Abnormal AMPAR-mediated synaptic plasticity, cognitive and autistic-like behaviors in a missense Fmr1 mutant mouse model of Fragile X syndrome

Fragile X syndrome (FXS) is the most frequent form of inherited intellectual disability and the best-described monogenic cause of autism. FXS is usually caused by a CGG-repeat expansion in the FMR1 gene leading to its silencing and the loss-of-expression of the Fragile X Mental Retardation Protein (FMRP). Missense mutations were also identified in FXS patients, including the recurrent FMRP-R138Q mutation. To investigate the mechanisms underlying FXS in these patients, we generated a knock-in mouse model (Fmr1R138Q) expressing the FMRP-R138Q protein. We demonstrate that the Fmr1R138Q hippocampus has an increased spine density associated with postsynaptic ultrastructural defects and increased AMPA receptor surface expression. Combining biochemical assays, high-resolution imaging and electrophysiological recordings, we also show that the mutation impairs the hippocampal long-term potentiation (LTP) and leads to socio-cognitive deficits in Fmr1R138Q mice. These findings reveal that the R138Q mutation impacts the postsynaptic function of FMRP and highlight potential mechanisms causing FXS in FMRP-R138Q patients.


Introduction
The formation of functional synapses in the developing brain is fundamental to establishing efficient neuronal communication and plasticity, which underlie cognitive processes. In the past years, synaptic dysfunction has clearly emerged as a critical factor in the aetiology of neurodevelopmental disorders including Autism Spectrum disorder (ASD) and Intellectual Disability (ID). X-linked ID (XLID) accounts for 5-10% of ID patients and is caused by mutations in genes located on the X chromosome.
The Fragile X Syndrome (FXS) is the most frequent form of inherited XLID and the first monogenic cause of ASD with a prevalence of 1:4000 males and 1:7000 females 1 . The majority of FXS patients exhibit mild-to-severe ID associated with significant learning and memory impairments, Attention Deficit Hyperactivity Disorder (ADHD) and autistic-like features 2-4 . To date, no effective therapeutic strategies are available.
FXS generally results from a massive expansion of the trinucleotide CGG (>200 repeats) in the 5'-UTR region of the FMR1 gene leading to its transcriptional silencing and consequently, the lack of expression of the encoded Fragile X Mental Retardation Protein (FMRP) 2, 5 . FMRP is an RNA-binding protein that binds a large subset of mRNAs in the mammalian brain and is a key component of RNA granules. These granules transport translationally-repressed mRNAs essential for the synaptic function along axons and dendrites 1, 3 . Neuronal activation triggers the local translation of these critical mRNAs at synapses allowing spine maturation and elimination, which are essential processes to shape a functional neuronal network in the developing brain. Accordingly, the lack of FMRP expression in FXS patients and Fmr1 knockout (Fmr1-KO) animal models leads to a pathological increase in immature dendritic protrusions due to a failure in the synapse maturation and/or elimination processes 6 . These defects correlate with significant alterations in glutamatergic αamino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR)-mediated synaptic plasticity, including Long-Term Depression (LTD) and Potentiation (LTP) 1, 7 . Consequently, these defects lead to learning and memory deficits and underlie the abnormal socio-emotional behaviours in Fmr1-KO mice 1,7 .
While the CGG-repeat expansion is the most frequent cause of FXS, other mutagenic mechanisms have been reported, including deletions, promoter variants, missense and nonsense mutations. To date, more than 120 sequence variants have been identified in the FMR1 gene. However, only three missense mutations (I304N, G266E and R138Q) have been functionally studied and showed an association with the aetiology of FXS 8- 13 . Among them, the R138Q mutation is of particular interest since it has been identified in three unrelated individuals presenting clinical traits of FXS. The first male patient sequenced displayed ID, anxiety and seizures 11, 13 , while the second presented the classical features of FXS including ID, ADHD, seizures and ASD 14 . Interestingly, a female with mild ID and attention deficits was recently identified bearing the R138Q mutation 15 .
The R138Q mutation does not affect the ability of FMRP to bind polyribosomes and repress the translation of specific target mRNAs 13 . In addition, the intracellular perfusion of a short N-terminal version of FMRP-R138Q in Fmr1-KO CA3 hippocampal neurons failed to rescue the action potential broadening, suggesting a functional alteration of the FMRP-R138Q truncated mutant form 13 .
However, the cellular and network alterations underlying the phenotype described in FMRP-R138Q FXS patients remains to be elucidated. Here, we have engineered a knock-in mouse model expressing the recurrent missense R138Q mutation in FMRP (Fmr1 R138Q ). We demonstrate that Fmr1 R138Q mice exhibits significant postsynaptic alterations in the hippocampus, including an increased dendritic spine density, AMPAR-mediated synaptic plasticity defects associated with severe impairments in their socio-cognitive performances.

