|Home | About | Journals | Submit | Contact Us | Français|
Zinc is vastly present in the mammalian brain and controls functions of various cell surface receptors to regulate neurotransmission. A distinctive characteristic of N-methyl-D-aspartate (NMDA) receptors containing a GluN2A subunit is that their ion channel activity is allosterically inhibited by a nano-molar concentration of zinc that binds to an extracellular domain called an amino terminal domain (ATD). Despite physiological importance, the molecular mechanism underlying the high-affinity zinc inhibition has been incomplete due to lack of a GluN2A ATD structure. Here we show the first crystal structures of the heterodimeric GluN1-GluN2A ATD, which provide the complete map of the high-affinity zinc binding site and reveals distinctive features from the ATD of the GluN1-GluN2B subtype. Perturbation of hydrogen bond networks at the hinge of the GluN2A bi-lobe structure affects both zinc inhibition and open probability supporting the general model where the bi-lobe motion in ATD regulates the channel activity in NMDA receptors.
Zinc has emerged as a fundamental modulator of brain functions in mammals in recent years. It is mostly present at axonal termini of various sets of glutamatergic neurons and is co-transmitted with L-glutamate in an activity dependent manner (Vergnano et al., 2014; Vogt et al., 2000; Westbrook and Mayer, 1987). The released zinc modulates activity of a number of ion channels, including N-methyl-D-aspartate (NMDA) receptors, which belong to the large family of ionotropic glutamate receptors (iGluRs) and are critically involved in neuronal plasticity and neurological disorders and diseases (Traynelis et al., 2010). NMDA receptors form obligatory heterotetrameric ion channels that are composed of GluN1 subunits (eight splice variants: 1–4a or b) and GluN2 subunits (A–D), and/or GluN3 subunits (A or B). The extracellular zinc binds and inhibits the NMDA receptors that contain GluN2A and GluN2B in a voltage-independent and allosteric manner at nano-molar and micro-molar potencies, respectively (Chen et al., 1997; Low et al., 2000; Paoletti et al., 1997; Rachline et al., 2005). The inhibition with the nano-molar potency or the high-affinity zinc inhibition in the GluN2A containing NMDA receptors plays critical roles in controlling pain sensation (Nozaki et al., 2011).
The NMDA receptor subunits are composed of four distinct domains, an amino-terminal domain (ATD), a ligand-binding domain (LBD), a transmembrane domain (TMD), and a carboxyl-terminal domain (CTD) (Figure 1). The ATDs have diverse primary sequences (35–57% identical among GluN2A–D (Figure S1A) and 22% identical among GluN1 and GluN3A–B) compared to LBD and TMD and define subtype-specific functions including the high-affinity zinc inhibition in the GluN1-GluN2A NMDA receptors as well as allosteric inhibitions by ifenprodil and micro-molar zinc in the GluN1-GluN2B NMDA receptors. The ATDs also play critical roles in controlling open probability as well as speeds of activation and deactivation in a subtype-specific manner (Gielen et al., 2009; Yuan et al., 2009a). Contrary to NMDA receptors, there is no apparent role of ATDs in directly controlling gating properties of the ion channels in non-NMDA receptors.
The structural studies of NMDA receptors have advanced moderately in recent years (Regan et al., 2015). The crystal structures of the isolated GluN2B ATD and the heterodimer of GluN1b and GluN2B ATDs (GluN1b-GluN2B ATD) showed that the micro-molar zinc binding site and the ifenprodil binding site reside at the cleft of the bi-lobed architecture of the GluN2B ATD (Karakas et al., 2009) and at the subunit interface of the GluN1b-GluN2B ATD (Karakas et al., 2011), respectively. The recent studies on the intact NMDA receptors show conformational movement of the GluN1-GluN2B NMDA receptors in various functional states stabilized by different combinations of ligands (Karakas and Furukawa, 2014; Lee et al., 2014; Tajima et al., 2016; Zhu et al., 2016). Contrary to the GluN1-GluN2B subtype, structural studies on the GluN1-GluN2A subtype except on LBD (Furukawa et al., 2005; Hackos et al., 2016; Hansen et al., 2013; Jespersen et al., 2014; Volgraf et al., 2016) have lagged due to technical difficulties associated with protein production. Consequently, the field has lacked any mean to structurally compare NMDA receptor subtypes and pinpoint subtype-specific elements in ATD. Here, we report the crystal structure of the heterodimers of GluN1b and GluN2A ATDs (GluN1b-GluN2A ATD) in the presence and absence of zinc. The structures unambiguously map the high-affinity zinc-binding site within the GluN2A ATD and show differences in conformations of the bi-lobed architecture and in the subunit orientation between the GluN1-GluN2A and the GluN1-GluN2B subtypes, thereby providing the novel structural insights into subtype-specific recognition of zinc and ifenprodil and ATD-mediated allosteric regulations.
