D-serine potently drives ligand-binding domain closure in the ionotropic glutamate receptor GluD2

SUMMARY

Despite their classification as ionotropic glutamate receptors, GluD receptors are not functional ligand-gated ion channels and do not bind glutamate. GluD2 receptors bind D-serine and coordinate trans-synaptic complexes that regulate synaptic plasticity. Instead of opening the ion channel pore, mechanical tension produced from closure of GluD2 ligand-binding domains (LBDs) drives conformational rearrangements for non-ionotropic signaling. We report computed conformational free energy landscapes for the GluD2 LBD in apo and D-serine-bound states. Unexpectedly, the conformational free energy associated with GluD2 LBD closure upon D-serine binding is greater than that for AMPA, NMDA, and kainate receptor LBDs upon agonist binding. This excludes insufficient force generation as an explanation for lack of ion channel activity in GluD2 receptors and suggests non-ionotropic conformational rearrangements do more work than pore opening. We also report free energy landscapes for GluD2 LBD harboring a neurodegenerative mutation and demonstrate selective stabilization of closed conformations in the apo state.

Graphical Abstract

eTOC Blurb

The GluD family of ionotropic glutamate receptors (iGluRs) lacks functional ion channel activity. Surprisingly, computed conformational free energies reveal that agonist binding in GluD2 receptors drives ligand-binding domain (LBD) closure more strongly than observed in other iGluRs. A neurodegenerative mutation within the GluD2 LBD stabilizes closed, active conformations.

INTRODUCTION

Ionotropic glutamate receptors (iGluRs) are ligand-gated ion channels that mediate neurotransmission. Enriched in postsynaptic termini of neurons, iGluRs are critical for fundamental neuronal processes such as learning, memory, long-term potentiation/depression, and synaptic plasticity (Contractor et al., 2011; Henley and Wilkinson, 2016; Iacobucci and Popescu, 2017). Signaling mediated by iGluRs is implicated in numerous neurological diseases, including stroke, traumatic brain injury, pain, amyotrophic lateral sclerosis, epilepsy, depression, schizophrenia, glioma, and HIV-associated neurocognitive disorder (Barria, 2019; Chang et al., 2012; Dickens et al., 2017; Lerma and Marques, 2013; Zhou and Sheng, 2013). The four subfamilies of iGluRs are the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors, N-methyl-D-aspartate receptor (NMDA) receptors, kainate receptors, and δ (GluD) receptors (Traynelis et al., 2010). Functional iGluRs are tetrameric and exhibit modular architecture, with each subunit consisting of four distinct domains: an extracellular amino-terminal domain (ATD), ligand-binding domain (LBD), transmembrane domain (TMD), and intracellular carboxy-terminal domain (CTD) (Greger and Mayer, 2019; Greger et al., 2017; Sobolevsky, 2015).

Despite their classification as iGluRs and structural similarity to AMPA, NMDA, and kainate receptors, GluD receptors are not ligand-gated ion channels and do not bind glutamate (Yuzaki, 2003). GluD receptors were long considered orphan receptors until D-serine was identified as an endogenous ligand for GluD2 receptors (Kakegawa et al., 2011; Naur et al., 2007). The Lurcher mutation is a widely studied variant of the GluD2 receptor that confers spontaneous ion channel activity and causes neurodegeneration (Zuo et al., 1997). GluD2 receptors can also become functional ion channels by LBD transplantation. Replacing the LBD of a GluD2 receptor with that of a GluK2 receptor creates a glutamate-gated ion channel chimera, suggesting that GluD2 LBDs function differently than other iGluR LBDs (Schmid et al., 2009). A longstanding question, therefore, is whether the free energy that becomes available for driving LBD closure upon agonist binding is insufficient for gating GluD receptors, and explains why they are not functional ion channels.

GluD receptors are the most functionally enigmatic among the iGluR subfamilies. Growing evidence indicates that GluD receptors are key bidirectional synaptic organizers, proteins in trans-synaptic complexes that initiate signaling in pre- and postsynaptic neurons to regulate synaptic plasticity (Yuzaki and Aricescu, 2017). GluD2 is enriched in the dendrites of cerebellar Purkinje cells receiving synaptic inputs from parallel fibers, which release Cbln1 that simultaneously binds presynaptic neurexins and postsynaptic GluD2 ATDs (Matsuda et al., 2010; Uemura et al., 2010). Burst stimulation of parallel fibers induces Bergmann glia to release D-serine, which binds GluD2 LBDs to enhance long-term depression and motor performance (Kakegawa et al., 2011). The neurexin-Cbln1-ATD trans-synaptic complex engages in allosteric interactions with LBDs, and this ATD-LBD coupling in GluD2 is critical for D-serine-dependent functions (Elegheert et al., 2016). It has been hypothesized that anchoring the GluD2 ATD to Cbln1 and neurexin restricts its motions, thereby efficiently funneling force generated by D-serine-induced GluD2 LBD closure towards the TMD and CTD for downstream signaling (Elegheert et al., 2016; Yuzaki and Aricescu, 2017). However, the structural mechanisms underpinning GluD2 LBD closure upon D-serine binding are unknown.