Results
The FXS R138Q mutation does not alter the total levels of FMRP mRNA and protein nor the neuroanatomy of the hippocampal formation.
To assess the pathophysiological impact of the recurrent R138Q mutation in vivo, we generated a specific knock-in mouse line expressing the R138Q FXS mutation using classical homologous recombination in murine C57BL/6 embryonic stem (ES) cells (Fig. 1a). The R138Q coding mutation was introduced in exon 5 by changing the CGA arginine codon into a CAA nucleotide triplet coding for a glutamine (R138Q: c.413G > A). The integrity of the FXS mutation in the Fmr1 R138Q mice was confirmed by genomic DNA sequencing (Fig. 1a). Fmr1 R138Q mice were viable, showed a standard growth and normal fertility and mortality rates ( Supplementary Fig. 1).
The expression pattern of the FMRP protein is developmentally regulated 16 . To assess whether the R138Q mutation affects the developmental profile of FMRP, we compared the FMRP protein levels in the brain of WT and Fmr1 R138Q mice at different postnatal (PND) ages (Fig. 1b). As expected from the literature 2,5 , the FMRP protein level peaked at PND10-15 (WT PND15: 1.330 ± 0.246) and then significantly decreased in the adult brain of WT mice (WT PND90: 0.441 ± 0.078). FMRP-R138Q protein levels showed a similar pattern in the developing Fmr1 R138Q brain ( Fig. 1b; Fmr1 R138Q PND15: 1.300 ± 0.303; Fmr1 R138Q PND90: 0.491 ± 0.072), indicating that the R138Q mutation does not alter the protein expression of the pathogenic FMRP.
Since FMRP is an RNA-binding protein regulating the local translation of a large number of mRNAs important to the synaptic function, we compared the total levels of a subset of its target mRNAs in 6 PND90 WT and Fmr1 R138Q male littermate brains by RT-qPCR (Fig. 1c). We found no significant differences in the total mRNA levels of the FMRP targets tested.
To determine whether the R138Q mutation alters the brain morphology, we next performed a Nissl staining on 20-µm-thick brain coronal slices from PND90 WT and Fmr1 R138Q male littermates. There were no apparent macroscopic defects in the Fmr1 R138Q brain and the structural organization of the hippocampus was preserved (Fig. 1d).

