in yellow, with P443 in sphere representation.

Collectively, these studies demonstrate that selected mutations, H71S, V75K, A79V, F80A, M86V, K90Q, L290A, P443A, I464V, V468I, and W472I, abolished the effects of NA-014, NA-010 and DA-038, suggesting these residues serve as shared molecular determinants of the allosteric modulatory effects of these analogs, being involved in cooperative/global signaling across the entire protomers.

3.6. Ten EAAT2 residues are sufficient to ensure binding to Site 1A and enable PAM-like stimulatory activity

To further validate key role of selected EAAT2 residues involved in NA-014 stimulation and NA-010 inhibition, we focused on residues computationally predicted and experimentally verified to either abolish the effect of NA-014 or decrease its potency. We selected from amongst this set of residues those sequentially different between EAAT1 and EAAT2: H71, M76, A79, K90, L290, A302, I303, F348, S444, and V468, and generated an EAAT1 construct termed “EAAT1-EAAT2 hybrid”, with 10 EAAT1 residues replaced by their EAAT2 counterparts: S74H, K79M, S82A, Q93K, V291L, E303A, M304I, I349F, Q445S, and −I469V (see the alignment in Supplemental Figure 2). Figure 6a shows the sequence alignment of EAAT1, EAAT2 and the hybrid. The topology diagram in Figure 6b shows the location of the substituted amino acids on the secondary structure. Glutamate uptake kinetics of the hybrid showed that its V_max was lower than those of WT EAAT1 and EAAT2, but not statistically different. The hybrid showed slightly reduced surface expression compared to WT EAAT1 and EAAT2, and its K_m values were comparable to both (Supplemental Figure 10).

Figure 6: Selected EAAT2 residues are sufficient and required for conferring PAM activity.

a, Sequence alignment of WT EAAT1 and EAAT2, highlighting the 10 residues substituted in EAAT1 to generate the EAAT1-EAAT2 hybrid. b, Location of these mutations shown in the topological diagram for EAATs. The incorporated mutations to EAAT1 are listed, and the corresponding EAAT2 mutant is written in parentheses. Purple color refers to EAAT2 mutants that abolished the effect of NA-014, and green to those that did not abolish the effect but increased the EC₅₀ of NA-014.

Dose-response assays showed that NA-014 increases EAAT2-mediated uptake (EC₅₀ of 1 nM [0.03, 54] and efficacy of 142 % [78, 206]), but has no effect on EAAT1 (Supplemental Figure 11a-b), as previously reported³⁰. We found that NA-014 increased glutamate uptake by the EAAT1-EAAT2 hybrid, with an EC₅₀ of 3 nM [0.04, 293] and efficacy of 163 % [139, 187]. The EC₅₀ and efficacies values of the EAAT1-2 hybrid were comparable to WT EAAT2 (unpaired Welch’s t-test, t(3.86) = 0.67, p = 0.54 for EC₅₀; t(2.56) = 1.28, p = 0.30 for efficacy; Figure 11b). These results show that 10 substitutions in EAAT1 are sufficient to generate EAAT2-like response to the PAM NA-014.

NA-010, also shown previously³⁰, does not affect EAAT1 activity but decreases EAAT2 activity (IC₅₀ > 100 μM, Supplemental Figure 11c). Interestingly we find that NA-010 shifts from a NAM to a PAM in the EAAT1-EAAT2 hybrid (EC₅₀ of 10 nM [3, 136] and efficacy 176 % [59, 292], Supplemental Figure 11d). Collectively, our results point to a dominant role of these PAM-associated 10 residues in enabling the binding of PAMs and for facilitating the transport enhancing activity of the PAMs. As we observed with a few EAAT2 mutants that elicited a change of activity of NA-010 from a NAM to a PAM (Table 6), this observation also suggests that very subtle changes in the interaction between the transporter and compound can determine whether the compound acts as a NAM or a PAM.

