An autism-associated mutation in GluN2B prevents NMDA receptor trafficking and interferes with dendrite growth

Autism spectrum disorders (ASDs) are a family of complex neurodevelopmental disorders characterized by restricted, repetitive behaviors and interests, impaired social interactions and difficulty with communication. ASD is highly heritable, and de novo protein-altering mutations are thought to cause a high percentage of ASD cases (Krumm et al., 2014). De novo mutations in the GRIN2B gene, which encodes the GluN2B subunit of NMDA receptors (NMDARs), have been identified in multiple probands with sporadic ASD but not in unaffected siblings or other control individuals (De Rubeis et al., 2014; Iossifov et al., 2014; Kenny et al., 2014; Myers et al., 2011; O'Roak et al., 2011, 2012b, 2014; Pan et al., 2015; Platzer et al., 2017; Sanders et al., 2015; Takasaki et al., 2016; Talkowski et al., 2012; Tarabeux et al., 2011; Yoo et al., 2012). The recurrence of de novo ASD-associated mutations in GRIN2B and rigorous analyses indicate that GRIN2B is a true ASD-associated gene (De Rubeis et al., 2014; Iossifov et al., 2014; O'Roak et al., 2012b; Stessman et al., 2017). However, we are only beginning to understand the impact of ASD-associated GRIN2B mutations on NMDAR function (Fedele et al., 2018; Liu et al., 2017; Vyklicky et al., 2018), and it is not yet clear how GRIN2B mutations alter neuronal development to cause ASD.

NMDARs are ionotropic glutamate receptors that are important for neuronal development and plasticity (Sanz-Clemente et al., 2013). These receptors are tetramers comprised of two GluN1 subunits and two GluN2 and/or GluN3 subunits. Within the assembled receptor, GluN2 subunits are necessary for glutamate-dependent activation and subcellular trafficking (Horak et al., 2014). While GluN2B-containing receptors can be found in the axon and presynaptic terminals in developing cortex and hippocampus (Berg et al., 2013; Charton et al., 1999; DeBiasi et al., 1996; Gill et al., 2015; Jourdain et al., 2007; Larsen et al., 2011, 2014; McGuinness et al., 2010), most GluN2B-containing receptors are found within dendrites.

GluN2B is essential for normal development. During postnatal development, most NMDARs in the cortex and hippocampus contain GluN2B (Komuro and Rakic, 1993; Monyer et al., 1994; Watanabe et al., 1992). GluN2B homozygous knockout mice die around birth (Kutsuwada et al., 1996), and genetic substitution of GluN2A (encoded by Grin2a) for GluN2B does not rescue this lethality (Hamada et al., 2014; Wang et al., 2011). Moreover, genetically replacing GluN2B with GluN2A results in reduced social exploration in surviving mice, even when total NMDAR currents are equivalent (Wang et al., 2011). In addition, abnormal expression or function of GluN2B-containing NMDARs has been observed in mouse models of ASD and ASD-related neurodevelopmental disorders, including Rett syndrome, Fragile X syndrome and fetal valproate exposure (for examples, see Asaka et al., 2006; Bostrom et al., 2015; De Filippis et al., 2015; Etherton et al., 2011; Krueger et al., 2011; Lee et al., 2008; Mierau et al., 2016; Roullet et al., 2010; Sceniak et al., 2016; Schutt et al., 2009; Stefanovic et al., 2015; Toft et al., 2016; Wang et al., 2018).

Symptoms and diagnosis of ASD frequently occur by the age of 3 (Bourgeron, 2009). This timing suggests that aspects of neuronal development that occur during this period underlie autism pathogenesis. Several studies have demonstrated roles for GluN2B in regulation of dendrite growth and patterning (Bustos et al., 2014; Espinosa et al., 2009; Ewald et al., 2008; Keith et al., 2019; Sepulveda et al., 2010), and abnormal dendrite architecture is often associated with ASD (Gao and Penzes, 2015; Lin et al., 2016; Varghese et al., 2017). Therefore, we hypothesized that ASD-associated mutations in GluN2B dysregulate elaboration of dendritic arbors.

To begin to understand how mutations in GluN2B contribute to ASD pathogenesis, we focused on a de novo single nucleotide variant mutation in GRIN2B that was identified in a patient with severe, early onset simplex ASD (O'Roak et al., 2011). This human mutation is predicted to produce a premature stop within the second extracellular loop of GluN2B, but the biological effects of this mutation have not yet been studied. Here, we evaluated the impact of this ASD-associated mutation in GluN2B on NMDAR function, receptor trafficking and cortical excitatory neuron development. We find that mutant receptors are completely non-functional and are trafficked neither to the cell surface nor into dendritic spines. Expression of ASD mutant GluN2B subunits led to dramatic defects in dendrite growth and branching. Dendrite maldevelopment occurred when mutant GluN2B was expressed in the presence of wild-type GluN2B. Importantly, this parallels the disease since ASD patients have mutation of only one allele of the gene. While dendritic arbors were smaller, dendritic spine density was unchanged, suggesting that total synapse number was reduced. Together, these data suggest that ASD mutations that severely disrupt NMDA receptor trafficking and function may contribute to ASD pathophysiology by limiting dendrite development, thereby disrupting normal circuit function.