Computing conformational free energy landscapes, or potentials of mean force (PMFs), describes how large-scale conformational stabilities of the LBD change upon ligand binding. These changes provide the useful work for ion channel gating (Lau and Roux, 2007). Utilized for opening the ion channel pore in AMPA, NMDA, and kainate receptors, this work is presumably transmitted to the ATD, TMD, and/or CTD of GluD receptors to drive conformational rearrangements for non-ionotropic functions (Elegheert et al., 2016; Yuzaki and Aricescu, 2017). Importantly, a causal relationship between ligand-induced LBD closure and conformational changes of other domains in GluD receptors exists – electrophysiological experiments demonstrate that D-serine abrogates spontaneous ion channel conductance in GluD2 receptors harboring the Lurcher mutation (Naur et al., 2007). Because it is difficult for experimental approaches to capture inherently transient or sparsely populated conformational states of iGluR LBDs, molecular dynamics (MD) simulations are valuable for estimating the populations of these states through conformational free energy calculations (Lau, 2019; Lau and Roux, 2007; Yu et al., 2016a). We have previously computed PMFs for AMPA, NMDA, and kainate receptor LBDs in apo- and ligand-bound conformations (Lau and Roux, 2007, 2011; Wied et al., 2019; Yao et al., 2013; Yu et al., 2016b).

Here we report the first PMFs for GluD receptor LBDs. Using all-atom umbrella sampling MD simulations, we computed PMFs for apo and D-serine-bound GluD2 LBDs. Notable similarities and differences in the free energy minima and free energy barriers to open conformations exist between GluD2 LBD PMFs and the PMFs of AMPA, NMDA, and kainate receptor LBDs. We found via ligand accessibility analysis that the conformational free energy attributable to agonist binding in the GluD2 LBD is greater than that to agonist binding in AMPA, NMDA, and kainate receptor LBDs. Our finding that agonist binding potently drives LBD closure in GluD2 LBDs is unexpected and excludes insufficient force generation from D-serine-induced LBD closure as an explanation for the lack of GluD2 ion channel activity. Additionally, our results suggest that conformational rearrangements for non-ionotropic functions of GluD2 receptors may require more work than ion channel pore opening. We also assessed the thermodynamic contribution of a clinically significant missense mutation, R710W, located in the GluD2 LBD that is associated with autosomal recessive early-onset cerebellar syndrome (Ali et al., 2017). Our results indicate that the R710W mutation selectively stabilizes semi-closed conformations of the apo GluD2 LBD. This stabilization reduces the conformational free energy made available for useful work upon D-serine binding.

RESULTS

Conformational dynamics of the GluD2 LBD

The crystal structures of apo and D-serine-bound GluD2 LBD (Naur et al., 2007) are shown in Figure 1A. To probe the conformational dynamics of the GluD2 LBD, we computed conformational PMFs for apo and D-serine-bound LBDs. The PMFs were computed by utilizing an umbrella sampling approach along a two-dimensional (2D) order parameter (ξ₁, ξ₂) that reports the extent of openness of the LBD clamshell (Figure 1B). ξ₁ and ξ₂ each correspond to a distance between the top lobe and the bottom lobe of the LBD. This approach has been previously employed to characterize large-scale conformational transitions in iGluR LBDs (Lau and Roux, 2007, 2011; Yao et al., 2013; Dai and Zhou, 2015; Yu et al., 2016b; Wied et al., 2019). Furthermore, a calculation of the 2D PMF, W(ξ₁, ξ₂), enables an estimation of the fraction of the conformational ensemble that populates a particular conformational state – the probability of observing a conformation (ξ₁, ξ₂) is proportional to exp[–W(ξ₁, ξ₂)/k_BT], where k_B and T are Boltzmann’s constant and temperature, respectively. The apo LBD exhibits a global free energy minimum at (ξ₁, ξ₂) = (12.3, 12.8 Å) (Figure 2A) while the D-serine-bound LBD features a global free energy minimum at a more closed conformation, (ξ₁, ξ₂) = (9.5, 9.7 Å) (Figure 2B). For reference, the apo (2V3T: B) and D-serine-bound (2V3U) LBD crystallographic conformations are (ξ₁, ξ₂) = (12.7, 10.1 Å) and (ξ₁, ξ₂) = (9.6, 7.4 Å), respectively. Both apo and D-serine-bound LBDs possess a single free energy minimum, but the free energy basin of the apo LBD is broader than that of D-serine-bound LBD. The standard deviations (SDs) of the PMFs are provided in Figure 2C,D.