Fmr1 R138Q mice show an increase in hippocampal spine density.
FMRP is essential to proper spine elimination and maturation 3 . A hallmark of the classical FXS phenotype is a pathological excess of long thin immature dendritic protrusions 17 , resulting from a failure in postsynaptic maturation and/or elimination processes. To understand if the R138Q mutation impacts spine maturation and/or elimination, we analyzed the morphology and density of dendritic spines in the Fmr1 R138Q hippocampus (Fig. 2). We first used attenuated Sindbis viral particles in WT and Fmr1 R138Q cultured hippocampal neurons at 13 days in vitro (13 DIV) to express free GFP and outline the morphology of dendritic spines 18, 19 . We then compared the density and morphology of dendritic spines 20h post-transduction (Fig.2a). Interestingly, while the length of dendritic spines was similar for both genotypes (WT: 1.563 ± 0.0717 µm; Fmr1 R138Q : 1.521 ± 0.06193 µm), Fmr1 R138Q neurons displayed a significant increase in spine density compared to WT neurons (WT: 6.06 ± 0.126 spines/10 µm; Fmr1 R138Q : 8.16 ± 0.207 spines/10 µm).
We also evaluated the characteristics of dendritic spines in the CA1 region of the hippocampus of PND90 WT and Fmr1 R138Q male littermates using Golgi-Cox staining ( Fig. 2b; Supplementary Fig. 2).
While there was no difference in spine length and width, Fmr1 R138Q hippocampal neurons displayed a significant increase in spine density both in basal ( Fig. 2b To characterize the effect of R138Q mutation on the composition of synapses, we compared the total protein levels of several pre-and postsynaptic proteins in brain homogenates prepared from WT and Fmr1 R138Q male littermates ( Fig. 3a; Supplementary Fig. 3). While the majority of proteins investigated in the Fmr1 R138Q brain showed levels similar to their WT littermates (GluA2 WT: 1 ± 0.14, Fmr1 R138Q : 1.206 ± 0.04; PSD95 WT: 1 ± 0.18, Fmr1 R138Q : 1.07 ± 0.04; GluN1 WT: 1 ± 0.06, Fmr1 R138Q : 1.027 ± 0.09), a significant increase in the total amount of the GluA1 AMPAR subunit was measured in the Fmr1 R138Q mice (GluA1 WT: 1 ± 0.11, Fmr1 R138Q : 1.51 ± 0.06). To further explore the 8 AMPAR defects in the Fmr1 R138Q brain, we compared the levels of surface-expressed GluA1 in DIV15 WT and Fmr1 R138Q cultured hippocampal neurons (Fig. 3b). Using surface-immunolabeling assays with specific anti-GluA1 antibodies, we showed that the surface levels of GluA1 in Fmr1 R138Q neurons were significantly increased (Mean surface GluA1 intensity, WT: 1 ± 0.087; Fmr1 R138Q : 1.42 ± 0.13; Surface cluster density, WT: 1 ± 0.049; Fmr1 R138Q : 1.22 ± 0.06). We confirmed this increase in AMPAR surface expression using cell surface biotinylation assays which showed that surface levels of both GluA1 and GluA2 AMPAR subunits were significantly higher in Fmr1 R138Q neurons than in WT cells Interestingly, while there was no alteration in the total levels of GluA2 in Fmr1 R138Q brain (Fig. 3a), the surface expression of GluA2 was significantly higher both in cultured