To investigate the cellular and molecular phenotypes resulting from patient GRIN2B mutations, we investigated a GRIN2B variant that was identified in a female proband (SSC proband #12681) who was clinically diagnosed with ASD and met the Autism Diagnostic Observation Schedule research criteria for ASD (O'Roak et al., 2011). As described by O'Roak and colleagues (2011), this proband's symptoms were severe, with a Calibrated Severity Score of 9 (with 10 corresponding to the most severe score possible). She showed impaired overall adaptive behaviors (6th percentile). Her communication abilities were ranked at the 5th percentile, and social-emotional behaviors at the 18th percentile. The proband exhibited comorbid intellectual disability, with a full-scale IQ of 63 and a non-verbal IQ of 65, both at the 1st percentile. The affected individual used language fluently, but did display language delay. Onset of ASD was early, prior to age 3, with possible regression. The proband had unaffected parents, each with no psychiatric history, and two siblings who were both unaffected. Importantly, this mutation has not been found in any unaffected individuals (O'Roak et al., 2011, 2012a, 2012b, 2014).

The single-nucleotide mutation identified in this proband (12681) was initially described by O'Roak et al. (2011) and expected to be a loss of function mutation (O'Roak et al., 2011, 2012a,b). This variant (c.2172-2A>G) is predicted to be protein truncating (O'Roak et al., 2012a). Therefore, we modeled the mutation in GluN2B cDNA by introducing a stop codon within the second extracellular loop after amino acid 724 (Fig. 1A), corresponding to the end of exon 10 of GRIN2B and the predicted truncation site. This construct is referred to here as GluN2B^724t.

Because it seemed possible that the conformational changes caused by the truncation could alter the affinity of anti-GluN2B antibodies for the mutant protein, we produced the mutation in GluN2B with a GFP tag inserted in the N-terminal domain, after the signal peptide, to allow reliable visualization of the mutant GluN2B. It is well-established that the template GFP–GluN2B construct expresses well, functions normally and traffics properly (Luo et al., 2002). When HEK-293 cells were transfected with GFP–GluN2B^724t or GFP–GluN2B^WT (wild-type), expression levels and transfection efficiencies for the two proteins appeared similar (Fig. 1B).

To determine whether this GluN2B mutation alters NMDA receptor function, we transfected HEK-293 cells with cDNAs encoding either wild-type GluN2B (GluN2B^WT) or the GluN2B^724t mutant, along with GluN1-1a, referred to as GluN1 hereafter (encoded by GRIN1) tagged with HcRed. Heterologous expression avoids the complications of endogenous wild-type GluN2B expressed in neurons. At 24–48 h after transfection, we used Fluo4 to image Ca²⁺ influx in response to brief perfusion with NMDA (100 µM) in the presence of glycine (10 µM) and absence of magnesium. Wild-type NMDA receptors consistently produced a profound NMDA-dependent increase in intracellular Ca²⁺ [Fig. 1B–D, n=80 cells from four independent experiments, maximal ΔF/F₀ (fold change in fluorescence, normalized to baseline): 0.796±0.009, mean±s.e.m.]. In stark contrast, NMDA failed to induce a Ca²⁺ response in all cells transfected with GluN2B^724t (Fig. 1B–D, n=40 cells from four independent experiments, maximal ΔF/F₀: −0.012±0.002, P=1.36×10⁻¹⁰). In all cases, NMDA-dependent Ca²⁺ influx was completely blocked in the same cells upon exposure to the NMDA receptor antagonist ketamine (Fig. 1B,C). These data suggest that GluN2B^724t is unable to form functional receptors.

GluN2B subunits have a well-established role in controlling the subcellular trafficking of NMDARs (Horak et al., 2014). Therefore, the observed impairments in receptor function could be due to disrupted subcellular trafficking. To test this hypothesis, we selectively labeled GluN2B at the cell surface in HEK-293 cells co-transfected with GluN1 and either GFP–GluN2B or GFP–GluN2B^724t (Fig. 2). To selectively visualize surface receptors, we labeled living, non-permeabilized cells with antibodies against the GFP tag, which was present on the N-terminal extracellular domain of GluN2B. Then, labeled cells were fixed and observed by confocal microscopy. No surface labeling was detected without GluN2B transfection or when antibodies to intracellular epitopes were used. In addition, no labeling was observed when cells were transfected with cytosolic GFP instead.

Surface expression of GluN2B was markedly different between GluN2B^724t and GluN2B^WT (Fig. 2). As expected, GluN2B^WT was abundant at the cell surface (Fig. 2A,B). Strikingly, GluN2B^724t was nearly absent from the cell surface (Fig. 2A,B). GluN2B^724t-expressing cells showed minimal surface expression [surface fluorescence normalized to wild-type surface fluorescence: 0.29±0.03 (GluN2B^724t), n=41 cells; P<0.0001; mean±s.e.m.] when compared to cells transfected with GluN2B^WT (1.00±0.12, n=64 cells). Similar results were observed using two distinct anti-GFP antibodies. The observed change in surface expression was not a result of reduced overall expression of GluN2B since total expression levels of GFP–GluN2B^WT and GFP–GluN2B^724t were not significantly different [Fig. 2C, total GluN2B expression, normalized to wild-type 1.00±0.16 (GluN2B^WT), 0.71±0.15 (GluN2B^724t); P=0.215]. Next, we measured the total expression of GluN1 in order to confirm that differences in GluN2B expression at the cell surface reflected the effects of the mutation rather than differences in co-transfection efficiency or GluN1 expression levels. As shown in Fig. 2D, GluN1 expression levels were not significantly different in wild-type and mutant cells [total GluN1 fluorescence, normalized to wild-type: 1.00±0.09 (GluN2B^WT, n=16 cells), 1.17±0.12 (GluN2B^724t, n=17 cells); P=0.2291]. These data indicate that the GluN2B^724t ASD-associated mutation prevents trafficking of NMDARs to the cell surface.