Our previous studies on the isolated GluN1b-GluN2B ATD constructs showed binding of zinc and ifenprodil with similar affinity to the intact GluN1b-GluN2B NMDA receptors demonstrating that the isolated ATD proteins are functionally relevant (Karakas et al., 2009, 2011). Here we subjected the GluN1 ATD from Xenopus laevis in combination with the GluN2A ATD from rat to our structural study due to higher degree of protein homogeneity compared to other orthologue combinations. The intact NMDA receptors composed of Xenopus laevis GluN1-1a and rat GluN2A show conservation of high-affinity zinc inhibition as assessed by electrophysiology (Figure S2), thus indicating that the structural study of Xenopus laevis GluN1 and rat GluN2A ATDs is relevant for deciphering functions of mammalian GluN1-GluN2A NMDA receptors. In this study, we used the Xenopus GluN1b, a splice variant with twenty amino acids encoded by Exon 5, over the Xenopus GluN1a for its permissiveness to high-resolution crystallography.
Recombinant expression of the GluN2A ATD proteins that are properly folded and secreted into insect or mammalian cell culture media relied on co-expression of the GluN1 ATD. To obtain sufficient amount of the GluN1b-GluN2A ATD proteins amenable to crystallographic studies, we made a construct where the GluN1b ATD and the GluN2A ATD are tethered by a linker containing fifty seven hydrophilic amino acid residues (57-link; Figure 1B, also see ‘Experimental Procedures’) with no predicted secondary structure and three Asn-linked glycosylation sites with an intention to further assist secretion. Overall, this extensive construct manipulation yielded approximately 1 mg of purified proteins per liter of insect cell culture. 57-link and purification tags were removed by thrombin digestion (Figure 1D) and the GluN1b-GluN2A ATD proteins were complexed to Fab fragments of the immunoglobulin that recognizes the GluN1 ATD (Figure 1C and 1D, also see ‘Experimental Procedures’).
The crystals of the GluN1b-GluN2A ATD – Fab complex soaked against the crystallization solution containing 1 mM EDTA (EDTA) or 1 mM zinc for two hours (Zn1) or twenty four hours (Zn2) showed x-ray diffraction with the Bragg spacing between 2.7 – 3.3 Å (Table S1). The structure of the Zn1-GluN1b-GluN2A ATD was first solved by molecular replacement using coordinates of the GluN1b ATD and the GluN2B ATD (PDB ID: 3QEL), and anti-dinitrophenyl-spin-label Fab (PDB ID: 1BAF) as multi-search probes. The structures of the Zn2-GluN1b-GluN2A ATD and the EDTA-GluN1b-GluN2A ATD were solved by molecular replacement using the coordinate of the Zn1-GluN1b-GluN2A ATD as a search probe. Our crystallographic analysis shows clear electron density for 689, 697, and 692 residues out of 750 possible residues for Zn1, Zn2, and EDTA crystals, respectively. Electron density for most of Exon 5 was invisible except for GluN1-Lys211 observed in the three GluN1b-GluN2A ATD structures (Zn1, Zn2, and EDTA). Electron density for GluN1-Pro210 was only visible in the Zn1-GluN1b-GluN2A ATD and EDTA-GluN1b-GluN2A ATD structures. The quality of electron density is particularly high at and around the zinc binding sites and the GluN1b-GluN2A subunit interface (Figure S3), thus, permitting us to conduct reliable structure-based functional analysis as described in the later sections.
The crystal structures of the GluN1b-GluN2A ATD show an ATD heterodimer and a Fab in the asymmetric unit, which reinforces the view that the hetero-tetrameric subunit arrangement of the intact GluN1-GluN2A NMDA receptor is likely a dimer of two GluN1-GluN2A heterodimers as previously demonstrated for the GluN1-GluN2B NMDA receptors (Karakas and Furukawa, 2014; Lee et al., 2014). Both the GluN1b ATD and the GluN2A ATD have bi-lobed architectures composed of upper (R1) and lower (R2) domains (Figure 2). Zinc binds to the R1–R2 interface of the GluN2A ATD and stabilizes the bi-lobe in a ‘closed’ conformation. While there are minor differences locally, the main chain carbons (Cαs) of the three GluN1b-GluN2A ATD structures (Zn1, Zn2, and EDTA) are superimposable to one another with root-mean square deviation (RMSD) of 0.802Å, 0.539Å, and 0.962Å, between Zn1 and Zn2, Zn1 and EDTA, and Zn2 and EDTA, respectively. The structure of the GluN1b ATD in the GluN1b-GluN2A ATD is almost identical to that observed in the crystal structures of the GluN1b-GluN2B ATD bound to ifenprodil (ifenprodil-GluN1b-GluN2B ATD) (Karakas et al., 2011) with RMSD of 0.696Å over 347 superimposable Cαs (Figure 3A). In contrast, the GluN2A ATD cannot be superimposed onto the zinc-bound GluN2B ATD (Zn-GluN2B ATD) (Karakas et al., 2009) even though the overall bi-lobed architectural feature is conserved while their R1 and R2 regions can be superimposed individually with RMSD values of 1.54 and 1.12 Å, respectively. This stems from the major difference in the extent of the R1–R2 separation between the Zn1-GluN2A ATD and the Zn-GluN2B ATD where the bi-lobed architecture of the Zn1-GluN2A ATD is ~13° more open compared to that in the Zn-GluN2B ATD (Karakas et al., 2009) (Figure 3B and 3D). That is, the bi-lobe of the GluN2A ATD is not capable of closing as much as the GluN2B ATD. This is evident from the apparent steric clash observed when the R1 and R2 lobes from the GluN2A ATD are individually superimposed onto the GluN2B ATD to simulate more ‘closed’ GluN2A ATD bi-lobe (Figure S4).
