Fragile X syndrome (FXS) is an inherited neurodevelopmental disorder with a wide variety of symptoms, including intellectual disability, seizures, hyperactivity, social anxiety, and others characteristic of autism spectrum disorders. It is caused by loss-of-function mutations in the RNA binding protein fragile X mental retardation protein (FMRP) (Pieretti et al., 1991; Kelleher and Bear, 2008; Lozano et al., 2014). The knockout of Fmr1 gene in mouse (Fmr1^−/y mice) exhibits the primary molecular and behavioral symptoms associated with FXS (Hou et al., 2006). Many pathological changes observed in FXS are thought to be a result of a modest increase in protein synthesis (Bear et al., 2004; Bhakar et al., 2012; Bhattacharya et al., 2012). Indeed, previous studies have shown a rescue of altered synaptic plasticity and some neurological symptoms by normalizing the rate of global mRNA translation (Dolen et al., 2007; Osterweil et al., 2010, 2013; Krueger and Bear, 2011; Qin et al., 2015). However, the mRNAs that are aberrantly translated remain to be identified.

In the search for genes expression deficiencies, quantitative genomic approaches are complicated by complex RNA profiles from individual cell types within a tissue. Genetically labeled cell types strategies now allow focusing on a pertinent cell type to identify specific gene expression (Heiman et al., 2008; Sanz et al., 2009). We chose the RiboTag approach (Sanz et al., 2009), which is a methodology for affinity purification of ribosome-bound mRNAs from genetically defined cell populations in the brain. The RiboTag mouse line expresses the ribosomal protein Rpl22 tagged with the hemagglutinin (HA) epitope in specific cell types by mating to a Cre recombinase-expressing mouse. HA-tagged ribosomes can be then purified from the target cell population and their associated mRNAs sequenced. This allows the comparison of translatome profiles in a genetically-identified cell population between mouse genotypes.

FMRP loss in Fmr1^−/y mice has been shown to cause abnormal synaptic and structural plasticity in CA1 pyramidal cells (Huber et al., 2002; Lauterborn et al., 2007; Hu et al., 2008; Meredith and Mansvelder, 2010; Busquets-Garcia et al., 2013), which in turn have been associated with impaired hippocampal function as well as cognitive deficits (Contractor et al., 2015; Radwan et al., 2016). We thus studied the mRNA translation in hippocampal CA1-pyramidal cells in Fmr1^−/y mice compared to wild-type littermates in order to identify those genes involved in this neurodevelopmental disorder. Differential analysis of ribosome-associated mRNA revealed up- and down-regulation of genes linked to plasticity-related functions. Among them, we found a decreased expression of Klk8 in Fmr1^−/y mice. KLK8, Kallikrein Related Peptidase 8 (also known as neuropsin), is a serine protease expressed focally in the limbic system (Chen et al., 1995), especially in hippocampal CA1 pyramidal cells, which drives early processes of memory acquisition and Schaffer collateral plasticity in adult mouse hippocampus (Tamura et al., 2006). Interestingly, KLK8 catalyzes the proteolysis of proteins from the extracellular matrix (Matsumoto-Miyai et al., 2003) and could thus control adhesion changes between pre- and postsynaptic neurons needed for stable synaptic plasticity. Here, we show that re-establishing KLK8 expression in Fmr1^−/y cultured hippocampal neurons restores normal dendritic spine maturation.

To determine the identity of any differentially translated mRNAs linked to the loss of FMRP in CA1 pyramidal neurons, we generated an Fmr1^−/y mouse line allowing us to perform the RiboTag approach (Sanz et al., 2009). For this purpose, we first generated a double transgenic mouse line, by crossing the RiboTag mouse line (Sanz et al., 2009) with the tamoxifen-inducible Wfs1-CreERT2 mouse line (Madisen et al., 2010) leading to the expression of the ribosomal protein Rpl22 tagged with the hemagglutinin (HA) epitope exclusively in Wolfram syndrome 1 (Wfs1)-expressing neurons (Luuk et al., 2008) (Figures 1, 2A). Double immunofluorescence analyses confirmed that HA expression was restricted to CA1 pyramidal cells (CaMKIIα) and absent from CA2 pyramidal cells (RGS14), GABAergic cells (GAD67), astrocytes (GFAP), and microglia (Iba1) (Figures 2A–C). As expected, expression of Wfs1 and the glutamatergic marker Slc1a1 was enriched after HA-immunoprecipitation on whole hippocampal lysates compared to the input fraction (which comprises all mRNAs from the initial homogenate) (Figures 2D,E). By contrast, gene expression markers for GABAergic cells (Gad1, Slc32a1), oligodendrocytes (Cnp), astrocytes (Gfap), microglia (Aif1), as well as interneuron-specific markers (Sst, Npy, Pvalb, Calb2, Kcnip1, Grm1) were all de-enriched (Figure 2E), validating the Wfs1-CreERT2:RiboTag mouse line to perform mRNA profiling in CA1 pyramidal cells. We therefore generated a triple transgenic mouse line (Wfs1-CreERT2:RiboTag:Fmr1^−/y) (Supplemental Figure S1) allowing to assess the impact of the loss of FMRP (Figure 2F) on the translatome of CA1 pyramidal cells.

