N-methyl-D-aspartate receptors (NMDARs), are key components of ionotropic glutamate receptors (Hansen, 2018). It forms a dihetero- or triheterotetramer composed of two identical obligatory GluN1 subunits and two identical or different variable subunits, GluN2 or GluN3 (Stroebel et al., 2018, Xiong et al., 2025). NMDAR activation requires the binding of the endogenous ligands glycine and glutamate, along with membrane depolarization to remove the Mg2+ blocker, thereby enabling activation (Iacobucci and Popescu, 2017). For this reason, some researchers refer to the NMDAR as a “coincidence detector” as it integrates both chemical and electrical signals. Upon activation, NMDARs allow Ca2+ to flow inward and Na+ to flow outward; inward Ca2+ subsequently activates a wide array of Ca2+ −dependent signaling pathways. Thus, NMDARs play a significant role in various pathological processes (Dupuis et al., 2023).

Owing to their importance, NMDARs are subject to precise regulation, including allosteric modulation, which includes both positive and negative modulation (Gibb, 2022). Allosteric modulators bind to sites distinct from orthosteric glutamate or glycine binding pockets, inducing conformational changes that can alter channel opening probability, ion conductance, or desensitization kinetics. Positive allosteric modulators (PAMs) enhance receptor activity, whereas negative allosteric modulators (NAMs) reduce it. This mechanism allows subtype-selective regulation by exploiting structural differences outside the highly conserved orthosteric site. Allosteric modulation provides a highly efficient and precise means of regulation without affecting neurotransmitter release or reuptake levels, making it an attractive method for therapeutic modulation of receptor function. Many receptor modulators have been discovered, including endogenous modulators such as protons and Zn2+, which target the NTD of the GluN1-GluN2A receptor (Jalali-Yazdi et al., 2018). Zn2+ has also been shown to modulate GluN1-GluN3A receptors allosterically via NTD (Madry et al., 2008). Additionally, spermine (Yi et al., 2019), a naturally occurring polyamine widely expressed in eukaryotic cells, is also allosteric modulators of NMDARs. Other endogenous modulators include neurosteroids such as pregnenolone sulfate (PS), which can potentiate certain NMDAR subtypes, and extracellular matrix components like glycosaminoglycans, which have been reported to influence receptor gating. These modulators participate in physiological processes such as synaptic plasticity and in pathological states including ischemia, epilepsy, and neurodegenerative diseases. Exogenously, researchers have discovered and developed various NMDAR allosteric modulators, including TCN-201 for GluN1-GluN2A (Steigerwald et al., 2022), Onfasprodil for GluN1-GluN2B (Gomez-Mancilla, 2023), and CIQ for GluN1-GluN2C and GluN1-GluN2D (Dembeck et al., 2024). Some traditional pore blockers are now also found to exhibit allosteric modulation, with recent studies showing that ketamine can allosterically modulate NMDARs (Abbott, 2024). Several of these compounds have entered clinical trials, such as GLYX-13 (rapastinel) and apimostinel, which act as functional modulators with potential antidepressants or cognitive-enhancing effects. The diversity of binding sites and mechanisms offers multiple therapeutic entry points, making the systematic characterization of NMDAR allosteric modulators essential for drug discovery.

Traditional structural biology studies have provided insights into the structural basis of certain NMDAR allosteric modulators (Jalali-Yazdi et al., 2018, Zhang et al., 2018). However, owing to limitations in resolution and resource requirements of conventional biology, many structural foundations of allosteric modulation remain unexplored. In particular, as a membrane protein, NMDAR is inherently difficult to crystallize. Moreover, many allosteric modulators exhibit transient binding and conformational heterogeneity upon interaction with the receptor, further complicating efforts to elucidate their structural mechanisms using traditional structural biology techniques. Protein structure prediction offers a new approach to address these issues. The Nobel Prize-winning AlphaFold (Abramson, 2024) and RoseTTAFold All-ATOM (Krishna, et al., 2024) utilize sequence depth analysis combined with evolutionary perspectives for protein structure prediction, achieving and demonstrating high accuracy. Additionally, some deep learning algorithms can predict protein-small molecule complexes based on AlphaFold’s capabilities.

In this study, we first used single-cell sequencing analysis to explore the temporal and spatial expression patterns of the NMDAR subunits. Our findings revealed significant diversity in the temporal, spatial, and cell type-specific distribution of NMDAR subunits and subtypes. Next, we employed AlphaFold Multimer and AlphaFold3 to predict the structures of GluN1-N2 diheteromeric NMDARs and compared them with the available cryo-EM structures. The results, along with validation using traditional biochemical methods, indicated that these predictions were highly accurate. Based on these structures, we utilized RoseTTAFold-All-Atom to systematically predict the binding sites and key interacting amino acids of most, if not all, of the known allosteric modulators. These results will provide a scientific basis for the treatment of diseases associated with NMDAR abnormalities.

Our predictions provide structural insights into the subtype-specific allosteric modulation of NMDARs by various small molecules, thereby addressing the limitations of traditional structural biology in this area. These results may contribute to structure-based drug development and help guide functional studies on NMDAR regulation. Moreover, this study demonstrated an integrative strategy combining protein structure prediction, single-cell transcriptomics, and biochemical validation, which may serve as a useful framework for investigating other complex membrane proteins.