Regulation of the NMDA receptor by its cytoplasmic domains: (how) is the tail wagging the dog?

Abstract

Excitatory neurotransmission mediated by N-methyl-D-aspartate receptors (NMDARs) is critical for synapse development, function, and plasticity in the brain. NMDARs are tetra-heteromeric cation-channels that mediate synaptic transmission and plasticity. Extensive human studies show the existence of genetic variants in NMDAR subunits genes (GRIN genes) that are associated with neurodevelopmental and neuropsychiatric disorders, including autism spectrum disorders (ASD), epilepsy (EP), intellectual disability (ID), attention deficit hyperactivity disorder (ADHD), and schizophrenia (SCZ). NMDAR subunits have a unique modular architecture with four semiautonomous domains. Here we focus on the carboxyl terminal domain (CTD), also known as the intracellular C-tail, which varies in length among the glutamate receptor subunits and is the most diverse domain in terms of amino acid sequence. The CTD shows no sequence homology to any known proteins but encodes short docking motifs for intracellular binding proteins and covalent modifications. Our review will discuss the many important functions of the CTD in regulating NMDA membrane and synaptic targeting, stabilization, degradation targeting, allosteric modulation and metabotropic signaling of the receptor.

1. Introduction

NMDA type ionotropic glutamate receptors play central roles in synapse development, synaptic transmission, plasticity, and are thus potential targets for treatment of a wide set of brain disorders. NMDARs are tetramers composed of two obligate GluN1 subunits along with two subunits of the GluN2 family (GluN2A-D)(Cull-Candy et al., 2001; Sanz-Clemente et al., 2013b); or a combination of GluN2 and GluN3 (GluN3A-B) subunits (Monyer et al., 1992; Schorge and Colquhoun, 2003; Ulbrich and Isacoff, 2007, 2008). NMDARs conduct both Na⁺ and Ca²⁺ ions, and calcium influx mediated by NMDARs is pivotal to neuronal signaling and plasticity. Because individual neurons normally express at least two different GluN2/3 subunits, native NMDARs can occur as di-heteromers, incorporating two GluN1 and two identical copies of a given GluN2 or GluN3 subunit, or as tri-heteromers, incorporating two GluN1 subunits with two distinct GluN2 subunits or one each of GluN2 and GluN3 subunits. Compelling evidence indicates that di-heteromers and tri-heteromers coexist within a single cell or even at a single synapse, conferring distinct biophysical and pharmacological properties to each receptor; and the different NMDARs mediate distinct calcium responses (Kellermayer et al., 2018; Paoletti et al., 2013; Stroebel et al., 2014; Tovar and Westbrook, 2017; Warming et al., 2019). One particularly curious example is NMDARs composed of two GluN1 and two GluN3A subunits, making a glycine-activated excitatory NMDAR that is impermeable to calcium (Grand et al., 2018). GluN subunit contribution to the synaptic population and NMDAR excitatory postsynaptic current (EPSC) depends on their activity-dependent interaction with intracellular molecules that regulate the assembly, transport, targeting, and anchoring of receptors (Aoki and Sherpa, 2017; Jeyifous et al., 2009; Lau and Zukin, 2007; Matsuzaki et al., 2004; She et al., 2012).

Rodent and human NMDAR subunit homologs are highly conserved, and receptors comprised of like subunits, demonstrate no major differences in the pharmacological and functional properties between species (Hedegaard et al., 2012). Each NMDAR subunit has a unique modular architecture with four semi-autonomous domains (Fig 1A, C). The ligand-binding domain (LBD) and ion channel pore organized by transmembrane domain (TMD) are the most conserved regions, as discussed extensively by Stroebel and Paoletti (Stroebel and Paoletti, 2020).

Figure 1.

A Schematic diagram of GluN1/GluN2 NMDAR subunits with relevant NMDAR modulators. Each NMDAR subunit has a unique modular architecture with four semi-autonomous domains: the extracellular amino-terminal domain (ATD), the extracellular ligand-binding domain (LBD), a large transmembrane domain (TMD) consisting of three transmembrane spanning segments (M1, M3, M4) and a pore forming re-entrant loop (M2), and an intracellular carboxyl-terminal domain (CTD). NMDAR blockers interact with different receptor domains as demonstrated. Listed NMDAR agonist and co-agonists: L-glutamate, D-serine and glycine; antagonists and blockers: CK7: 7-chlorokynurenate, APV: (2R)-amino-5-phosphonovaleric acid, or (2R)-amino-5-phosphonopentanoate; Ifenprodil: NP-120; MK-801 also known as Dizocilpine; Magnesium and Zink. B The histogram demonstrates length in the number of amino acids of each NMDAR subunit CTD and its splice variant CTD in humans (count from 1aa). The GluN CTDs exhibit great diversity among glutamate receptor subunits. C Linear representation of GluN subunit architecture that is common to all NMDA subunits.

The CTD exhibits great diversity amongst glutamate receptor subunits, varying both in sequence and length (Fig 1B). These long CTDs, ranging from 50 amino acids (aa) in GluN1 to 650 aa in GluN2B, allow for a multiplicity of intracellular interactions that affect the biophysical, pharmacological, and signaling attributes of NMDARs. Critical roles for the CTDs were revealed in the findings that transgenic mice expressing GluN2B with truncated CTD subunits died perinatally (Sprengel et al., 1998). Electrophysiological studies in cultured neurons from GluN2B∆CTD mice showed a complete absence of synaptic GluN2B-containing NMDARs (Mori et al., 1998; Sprengel et al., 1998). Similarly, mice expressing GluN2A with a truncated CTD were ataxic, exhibited impaired NMDAR localization to synapses (Steigerwald et al., 2000), and their synapses failed to undergo long-term potentiation (LTP), a use-dependent increase in synaptic efficacy (Sprengel et al., 1998). These studies provided the first hints that the GluN2A and GluN2B CTDs might regulate function by controlling NMDAR trafficking and localization (Wenthold et al., 2003), as discussed in detail below. Whole exome sequencing of patients with different neurodevelopmental and neuropsychiatric disorders has identified numerous variants in the NMDAR subunit-encoding GRIN genes, and about 1 in 5 of these variants are located in the CTD coding regions (XiangWei et al., 2018).

The CTD of the obligatory NMDAR subunit, GluN1, has key roles in the assembly and trafficking of NMDARs (Fukaya et al., 2003; Mu et al., 2003). The GluN2 subunit CTDs also interact extensively with cellular binding partners to control NMDAR trafficking to the surface, localization to synaptic and extra synaptic sites, and activation of downstream signaling cascades (Okabe et al., 1999; Prybylowski et al., 2005; She et al., 2012; Steigerwald et al., 2000; Yi et al., 2007). By comparison, interactions with the GluN3 CTD have been comparatively understudied (Otsu et al., 2019). In this review, we discuss the known functions of the NMDAR subunit CTDs, and how they govern NMDAR function in an activity dependent manner.

2. NMDA receptor subunit cytosolic tail domain composition, variation, and modifications

Genes encoding the human GluN1, GluN2A, GluN3A and GluN3B subunits undergo a large diversity of alternative mRNA splicing events. Most human-specific GluN splice variants occur within the amino terminal domain (ATD) and CTD (Fig 1A–C) (Laurie and Seeburg, 1994; Lee et al., 2014; Zukin and Bennett, 1995). Hence, we will use examples from all human GluN splice variants in the CTD and describe all relevant functional residues in them. In addition, GluN CTDs are subject to extensive covalent modifications, including phosphorylation, ubiquitination, and palmitoylation, which also impact receptor stability, trafficking and synaptic localization (Goebel-Goody et al., 2009; Harrison et al., 2007; Lee, 2006; Lussier et al., 2015). We will discuss these modifications to the CTDs in the context of how they impact interactions with other protein and lipid partners and how these impact overall NMDAR function.

2.1. The GluN1 CTD undergoes extensive alternative splicing in human.

The GluN1 subunit undergoes alternative splicing to generate 8 distinct isoforms via alternative inclusion of exon 5 in the ATD (GluN1–1b), and exons 21 and 22 in the CTD (GluN1-xa, where x is number 1 to 4 depending on the splice variant). Following exon 20, termed the C0 “cassette”, distinct GluN1 isoforms are generated by alternative splicing of exon 21 (encoding C1) and 22 (encoding C2 or C2’). Exon 22 is alternatively spliced into either C2 (38aa) or C2’ (22aa) (Zukin and Bennett, 1995). GluN1–1a carries cassettes, C0-C1-C2, making it the longest of the splice variants. GluN1–2a contains C0-C2, GluN1–3a consists of C0-C1-C2’, and GluN1–4a has C0-C2’ (Fig 2A) (Sugihara et al., 1992; Zukin and Bennett, 1995). Alternative splicing of exons 21 and 22 does not impact the functional properties nor pharmacological sensitivities of NMDARs, but the different CTDs mediate distinct sets of interactions with cellular binding partners that impact receptor trafficking to the cell surface, including both synaptic and extrasynaptic localization (Mu et al., 2003; Scott et al., 2001). For example, the GluN1 regulatory phosphorylation site (Y⁸³⁷), located in C0, is found in all splice variants, where it regulates NMDAR internalization. Two splice variants, GluN1–1a and GluN1–3a, have 4 additional phosphorylation sites in the C1 cassette (Tingley et al., 1993) (Fig1). The distinct GluN1 variant expression patterns in the rodent brain suggest that they differentially impact neuronal function (Laurie and Seeburg, 1994; Monyer et al., 1994). Additionally, changes in the expression levels of GluN1–3a and GluN1–4a variants in response to decreased neuronal activity suggest that splicing may be coordinated to adjust the regulation of NMDAR function in response to activity (Mu et al., 2003; Xie and Black, 2001).