Functional alterations in synaptic transmission in the Fmr1 R138Q hippocampus.
We showed that the recurrent FXS R138Q missense mutation leads to an increase in AMPAR surface expression. To assess whether this increase occurs at least in part synaptically, we performed STimulated Emission Depletion (STED) microscopy on surface GluA1-labelled WT and Fmr1 R138Q hippocampal neurons (Fig. 4a). Interestingly, we measured an increase in the mean level of surfaceexpressed GluA1 clusters in Homer1-labelled Fmr1 R138Q synapses (Fig. 4a, WT: 2.015±0.094; Fmr1 R138Q : 2.582±0.106). These data indicate that the missense R138Q mutation leads to a significant increase in available postsynaptic AMPARs in the Fmr1 R138Q hippocampus.
Finally, we performed whole cell patch-clamp recordings in CA1 neurons from hippocampal slices of PND90 WT and Fmr1 R138Q male littermates to assess whether the increase in surface-expressed AMPARs measured in the Fmr1 R138Q hippocampus is associated with alterations in glutamatergic transmission ( Fig. 4b-h). We showed that the amplitude of AMPAR-mediated miniature Excitatory PostSynaptic Currents (mEPSCs) is significantly enhanced in Fmr1 R138Q mice (Fig. 4b,c; WT: 17.09 ± 0.539 pA; Fmr1 R138Q : 18.97 ± 0.405 pA). Interestingly, we did not measure any significant differences in the frequency of mEPSCs (Fig. 4d,e; WT: 0.178 ± 0.040 Hz; Fmr1 R138Q : 0.143 ± 0.026 Hz) or the kinetics of these events (Fig. 4f,g) between the two genotypes. Altogether these data confirm the synaptic increase in functional surface-expressed AMPARs in Fmr1 R138Q mice and that the R138Q FXS mutation leads to important postsynaptic defects.
The hippocampal long-term potentiation is impaired in Fmr1 R138Q mice.
We next investigated the consequences of the R138Q mutation in hippocampal plasticity. Since the level of surface-expressed AMPARs is increased in the Fmr1 R138Q hippocampus, we wondered whether a further increase in synaptic AMPARs could be triggered upon the expression of long-term potentiation (LTP; Fig. 5). We first combined surface biotinylation assays with the chemical induction of LTP (cLTP 20 ) in WT and Fmr1 R138Q hippocampal neurons (Fig. 5a-c). In line with the literature 20 , the cLTP treatment triggered the increase in both GluA1 and GluA2 surface expression in WT neurons (WT GluA1 cLTP: 1.578 ± 0.107; WT GluA2 cLTP: 1.203 ± 0.008). In contrast, the surface levels of AMPARs upon cLTP were not increased but rather unexpectedly decreased in Fmr1 R138Q hippocampal neurons (Fmr1 R138Q GluA1 cLTP: 0.490 ± 0.073; Fmr1 R138Q GluA2 cLTP: 0.899 ± 0.157).
We confirmed these results using surface immunolabeling assays on WT and Fmr1 R138Q hippocampal neurons in basal and cLTP-induced conditions (Fig. 5d,e). Indeed, while the level of surface GluA1 was significantly increased following cLTP in WT neurons (Mean surface GluA1 intensity, WT: 1.427 ± 0.129 vs basal; Surface cluster density, WT: 1.308 ± 0.08596 vs basal), the surface level of GluA1 in Fmr1 R138Q neurons was surprisingly decreased (Mean surface GluA1 intensity, Fmr1 R138Q : 0.6354 ± 0.08647 vs basal; Surface cluster density, Fmr1 R138Q : 0.7455 ± 0.08227 vs basal). These data thus indicate that the hippocampal plasticity is severely impaired in the Fmr1 R138Q mice.
In line with the data obtained on hippocampal cultures, BS3-crosslinking assays showed that the induction of cLTP in Fmr1 R138Q hippocampal slices led to a significant reduction in the levels of surface-expressed AMPARs ( Fig. 5f-h, Fmr1 R138Q GluA1 cLTP: 0.706 ± 0.008 vs basal; Supplementary Altogether, these data reveal that the induction of cLTP does not promote any increase in the surface levels of AMPARs in the Fmr1 R138Q hippocampus. Therefore, to determine if the expression of the FMRP-R138Q mutant also physiologically impacts the AMPAR-mediated responses, we induced LTP by high frequency stimulation (HFS) in acute WT and Fmr1 R138Q hippocampal slices and recorded the postsynaptic responses in CA1 neurons ( Fig. 5i-k). First, we tested the impact of the R138Q mutation on the CA3 to CA1 synaptic transmission and did not find any significant differences with the WT responses in Input/Output curves established following the stimulation of the Schaffer collaterals ( Supplementary Fig. 5). This indicates that the connectivity between the pre-and postsynaptic sites is preserved in the Fmr1 R138Q hippocampus. Next, we found that the induction of LTP by HFS was evoked as expected in the WT hippocampus (WT LTP: 157.2 ± 6.52 vs basal) but was drastically reduced in Fmr1 R138Q male littermates (Fmr1 R138Q LTP: 122.4 ± 10.12 vs basal), in line with the impaired LTP seen in both biochemical and imaging experiments ( Fig. 5a-h). In addition, we did not measure any significant difference in the mean fiber volley (FV) slope between genotypes ( Fig. 5i-k), indicating that the impaired LTP in the Fmr1 R138Q hippocampus arises from postsynaptic impairments rather than from presynaptic alterations.

Fmr1 R138Q mice display ID-and ASD-like features.
The R138Q mutation has been identified in both male and female patients 11,13,15 . Since the mutation affects the surface levels of AMPARs and directly impacts synaptic plasticity in the hippocampus, we investigated whether male and female Fmr1 R138Q mice display altered cognitive and/or social performances (Fig. 6). To avoid possible pitfalls due to an impact of the R138Q mutation on locomotion, we first tested PND40-45 mice using the open field test to detect any potential motor defects. We did not measure any significant differences in the number of crossings between the two genotypes demonstrating that there is no motor alteration in Fmr1 R138Q mice (Supplementary Fig.   6a,b).
The communication skills are altered in some FXS patients and Fmr1-KO mice 19 . We therefore compared the ultrasonic vocalization (USV) profile in PND7 WT and Fmr1 R138Q pups removed from the nest and found a ~50% decrease in the number of USVs in Fmr1 R138Q compared to WT animals in both genders (Fig. 6a Next, we evaluated the cognitive performance in PND40-45 males and females with the novel object recognition test (Fig. 6c,d). We found a profound deficit for both genders as the Fmr1 R138Q mice spent significantly less time than WT animals exploring the novel object (Fmr1 R138Q  32.84 ± 6.63%). These data indicate that the R138Q mutation substantially alters the cognitive function in the Fmr1 R138Q mice.
About 25% of FXS patients presents ASD traits, including social avoidance and decreased social skills. Therefore, to evaluate the impact of the R138Q mutation in ASD-like behaviors, we tested sociability in both male and female WT and Fmr1 R138Q mice using the three-chamber social arena (Fig. 6e,f). As expected, both WT genders spent significantly more time sniffing the novel mouse rather than the empty cage. We showed that both the time spent sniffing the social stimulus and the ability to discriminate between the novel mouse and the empty cage were dramatically reduced in Fmr1  Altogether, our behavioral data reveal that both male and female Fmr1 R138Q mice display important communicative, cognitive and social deficits that correlate with severe postsynaptic alterations.

