l-Serine dietary supplementation is associated with clinical improvement of loss-of-function GRIN2B-related pediatric encephalopathy

See allHide authors and affiliations

Science Signaling  18 Jun 2019:
Vol. 12, Issue 586, eaaw0936
DOI: 10.1126/scisignal.aaw0936

Treating NMDA receptor deficiency with a dietary supplement

Patients with Rett-like syndrome show impaired neuronal, motor, cognitive, and social development. Soto et al. studied the etiology and treatment of the disorder in a 5-year-old patient who had a mutation in GRIN2B, the gene encoding the NMDA receptor subunit GluN2B. Expressing this mutant in cultured neurons impaired both electrophysiological activity and dendritic morphology, but these features were improved with the application of the known NMDA receptor agonist, d-serine. d-Serine administration has some toxicity in rodents, but its stereoisomer l-serine is a natural, nonessential amino acid found in various foods. Adding l-serine powder to the patient’s food or drink over the course of more than a year increased the amount of d-serine in the plasma and cerebrospinal fluid and markedly improved all neurodevelopment assessments, suggesting that this rather simple treatment might be extended to various patients with disorders associated with NMDA receptor deficiency.


Autosomal dominant mutations in GRIN2B are associated with severe encephalopathy, but little is known about the pathophysiological outcomes and any potential therapeutic interventions. Genetic studies have described the association between de novo mutations of genes encoding the subunits of the N-methyl-d-aspartate receptor (NMDAR) and severe neurological conditions. Here, we evaluated a missense mutation in GRIN2B, causing a proline-to-threonine switch (P553T) in the GluN2B subunit of NMDAR, which was found in a 5-year-old patient with Rett-like syndrome with severe encephalopathy. Structural molecular modeling predicted a reduced pore size of the mutant GluN2B-containing NMDARs. Electrophysiological recordings in a HEK-293T cell line expressing the mutated subunit confirmed this prediction and showed an associated reduced glutamate affinity. Moreover, GluN2B(P553T)-expressing primary murine hippocampal neurons showed decreased spine density, concomitant with reduced NMDA-evoked currents and impaired NMDAR-dependent insertion of the AMPA receptor subunit GluA1 at stimulated synapses. Furthermore, the naturally occurring coagonist d-serine restored function to GluN2B(P553T)-containing NMDARs. l-Serine dietary supplementation of the patient was hence initiated, resulting in the increased abundance of d-serine in the plasma and brain. The patient has shown notable improvements in motor and cognitive performance and communication after 11 and 17 months of l-serine dietary supplementation. Our data suggest that l-serine supplementation might ameliorate GRIN2B-related severe encephalopathy and other neurological conditions caused by glutamatergic signaling deficiency.


Rett syndrome (RTT; Online Mendelian Inheritance in Man, OMIM: 312750) is a neurodevelopmental disorder affecting 1 in 10,000 live female births (1, 2). Clinical manifestations include microcephaly, loss of achieved psychomotor abilities, intellectual disability (ID), and autistic behaviors (3). Whereas most cases of typical RTT harbor loss-of-function mutations in the X-linked gene encoding methyl-CpG–binding protein 2 (MECP2) (4), mutations in genes encoding cyclin-dependent kinase–like 5 (CDKL5) and Forkhead box G1 (FOXG1) have also been identified (5). Rett-like syndrome mostly affects patients exhibiting symptoms that are similar to those seen in patients with RTT; however, the genetic and molecular etiologies of this rare disease are different from those associated with RTT.

A growing number of genetic and functional studies are unraveling the complex scenario and molecular players involved in neurodevelopmental disorders and, in particular, in Rett-like syndrome. It has been shown that the dysregulation of synaptic proteins can lead to neurodevelopmental disorder (68). Genes encoding for the N-methyl-d-aspartate receptor (NMDAR) could play critical roles in the dysfunction of glutamatergic transmission associated with RTT. Functionally, NMDARs play critical roles in neurogenesis, synaptogenesis, and synaptic plasticity processes. Early in development, NMDAR subunit GluN2B expression is particularly high (9). Accordingly, it has been proposed that GRIN2B gene disturbance might markedly compromise critical steps of neuronal, synaptic, and brain circuitry development (10). Moreover, discrete de novo mutations of GRIN2B gene have been associated with neurodevelopmental disorders (1113) such as early infantile epileptic encephalopathy-27 (EIEE27; OMIM: 616139) (14) and autosomal dominant mental retardation (MRD6; OMIM: 613970) (1517).

In this study, we investigated the effects of a de novo missense mutation in the GRIN2B gene in a patient with Rett-like syndrome and severe encephalopathy. Functional studies showed that channel gating is altered in mutant NMDARs markedly reducing NMDAR-mediated currents. Dietary supplementation with l-serine—the precursor of d-serine, an endogenous NMDAR coagonist (18, 19)—during 17 months was associated with ameliorated intellectual, communication, and motor deficits in the patient. These results support the pathogenicity of GRIN2B mutation and suggest that enhancing NMDAR activity using l-serine dietary supplementation can have therapeutic benefits in certain neurodevelopmental disorders associated with NMDAR hypofunctionality.


Patient clinical symptomatology and genetic studies

The patient was a girl born after an uneventful pregnancy and with no family history of neurodevelopmental disorders. She was referred to our clinic at 1 year old, and the primary clinical examination showed a psychomotor delay with severe hypotonia (with the presence of osteotendinous reflexes and devoid of pyramidal signs) and an inability to hold up her head and to sit upright. Behaviorally, she had an overall “absence,” as manifested by a poor visual contact, an impairment in social interaction, and no interest in the environment. Along with these alterations, the patient showed high irritability with sleep disturbances. Considering these symptoms, together with the presence of “hand-washing” stereotypies, the girl was tentatively diagnosed with RTT rather than Rett-like phenotype. Cytogenetic analysis, brain magnetic resonance imaging, and neurometabolic analysis did not show abnormalities. At 2.5 years old, she was less irritable and had developed the capacity to hold up her head. Behaviorally, she had slightly improved social interaction, and 1 year later, her sleeping pattern was ameliorated. At that age (3.5 years old), an electroencephalogram (EEG) indicated the presence of epileptiform alterations of brain activity, she was treated with valproic acid, and later, the treatment was changed to levetiracetam to prevent changes in irritability. At 5 years and 10 months old, the patient’s adaptive behavior was assessed by the Vineland test, with scores indicative of a mental age below 1 year old (Fig. 5C).

As noted, the patient was tentatively diagnosed with RTT-like phenotype. Because no mutations of RTT candidate genes (MECP2, CDKL5, and FOXG1) were detected, whole-exome sequencing was performed (20). After genetic data filtering against parental variants and then against a pool of controls, we identified a de novo heterozygous missense mutation in GRIN2B gene coding for the GluN2B subunit of NMDARs, resulting in an amino acid substitution of a proline (Pro) residue by a threonine (Thr) at GluN2B subunit residue 553 (Fig. 1A).

Fig. 1 Identification of GRIN2B(P553T) mutation associated with the case study and predicted structural consequences.

(A) Left: Familial pedigree of the GRIN2B(P553T) case study. Right: Chromatograms of GRIN2B(c.1657C > A) mutated nucleotide (indicated by an arrow) using forward and reverse oligonucleotides. (B) Structure of heterotetrameric (GluN1)2-(GluN2B)2 NMDAR [Protein Data Bank (PDB) ID: 4PE5], according to Karakas and Furukawa (61), showing the N-terminal domain (NTD), the ligand-binding domain (LBD), and the transmembrane domain (TMD; containing the mutated amino acid P553T). Bottom: Magnification of the transmembrane domain, showing the topological position of Pro553 residue at the beginning of the M1 (P553; green) of the transmembrane domain, facing Phe653 residue (F653; light blue) at M3 (blue). (C) Top: Transmembrane domain structural molecular model of wild-type (GluN1)2-(GluN2B)2 receptor (from the extracellular domain). Inset: Magnification of residues Pro553(M1)-F653(M3) proximity. Bottom: Transmembrane domain structural molecular model of mutant (GluN1)2-[GluN2B(P553T)]2 receptor (from the extracellular domain). Inset: Magnification of residues Pro553(M1)-F653(M3) distance. (D) GluN2B protein sequence alignment around residue Pro553. Representative sequences from a larger alignment containing 147 proteins from 12 metazoan species spanning seven phyla are shown. Displayed protein sequences are from the following species: Homo sapiens GluN2B (UniProt ref. Q13224), Mus musculus GluN2B (UniProt ref. Q01097), Danio rerio GluN2Bb (Ensembl ref. ENSDARG00000030376), Branchiostoma belcheri 254360R.t1 (from the database B.belcheri_v18h27.r3_ref_protein included in LanceletDB Genome browser; Sun Yat-sen University), Saccoglossus kowalevskii Sakowv30010297m (Metazome database), and Strigamia maritima SmarNMDAR2b, Apis melifera GB48097, Capitela teleta CapteT179505, and Lottia gigantea LotgiT137890 (all from Ensembl Metazoa).