Figure 1.

GluD2 Ligand-Binding Domain Structure.

(A) Crystal structures of apo (light blue; PDB 2V3T) and D-serine-bound (dark blue; D-serine in orange; PDB 2V3U) GluD2 LBD, overlaid.

(B) 2D order parameter (ξ₁, ξ₂) describing the conformational state of GluD2 LBD. ξ₁ and ξ₂ each indicate a distance, represented by dashed lines, between the center-of-mass of the residues shown in spheres.

Figure 2.

Conformational Free Energy Landscapes for WT GluD2 LBDs.

(A) Apo and (B) D-serine-bound 2D PMFs. Contour lines correspond to a difference of 1 kcal/mol. Yellow dots show conformations of GluD2 PDB structures.

(C) Apo and (D) D-serine-bound 2D PMF SDs. Contour lines correspond to the respective PMF, while colors correspond to the SD calculated via block averaging.

How do the GluD2 LBD PMFs compare with the PMFs previously computed for AMPA, NMDA, and kainate receptor LBDs (Lau and Roux, 2007; Wied et al., 2019; Yao et al., 2013)? Superposition of one-dimensional (1D) projections of the 2D PMFs onto an order parameter termed ξ₁₂, which is an average of ξ₁ and ξ₂, is a convenient method to compare PMFs from different iGluR LBDs. Structural similarities between iGluR subtypes permit approximate comparisons of ξ₁₂. The 1D PMFs and associated SDs for apo and D-serine-bound LBDs are shown in Figure 3A,B. The PMF of apo GluD2 LBD resembles those of apo GluA2 and GluN2A LBDs in that they contain one broad free energy basin in open conformations and are relatively unstable in closed conformations (Figure 3C). Apo GluN2A, however, is more stable in closed conformations than either apo GluD2 or GluA2 by about 4 or 3 kcal/mol, respectively. This is in contrast to the bistable PMFs of apo GluN1, GluN3A, and GluK2 LBDs. The slope of the PMF of D-serine-bound GluD2 LBD is similar to those of glutamate-bound GluA2 and glycine-bound GluN1 LBDs (Figure 3D). The slopes of glutamate-bound GluN2A, glycine-bound GluN3A, and glutamate-bound GluK2 are shallower. These results indicate that D-serine-bound GluD2 LBD exhibits a relatively high free energy barrier to accessing open conformations.

Figure 3.

1D PMFs for WT GluD2 LBDs.

(A) Apo and (B) D-serine-bound 1D PMFs. 2D PMFs were projected per ξ₁₂ = (ξ₁ + ξ₂)/2. ξ₁₂ for the 1D PMFs are adjusted such that the value is set to 0 Å where the D-serine-bound LBD has its global free energy minimum. SDs are shown in green and calculated via block averaging.

(C) Apo 1D PMFs for GluD2 (red), GluA2 (green), GluN1 (turquoise), GluN2A (blue), GluN3A (purple), and GluK2 (black) LBDs. PMFs were aligned such that ${\xi }_{12}^{\text{'}}$ is set to 0 Å where the respective ligand-bound LBD has its global free energy minimum.

(D) 1D PMFs for D-serine-bound GluD2 (red), glutamate-bound GluA2 (green), glycine-bound GluN1 (turquoise), glutamate-bound GluN2A (blue), glycine-bound GluN3A (purple), and glutamate-bound GluK2 (black) LBDs.

R710W mutation

While specific residues involved in key interactions that govern conformational free energies of cleft closure have been identified in other iGluR LBDs (Lau and Roux, 2007; Wied et al., 2019), it remains unclear which residues within GluD receptor LBDs are critical for facilitating cleft closure. To ascertain determinants of thermodynamic stability of the closed-cleft conformation, we computed PMFs for “R710W” GluD2 LBD in apo and D-serine-bound states (this mutation is R687W in the mature polypeptide; “R710W” includes the signal sequence and is used in the clinical literature). Recently, whole-exome sequencing, magnetic resonance imaging, and clinical assessment identified c.2128C>T (p.R710W) as a novel homozygous missense variant in GRID2, the gene encoding for GluD2, that is associated with autosomal recessive early-onset cerebellar syndrome and segregates with disease in a consanguineous family (Ali et al., 2017). The R710W mutation is situated in a helix in the bottom lobe of GluD2 LBD (Figure 4A) and highly conserved across multiple species (Figure 4B). It is unknown whether the mutation affects ligand-binding in the GluD2 LBD.