We next used high-throughput RNA sequencing (RNAseq) to perform genome-wide analyses of CA1 tagged ribosome-bound mRNAs in Fmr1^−/y mice and wild-type littermates (GEO; http://www-ncbi-nlm-nih-gov.insb.bib.cnrs.fr/geo/). Using a P value for a FDR < 0.05, our analysis identified 78 genes whose expression was different between the two genotypes. Among them, 49 genes were significantly downregulated and 29 upregulated (Figure 3A). Differentially expressed genes including downregulated (Serpina3n, Klk8, Efcab6), upregulated (Itih3, Igfbp2, Cml3) as well as equally expressed (Neurog2) genes were confirmed by qRT-PCR (Figure 3B), validating the reliability of gene quantification by the RNAseq approach. To gain insight into the nature of the differentially translated mRNAs in CA1 pyramidal neurons, we compared our data with the 842 FMRP-bound mRNAs previously identified from mouse brain polyribosomes (Darnell et al., 2011). Our cross-analysis revealed that 15% (12 out of 78) of these genes were FMRP-bound mRNAs. Given that 4.7% of the whole genome encodes for FMRP-binders (842 out of 17,823) this comparative analysis highlights a three-fold enrichment in FMRP-binders in our screen. Interestingly, among the 12 FMRP-bound mRNAs identified, 11 were downregulated (Kcnq3, Dlgap3, Cacna1g, Frmpd4, Sipa1l1, Fasn, Slc8a1, Itpr1, Atmin, Cdc42bpa, Dennd5a), while only Atp1a2 was upregulated in Fmr1^−/y mice compared to wild-type littermates (Figure 3C). In addition, our study also revealed altered mRNAs translation of new sets of genes belonging to gene families known to be the target of FMRP (Figure 3D). Furthermore, 85% (66 out of 78) of the differentially translated mRNAs identified were FMRP-unbound mRNAs, suggesting that the loss of FMRP has a more widespread impact on mRNAs translation (Figure 3C). Together, these results reveal that FMRP controls the translation as well as the repression of a subset of mRNAs in CA1 pyramidal neurons.

To determine whether differentially translated mRNAs belong to specific physiological categories, we next performed Gene Ontology (GO) analysis, which included Biological Process, Cellular Component, and Molecular Function, as well as KEGG pathways analyses to map molecular interaction and reaction networks (Figure 2E). Notably, the most enriched GO categories comprised terms related to spine formation/maturation including synapses, dendritic spines, actin cytoskeleton, cell junctions and extracellular matrix (ECM) regulation (Figure 3E, Supplemental Figures S2–S6, Supplemental Table 3), suggesting a potential contribution in the aberrant spine maturation described in CA1 pyramidal neurons (Bilousova et al., 2009; He and Portera-Cailliau, 2013).

