X-linked neonatal-onset epileptic encephalopathy associated with a gain-of-function variant p.R660T in GRIA3

The X-linked GRIA3 gene encodes the GLUA3 subunit of AMPA-type glutamate receptors. Pathogenic variants in this gene were previously reported in neurodevelopmental diseases, mostly in male patients but rarely in females. Here we report a de novo pathogenic missense variant in GRIA3 (c.1979G>C; p. R660T) identified in a 1-year-old female patient with severe epilepsy and global developmental delay. When exogenously expressed in human embryonic kidney (HEK) cells, GLUA3_R660T showed slower desensitization and deactivation kinetics compared to wildtype (wt) GLUA3 receptors. Substantial non-desensitized currents were observed with the mutant but not for wt GLUA3 with prolonged exposure to glutamate. When co-expressed with GLUA2, the decay kinetics were similarly slowed in GLUA2/A3_R660T with non-desensitized steady state currents. In cultured cerebellar granule neurons, miniature excitatory postsynaptic currents (mEPSCs) were significantly slower in R660T transfected cells than those expressing wt GLUA3. When overexpressed in hippocampal CA1 neurons by in utero electroporation, the evoked EPSCs and mEPSCs were slower in neurons expressing R660T mutant compared to those expressing wt GLUA3. Therefore our study provides functional evidence that a gain of function (GoF) variant in GRIA3 may cause epileptic encephalopathy and global developmental delay in a female subject by enhancing synaptic transmission.

Thus far, about 20 pathogenic variants in GRIA3 have been reported, as summarized in a recent paper [12]. The majority of affected individuals are males with the pathogenic variant inherited from unaffected mothers [12,13]. However, pathogenic alterations in GRIA3 have been reported in two female individuals as well. An early work found that a female with bipolar disorder and intellectual disability carried a balanced translocation affecting GRIA3 [14]. Recently, a young female with early-onset epileptic encephalopathy was reported to harbor a de novo variant, p.A248V [15]. However, these two papers presented no functional testing that could support the clinical findings. Therefore functional analyses are needed to explore if and why single nucleotide variants in GRIA3 cause disease in females.
In this study, a de novo variant in GRIA3 was identified by clinical whole genome sequencing in a 1-year-old female with severe epilepsy and global developmental delay. We then systemically studied the physiological function of this mutant in HEK cells and neurons to determine the potential impact of this variant on protein function. The mutant GLUA3_R660T displayed slower desensitization and deactivation kinetics in HEK cells and neurons, suggesting it is a Gain-of-Function variant. Our data demonstrates the R660T mutant causes prolonged activity of AMPARs in brain, explaining the potential mechanisms in epileptogenesis. when asleep. She was intubated for 20 days and remained in the neonatal intensive care unit for 30 days. An electroencephalography confirmed a diagnosis of epilepsy ( Fig 1A) but due to limited access to her medical files, we were unable to obtain further clinical data regarding the neonatal period. She did not pass the newborn hearing screen but at 13 months her follow-up brainstem audiogram was normal.
At five months of life, she was treatment-resistant despite being on three anticonvulsants (vigabatrin, phenobarbital and valproate) and still had many daily brief myoclonic jerks along with several weekly bilateral tonic clonic seizures. She remained hypertonic and had brisk deep tendon reflexes. She did not show any facial dysmorphism but had a large inguinal hernia. The hernia recurred after initial repair. An ophthalmological examination showed cortical vision impairment and nystagmus. A brain magnetic resonance imaging performed in the newborn period was of poor quality but showed some thinning of the corpus callosum.
She is kept on levetiracetam, valproate acid, vigabatrin, cannabis oil, and clobazam but remains treatment-resistant. She still has both myoclonic jerks and bilateral tonic-clinic seizures. Nevertheless, epilepsy burden has reduced from 30 to 10 seizures per day and convulsions are also shorter in duration. She is unable to feed orally and depends on a gastrostomytube. In addition, she presents with profound developmental delay. Currently, she has some head control but cannot support herself in a seated position and has severe growth restriction.
Whole genome sequencing revealed a heterozygous de novo missense variant in the GRIA3 gene (NM_000828.4:c.1979G>C; p.R660T) (Fig 1B) when the patient was 1-year old. This variant affects a highly conserved amino acid ( Fig 1C) and is located in the linker between the third transmembrane domain (M3) and the S2 extracellular domain of glutamate binding domain of AMPAR subunit (Fig 1C and 1D) [16]. Bioinformatic predictions are in favor of its pathogenicity (SIFT pathogenic (score 0); polyphen2, probably damaging (score 0,985). This variant was absent from the Genome Aggregation Database (gnomAD) and was not present the ClinVar or HGMDpro databases at the time of analysis. No additional variants of clinical interest in genes associated with epilepsy or other aspects of the patient's phenotype were identified.