Discussion
Synaptic transmission and/or plasticity defects have been clearly linked to the development of many, if not all, neurological disorders. Therefore, a better understanding of the pathways underlying these alterations is essential to develop strategies to rescue the identified dysfunctions and design innovative targeted therapies to treat these diseases. Here, we generated and characterized a novel mouse model for FXS expressing the recurrent R138Q missense mutation in the FMRP protein. We show for the first time that the R138Q mutation leads to an increase in spine density, alterations in the post-synaptic organization and an impaired LTP in the Fmr1 R138Q hippocampus. The direct consequence of this plasticity defect is an abnormal socio-cognitive behavior in Fmr1 R138Q mice that resembles the ID and ASD-like traits described in FXS patients bearing the R138Q mutation. Altogether, our data validate the Fmr1 R138Q mouse line as a compelling preclinical model to investigate the molecular mechanisms underlying the pathology.
To date, only two studies have provided some insights into the impact of the R138Q mutation in neuronal function 13,21 . FMRP is known to participate in the regulation of AMPAR trafficking 1,7 , which is critical to maintain the synaptic function. Alpatov and colleagues 21 investigated the impact the R138Q mutation on the basal trafficking of AMPAR and reported that the exogenous expression of FMRP-R138Q does not impact the constitutive endocytosis of AMPAR in a Fmr1-KO background.
Here, we clearly demontrated that the R138Q missense mutation leads to important postsynaptic alterations including increased levels of synaptic AMPARs and an impaired LTP in the Fmr1 R138Q hippocampus. This revealed a previously unsuspected impact of the mutation on the postsynaptic targeting and activity-dependent trafficking of AMPARs in the hippocampus.
In a second study aiming at investigating the functional impact of the R138Q mutation, Myrick and colleagues reported that the exogenous overexpression of a truncated (FMRP1-298) version of the FMRP-R138Q mutant fails to rescue action potential (AP) broadening in Fmr1-KO neurons 13 , which correlates with an increased presynaptic release 22 . However, the presynaptic release per se was not assessed in this work 13 . The authors also showed that the R138Q mutation disrupts the interaction of the short FMRP1-298 form with the β4 subunit of BK channels, thus underlying AP duration 13, 22 . To conclude, they hypothesized that an alteration of the presynaptic function is likely responsible for the ID and seizures exhibited by the first FXS R138Q patient 13 . In contrast, the present data did not reveal any obvious physiological impairment linked to the presynaptic function in the Fmr1 R138Q mouse model. One possibility explaining this difference could be that Myrick and colleagues used the short N-terminal FMRP1-298 fragment of FMRP 13 rather than the full length of the protein in their study. Furthermore, BK channels are localized both at pre-and postsynaptic sites 23 . Consequently, alterations in FMRP/BK channel interaction may not only be linked to a presynaptic impairment, but also to potential postsynaptic defects. Future studies are still needed to assess the precise impact of the R138Q mutation on the presynaptic function in vivo.
Since most of the knowledge on FXS derives from studies using Fmr1-KO models, it is important to compare the phenotype reported in these models with our data on the Fmr1 R138Q mice. Because the R138Q mutation occurs in the FMR1 gene, it has been unequivocally associated with the development of an FXS-like pathology. It is important to stress here that the three unrelated FMRP-R138Q patients show markedly variable phenotypes 11,13-15 , indicating that the same mutation leads to different clinical features ranging from mild symptoms to a full, complex classical FXS phenotype.
Even if some of the defects measured in Fmr1 R138Q animals are different from those measured in  (Fig. 4).
Finally, we have shown that the NMDAR-mediated LTP is severely impaired in the Fmr1 R138Q hippocampus. While LTP impairments have also been reported in the Fmr1-KO hippocampus 30,31 , the molecular mechanisms underlying these defects in Fmr1 R138Q mice are likely different from those in the Fmr1-KO background. We also showed that Fmr1 R138Q mice display reduced social interaction in the three-chamber test, which is reminiscent of autistic-like features. Fmr1 R138Q mice are also unable to discriminate between a familiar and a novel object demonstrating that learning and memory processes are impaired in these animals. Similar cognitive and socio-emotional deficits have been reported in the Fmr1-KO mice in the same behavioural tasks 32,33 . Altogether, our data indicate that different mutations in the FMR1 gene engage distinct molecular mechanisms leading to similar pathological conditions. Therefore, the identification of additional FMRP-R138Q patients will undoubtedly help to shed light on the phenotypic and mechanistic similarities and differences with the classical FXS pathology.
The Interestingly, the excess of available AMPARs in the Fmr1 R138Q brain may correlate with the intractable seizures observed in the first-reported patient carrying the R138Q mutation 13 . Therefore, it would be of interest to assess whether reducing the overactivation of AMPARs could produce any beneficial effects in this mouse model. For instance, testing FDA-approved antiepileptic drugs such as the AMPAR antagonist Perampanel 34 to reduce the activity of these glutamate receptors could be an efficient way to treat the epileptic manifestations in FXS patients carrying the R138Q mutation. Preclinical studies are now required to first investigate the epileptic activity in the Fmr1 R138Q mouse line and then, determine if Perampanel can correct the altered AMPAR function as well as the socio-cognitive behaviours in these mice.
Generating preclinical models for brain disorders is an essential step toward the development of efficient therapies to treat human diseases. In this context, the Fmr1 R138Q mouse line certainly represents a unique preclinical model to test the efficacy of new molecules and/or the repurposing of existing FDA-approved drugs to correct the altered pathways identified in these mice. This may also lead in the near future to the development of clinical studies to assess the potential benefits of these drugs in FXS patients. Therefore, gaining insights into this complex neurodevelopmental disorder may lead to innovative therapeutic strategies for these particular cases of FXS.
Nevertheless, since AMPAR alterations as well as synaptic defects have been linked to other neurodevelopmental disorders, these approaches might be extended to classical FXS patients and more generally to patients presenting ID and ASD in which the AMPAR pathway is altered.