The last transmembrane domain and the C-terminal cytoplasmic tail of GluN2B both play a role in trafficking of GluN2B to the cell surface (Chung et al., 2004; Hawkins et al., 2004; Horak et al., 2008; Mori et al., 1998; Prybylowski et al., 2005; Sans et al., 2003; Steigerwald et al., 2000), and both are missing in the GluN2B^724t mutant. Therefore, we wondered whether the defects in GluN2B trafficking could be rescued by restoring the fourth transmembrane domain and cytoplasmic tail. To test this, we created a mutant that still lacked a significant portion of the deleted sequence in the extracellular loop but maintained the transmembrane domain and cytoplasmic tail. Interestingly, this deletion mutant, GluN2B^Δ724-786, was not trafficked to the cell surface (Fig. 2E,F; surface fluorescence normalized to wild-type: GluN2B^Δ724-786: 0.27±0.04, n=60 cells; wild-type: 1.0±0.12, n=55 cells, P<0.0001), phenocopying what was seen with the truncation mutant. Therefore, loss of a portion of the extracellular loop is sufficient to disrupt GluN2B trafficking.

The lack of trafficking of GluN2B to the cell surface could be a result of impaired co-assembly with the obligatory subunit, GluN1. Therefore, we next asked whether GluN2B^724t interacts with GluN1. To test this, lysates of HEK cells transfected with GluN1 and either GFP–GluN2B^724t or GFP–GluN2B^WT were subjected to immunoprecipitation with anti-GFP antibodies followed by immunoblotting for GluN1 and GluN2B (Fig. 2G). GluN1 co-precipitated with both wild-type and mutant GluN2B, suggesting that GluN2B^724t can co-assemble with GluN1.

Since the mutation prevented trafficking of GluN2B to the cell surface in HEK-293 cells, we next asked whether GluN2B^724t receptors are trafficked appropriately within neurons. To do so, we transfected wild-type rat cortical neurons with GFP–GluN2B^WT or GFP–GluN2B^724t. When the localization of GluN2B was examined in developing excitatory neurons, GluN2B^WT puncta were distributed throughout the dendritic arbor; however, GluN2B^724t failed to localize to dendritic puncta (Fig. 3A). Instead, GluN2B^724t was restricted to the soma and occasional proximal dendrites. In some cases, faint GFP–GluN2B^724t signal was observed a bit further into the dendritic tree, but in these cases the signal was diffuse rather than punctate. Within dendrites, GFP–GluN2B^WT puncta were often associated with dendritic spines, whereas mutant GFP–GluN2B puncta were not observed in spines (Fig. 3B).

The mutation also prevented trafficking to the surface of neurons. When neurons were transfected with either with GFP–GluN2B^WT or GFP–GluN2B^724t followed by live labeling with anti-GFP antibodies, GluN2B^WT was observed in discrete puncta at the surface of dendrites, as expected (Fig. 4A). In contrast, GluN2B^724t was not observed at the cell surface (Fig. 4A). Quantification of the level of surface GFP–GluN2B verified that the mutation prevented trafficking of GluN2B to the cell surface [Fig. 4B; surface GluN2B fluorescence, normalized to wild-type: 1.00±0.20 (GluN2B^WT), 0.11±0.05 (GluN2B^724t); P=2.95×10⁻⁴; n=13 GluN2B^724t and 14 GluN2B^WT neurons; mean±s.e.m.]. The lack of GluN2B^724t trafficking to dendrites and the neuronal cell surface was not due to reduced expression of the mutant protein since neurons chosen for analysis expressed GFP at similar moderate levels [Fig. 4C; total GluN2B, normalized to wild-type: 1.00±0.07 (GluN2B^WT), 1.18±0.22 (GluN2B^724t); P=0.4397].

The experiments above were performed in wild-type neurons expressing endogenous wild-type GluN1 and GluN2B. Unexpectedly, wild-type neurons that expressed the GluN2B^724t mutant had profoundly abnormal dendritic architectures (Fig. 5A). In contrast, neurons expressing GFP–GluN2B^WT had normal dendritic arbors (Fig. 5A). Two types of abnormal dendrite growth were observed when mutant GluN2B was present. In some mutant neurons, the dendrites appeared to be underdeveloped, with shorter and fewer branches than wild-type neurons (Fig. 5A,B). In other mutant neurons, dendrites were highly dysmorphic, often appearing as a mass of hairy extensions or lamellipodia-like structures protruding from the soma rather than bona fide branches. Neurons expressing GluN2B^WT almost always appeared normal and never appeared dysmorphic (Fig. 5B, 83.33% normal, 8.33% underdeveloped, 0% dysmorphic, n=12 neurons). Strikingly, neurons expressing GluN2B^724t rarely appeared normal (Fig. 5B; 6.25% normal, 43.75% underdeveloped, 31.25% dysmorphic, n=16 neurons). The abnormal structures of dysmorphic neurons typically lacked the dendrite marker MAP2 (Fig. S1), although MAP2 could be seen at the bases of some structures. MAP2 was observed in structures that otherwise appeared to be short dendrites that were thicker and longer than normal filopodia. Importantly, the observed effects of the GluN2B mutation on dendrite growth were cell autonomous since neurons were sparsely transfected, so each transfected neuron was surrounded by untransfected neurons.