GLUA3_R660T mutant slowed deactivation and desensitization kinetics
To evaluate the effects of the GLUA3_R660T on functional properties of AMPA receptors, we transfected HEK293 cells with expression constructs of human GLUA3 cDNA or R660T variant. To visit the transfection, EGFP was linked to GLUA3 by an Internal Ribosome Entry Site (GLUA3-IRES-EGFP) to express GLUA3 and EGFP separately. Outside-out patched were excised and saturating concentration of glutamate (10 mM) was applied by a fast piezoelectric system [17]. It is known that glutamate release into synaptic cleft is with the concentration of millimolar and reabsorbed with time frame about 1 ms under normal physiological condition [18]. Therefore, the deactivation kinetics recorded by a 1 ms brief application of glutamate mimics the synaptic released glutamate in brain. The deactivation kinetics was slower for GluA3_R660T than wt GLUA3 (Fig 2A). When the extracellular glutamate clearance is impaired under pathophysiological conditions such as in epilepsy, the postsynaptic glutamate receptors are exposed in high concentration of glutamate for prolonged period of time. The desensitization kinetics recorded by prolonged exposure (500 ms) to glutamate would mimic this condition. The desensitization kinetics of R660T was also slower than that of wt GLUA3 ( Fig 2B). Furthermore, R660T produced a substantial non-desensitized steady-state current, which was about 30% of the peak current and absent in wt GLUA3 ( Fig 2B).
We further wondered whether the R660T had an effect on single channel conductance. Non-stationary fluctuation analysis (NSFA) provides a convenient method to measure single channel conductance [19]. We found that neither single channel conductance nor the peak open probability was affected by R660T (Fig 2C-2E).

GLUA3_R660T mutant slows gating kinetics of GLUA2/A3 receptor
In the brain, GLUA3 forms heteromeric Ca 2+ -impermeable AMPARs with GLUA2 [5]. The current-voltage (I-V) curves of GLUA3 homomers are rectifying and GLUA2/A3 heteromers are linear [20][21][22]. To examine whether the R660T variant affects the capability of GLUA3 to form heteromeric receptors, the mutant was co-expressed with edited GLUA2 subunits (GLUA2R) in HEK cells. In the absence of GLUA2, the rectification of the mutant was the same as wt GLUA3 (Fig 3A). In the presence of GLUA2, wt GLUA3 showed a linear IV-curve, indicating GLUA2/A3 are the major receptors under the condition. GLUA3_R660T coexpression with GLUA2 also showed a linear IV-curve with identical rectification index as the wt GLUA2/A3, demonstrating that the mutant did not change its capability to form heteromeric receptors with GLUA2 ( Fig 3B). We then examined the gating kinetics of GLUA2/A3 heteromers, and found that the deactivation and desensitization kinetics of GLUA2/A3 receptors were also slowed by the R660T mutant (Fig 3C and 3D). In addition, GLUA2/A3_R660T displayed non-desensitized steady state currents during prolonged exposure to glutamate ( Fig  3D). In summary, these data suggested that the mutant primarily slowed AMPAR gating kinetics.
We also examined cornichon family AMPA receptor auxiliary protein 2 (CNIH2), another auxiliary subunit of AMPARs, on the gating kinetics on GLUA3_R660T. We found that the deactivation and desensitization of GLUR3 with or without GLUA2 were slowed by CNIH2 (S1 Fig), while R660T further enhanced the slowing.

