FMRP-Regulated Alternative Splicing is Multifactorial and Resembles Splicing Control by MBNL1

Fragile X Syndrome (FXS) is a neurodevelopmental disorder that is often modeled in Fmr1 knockout mice where the RNA binding protein FMRP is absent. Here we show that in Fmr1-deficient mice, RNA mis-splicing occurs in several brain regions and peripheral tissues. To assess molecular mechanisms of splicing mis-regulation, we employed N2A cells depleted of Fmr1. In the absence of FMRP, RNA-specific exon skipping events are linked to the splicing factors hnRNPF, PTBP1, and MBNL1. FMRP regulates the translation of Mbnl1 mRNA as well as Mbnl1 RNA auto-splicing. Elevated Mbnl1 auto-splicing in FMRP-deficient cells results in the loss of a nuclear localization signal (NLS)-containing exon. This in turn alters the nucleus-to-cytoplasm ratio of MBNL1. This re-distribution of MBNL1 isoforms in Fmr1-deficient cells could result in downstream splicing changes in other RNAs. Indeed, further investigation revealed that splicing distruptions resulting from Fmr1 depletion could be rescued by overexpression of nuclear MBNL1. Altered Mbnl1 auto-splicing also occurs in human FXS post-mortem brain. These data suggest that FMRP-controlled translation and RNA processing may cascade into a general dys-regulation of splicing in Fmr1-deficient cells.


Introduction
Fragile X Syndrome (FXS) is a neuro-developmental disorder characterized by mild to severe intellectual disability, speech and developmental delays, social impairment, perseveration, aggression, anxiety, and other maladies. FXS lies on the autism spectrum and is the most common single gene cause of autism. FXS is caused by an expansion of 200 or more CGG triplets in the 5' untranslated region (UTR) of FMR1, which in turn induces DNA methylation and gene silencing. Loss of the FMR1 gene product FMRP results in the disruption of neuronal circuitry and synaptic efficacy, which produces an array of neuro-pathological conditions [1][2][3]. FMRP, an RNA binding protein present in probably all cells is frequently studied in mouse hippocampus, where several studies show that it represses protein synthesis [4][5][6][7]. This observation, in conjunction with results showing that FMRP co-sediments with polysomes in sucrose gradients [8,9] and that in UV CLIPs (crosslink-immunoprecipitation) mostly to coding regions of mRNA [5,[10][11][12] suggests that it inhibits translation by impeding ribosome translocation. Indeed, it is now clear that at least one activity of FMRP is to stall ribosomes [5,7,[13][14][15]. How this occurs is unclear, but it could involve codon bias or optimality [16,17], impairment of ribosome function [18], or formation of translationally quiescent subcellular granules [13].
One group of FMRP target RNAs encodes chromatin modifying enzymes [5,19,20]. The synthesis of several of these enzymes is inhibited by FMRP; in its absence, excessive levels of these chromatin proteins alter the epigenetic landscape, which in turn impairs cognitive function [19]. A few mRNAs encoding epigenetic factors associate with FMRP-stalled ribosomes [15]. One of these, Setd2, encodes an enzyme that establishes the histone modification H3K36me3, which is most often located in gene bodies [21,22]. In Fmr1-deficient mouse brain, SETD2 protein levels are elevated, which in turn alter the distribution of H3K36me3 chromatin marks. H3K36me3 has been linked to alternative pre-mRNA splicing [23][24][25], and indeed there is some correlation between the genes with recast H3K36me3 and altered splicing in Fmr1-deficient mouse hippocampus [15]. The observation that Fmr1-deficiency results in hundreds of mis-splicing events prompted us to investigate both the prevalence and mechanism of FMRP-regulated nuclear pre-RNA processing.
We find that mis-splicing, mostly exon skipping, is widespread in Fmr1-deficient mice and occurs in all brain regions and peripheral tissues examined. To determine how FMRP might regulate splicing, we depleted Fmr1 from mouse N2A cells, which resulted in hundreds of mis-splicing events. We focused on specific exons in three RNAs that are aberrantly skipped or included in