Figure 4.

Conformational Free Energy Landscapes for the R710W GluD2 LBD.

(A) Location of the R710W mutation.

(B) Sequence alignment of the GluD2 protein.

(C) Apo and (D) D-serine-bound 2D PMFs.

(E) Apo and (F) D-serine-bound 2D PMF SDs.

(G) Lowest three contour levels of apo WT (black) and apo R710W (red) PMFs.

(H) Lowest three contour levels of D-serine-bound WT (black) and D-serine-bound R710W (red) PMFs.

The PMFs for apo (Figure 4C) and D-serine-bound (Figure 4D) R710W LBDs feature global free energy minima at (ξ₁, ξ₂) = (11.7, 12.4 Å) and (9.4, 9.8 Å), respectively. The SDs of apo and D-serine-bound R710W PMFs are provided in Figure 4E,F. Interestingly, compared with the apo WT PMF, the apo R710W PMF exhibits a broader free energy basin that extends toward more closed conformations of the LBD (Figure 4G). Contrastingly, the free energy basin of the D-serine-bound R710W PMF is virtually identical to that of the D-serine-bound WT PMF (Figure 4H). Taken together, these results suggest a mechanism whereby the R710W mutation selectively stabilizes more closed conformations of the apo GluD2 LBD. This may result in “leaky” activation of the receptor in the absence of D-serine and/or the receptor less readily binding D-serine because a larger fraction of the LBD conformational ensemble is closed.

Free energy difference between open and closed GluD2 LBDs

How can a given GluD2 LBD conformation be defined as open or closed? An approach to differentiate these states is to define closed states as those that do not permit entry or exit of D-serine into or out of its binding pocket. The program CAVER (Chovancova et al., 2012) was used to identify potential passageways for D-serine between the protein surface and its binding pocket. Upon evaluating multiple GluD2 LBD conformers, a single access tunnel was identified. Y473 in the upper lobe and A663 in the lower lobe were selected as bottleneck residues that form the most spatially restricted point along the tunnel (Figure 5A). These residues are in nearly analogous locations to bottleneck residues previously utilized for the GluK2 LBD (Wied et al., 2019). Based on an estimated diameter of D-serine, at least 11 Å between Y473 and A663 Cγ atoms is considered to be required for the passage of D-serine into and out of its binding pocket. Therefore, conformations with bottleneck distances greater than 11 Å are classified as open, and all others are classified as closed. Plots of average bottleneck distance versus (ξ₁, ξ₂) for WT and R710W LBDs in both apo and D-serine-bound states are shown in Figure 5B–E. Standard deviations of the bottleneck distances are shown in Figure 6A–D. Statistics for the plots were calculated using ~160,000 snapshots extracted from the simulations using VMD (Humphrey et al., 1996).

Figure 5.

Bottleneck Analysis for WT and R710W GluD2 LBDs.

(A) Residues defining the bottleneck to the D-serine binding site. Access tunnels are shown in dark blue.

(B–E) Average bottleneck distance at each (ξ₁, ξ₂) overlaid with the corresponding PMF contours for (B) apo WT, (C) D-serine-bound WT, (D) apo R710W, and (E) D-serine-bound R710W LBDs. Dashed boxes in upper right corners approximately demarcate open conformations (average bottleneck distance ≥ 11 Å).

Figure 6.

Bottleneck Analysis SDs for WT and R710W GluD2 LBDs.

(A–D) SD of bottleneck distance at each (ξ₁, ξ₂) overlaid with the corresponding PMF contours for (A) apo WT, (B) D-serine-bound WT, (C) apo R710W, and (D) D-serine-bound R710W LBDs.