Figure 2.

A Diagram of GluN1A CTD splice variants and their covalent modifications. GluN1A-1 carries all cassettes- C0-C1-C2 - and is the longest of the splice variants, GluN1A-2 contains C0-C2, GluN1A-3 consists of C0-C1-C2’, and GluN1A-4 has C0-C2’ (cassettes are color coded). Within the GluN1A CTD a total of 5 phosphorylation sites are noted, with one located in C0 and four localizing in C1 (covalent modifications are color coded in caption by kinase). B GluN1 splice variants ER retention signals (in red) and protein binding motifs (color coded as in caption); GluN1A cassettes separated by dashed lines. C Schematic representation of all human GluN2 and GluN3 CTD covalent modifications and protein interaction motifs (with Y axes showing aa number on the left). The GluN2A-S splice variant is included for reference, with the differing sequence of 23 aa marked in black. The GluN2A and GluN2B CTDs have many identified sites for covalent modifications associated with function (color-coded), but more were identified by phosphoproteomics (white circles). The GluN2C and GluN2D CTDs are known to have only 2 and 1 phosphorylation sites, respectively. Much less is known about the effects of covalent modifications and protein interactions of GluN2D, GluN3A and GluN3B subunits, although phosphoproteomics has identified several sites. All discussed interactions, and some additional protein interactions, are marked on the tails and are color coded. CTD variants listed in Table 1 are marked accordingly by red stars on the main GluN splice variants. GluN2 CTDs are aligned and numbered according to UniProtKB data base with the following accession numbers: GluN1–1a: Q05586–1; GluN1–2a: Q05586–3; GluN1–3a: Q05586–4; GluN1–4a: Q05586–2 ; GluN2A-F: Q12879–1; GluN2A-S: Q12879–2; GluN2B: Q13224; GluN2C: Q14957; GluN2D: O15399; GluN3A: Q8TCU5; GluN3B: O60391.

2.2. GluN2A is also alternatively spliced in humans.

Human GluN2A is expressed as two alternative splice isoforms, termed either full or short (GluN2A-F and GluN2A-S), with GluN2A-S having a shorter CTD of 443 aa as compared to 626 aa in GluN2A-F (Fig 1B)(Warming et al., 2019). Western blots of postmortem human cortex found that GluN2A-S is present at about one third the level of GluN2A-F, whose expression remains stable throughout different ages. The GluN2A-S splice variant is missing 183 aa of the full subunit CTD, which is replaced with new sequence of 23 aa. Based on this, GluN2A-S lacks the following sites: Protein kinase C (PKC) and Src family kinase (SFK) phosphorylation sites, as well as SH3 (Src homology 3) binding motif (Fig 2C) (Cousins et al. 2009). These sites impact receptor localization and may affect synaptic plasticity, raising questions of how the loss of these sites impact the trafficking and function of GluN2A-S-NMDARs (Al-Hallaq et al., 2007; Warming et al., 2019).

2.3. GluN3A and GluN3B subunits are alternatively spliced in rodents.

Two GluN3A splice variants have been observed in rodent brain, but not in humans (Sasaki et al., 2002; Sun et al., 1998). The longer form, termed GluN3A-Long (GluN3A-L), contains an insertion of 20 aa at the start of its CTD. Sequence analyses have identified potential regulatory PKC, Protein kinase A (PKA), and calmodulin dependent kinase II (CaMKII) phosphorylation sites within this insertion (Sun et al., 1998). The expression patterns of the GluN3A-L and the shorter isoform (GluN3A-S) are region-specific, suggestive of possible functional diversity among GluN3A- NMDAR.

GluN3B has 5 splice variants found in rodents, which originate by alternating splicing. Two GluN3 splice variants lack large parts of the protein including the transmembrane M3 and M4 domains and most of the CTD. GluN3B^∆600 lacks exons 4 to 8, which removes transmembrane domain M4 and most of the CTD. The GluN3B^∆1125 variant has spliced out exons 3 to 9, removing transmembrane domains M2, M3, M4 and most of the CTD (de Jesus Domingues et al., 2011). Surprisingly, these GluN3 splice variants assemble with GluN1 to form functional NMDARs, albeit with altered peak calcium influx (de Jesus Domingues et al., 2011). Notably, the GluN3B^∆600 and GluN3B^∆1125 splice variants are only expressed in white matter glial cells. However, due to the relative dearth of studies on the GluN3B CTD function, it is not clear how loss of the CTD changes the receptor.

3. NMDA receptor CTD protein interactions and their roles in receptor function.

NMDAR subunit CTDs physically associate with dozens of different proteins and protein complexes, including interactions with other ion channels, adhesion receptors, and signaling proteins (Frank and Grant, 2017; Frank et al., 2016; Husi et al., 2000a). These interactions are crucial for proper NMDAR localization and function, as well as proper activation of downstream signaling cascades. Many of these NMDAR-interacting proteins (calmodulin (CaM), CaMKII, PKA, PKC, calcineurin (PP2B), α-actinin and many more) bind directly to the GluN CTD, or interact through scaffolding proteins, such as PSD-95, that organize larger protein complexes (Frank et al., 2017, 2016; Husi et al., 2000b). Hence, GluN subunit CTDs can serve as a molecular hub for convergence of regulatory and signaling events.

NMDARs localize to both synaptic and extrasynaptic sites (Yan et al., 2020; Zhou et al., 2015). Synaptic and extrasynaptic NMDARs play different roles in neuronal development, excitatory synaptic transmission, learning and memory, and their dysfunction contributes differently to neurological disorders (Chen et al., 2008; Gladding and Raymond, 2011; Grol et al., 2012; Mao et al., 2009; Yan et al., 2020). Synaptic versus extrasynaptic NMDAR localization is believed to be controlled by the membrane-associated guanylate kinases (MAGUK) protein family. Examples of MAGUKs include PSD-95/SAP-90, PSD-93/chapsyn-110, SAP-97/hDlg, and SAP-102. MAGUKs are crucial scaffolds in organizing protein complexes in the post synaptic density (PSD) (Frank et al., 2016; Kim and Sheng, 2004; Sheng and Sala, 2001; Won et al., 2017). MAGUKs have PDZ (post-synaptic density-95/discs-large/zona-occludens-1) domains that bind PDZ-binding motifs (ES[D/E]V) which are found in the GluN2 and GluN1–3/4a variants CTDs (Fig 2A)(Cousins et al., 2009). However, MAGUKs have SH3 domains and can also bind SH3 domain-binding motifs (typical PXXP or atypical YXXY, Fig 2C) found in GluN2A and GluN2B, but the implications of this interaction interface remain to be studied (Cousins et al., 2009; Cousins and Stephenson, 2012). GluN3 subunits do not have PDZ motifs, suggesting they do not interact with MAGUKs.

Some MAGUKs have overlapping roles in regulating NMDARs. For example, knockdown of PSD-95 and PSD-93 did not significantly impact synaptic NMDAR signaling (Elias et al., 2006; Elias and Nicoll, 2007), while genetic or short hairpin-mediated knockdown of PSD-95, PSD-93 and SAP-102 strongly decreased NMDAR synaptic signaling in cultured neurons and in vivo (Chen et al., 2015; Elias and Nicoll, 2007; Frank et al., 2016). Other proteins, such as Rabphilin3A (Rph3A) which binds to the GluN2A CTD (1349–1389aa, Fig 2B, 3) appear to cooperate with PSD-95 to localize NMDARs at the synapse (Stanic et al., 2015). Both PSD-95 and PSD-93 are required to anchor GluN2B, but not GluN2A-NMDAR, by creating a tripartite complex with CTD (Fig 3)(Frank et al., 2016). This evidence suggests that anchoring complexes bind to CTD by organizing multi-protein complexes.

Figure 3.