To quantify the effects of GluN2B^724t on dendrite extension and branching, we performed Sholl analysis. In Sholl plots, there was a clear distinction between the dendritic architecture of neurons expressing GFP–GluN2B^WT and those expressing GFP–GluN2B^724t (Fig. 5C). While the number of branches close to the soma appeared to be similar, expression of mutant GluN2B dramatically reduced the number of branches found further from the soma. Abnormal dendrite morphology could result from changes in the number of dendrite branches, the length of dendrite segments, or both. To distinguish between these possibilities, we quantified these measures in several ways. GluN2B^724t mutant neurons had significantly fewer overall branches [Fig. 5D, 29.17±2.01 (wt) and 24.07±1.51 (724t) branch points per neuron; P=0.005, n=12 (WT) and 15 (724t) neurons; mean±s.e.m.] and fewer terminal branches [Fig. 5E, 36.0±1.87 (wt) and 30.06±1.67 (724t) terminal dendrites; P=0.0414], supporting a defect in branching. Mutant neurons also had reduced dendrite length (Fig. 5F, P<0.0001). GluN2B^WT and GluN2B^724t-expressing cells appeared to be similar with regard to the number of primary branches (Fig. 6A). However, clear differences were observed in intermediate (e.g. tertiary) and terminal dendrite branches (Fig. 6A). Deficits in dendrite length were observed throughout the dendritic arbor, regardless of branch order [Fig. 6B, primary: 38.09±3.70 μm/dendrite segment (wt) and 26.29±1.20 μm (724t), P=0.0014; secondary: 41.14±3.70 μm (wt) and 24.35±2.20 μm (724t), P=8.5×10⁻⁵; tertiary: 50.31±1.20 μm (wt) and 38.67±1.57 μm (724t) terminal: 57.70±3.49 μm (wt) and 42.64±4.05 μm (724t), P=4.16×10⁻⁶; n=12 (wt) and 15 (724t) neurons; mean±s.e.m.]. The pervasiveness of defects in the dendritic arbors of mutant neurons indicates a severe impairment in dendrite development.

The above results show that effect of this ASD-associated mutation in GRIN2B appears to be due to the mutant GluN2B interfering with dendrite growth. The alternative possibility is that overexpression of GluN2B^WT enhances dendrite growth, while the mutant is unable to promote growth. In support of the former hypothesis, the prevalence of highly abnormal and dysmorphic dendritic structures upon expression of GluN2B^724t suggests that the mutant protein produces a detrimental effect. In addition, when neurons were transfected with GFP–GluN2B^WT and then labeled with antibodies against GluN2B, the fluorescence intensities of GluN2B from transfected cells and neighboring untransfected neurons indicated that GluN2B^WT was not substantially overexpressed in dendrites, consistent with previous reports (Barria and Malinow, 2002; Philpot et al., 2001). Moreover, in contrast to what was seen for GFP–GluN2B^724t, neurons transfected with GFP–GluN2B^WT appeared similar to neurons transfected with only GFP (Fig. S2). These observations argue against the latter hypothesis. Together, these data suggest that this ASD-associated truncated GluN2B mutant interferes with normal dendrite development, even in the presence of wild-type GluN2B.

To determine whether loss of the extracellular loop was sufficient to produce the underdeveloped and dysmorphic dendrite phenotypes, we also evaluated the effects of the GluN2B^Δ724-786 mutant on dendrite architecture. The effects of the deletion mutant were qualitatively and quantitatively indistinguishable from those of GluN2B^724t (Fig. 5A–F; for all comparisons of GluN2B^Δ724-786 and GluN2B^724t mutants, P>0.3). Both underdeveloped and dysmorphic dendritic arbors were prevalent in GluN2B^Δ724-786 -expressing neurons (Fig. 5A,B). When compared to GFP-GluN2B^WT neurons, GluN2B^Δ724-786 neurons also had fewer branch points (Fig. 5D, 20.80±1.52 μm, P=0.0035; mean±s.e.m.), fewer terminal dendrites (Fig. 5E, 28.0±1.96 μm, P=0.0078) and reduced dendrite length (Fig. 5F, P<0.0001; n=15 GluN2B^Δ724-786 neurons and 12 GluN2B^WT neurons).

The mTOR pathway promotes dendrite growth (Feliciano et al., 2011; Fraser et al., 2008; Gallent and Steward, 2018; Jaworski et al., 2005; Kwon et al., 2006; Meikle et al., 2007; Zhou et al., 2009), and mutations in components of the mTOR pathway can lead to ASD (Gao and Penzes, 2015; Lin et al., 2016; Magdalon et al., 2017). In addition, NMDAR activation, and in particular activation of GluN2B-containing receptors, has been linked to mTOR activation (Burket et al., 2015; Iafrati et al., 2014; Miller et al., 2014; Wang et al., 2011). Therefore, an important question is whether mutations in GluN2B disrupt dendrite development by interfering with mTOR signaling. Phosphorylation of p70 ribosomal S6 kinase (pS6K) is a standard assay for activation of mTOR signaling, so we tested whether mTOR activity was altered upon expression of GluN2B^724t. To account for potential variability across networks, pS6K was quantified by calculating the ratio of pS6K immunofluorescence in GluN2B-transfected neurons versus neighboring untransfected neurons. We analyzed mTOR signaling within the cell body since most of the pS6K signal was localized to the soma. There was no statistically significant difference in the pS6K intensity ratio in GluN2B^724t neurons when compared to GluN2B^WT neurons [Fig. 7A,B; ratio of fluorescence in transfected and untransfected cells: 0.94±0.12 (wt) and 0.92±0.17 (724t); P=0.955; WT: n=8 transfected and 21 untransfected neurons, 724t: n=14 transfected and 40 untransfected neurons; mean±s.e.m.].

Inhibition of mTOR results in an overall reduction in cell size (Backman et al., 2001; Feliciano et al., 2011; Gallent and Steward, 2018; Kwon et al., 2001; Meikle et al., 2007; Tavazoie et al., 2005). If GluN2B^724t alters dendrite growth by disrupting mTOR signaling pathways, we would also expect to observe a reduction in soma size. The soma size was not significantly different between pyramidal-shaped neurons that expressed GluN2B^WT or GluN2B^724t (Fig. 7C; GluN2B^WT: 299.112±109.73 μm², n=13 and GluN2B^724t: 245.8066±59.94 μm², n=15; P=0.1162). Together, these data suggest that GluN2B^724t does not persistently decrease mTOR signaling. It remains possible that transient changes in mTOR signaling occurred earlier in development and contribute to the GluN2B^724t-dependent reduction in dendrite growth; however, dendrite length was reduced in the newest (terminal) branches, suggesting that the abnormal signaling mechanisms that are directly responsible for the dendrite growth phenotype persists through the age at which mTOR activation was evaluated here.