GLUA3_R660T variant slows evoked AMPAR-EPSCs in hippocampal neurons
We then wondered if the GLUA3 variant affects the AMPAR function in vivo. To test this possibility, we transfected wt or mutant GLUA3 into hippocampal neurons via in utero electroporation. Acute hippocampal slices were prepared from P21-P28 puppies. Synaptic functions of glutamate receptors were analyzed by dual whole-cell recordings on a transfected neuron and a neighboring control neuron simultaneously ( Fig 6A). We found that overexpression of wt GLUA3 reduced the peak amplitudes of evoked AMPAR-EPSCs (AMPAR-eEPSCs) while the GLUA3_R660T did not (Fig 6B-6E). However, the AMPAR-mediated charge transfer was increased by R660T compared to the control neurons (Fig 6F), indicating that the receptor channel may open longer. Indeed, analysis of the decay kinetics of AMPAR-eEPSCs showed that wt GLUA3 speeded up the decay of AMPAR-eEPSCs while the R660T slowed it ( Fig 6G). Consistently, the mEPSCs were speeded up by wt GLUA3 while slowed by R660T (Fig 6H). Paired pulse ratio (PPR), a measure of the release probability of presynaptic neurotransmitters in neurons, was unaltered by overexpression of GLUA3 and R660T (Fig 6I). NMDAR-eEPSCs were unaltered in GLUA3 and R660T transfected neurons (Fig 6J-6M). These data thus provide evidence supporting that the R660T variant slows the decay of synaptic AMPARs in vivo.