Schematic diagram of NMDAR trafficking from the endoplasmic reticulum to the synapse. In the canonical secretory pathway (marked by blue arrows), assembled NMDARs are trafficked from the ER to the TGN, where they are sorted and packaged into transport vesicles with motor proteins to be delivered on microtubules to the plasma membrane near synapses. NMDAR association with MAGUKs and the exocyst complex starts in the somatic ER. GluN1A-3/4 splice variants can promote assembled NMDARs to exit from the ER by associating with COPII. Part of the NMDARs are diverted from the somatic ER into a specialized dendritic ER sub-compartment (dER) that bypasses TGN, merging instead with dGolgi outposts on their path to the synapse (marked by orange arrows). This alternative path is usually exerted by receptors transported inside the dER all the way to the dGolgi outposts. Accumulation of the receptors in the dER allows for faster local processing in the dGolgi and synaptic delivery. Another non-canonical trafficking path was demonstrated for GluN2-NMDARs where they bypass the TGN, through vesicles transported by motor proteins along microtubules, to reach the dGolgi. (following green arrows). Sorting of the GluN2-NMDARs into this pathway is coordinated by SAP-97:CASK and involves a KIF17-containing complex and possibly other KIFs as KIF5b.

Some observations also hint that GluN2A and GluN2B CTD interactions with the MAGUKs play distinct roles in stabilizing receptors in the PSD (Prybylowski et al., 2005; Sans et al., 2000; Zhang and Diamond, 2009). For example, acute disruption of GluN2 CTD:MAGUK interaction differentially affects GluN2B- and GluN2A-NMDAR organization in the PSD. In the absence of MAGUK interactions, synaptic GluN2B-NMDAR clusters reorganized their clusters within the PSD without affecting signaling and total receptor numbers, while synaptic GluN2A-NMDAR levels greatly decreased (Kellermayer et al., 2018). These subunit-specific effects support the idea of differential control of the NMDAR by MAGUKs that depends on NMDAR subunit composition. Interestingly, the PDZ-containing protein GIPC (GAIP-interacting protein), interacts with both GluN2A and GluN2B CTDs solely in extrasynaptic spaces (Fig 3)(Yi et al., 2007).

Tak together, NMDAR synaptic/extrasynaptic localization is actively coordinated in a subunit-specific manner by MAGUKs and other proteins through binding to the GluN CTDs.

4. NMDA receptor trafficking is orchestrated by GluN CTDs.

The transport, anchoring, and recycling of NMDARs within neurons determines their surface levels and localization – key elements that govern their function. Indeed, disruption of normal NMDAR surface localization may be a contributing factor to NMDAR dysregulation in autism, Alzheimer’s Disease, Parkinson’s disease, and schizophrenia (Burnashev and Szepetowski, 2015; Mohn et al., 1999; Okabe et al., 1999; Standley et al., 2000). In contrast, increased GluN2B-NMDAR function in transgenic mice increases learning and memory (Cui et al., 2011; Sasaki et al., 2002; Tang et al., 2001, 1999) and improves age-related memory decline and impaired cognitive functions in mouse models of depression and neurodegenerative disorders (Hu et al., 2017; D. Wang et al., 2014). Given their broad involvement in cognitive and neuropathological processes, it is important to understand the mechanisms regulating NMDAR trafficking. Not surprisingly, these processes are closely managed by a large set of proteins that dynamically interact with the NMDAR subunit CTDs.

4.1. GluN1 CTDs control exit and retention in the endoplasmic reticulum (ER).

GluN subunits are co-translationally inserted across the ER membrane. Most unassembled GluN1 and GluN2–3 subunits are retained in the ER pending their assembly into hetero-tetramers (Fukaya et al., 2003; Pérez-Otaño et al., 2001; Schüler et al., 2008). This retention is mediated by a series of discrete ER retention signals that mediate interactions with resident ER chaperone proteins that ensure correct folding and proper subunit stoichiometry of heteromeric receptors (Ellgaard et al., 1999; Teasdale and Jackson, 1996). The proper assembly of subunits into hetero-oligomeric complexes masks the retention signals that keep unassembled subunits in the ER; and allows assembled receptors to traffic to the Golgi and ultimately to the cell surface. Hence, ER retention is a critical first step in heteromeric NMDAR assembly.

Analyses of the differential trafficking of GluN1 isoforms have revealed key features of the CTD that regulate NMDAR movement from the ER to the cell surface. For example, when expressed as single subunits in a heterologous cell line, GluN1–1a is retained in the ER, while the GluN1–2/−3/−4a splice variants reach the membrane surface even without a partner GluN2 subunit (Okabe et al., 1999; Scott et al., 2001). Subsequent analyses of the localization of different GluN1 isoforms and mutants thereof have revealed key features that implicate GluN1 early trafficking in the ER.

Key features of the GluN1 CTD both promote and impair NMDAR trafficking from the ER to the surface. GluN1 mutants in which the CTD is terminated at aa 851 exhibits robust surface staining, while a mutant truncated at aa 838 is retained in the ER. These data suggest that some determinant signal in the C0 cassette enhances GluN1 surface targeting, most likely by loss of an ER retention signal (Horak and Wenthold, 2009). Alternatively, aa 838–851 are part of a larger domain that must be properly folded for the subunit to exit the ER. In contrast, the C1 cassette (present in GluN1–1/3a; Fig 2A) contains two powerful ER retention signals: RRR (aa 915–917) and KKK (aa 898–900) motifs (Fig 2A). Mutational studies indicate that each signal is sufficient to independently serve as an ER retention signal (Horak and Wenthold, 2009; Standley et al., 2000). The C1 cassette also contains three serine residues just after the RRR (893–895aa) motif. It was thought that phosphorylation by PKC and PKA of the Ser⁸⁹⁰ and Ser⁸⁹⁶ can negate the C1 retentions in the chimeric proteins containing the GluN1–1a CTD (Scott et al., 2001). However, subsequent work revealed phosphorylation of these serines was not sufficient to overcome function of both C1 retention signals in the full length GluN1–1a (Horak et al., 2014; Horak and Wenthold, 2009). This difference between the impact of phosphorylation on retention in chimeric proteins versus full length GluN1–1a may be due to the presence of additional retention signals in the M3 of GluN1 (Horak et al., 2008).

Chimeric transmembrane proteins containing the GluN1–2 CTD (Fig 2A) in heterologous cells are strongly retained in the ER, as compared to a truncated GluN1-C0 construct (851-STOP), suggesting a possible retention signal in the C2 cassette (Standley et al., 2000). In addition, GluN1–1a (Fig 2A) traffics to the surface even when expressed alone, (Horak and Wenthold, 2009; Standley et al., 2000). This suggests that C2’ can overcome the retention signals in the C1, and thus promote exit from the ER. Thus, C2 and C2’ play opposing roles in receptor exit from the ER. The accumulated data indicate that GluN1 subunits, with varying CTDs arising from alternative splicing, play key roles in NMDAR trafficking.

4.2. Regulation of ER retention of other GluN subunits by their CTDs.

In contrast to GluN1, GluN2 and GluN3 subunit CTDs have less prominent roles in regulating NMDAR exit from the ER. In fact, retention signals have not been reported in the GluN2A, C, D or GluN3 CTDs.

GluN3 CTDs do not contain ER retention signals, but their presence can enhance trafficking of tri-heteromeric receptors from the ER. The region between aa 952 and 985 in GluN3B is required for assembled GluN1:GluN3B-NMDARs to exit the ER and reach the surface (Matsuda et al., 2003). Regulation of the trafficking of the newly discovered GluN1:GluN3A-NMDARs has not yet been extensively studied (Grand et al., 2018). Together, these studies indicate that regulation of NMDAR exit/retention from the ER is mainly regulated by the signals in GluN1 CTD but can also be affected by determinants in specific GluN2–3 subunits.

4.3. GluN CTD interactions with binding partners promotes forward trafficking of NMDARs from the ER.

In the canonical secretory pathway, assembled NMDARs are trafficked from the ER to the somatic trans-Golgi network (TGN), where they are sorted and packaged into transport vesicles to be delivered to the plasma membrane (Ramirez and Couve, 2011). This canonical delivery pathway employs kinesin (KIF) or dynein-mediated trafficking on microtubules in axons and dendrites for long-range delivery (Guillaud et al., 2003; Valenzuela et al., 2014), while trafficking on actin is responsible for short-distance transport inside the spines. Part of the NMDARs are diverted from the somatic ER into a specialized dendritic ER sub-compartment (dER) that bypasses the TGN, merging instead with dendritic Golgi (dGolgi) outposts on their path to the synapse (Grieve and Rabouille, 2011; Jeyifous et al., 2009; Jo et al., 1999). This alternative path (non-canonical NMDAR transport) can be used to traffic receptors from the dER all the way to the dGolgi outposts. Accumulation of the receptors in the dER allows for faster local processing in the dGolgi and synaptic delivery. Recently, another non-canonical transport path was demonstrated for GluN2B-NMDARs where they would reach dER bypassing TGN by means of vesicles transported by protein motors along microtubules (Jeyifous et al., 2009). This alternative path seems to carry NMDAR-containing vesicles at a greater speed to more efficiently promote insertion of NMDARs at the PSD (Horton and Ehlers, 2003; Jeyifous et al., 2009). Assembled NMDARs do not exit the ER as discrete entities, but rather are assembled into large macromolecular transport packets with MAGUKs and are transported as a single unit (Georges et al., 2008; Kapitein and Hoogenraad, 2011; Konietzny et al., 2017).