Since the ASD-associated mutation reduced dendrite arbor length and complexity, we next asked whether development of dendritic spines was also abnormal. To do so, we imaged tdTomato-filled neurons transfected with either GluN2B^WT or GluN2B^724t then measured dendritic spine densities (Fig. 8). Although the overall dendritic spine density for GluN2B^724t neurons generally appeared to be lower than GluN2B^WT neurons (Fig. 8A–C), the apparent difference was not significant (P=0.27, one-way ANOVA). The average spine density per 40 μm of dendrite length for GluN2B^WT-expressing neurons was 5.7±0.8 (n=7 dendrites; mean±s.e.m.) and for GluN2B^724t-expressing neurons was 4.5±0.6 (n=15 dendrites; Fig. 8C). Since the GluN2B^724t mutation reduced dendrite lengths without an increase in spine density, we infer that GluN2B^724t neurons have fewer synapses overall, which may contribute to ASD pathology.

As neurons develop, dendritic spine morphology changes, and it is thought that these changes in morphology reflect spine maturation (Berry and Nedivi, 2017). For example, it is thought that filopodia and stubby spines are immature structures, while mushroom spines are mature. Therefore, we examined spine morphology upon expression of either GluN2B^WT or GluN2B^724t. Spines were classified as filopodial, thin, stubby or mushroom. When wild-type and mutant neurons were compared, we did not observe significant differences in the distribution of spines among mature and immature morphological classes (Fig. 8D; GluN2B^WT: n=39 spines, GluN2B^724t: n=64 spines). For example, there was no change in either immature forms, such as filopodia and stubby spines (P=0.6076 and 0.1546, respectively), or in mushroom spines, a mature morphology (P=0.4272). In contrast, the percentage of spines classified as thin was elevated in neurons expressing GluN2B^724t (Fig. 8D; GluN2B^WT: 12.5±6.3% of spines, n=39 spines, GluN2B^724t: 40.7±8.0% of spines, n=64 spines; P=0.0354 by ANOVA; mean±s.e.m.).

Human and mouse studies have suggested that NMDA receptor dysfunction is a common pathogenic mechanism in ASD (Gao and Penzes, 2015; Uzunova et al., 2014). As a result, the ‘NMDAR dysfunction hypothesis’ of ASD has been proposed (Lee et al., 2015). In most studies of ASD model animals, abnormal NMDAR signaling or expression has been observed as a consequence of mutations in other genes (Gao and Penzes, 2015; Uzunova et al., 2014). Therefore, it is unclear whether altered NMDAR signaling causes ASD or, instead, is an epiphenomenon that does not itself lead to ASD. Here, we directly tested whether NMDARs play a pathogenic role in ASD by investigating the effects of ASD-associated mutations within the GluN2B subunit of NMDARs. We find that an ASD-associated de novo mutation eliminates receptor trafficking to the cell surface but not assembly with wild-type receptor subunits, and homozygous mutant receptors completely lack Ca²⁺ currents. Expression of ASD mutant GluN2B in cortical neurons resulted in aberrant dendrite arborization. Mutant GluN2B was sparsely expressed in neurons; therefore, the observed effects on dendrite growth were cell autonomous. Overall, our results suggest that ASD-associated GluN2B mutations contribute to ASD pathogenesis by interfering with NMDAR signaling and trafficking, which in turn disrupts development of dendrites. Therefore, this study provides strong evidence that NMDA receptor dysfunction plays a direct pathogenic role in ASD by disrupting neuronal development.

Our results indicate that ASD-associated mutations in GluN2B impair dendrite growth and patterning in cortical excitatory neurons. Consistent with our findings, several reports have shown GluN2B-containing receptors are required for normal NMDAR-dependent dendrite growth and refinement, since direct manipulation of GluN2B expression or genetic swapping of GluN2A with GluN2B is sufficient to induce abnormal dendrite architecture in a variety of neuron types (Bustos et al., 2014; Espinosa et al., 2009; Ewald et al., 2008; Keith et al., 2019; Sepulveda et al., 2010). Interestingly, when GluN2B-knockout neurons were compared to GluN2B heterozygous neurons in vivo, the patterning of dendrites was abnormal in knockout neurons but dendrite length and branching were unaltered in granule cells and cortical layer 4 spiny stellate cells in barrel cortex (Espinosa et al., 2009). In the case of the ASD-associated mutation studied here, both the length of dendrites and the number of dendrite branches were affected when the mutant receptor was co-expressed with wild-type receptors.

Where does this difference in phenotype come from? Based on the effects of NMDARs and signaling pathways that act downstream of NMDARs, it has been proposed that regulation of dendritogenesis by NMDARs is cell type specific (Lohmann and Wong, 2005). The present study targeted cortical pyramidal neurons for analysis, while the knockout study characterized layer 4 spiny stellate neurons from barrel cortex and dentate gyrus granule cells (Espinosa et al., 2009). Therefore, the cortical neurons studied here could have a stronger dependence on GluN2B-containing receptors, while GluN2B could play less of a role in spiny stellate and granule cells. Importantly, it is widely accepted that cortical pyramidal neurons contribute to ASD pathophysiology (Gao and Penzes, 2015; Varghese et al., 2017), making them an ideal model to investigate the effects of GluN2B mutations.