RNA splicing mis-regulation in Fmr1 KO brain
Gene expression and RNA splicing are mis-regulated in the Fmr1-deficient mouse hippocampus [15] and FXS patient derived blood samples [26]. To determine whether this mis-regulation occurs in other brain regions and in peripheral tissues from mice, we sequenced RNA from (n=3, 2-3 month old) WT and Fmr1 KO hippocampus, cerebellum, and cortex, as well as liver, muscle, and testis (Fig 1A). Volcano plots show that hundreds of RNAs are up or down-regulated in Fmr1 KO cortex although fewer RNAs were similarly mis-regulated in hippocampus and cerebellum (log2FC > 0.2 or < -0.2, FDR < 0.05, n=3) (Fig 1B). A Venn diagram shows that a significant group of RNAs, mostly encoding proteins involved in synapse or cell junction formation, was shared between hippocampus and cortex (Fig 1C). In the cortex, many upregulated RNAs encode proteins involved in RNA processing for biogenesis, while down regulated RNAs code for proteins mediating membrane potential and synapse organization (Fig 1D). Analysis of these RNA seq datasets demonstrates that hundreds of RNAs are mis-spliced, mostly exon skipping, in the Fmr1deficient hippocampus, cortex, and cerebellum (p-value < 0.05, delta percent spliced-in [|delta PSI|] > 0.05, n=3) (Fig 1E). For the hippocampus and cortex, the percent of exons spliced in has a median of about 20% Fig 1F). In the cortex and hippocampus, RNAs displaying differential exon skipping between the two genotypes encode proteins involved in synapse organization and development and JNK signaling, respectively (Fig 1 and S1A). (B) Volcano plots of differential gene expression comparing WT and Fmr1-deficient cortex (CTX), hippocampus (HC), and cerebellum (CB). The numbers refer to those RNAs that are up or downregulated between the two genotypes (n=3, FDR < 0.05).
(C) Venn diagram comparing differential RNA levels from WT and Fmr1 KO HC, CTX, and CB (hypergeometric test, ***p < 0.001). GO terms for cellular components and adjust p-value for overlapped RNAs are indicated. Dcun1d2 constitutive Exon 2 was amplified to compare total mRNA levels between the genotypes and mean ± S.D is shown (Student's t-test, *p < 0.05, **p < 0.01).

Aberrant RNA splicing in Fmr1 KO peripheral tissues
Because FMRP is expressed in probably all tissues, we examined RNA splicing in WT and Fmr1 deficient liver, muscle (gastrocnemius), and testis. As with the brain, hundreds of RNAs are up or down regulated in FMRP KO peripheral tissues relative to WT (FDR, p < 0.05, n=3) (Fig 1, S1B), which may be somewhat surprising because relative Fmr1 levels (in transcripts per million, TPM) in these tissues are about one-tenth the amount in the brain (S1C). In the liver, RNAs that are up or down regulated in FMRP KO relative to WT encode factors involved in various metabolic processes and catabolic and phosphorylation events, respectively (S1D). In muscle, up or down regulated RNAs encode factors involved in extra cellular matrix organization and mitochondrial function, respectively (S1E). In testis, up or down regulated RNAs encode factors involved in cell division, and reproductive system development, respectively (S1F).
In FMRP KO peripheral tissues, splicing mis-regulation is widespread; in the brain there are mostly skipped exons but many mutually exclusive exons as well (p < 0.05, |delta PSI| > 0.05, n=3) (Fig 1H). In the liver and muscle, RNAs with differential exon skipping between the two genotypes encode chromatin modifying enzymes and Wnt signaling components, respectively (S1G). Comparison of the RNAs from all brain regions and peripheral tissues that display significantly different exon skipping between the two genotypes shows a remarkable degree of overlap (Fig 1I). For example, nearly 20% of RNAs with skipped exons in hippocampus are the same as in cortex, which might be expected. However, ~10% of RNAs with skipped exons in the liver also exhibit exon skipping in the hippocampus. In this same vein, ~9% of RNAs with skipped exons in the testis also show exon skipping in the cortex. These data indicate that if FMRP regulates exon skipping in one type of tissue (e.g., the brain), it is likely to do so in another tissue (e.g., liver).

FMRP-regulated splicing in N2A cells
To investigate the mechanism of FMRP-mediated splicing, we surmised that using a single cell type approach would be more efficacious compared to a tissue containing multiple cell types.
(G) pFlare system for assessing exon skipping and inclusion. Mapt exon 4 was inserted into pFlareA. When the exon is skipped, GFP is expressed; when the exon is included, RFP is expressed. This plasmid, as well as an empty pcDNA plasmid or one that expresses mouse