4.4. GluN1 CTD variants promote forward trafficking of NMDARs via protein-protein interactions that are regulated by activity.

GluN1–3/4a receptor subunits have increased mRNA splicing early in development and under conditions of prolonged NMDAR block by APV (2R)-amino-5-phosphonovaleric acid, or (2R)-amino-5-phosphonopentanoate) (Hong et al., 2015). GluN1–3/4a CTDs variants contain the PDZ-binding motif and VV in the C2’, which interact with MAGUKs and COPII, respectively. COPII is a coatomer that promotes formation of ER vesicles for transport to the TGN, and is thus found to promote GluN1–3a-NMDAR export from the ER in the conventional trafficking pathway (Grieve and Rabouille, 2011; Mu et al., 2003). Also, GluN1–3a interaction with SAP-97 expedites receptor exit from the ER.

4.5. GluN2 subunit CTD interactions regulate canonical NMDAR trafficking.

The GluN2 subunit CTDs play key regulatory roles in the transport, sorting, and synaptic targeting of native GluN1-GluN2 di-heteromeric NMDARs. The EHL motif, proximal to the M4 domain, is preserved in all GluN2 CTDs and required for the exit of the assembled receptor from the ER (Fig 2B, 3) (Hawkins et al., 2004; Yang et al., 2007). Additionally, GluN2 subunits associate with MAGUKs in the ER via their PDZ motif (Standley et al., 2012). This MAGUK:GluN2 association is implicated in forward trafficking of NMDARs. SAP-102 binds both GluN2A and GluN2B CTDs and these binding interactions serve as a nexus to recruit proteins of the exocyst complex. These proteins include mPins and Sec8/Sec6 proteins, also called L-G-A repeat-enriched proteins (LGNs)(Sans et al., 2003, 2000), which mediates tethering of NMDAR-containing and other vesicle cargo to the specific part of the membrane prior to vesicle fusion (Fig 3)(Sans et al., 2003, 2000). Disruption of the SAP-102:Sec8 interaction interface decreases NMDAR cell surface delivery in neurons, demonstrating its importance in promoting forward trafficking of the receptor (Sans et al., 2005, 2003).

4.6. GluN2 CTD coordination of non-canonical NMDAR transport.

GluN2A-NMDARs are mainly thought to be transported via a non-canonical path involving dER and dGolgi. Indeed, BiP, an ER chaperone, selectively interacts with GluN2A and promotes assembly of GluN2A- NMDAR within dER and its subsequent transport to dGolgi, leading to increased GluN2A-NMDAR surface levels after increase in synaptic activity (Fig 3, path marked by yellow arrows) (Zhang et al., 2015). Interactions of GluN2B-NMDARs with SAP-97:CASK (mLin2) in the ER can promote direct trafficking to the dGolgi that bypasses the TGN (Fig 3, green arrows)(Jeyifous et al., 2009; Setou et al., 2000). Notably GluN2B:SAP-97:CASK vesicles travel along this non-conventional pathway with higher velocities (0.7μm/s) than vesicle trafficking via the regular non-conventional pathway by dER (0.2–0.3μm/s). These speeds are most similar to those of GluN2B-cargoes transported by KIF17 microtubular motors (Guillaud et al., 2003; Jeyifous et al., 2009; Setou et al., 2000). Indeed, knockdown of SAP and CASK decrease GluN2B-NMDAR trafficking speed and virtually eliminate NMDAR levels in the dGolgi (Guillaud et al., 2003; Jeyifous et al., 2009; Setou et al., 2000). Together, these data suggest that GluN2B-NMDAR vesicles are sorted by SAP-97:CASK into a distinct non-canonical trafficking path in which KIF17 transports the vesicles directly from the ER to the dGolgi. In addition, KIF5b and KIF3b were recently demonstrated to localize with GluN2A and GluN2B-NMDARs in the soma and in dendrites in vivo. However, it remains unclear whether KIF5b or KIF3b preferentially participates in canonical or non-canonical transport of NMDARs (Alsabban et al., 2020; Lin et al., 2019).

4.7. GluN CTDs palmitoylation is necessary for forward trafficking and functional synaptic incorporation of NMDARs.

Two clusters of cysteine residues in both the GluN2A and GluN2B CTDs (Cluster I: GluN2A-C⁸⁴⁸-C⁸⁷⁰; GluN2B-C⁸⁴⁹-C⁸⁷¹; Cluster II: GluN2A-C¹²¹⁴-C¹²³⁹; GluN2B-C¹²¹⁵-C¹²⁴⁵; Fig 2B) undergo palmitoylation, which is essential for NMDAR forward trafficking. (Hayashi et al., 2009; Mattison et al., 2012). The addition of palmitoyl moieties helps to anchor the NMDARs to the inner part of the cell membrane, and regulate surface levels and internalization (Salaun et al., 2010). Mutagenesis of the Cluster I cysteine to serine residues in GluN2B reduced NMDAR surface expression by increasing constitutive receptor internalization (Hayashi et al., 2009; Mattison et al., 2012). Cluster I palmitoylation in GluN2A and GluN2B subunits is required for phosphorylation of neighboring tyrosine (Y⁸⁴²) phosphorylation by Src. The Y⁸⁴² is a central determinant of the AP-2 binding/internalization motif (Y⁸⁴²WKL-GluN2A; Y⁸⁴³WQE-GluN2B; Fig 2B), and phosphorylation of this site decreases receptor internalization. Palmitoylation of the cluster II in the GluN2A and GluN2B CTDs causes NMDAR accumulation in the Golgi apparatus, suggesting that surveillance of this modification serves as a quality control step in receptor maturation (Hayashi et al., 2009; Kang et al., 2008). Mutations of the cluster II to non-palmitoylatable residues increases extra synaptic NMDAR levels but did not affect the synaptic pool. Hence, palmitoylation may slow NMDAR exit from Golgi apparatus, leading to net increases in surface levels to normalize synaptic levels of the receptors.

5. Covalent modifications and protein interactions regulate experience-dependent NMDA receptor signaling.

NMDARs are highly mobile, and their relative levels within synapses change with development (Lussier et al., 2015; McKay et al., 2018; Sanz-Clemente et al., 2013c, 2010), sensory experience (Matta et al., 2011; Philpot et al., 2003; Smith et al., 2009), and synaptic plasticity (Barth and Malenka, 2001; Bellone and Nicoll, 2007; Dupuis et al., 2014). Earlier studies, using the irreversible NMDAR blocker MK-801, showed that once blocked, synaptic NMDAR-mediated currents can recover in under 7 minutes (Tovar and Westbrook, 2002). These observations suggested that NMDARs undergo rapid synaptic replenishment. Insertion and internalization of NMDARs both occur predominantly outside the PSD (Kennedy et al., 2010). Imaging of single NMDAR molecules and fluorescence recovery after photobleaching indicate that NMDARs enter the synapse via lateral diffusion from extrasynaptic sites (Dupuis et al., 2014; Groc et al., 2007; Kellermayer et al., 2018). This process may be regulated via covalent modifications of NMDAR subunit CTDs and their interactions with intracellular proteins.

5.1. GluN CTD modifications dramatically regulate NMDA receptor endocytosis and localization in activity-dependent manner.

The Y⁸³⁷ in the GluN1 CTD and Y^840−843 in GluN2(A-D) subunit CTDs can undergo tyrosine phosphorylation. Each tyrosine is embedded within a YXXφ motif, which mediates binding of the clathrin adaptor AP-2 to trigger endocytosis (Nakatsu and Ohno, 2003; Scott et al., 2004; Vissel et al., 2001). Phosphorylation of these tyrosines by SFKs disrupts AP-2 binding to prevent endocytosis (Scott et al., 2004; Vissel et al., 2001). Repetitive application of NMDAR agonists reduces NMDAR currents by triggering receptor internalization (Nakatsu and Ohno, 2003; Scott et al., 2004; Vissel et al., 2001). This activity-dependent NMDAR internalization requires dephosphorylation of at least GuN2 subunit tyrosines (Vissel et al., 2001). Further control over the balance of NMDAR endo- and exocytosis is mediated by additional residues in the GluN2A and GluN2B CTDs. For example, phosphorylation of S¹⁰⁴⁸ on the GluN2A CTD by the Dyrk1a kinase hinders internalization and increases GluN1:GluN2A-NMDAR surface levels, both in heterologous cells and primary cortical neurons (Grau et al., 2014). In GluN2B, Y¹⁴⁷² (Y¹⁴⁷⁴ in human) lies within the YEKL motif, a second endocytic motif unique to the GluN2B CTD. Like the other tyrosine-based motifs cited above, phosphorylation of Y¹⁴⁷² disrupts AP-2 binding (Prybylowski et al., 2005). Notably, GluN2B Y¹⁴⁷² phosphorylation is both activity-dependent and promoted by receptor interactions with the PSD-MAGUKs, as they directly bind SFKs, keeping them close for NMDAR interactions (Lavezzari et al., 2003; Prybylowski et al., 2005; Zhang et al., 2008).