What fraction of an equilibrium ensemble of LBD conformations is open versus closed? The relative probability of an LBD occupying each state is governed by the ratio of the partition functions for each state,

$$\frac{{\Omega }_{\mathit{\text{open}}}}{{\Omega }_{\mathit{\text{closed}}}}=\frac{{\int }_{\mathit{\text{open}}}{e}^{-W\left({\xi }_1,{\xi }_2\right)/k_{B}T}d{\xi }_{1}d{\xi }_2}{{\int }_{\mathit{\text{closed}}}{e}^{-W\left({\xi }_1,{\xi }_2\right)/k_{B}T}d{\xi }_{1}d{\xi }_2},$$

where W(ξ₁, ξ₂) is the PMF, and the region of integration corresponds to (ξ₁, ξ₂) in which the bottleneck distance is either ≥ 11 Å (open) or < 11 Å (closed). Thus, the conformational free energy difference between open and closed states is given by

$$\Delta G_{\mathit{\text{conf}}}=G_{\mathit{\text{open}}}-G_{\mathit{\text{closed}}}=-k_{B}T\text{ln}\left[\frac{{\Omega }_{\mathit{\text{open}}}}{{\Omega }_{\mathit{\text{closed}}}}\right].$$

The free energy by which D-serine binding stabilizes the closed LBD population is quantified by

$$\Delta \Delta G_{\mathit{\text{conf}}}=\Delta G_{\mathit{\text{conf}}}^{D-\mathit{\text{ser}}}-\Delta G_{\mathit{\text{conf}}}^{\mathit{\text{apo}}}.$$

Values of ΔG_conf and ΔΔG_conf for WT and R710W LBDs are provided in Table 1. Bound D-serine stabilizes the closed-cleft WT LBD by 14.4 kcal/mol relative to its apo state. In the R710W LBD, bound D-serine stabilizes the closed-cleft by 12.1 kcal/mol relative to its apo state. The lower conformational free energy for R710W LBD is expected since the apo PMF features a broader free energy basin toward closed-cleft conformations, which results in a greater ΔG_conf,apo value and a lower ΔΔG_conf value.

Table 1.

Comparison of ΔG_conf and ΔΔG_conf values across iGluR subfamilies.

	GluD2	GluD2 R710W	GluA2	GluN1	GluN2A	GluN3A	GluK2**
ΔGconf,ligand (kcal/mol)*	12.9	15.5	10.1	1.7	2.9	1.2	2.6
ΔGconf,apo (kcal/mol)	−1.5	3.4	0.0	−0.9	0.1	−0.7	−0.2
ΔΔGConf (kcal/mol)	14.4	12.1	10.1	2.6	2.8	1.9	2.8

^*Ligand is D-serine (GluD2), glutamate (GluA2, GluN2A, GluK2), and glycine (GluN1, GluN3A).

^**Values obtained from Wied et al., 2019.

Comparison with other iGluR LBDs

How does the conformational free energy attributable to ligand-binding in GluD2 LBD compare with that of other iGluR LBDs? While ΔΔG_conf calculations have been made for the GluK2 LBD (Table 1) (Wied et al., 2019), ΔΔG_conf values for AMPA and NMDA receptor LBDs have not been calculated. Hence, we performed ligand accessibility analysis using the bottleneck method on previously computed PMFs for GluA2, GluN1, GluN2A, and GluN3A LBDs (Lau and Roux, 2007; Yao et al., 2013). Bottleneck distance cutoffs of 10 Å and 12 Å were applied to glycine and glutamate, respectively, to delineate open and closed states (Wied et al., 2019; Yu et al., 2016b). The bottleneck residues and plots of average bottleneck distance, defined as the distance between Y448 and L648 Cγ atoms, versus (ξ₁, ξ₂) (Lau and Roux, 2007) for GluA2 LBD in both apo and glutamate-bound states are shown in Figure 7A–C. The bottleneck residues and plots of average bottleneck distance, defined as the distance between F461 and W708 Cγ atoms, versus (ξ₁, ξ₂) (Yao et al., 2013) for GluN1 LBD in both apo and glycine-bound states are shown in Figure 7D–F. The bottleneck residues and plots of average bottleneck distance, defined as the distance between Cα of G460 and Cβ of N664, versus (ξ₁, ξ₂) (Yao et al., 2013) for GluN2A LBD in both apo and glutamate-bound states are shown in Figure 7G–I. The bottleneck residues and plots of average bottleneck distance, defined as the distance between Cγ of Y582 and Cγ1 of V774, versus (ξ₁, ξ₂) (Yao et al., 2013) for GluN3A LBD in both apo and glycine-bound states are shown in Figure 7J–L. Standard deviations of the bottleneck distances are shown in Figure 8A–H.

Figure 7.

Bottleneck Analysis for AMPA and NMDA receptor LBDs.

(A) Residues defining the bottleneck to the glutamate binding site of GluA2. Access tunnels are shown in blue.