Discussion
Since the first description, about 20 GLUA3 pathogenic variants have been reported, including a balanced translocation in which the breakpoints disrupted the GRIA3 gene, a deletion, duplications, and missense variants. Most of them are found in males with X-linked ID together possibly associated with dysmorphic features or epilepsy, and are inherited from unaffected mothers [12,13]. Two de novo variants have been reported in females, one with bipolar symptom and ID who carries a balanced translocation involving GRIA3 [14], and another with epilepsy for whom a de novo p.A248V variant was identified [15]. Our case is the third affected female to be reported. She presented with a devastating neurological phenotype compatible with a developmental and epileptic encephalopathy. Symptoms included treatment resistant neonatal-onset epilepsy, congenital hypertonia, and severe developmental delay.
The variant affects a highly conserved amino acid (R660T) and is located in the extracellular linker 2 of AMPAR, between M3 transmembrane domain and S2 glutamate binding domain [16]. Recombinant expression in HEK cells showed that the R660T variant causes slowing of deactivation and desensitization of GLUA3 homomeric receptors as well as GLUA2/A3 heteromeric receptors. By co-expression with TARP γ-2 and CNIH2, we also demonstrate that the slowing of channel kinetics are further enhanced by AMPAR auxiliary subunits TARPs and cornichons [19,32]. When overexpressed in both cultured CGNs and hippocampal neurons, the variant slows the decay kinetics of miniature and evoked AMPAR-EPSCs, consistent with the observations in HEK cells. Interestingly, an early study demonstrates that mutations on the corresponding sites in rodent GluA1, GluA2 and kainate-type glutamate receptor GluK2 slow the ion channel kinetics [33]. Our study on GLUA3_R660T thus provides further homomeric receptors. Up panel, Desensitization curves were recorded while the holding potential was elevated from -100 mV with a step of 20 mV to +100 mV. Insert shows the voltage protocol. Dark area represents drug application (10 mM glutamate for 200 ms). Low panel, I-V curve. Peak currents were normalized to the absolute value of the peak current amplitudes recorded at -100 mV. (B) I-V relationship for GLUA2/A3 and GLUA2/A3_R660T. (C) The deactivation of GLUA2/A3 receptors was slowed by R660T variant. GLUA2/A3, 0.9 ± 0.1 ms, n = 12; GLUA2/A3_R660T, 1.5 ± 0.2 ms, n = 11; �� p = 0.0083. (D) Up, the sample traces for desensitization of GLUA2/A3 and GLUA2/A3_R660T. Low left, statistics of weighted τ des . GLUA2/A3, 6.7 ± 0.7 ms, n = 7; GLUA2/A3_R660T, 12.8 ± 0.9 ms, n = 10; ��� p < 0.001. Low right, statistics of steady state currents. GLUA2/A3, 0.9 ± 0.3%, n = 7; GLUA2/A3_R660T, 12.6 ± 2.6%, n = 9; �� p = 0.0015. Data are presented as mean ± SEM. Unpaired t-test was used for data analysis. It is widely accepted that glutamate-medicated hyperexcitability of neural circuits plays a causative role in seizure generation [34,35]. Intracerebral injection of glutamate or glutamate receptor agonists into laboratory animals causes epileptic seizures [36]. Disturbance of extracellular glutamate clearance also causes epilepsy [37]. Cyclothiazide, a potent AMPAR desensitization blocker, induces seizure in rodents [38], suggesting that enhancing postsynaptic glutamate receptor function also leads to epilepsy. Recently variants in NMDA receptors and AMPARs have been reported to cause epilepsy [39,40]. Such AMPAR dynamic anomalies could be secondarily reinforced and worsened as epileptic seizures cause fast release and extracellular accumulation of glutamate, which further induces excitotoxicity and neural damage [9,35].
https://doi.org/10.1371/journal.pgen.1009608.g004  providing tailored molecular diagnosis enabling precision medicine in approximately one quarter of patients, illustrating the enormous utility of genetic testing for therapeutic decisionmaking [41,42]. A major goal of genetic studies is the identification of novel drug targets and tailored therapies based on the etiology of disease. The discovery of specific genetic variants has also helped us to repurpose drugs with specific actions which may have been used in entirely unrelated conditions.
In the case of GLUA3 loss-of-function variants there are no FDA or MDA approved treatment options, however in the case of GoF variants, one drug is available: Perampanel (PER) which is an orally active, selective, non-competitive alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor agonist [43]. Although there are several publications describing the safety and efficacy of PER for individuals with epilepsy [43][44][45], there are no similar data available in individuals with GRIA3 GoF variants. An empirical antiepileptic treatment in monogenic disorders including GRIA3-deficiency is often found ineffective, may cause unwanted side effects and thus ultimately result in a diminished quality of life. This is also demonstrated by our case which highlights the urgency for precision treatment options. The GoF property of GLUA3_R660T suggests it enhances and expands excitatory glutamate signaling in brain. Functional analysis on the mutants would lead to some clues on the therapy strategy. Our study suggests that the slow decay kinetics is likely the most relevant characteristic of the mutant channel. Since auxiliary subunit TARPs and cornichons are likely the most important factors in regulating gating kinetics, therapies targeting those auxiliary subunits may represent a promising field.

Ethics statement
The study was conducted in agreement with the Declaration of Helsinki and approved by the Ethics Committee of the Strasbourg University Hospital, approval number CE-2021-72. Formal consent for genetic testing and publication was obtained from the parents. The detailed information about seisure semiology, neurologic examination (EEG and MRI) and treatment outcomes were collected following interview with parents and by reviewing the proband medical files.
To identify candidate variants of potential clinical interest, variants were filtered and considered based on multiple factors including population allele frequency, variant consequence, evolutionary conservation, occurrence in a gene with a well-established gene-disease relationship, occurrence in a gene whose disease association overlaps with the patient's reported phenotype, and inheritance mode, as appropriate. Clinical interpretation was performed on variants and CNVs of interest in accordance with the American College of Medical Genetics and Genomics guidelines.
The TruGenome Undiagnosed Disease test was developed and its performance characteristics determined by Illumina Clinical Services Laboratory (CLIA# 05D1092911/CAP# 7217613). It has not been cleared or approved by the U.S. Food and Drug Administration. Pursuant to the requirements of CLIA '88, this laboratory test has established and verified the test's accuracy and precision. cDNA constructs cDNAs encoding human GLUA3 was subcloned into the NheI and XhoI restriction sites of the vector pCAGGS-IRES-EGFP. Human GLUA2, TARP γ-2 and CNIH2 were subcloned into the vector pCAGGS-IRES-mCherry. Coexpression of AMPARs and the auxiliary subunits were identified by the fluorescence of EGFP and mCherry. GLUA3_R660T was made by overlapping PCR and confirmed by Sanger sequencing.