GluN2A and GluN2B are also subject to regulation by the Cdk5 kinase via phosphorylation of their CTDs. Cdk5 phosphorylates S¹²³² on GluN2A CTD in vitro and in vivo, which correlates with increased GluN2A-NMDAR activity (Wang et al., 2003). These phenomena contribute to our understanding of the mechanism underlying the death of hippocampal neurons via increased Cdk5 activity after ischemic stroke, as it was found that Cdk5 inhibition provides neuroprotection from NMDAR-mediated neuron death in this model (Wang et al., 2003). By contrast, Cdk5 phosphorylation of S¹¹¹⁶ on GluN2B CTD prevents surface expression (Plattner et al., 2014). In this regard, Cdk5 phosphorylation of GluN2B decreases Src-mediated phosphorylation of GluN2B at Y¹⁴⁷², increasing AP-2-mediated endocytosis (Zhang et al., 2008). In addition, GluN2B, Cdk5, and the calcium-dependent protease calpain form a complex in the PSD, and Cdk5 promotes cleavage and clearance of GluN2B via calpain activation (Hawasli et al., 2007). Therefore, Cdk5 can positively and negatively regulate GluN2A- and GluN2B-NMDARs, respectively.

S¹³⁰³ on the GluN2B CTD serves as the convergence point of different modulators as it can be phosphorylated by multiple kinases (PKC, death associated protein kinase (DAPK1) and CAMKII) depending on the activity context. Its phosphorylation disrupts CaMKII binding and decreases GluN2B-NMDAR internalization, thus increasing surface levels in neurons (Strack et al., 2000).

S¹⁴⁸⁰ located within GluN2B CTD PDZ binding motif is phosphorylated by Casein kinase 2 (CK2) to control GluN2B localization within the membrane (Sanz-Clemente et al., 2010). Phosphorylation of the S¹⁴⁸⁰ disrupts binding of the CTD with MAGUK:SFK complex, leading to GluN2B-NMDAR lateral diffusion into extrasynaptic regions. This also reduces SFK-mediated Y¹⁴⁷² phosphorylation, leading to receptor endocytosis (Hee et al., 2004; Lavezzari et al., 2003; Sanz-Clemente et al., 2013c, 2010). Notably, association of the GluN2B CTD with protein phosphatase 1 (PP1) dephosphorylates S¹⁴⁸⁰, allowing GluN2B-NMDAR to move back into the PSD (Chiu et al., 2019). Note that CK2 and PP1 do not directly bind to the CTD but are recruited by other proteins, such as CaMKII and spinofillin, that are bound in an activity-dependent manner (Baucum et al., 2013; Chiu et al., 2019; Sanz-Clemente et al., 2013a). These data suggest that associations of CK2 and the PP1 antagonize one another to control the synaptic vs extra-synaptic levels of GluN2B-NMDARs.

Similar to its GluN2 subunit cousins, the GluN3A CTD contains a YWL motif (Y⁹⁷¹ in mouse, Y⁹⁸⁰ in human) that mediates AP-2 binding and internalization. Y⁹⁷¹ is subject to phosphorylation, but unlike for GluN2B, it increases AP-2 interaction and GluN3A-NMDAR endocytosis (Chowdhury et al., 2013). The GluN3A CTD binds the endocytic adaptor PACSIN1/syndapin1 to mediate activity-dependent endocytosis (Pérez-Otaño et al., 2006). Together, these data nicely exhibit how GluN subunits are subject to exquisite activity-dependent control of their endocytosis and degradation mediated by their CTDs.

5.2. GluN CTDs determine the NMDAR fate after endocytosis.

Following endocytosis, GluN2A-NMDARs are preferentially targeted for degradation, whereas GluN2B-NMDARs, including tri-heteromers, are often recycled back to the membrane. GluN1 (Y⁸³⁷WK) and GluN2A (Y⁸⁴²WK) CTD motifs, proximal to the M4, direct them to late endosomes for degradation (Scott et al., 2004; Vissel et al., 2001). The S¹⁴⁵⁹ phosphorylation on GluN2A CTD tags GluN2A-NMDARs for recycling back to the synapse by SNX-27 retromer binding (Ling et al., 2021; Mota Vieira et al., 2020). This serine is phosphorylated in early endosomes by CaMKIIα after an increase in synaptic activity. Phosphorylation of the S¹⁴⁵⁹ decreases PSD-95 binding, suggesting that this residue needs to be dephosphorylated to reach the PSD (Mota Vieira et al., 2020). GluN1-C0 has one additional endocytic motif VWRK⁸⁶¹, in which ubiquitination of the lysine is needed to activate NMDR endocytosis (Scott et al., 2004).

Endocytosis of GluN2B by means of the second YEKL motif requires Y¹⁴⁷² dephosphorylation, and later is mostly recycled back to the surface (Jurd et al., 2008; Prybylowski et al., 2005; Roche et al., 2001; Scott et al., 2004). However, GluN2B-NMDAR di-heteromer endocytosis mediated by the first AP-2 binding motif, proximal to the M4 domain in GluN2B, or via the GluN1-C0 endocytic signals (at Y⁸⁴² see 5.1) allows for Y¹⁴⁷² on GluN2B CTD to remain phosphorylated after endocytosis. Subsequently this phosphorylation- Y¹⁴⁷² can promote interaction with the E3 ligase, Mind bomb 2 (Mib2) at aa 1170–1482, and ubiquitinate the CTD, thus tagging it for proteasomal degradation in late lysosomes (Jurd et al., 2008; Prybylowski et al., 2005). Notably, GluN2B:GluN2A tri-heteromeric NMDARs are preferentially recycled, which is coordinated by GluN2B CTD (Tang et al., 2010). The GluN2B CTD can also interact with the ubiquitin E3 ligase Cbl-b to regulate GluN2B-NMDAR levels in neurons of the spinal cord (Zhang et al., 2020). The GluN2D CTD is ubiquitinated by Nedd4–1, leading to receptor degradation (Gautam et al., 2013). Together, these findings illustrate how covalent modifications to GluN CTDs impact protein interactions central to regulating the surface and synaptic localization of NMDARs in a subunit- and activity-specific manner.

5.3. The GluN2A and GluN2B CTDs impact NMDAR synaptic localization during development.

During early postnatal development in the synapses of many brain regions, NMDARs switch their subunit composition from primarily GluN2B- to predominantly containing GluN2A-NMDAR (Barth and Malenka, 2001; Dumas, 2005). The replacement of GluN2B with GluN2A subunits is not complete, as GluN2B is still found in many regions of the adult brain, including subcortical regions as well as the hippocampus, often as GluN2A:GluN2B tri-heteromers (Barth and Malenka, 2001; Tovar and Westbrook, 1999; Xiao et al., 2016). These changes in synaptic NMDAR subunit composition are driven partly by differential transcription of the GluN2 subunits, but also by developmental changes in the covalent modifications to GluN2A and GluN2B. Within first 2 weeks of postnatal period in mice and in response to enhanced sensory experience, mRNA levels for GluN2A increase in the developing visual cortex, accompanied by increases in synaptic GluN2A-NMDARs (Matta et al., 2011; McKay et al., 2018; Philpot et al., 2003; Smith et al., 2009). This increase in synaptic GluN2A-NMDARs is not observed in dark-reared animals, but is reversible once the animals are exposed to light (Barria and Malinow, 2002; Matta et al., 2011; Philpot et al., 2003; Smith et al., 2009).

The impact of activity on NMDAR levels at the synapse is under the control of activity-dependent signaling pathways that act on NMDAR trafficking via interactions with the GluN2 CTDs. mGluR activated phospholipase C (PLC) leads to accumulation of the Rab3A, which together with PSD-95, increases GluN2A-NMDARs localization in the synapse (Franchini et al., 2019; Stanic et al., 2015). Activity-dependent binding of CAMKIIα to GluN2B CTD region 1290–1310 aa recruits CK2 to phosphorylate S¹⁴⁸⁰, resulting in removal of GluN2B-NMDARs from the PSD, as discussed above.

Other developmental cues may also regulate NMDAR turnover, including maturation of the hyaluronic acid-based extracellular matrix, which correlates with decreased Y¹⁴⁷² phosphorylation, a trigger for endocytosis (Schweitzer et al., 2017). As cited above, GluN2B CTD ubiquitination mediates its removal from the synapse in the spinal cord during neuronal maturation (Zhang et al., 2020). Thus, developmental GluN2A- to -GluN2B exchange is experience-dependent and mediated both by changes in subunit expression levels and tightly coordinated GluN2B-NMDAR removal via CTD-mediated interactions.