The GluN1b and GluN2A ATDs form a heterodimer mostly at the R1 regions with buried surface area of 1,144 Å2. Our study reveals the major difference in the pattern of the inter-GluN1-GluN2 subunit orientation between the GluN1b-GluN2A ATD and the ifenprodil-GluN1b-GluN2B ATD, which is characterized by a ~12° rotation (Figure 3C and 3D). To validate physiological relevance of the heterodimeric assembly pattern observed in our crystal structures, we engineered cysteine mutants at the subunit interface and tested for disulfide bond formation in the context of the intact hetero-tetrameric GluN1b-GluN2A NMDA receptor ion channel. We reasoned that the engineered cysteines at the subunit interface should form spontaneous disulfide bonds if the pair of mutated cysteine resides are proximal to each other. Specifically, we incorporated two pairs of cysteine mutants at distinct locations which we named Site-I (GluN1b-Leu341Cys/GluN2A-Ser209Cys) and Site-II (GluN1b-Phe113Cys/GluN2A-Ala108Cys) (Figure 4A), expressed and purified the mutant GluN1b-GluN2A NMDA receptors, and conducted Western blot analysis under a non-reducing condition to detect band shifts representative of the disulfide bond formation. Site-I is located at the interface between GluN1b R1 and GluN2A R2 whereas Site-II is at the interface between GluN1b R1 and GluN2A R1. In those two sites, the disulfide bonds are formed only when the cysteine mutants of both GluN1b and GluN2A are co-expressed but not when a combination of the cysteine mutant of one subunit is co-expressed with the wild type (WT) of the other subunit indicating that disulfide bonds are formed specifically by the engineered cysteines (Figure 4B). Thus, the biochemical data above indicates that the GluN1b-GluN2A heterodimeric arrangement observed in the crystal structure is the physiological representation of the intact GluN1b-GluN2A NMDA receptor ion channel. It is noteworthy that inter-subunit disulfide bonds have been shown to form at the similar locations to Site-I and Site-II within the GluN1b-GluN2B ATD implying that the pattern of the subunit arrangement is roughly similar between GluN1b-GluN2A ATD and GluN1b-GluN2B ATD (Karakas et al., 2011; Tajima et al., 2016).
The heterodimeric GluN1-GluN2B subunit interface within ATD is where phenylethanolamine compounds such as ifenprodil bind in the GluN1-GluN2B subtype-specific manner (Karakas et al., 2011), however, the mechanism of this subtype-specific binding has remained obscure due to lack of ATD structures from any NMDA receptor subtypes except the GluN1-GluN2B (Figure 5). While no binding of ifenprodil to the GluN1-GluN2C and GluN1-GluN2D subtypes may be explained by non-conservation of primary sequences, all but one residue, GluN2A-Met112, is conserved in GluN2A at the region corresponding to the ifenprodil binding site (Karakas et al., 2011). We have previously altered GluN2A-Met112 to the corresponding residue in GluN2B, isoleucine, and observed no gain of ifenprodil sensitivity (Karakas et al., 2011). Here, our structure show that the ~12° inter-subunit rotation in the GluN1b-GluN2A ATD compared to the ifenprodil-GluN1b-GluN2B ATD described above results in shortening of the distance between GluN1-α3 and GluN2-α2′, thereby eliminating the cavity space for compound binding (Figure 5C). Indeed, the GluN1b-GluN2A ATD structures show a cavity with insufficient volume (~130 Å3) (Figure 5F) to accommodate the ifenprodil molecule that has the volume size of 324 Å3. The measured volume of the inter-subunit cavity is 697 Å3 (Figure 5D) in the ifenprodil-GluN1b-GluN2B ATD.
To determine if the cavity in the subunit interface in the GluN1b-GluN2B ATD is inherently present or formed exclusively by binding of ifenprodil, we obtained a crystal structure of the GluN1b-GluN2B ATD in the absence of ifenprodil (Figure S5, Table S1). This structure (apo2-GluN1b-GluN2B ATD) is distinct from our recently published apo-GluN1b-GluN2B ATD (Tajima et al., 2016) in that the bi-lobe of GluN2B ATD is ‘closed’ and the inter-GluN1-GluN2 subunits are arranged in a similar manner to the ifenprodil-GluN1b-GluN2B ATD, which is an indication that the conformation of the GluN1b-GluN2B ATD can intrinsically fluctuate as previously predicted (Gielen et al., 2009). Importantly this new apo-state structure with the ‘closed’ bi-lobe (apo2-GluN1b-GluN2B ATD) retains the subunit interface cavity with sufficiently large volume (550 Å3) to accommodate ifenprodil demonstrating that the presence of the large cavity is intrinsic to the GluN1b-GluN2B ATD (Figure 5E). In contrast, the apo-GluN1b-GluN2B ATD with the ‘open’ GluN2B ATD bi-lobe (Tajima et al., 2016) has little or no cavity space indicating that the ifenprodil binding site is created only when the GluN2 ATD bi-lobe is sufficiently ‘closed.’