One key process in controlling structural synaptic plasticity is the ECM regulation. In particular, ECM proteases activity regulates cell adhesion at synapses and defects in proteases expression or function have been linked to FXS (Gkogkas et al., 2014; Sidhu et al., 2014). Interestingly, KLK8, a protease from the ECM displayed the most significant differential expression in our RNAseq analysis (just after Fmr1) together with Serpina3n, a serine-protease inhibitor regulating KLK8 activity, among other targets. Hence, to test for a causal link between mRNA dysregulation and impaired spine morphology in Fmr1^−/y mice, we selected the Klk8 gene as candidate (Figure 3A). Its mRNA down-regulation in CA1 pyramidal cells from Fmr1^−/y mice was confirmed by qRT-PCR (Figure 3B). In wild-type mice, Klk8 gene encodes for a trypsin-like serine protease exclusively expressed in limbic areas, with the highest expression in pyramidal neurons of the hippocampal CAl-CA3 subfields and to a lesser extent CA2 (Chen et al., 1995) (Figure 4A, http://mouse.brain-map.org). The analysis at the protein level by Western blot performed on hippocampal homogenates confirmed the decreased expression in Fmr1^−/y mice (Figures 4B,C). Upon neuronal activity, KLK8 triggers proteolytic cleavage of cell adhesion molecules (CAM), a mechanism contributing to dendritic spine maturation and stability (Matsumoto-Miyai et al., 2003). The blocking of KLK8 by a neutralizing antibody (mAbB5) resulted in a concomitant reduction in neuronal activity-induced L1-CAM processing and the amplitude of Schaffer collateral LTP (Komai et al., 2000; Matsumoto-Miyai et al., 2003). We therefore tested whether the density and the morphology of dendritic spines were affected by KLK8 activity. Notably, a significant reduction of the proportion of mature stubby- and mushroom-shaped spines was observed after application of the activity-neutralizing anti-KLK8 antibody (mAbB5) on primary culture of wild-type hippocampal neurons (Figures 4D,E). Because these structural changes are reminiscent of the immature protrusion phenotype observed in Fmr1^−/y mice (Figure 4F), we next assessed whether low levels of KLK8 might contribute to dendritic spine defects resulting from FMRP loss. Spine density and morphology were analyzed in FMRP-deficient neurons re-expressing KLK8 or not. As previously reported, hippocampal neurons cultured from Fmr1^−/y mice develop more immature spines (Figures 4F,G). In contrast, a significant increase in the number of mushroom-shaped spines was found in FMRP-deficient neurons re-expressing KLK8 (Figures 4F,G). Together these results demonstrate that KLK8 activity contributes to dendritic spine maturation and that aberrant synaptogenesis observed in Fmr1^−/y mice relies in part on altered levels of KLK8.

Enhanced global mRNA translation has been previously reported in the hippocampus of Fmr1^−/y mice (Qin et al., 2005; Dolen et al., 2007; Osterweil et al., 2010). Hence, numerous efforts have converged these last years in understanding which specific set of mistranslated mRNAs in absence of FMRP could cause neurological deficits including learning disabilities and cognitive impairment. A classical approach has consisted in the identification, by crosslinking immunoprecipitation, of FMRP interactions with rodent brain polyribosomal mRNAs, assuming that FMRP binders could play important functions in synaptic plasticity. Although studies efficiently used this strategy to identify gene candidates relevant to synaptic functions (Darnell et al., 2005, 2011; Tabet et al., 2016), this method was restricted to FMRP targets, ignoring the functional importance of mRNAs that do not bind FMRP. By opposition and without assumption, in the present work we identified genes whose expression was modified in the Fmr1^−/y adult mice compared to control mice, as a snapshot integrating all dysregulated mRNAs. Interestingly, our study shows that the loss of FMRP has a widespread impact on mRNAs translation since both FMRP-bound and unbound mRNAs are differentially translated. These expression changes are restricted to a subset of 78 genes. Noteworthy, both up- and down-regulation of genes are important to fully comprehend the ethiology of this neurodevelopmental disorder.

Our study provides evidence about the molecular consequences associated with FMRP loss. We show that 15% of the mRNAs differentially regulated in Fmr1^−/y compared to control mice are FMRP-binders. Hence, the proportion of transcripts that bind FMRP in this list of genes differentially regulated is three-fold enriched compared to their proportion in the whole genome, comforting the function of FMRP as direct regulator of mRNA translation. In absence of FMRP, all but one of the identified FMRP-bound mRNAs were down-regulated. This result is consistent with previous data suggesting that FMRP would inhibit translation by stalling ribosomal translocation on its mRNA interactors, increasing their accumulation (although not translated) at synapses (Darnell et al., 2011). Furthermore, a bi-functional role of FMRP has been recently proposed to suppress or de-repress its target transcripts through the miRNA pathway by modulating RNA secondary structure (Kenny and Ceman, 2016). In our study, the majority of FMRP-binders were not found to be differentially regulated in CA1 pyramidal cells from adult mice hippocampi, suggesting a possible dysregulated expression during neurodevelopment. Finally, our data also highlights the preponderance (85%) of non-FMRP binders among the genes differentially regulated between adult Fmr1^−/y and control mice. The subset of altered translated mRNAs found herein, indirectly linked to the absence of FMRP and consecutive to neurodevelopmental adaptations, expands the list of pertinent gene candidates that could counteract neuronal plasticity defects, learning disabilities and cognitive impairments of FXS.