6. Role of GluN2A and GluN2B CTDs in synaptic plasticity and adaptation

Long-term synaptic plasticity is a common term used for a long-lasting experience-dependent change in the efficacy of synaptic transmission. NMDARs have a high permeability to calcium, which is essential for the induction of either long-term potentiation (LTP, high calcium influx, high frequency stimulation) or long-term depression (LTD, lower calcium, low frequency stimulation), two prominent forms of neuronal plasticity (Barrionuevo et al., 1980; Jacobs et al., 2015; Migaud et al., 1998; Morishita et al., 2007). Calcium influx is critical to trigger intracellular signaling that ultimately mediates recruitment or removal of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid glutamate receptors (AMPAR) to/from the PSD. Changes in the quantity or subcellular location and specific CTD interactions of NMDARs can alter the calcium dynamics in the spine, affecting the future induction of plasticity (Lüscher et al., 1999). Here we will focus on specific roles for GluN subunit CTDs in mediating synaptic plasticity.

6.1. LTP in immature neurons and role of GluN2 CTDs in its regulation.

LTP induction in immature neurons requires activation of both NMDARs and group 1 metabotropic glutamate receptors (mGluRs). Once activated, mGluRs trigger downstream signaling via PKC, PKA and PLC leading to insertion of AMPAR and increase in the synaptic activity (Esteban et al., 2003; Ling et al., 2002; Luchkina et al., 2013; Matta et al., 2011). Shortly after LTP is activated in immature neurons, GluN2A- and GluN2B-NMDARs undergo fast redistribution, which is implicated in the GluN2A to GluN2B developmental exchange via mechanisms discussed above (Bellone and Nicoll, 2007; Dupuis et al., 2014; Kellermayer et al., 2018).

To study functional implications of this subunit specific NMDAR reorganization acutely during LTP in immature synapses, Kellemayer et al. used subunit specific GluN2A and GluN2B PDZ binding ligands which disrupt GluN2-PDZ:MAGUK binding. Notably these specific ligands increased the ratio of GluN2A-NMDARs to GluN2B-NMDARs, without changing overall NMDA amplitude currents of the synapse (Kellermayer et al., 2018). This increased GluN2A/GluN2B ratio strongly affected induction of LTP, as increase of the ratio increased the LTP strength, while a decrease in ratio decreased LTP strength (Gardoni et al., 2009; Kellermayer et al., 2018). This suggests that GluN2A- to GluN2B-NMDAR synaptic ratio can fine tune LTP, and this ratio is controlled by their PDZ interactions. These data are also in agreement with other research demonstrating that the potential to induce LTP decreases with age in brain areas linked to decrease in GluN2B-NMDAR levels (Tang et al., 1999; Teyler et al., 1989).

6.2. Roles for the GluN2 CTDs in LTP and LTD in mature neurons.

In mature neurons, GluN2A-and GluN2B-NMDAR numbers and organization are stable during LTP induction. LTP in mature neurons is thought to be facilitated by calcium influx and CaMKIIα and CaMKIIβ (Bayer et al., 2001). Calcium influx through NMDAR activates CaMKIIα, which also binds the GluN1 and GluN2B CTDs. Notably, GluN2B has two CaMKIIα binding sites, one (aa 1290–1310) that is calmodulin-dependent, and a second (aa 839–1120), that requires CaMKIIα auto-phosphorylation at T²⁸⁶ to stabilize interaction with GluN2B (Bayer et al., 2001). Genetic manipulations that disrupt GluN2B CTD:CaMKIIα binding in vivo by point mutations on the GluN2B CTD severely impairs LTP by preventing synaptic strengthening (Barria and Malinow, 2005; Incontro et al., 2018; J. Q. Wang et al., 2014). This impairment is reversible and can be rescued by expression of a chimeric protein containing GluN2A with a GluN2B CTD, or by restoring GluN2B:CAMKIIα interaction directly by re-expressing intact GluN2B (Barria and Malinow, 2005; Incontro et al., 2018). In contrast, the converse chimeric GluN2B subunit with a GluN2A CTD chimera does not support LTP induction (Borovac et al., 2018). These data point to the GluN2B CTD as the requirement for LTP in mature neurons. While its CTD is not essential for LTP, GluN2A is still critical for LTP induction as it mediates calcium influx to initiate CaMKII accumulation in the PSD and trigger signaling events that lead to rapid recruitment of AMPARs into the synapse (Borovac et al., 2018; Franchini et al., 2019).

Because NMDAR-dependent calcium influx is required for both LTP and LTD, the synapse must have a mechanism for deciding whether to potentiate or depress its activity. Low frequency activation of the synaptic NMDARs leads to low calcium influx, triggering LTD. Interestingly, CaMKIIα is also required for LTD induction. In response to low calcium influx, CaMKIIα auto-phosphorylates not only T²⁸⁶ but also Y³⁰⁵/T³⁰⁶, switching it to an LTD promotion state (Bayer et al., 2001; Coultrap et al., 2014; Pi et al., 2010; Strack and Colbran, 1998). In this state, CaMKIIα reduces synaptic strength by phosphorylating AMPAR subunits, marking them for endocytosis (Coultrap et al., 2014). In addition to CAMKIIα, LTD requires activation of the phosphatases PP2B and PP1 (Chiu et al., 2019). PP2B-dependent de-phosphorylation of DAPK1 activates its kinase activity to phosphorylate the GluN2B CTD at S¹³⁰³ (Goodell et al., 2017). This phosphorylation prevents GluN2B CTD:CaMKIIα binding and accumulation in the PSD, and strengthens its interaction with DAPK1 (Goodell et al., 2017). In fact, inhibition of DAPK1 activity completely blocks LTD in acute hippocampal slices (Goodell et al., 2017). While DAPK1 interacts with GluN2B CTD during basal signaling, it is replaced by activated CaMKIIα in response to calcium influx following strong LTP inducing stimuli. Hence, the balance of LTD versus LTP depends in part on the activity level controlling CaMKIIα and the relative interactions of the GluN2B CTD with CAMKIIα versus DAPK1.

7. Allosteric channel activity modulation and metabotropic signaling is coordinated by GluN CTDs.

The GluN2 subunit CTDs represent one third of their total size and, as discussed above, is subject to significant covalent modification and binding interactions. The GluN CTD is an intrinsically disordered domain with partial secondary structure (Choi et al., 2011; Romero et al., 2001). It is not clear how covalent modifications of the CTD could affect channel gating properties, or how agonist binding could signal through the CTD. However, increasing evidence suggests that CTD can be perturbed by agonist binding to the LBD and that the CTD can impact channel gating. We discuss the evidence for these regulatory interactions and possible underlying mechanisms here.

7.1. The GluN2 CTDs disordered state perturbations can coordinate allosteric modulation of the NMDARs.

Deletion of the GluN2A and GluN2B CTDs, depletion of PKA and PKC activities, or disruption of the GluN2B:PSD-95 interaction all result in changes of function of NMDARs as an ion channel (Maki et al., 2012; Sapkota et al., 2019; Sornarajah et al., 2008). Some studies suggest that the membrane proximal 100 aa of GluN2 CTDs regulate channel gating properties of GluN2A and GluN2B-NMDARs (Maki et al., 2013, 2012). This phenomenon is proposed to be transduced by direct connection of the CTD to M4, or interactions of the CTD with intracellular segments between M1–2 and M2–3 (Maki et al., 2012). Many phosphorylation sites on the CTD regulate NMDA channel gating but how this occurs is unclear (Aman et al., 2014; Bah et al., 2015; Fang et al., 2015; Jones and Leonard, 2005; Taniguchi et al., 2009; Yu et al., 2016). Phosphorylation can induce folding in some disordered proteins, and it may induce folding of the GluN CTD into conformations that regulate channel properties of the NMDAR (Bah et al., 2015). Single molecule studies strongly suggest that Src-induced GluN2 CTD phosphorylation on Y¹³³⁶ and Y¹⁴⁷² promotes CTD transition into an extended state (Choi et al., 2011). In contrast, replacement of multiple proline residues in the GluN2B CTD, which are often enriched in disordered proteins, turn it to a more condensed state (Choi et al., 2013). Curiously, both proline depletion and Src phosphorylation decrease zinc dependent GluN2B-NMDAR current inhibition without affecting other receptor gating functions (Choi et al., 2013, 2011; Traynelis et al., 1998; Zheng et al., 1998). One possible explanation is that the GluN2B CTD has non-consecutive sequences that exert allosteric effects on the ATD to impact receptor gating properties. In summary, these discoveries suggest that conformational dynamics within the intrinsically disordered GluN CTD can affect the allosteric regulation of NMDAR gating. However, more studies are required to identify specific features in the CTD that are critical to mediate these effects.