(B, C) Average bottleneck distance at each (ξ₁, ξ₂) overlaid with the corresponding PMF contours for (B) apo and (C) glutamate-bound GluA2 LBDs. Dashed boxes in upper right corners approximately demarcate open conformations (average bottleneck distance ≥ 12 Å).

(D) Residues defining the bottleneck to the glycine binding site of GluN1. Access tunnels are shown in purple.

(E, F) Average bottleneck distance at each (ξ₁, ξ₂) overlaid with the corresponding PMF contours for (E) apo and (F) glycine-bound GluN1 LBDs. Dashed boxes in upper right corners approximately demarcate open conformations (average bottleneck distance ≥ 10 Å).

(G) Residues defining the bottleneck to the glutamate binding site of GluN2A. Access tunnels are shown in blue.

(H, I) Average bottleneck distance at each (ξ₁, ξ₂) overlaid with the corresponding PMF contours for (H) apo and (I) glutamate-bound GluN2A LBDs. Dashed boxes in upper right corners approximately demarcate open conformations (average bottleneck distance ≥ 12 Å).

(J) Residues defining the bottleneck to the glutamate binding site of GluN3A. Access tunnels are shown in black.

(K, L) Average bottleneck distance at each (ξ₁, ξ₂) overlaid with the corresponding PMF contours for (K) apo and (L) glycine-bound GluN3A LBDs. Dashed boxes in upper right corners approximately demarcate open conformations (average bottleneck distance ≥ 10 Å).

Figure 8.

Bottleneck Analysis SDs for AMPA and NMDA receptor LBDs.

(A–H) SD of bottleneck distance at each (ξ₁, ξ₂) overlaid with the corresponding PMF contours for (A) apo GluA2, (B) glutamate-bound GluA2, (C) apo GluN1, (D) glycine-bound GluN1, (E) apo GluN2A, (F) glutamate-bound GluN2A, (G) apo GluN3A, and (H) glycine-bound GluN3A.

Values of ΔG_conf and ΔΔG_conf for GluN1, GluN2A, and GluN3A LBDs are provided in Table 1. Bound glutamate stabilizes the closed-cleft GluA2 LBD by 10.1 kcal/mol relative to its apo state. Bound glycine stabilizes the closed-cleft GluN1 LBD by 2.6 kcal/mol relative to its apo state. Bound glutamate stabilizes the closed-cleft GluN2A LBD by 2.8 kcal/mol relative to its apo state. Bound glycine stabilizes the closed-cleft GluN3A LBD by 1.9 kcal/mol relative to its apo state. The ΔΔG_conf value for the GluD2 LBD is greater than that for the GluA2, GluN1, GluN2A, GluN3A, and GluK2 LBDs. The ΔΔG_conf value for the GluA2 LBD is greater than that for the GluN1, GluN2A, GluN3A, and GluK2 LBDs. These differences are attributable to the high ΔG_conf values of agonist-bound GluD2 and GluA2 LBDs. Mechanistically, this suggests that GluD and AMPA receptor LBDs utilize similar protein-agonist interactions that are not present in NMDA and kainate receptor LBDs to significantly stabilize closed-cleft conformations.

DISCUSSION

To characterize the structural thermodynamics of GluD2 LBDs, we computed conformational free energy landscapes for the GluD2 LBD in apo and D-serine-bound states. Unexpectedly, we found via ligand accessibility analysis that the conformational free energy attributable to agonist binding in GluD2 LBD is significantly greater than that to agonist binding in GluA2, GluN1, GluN2A, GluN3A, and GluK2 LBDs. Corroborating LBD transplantation studies demonstrating that the GluD2 LBD functions differently than other iGluR LBDs (Schmid et al., 2009), our results indicate that the unique properties harbored by the GluD2 LBD may in part be due to distinct structural thermodynamic mechanisms. Force produced from agonist-induced LBD closure typically drives iGluR activation and ion channel opening, thus the finding that GluD2 LBD exhibits a significantly higher conformational free energy of agonist-binding is surprising and excludes insufficient force generated from agonist-induced LBD closure as a reason for the absence of GluD2 ion channel activity. Furthermore, our findings suggest that the conformational rearrangements in the ATD, TMD, and/or CTD of GluD2 receptors for non-ionotropic functions in aggregate require more work than ion channel pore opening.