Molecular modeling of mutant (GluN1)2-[GluN2B(P553T)]2 receptor

To identify the structural changes induced by GluN2B(P553T) mutation, molecular models of wild-type and mutant receptors were generated by homology modeling (using crystal structures of NMDAR at closed state) and molecular dynamics simulation. The Pro553 residue in GluN2B is located at the beginning of the transmembrane helix 1 (M1; Fig. 1B). This model suggested a role for Pro553 in breaking M1 and bending it toward M3 (Fig. 1C, top), thereby enabling the interaction of Pro553 with Phe653 (M3) of the GluN2B subunits. In addition, according to this model, the residue Pro557 (M1) is also bending M1 to M3 through its Pro kink, allowing the interaction with Phe654 (M3) of the GluN1 subunit. This model predicted that the P553T mutation would prevent M1 bending, disrupting M1-M3 interaction (Fig. 1C, bottom), which in turn would bring M3 closer to the center of the pore and consequently induce a more closed channel conformation, altering the gating properties of the mutant receptor.

On the basis of this structural model, we investigated the evolutionary conservation of the GluN2B Pro553 residue and its predicted interacting residue Phe653. We found that the Pro553 motif was highly conserved across species (Fig. 1D). The GluN2B Phe653 and GluN1 Phe654 residues are within the SYTANLAAF motif (fig. S4), the most highly conserved motif among mammalian ionotropic glutamate receptors (iGluRs) (21, 22). Multiple sequence alignments of 147 metazoan iGluRs showed a high conservation of these residues (the Pro553 and Phe653 residues were detected in 144 and 131 iGluR protein–encoding gene sequences, respectively; fig. S4), supporting a potentially critical role of the Pro553 residue in GluN2B in NMDAR channel activity.

Heteromerization and trafficking of GluN1-GluN2B(P553T) receptors

To assess the effects of GluN2B(P553T) on the oligomerization and trafficking of NMDARs, we cotransfected human embryonic kidney (HEK) 293T cells with GluN1 and hemagglutinin (HA)–tagged GluN2A, together with either green fluorescent protein (GFP)–tagged wild-type GluN2B (GFP-GluN2Bwt) or the P553T mutant [GFP-GluN2B(P553T)]. Biochemical analysis showed that the protein abundance of GluN2B(P553T) was similar to that of GluN2Bwt in these cells (fig. S1A). Further, coimmunoprecipitation experiments showed the presence of GluN1 and HA-GluN2A subunits in anti-GFP pulldown complexes, indicating that the mutant GluN2B(P553T) subunit interacted—in similar abundance as GluN2Bwt—with GluN1 and/or GluN2A subunits (fig. S1B). Immunofluorescence analysis showed that the missense mutation on Pro553 does not abolish the trafficking of GluN1-GluN2B(P553T) to the surface of COS-7–transfected cells (fig. S1C), as previously reported for another GRIN2B missense variant affecting the same amino acid position, GluN2B(P553L) (23). In primary cortical murine neurons, the dendritic surface:intracellular abundance ratio of transfected GluN2B(P553T) was normal at days in vitro 7 (DIV7) and DIV11, with a slight decrease in DIV16 (fig. S1D).

Biophysical assessment of GluN2B(P553T) subunit–containing NMDARs

We performed patch-clamp experiments to evaluate the biophysical properties of GluN2B(P553T)-containing NMDARs in transfected HEK-293T cells. After a fast glutamate (1 mM) and glycine (50 μM) application, NMDAR-mediated current amplitudes were significantly reduced in HEK-293T cells expressing GluN1-GluN2B(P553T) receptors Fig. 2, A to C), whereas voltage-dependent channel blockade by extracellular Mg2+ was spared (Fig. 2, D to F). Because the P553T mutation is located in the vicinity of the agonist binding site and the channel pore in GluN2B, we explored the possible effects of the mutation on channel kinetics. Electrophysiological recordings showed a significantly faster deactivation rate in mutant receptors (Fig. 2, G and H) and a faster desensitization thereof, quantified upon 5-s duration jumps (Fig. 2, I and J). Moreover, in agreement with aforementioned modeling predictions, nonstationary fluctuation analysis (NSFA) (24) showed a reduction of the single-channel conductance in mutant receptors (Fig. 2, K to M) and a reduced open probability (Fig. 2N). Together with the biochemical data indicating normal expression and oligomerization, we concluded that GluN1-GluN2B(P553T) receptors are intrinsically hypofunctional.

Fig. 2 Biophysical characterization of (GluN1)2-(GluN2B(P533T))2channel properties.

(A) Representative whole-cell currents evoked by rapid application of 1 mM glutamate + 50 μM glycine (0.5-s duration; −60 mV) in HEK-293T cells expressing GluN1-GluN2B (black trace) or GluN1-GluN2B(P533T) (red trace) receptors. n = 19 and 21 cells from six and five experiments, respectively. (B) Average of raw peak currents from HEK-293T cells expressing GluN2B and GluN2B(P533T). n = 19 and 21 cells from six and five experiments, respectively. ***P < 0.001 by Mann-Whitney U test. (C) Normalized peak currents (in pA/pF) in HEK-293T cells expressing GluN1-GluN2B and GluN1-GluN2B(P533T), with values from a representative experiment superimposed. Data are from six and five experiments, respectively. ****P < 0.0001 by Mann-Whitney U test. (D) Traces recorded at −60 mV in an HEK-293T cell expressing GluN1-GluN2B(P533T) with Mg2+ block of the NMDAR. Data are representative of five and seven cells from three independent cultures. (E) Percentage of current blocked at −60 mV by Mg2+ (1 mM) for GluN2Bwt- and GluN2B(P533T)-containing NMDARs. Single-value experiments are denoted as open circles for each condition. n = 5 and 7 cells, respectively, from three independent experiments per condition. n.s. (not significant) by Mann-Whitney U test. wt, wild-type. (F) Current-voltage relationship for GluN2B- and GluN2B(P533T)-containing NMDARs. n = 3 and 4, respectively, from two independent experiments. (G) Representative peak-scaled responses to 1 mM glutamate + 50 μM glycine (0.5-s agonists application; −60 mV) for GluN1-GluN2B (black trace) and GluN1-GluN2B(P553T) (red trace). n = 16 and 17 cells from six and five experiments, respectively. (H) Average deactivation time constant (τw; fitted to a double exponential) fitted from tail currents for GluN1-GluN2B and GluN1-GluN2B(P553T). Values from a representative experiment are shown as open circles for each condition. n = 16 and 17 cells from six and five independent experiments per condition, respectively. ****P < 0.0001 and n.s. by Mann-Whitney U test. (I) Representative peak-scaled responses to 1 mM glutamate + 50 μM glycine (long jumps of 5-s duration; −60 mV) in HEK-293T cells expressing GluN1-GluN2B or GluN1-GluN2B(P553T), for the comparison of desensitization rates. n = 14 cells from three independent experiments. (J) Desensitization weighted time constant (τw) for GluN2Bwt and GluN2B(P553T). Values from a representative experiment are shown as open circles for each condition. n = 14 from three independent experiments. **P < 0.01 by Mann-Whitney U test. (K and L) Whole-cell currents activated by rapid application of 1 mM glutamate + 50 μM glycine (0.5 s; −60 mV) from HEK-293T cells expressing GluN1-GluN2Bwt (K) or GluN1-GluN2B(P553T) (L). Gray traces represent single responses, and black lines are the average of 69 (wild-type) or 33 (P553T) responses. Insets: Current variance versus mean current plot calculated from the deactivating tail current. (M and N) Bar graph showing single-channel conductance values (M) and peak open probability (N) in GluN1-GluN2Bwt– and GluN1-GluN2B(P553T)–containing NMDARs expressed in HEK-293T cells. n = 12 and 9 cells, respectively, from four independent experiments. *P < 0.05 by Mann-Whitney U test. Single cells are shown as open circles superimposed to bar graph.