7.2. The GluN1 CTD interaction with the actin cytoskeleton controls NMDAR calcium-dependent desensitization.

It is established that NMDAR currents are suppressed by increases in intracellular calcium, a function called calcium-dependent desensitization or inactivation (CDI) (Rycroft and Gibb, 2004; Yue et al., 1990; Zhang et al., 1998). This NMDAR CDI appears to be a critical mechanism to limit cellular calcium overload, preventing NMDAR-mediated neurotoxic effects. Current models suggest that CDI is regulated by calmodulin (CaM) binding to the GluN1 CTDs. The GluN1 CTD has two CaM binding sites (in C0 and C1; Fig 2C). Deletion or mutation of the CaM sites abolishes CDI (Ehlers et al., 1996), thus they are required for CDI. Calcium activated CaM binding to the C0 site competes with the cytoskeletal protein α–actinin for binding GluN1 and hence, CDI correlates with release of the NMDAR from cytoskeletal tethering (Krupp et al., 1999; Merrill et al., 2007). However, CaM binds to C1 with higher affinity than to C0, which may contribute to the stronger CDI in C1-containing GluN1 variants (Ehlers et al., 1996; Hisatsune et al., 1997; Iacobucci and Popescu, 2020). In addition, phosphorylation of the C1 region by PKC inhibits CaM binding, suggesting an additional point of CDI modulation (Hisatsune et al., 1997). NMDAR release from cytoskeletal tethering can affect the receptor in two ways, by either releasing its localization at the synapse or by directly gating receptor channel opening (Allison et al., 1998; Paoletti and Ascher, 1994; Rosenmund and Westbrook, 1993; Shaw and Koleske, 2021). Although CDI is mediated by calmodulin binding to the constitutive GluN1 subunit, NMDARs with different GluN2 subunits exhibit different propensities to CDI, ranging from high in GluN2A-NMDARs to much less in GluN2B-NMDARs and almost none in GluN2C/D-NMDARs (Iacobucci and Popescu, 2020; Krupp et al., 1999; Villarroel et al., 1998). Interestingly, Iacobucci and Popescu recently demonstrated that GluN2B-NMDARs undergo the same level of CDI as GluN2A-NMDARs when robust calcium is supplied intracellularly and not via NMDARs (Iacobucci and Popescu, 2020). Hence, different CDI propensity is explained by the lower open channel probability of GluN2B-NMDARs than for GluN2A-NMDARs, which does not allow for efficient calcium influx to activate CaM dependent CDI (Iacobucci and Popescu, 2020).

7.3. GluN CTD positioning change and protein interaction transduce metabotropic NMDAR signaling.

Several studies indicate that NMDAR ligand binding can induce signaling events, mediated by the CTD, that occur without channel activation and thus without ion influx (Aow et al., 2015; Nabavi et al., 2013; Vissel et al., 2001). These events are critical for activity-dependent NMDAR endocytosis and NMDAR-dependent LTD.

Förster (or Fluorescence) Resonance Energy Transfer (FRET) imaging of NMDARs tagged with fluorescent proteins on their CTDs show that ligand binding causes a change in intracellular positioning of the GluN1 subunit CTDs (Aow et al., 2015; Dore et al., 2015). This movement is blocked by the competitive NMDAR blocker APV, but not by the inhibitors 7CK (7-chlorokynurenate) or MK-801, which block ion influx by binding the LBD or plugging the ion pore, respectively (Fig 1A)(Dore et al., 2015). This demonstrates that GluN1 CTD conformation can be changed by agonist binding to its LBD. This CTD movement may be important to initiate signaling events, explaining how the NMDARs can signal even when ion flux is blocked (Aow et al., 2015; Nabavi et al., 2013). Such a mechanism might explain how LTD can occur solely through NMDAR metabotropic signaling (Aow et al., 2015; Carter and Jahr, 2016; Nabavi et al., 2013; Stein et al., 2015). Interestingly, NMDAR-mediated metabotropic LTD may be mediated by interactions between PP1 and GluN2B CTD, following glutamate-induced CTD conformational changes (Aow et al., 2015). However, more studies are required to clarify exact initiation of downstream cascades evoking NMDA metabotropic LTD and subsequent spine shrinkage.

In summary, the GluN CTDs contribute to important physiological functions as part of the NMDA receptor. These functions include quality control of the subunit/receptor assembly, trafficking to the membrane and ER retention of unassembled subunits, NMDAR synaptic and extrasynaptic membrane localization, activity-dependent regulation of surface number via control of exo-/endocytosis, allosteric regulation of receptor activity, and metabotropic signaling, likely by acting as a platform for assembly of signaling modules (Fig 4 A).

Fig. 4.

GluN CTD physiological functions and implications to pathophysiological conditions in neurological disorders. A Summary diagram of the GluN CTDs implication to NMDAR physiological function. B Diagram of GRIN family gene variants localizing in the CTDs. C Table summarizing numbers of the disease associated rare variants found in the GluN CTD vs total number of variants in other GluN domains. D Specific GluN CTD variants implicated in neurodevelopmental and neuropsychiatric disorders colored by group. Groups divided as ASD, ID and MR; next group includes all types of epilepsies, EP and ID; next group includes SCZ only and the last group includes ANS and other neurologic disorders. Variants found in multiple disorders are counted in each group they were found.

8. NMDAR CTD genetic variants are associated with human brain disorders.

Rare variants of NMDARs are found in patients with neuropsychiatric and neurodevelopmental disorders. So far, at least 282 GRIN variants associated with pathophysiological outcomes have been identified by genome sequencing (Hu et al., 2016; Perszyk et al., 2020; XiangWei et al., 2018). Variants are most commonly found in the GluN2A and GluN2B subunits, and less frequently in GluN2C/D and GluN3A/B subunits (Hu et al., 2016; Lemke et al., 2016; Perszyk et al., 2020; XiangWei et al., 2018). Unsurprisingly, a large number of those variants localized to the highly conserved ABD and TMD regions of GluN subunits (Fig 1A,C) but many (about 21%) were found within the CTDs (Fig 4B, Table 1) (Hu et al., 2016; Perszyk et al., 2020; XiangWei et al., 2018). The GluN1 and GluN2A CTD variants are most commonly associated with epilepsy, but also with intellectual disability (ID) and autism spectrum disoders (ASD). In contrast, variants within the GluN2B-D and GluN3A/B subunit CTDs are associated with ID, ASD, and schizophrenia (SCZ) (see Table 1 and Fig 2 and 4 C,D for details) (Bramswig et al., 2015; Duguid and Smart, 2009; Hu et al., 2016; Lal et al., 2015; Lemke et al., 2016, 2013; Lesca et al., 2013; Liu et al., 2017; Pankratov et al., 2002; Tarabeux et al., 2011; von Stülpnagel et al., 2017). Notably, individuals carrying the same mutation can experience different disease outcomes, such as in the case of GluN2A-N1397 frame shift mutation, which is associated with EP, ID, and ASD (Bramswig et al., 2015). This finding is consistent with the significant overlap in genetic vulnerability to these disorders (Anttila et al., 2018). Functional analysis of these variants is lacking, and hence there is no comprehensive understanding of how disease-associated variants of the GluN CTDs impact overall NMDA receptor properties and function. As such, functional assessment of the GluN CTDs is needed.

Table 1.

GluN CTD variants implicated in the pathophysiology of human neuropsychiatric and neurodevelopmental disorders. Many variants implicate to multiple phenotypes, and these are only a snapshot of the current literature.