4. Discussion

This study explored compounds that differentially modulate EAAT activity to enhance our understanding of EAAT modulation mechanisms. We selected three compounds, an EAAT2-selective PAM, NA-014; an EAAT2-selective NAM, NA-010; and a pan EAAT1-3 PAM, DA-038. Docking simulations pointed to four binding sites, Sites 1A, 1B (or Site 4), 2, and 3, and key residues therein, selectively binding those compounds in the OF or IF states of EAAT2. Both PAMs were observed to bind to Site 1A (strongest preference; −5.6 kcal/mol, based on PRODIGY-LIG). They were also predicted to bind Site 3, irrespective of the conformational state of the transporter, and show a tendency to occasionally bind to Site 1B in the OF state and Site 2 in the IF. In contrast, the NAM NA-010 bound to Site 1A only in the OF state and bound to Site 1B/4 in both states, and to Site 2 in the IF EAAT2. These distinct preferences of PAMs and NAM may underlie their different pharmacological activities. Out of residues that were predicted to interact with the modulators several stood out: M76, A79, M86, K90 on TM2, L295, G298 and A302 on TM5; A302, F348, F352 and W355 on HP1; G438-S444 on HP2; and V468 and R476 on TM8. These were mostly located at or near Site 1A, at the interface between the scaffold and transport domain and exhibited strong binding affinities in our computations (Supplemental Table 1). Site 1B was occupied by detergents or lipids in EAAT1 experiments. ¹² In addition, Site 2 overlaps with an inhibitor (WAY213613) binding site in the IF structure,⁴² Sites 3 and 4 are likely to bind to lipids based on the resolved IF EAAT2 structure.

The above computationally predicted key residues were experimentally assessed through point mutagenesis and glutamate uptake assays. We classified the effects of mutations on the activity of the compounds in three groups: complete abolishment of the modulatory effect; no abolishment but decrease in the potency (manifested by increased EC₅₀); and no effect compared to WT. The former group of residues were interpreted as important for allosteric modulation. Similarly, if the EC₅₀ was altered by a mutation, we considered this residue to be important to activity and binding, at least indirectly. When mutants did not alter the action of the compound, these residues were estimated to not be involved in the compound binding or action. Out of 24 point mutations examined, ~ 80% either abolished the effect of all three compounds or increased EC₅₀, which supports the significance of computationally predicted residues in binding the modulators or coordinating their allosteric effects.

Although some mutants investigated did not abolish the effect of NA-014 (Supplemental Table 2, yellow regions), they exhibited functional differences, pointing to a role in the transport cycle. Most of the residues predicted to bind this PAM were located on TM2 and TM5 (and a few on HP1, HP2 and TM8), all facing the interface between the scaffold and transport domains (Figure 6a-b). They are positioned in a region crucial for the elevator-like sliding/twisting movement of the transport domain,^14,15,45-48 past the trimerization/scaffold domain. Mutations at those sites can thus readily impact the function in either direction, facilitating or obstructing the movement, which would be reflected on altered actions of the modulators. For example, the PAM activity could ease the ‘sliding’ of the transport domain between its OF and IF states.^{12,37,42,43,49,50}

In principle, different sites may be accessible during the reconfiguration of the transporter during the transport cycle,^37,42,49,50 and as such we cannot rule out the occasional occurrence of binding to other sites, some of which were experimentally confirmed to be influential (i.e. Sites 1B/4 and 3). This may also originate from their spatial proximity to Site 1A. Future studies using (time-resolved) cryo-EM and smFRET⁵¹ could provide additional information on the kinetics of global transitions under different conditions. While direct structural confirmation of compound binding at Site 1A will require future cryo-EM or structural studies, the convergence of docking predictions, mutagenesis data, and functional analyses strongly supports this region as a primary determinant of PAM activity.