Because GluN2B subunits can assemble into both (GluN1)2-(GluN2B)2 heterodimers and (GluN1)2-GluN2A-GluN2B heterotrimers (25), we explored whether the GluN2B(P553T) mutation may also impair heterotrimeric receptors in HEK-293T cells (26). NMDA current amplitudes were not significantly reduced in GluN2B(P553T)-containing heterotrimers, although their deactivation and desensitization rates were increased similar to those of GluN1-GluN2B(P553T) heterodimers (fig. S2, A and B). However, administration of 100 μM d-serine potentiated GluN1-GluN2A-GluN2B(P553T) heterotrimers more strongly than wild-type heterotrimers (fig. S2, A to C). Likewise, the desensitization and deactivation kinetics of triheteromeric mutants were increased compared with controls (fig. S2, D to H), recapitulating some of the effects observed in the mutant (GluN1)2-[GluN2B(P553T)]2 diheterodimers.

Evaluation of d-serine effect in GluN1-GluN2B(P553T) receptors

Next, we sought to enhance the activity of mutant NMDARs using d-serine, an endogenous NMDAR coagonist. In agreement with a previous report (19), d-serine administration dose-dependently increased NMDAR-mediated currents in HEK-293T cells (Fig. 3, A and B). The relative increase mediated by d-serine was stronger in HEK-293T cells expressing GluN1-GluN2B(P553T) than in cells expressing GluN1-GluN2Bwt (Fig. 3, A and B). Although not reaching GluN1-GluN2Bwt–mediated current density, hypofunctional GluN1-GluN2B(P553T) receptor–mediated currents were significantly increased in d-serine coapplication (Fig. 3C). Further, a similar increase was observed in the presence of a high dose of glycine (100 μM), as well as with a stronger potentiation in GluN1-GluN2B(P553T)–expressing cells compared with GluN1-GluN2Bwt–expressing HEK-293T cells (Fig. 3, D to E). The differential potency of d-serine and glycine might result from a reduced affinity for GluN1-GluN2B(P553T), leading to the enhanced potentiation and faster deactivation/desensitization rates at higher concentrations. Alternatively, because glutamate binding increases the dissociation rate of glycine/d-serine coagonist with NMDARs (27), these changes in d-serine potency might be explained by altered glutamate affinity. Concentration-response experiments showed no changes in d-serine median effective concentration (EC50; Fig. 3F), whereas glutamate EC50 increased by sevenfold in GluN1-GluN2B(P553T) receptors (Fig. 3G).

Fig. 3 d-Serine coapplication effect on wild-type and mutant (GluN1)2-[GluN2B(P533T)]2NMDARs.

(A) Representative traces evoked by physiological concentrations of 1 mM glutamate + 1 μM glycine (0.5 s; −60 mV) from GluN1-GluN2Bwt– or GluN1-GluN2B(P553T)–expressing HEK-293T cells, either in the absence (black traces) or in the presence (red traces) of d-serine at different concentrations. n ≥ 6 cells from at least two independent experiments. (B) Average peak current evoked in transfected HEK-293T cells by application of 1 mM glutamate + 1 μM glycine in the presence of different d-Ser concentrations (gray bars) normalized to that of 1 mM glutamate + 1 μM glycine without d-Ser (white bars). Numbers inside the bars denote the recordings for each condition, from at least two independent experiments. *P < 0.05 and **P < 0.01 by Mann-Whitney U test. (C) Raw peak current responses from data shown in (B), indicating the percentage increase in current due to 100 μM d-Ser. n = 10 cells from at least two experiments. (D) Representative recordings in HEK-293T cells expressing GluN1-GluN2Bwt or GluN1-GluN2B(P553T) receptors evoked by physiological applications (5 s; −60 mV) of 1 mM glutamate + 1 or 100 μM glycine. n = 8 cells per condition from five independent experiments. (E) Bar graph representing peak current potentiation induced by high glycine concentration (100 μM versus 1 μM) with coapplication of 1 mM glutamate. n = 8 from five independent experiments for each condition. *P < 0.05 and n.s. by Student’s t test. (F) d-Serine concentration-response curves recorded in GluN1-GluN2Bwt and GluN1-GluN2B(P553T) NMDARs expressed in HEK-293T cells, constructed from responses to 1 mM glutamate + the specified d-serine concentration in the absence of glycine. n = 5 cells per construct from two independent experiments. (G) Glutamate concentration-response curves recorded in GluN1-GluN2Bwt and GluN1-GluN2B(P553T) NMDARs expressed in HEK-293T cells. The dose-response curve was constructed from responses to 1 mM d-serine plus the specified glutamate concentration in the absence of glycine. n = 5 and 7 cells, respectively, from two independent experiments.

Evaluation of GluN2B(P553T) variant neuronal outcomes

To evaluate the neuronal impact of GluN2B(P553T), we overexpressed GluN2B(P553T) in primary hippocampal neurons and measured morphological parameters, synaptic plasticity processes, and NMDAR-mediated synaptic currents. Sholl analysis of dendrites labeled with GFP-GluN2B(P553T) or GFP-GluN2Bwt indicated similar distributions of the subunit across the dendritic arbor (Fig. 4A). Nevertheless, spine density was significantly reduced in neurons expressing GluN2B(P553T), resulting from a decrease in the different spine subtypes (Fig. 4A). Further, immunofluorescence analysis of the GluA1 AMPA receptor (AMPAR) subunit, which is overexpressed in RTT murine models (28), revealed a significant increase in GluA1 in DIV11 neurons overexpressing GluN2B(P553T) (fig. S3). Overall, these morphological and molecular changes indicated deficient spine development in hippocampal neurons expressing GluN2B(P553T).

Fig. 4 Synaptic outcomes and d-serine effects on Gly-induced chemical long-term potentiation (Gly-cLTP) in GluN2B(P553T)-expressing primary hippocampal neurons.

(A) Top left: Representative images of murine primary hippocampal neurons transfected with GFP-GluN2Bwt or GFP-GluN2B(P553T). Insets: Immunodecoration to visualize spines, indicated by yellow arrowheads. Top right: Immunofluorescence detection of GFP-GluN2B to analyze dendritic arborization by Sholl analysis. n = 16 to 18 neurons per condition from three independent experiments; F = 0.71, P = 0.884 by repeated measures two-way analysis of variance (ANOVA) and Bonferroni post hoc test. Bottom: Quantification of spine density and morphology in basal neurons and neurons treated with glycine alone or glycine and 100 μΜ d-Ser. n = 27 to 48 dendrites per condition from three independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001, and n.s. by Student’s t test or Mann-Whitney U test for parametric or nonparametric analyses, respectively. (B) Representative traces from spontaneous activity–dependent NMDAR-mediated EPSCs recordings from GluN2Bwt- or GFP-GluN2B(P553T)–transfected murine hippocampal neuronal cultures, recorded at −70 mV in the presence of 50 μM NBQX (2,3-dihydroxy-6-nitro-7-sulfamoylbenzo[f]quinoxaline 6-nitro-7-sulfamoylbenzo[f]quinoxaline-2,3-dione) and 10 μM picrotoxin and in the absence of tetrodotoxin, both under basal conditions (top traces) or after 100 μM d-serine application (bottom traces). Graphs show mean amplitudes (left) and mean time of interevent intervals (right) of the EPSCs recorded. n = 8 and 6 neurons, respectively, from three independent experiments. **P < 0.01 and #P < 0.05 by Mann-Whitney U test. (C) Representative images and immunofluorescence analysis of surface abundance of the AMPAR subunit GluA1 (red) in murine primary hippocampal neurons (green) at DIV16 that had been transiently transfected at DIV11 with GFP-GluN2Bwt or with GFP-GluN2B(P553T) assessed under basal conditions, after Gly-cLTP, and after simultaneous Gly-cLTP induction and 100 μM d-serine application (Gly + d-Ser cLTP). A.U., arbitrary units. Data are presented relative to the basal condition as means ± SEM from n = 30 to 40 spines per dendrite from 14 to 40 dendrites and three to seven neurons per condition, obtained from three independent experiments. **P < 0.01 and ***P < 0.001 by Student’s t test.

Patch-clamp recordings revealed a decrease in the amplitude of spontaneous excitatory postsynaptic currents (EPSCs) mediated by NMDARs in neurons overexpressing mutant GluN2B(P553T) compared with GluN2Bwt (Fig. 4B, top traces and left bar graph), directly demonstrating an effect of this mutation in synaptic NMDARs. Because d-serine administration enhanced the activity of GluN2B(P553T)-containing NMDARs in heterologous cells (see above), we assessed its effect in primary neurons. The addition of 100 μM d-serine similarly increased EPSCs frequency in GluN2Bwt- and GluN2B(P553T)-overexpressing neurons (measured as a shortening of the interevent interval; Fig. 4B, bottom traces and right bar graph). In contrast to heterologous cell lines data, 100 μM d-serine addition did not increase EPSC amplitudes (Fig. 4B, left bar graph), perhaps due to the recruitment of new synapses or increased desensitization at higher frequencies, thus masking a putative effect on EPSC amplitudes. The rate of recovery from desensitization for NMDARs is quite slow, spanning several seconds (29).