Abnormal dendrite morphogenesis appears to represent a common cellular phenotype seen in individuals with ASD and ASD animal models. Analysis of the brains of individuals with ASD has frequently revealed decreased dendritic arborization (Lin et al., 2016; Varghese et al., 2017). In addition, defective dendrite development has been reported in several mouse models of syndromic ASD and valproate exposure, which dramatically increases the risk for ASD (Lin et al., 2016; Martínez-Cerdeño, 2017; Varghese et al., 2017). Interestingly, ASD-associated mutations in CaMKII interfere with dendrite arborization and prevent binding of CaMKII to GluN2B (Stephenson et al., 2017). Here, we found that dendrite development was impaired by mutations in GluN2B that lead to non-syndromic ASD. Since confounding comorbidities are minimized in non-syndromic ASD, observing this neurodevelopmental phenotype in a model of non-syndromic ASD provides strong support for a role for dendrite maldevelopment in ASD pathophysiology.

Here, the ASD mutation analyzed leads to a deletion of the cytoplasmic tail of GluN2B, along with the fourth transmembrane domain and a substantial portion of the S2 extracellular loop. It has been shown that the C-terminus of GluN2 is important for regulation of dendrite growth and branching by GluN2B-containing receptors (Bustos et al., 2014; Keith et al., 2019). Residues in the C-terminus of GluN2B and GluN2A are also essential for NMDAR trafficking to dendrites and synapses (Chung et al., 2004; Hawkins et al., 2004; Horak et al., 2008; Mori et al., 1998; Prybylowski et al., 2005; Sans et al., 2003; Steigerwald et al., 2000). In GluN2A, the analogous fourth transmembrane domain (M4) contains residues important for channel gating, and these residues are conserved in GluN2B (Ren et al., 2003). Moreover, receptors with deletion of the M4 domain from GluN2A do not form functional channels (Schorge and Colquhoun, 2003). Therefore, we asked whether the cytoplasmic tail and/or M4 mediate the effects of this ASD mutation on NMDAR trafficking, function and dendrite development. Unexpectedly, by analyzing the effects of deletion mutants, we found that these regions were not necessary for the observed deficits. Instead, deletion of a portion of the S2 extracellular loop was sufficient to reproduce all of the phenotypes observed with the truncation mutant. Consistent with our observations, this portion of the extracellular loop includes residues that are involved in ligand binding, and glutamate binding is necessary for trafficking of this protein through the secretory pathway (She et al., 2012).

Recent genetic analysis of de novo mutations has identified additional ASD-associated mutations in GluN2B. Consistent with the observations reported here, several of these mutations produce NMDA receptors with reduced or abolished channel properties or trafficking to the cell surface in heterologous expression systems (Adams et al., 2014; Fedele et al., 2018; Liu et al., 2017; Swanger et al., 2016; Vyklicky et al., 2018). It will be important in future studies to compare the effects of distinct ASD mutations on NMDAR trafficking and function within neurons and to determine whether abnormal dendrite development is a common phenotypic outcome of ASD-associated GluN2B mutations.

To date, studies have identified more than 25 disease-associated variations within the S2 extracellular ligand-binding loop and the linker sequences that connect S2 to transmembrane domains 3 and 4 (SFARi gene, https://gene.sfari.org, July 2019). These mutations may produce cellular phenotypes that are similar to those presented here for GluN2B^724t and GluN2B^Δ724-786. In particular, analogous phenotypes seem likely for variants with predicted truncations within this region: c.2131C>T, which produces a premature stop at amino acid 711 (Kenny et al., 2014) and c.2384_2391del, which produces a premature stop at amino acid 795 (Stessman et al., 2017). A variant c.2539C>T, predicted to truncate GluN2B at amino acid 847 within the fourth transmembrane domain (Platzer et al., 2017) and the splice site variant c.2360-2A>G (Endele et al., 2010) may result in similar effects, as well. Moreover, in future studies, it will be interesting to determine whether any mutation that interferes with glutamate-binding produces deficits in subcellular trafficking and dendrite morphogenesis, as observed for GluN2B^724t.

The detrimental effect of mutant GluN2B on dendrite growth in the presence of two copies of normal wild-type GRIN2B raises the question of whether ASD-associated GRIN2B mutations act via a dominant-negative mechanism, a gain of toxic function or haploinsufficiency. As individuals with ASD-associated de novo mutations are heterozygous for the disrupting mutation, it is widely assumed that these mutations will result in haploinsufficiency for the mutated gene. However, our data demonstrate that mutant GluN2B impairs dendrite growth in the presence of two copies of normal wild-type GRIN2B. Furthermore, loss of one GRIN2B allele in heterozygous knockout mice does not significantly reduce GluN2B-dependent post-synaptic responses (Espinosa et al., 2009). Together, these observations argue against a simple haploinsufficiency mechanism. To distinguish between dominant-negative mechanisms and toxic gain-of-function, it will be necessary to determine whether wild-type NMDAR function or localization are impaired upon expression of mutant GluN2B. In addition, it will be important to consider both ionotropic and unconventional, ion flux-independent functions of NMDARs (Burket and Deutsch, 2019; Dore et al., 2017; Keith et al., 2019; Montes de Oca Balderas, 2018; Keith et al., 2019; Montes de Oca Balderas, 2018; Dore et al., 2017). Regardless of the specific mechanism involved, the differences between expression of GluN2B bearing a specific ASD mutation and heterozygous deletion of GRIN2B underscores the importance of evaluating specific ASD-associated mutations when investigating etiological mechanisms of ASD.

We observed reduced dendrite arbor length and complexity in GluN2B^724t-expressing neurons, which is expected to alter neural circuit function by changing dendrite cable properties (Lefebvre et al., 2015). Future studies should be aimed at directly measuring the effects of GluN2B mutations and the resulting dendrite maldevelopment on dendrite cable properties. In addition, this study focuses on cortical excitatory neurons since they are the principal neurons responsible for carrying feed-forward information within the cerebral cortex. In order to understand the ultimate impact on circuit function, it will also be important in the future to determine whether similar changes are observed in cortical inhibitory neurons and whether excitatory/inhibitory balance is shifted.