A few mutants were investigated in previous studies: (i) S441G exhibited here a higher substrate affinity (decreased K_m), consistent with a previous study,⁴² (ii) G298A, reported previously to increase V_max compared to WT EAAT2.²⁸ The effect of NA-014 was not altered in the G298A, compared to WT EAAT2, suggesting no additive or synergistic effect by the combination of the ‘gain-of-function’ mutant and the PAM compound. Similar results were also seen with the early analog, GT949,²³ suggesting that GT949, NA-014 and G298A may share a molecular mechanism, possibly by reducing the energy barrier necessary to bring the transport domain from the OF to IF state,^46,52,53 (iii) Cryo-EM studies revealed that the selective EAAT2 inhibitor WAY213613 were not abolished by I464V and V468I.³⁷ In contrast, our data indicated that these mutants, predicted to line Site 2 of IF EAAT2 (Supplemental Table 1), abolished the effects of NA-014, NA-010 and DA-038, supporting the notion that our compounds act through an allosteric mechanism, distinct from the inhibitory mechanism of WAY213613; (iv) A previous study,⁵⁴ reported that M76A and F80A in TM2 of EAAT2 decreased uptake and surface expression. In our hands, these mutants showed V_max and surface expression values comparable to WT, possibly reflecting differences in experimental conditions, such as the use of HeLa cells instead of COS-7 cells.

Another study observed a lack of effect for GT949 in a JumpIn-EAAT2 HEK cell line.⁵⁵ This may indicate that factors specific to COS-7 cells, such as the lipid composition of the cell membrane^56,57 or protein composition of the cell,^58,59 may influence this effect. However, our previous publications also evaluated the effect of our compounds on primary glia cultures, that endogenously express EAAT2 transporters, and found that the effect is consistent with the effects in COS-7 cells,^22,30 and the mutagenesis results of the current study give us confidence that NA-014 function as an EAAT2 PAM by interaction with EAAT2. However, future studies using smFRET, cryo-EM and molecular dynamic simulations could evaluate how different cell lines and, more specifically, lipid composition and protein interactions impacts transporter activity and allosteric modulation.

Uptake experiments with mutants L295A, G298A, F352L and R476K showed that NA-010 switched activity from a NAM to a PAM. As G298A, F352L and R476K did not abolish the effect of NA-014, this suggests that these residues may be important for conferring NAM activity. On the other hand, L295A abolishes the effect of NA-014, suggesting that this residue is important for both NA-014 and NA-010, in addition to having a role in determining modulatory activity. Structurally, L295 is located very close to Site 1A, and appears to play a direct role in stabilizing and facilitating NA-014 interactions. Therefore, L295A mutation likely perturbs structural changes necessary for NA-014 interactions that are important for mediating activation leading to a loss of function (abolishing effect). However, in case of NA-010, substituting with alanine, that likely weakens domain packing and increases flexibility at this junction, results in a mutant that can now accommodate NA-010 in a subtly different way that enables the interaction to enhance glutamate transport activity. For DA-038, we found that A302I, W355L, T361A and S444A did not abolish its effect on EAAT2. As A302I and S444A mutations did not abolish the effect of NA-014 but decreased its potency for transport augmentation (increased EC₅₀), this suggests that the residues A302 and S444 are important for mediating the effect of NA-014, but not that of DA-038 in EAAT2. This implies that DA-038 acts as an EAAT2 PAM through interaction with different residues from NA-014.

Based on the glutamate uptake assays with NA-014, we identified ten EAAT2 mutations that abolished the effect or decreased the potency of NA-014 and were not conserved between EAAT1 and EAAT2. To investigate the role of these amino acids as molecular determinants of NA-014 activity, we created the EAAT1-EAAT2 hybrid, an EAAT1 construct in which these 10 EAAT2 residues replaced their EAAT1 counterparts. Strikingly, NA-014 (which was not an EAAT1 PAM) enhanced the activity of the hybrid. This ‘gain-of-function’ strongly supports the significance of these ten residues as determinants of the PAM activity of NA-014. Additional combination of mutants may be necessary to further establish and/or narrow down the molecular determinants of the observed behavior.

Because EAATs are the principal regulators of extracellular glutamate and serve functions beyond preventing excitotoxicity they are potential therapeutic targets. By shaping synaptic glutamate kinetics, EAATs influence AMPA versus NMDA receptor activation, limit spillover to extrasynaptic receptors, and define the timing and precision of excitatory signaling that governs synaptic plasticity (LTP vs. LTD). Thus, EAATs actively modulate glutamatergic transmission.

From a therapeutic standpoint, allosteric modulators offer an advantage over EAAT expression enhancers, as they preserve physiological glutamate transients while tuning transporter function. Because EAATs undergo multiple conformational transitions during the transport cycle, allosteric modulators can fine-tune clearance kinetics without altering expression levels. By adjusting transport rates and conformational state occupancy, these compounds may differentially influence synaptic versus extrasynaptic signaling, providing refined and potentially reversible control over glutamate dynamics.