Assessment of the effect of high-dose d-serine in GluN2B(P553T)-expressing neurons

Sustained activation of synaptic NMDARs modifies postsynaptic biochemical content (synaptic recruitment of AMPARs), induces morphological changes (synapse enlargement), and elicits LTP (3032). Therefore, we assessed the ability of GluN2B(P553T)-expressing neurons to support this form of synaptic plasticity. Transfected primary hippocampal cultures were treated either with 200 μM glycine (“Gly-cLTP”) or simultaneously with 200 μM glycine and 100 μM d-serine (“Gly + d-Ser cLTP”). The analysis of dendritic spines density showed a slight significant increase in Gly-cLTP condition (Fig. 4A). Coapplication of 100 μM d-serine resulted in a more significant increase in spine density in GluN2B(P553T)-expressing neurons, with a number of stubby, mushroom, and thin-shape spines similar to neurons expressing GluN2Bwt under basal conditions (Fig. 4A). Immunofluorescence analysis showed that Gly-cLTP significantly increased surface GluA1 abundance at spine-like structures in GluN2Bwt-transfected neurons (Fig. 4C, left images and bar graph). In contrast, GluA1 surface abundance was not significantly increased in GluN2B(P553T)-expressing neurons (Fig. 4C). These defects were rescued by simultaneous administration of glycine and d-serine (Fig. 4C), indicating that d-serine can also facilitate a major potentiation of NMDAR-dependent plasticity in the context of GluN2B(P553T) mutation.

Evaluation of biochemical and clinical effects of l-serine dietary supplementation in pediatric patient with GluN2B(P553T) variant

The partial restoration of mutant NMDAR function by d-serine prompted us to translate the therapeutic strategy to the clinical practice. Because d-serine use was still under investigation at the beginning of the study, the patient was supplemented with l-serine, an approved nutraceutical amino acid acting as the endogenous d-serine precursor. We hypothesized that l-serine supplement might increase serine racemase substrate in the brain (33, 34), raising d-serine brain levels that might potentiate hypofunctional NMDARs. At 5 years and 10 months of age, the patient was administered l-serine (500 mg/kg per day) and continued to date (17 months at the time of writing this report). Ultraperformance liquid chromatography–tandem mass spectrometry (UPLC-MS/MS) analysis showed that l-serine treatment was associated with a 4.4-fold increase in the abundance of d-serine in the patient’s plasma (Fig. 5A), whereas d-enantiomers of branched-chain amino acids (valine, isoleucine, and leucine) and plasma amino acids with a potential NMDAR modulatory role (such as glycine, taurine, and cysteine) showed no notable alterations (Fig. 5A). Furthermore, considering that l-serine is involved in sphingolipid biosynthesis, an untargeted UPLC-MS lipidomic analysis was conducted and revealed a strong alteration of the sphingolipidomic profile of the patient, which was not altered by l-serine dietary supplementation (table S1). Because d-serine racemase is strongly expressed in neurons (33), we reasoned that the increased d-serine plasma levels most probably reflect increased d-serine abundance in the brain of the patient. UPLC-MS analysis showed highly increased d-serine levels in cerebrospinal fluid (CSF) of the treated patient (Fig. 5B).

Fig. 5 Biochemical and behavioral assessment of GRIN2B(P553T) patient after l-serine dietary supplementation.

(A) Top: Analysis of amino acid plasma concentration of d-l-serine, d-l-valine, d-l-isoleucine, d-l-leucine, glycine, taurine, and cysteine. Plasma samples analyzed correspond to the patient before and after l-serine treatment (light red and dark red dots, respectively) and the mean of controls (black lines; whiskers representation of mean-min-max values). Bottom: Chromatographic profiles of d-serine and l-serine enantiomers in plasma samples from the control individual (gray trace), the patient before treatment (light red trace), and the patient with l-serine dietary supplementation (dark red trace). (B) Top: Overlapping chromatographic profiles of d-serine and l-serine enantiomers (left and right, respectively) in CSF from a control (gray trace) and from the patient with GRIN2B(P553T) mutation, after l-serine dietary supplementation (red trace). Bottom: Serine enantiomers concentration analysis in CSF of controls (gray dots; whiskers representation of mean-min-max values; gray traces in the chromatogram) and the patient with l-serine dietary supplement (red dots and traces). (C) Clinical manifestations of the patient harboring a de novo GRIN2B(P553T) mutation, before and after 17 months of l-serine dietary supplementation.

Together with the biochemical profiling, a neuropsychological assessment was performed. Before treatment, the patient (5 years and 10 months old) had few communication skills, poor eye contact, did not pay attention to the activities of the parents, teachers, and schoolmates, did not have verbal language (any words or sounds), did not show the use of symbolic play, and could not move from the prone position to the sitting position, although she could then sit without support. The Vineland standard score (SS) test before treatment were 42 for communication, 36 for daily living skills, 49 for socialization, and 31 for motor skills (mean SS of 100 with an SD of 15), with an overall SS of 38.

After 11 months of l-serine dietary supplementation, although the EEG was still aberrant, the behavior of the patient was notably improved, as shown by the Vineland Adaptive Behavior Scale (VABS). The Vineland SSs were 42 for communication, 34 for daily living skills, 51 for socialization, and 38 for motor skills. This midterm assessment showed that her motor skills were strongly improved; she was able to stand from sitting and started to walk with external help. Behaviorally, the patient started to communicate with gestures (such as by extending her hands to be held), and her display of facial expressions increased. She started to react by turning her head when called by her name, had improved visual contact, and started to smile proactively. Overall, the clinical assessment indicated an attenuation of both motor and cognitive impairments.

After 17 months of treatment, the patient remarkably improved communication, social, and motor skills (Fig. 5C). The Vineland SSs (motor assessment cannot be performed beyond 7 years old) after l-serine treatment in the patient were 48 for communication, 48 for daily living skills, and 50 for socialization. Behaviorally, she was then interested in faces and had a persistent eye contact, followed the activities of surrounding humans with interest, was able to stretch her arms out to be held, turn her head when called, laugh at funny situations, and was generally (by outward assessments) happier. She could then also imitate toys and animal sounds, creep on the floor, and move from a prone to a sitting position, as well as from a sitting to standing position with support. She had also improved her sleeping pattern, was able to sit down without help, and notably could walk using an orthopedic walker. An EEG showed epileptiform alterations, but clinical seizures were not present. Despite pharmacological interventions to target them specifically (valproate first and levetiracetam later), the epileptiform alterations detected in the EEG were still present; thus, these treatments were stopped. Nonetheless, overall, the clinical assessments indicated an improvement in cognitive, communicative, and motor impairments associated with l-serine dietary supplementation.


NMDARs are critical players of glutamatergic neurotransmission and are fundamental actors in neuritogenesis, synaptogenesis, and synaptic plasticity processes. Currently, upon the development of next-generation sequencing, there is a growing body of data implicating de novo mutations of iGluRs (7, 8) in mental and behavioral disorders. Several mutations in subunits of the NMDAR have been related to neurodevelopmental diseases (1113). However, these data require functional validation to unveil whether these mutations are really pathogenic. In the present study, we identified a GRIN2B(P553T) missense de novo mutation of the GluN2B subunit of the NMDAR in a patient with Rett-like syndrome with severe encephalopathy. This mutation resulted in the exchange of a highly evolutionary conserved Pro into a Thr. A mutation affecting the same amino acid position GluN2B(P553L) was previously described in a patient from a cohort of individuals with ID (16). This patient exhibited phenotypic alterations similar to those of the case in the present study, including severe ID, hypotonia, and no speech. The phenotypic similarity between these two cases provides a strong evidence for a pathogenic role of GluN2B(Pro553) mutations under these neurodevelopmental conditions.

Electrophysiological studies confirmed the in silico structural studies predicting the hypofunctionality of mutant NMDARs. In addition to a reduction of the channel conductance and peak open probability, we found a significant increase in both the desensitization and the deactivation rates of GluN1-GluN2B(P553T) receptors. Whether the P553T mutation located far from the binding site is altering deactivation kinetics, an intrinsic property of the receptor, is an incognita, but electrophysiological concentration-response experiments indicate a decreased glutamate binding affinity of mutant NMDARs, triggering a decreased receptor efficacy that seems to be physiologically relevant. NMDAR function is largely determined by the high amount of Ca2+ influx, which is mostly dependent on channel kinetics, particularly the rates of desensitization and deactivation. Thus, we can speculate that kinetics changes detected on GluN2B(P533T) mutant receptors might be limiting Ca2+ influx, which in turn would alter Ca2+-mediated signaling pathways and synaptic plasticity.