Understanding the precise mechanisms of action of individual mutations in strong candidate ASD genes is essential to unraveling the complex etiology that leads to ASD. Studying individual mutations will help the field move toward identifying core neurodevelopmental phenotypes that characterize ASD or subtypes of ASD. In addition, understanding common phenotypes will be informative with regards to which potential therapeutic approaches might be beneficial for individual patients. For example, augmenting the activity of GluN2B-containing receptors might be beneficial in the case of mutations that severely disrupt NMDAR function, signaling pathways activated by NMDARs or dendrite development.

All animal studies were conducted according to approved protocols from the Case Western Reserve University Institutional Animal Care and Use Committee or the Central Michigan University Institutional Animal Care and Use Committee, in compliance with the National Institutes of Health guidelines for care and use of experimental animals.

pEGFP-GluN2B was Addgene plasmid #17925 (deposited by Stefano Vicini; Luo et al., 2002). GFP is inserted in the N-terminal extracellular domain after the signal peptide. GluN1–HcRed and GluN1–CFP expression plasmids were from Phillip Washbourne. Untagged GluN2B, GluN2A and GluN1 plasmids were gifts of John Woodward. The tdTomato-N1 plasmid was Addgene plasmid #54642 (deposited by Michael Davidson and Roger Tsien).

Mutant GluN2B cDNA constructs were made from pEGFP-GluN2B using the Q5 site-directed mutagenesis kit (New England Biolabs). For GluN2B^724t, a stop site was introduced using the following primers, designed using NEBaseChanger (New England Biolabs): 5′-GAAAACAGGGTACGCTTGATGCATTC-3′ and 5′-AGGGAGAGCAATGCATCATC-3′. Mutations were confirmed by sequencing.

HEK-293 cells (ATCC, low passage) were cultured in Dulbecco's modified Eagle's medium (Invitrogen or HyClone) supplemented with 10% fetal bovine serum (Invitrogen or HyClone), penicillin and streptomycin (Invitrogen) and maintained at 5% CO₂. HEK cells were used at a low passage number and displayed appropriate morphology. For imaging, HEK cells were plated on poly-lysine-coated glass coverslips (Carolina Scientific). HEK-293 cells were transfected using Lipofectamine 2000 (Invitrogen) in the absence of antibiotics, essentially as previously described (Bury and Sabo, 2014; Gill et al., 2015). To prevent excitotoxicity, kynurenate (1 mM) or ketamine (1 μg/ml) was added at the time of transfection. Cells were used in assays at 24–72 h after transfection.

Neurons and astrocytes were derived from the cortices of wild-type Sprague Dawley rats (Charles River or Taconic Biosciences) of both sexes at 0–1 days postnatal. Neurons were dissociated and grown on a confluent monolayer of astrocytes, as previously described (Bury and Sabo, 2011, 2014; Sceniak et al., 2012). Neurons were maintained at 5% CO₂ in Neurobasal medium (Invitrogen) with B27 supplement (Invitrogen). Neurons were transfected at 2 DIV using the Ca²⁺ phosphate method (Berry et al., 2012) with 2.6 µg GluN2B DNA/18 mm coverslip. Neurons were co-transfected with tdTomato (0.4 µg DNA/coverslip) to fill transfected cells. For some surface localization experiments, neurons were transfected at 3 DIV with GFP–GluN2B expression plasmid. Transfection efficiency was sufficiently low that a large majority of transfected neurons were excitatory neurons, based on pyramidal-shaped somas and spiny dendrites.

HEK-293 cells were loaded with 5 µM Fluo4-AM (Invitrogen) for 30 min at room temperature. Imaging was performed in Mg²⁺-free artificial cerebrospinal fluid (ACSF; 120 mM NaCl, 3 mM KCl, 2 mM CaCl₂, 30 mM D-glucose, 20 mM HEPES and 0.2% sorbitol, pH 7.3). For each condition, 60 images were collected at a rate of 1 frame per second. Stimulation was via perfusion with ACSF containing NMDA (100 µM) in the presence of glycine (10 µM). After washout, stimulation was repeated in the presence of ketamine (1 μg/ml). Only healthy cells, based on morphology in differential interference contrast images, were selected for analysis, and health assessments were made while the researcher was blind to fluorescence information.

Immunocytochemistry was generally performed on neurons at 10–14 DIV. In most cases, neurons were fixed with 4% paraformaldehyde and 0.32 M sucrose in phosphate-buffered saline and then permeabilized using 0.1% Triton X-100 in PBS. For GluN1 immunolabeling, neurons were fixed in −20C methanol then permeabilized in Triton X-100 (0.025%). After blocking with 10% horse serum or bovine serum albumin, neurons were labeled with primary antibodies, applied overnight at 4°C in the presence of 3% horse serum or bovine serum albumin. Primary antibodies were chicken anti-GFP (1:2500; AB16901, Millipore), rabbit anti-GFP (1:2000; A-11122, Invitrogen); mouse anti-GluN1 (1:100; clone 54.1, Millipore); rabbit anti-GluN2B (1:300; AGC-003, Alomone); rabbit anti-phosphorylated S6 (1:800; D68F8, Cell Signaling Technologies); and chicken anti-MAP2 (1:2000; AB5543, Millipore) antibodies. Secondary antibodies were Alexa Fluor conjugates (Invitrogen). Coverslips were mounted in Fluoromount (Thermo Fisher Scientific, Pittsburgh, PA) containing 1,4-diazabicyclo[2.2.2] octane (DABCO) (Sigma-Aldrich, St Louis, MO).