Our structural analysis unambiguously reveals the molecular organization of the zinc-binding site responsible for the high-affinity zinc inhibition. Zinc anomalous difference Fourier maps unambiguously show one zinc-binding site at the inter-R1–R2 cleft of the GluN2A ATD in both the Zn1-GluN1b-GluN2A ATD and the Zn2-GluN1b-GluN2A ATD. In the Zn2-GluN1b-GluN2A ATD, we noticed the presence of another zinc-binding site within the R1 region of the GluN1b ATD (Figure S6). The residues for this second zinc binding site within the GluN1b ATD, GluN1b-His67, -His94, and -Asp100, are conserved amongst multiple species, however, mutating those residues has little or no effect on the high-affinity zinc inhibition or the micromolar affinity zinc inhibition, which has previously been shown to occur independent of voltage and of GluN2A ATD (Fayyazuddin et al., 2000) (Figure S6). The electron density for the zinc-binding site within GluN2A ATD unambiguously shows that zinc is coordinated by the four residues, GluN2A-His44, -His128, -Glu266, and -Asp282 (Figure 6A and 6B). The zinc-recognition pattern in the GluN2A ATD is different from the one previously observed in the GluN2B ATD with only two directly coordinating residues (Karakas et al., 2009) (Figure 6D). Notably, the loop that contains GluN2A-His44 (Zn-loop) has a distinct length and structure from the GluN2B ATD (Figure 6C–E). Extensive mutagenesis studies in the past demonstrated that altering GluN2A-His44, GluN2A-His128, and GluN2A-Glu266, but not GluN2A-Asp282 to alanine reduces potency and efficacy of the zinc inhibition (Choi and Lipton, 1999; Fayyazuddin et al., 2000). Instead, the GluN2A-Asp283Ala mutation was shown to alleviate the high-affinity zinc inhibition, which led to the conclusion that GluN2A-Asp283 but not GluN2A-Asp282 is involved in direct zinc coordination (Stroebel et al., 2011). In order to validate our structural observation that GluN2A-Asp282 is directly involved in coordination of zinc and therefore high-affinity zinc inhibition, we mutated GluN2A-Asp282 to alanine, histidine, and glutamate and measured zinc-inhibition by electrophysiology and found that site-directed mutation of GluN2A-Asp282 dramatically affected zinc inhibition in our hands (Figure 6F and 6G). Together, the combination of our mutagenesis results and the previous results (Choi and Lipton, 1999; Fayyazuddin et al., 2000; Low et al., 2000; Paoletti et al., 2000) support structural observation that side chains of GluN2A-His44, GluN2A-His128, and GluN2A-Glu266, and GluN2A-Asp282 are direct coordinators of zinc.
To understand a plausible pattern of zinc association and dissociation with the GluN2A ATD, zinc in the Zn1-GluN1b-GluN2A ATD was removed by exhaustively soaking crystals against the crystallization solution containing 1 mM EDTA (EDTA-GluN1b-GluN2A ATD). As a quality control of the experiment, we collected x-ray diffraction data at the zinc peak wavelength (1.28 Å) and observed no signal in the anomalous difference Fourier map to confirm absence of zinc in the crystal. While the crystal packing kept the overall bi-lobe conformation to that observed in the Zn1- or Zn2-GluN1b-GluN2A ATD, the EDTA-GluN1b-GluN2A ATD revealed a significant structural change in Zn-loop that contains the zinc coordinating residue, GluN2A-His44, uniquely present in GluN2A (Figure 6 and Figure S7). Zn-loop is ordered and directed toward the zinc-binding site at the cleft in the Zn1- or Zn2-GluN1b-GluN2A ATD. In the EDTA-GluN1b-GluN2A ATD, Zn-loop flaps away from the bi-lobe cleft and the electron density for the side chain of GluN2A-His44 becomes disordered. This result implies that the high-affinity zinc binding follows an induced fit mechanism involving rearrangement of Zn-loop. It is noteworthy to mention that this local change in the Zn-loop conformation is not caused by the crystal packing. Despite extensive efforts, we were unable to capture the bona-fide apo-state with a plausible open-cleft ATD as we recently observed in the GluN2B ATD (Tajima et al., 2016) due to unsuccessful attempt to crystallize the GluN1b-GluN2A ATD proteins in the complete absence of zinc.
To gain mechanistic insights into the high-affinity zinc inhibition, we evaluated the role of residues that are not directly coordinating zinc but affecting zinc-inhibition. Two residues, GluN2A-Asp105 in R1 and GluN2A-Lys233 in R2 located at the R1–R2 interface of the ATD bilobes, form a salt bridge as evident from the strong electron density in all of the structures (Figure 7A). Site-directed mutagenesis of those two residues, GluN2A-Asp105Ala and GluN2A-Lys233Arg, has been previously reported to show significant effects on both potency and efficacy of high-affinity zinc inhibition (Fayyazuddin et al., 2000; Paoletti et al., 2000), which originally led to the hypothesis that those residues may be involved in zinc binding. With the GluN1b-GluN2A ATD structures in our hands, the above result led us to newly speculate that the polar interaction network at the GluN2A R1–R2 interface may be a critical element for mediating allosteric regulation by ATD. To buttress this hypothesis, we first incorporated GluN2A-Lys233Ala to completely break the salt bridge with GluN2A-Asp105 and measured zinc sensitivity by electrophysiology. The GluN2A-Lys233Ala mutation abolished the high-affinity zinc inhibition to an extent that is comparable to GluN2A-Lys233Arg (Figure 7C) supporting the view that the inter-R1–R2 interaction is necessary for eliciting high-affinity zinc inhibition. To further understand the role of polar interactions at the GluN2A R1–R2 interface in zinc inhibition, we tested the effect of mutating GluN2A-Asn264 that is located in the R2 lobe and is in position to form a hydrogen bond with GluN2A-Asp105 located in the R1 lobe. While GluN2A-Asn264Ala showed no strong effect in the efficacy, the potency for the zinc-inhibition lowered by ~2-fold (Figure 7C). Placement of a bulky side chain by the GluN2A-Asn264Trp mutation robustly abolished the zinc-inhibition in a similar manner to the GluN2A-Lys233Ala mutation (Figure 7C).