Overall, these changes markedly reduce NMDAR-mediated currents and might be underlying the severe phenotype of the patient. GluN2B subunits are highly expressed at embryonic and initial postnatal stages, playing a critical role in neurodevelopmental processes (35). Consequently, the hypofunctionality of this major subtype of NMDARs might certainly affect neuritogenesis and synaptogenesis, leading to altered synaptic transmission. This hypothesis is supported by the morphological and biochemical findings, showing a significant decrease in spine density together with increased levels of GluA1 subunit of AMPARs. Pozzo-Miller’s group (28, 36) has reported similar synaptic outcomes in patients with RTT and in primary murine neuronal cultures deficient or harboring mutations of MeCP2. Therefore, our data suggest that NMDAR-induced alterations of glutamatergic synapses might be involved in the pathophysiology of classical RTT condition.

In agreement with the proposed functional impact of de novo mutations of GluN2B subunit, previous studies have associated de novo mutations of GRIN2B with severe phenotypic alterations (6, 14, 15, 37). Lemke and collaborators (13, 38) described the functional consequences of GRIN2B mutations in patients with West syndrome and in individuals with ID with focal epilepsy. In the latter study, the patient with ID and focal epilepsy had a missense mutation in the extracellular glutamate–binding domain. De novo GRIN2B mutations lead to a gain of function, either significantly reducing Mg2+ block and increasing Ca2+ permeability (N615I and V618G mutations, affecting the M2 domain in the pore of the channel) or increasing the apparent glutamate binding affinity (R540H mutation, within the extracellular S1 domain). These gain-of-function mutations point out the important role of facilitated NMDAR signaling in epileptogenesis, with further therapeutic strategies consisting on the selective blockade of mutant leak/hyperactive channels. In contrast, loss-of-function mutations might be causing a hypoglutamatergic function that could be potentially rescued by increasing NMDAR activity with a therapeutic purpose.

The glutamatergic synapse is an extremely sophisticated system where a plethora of molecular actors reside and interact to finely tune neurotransmission. However, under pathological conditions, the dysregulation of critical players might compromise glutamatergic neurotransmission, resulting on an enhancement or a reduction of glutamate signaling (8). In the present work, in silico and in vitro experiments concluded that GluN2B(P553T)-containing NMDARs are hypofunctional. Thus, we envisioned that enhancing NMDAR activity might recover normal glutamatergic neurotransmission and attenuate clinical manifestations. To this end, we evaluated the effect of d-serine, a coagonist of the GluN1 subunit, coapplied with physiological glycine concentrations (39). Our findings indicate an enhancement of mutant GluN1-GluN2B receptor activity, suggesting that the structural changes induced by the mutation are not transduced to GluN1 ligand-binding domain. The ability of d-serine to activate wild-type GluN1-GluN2B receptors (shown in this work) and all NMDAR subtypes (40, 41) should have a general effect on glutamatergic function. Because l-serine is a nutraceutical serine enantiomer already used in pediatric care, the patient was treated with l-serine dietary supplementation. This dietary supplement resulted in increased d-serine concentration in the patient’s plasma, concomitant with strongly increased d-serine amounts in the CSF, in agreement with a previous report (42). Because l-serine is the substrate of serine racemase in the brain (34), our data support the hypothesis that l-serine supplement increases l-serine availability in the brain, which in turn promotes l-serine conversion to d-serine in the brain (43), likely potentiating hypofunctional NMDARs. The beneficial effect of d-serine in healthy individuals has been described in a clinical trial (44). In that work, Heresco-Levy and colleagues showed the procognitive effects of d-serine through NMDAR function and, as we propose in the present study, intended the development of NMDAR glycine site strategies for treating synaptopathies. In agreement with this, d-serine deficits have been associated with aging in rats, with functional rescue observed after exogenous d-serine administration (45, 46). In addition, serine deficiency disorders also provoke neurological phenotypes (psychomotor retardation, microcephaly, and seizures in newborns and children) that can be safely treated by serine oral replenishment (47). Together, these works and ours indicate that, independent of the molecular etiology (whether serine racemase deficit or NMDAR hypofunctionality), serine-potentiated NMDAR activity can partially rescue hypoglutamatergic function. In addition to the beneficial effect of l-serine–mediated increase in d-serine levels, the effect of some other l-serine metabolites and the neurodevelopmental factors might be considered. Regarding the former, our biochemical studies have shown that additional l-serine–derived amino acids and sphingolipids that potentially modulate the NMDAR were not modified by l-serine dietary supplementation. Regarding the neurodevelopmental aspects, the improved patient’s condition might also be influenced by the developmental changes in GluN2 subunit expression, increasing the GluN2A:GluN2B ratio during development (9). Our findings indicate that mutant heterotrimeric NMDARs are less affected than heterodimeric mutant receptors. Therefore, in addition to the NMDAR-potentiating effect of l-serine treatment, GluN2 subunits, the developmental switch might also contribute to improve the clinical symptoms of the patient to some extent.

In summary, our data represent a proof-of-concept study to identify the pathogenicity of de novo mutations of NMDARs and the development of precision therapeutic strategies. The methodological pipeline developed along this study might be further implemented to functionally stratify de novo iGluR mutations associated with synaptic dysfunctions and to define therapeutic strategies. Moreover, they support the use of l-serine as a dietary supplement for the enhancement of glutamatergic neurotransmission and/or excitatory or inhibitory neurotransmitter imbalance that are associated with a large spectrum of neurological disorders.


Patient neurodevelopmental and adaptive behavior assessments before and after dietary l-serine supplement

The study was approved by the appropriate informed consent of the patient’s parents. The VABS-II (48) semistructured interview, allowing the assessment of four domains of adaptive behavior (communication, daily living skills, socialization, and motor skills), was conducted by a trained neuropsychologist before (at 5 years 10 months old) and after 11 and 17 months of l-serine supplementation (at 6 years and 9 months old and 7 years and 3 months old, respectively). Initially, l-serine dose started at 250 mg/kg per day (for 4 weeks), and upon the confirmation of a lack of side effects, the dose was increased to 500 mg/kg per day (divided into three dietary supplements of l-serine powder, mixed with food and/or drinks) and maintained along the extent of the trial.

Whole-exome sequencing

Coding regions were captured using the TruSeq DNA Sample Preparation and Exome Enrichment Kit (Illumina, San Diego, CA, USA), and paired-end 100 × 2 sequences were sequenced with the Illumina HiScanSQ system at National Center for Genomic Analysis in Barcelona (Catalonia, Spain). The overall coverage statistics for the trio are 381.451, 433.847, and 31.635 for the patient, the mother, and the father, respectively.

Bioinformatic pipeline

Sequence reads were aligned to the Genome Reference Consortium Human Genome Build 37 (GRCh37) hg19 using the Burrows-Wheeler Aligner (49). Properly mapped reads were filtered with SAMtools (1, 2), which was also used for sorting and indexing mapping files. Genome Analysis Toolkit (GATK) (3) was used to realign the reads around known indels and for base quality score recalibration. Once a satisfactory alignment was achieved, identification of single-nucleotide variants (SNVs) and indels was performed using GATK standard multisample variant calling protocol, including variant recalibration (4). For the final exome sequencing analysis report, the Annotate Variation (ANNOVAR) (5) tool was used to provide additional variant information to ease the final selection of candidates. In particular, minor allele frequency (MAF), obtained from dbSNP (Single-Nucleotide Polymorphism Database) (50, 51) and 1000 Genomes Project (68), was provided to help to select previously undescribed variants in healthy population.

De novo SNVs and small insertion or deletion events

To identify de novo SNVs and small insertion and/or deletion events, the patient’s variants were filtered first against parental variants and then against a pool of controls made up by all healthy parents included in the study. SIFT (52) and PolyPhen-2 (Polymorphism Phenotyping v2) (53) damage scores were computed to predict putative impact over protein structures. The successive application of quality control filters and the prioritization by the parameters with potential functional impact was used to construct a list of candidate genes (and variants) ranked by its uniqueness in the cases (or very low frequency in the control population, as derived from the MAFs) and the putative potential impact.

Sanger sequencing

The variants were validated by Sanger sequencing using BigDye Terminator v3.1 Cycle Sequencing Kit (Life Technologies, Grand Island, NY, USA) in an Applied Biosystems 3730/DNA Analyzer (Applied Biosystems, Life Technologies, Grand Island, NY, USA) using the following set of primers: 5′-TACAATCTAACCTAGGCCCTGG-3′ (forward) and 5′-TGGATATGCTAGGGAAAATGCAG-3′ (reverse). The raw data were analyzed with CodonCode Aligner software (CodonCode Corporation, Centerville, MA, USA).