	Subunit	Protein change	Variant Type	Disease relevance	Citation
1.	GluN1	R844C	missense	ID, EPI	(Lemke et al., 2016)
2.	GluN2A	G858R	missense	ASD, ID	(Stessman et al., 2017)
3.	GluN2A	R865	frame shift	RE, ID	(Dimassi et al., 2014)
4.	GluN2A	I876T	missense	EPI	(Lal et al., 2015)
5.	GluN2A	I904F	missense	EPI, ID, FS	(Venkateswaran et al., 2014)
6.	GluN2A	S913	frame shift	EPI, ID	(von Stülpnagel et al., 2017)
7.	GluN2A	D933N	missense	EPI, LKS	(Addis et al., 2017; Lesca et al., 2013)
8.	GluN2A	T943	stop	CSWSS, MR, FS, CTS, RE	(Lemke et al., 2013)
9.	GluN2A	Q950	stop	ANS	(Retterer et al., 2016)
10.	GluN2A	V967L	missense	TLE, ABPE, BECTS	(Lal et al., 2015; Lemke et al., 2013)
11.	GluN2A	A968T	missense	SCZ	(Grozeva et al., 2015)
12.	GluN2A	N976S	missense	ABPE, CSWS	(Addis et al., 2017; Lemke et al., 2013)
13.	GluN2A	S1025T	missense	EPI, ID	(von Stülpnagel et al., 2017)
14.	GluN2A	T1064A	missense	SCZ	(Tarabeux et al., 2011)
15.	GluN2A	N1076K	missense	LKS	(Endele et al., 2010; Lemke et al., 2013)
16.	GluN2A	D1251N	missense	RE, AE	(Lesca et al., 2013)
17.	GluN2A	A1276G	missense	CSWSS, BECTS	(Lesca et al., 2013)
18.	GluN2A	R1281Q	missense	ID	(Grozeva et al., 2015)
19.	GluN2A	R1376S	missense	EPI, ID	(von Stülpnagel et al., 2017)
20.	GluN2A	I1379V	missense	AE	(Lal et al., 2015; Lemke et al., 2013)
21.	GluN2A	Y1387	stop	CSWSS, ASD, BCE	(Lesca et al., 2013)
22.	GluN2A	N1397	frame shift	EPI, ID, ASD	(Bramswig et al., 2015)
23.	GluN2B	R847	stop	ASD, ID	(Firth et al., 2009)(Platzer et al., 2017)
24.	GluN2B	I864	frame shift	ANS	(Retterer et al., 2016)
25.	GluN2B	L976	frame shift	ID	(Platzer et al., 2017)
26.	GluN2B	Y1004	stop	ASD, ID	(Platzer et al., 2017)
27.	GluN2B	F1011L	missense	ID	(Grozeva et al., 2015)
28.	GluN2B	Q1014R	missense	SCZ	(Tarabeux et al., 2011)
29.	GluN2B	G1026S	missense	ASD, SCZ	(Hu et al., 2016; Tarabeux et al., 2011)
30.	GluN2B	R1099	frame shift	ASD, ID	(Rauch et al., 2012)
31.	GluN2B	R1099H	missense	ASD	(Takasaki et al., 2016)
32.	GluN2B	R1111H	missense	ID	(Platzer et al., 2017)
33.	GluN2B	Y1155	stop	ASD, ID	(Stessman et al., 2017)
34.	GluN2B	T1228M	missense	ASD	(Pan et al., 2015)
35.	GluN2B	R1241Q	missense	ASD,ID	(Stessman et al., 2017)
36.	GluN2B	A1267S	missense	ID, SCZ	(Endele et al., 2010; Williams et al., 2012)
37.	GluN2B	T1273K	missense	ASD	(Pan et al., 2015)
38.	GluN2B	K1292R	missense	SCZ	(Takasaki et al., 2016)
39.	GluN2B	K1293R	missense	AD	(Andreoli et al., 2014)
40.	GluN2B	M1331I	missense	ID	(Endele et al., 2010)
41.	GluN2B	M1339V	missense	ASD	(Pan et al., 2015)
42.	GluN2B	N1352	frame shift	SCZ	(Tarabeux et al., 2011)
43.	GluN2B	S1415L	missense	ASD	(Liu et al., 2017; Tarabeux et al., 2011)
44.	GluN2B	L1424F	missense	SCZ	(Myers et al., 2011; Tarabeux et al., 2011)
45.	GluN2B	G1436A	missense	ID	(Chen et al., 2017)
46.	GluN2B	I446T	missense	ID	(Grozeva et al., 2015)
47.	GluN2B	S1452F	missense	SCZ	(Tarabeux et al., 2011)
48.	GluN2C	I863T	missense	ASD	(Tarabeux et al., 2011)
49.	GluN2C	S992F	missense	ID	(Hamdan et al., 2011)
50.	GluN2C	S995L	missense	ASD, SCZ	(Tarabeux et al., 2011)
51.	GluN2C	E1048	insert	ID	(Tarabeux et al., 2011)
52.	GluN2C	H1187	frame shift	ID	(Hamdan et al., 2011)
53.	GluN2C	A926T	missense	ASD	(Tarabeux et al., 2011)
54.	GluN2C	A982P	missense	SCZ	(Tarabeux et al., 2011)
55.	GluN2C	H1187-G1193	deletion	ID, SCZ	(Tarabeux et al., 2011)
56.	GluN2D	A901I	missense	SCZ	(Tarabeux et al., 2011)
57.	GluN2D	A926T	missense	ASD	(Tarabeux et al., 2011)
58.	GluN2D	A982P	missense	SCZ	(Tarabeux et al., 2011)
59.	GluN2D	G1317S	missense	ID	(Tarabeux et al., 2011)
60.	GluN3A	G898W	missense	ASD	(Tarabeux et al., 2011)
61.	GluN3A	R1111G	missense	ASD	(Tarabeux et al., 2011)
62.	GluN3B	E919D	missense	ASD	(Tarabeux et al., 2011)

Abbreviations used in the table: ASD, autism spectrum disorder or features of; ANS, abnormal nervous system; ABPE, atypical benign partial epilepsy; AE, absence epilepsy; AD, Alzheimer’s Disease.; BP, bipolar disorder; BCE, benign childhood epilepsy; CTS, centro-temporal spikes; CSWSS, continuous spike and slow wave during sleep; DD, developmental delay; EPI, epilepsy, focal or generalized seizures; HPT, hypotonia; ID, intellectual disability (includes non-verbal); IS, infantile spasms; LKS, Landau-Kleffner syndrome; MC, microcephaly; MD, movement disorder; SCZ, schizophrenia; WS, West Syndrome; TLE, temporal lobe epilepsy; BECTS, benign epilepsy with centro-temporal spikes; FS, febril seizures; RERolandic epilepsy; MR, mental retardation. Our table adopted data on GluN1, GluN2A/B CTD variants previously compiled by Warnet et.al., 2020and expanded with CTD variants found in GluN2C/D and GluN3A/B subunits (Warnet et al., 2020). All variants were confirmed using following data bases https://www.ncbi.nlm.nih.gov/clinvar/ and http://functionalvariants.emory.edu/database/index.html.

A few GluN2B CTD variants have been studied in detail and evidence suggests that these variants impact protein-protein interactions and surface expression. For example the GluN2B CTD mutations S1415L, L1424F, and S1452F found in patients with SCZ and ASD, all reduce interactions with MAGUKs in vitro (Liu et al., 2017; Myers et al., 2011; Tarabeux et al., 2011). Notably, two variants (S1415L, L1424F) lie within regions with no clear function, while S1252F lies just upstream of a P-X-X-P (1454–1458aa, Fig. 2C) thought to be a binding site for MAGUKs and other SH3 domain-containing proteins (Cousins et al., 2009). Electrophysiological data, in vitro, show the variant GluN1:GluN2B L1424F exhibit a modest decrease of glutamate potency, while GluN1:GluN2B S1452F receptor variants exhibit slightly increased glycine potency. The major pharmacological properties remained unchanged in NMDARs containing any of these three variants (Liu et al., 2017). However, only GluN2B-S1415L impacted NMDAR surface expression, synaptic currents, and spine density in a mouse model, suggesting that this mutation may contribute to the ASD phenotype identified in the patient (Liu et al., 2017; Myers et al., 2011; Tarabeux et al., 2011). Liu et. al. suggest that the upstream variants can still affect SH3 motif binding. These observations suggest that variants may impact distinct molecular interactions to cause pathological phenotypes and that more functional studies on disease associated GluN CTD variants are required to understand functional consequences to develop strategies to correct dysfunction. For example, pharmacological chaperones have been proposed, for variants that impact surface expression, to facilitate NMDA receptor biogenesis and improve trafficking to the cell surface. Building on a better understanding of CTD function, it may one day be possible to selectively target CTD variants for therapeutic intervention.

9. Conclusion and future directions

NMDARs have been extensively studied for over 30 years since their first identification. By far, the most studied are GluN2A- and GluN2B-NMDARs, due to their prevalence in the cortex and hippocampus, and high number of implications in neurological disorders. These studies suggest that GluN2A- and GluN2B-NMDARs are trafficked through different pathways, play specific roles in fine tuning NMDAR-dependent LTP and LTD, and differentially participate in signaling cascades. GluN CTDs is the most diverse subunit component, and these diversities play essential regulatory roles in the assembly, trafficking, localization, gating, and signaling functions of NMDARs as summarized on Fig. 4. Metabotropic signaling by NMDARs is transduced by extracellular domain interactions that somehow engage CTD downstream signaling pathways. These observations raise key unresolved questions about GluN CTD structure and how interactions with binding partners are impacted by covalent modifications. It is clear that the variation in NMDAR CTD structure and their many modes of regulation greatly impact the functional abilities of the NMDAR family.

Highlights:

• GluN subunit CTDs serve as a molecular hub for convergence of regulatory and signaling events.

• NMDA activity dependent regulation of synaptic activity/plasticity is conveyed through CTD interactions and covalent modifications.

• Despite being the most diverse subunit component, GluN CTDs are highly conserved between species, with variations contributed through mRNA splicing.

• GluN CTD modifications and interactions can affect NMDAR gating and metabotropic signaling despite being intrinsically disordered.

• NMDAR CTD genetic variants are associated with human brain disorders.