Mouse lines
Fmr1 R138Q mice were successfully generated at the 'Institut Clinique de la Souris' (ICS; Illkirch-Graffenstaden, France) using standard procedures of homologous recombination in murine embryonic C57BL/6 stem (ES) cells (Fig. 1a). Fmr1 R138Q mice were backcrossed for more than 10 generations into the C57BL/6J genetic background (Janvier, St Berthevin, France). All animals were handled and treated in accordance with the European Council directives for the Care and Use of Laboratory animals and following the ARRIVE guidelines. Mice had free access to water and food. Mice were exposed to a 12h light/dark cycle and the temperature was maintained at 23 ± 1°C.

Histology and Nissl staining
Brains of PND90 WT and Fmr1 R138Q male littermates were carefully dissected and frozen in precooled isopentane in liquid nitrogen (-60°C) for 1 min and stored at -20°C until cryostat sectioning. 20-μm sections containing the hippocampus were then quickly fixed in 4% formaldehyde in PBS.
Hippocampal sections were then Nissl stained to reveal the gross architecture of the hippocampus.
Fixed sections were first immersed in a cresyl violet solution (5 g/L cresyl violet, 0.3% acetic acid) for 2-3 min, then quickly washed in distilled water and 70% ethanol/0.05% acetic acid. After dehydrating in 100% ethanol, sections were mounted with Entellan and imaged with a 4x lens on a Leica DMD optical microscope.