HEK cells
HEK293T cells were cultured in a 37˚C incubator supplied with 5% CO 2 . Transfection was performed in 35-mm dishes using lipofectomine2000 reagents (Invitrogen). When coexpression was carried out, the ratio of GLUA3 to GLUA2 cDNA was 1:1. The ratio between GLUAs and CNIH2 or TARP γ-2 was 1:1. NBQX (100 μM) was included in culture media to block AMPAR-induced cytotoxicity. Cells were dissociated with 0.05% trypsin and plated on coverslips pretreated with poly-D-lysine 24 h post transfection. Recording was performed 4 h after plating.

Animals
ICR mice were obtained from Gempharmatech Inc. (Nanjing). The mice were maintained in the core animal facility of the Model Animal Research Center (MARC) at Nanjing University with the room temperature and the light-cycle automatically controlled (25±1˚C; 12 hrs for light and 12 hrs for dark). Mice had free access to food and water. Mouse breeding and experiments were conducted under an IACUC approved protocol. All the experiments were performed in accordance with the Guide for the Care and Use of Laboratory Animals of Nanjing University.

In Utero electroporation
In utero electroporation was performed as recently described [47]. E14.5-15.5 pregnant mice were anesthetized with a mixture of ketamine (100 mg/kg body weight) and xylazine (5 mg/kg body weight) via intraperitoneal injection. The mice were then subjected to a surgical procedure to expose the uterus. Each embryo was injected with about 2 μL of plasmid DNA mixed with Fast-Green into the left lateral ventricle via a beveled glass micropipette. The embryos were then electroporated with five 42-V pulses of 50 ms, delivered at 1 Hz, using platinum tweezertrodes in a square-wave pulse generator (BTX, Harvard Apparatus). Following electroporation, the embryos were placed back into the abdominal cavity, and the muscle and skin were sutured. Mice were injected i.p. with carprofen (5mg/kg body weight) and monitored until fully awake.

Non-stationary fluctuation analysis
Non-stationary fluctuation analysis was used to analyze AMPAR channel properties. Responses to 1 ms pulse of glutamate (10 mM) were recorded with the frequency of 0.2 Hz for at least 100 traces. The average variance (σ 2 ) was calculated from all the traces and fitted with σ 2 = iI-I 2 /N + σ 0 2 where i is single channel current, I is the mean current, N is the number of available channels in the patch and σ 0 2 is the variance of background noise. The probability of opening of the receptor channels in patch at any given point in time is determined by P open = I/iN. The single channel conductance (γ) was calculated with γ = i/(Vh-V r ), where V h is the holding potential of -70 mV, V r is the reverse potential for AMPARs assumed to be 0 mV. GLUA2/A3_R660T/CNIH2, 16.7 ± 1.7 ms, n = 9; ��� p < 0.001. (D) Up, the sample traces for desensitization of GLUA2/A3/CNIH2 and GLUA2/A3_R660T/CNIH2. Low left, statistics of weighted τ des . GLUA2/A3/CNIH2, 33.0 ± 2.0 ms, n = 9; GLUA2/A3_R660T/CNIH2, 52.4 ± 6.1 ms, n = 9; �� p = 0.008. Low right, statistics of steady state currents. GLUA2/A3/ CNIH2, 17.5 ± 2.9%, n = 9; GLUA2/A3_R660T/CNIH2, 66.8 ± 4.3%, n = 9; ��� p < 0.001. Data are presented as mean ± SEM. Unpaired t-test was used for data analysis.