In silico prediction of mutation impact

The following four in silico prediction tools of functional mutation impact were used: SIFT, PolyPhen-2, Protein Variation Effect Analyzer (PROVEAN) (54), and MutationTaster2 (55). SIFT is a sequence homology–based tool that predicts variants as neutral or deleterious using normalized probability scores. Variants at position with a normalized probability score less than 0.05 are predicted to be deleterious, and a score greater than 0.05 is predicted to be neutral (56). PolyPhen-2 uses a combination of sequence and structure-based attributes and naive Bayesian classifier for the identification of an amino acid substitution and the effect of mutation. The output results of probably damaging and possibly damaging were classified as deleterious (≥0.5) and the benign level being classified as tolerated (≤0.5). PROVEAN uses a region-based delta alignment score, which measures the impact of an amino acid variation not only based on the amino acid residue at the position of interest but also on the quality of sequence alignment derived from the neighborhood flanking sequences. Variants with a PROVEAN score lower than −2.5 are predicted to be deleterious (57). MutationTaster2 also uses a Bayes classifier to generate predictions but includes all publicly available single-nucleotide polymorphisms and indels from the 1000 Genomes Project, as well as known disease variants from ClinVar (58) and the Human Gene Mutation Database (HGMD) (59). Alterations found more than four times in the homozygous state in 1000 Genomes or in HapMap (60) were automatically regarded as neutral. Variants marked as pathogenic in ClinVar were automatically predicted to be potentially disease causing.

Molecular modeling of wild-type and mutant (GluN1)2-(GluN2B)2 receptors

An initial homology model was constructed for the transmembrane domain of human (GluN1)2-(GluN2B)2 receptor using the coordinates of the crystal structure of rat receptor (PDB ID: 4PE5) (61). MODELLER v16 (62) was used to model the lacking residues of the loop connecting M1 and M3. The side-chain conformations for those residues were positioned according to SCWRL (63). The backbone conformation from residues 551 to 556 was modeled with MODELLER v16 software (62). Protein complexes were embedded in a model membrane containing 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) lipids, water molecules, and 0.15 M NaCl in a rectangular box (64). The initial system was energy-minimized and then subjected to molecular dynamics equilibration (25 ns). Subsequently, the system was subjected to a production stage, extending 250 ns. All the simulations were performed with GROMACS 4.5 simulation package (65).

Phylogenetic analysis of GluN2B(P553)-containing domain

Genes coding for iGluRs subunits from different species belonging to the phylum metazoan were identified using the BLASTP tool and human iGluRs as queries. Subject sequences with an E value below 0.05 were selected as possible homologs. These were reblasted against the National Center for Biotechnology Information “nonredundant protein sequences (nr)” database to establish their correspondence with glutamate receptors. These searches were performed against the Ensembl Metazoa database ( for L. gigantea, C. teleta, S. maritima, A. melifera, Nematostella vectensis, Mnemiopsis leidyi, and Trichoplax adhaerens; Metazome database ( for S. kowalevskii; and alternative polyadenylation sites database (APAsdb) ( for B. belcheri. Sequences from porifera species were taken from Riesgo’s group database (66). All vertebrate sequences were obtained from the Ensembl Metazoa database. A final set of 147 iGluR sequences from 12 metazoan species spanning seven phyla was used in the phylogenetic analysis. Sequence multiple alignment was made using MUSCLE tool (67) included in the MEGA6 software (68). The alignment was made with coding DNA sequences, using “codons” option, which allows maintaining nucleotide triplets coding for amino acids. Using MEGA6, we established that GTR + I + G (general time reversible with gamma rates and a proportion of invariant sites) was the best evolutionary model to use in the phylogenetic tree. The phylogenetic tree was constructed using the Bayesian inference method, with MrBayes 3.2 software (69). Arabidopsis thaliana iGluR sequences were obtained from the TAIR (The Arabidopsis Information Resource) database ( and used as an external group of the tree. Two simultaneous trees were analyzed, and 3,630,000 iterations were run until both trees converged.


The expression plasmids for rat GluN1 and GFP-GluN2B were provided by S. Vicini (Georgetown University Medical Center, Washington, USA) (70). The plasmids used for the analysis of triheteromeric NMDARs were provided by P. Paoletti (École Normale Supérieure Paris, France, EU). Nucleotide change (the mutation of GluN2B serine at amino acid position 553 to a Thr residue) was achieved by oligonucleotide-directed mutagenesis, using the QuikChange site–directed mutagenesis kit according to the manufacturer’s instructions (Stratagene). The plasmid generated by site-directed mutagenesis was verified by DNA sequencing.

Coimmunoprecipitation experiments

For immunoprecipitation of heterologously expressed GluN1, HA-GluN2A, and GFP-GluN2B (wild-type or mutant), transfected HEK-293T cells were washed in cold phosphate-buffered saline (PBS) and subsequently solubilized in radioimmunoprecipitation assay buffer supplemented with protease and phosphatase inhibitor cocktail (PPIC), for further immunoprecipitation with anti-GFP. The homogenates were clarified by centrifugation at 4°C for 10 min at 16,000g. After preclearing the soluble lysates for 1 hour at 4°C with equilibrated protein G Sepharose, they were incubated overnight at 4°C with 10 μg of an anti-GFP mouse monoclonal antibody (Ab) (1:200; catalog no. 9777966, Clontech). Nonspecific mouse immunoglobulin G (catalog no. I5381, Sigma-Aldrich) was used as a control for specificity. The immunocomplexes were incubated with protein G Sepharose for 2 hours at 4°C, and the beads were then washed twice with lysis buffer and once with PBS. The bound proteins were eluted in Laemmli’s buffer and analyzed by Western blot.

Western blot analysis

For protein extraction, cells were washed once with PBS and scraped off the plate in 400 μl of lysis buffer [50 mM Hepes (pH 7.4), 150 mM NaCl, 2 mM EDTA, 1% NP-40, and PPIC]. After 10 min of incubation at 4°C, the cell debris was pelleted at 15,000g, the solubilized proteins were collected, and the protein concentration was determined using a bicinchoninic acid assay (BCA). Proteins were separated by 8% SDS–polyacrylamide gel electrophoresis and transferred to nitrocellulose membranes (Amersham), which were then blocked with 10% skimmed milk in 10 mM tris-HCl (pH 7.5)/100 mM NaCl [tris-buffered saline (TBS)] plus 0.1% Tween 20 (TBS-T). The membranes were probed overnight at 4°C with the primary Ab of interest (diluted in TBS-T + 5% skimmed milk) directed against GluN1 (1:500; catalog no. 05-432, Millipore), HA epitope (catalog no. MMS-101R-500, Covance Inc.), and GFP (Clontech). Ab binding was detected with an anti-mouse or anti-rabbit Ab coupled to horseradish peroxidase (Dako) for 1 hour at room temperature (RT), and the immunocomplexes were visualized by chemiluminescence (ECL detection system, Pierce), following the manufacturer’s instructions. Immunosignals were analyzed densitometrically with ImageJ software [National Institutes of Health (NIH), USA].

Cell culture and transfection

HEK-293T and COS-7 cell lines were obtained from the American Type Culture Collection and maintained at 37°C in Dulbecco’s modified Eagle’s medium, supplemented with 10% fetal calf serum and antibiotics [penicillin (100 U/ml) and streptomycin (100 μg/ml)]. Furthermore, d-2-amino-5-phosphonopentanoic acid (Abcam) was added to the medium (final concentrations of 200 or 500 μM for HEK-293T and COS-7 cells, respectively) to avoid excitotoxicity. Transient transfection of HEK-293T cells was achieved by the calcium phosphate method (Clontech), and cell extracts were obtained 48 hours after transfection. COS-7 cells were transfected with Lipofectamine 2000 (Invitrogen), following the manufacturer’s instructions, and cells were fixed 24 hours after, for further immunofluorescence analysis.