Previous studies have shown that ATD in NMDA receptor is a critical locus for subtype-specific gating control (Gielen et al., 2009; Mony et al., 2011; Yuan et al., 2009a), thus, we next questioned the role of the above R1–R2 interaction in channel open probability (Popen). Toward this end, we estimated open probability by an indirect method (Jones et al., 2002) where the GluN1-Ala652Cys mutant is co-expressed with the GluN2A mutants, GluN2A-Lys233Ala, GluN2A-Lys233Arg, GluN2A-Asn264Ala, and GluN2A-Asn264Trp and covalently modified by 2-aminoethylmethanethiosulphonatehydrobromide (MTSEA) to lock the open channel. The MTSEA modified receptor is assumed to have Popen = 1.0 and thus, the degree of potentiation of the maximal agonist response would be inversely related to Popen. All of the R1–R2 interface mutants, GluN2A-Lys233Ala, GluN2A-Lys233Arg, GluN2A-Asn264Ala, and GluN2A-Asn264Trp, showed 1.3~2-fold higher estimated open probabilities compared to the wildtype indicating that destabilization of the R1–R2 interface in GluN2A ATD favors activation of the NMDA receptor channel (Figure 7D and Figure S8A). Indeed, the equivalent mutants in GluN2B, GluN2B-Lys234Ala, also has ~2-fold increase in the estimated Popen (Figure 7B and 7E and Figure S8B), showing that the correlation between the strength of the R1–R2 interaction and the gating properties of NMDA receptor ion channels is applicable to both the GluN1-GluN2A and the GluN1-GluN2B subtypes.
Recombinant expression of eukaryotic multimeric glycoproteins or glyco-membrane proteins such as NMDA receptor ATDs, in general, poses a challenge due to requirement of those proteins to interact with each other with a defined stoichiometry and to be posttranslationally modified to pass through the secretory pathways involving the endoplasmic reticulum (ER) and the Golgi apparatus. Production of the GluN2A ATD proteins required co-expression of the GluN1 ATD to mask the ER retention signal present within the GluN2A ATD (Qiu et al., 2009) for proper trafficking and secretion. A number of oligomeric proteins such as GABAB receptors have a similar characteristic where trafficking to the plasma membrane occurs only when the two subunits, GBR1 and GBR2, interact with each other to mask the ER retention signal (Margeta-Mitrovic et al., 2000; Pagano et al., 2001). The ~5-fold higher expression level observed in the GluN1b-GluN2A fusion approach compared to the co-expression approach may attribute largely to efficient masking of the ER retention signal due to the facilitated GluN1-GluN2A interaction. Another potential factor may be the addition of three designed Asn-linked glycosylation sites in 57-link that is used to fuse GluN1b and GluN2A ATDs together (Figure 1). Asn-linked glycosylation have been previously shown to increase a secretion level in mammalian and yeast cells (Liu et al., 2009). Furthermore, this fusion approach ensured a 1:1 stoichiometry of the two subunits at a translational level, thus, minimizing a possibility for heterogeneous subunit association (e.g. GluN1 and GluN2A homodimers). Overall, our expression strategy that uses 57-link may be generally applicable to expression of other hetero-multimeric glycoproteins.