For surface labeling of HEK cells, cells were incubated with rabbit anti-GFP (1:2000, A11122, Invitrogen) and then Alexa Fluor 633-conjugated anti-rabbit-IgG (Invitrogen) antibodies for 30 min at 4°C or chicken anti-GFP followed by Alexa Fluor 550-conjugated anti-chicken-IgY antibodies. The cells were then fixed for 15 min in 4% paraformaldehyde in phosphate-buffered saline (PBS) containing 4% sucrose. For surface labeling of neurons, primary antibodies were applied to living neurons prior to fixation, and secondary antibodies were applied after fixation. Neuronal surface immunofluorescence images were collected after 6–13 DIV.

Fluorescence was not observed in the absence of primary antibody. Labeling was absent with anti-GFP antibodies when cells were not transfected. Untransfected HEK-293 cells did not display fluorescence with anti-GluN1 and -GluN2B antibodies.

Cells were imaged with a Nikon C1 Plus confocal system with an Eclipse Ti-E microscope. Lasers were 488 nm argon, 543 nm helium-neon and 633 nm helium-neon. Detection filters were 515/30 nm bandpass, 590/50 nm bandpass, and 650 LP. Some images were collected using a Nikon C2 plus confocal system with an Ni-E microscope, 405, 488, 561 and 640 nm diode lasers and 525/50 nm bandpass, 600/50 nm bandpass, and 660 LP detection filters. For live time-lapse imaging, a perfect focus system was used to maintain focus. For both systems, objectives used were 60× (1.4 NA), 20× (NA 0.75) or 40× (0.95 NA) Plan Apo objectives.

Large neurons with pyramidal somas were chosen for analysis. These were likely excitatory neurons since they were pyramidal and spiny. Only healthy cells were imaged, as determined by the appearance of the soma under differential interference contrast and a smooth axon without fragmentation or blebbing. Within a given experiment, neurons were also chosen with similar transfection levels, as judged by tdTomato and GFP fluorescence. Stacks of images (Z-stacks) were taken in order to span the entire cell volume. Gains were set to avoid saturation and kept constant during imaging of each experiment when intensities, morphologies or areas were quantified. All channels were imaged separately and sequentially to avoid bleed through. Images were taken with pixel sizes ranging from 90 (for synaptic structures) to 207 nm (for imaging complete dendritic arbors).

Sholl analysis was conducted essentially as previously described (Kutzing et al., 2010; Langhammer et al., 2010; O'Neill et al., 2015). Dendrites were manually traced in NeuronJ based on cytosolic tdTomato fills, while the user was blind to the condition. Traces were transferred to BonFire for further analysis. Dendrites were considered to be extensions that were at least 10 µm long. Primary dendrites emanate directly from the soma, while remaining orders are defined by the successive branch points between the soma and the relevant dendrite.

Dendritic spines were measured in Z-projections from secondary and, occasionally, tertiary dendrite branches of pyramidal neurons. Dendrite segments were generally clear from intersecting branches and at least 40 µm in length. Spines were analyzed by observers blind to the experimental condition using ImageJ/Fiji. Dendritic spines were counted in a semi-automatic manner, as previously described (Orlowski and Bjarkam, 2012). Briefly, images of tdTomato fills were automatically thresholded using the ImageJ ‘Threshold’ function. Then, dendrites and spines were automatically detected by using two sequential applications of the ‘Skeletonize’ function in ImageJ. Dendritic spine lengths and widths were measured manually using ImageJ. Any dendritic protrusions over 10 μm in length or not clearly focused within the imaged volume were discarded from the analysis. Spine morphologies were classified as filopodial, long thin, thin, stubby, mushroom or branched based on length, width and length:width ratios, as previously described (Risher et al., 2014). Briefly, filopodia were defined as having lengths greater than 2 µm; stubby spines had length:width ratios less than 1; mushroom spines had widths greater than 0.6 µm; and thin spines had lengths less than 2 µm and length:width ratios greater than 1.

Transfected cells were lysed on ice for 30 min in lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 2 mM EGTA, 1% Triton X-100, and protease and phosphatase inhibitors). The 20,000 g supernatant was then incubated with rabbit anti-GFP antibodies followed by protein A/G agarose beads that had been pre-washed with lysate. Bound proteins were subjected to SDS-PAGE, using 4–20% Tris-glycine gels (Life Technologies), then transferred to nitrocellulose membranes. After blocking with 10% non-fat dry milk in Tris-buffered saline with Tween-20, membranes were probed with rabbit-anti-GFP, mouse anti-GluN1 and IRDye-conjugated secondary antibodies (LiCor, Lincoln, NE) or HRP-conjugated (BioRad) secondary antibodies. Anti-GFP and anti-GluN1 antibodies are described above. Blots were imaged using a LiCor Odyssey imaging system or chemiluminescence.

Images were analyzed using custom-written macros and functions in ImageJ and Matlab (Mathworks). These macros are available via email request. Maximum intensity projections were created, and individual transfected cells or regions of interest were outlined based on fluorescent fills and image statistics. Mean fluorescence intensity was measured for each channel quantified. Fluorescence intensity data were analyzed for statistical significance using ANOVA, with Fisher's PLSD post-hoc tests. Multiple comparisons corrections were applied where appropriate (Figs 5 and 6) by using the ‘multcompare’ function in Matlab with Bonferroni corrections. Spine densities were compared by one-way ANOVA, and spine morphology types were compared through ANOVA with Bonferroni's multiple comparisons correction.

We thank Jacob Bahry for helpful reading of the manuscript. We thank Sahil Patel, Natuwa Basalirwa and Jessica Lahr for technical assistance.