To prepare dissociated mouse hippocampal neuron cultures, mouse embryos (embryonic day 18) were obtained from pregnant CD1 females, the hippocampi were isolated and maintained in cold Hank’s balanced salt solution (HBSS; Gibco) supplemented with 0.45% glucose (HBSS-glucose). After carefully removing the meninges, the hippocampi were digested mildly with trypsin for 17 min at 37°C and dissociated. The cells were washed three times in HBSS and resuspended in Neurobasal medium supplemented with 2 mM GlutaMAX (Gibco) before filtering in 70-μm mesh filters (BD Falcon). The cells were then plated onto glass coverslips (5 × 104 cells/cm2) coated with poly-l-lysine (0.1 mg/ml; Sigma-Aldrich), and 2 hours after seeding, the plating medium was substituted by complete growth medium, consisting on Neurobasal medium supplemented with 2% B27 (Invitrogen) and 2 mM GlutaMAX, and the coverslips were incubated at 37°C in a humidified 5% CO2 atmosphere. Every 3 to 4 days, half of the conditioned medium was removed and replaced by fresh growth medium. Primary cultures were transfected with 0.8 μg of DNA (Lipofectamine 2000, Invitrogen) on DIV4, DIV7, or DIV11 for further surface expression analysis of GFP-GluN2B constructs and endogenous GluA1. All the experimental procedures were carried out according to European Union guidelines (Directive 2010/63/EU) and following protocols that were approved by the Ethics Committee of the Bellvitge Biomedical Research Institute (IDIBELL).

Immunofluorescence analysis of surface NMDARs

The surface-to-total expression of NMDARs in COS-7 cells was performed as previously described (71). Briefly, cells were washed twice with PBS before they were fixed with 4% paraformaldehyde (PFA). Surface expression of GFP-GluN2B constructs was detected using an Ab against GFP (1:1000; catalog no. A11122, Life Technologies) that recognizes the extracellular epitope of heterologously expressed receptors and that was visualized with an Alexa Fluor 555–conjugated goat anti-rabbit Ab (1:500; catalog no. A-31851, Thermo Fisher Scientific). The total pool of receptors was detected by the fluorescent signal emitted by the GFP-GluN2B construct.

To analyze the surface expression of the transfected NMDARs in primary hippocampal neuronal cultures, cells were washed twice with PBS and fixed with 4% PFA in PBS containing 4% sucrose. The surface expression of GFP-GluN2B constructs was detected by incubating with anti-GFP (1:1000; catalog no. A11122, Life Technologies) during 1 hour at RT and visualized with an Alexa Fluor 488–conjugated goat anti-rabbit Ab (1:500; catalog no. A-11078, Thermo Fisher Scientific). The intracellular pool of receptors was identified by permeabilizing cells with 0.1% Triton X-100 and labeling them with a rabbit anti-GFP–Alexa Fluor 555–conjugated Ab (1:2000; catalog no. A-31851, Thermo Fisher Scientific).

Fluorescence was visualized with a Leica TCS SL spectral confocal microscope (Leica Microsystems, Wetzlar, Germany) using a Plan-Apochromat 63×/1.4–numerical aperture immersion oil objective (Leica Microsystems) and a pinhole aperture of 114.54 or 202 μm (for surface receptors). To excite the different fluorophores, the confocal system is equipped with excitation laser beams at 488 and 546 nm. In each experiment, the fluorescence intensity was measured in 10 to 15 dendrites from at least two to three pyramidal neurons (or in 10 to 15 COS-7 cells) per condition. Fluorescence was quantified using Adobe Photoshop CS5 software (Adobe Systems Inc.), and the results are represented as the means ± SEM of the ratio of surface:intracellular (primary cell culture) or surface:total (COS-7 cells) GluN2B immunofluorescence signal, analyzing at least three independent experiments.

Morphological analysis of dendritic arborization and spines

GFP-GluN2Bwt– and GFP-GluN2B(P553T)–transfected neurons were immunolabeled, and Z-stack images were acquired. The resulting maximum projections were analyzed using “Sholl analysis” ImageJ plugin (72). Dendrites were manually traced with Neuron Studio software ( Dendritic spines from tertiary neurites were counted and classified into morphological categories (thin, mushroom, and stubby), using Neuron Studio software automatic analysis, followed by manual revision to discard artifacts and/or spines counts redundancy.

Chemical LTP–induced recruitment of AMPARs in the cell surface and synaptic morphology changes

Gly-cLTP assay was performed at DIV14 on primary hippocampal neurons transiently transfected at DIV11 with either GFP-GluN2Bwt + GFP (4:1 ratio) or GFP-GluN2B(P553T) + GFP (4:1 ratio), adapting the protocols previously described (31, 32). Briefly, basal conditions (nonstimulated neurons) consisted in the incubation of primary cultures with Krebs-Ringer solution supplemented with 1 mM Mg2+ and 1 μM tetrodotoxin. For Gly-cLTP induction, cells were briefly washed in 20 μM bicucculine (BIC) + 20 μM strychnine and then incubated for 5 min in Krebs-Ringer solution supplemented with 20 μM BIC + 1 μM strychnine and 200 μM Gly (100 μM d-serine supplement, for Gly + d-Ser condition). The solutions were replaced by a medium supplemented with 20 μM BIC, 20 μM strychnine, and 1 mM Mg2+. After 20 min of incubation at 37°C (GluA1 surface recruitment studies) or 35 min of incubation at 37°C (dendritic spines analysis), cells were fixed with ice-cold 4% PFA in PBS containing 4% sucrose. For immunofluorescence analysis of GluA1 surface recruitment, neurons were incubated at RT for 30 min with anti-GluA1 Ab (1:200; catalog no. MAB2263, Millipore) and incubated with Alexa Fluor 555–conjugated goat anti-mouse Ab (1:500; catalog no. A31570, Molecular Probes). After permeabilization/blocking, cells were incubated with rabbit anti-GFP (1:1000; catalog no. A11122, Life Technologies) and then with Alexa Fluor 488–conjugated goat anti-rabbit Ab (1:500; catalog no. A-11078, Thermo Fisher Scientific), and washed and mounted for confocal microscopy analysis. Secondary dendritic processes expressing the GFP-GluN2B constructs were analyzed by quantifying the fluorescence intensity of GluA1 puncta with ImageJ software (NIH), and tertiary dendritic processes were included for spines density and morphology analysis.

Electrophysiological recordings of diheteromeric and triheteromeric NMDAR-mediated whole-cell currents in HEK-293T cells

Electrophysiological recordings were obtained 18 to 24 hours after transfection, perfusing the cells continuously at RT with an extracellular physiological bath solution: 140 mM NaCl, 5 mM KCl, 1 mM CaCl2, 10 mM glucose, and 10 mM Hepes, adjusted to pH 7.42 with NaOH. Glutamate (1 mM; Sigma-Aldrich), in the presence of glycine (1, 50, or 100 μM depending the experiment type; Tocris) and d-serine (0 to 300 μM) was applied for 0.5 s by piezoelectric translation (P-601.30; Physik Instrumente) of a theta-barrel application tool made from borosilicate glass (1.5 mm outside diameter; Sutter Instruments), and the activated currents were recorded in the whole-cell configuration at a holding potential of −60 mV, acquired at 5 kHz and filtered at 2 kHz by means of Axopatch 200B amplifier, Digidata 1440A interface, and pClamp10 software (Molecular Devices Corporation). Electrodes with open-tip resistances of 2 to 4 megohms were made from borosilicate glass (1.5 mm outside diameter, 0.86 mm outside diameter; Harvard Apparatus), pulled with a PC-10 vertical puller (Narishige), and filled with intracellular pipette solution containing 140 mM CsCl, 5 mM EGTA, 4 mM Na2ATP, 0.1 mM Na3GTP, and 10 mM Hepes, adjusted to pH 7.25 with CsOH. Glutamate and glycine-evoked currents were expressed as current density (in pA/pF; maximum current divided by input capacitance as measured from the amplifier settings) to avoid differences due to surface area in the recorded cells.

The kinetics of deactivation and desensitization of the NMDAR responses were determined by fitting the glutamate/glycine-evoked responses at Vm − 60 mV to a double-exponential function to determine the weighted time constant (τw,des)τw,des=τf(AfAf+As)+τs(AsAf+As)where Af and τf are the amplitude and time constant of the fast component of desensitization, respectively, and As and τs are the amplitude and time constant of the slow component of desensitization, respectively.

To infer single-channel conductance values from macroscopic deactivating currents, we used NSFA as previously described (24). The single-channel current (i) was calculated by plotting the ensemble variance against mean current (Ī) and fitting with Sigworth parabolic function (73)σ2=σB2+(iI(I2N))where σ2B is the background variance and N is the total number of channels contributing to the response. The weighted-mean single-channel conductance was determined from the single-channel current and the holding potential of −60 mV.