Our GluN1b-GluN2A ATD structures represent the first of NMDA receptor ATD other than the GluN1b-GluN2B ATD, thus, provide the novel opportunity to define molecular elements for subtype-specificity by structural comparison. The current study answers the following two fundamental questions related to the NMDA receptor ATD pharmacology: 1) how does GluN2A ATD elicit ~80–200-fold higher zinc potency than GluN2B ATD?; and 2) why is ifenprodil specific to GluN2B ATD? Our crystal structures show that the high-affinity zinc binding site in GluN2A is formed by direct coordination of the four residues, GluN2A-His44, -His128, -Glu266, and -Asp282 located at the R1–R2 interface (Figure 6). While GluN2A-His44, -His128, and -Glu266 were correctly predicted to participate in the high-affinity zinc inhibition by extensive mutagenesis (Fayyazuddin et al., 2000; Low et al., 2000; Paoletti et al., 2000), identifying the fourth residue, GluN2A-Asp282, was difficult even with the availability of the GluN2B ATD structure (Karakas et al., 2009) to guide mutagenesis. This challenge stems from the structural difference between the hinge loop that contains GluN2A-Asp282 and the equivalent loop in GluN2B (Figure 6C and 6D), which is not predictable by comparison of primary sequences (Figure 6E and Figure S1A). Nevertheless, the zinc binding site in GluN2A formed by the four directly coordinating residues is in stark contrast to the micro-molar zinc binding site in GluN2B where only two residues, GluN2B-His127 and GluN2B-Glu284, are participating in direct zinc coordination (Karakas et al., 2009). Thus, we show here that the number of coordinating bonds is the major contributing factor for the difference in the zinc-inhibition potency between GluN2A and GluN2B. Consistent with this observation, the GluN2A-His44Ala and -Glu266Ala mutants have been reported to show micro-molar zinc inhibition potency (IC50 = 1.1 μM and 1.4 μM for GluN2A-His44Ala and –Glu266Ala, respectively) similar to that observed in the GluN1-GluN2B NMDA receptors (IC50 = 0.8 μM) (Rachline et al., 2005). The GluN1b splice variant containing the Exon 5 encoded sequences is known to lower potency of voltage-independent zinc inhibition by approximately 10-fold when combined with GluN2A (Traynelis et al., 1998). This significant effect of Exon 5 on the zinc potency is likely indirect since there is no evidence of Exon 5 – GluN2A interaction in our crystal structure. Instead, the effect is caused by alleviation of proton sensitivity, which is coupled to zinc sensitivity, by Exon 5 as previously shown (Traynelis et al., 1998).
The mechanism underlying lack of ifenprodil binding in GluN2A whose primary sequence is highly similar at and around the binding pocket to GluN2B has been enigmatic (Karakas et al., 2011). Our structural study showed that the difference in the architecture of the subunit interface between the GluN1b-GluN2A ATD and the GluN1b-GluN2B ATD accounts for specific binding of ifenprodil to the GluN1-GluN2B ATD (Figure 5). Specifically, in the GluN1b-GluN2A ATD, the space in the equivalent locus to the ifenprodil binding site at the subunit interface is mostly eliminated due to the ~12° rotation in the GluN1-GluN2 orientation compared to that of the GluN1b-GluN2B ATD (Figure 3C and 3D and Figure 5). This ‘skewing’ of the GluN1-GluN2 orientation and the cavity disruption at the subunit interface are permitted by the intrinsic nature of the GluN2A ATD whose ‘fully-closed’ zinc-bound bi-lobed architecture is ~13° more ‘open’ than that of the ‘fully-closed’ GluN2B ATD bound to zinc or ifenprodil (Figure 3B). That is, the space in the GluN1-GluN2 subunit interface is more ‘narrowed’ when the bi-lobe of the GluN2 ATD is more open. Thus, the lack of ifenprodil binding in the GluN1-GluN2A NMDA receptors is caused by the difference in the pattern of the GluN1-GluN2 interaction, which attributes to an inability of the GluN2A ATD cleft to ‘close’ as much as GluN2B ATD (Figure 5G and Figure S4).
The ATDs in NMDA receptors control gating properties including Popen and speeds of deactivation (Gielen et al., 2009; Yuan et al., 2009a). The difference in the pattern of bi-lobe opening and GluN1-GluN2 subunit interaction between the ifenprodil-GluN1b-GluN2B ATD (Karakas et al., 2011) and the Zn1-GluN1b-GluN2A ATD is reminiscent of that observed between the ifenprodil-GluN1b-GluN2B ATD and the apo-GluN1b-GluN2B ATD (Tajima et al., 2016). We recently showed that the two types of conformational movements, ‘opening’ of the GluN2B ATD bi-lobe and the rotation of the GluN1-GluN2B inter-subunit orientation along with agonist binding, are associated with channel gating (Tajima et al., 2016). By following this analogy, it is reasonable to speculate that the intrinsic differences in the extent of bi-lobe opening and the pattern of subunit interaction between GluN1-GluN2A and GluN1-GluN2B NMDA receptors may account for the differences in the gating properties between the two (e.g. the GluN1-GluN2A NMDA receptors have higher channel open probability than the GluN1-GluN2B NMDA receptors). Similar to the gating mechanism in the GluN1-GluN2B NMDA receptors, we speculate activation of the GluN1-GluN2A NMDA receptors accompanies opening of the GluN2A ATD bi-lobe and the GluN1-GluN2 subunit rotation (Figure 8). Allosteric inhibitors, zinc and ifenprodil both stabilize ‘closed’ conformation of the ATD bi-lobes (Karakas et al., 2009) to place the channel into the ‘non-active’ state. Supporting this view is our experiment showing that perturbation of the R1–R2 interface or disfavoring bi-lobe closure, by mutating residues such as GluN2A-Lys233, GluN2A-Asn264 to break inter-R1–R2 polar interactions alleviates high-affinity zinc-inhibition and increases Popen (Figure 7). The bi-lobe opening and closing in the GluN2A ATD is further supported by the study where the pattern of luminescence resonance energy transfer between two labeled residues in R1 and R2 estimates a greater distance in the apo-state than in the zinc-bound state in GluN2A (Sirrieh et al., 2013).