Functionally, EAAT2 NAMs modestly slow glutamate clearance and may enhance synaptic glutamate availability in hypoglutamatergic states. Schizophrenia, for example, is associated with reduced glutamatergic tone and NMDA receptor hypofunction, leading to impaired cortical output and cognitive deficits.^60-66 Unlike direct NMDA-targeting strategies, which have been limited by adverse effects, EAAT2 NAMs may modestly elevate synaptic glutamate while maintaining physiological regulation. Similar considerations may apply to Rett syndrome ^67,68 and to mood or obsessive–compulsive disorders linked to reduced glutamatergic signaling.⁶⁹ Thus, disorders characterized by reduced glutamatergic tone may benefit from NAMs. In addition, NAMs may serve as mechanistic tools to model transporter hypofunction.

Conversely, conditions associated with excessive extracellular glutamate, including neurodegenerative disorders, epilepsy, chronic pain, and substance use disorders, may benefit from EAAT2 PAMs, which accelerate glutamate clearance, reducing pathological spillover and extrasynaptic NMDA receptor activation. Genetic studies linking SLC1A2 ^70,71 or SLC1A3 ^72,73 mutations to epilepsy and ataxia further support this rationale. Many of these variants reduce transporter function or impair glutamate clearance, promoting neuronal hyperexcitability and excitotoxic stress. In such cases, PAMs may enhance residual transporter activity and restore glutamate homeostasis, suggesting that gain-of-function modulation of EAAT activity may be particularly relevant for these genetically defined epilepsies and ataxias.

In addition to substrate transport, EAATs possess an uncoupled anion conductance that can be altered independently of glutamate transport. Several disease-associated variants enhance anion currents in EAAT1,^74-76 or EAAT2.^77,78 For some of these diseases, pathogenetic links between gain-of-anion channel function and neurological symptoms have been reported.⁷⁹ The extent to which the PAMs and NAMs described here influence transporter-associated anion currents remains unknown. Because transport and anion gating are mechanistically coupled yet partially separable processes,^10,50 it is conceivable that allosteric modulators could differentially affect substrate flux and anion conductance. Determining whether these compounds modulate anion currents will be essential for evaluating their therapeutic suitability in disorders involving altered anion channel activity.

Overall, our studies contribute to the current knowledge on the mechanism of allosteric modulators compounds at biochemical, structural, and molecular levels. This may inform and support the future discovery of new, improved EAAT allosteric modulators that could lead to breakthrough therapeutics for multiple neurological and neuropsychiatric disorders in which glutamate dysregulation is implicated.

Supplementary Material

This article has supplemental material available at molpharm.aspetjournals.org.

Abbreviations

CNS

Central nervous system

cryo-EM

Cryo-electron microscopy

EAAT2

Excitatory Amino Acid Transporter 2

EAAT1

Excitatory Amino Acid Transporter 1

EAAT3

Excitatory Amino Acid Transporter 3

EC

extracellular conformation

ETB-TBOA

(2S,3S)-2-amino-3-[(3-{4-[(1,2-3H2)ethyl]benzamidophenyl)methoxy]butanedioic acid

HP

Hairpin

IC

intracellular conformation

IF

inward facing

NAM

Negative allosteric modulator

OF

outward facing

PAM

Positive allosteric modulator

PBS

Phosphate-buffered saline buffer

smFRET

single molecule fluorescence resonance energy transfer

TFB-TBOA

(3S)-3-[[3-[[4-(Trifluoromethyl) benzoyl]amino]phenyl]methoxy]-L-aspartic acid

TM

Transmembrane

WT

wild type

Data availability

Previously published structural models are available from the PDB under accession codes 5LLU, 7XR6 and PDB: 7VR8. Plasmid backbones designed in this study are available on Addgene (CMV-hEAAT2, 32814; CMV-hEAAT1, 32813), pCMV5 was a gift from Susan Amara (NIH). All other mutant plasmids and data are available upon request.