MD simulations of the full-length GluD2 receptor are needed to elucidate the structural details of non-ionotropic conformational changes and further probe why there is a lack of ion channel activity. Recently, a cryo-EM structure of the full-length GluD2 receptor was reported (Burada et al., 2020a). Our conclusion that non-ionotropic functions in aggregate require more work than ion channel pore opening assumes, based on how other iGluRs function, that force is indeed transmitted to the ATD, TMD, and/or CTD. However, it is possible in the full-length GluD2 receptor that the force transfer is impeded at the “extra-LBD” level, for example by a linker region or dimer interface. Our finding that GluD2 LBD generates relatively high conformational free energy could support a “distortion-inactivation” model whereby a sustained force may be too large. This model may explain the experimental observation that D-serine binding inactivates spontaneous ion channel conductance in GluD2 receptors containing the Lurcher mutation (Naur et al., 2007). Alternatively, it remains possible that the conformational free energy generated by GluD2 LBD is insufficient for pore opening due to an absence of gating. GluD ion channel activity is gated by metabotropic glutamate receptor signaling in certain cells by an unknown mechanism (Ady et al., 2014; Benamer et al., 2018; Gantz et al., 2020).

Because numerous questions regarding the physiological role and pharmacology of GluD receptors remain unanswered, the identification of novel ligands for the GluD2 LBD is critical both for elucidating molecular mechanisms and developing therapeutic avenues. For example, 7-chlorokynurenic acid was found to bind the GluD2 LBD with distinct conformational, thermodynamic, and functional properties (Kristensen et al., 2016). Computing conformational free energy landscapes of GluD2 LBD bound to novel ligands may further illuminate structure-function relationships exhibited by the full-length GluD2 receptor. For example, cryo-EM structures of full-length GluD1 and GluD2 receptors bound to 7-chlorokynurenic acid were recently reported (Burada et al., 2020b, 2020a).

Relating ligand accessibility as a function of cleft closure reveals that D-serine stabilizes the closed-cleft GluD2 LBD more than glutamate or glycine stabilize closed-cleft AMPA, NMDA, and kainate receptor LBDs. It is important to note that the conformational free energies computed here constitute only a partial contribution to the total free energy of agonist binding because agonist-docking contributions were not considered (Lau and Roux, 2011). Previously, isothermal titration calorimetry (ITC) measurements of D-serine binding to the GluD2 LBD have obtained a ΔG of approximately −4 kcal/mol (Naur et al., 2007; Tapken et al., 2017). Because we computed a conformational free energy contribution of −14.4 kcal/mol upon D-serine binding to the GluD2 LBD, the ligand-docking contribution is likely endergonic and suggests that D-serine preferentially stays in bulk solvent.

The PMFs we computed for the clinically significant GluD2 variant R710W demonstrates a selective stabilization of more closed-cleft conformations in the apo state, ultimately resulting in a lower conformational free energy attributable to D-serine binding. Given the distant location of R710 from the inner cleft and lack of apparent interactions that directly affect cleft closure, our results suggest that long-range allosteric mechanisms may thermodynamically govern conformations sampled by the GluD2 LBD. Interestingly, experimental evidence also suggests that allosteric mechanisms can regulate the structural thermodynamics of the GluD2 LBD. Mutating binding pocket residues have little effect on the ΔG, measured by ITC, of D-serine binding to the GluD2 LBD, whereas exchanging the hinge region of the GluD2 LBD with that of the GluN1 LBD significantly lowers ΔG (Tapken et al., 2017). Ultimately, in vivo studies are required to ascertain the pathogenicity of this variant.

STAR METHODS

Resource Availability

Lead Contact

Further information and requests for resources and reagents related to this work should be directed to and will be fulfilled by the Lead Contact, Albert Y. Lau (alau@jhmi.edu).

Materials Availability

This study did not generate new unique reagents.

Data and Code Availability

The datasets supporting the current study have not been deposited in a public repository because of their large size but are available from the Lead Contact on request.