The current as well as many other studies in the field show that the NMDA receptor ATDs are highly exposed domain that are robustly involved in controlling the ion channel activities. Those intrinsic characteristics of ATDs are manipulated by natural pressures including genetic mutations, proteolysis, and antibody binding to alter functions of NMDA receptors. Recent examples of de novo ATD mutations are GluN2A-Pro79Arg and GluN2A-Arg370Trp, which result in altered zinc-sensitivity and cause childhood epilepsy and cognitive deficit (Serraz et al., 2016). Another case is proteolysis of the GluN2A ATD by a serine protease, plasmin, at GluN2A-Lys317, which results in a loss of zinc-sensitivity (Yuan et al., 2009b). Finally, the autoimmune antibodies causing anti-NMDA receptor encephalitis have been reported to recognize the GluN1 ATD and affect Popen and the surface expression level of the NMDA receptors (Dalmau et al., 2008). It is highly conceivable that the above factors alter conformational dynamics of ATDs and thus, affects the ion channel properties. The precise mechanism underlying such changes in ATD conformations and ion channel activities by external factors is another aspect of ATD biology that needs attention.
Overall, the current study provides the molecular insights into zinc recognition and the high-affinity zinc inhibition in the GluN2A-containg NMDA receptor. This study sets the field one step closer to tackling the next question regarding transduction of conformational signal from ATD to LBD onto TMD and vice versa in the GluN2A-containing NMDA receptors as indicated by electrophysiological studies showing strong allosteric coupling between ATD and LBD (Gielen et al., 2008; Zheng et al., 2001). The structural insights gained here clearly establish an initial guideline for studying molecular mechanisms of subtype-specificity and allosteric regulation in NMDA receptors.
All experiments were conducted using protocols approved by the Cold Spring Harbor Laboratory Animal Care and Use Committees.
The DNA sequence encoding Xenopus laevis GluN1 Isoform 1b ATD (Asp23 to Glu408) containing Cys22Ser, Asn61Gln and Asn371Gln mutations fused to that encoding Rattus norvegicus GluN2A ATD (Gly30 to Tyr393) via the 57 amino acid linker (57-link), which includes two thrombin sites and two histidine tags (Figure 1). 57-link was designed so that it is sufficient to fulfill the distance gap between the C-terminal end of GluN1 ATD and the N-terminal end of GluN2B ATD in the crystal structure of ifenprodil-GluN1b-GluN2B ATD (Karakas et al., 2011; Tajima et al., 2016). The GluN1b-GluN2A ATD fusion protein was expressed as a secreted protein using the High Five cells (Trichoplusia ni)/baculovirus system. The High Five cell culture (1.8 ×106 cells/ml) grown in ESF921 medium (Expression Systems) was infected with the recombinant virus harboring the GluN1-GluN2A ATD fusion and harvested after 48 h.
Monoclonal antibodies (mouse immunoglobulin-γ (IgG)) that bind to GluN1 ATD (RRID: AB_2629508) from Xenopus laevis and Rattus norvegicus were made by immunizing mice with the purified intact GluN1a-GluN2B NMDA receptor (Tajima et al., 2016). The IgGs (IgG29) were purified from hybridoma cell culture supernatant by Protein-A Sepharose (GE healthcare). The Fab fragments (Fab29) of IgG29 were obtained by papain proteolysis followed by re-chromatographing onto Protein-A Sepharose to remove the Fc fragments.
The High Five cell culture medium containing secreted GluN1b-GluN2A ATD fusion proteins was collected, concentrated, and dialyzed against 20 mM Tris-HCl (pH 8.0) and 200 mM NaCl using tangential flow filtration. The concentrated sample supplemented with 5 mM imidazole was load onto a Chelating Sepharose (GE Healthcare) conjugated with cobalt (Co-NTA), washed with 20 Column Volumes (CV) of 40 mM imidazole, 10 CV of 80 mM imidazole and eluted with 10 CV of 300 mM imidazole. The octa-histidine tags and 57-link were removed by thrombin digestion at 18°C for 20 h. The proteins were concurrently de-glycosylated by endoglycosidase, EndoF1, at a 1:1 (w:w) ratio of GluN1b-GluN2A ATD and EndoF1 in 20 mM Tris-HCl and 200 mM NaCl at pH 8. The digested samples were recharged onto the Cobalt-NTA at 20 mM imidazole and the flow-though was collected. At this step Fab29 is added to the protein at a 1:2.25 molar ratio (GluN1b-GluN2A ATD:Fab). The mixture was subjected to a size-exclusion chromatography (Superdex200) in the presence of 10 μM ZnCl2 and the peak fraction containing the GluN1b-GluN2A ATD-Fab complex was collected. Expression and purification of GluN1b-GluN2B ATD was done as described previously described (Karakas et al., 2011) with the exception that there was no addition of ifenprodil or any other ligands during the purification process.