NMDAR agonists dose-response experiments

To determine the affinity for d-serine or glutamate in GluN2B- and GluN2B(P553T)-transfected cells, concentration-response curves were constructed from whole-cell currents elicited by rapid jumps of 0.5-s duration at different concentrations of the coagonist d-serine (10−9 to 10−4 M) in the presence of 1 mM glutamate or at different concentrations of the agonist glutamate (10−8 to 10−2 M) in the presence of 1 mM d-serine. Concentration-response curves were fitted individually for every cell using the Hill equationI=Imax1+(EC50[A])nHwhere Imax is the maximum current, [A] is the concentration of d-serine, nH is the slope (Hill) coefficient, and EC50 is the concentration of d-serine or glutamate that produces a half-maximum response. Each data point was then normalized to the maximum response obtained in the fit. The average of the normalized values with their SEM were plotted together and fitted again with the Hill equation. The minimum and maximum values were constrained to asymptote 0 and 1, respectively.

Electrophysiological recordings of NMDA EPSCs in hippocampal neurons

Spontaneous activity–dependent NMDAR-mediated EPSCs were recorded in cultures of hippocampal neurons (DIV16). Whole-cell recordings were obtained from transfected neurons (at DIV12) with GluN2B-GFP or GluN2B(P553T). Extracellular solution contained 140 mM NaCl, 3.5 mM KCl, 2 mM CaCl2, 20 mM glucose, and 10 mM Hepes, (Mg2+-free) adjusted to pH 7.42 with NaOH. To isolate NMDAR component, 50 μM NBQX and 100 μM picrotoxin were added to block AMPAR and γ-aminobutyric acid type A receptor–mediated PSCs, respectively. Intracellular pipette solution contained 116 mM K-gluconate, 6 mM KCl, 8 mM NaCl, 0.2 mM EGTA, 2 mM MgATP, 0.3 mM Na3GTP, and 10 mM Hepes, adjusted to pH 7.25 with KOH. QX-314 (2.5 mM) was included into the pipette solution to block action potential firing. EPSCs were acquired at 5 kHz and filtered at 2 kHz as described for cell lines at a holding potential of −70 mV. EPSCs were measured in 5-min periods in the presence of NBQX (baseline) and NBQX + d-serine (100 μM), as indicated. After d-serine treatment, 2-amino-5-phosphonopentanoic acid (50 μM) was added to validate that EPSCs recorded were NMDAR-mediated. pClamp10/Clampfit10.6 (Molecular Devices) was used to record, detect, and analyze the amplitude, interevent interval, and charge transfer (as area under the curve, in pA*ms) from single EPSCs.

Quantitative analysis of total and stereoselective amino acids in human samples

Amino acid analysis was performed on an ACQUITY UPLC H-class instrument (Waters Co., Milford, MA, USA) with a reversed-phase C-18 column using water and acetonitrile, 0.1% formic acid as mobile phases (run time, 9 min). The detection was performed with a Xevo TQD triple-quadrupole mass spectrometer (Waters Co., Milford, MA, USA) using positive electrospray ionization in the multiple reaction monitoring mode. For the quantification of l– and d–amino acids enantiomers of human biological fluids (plasma and CSF), a UPLC-MS/MS–based method was performed, using N-(4-nitrophenoxycarbonyl)-l-phenylalanine 2-methoxyethyl ester [(S)-NIFE method], as previously reported (74). Briefly, EDTA-anticoagulated plasma and CSF samples were collected from control age-matched individuals (normal diet) and from the patient (before and after l-serine dietary supplementation, for plasma analysis; after l-serine supplement, for CSF analysis), as previously described (75). All samples were stored at −80°C until use. Plasma and CSF samples were mixed with internal standard solution. After 10 min of incubation at 4°C, ice-cold acetonitrile was added, and the mixture was incubated for >15 min on ice. Precipitates were removed by centrifugation, and the supernatant was evaporated to dryness under a stream of nitrogen, using a heating block set to 40°C. The residue was dissolved in water, followed by the addition of sodium tetraborate and (S)-NIFE solution in acetonitrile. After 10 min of incubation at RT, the reaction was terminated by hydrogen chloride addition. The derivatized and filtrated samples were immediately separated on a 100 mm by 2.1 mm ACQUITY 1.7-μm BEH C18 column, using an ACQUITY UPLC system coupled to a Xevo tandem MS (Waters Co., Milford, MA, USA).

Untargeted sphingolipidomic studies in human samples

Plasma samples were processed and analyzed as previously described (76). Briefly, total lipid extract was obtained from 100 μl of plasma, using modified Bligh and Dyer extraction (77). Plasma total lipid extracts were separated on an Dionex UltiMate 3000 UPLC system (Thermo Fisher Scientific, San Jose, CA, USA) using a kinetex C8 150 × 2.1 mm, 2.6-μm column (Phenomenex, Sydney, NSW, Australia). After injection of the samples (10 μl), the column effluent was directly introduced into the heated electrospray ionization source of an Orbitrap Fusion mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA), and analysis was performed in positive ionization mode. The relative quantification of lipid species was obtained from the area of their corresponding individual chromatographic peaks.

Statistical analysis

Comparison between experimental groups was evaluated using InStat software (GraphPad Software Inc.), applying a one-way ANOVA, followed by a Bonferroni post hoc test for multiple comparisons or a repeated measures two-way ANOVA for Sholl analysis. For single comparisons, either Student’s t test (for parametric data) or Mann-Whitney U test (for nonparametric data) was used. Data are presented as means ± SEM from at least three independent experiments.


Fig. S1. Protein interactions and cellular trafficking of GluN2Bwt- and GluN2B(P553T)-containing NMDARs.

Fig. S2. Altered biophysical properties of heterotrimeric GluN1-GluN2A-GluN2B(P553T) NMDARs.

Fig. S3. GluN2B(P553T) mutation alters GluA1 abundance in hippocampal neurons.

Fig. S4. Alignment of eumetazoan iGluRs showing the residues conservation of Pro553 and Phe653.

Table S1. Untargeted analysis of plasma sphingolipid profile in the GRIN2B(P553T) patient before and after l-serine dietary supplementation.

References (78, 79)


Acknowledgments: We acknowledge the help from N. Verhoeven and J. Gerrits (Universitair Medisch Centrum Utrecht, The Netherlands), as well as R. Artuch and A. Ormazábal (Hospital Sant Joan de Déu, Barcelona, Spain), for conducting the analytical measurements of amino acids in biofluids. We thank S. Jurado (Instituto de Neurociencias, Alicante, Spain) for guidance in cLTP experiments and the “Medicinal Computational Laboratory” from Universitat Autònoma de Barcelona for providing computing facilities. We also acknowledge P. Paoletti (ENS, Paris, France) for providing the plasmids allowing the studies with triheteromeric NMDARs. Funding: This work was supported by the grants PI16/00851 [ISCIII, cofunded by European Regional Development Fund (ERDF), a way to build Europe], La Marató (project no. 20140210), PCIN-2014-105 (MINECO), and Miguel Servet Program (CPII16/00021, ISCIII) to X.A.; BFU2017-83317-P (MICINN) to D.S.; PI15/01082 (ISCIII) to À.G.-C.; SAF2016-77830-R to M.O.; BFU2012-34398 and BFU2015-69717-P (MINECO), Career Integration Grant (ref. 304111), Marie Curie Intra-European Fellowship (ref. 221540), and Ramón y Cajal Fellowship (RYC-2011-08391p) to À.B.; SAF2012-40102 (MINECO/FEDER), FP7 Marie Curie CIG grant (#631035), and Ramón y Cajal (RYC-2011-08026) to C.S.; PI17/00296 and RETIC RD16/0008/0014 to X.G.; and MetaboHUB project (ANR-11-INBS-0010) to B.C. C.A. received an FPI contract (MINECO). X.A., C.S., and À.B. also benefit from the financial support of AGAUR (SGR14-297). A.S.-G. benefits from a Fundación Tatiana Pérez de Guzmán el Bueno PhD fellowship. We thank CERCA Programme/Generalitat de Catalunya for institutional support. Author contributions: X.A. designed the study, in collaboration with À.G.-C., D.S., M.O., X.G., and C.S. D.S., X.G., R.G.-D., and E.G.-B. conducted the electrophysiological recordings. M.O. developed the structural model and performed the bioinformatic analysis. À.B. and D.R.-V. performed the phylogenetic analysis. C.S., C.A., C.G., and M.G.d.S. performed the cLTP analysis. C.G., A.S.-G., and S.L. performed the morphological analysis. E.C.-V. and B.C. performed the lipidomic studies, A.L.-S. performed the behavioral assessments. V.F.-D. and F.C. performed the pharmacological studies. J.A. generated and validated the genetic data. À.G.-C. coordinated the clinical part of the study. X.A. wrote the manuscript with the help of À.G.-C., D.S., M.O., and all authors proofread the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials

Stay Connected to Science Signaling

Navigate This Article