Method Details

Molecular Dynamics Simulations

Methods employed in this study are similar to those we have previously used to characterize the structural thermodynamics of other iGluR LBDs (Lau, 2019; Lau and Roux, 2007, 2011; Wied et al., 2019; Yao et al., 2013; Yu et al., 2016a, 2016b). Apo GluD2 LBD (PDB: 2V3T, chain B) and D-serine bound GluD2 LBD (PDB: 2V3U) were used (Naur et al., 2007). MD simulations were performed using CHARMM (Brooks et al., 2009), NAMD (Phillips et al., 2005), and the TIP3P water model (Jorgensen et al., 1983). Residue numbering refer to the mature polypeptide sequence (without signal peptide) (Sobolevsky et al., 2009). All residues not present in the crystal structure were built in using MODELLER (Webb and Sali, 2016), and sidechain conformations were modelled with SCWRL4 (Krivov et al., 2009). D-serine was parameterized by changing the internal coordinate signs of L-serine. Crystallographic waters buried within the LBD were included in the simulations, whereas those situated outside the LBD were excluded. A 90 × 70 × 60 ų orthorhombic water box with ~150 mM NaCl was used for all simulations, and the net charge of the system was set to zero by adjusting the number of ions (WT: Na⁺ = 33, Cl⁻ = 26; R710W: Na⁺ = 33, Cl⁻ = 25). For simulations of D-serine-bound GluD2 LBDs, the ligand was confined within the binding pocket by a half-harmonic restraint that kept the α-carboxyl oxygens of the ligand within 2.8 Å of the guanidine group of R507 (PDB numbering: 92). Initial conformations for umbrella sampling, based on the two-dimensional order parameter (ξ₁, ξ₂), described below, were generated using targeted MD (Lau and Roux, 2007). Equilibration was conducted in the NVT ensemble with applied restraints, which were gradually released, on backbone and sidechain atoms. Production runs were conducted in the NPT ensemble at 1 atm and 300 K using the method of the Langevin piston (Feller et al., 1995). Electrostatic interactions were calculated via the particle mesh Ewald (PME) algorithm (Essmann et al., 1995).

Quantification and Statistical Analysis

Free Energy Computations

Umbrella sampling was performed using a two-dimensional order parameter, (ξ₁, ξ₂), which specifies distances between the top and bottom lobes of the LBD. The top lobe was defined as residues 417–516 and 748–790 (PDB numbering: 1–101 and 223–265; includes an extra N-terminal glycine present in the structure), and the bottom lobe was defined as residues 522–528 and 641–743 (PDB numbering: 107–218; includes a Gly-Thr linker present in the structure). The first coordinate, ξ₁, is defined as the distance between the center-of-mass of all non-hydrogen atoms in residues 501–503 (PDB numbering: 86–88) in the top lobe and the center-of-mass of all non-hydrogen atoms in residues 663–664 (PDB numbering: 138–139) in the bottom lobe. The second coordinate, ξ₂, is defined as the distance between the center-of-mass of all non-hydrogen atoms in residues 426–428 (PDB numbering: 11–13) in the top lobe and the center-of-mass of all non-hydrogen atoms in residues 702–703 (PDB numbering: 177–178) in the bottom lobe. A force constant of 2 kcal/mol/Ų was used for ξ₁ and ξ₂. The order parameter spans the relevant regions of conformational space surrounding the open- and closed-cleft conformations observed in crystal structures (Lau and Roux, 2007).

233 windows were utilized for umbrella sampling. Each window was sampled for 1.4 ns for a total of 326.2 ns. Umbrella sampling distribution functions were unbiased and recombined by the weighted histogram analysis method (WHAM) (Kumar et al., 1992; Souaille and Roux, 2001) to obtain the PMFs.

1D projections of the 2D PMFs were generated by computing the PMF along a hybrid order parameter, ξ₁₂, that is an average of ξ₁ and ξ₂. Each ξ₁ and ξ₂ pair was mapped to a bin in ξ₁₂ with a width of 0.1 Å, and the values from the 2D unbiased distribution in each bin was summed to create a 1D unbiased distribution. The 1D PMF, W(ξ₁₂), is given by

$$W\left({\xi }_{12}\right)=-k_{B}T\text{ln}\left(\frac{\rho \left({\xi }_{12}\right)}{{\sum }_{j=1}^N{\rho }_j\left({\xi }_{12}\right)}\right)$$

where ρ(ξ₁₂) is the 1D unbiased distribution function, and ${\sum }_{j=1}^N{\rho }_j\left({\xi }_{12}\right)$ is the partition function.

Uncertainty in the 2D PMFs was calculated by block averaging (Grossfield and Zuckerman, 2009). The simulation trajectory for each umbrella sampling window was divided into ten equally sized “blocks,” and WHAM was performed on each block to generate ten block PMFs. The standard deviation of the block PMFs was computed. Uncertainty in each 1D PMF was calculated based on the computed standard deviation (SD) in the 2D PMF, i.e., 100 randomly generated 2D PMFs, based on the 2D SD, were projected to 1D PMFs, and the 1D SD was calculated.

Structures presented in this study were prepared in PyMOL, and plots were generated in Gnuplot (www.gnuplot.info). Sequence alignment was performed using manually annotated UniProt entries via Clustal Omega (Sievers et al., 2011).

Highlights.

• D-serine binding to the GluD2 receptor strongly drives ligand-binding domain closure

• A neurodegenerative mutation in GluD2 stabilizes activated states of the LBD

• Ligand accessibility analysis quantifies LBD conformational thermodynamics