The GluN1b-GluN2A ATD-Fab protein was crystallized by hanging-drop vapor diffusion at 17°C by mixing the protein (12–13 mg/ml) at a 1:1 or 2:1 volume ratio with a reservoir solution containing 1.8–1.9 M ammonium Sulfate, 2.5–3.0% isopropanol. Crystals normally appeared after 3–4 days and continue growing for up to 2 weeks. Crystals were frozen in the crystallization buffer supplemented with 20% glycerol. The GluN1b-GluN2B ATD was crystallized by hanging-drop vapor diffusion at 17°C by mixing the protein (10 mg/ml) at 2:1 volume ratio with a reservoir solution containing 20% PEG 3350, 1.4 M sodium formate, 0.3 M sodium thiocyanate and 0.1 M sodium acetate pH 5.0. Crystals were frozen in the presence of 5% glycerol. X-ray diffraction data was collected at the ID23-B and ID23-D beamlines at the Advanced Photon System (APS) in the Argonne National Laboratory and at the X25 beamline at the National Synchrotron Light Source at Brookhaven National Laboratory. Anomalous diffraction data was collected at the wavelength of 1.27 Å (zinc peak) whereas native data set was collected at 1.0 Å. Structures were solved by the molecular replacement method using GluN1b ATD, GluN2B ATD (PDB code: 3QEL) and anti-dinitrophenyl-spin-label Fab (PDB ID: 1BAF) as molecular search probes using the program PHASER. Structural refinement and model building was performed with PHENIX and COOT, respectively.
Recombinant GluN1 ΔCTD (Met1-Gln868) - GluN2A ΔCTD (Ala58-Gly883 fused to the Xenopus GluN1 signal peptide) constructs with all free cysteines mutated to serine were expressed in Spodoptera frugiperda (Sf9) cells. The cell pellets were solubilized in a buffer containing 20 mM HEPES-NaOH (pH 7.0), 200 mM NaCl, 10 mM Glycine, 10 mM Glutamate, 0.5% Lauryl Maltose Neopentyl Glycol (LMNG; Anatrace) and 1 mM Phenylmethylsufonyl floride at 4°C for 1 h. The solubilized membranes were isolated by centrifugation (~125,000g) for 40 min and the supernatant was loaded onto Step-Tactin Sepharose (IBA), washed with a buffer containing 20 mM HEPES-NaOH (pH 7.0), 200 mM NaCl, 10 mM Glycine, 10 mM Glutamate, 0.002% Lauryl Maltose Neopentyl Glycol, 10 μM ZnCl2 and eluted with the same buffer containing 3 mM desthiobiotin. Western blotting was carried out on samples with and without 100 mM beta-mercaptoethanol using 7% SDS-polyacrylamide gel electrophoresis, nitrocellulose blotting membrane (GE Healthcare), and mouse monoclonal anti-GluN1 antibody (MAB1586, Millipore, RRID: AB_2279138) or anti-GluN2A (SAB5200888, Sigma-Aldrich, RRID: AB_2629501), followed by HRP-conjugated anti-mouse secondary antibodies (GE Healthcare). The ECL detection kit (GE Healthcare) was used to visualize the antigens.
Recombinant GluN1-1a-GluN2A NMDA receptors for the wildtype and mutants were expressed by co-injecting 0.1 ng and 1 ng of cRNAs encoding GluN1-1a and GluN2A at a 1:1 (w/w) ratio into defolliculated Xenopus laevis oocytes. The two-electrode voltage-clamp recordings were performed using agarose-tipped microelectrodes (0.4–1.0 MΩ) filled with 3 M KCl at a holding potential of −60 mV for zinc dose–response. The bath solution contained 5 mM HEPES, 100 mM NaCl, 10 mM Tricine and 0.3 mM BaCl2 at pH 7.3 (adjusted with KOH). Peak currents were measured by adding 100 μM each of the agonists, glycine and L-glutamate, and high-affinity zinc inhibition was monitored in the presence of the agonists and various concentrations of ZnCl2. The free zinc concentration in the above buffer was estimated as: [Zinc]free = 1/200 [Zinc]added as done previously (Fayyazuddin et al., 2000). For the MTSEA experiments, the glycine/glutamate-induced currents were measured in the presence and absence of MTSEA (0.2 mM). Data sets were acquired and analyzed by the program Pulse (HEKA). Dose–response curves were plotted and fit using Kaleida graph (Synergy software). Statistical significance for MTSEA experiments was estimated by the Two-sample Kolmogorov-Smirnov test using Matlab R2015b.
We thank staff at the 23-ID beamlines at the Advanced Photon System in the Argonne National Laboratory and the X25 beamline at the National Synchrotron Light Source in the Brookhaven National Laboratory. We also thank Pierre Paoletti for providing important comments on this work. This work was supported by the National Institutes of Health (MH085926 and GM105730), the Stanley Institute of Cognitive Genomics, and the Robertson Research Fund of Cold Spring Harbor Laboratory (all to H.F.). A.R.H. is funded by the Genentech Foundation Fellowship and the Starr Foundation. The structural coordinates related to this work have been deposited to the Protein Data Bank with the entry 5TPW, 5TQ2, 5TQ0, and 5TPZ for Zn1-GluN1b-GluN2A ATD, Zn2-GluN1b-GluN2A ATD, EDTA-GluN1b-GluN2A ATD, and apo2-GluN1b-GluN2B ATD, respectively.
Author contributionsARH and HF jointly contributed to project design. ARH and EK expressed and purified proteins and did x-ray crystallography. ARH conducted electrophysiology. NS characterized ATD proteins and antibodies crucial for the x-ray crystallographic study. ARH and HF were involved in manuscript preparation.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.