Analyzing gene functions, we further demonstrated that those differentially translated mRNAs belong to specific physiological categories, including synapses, dendritic spines, calcium signaling, actin cytoskeleton, cell junctions, and extracellular matrix regulation, together supporting connectivity-related processes. Calcium signaling is critical for induction of excitatory synaptic plasticity. FXS models display both calcium transient and calcium store release defects (Tessier and Broadie, 2011). This unreliable calcium signaling causes plasticity alterations (Meredith et al., 2007). Here, we identified genes related to calcium signaling that might be involved in plasticity-related disorders. Thus, we found that Itpr1 (IP3 receptor 1), which controls calcium release from intracellular stores and is an autism spectrum disorder-associated gene (Xu et al., 2012), is dysregulated in CA1 pyramidal neurons of Fmr1^−/y mice. Our RNAseq also revealed the Slc8a1 and Cacna1g genes, which code for a sodium/calcium exchanger and a voltage-dependent T-type calcium channel subunit respectively, and are both found to be mis-expressed in human neural progenitor cells from FXS patients (McMillan et al., 2012). Dysregulation of these three genes might thus participate to aberrant functional plasticity phenotypes. Alteration of calcium signaling dynamics in FXS models also suggests that activity-driven calcium signaling would couple structural and functional developmental changes in the FXS disease state (Doll and Broadie, 2014). Consistent with a role on spine morphology, we also identified genes involved in regulation of actin cytoskeleton such as Arhgef6 (Rac/Cdc42 guanine nucleotide exchange factor 6), Fgf11 (fibroblast growth factor 11), and Grlf1 (glucocorticoid receptor DNA binding factor 1) which could be responsible for elevated density of elongated spines in FXS patients (Irwin et al., 2001).

Dendritic spine defects resulting from FMRP loss have also been well-documented in mice (Comery et al., 1997; Portera Cailliau and Yuste, 2001). Our study shows a prevalence of immature spines in cultured hippocampal neurons from Fmr1^−/y mice compared to control mice, in agreement with previous data suggesting that the most consistent trouble with spines in FXS would be a deficiency in activity-dependent spine plasticity and maturation (He and Portera-Cailliau, 2013). Among structural plasticity-related functions, genes expressed at the ECM, linking pre- and postsynaptic elements, are of major interest. In particular, neural ECM proteases control learning and synaptic plasticity (Tsilibary et al., 2014). We found that in absence of FMRP, Klk8, a gene coding for an ECM serine protease highly expressed in CA1 pyramidal cells of the hippocampus (Chen et al., 1995), displayed one of the most significant changes in gene expression. Importantly, we show that reduced levels of KLK8 in CA1 pyramidal neurons contribute to the abnormal spine morphology observed in Fmr1^−/y mice by preventing the maturation of mushroom-shaped spines, further supporting the role of proteins involved in ECM regulation, such as the metalloproteinase 9 in the dendritic spine abnormalities associated with FMRP loss (Janusz et al., 2013; Gkogkas et al., 2014; Sidhu et al., 2014). In addition to its role in structural plasticity, KLK8 has also been reported to play a role in synaptic plasticity in the CA1 subfield as well as in some hippocampus-dependent learning (Tamura et al., 2006; Ishikawa et al., 2011). The loss of KLK8 has also been shown to predispose to seizures (Davies et al., 2001). Future studies will determine whether altered KLK8 expression also participates in the impaired synaptic plasticity, cognitive deficits and epileptiform discharges observed in Fmr1^−/y mice.

By using a cell-type specific RNAseq approach, our study shows that not only are both FMRP-bound and unbound mRNAs differentially translated, but also that this altered translation is bidirectional and restricted to a subset of genes. These findings are in agreement with a recent study describing a cell-type-specific translation profiling in juvenile Fmr1^−/y mice, using a BAC-TRAP strategy where the cholecystokinin (Cck) promoter drives EGFP-L10a expression (Thomson et al., 2017). Interestingly, the translatome profile identified by Thomson et al. strongly differs from ours. Several experimental conditions could explain this apparent discrepancy: the age of mice (25–32 days vs. 2–6 months), their genetic background (hybrid FVB × C57BL/6J vs. C57BL/6J mice), the promoter-driven expression of the tagged ribosomal subunit (Cck vs. Wfs1), the sequencing depth (50 or 75 bp, paired end vs. 50 bp single read) as well as the default false discovery rate (FDR) chosen for gene analyses (DESeq2 at the FDR of 0.1 vs. 0.05). Moreover, since heterogeneous ribosomes preferentially translate distinct subpools of mRNAs (Shi et al., 2017), the nature of the tagged ribosomal subunit (Rpl10a vs. Rpl22) has also to be carefully taken into consideration while performing translation profiling.

To conclude, our study identified gene candidates involved in neuronal connectivity-related functions, which should together support hippocampus-dependent processes. These genes could be new therapeutic targets to rescue physiological neuronal plasticity in FXS.

LC conceived and supervised the breeding strategy, designed, performed and analyzed all in vitro experiments. NB and EP performed biochemical and histological experiments. JV performed qRT-PCR. SR and DS performed the RNAseq and bioinformatic analyses. EV, JP, and EP conceived, designed, led the project and wrote the manuscript.