Golgi-Cox staining
WT and Fmr1 R138Q male littermates at PND90 were deeply anesthetized by an intraperitoneal injection of 50 mg/kg sodium pentobarbital. Mice were then transcardially perfused with a 0.9% NaCl solution (w/v) to remove blood from the vessels. Brains were stained with the FD Rapid GolgiStain Kit (FD NeuroTechnologies) following the manufacturer's instructions. Briefly, tissues were quickly rinsed in ddH2O and immersed in the impregnation solution for 14 days at RT in the dark. Brains were then transferred to the tissue-protecting solution and kept in the dark at RT for a further 72h. 100-μm coronal sections mounted in the tissue-protecting solution on coated slides (2% gelatin, 1% KCr(SO4)2) were then air dried at room temperature (RT) for 2h in the dark. After two 4-min washes in ddH2O, sections were stained in an ammonium and sodium thiosulfatecontaining solution for 10 min. Sections were then dehydrated successively in ethanol with increasing percentage (50 to 100%) and finally, in xylene for 4 min and mounted in Entellan. Images were acquired in the following 48h using an Axiovert200M videomicroscope (Zeiss) with a 100x oil immersion lens. Z-series of randomly selected basal and apical secondary dendrites from CA1 hippocampal neurons were processed with the Extended Depth of Field plugin of ImageJ. Dendritic spine length, width and density were measured in ImageJ and data imported in GraphPad Prism software for statistical analysis.

Electron microscopy
WT and Fmr1 R138Q male littermates at PND90 were deeply anesthetized by an intraperitoneal injection of 50 mg/kg sodium pentobarbital. Mice were transcardially perfused with a 0.9% NaCl solution and then with 2.5% glutaraldehyde and 2% paraformaldehyde in 0.15 M sodium cacodylate buffer (pH 7.4). Brains were post-fixed for an additional 24-48h at 4°C. Hippocampi were then manually dissected from 100 µm-thick vibratome sections. After washing, they were further fixed in 2% osmium tetroxide, rinsed, stained with 1% uranyl acetate in water for 45 min, dehydrated and Excitatory synapses in the apical dendrite layer of the hippocampal CA1 region were selected for analyses based on the presence of a cluster of presynaptic vesicles, a defined synaptic cleft and an electron-dense postsynaptic density (PSD). The density of excitatory synapses, as well as the length and thickness of the PSD were computed. The average thickness of the PSD was calculated as described in 37, 38 . The estimation of the density of excitatory synapses per μm 3 of hippocampus was computed using a size-frequency stereological method 38,39 .

Mouse brain lysate preparation
Brain lysates from WT and Fmr1 R138Q littermates from the indicated developmental stages (PND3-90) were prepared as previously described 18 . Briefly, freshly dissected brains were transferred in 5 volumes (w/v) of ice-cold sucrose buffer (10 mM Tris-HCl pH 7.4, 0.32 M sucrose) supplemented with a protease inhibitor cocktail (Sigma, 1/100) and homogenized at 4°C using a Teflon-glass potter.
Nuclear fraction and cell debris were pelleted by centrifugation at 1,000g for 10 min. The postnuclear S1 fraction (supernatant) was collected and protein concentration measured using the BCA protein assay (Biorad).

Total mRNA analysis
Total RNA was extracted from PND90 WT and
Briefly, cRNAs were generated from the pSinRep5 plasmid (Invitrogen) containing the sequence coding for eGFP and from the defective helper (pDH-BB) plasmid using the Mmessage Mmachine SP6 solution (Ambion). Both cRNAs were then electroporated into BHK21 cells. Pseudovirions present in the culture medium were harvested 48h after electroporation and ultracentrifugated on a SW41Ti. Aliquots of concentrated Sindbis particles were titrated and stored at -80°C until use.
Neurons were transduced at a multiplicity of infection (MOI) of 0.1 and incubated at 37°C under 5% CO2 for 20h until use.
Standard β3-tubulin controls were included to ensure that intracellular proteins were not BS3crosslinked. Bands were quantified using ImageJ software (NIH). A surface/intracellular ratio was performed to analyze the levels of surface GluA1 and GluA2 expression.

Analysis of synaptic AMPAR cluster
Deconvolved Z-stack images of confocal Homer1-Alexa594/STED GluA1-StarRed were analyzed using a home-made macro program in FIJI software 45 to quantify the levels of surface-expressed GluA1 on dendritic spines. Deconvolved Z-stacks were first converted to 2D images by a sum projection. The Homer1 image was then segmented (Median filtering, intensity thresholding, Watershed) to delineate mature dendritic spines with a diameter criterion (>250 nm).
Corresponding Spine Regions of Interest (SROI) obtained were first used to measure the intensity of the surface GluA1 staining below these SROI on GluA1 images. Then the number of receptors per spine was counted by an individual SROI screening on GluA1 images: each SROI was duplicated to give a small surface GluA1 image per spine and segmented. The number of surface-associated GluA1 clusters detected per spine in secondary dendrites was incremented according to a size criterion, assuming a maximum size of 90 nm. Data were then imported in GraphPad Prism software for statistical analysis. resistance were continuously monitored throughout the recordings and, if either of these parameters varied by more than 20%, the experiment was discarded. At least 100 events were obtained from each cell. Experiments and analysis were done blind to the genotype. Recordings were analyzed using the Clampfit software by applying a threshold search to detect spontaneous events. The threshold for AMPAR mEPSCs detection was set at -9 pA to exclude electric noise contamination.

Behavioural tasks
All experiments were performed between 9:00 a.m. and 4:30 p.m. Both male and female WT and Locomotor activity. At PND40-45, during the habituation session of the novel object recognition test, locomotor activity was calculated using a grid that divides the arena into equally sized squares and that is projected over the recordings. The number of line crossings made by the animal was quantified to assess its motor activity.
The Isolation-induced ultrasonic vocalizations (USVs) test. The test was performed as previously described 19 . Briefly, each pup (PND7) was individually removed from the nest and placed into a black Plexiglas arena, located inside a sound-attenuating and temperature-controlled chamber. USVs from the pups were detected for 3 min by an ultrasound microphone (Avisoft Bioacoustics, Germany) sensitive to frequencies between 10 and 250 kHz and fixed at 10 cm above the arena. Pup axillary temperature was measured before and after the test by a digital thermometer. The emission of USVs was analyzed using Avisoft Recorder software (Version 5.1).
Novel object recognition test. The novel object recognition test was performed at PND40-45. The test consisted of three phases: habituation, training and test. In the habituation phase, the animals are allowed to explore an empty arena (a Plexiglas arena measuring 40 × 40 × 40 cm 3 ) for 5 min.
Twenty-four hours later, on the training trial, each mouse was individually placed into the arena containing two identical objects (A1 and A2), equidistant from each other, and allowed to explore the objects for 10 min. After 1h, during the test phase, one copy of the familiar object (A3) and a new object (B) were placed in the same location as during the training trial. The time spent exploring each object was recorded for 5 min. The discrimination index was calculated as the difference in time exploring the novel and the familiar objects, expressed as the percentage ratio of the total time spent exploring both objects 44 .
Three-chamber test. This test was performed at PND40-45. The apparatus was a rectangular three-

Data manipulation and statistical analysis
Statistical analyses were calculated using GraphPad Prism (GraphPad software, Inc) or Sigma Plot (13.0; Systat Software, Inc, USA) software. All data are expressed as mean ± s.e.m. Unpaired t-test (Figs. 2c, 3a, 4, 5k and Supplementary Figs. 1b, 3), Ratio t-test (Figs. 3c,d and 5b,c,h) or nonparametric Mann-Whitney test (Figs. 2a,b, 3b, 5e and supplementary Fig. 2) were used to compare 31 medians of two data sets. Statistical significance for multiple comparison data sets was computed two-way ANOVA with Sidak's post-test (Fig. 1b). Behavioral data ( Fig. 6 and Supplementary Fig. 4) were analyzed by two-way ANOVA including genotype and sex as factors with a student-Newman-Keuls post hoc test for individual group comparisons. Normality for all groups was verified using the Shapiro-Wilk test. For electrophysiological data, distributions were analyzed by a Kolmogorov-Smirnov test (Fig. 4). *p<0.05 was considered significant.

Data availability
All relevant data are available from the corresponding author upon reasonable request.      a. Representative immunoblots showing the levels of the indicated synaptic proteins in brain homogenates from PND21 WT and Fmr1 R138Q brains. GAPDH was used as a loading control.

Figures and legends
Quantification shows the mean ± s.e.m. of the total levels of the indicated proteins. Unpaired t-test. n.s., not significant. *p<0.05. b. Representative secondary dendrites from TTX-treated WT and