Rescue of mis-regulated splicing by FMRP replacement
To confirm FMRP control of splicing by an entirely different method, we used CRISPR/Cas9 gene editing to delete 7 nucleotides from exon 3 of Fmr1, which causes a reading frame shift to a stop codon resulting in nonsense mediated mRNA decay (Fig 2D; S2B) and a complete loss of FMRP ( Fig 2E). In these KO cells, loss of Mapt exon 4 skipping was nearly identical as observed with siFmr1 knockdown of Fmr1 (Fig 2F). We next generated a reporter construct where Mapt exon 4 and its flanking intron sequences were inserted into the pFlareA plasmid, which contains GFP and RFP sequences. Here, if Mapt exon 4 is skipped, an A nucleotide will generate a start codon when juxtaposed to a TG dinucleotide following splicing to the GFP reading frame and will express GFP. If Mapt exon 4 is included, RFP will be expressed. This plasmid, together with an FMRPexpressing plasmid or an empty control plasmid, were transfected into normal or Fmr1 KO N2A cells and green/red fluorescence intensity was analyzed by flow cytometry (Fig 2G). The quantification of Mean Fluorescence Intensity (MFI) for both mCherry and GFP was conducted through flow cytometry analysis in both the control and Fmr1 KO cells (Fig 2H). As depicted in Figure 2H, the inclusion of Mapt exon 4 was elevated in the KO in comparison to the control.
Interestingly, this inclusion was reversed upon the introduction of an FMRP overexpression plasmid. Moreover, the targeted deletion of a specific MBNL1 binding site, situated near exon 4, exhibited a reduction in Mapt exon 4 inclusion within the Fmr1 KO cells. Intriguingly, this binding site deletion showed no discernible effect on the control cells (S2C). The western blot shows the expression level of FMRP relative to GAPDH. The "rescuing" ectopic FMRP was expressed at ~10% of endogenous FMRP levels. In the FMRP KO cells, Mapt exon 4 in the reporter was more included relative to that observed in control cells, which replicates the data with endogenous Mapt exon 4 with both siFmr1 depletion (Fig 2C) and CRISPR/Cas9-edited Fmr1 KO cells (Fig 2F), albeit not to the same extent. Importantly, ectopic expression of FMRP in the KO cells restored Mapt exon 4 inclusion to control cells levels, demonstrating the reversibility of the exon skipping that is FMRP-dependent.

FMRP regulation of splicing factor activity
To identify splicing factors that might be regulated by FMRP, we focused on exons in three RNAs that are skipped or included in Fmr1-deficient cells and used the SFMap database [27] to identify potential splicing factor binding sites. Mapt exon 4, which is more included in Fmr1-deficient cells relative to control cells, is flanked by binding sites for splicing factors MBNL1, PTBP1, hnRNPF, and hnRNPQ (Fig 3A). We depleted the RNAs encoding each of these splicing factors as well as Fmr1 (S3A-H). Depletion of Mbnl1 resulted inclusion of exon 4 even more so compared to Fmr1 depletion. A double depletion of both Mbnl1 and Fmr1 caused even greater exon inclusion than the single depletions (Fig 3B). Constitutive Mapt4 exon 15 was unaffected by these depletions (S3I). Depletion of Ptbp1 also resulted in a greater inclusion of Mapt exon 4 than Fmr1 depletion.
(E) Summary of exon inclusion/skipping following Fmr1 and/or splicing factor depletion from N2A cells.

________________________________________________________________________
We next examined App exon 8, which is also flanked by MBNL, PTBP, hnRNPF, and hnRNPQ binding sites, is skipped more frequently upon Fmr1 depletion compared to control. Mbnl1 depletion caused App exon 8 skipping at the same frequency as Fmr1 depletion. A double depletion of Mbnl1 and Fmr1 was not additive for exon 8 skipping (Fig 3C). Depletion of hnRNPF, however, caused increased skipping of App exon 8 similar to that observed when Fmr1 was depleted. A double depletion was not additive for exon skipping. hnRNPQ depletion did not result in any change in App exon 8 skipping. Depletion of these factors had little effect on skipping/inclusion of constitutive App exon 2 (S3Q-T).
Finally, we examined Tnik exon 21, which is flanked by the same splicing factor binding sites, was included more frequently when Fmr1 is depleted (Fig 3D). While Mbnl1 depletion had no effect on Tnik exon 21 skipping/inclusion, depletion of both Ptbp1 and hnRNPF caused greater inclusion relative to controls (Fig 3D). Depletion of these factors had little effect on Tnik constitutive exon 25 (S3Q-T). A summary of all these data demonstrates that FMRP regulation of certain splicing factors influences inclusion or skipping of specific exons (Fig 3E).
To assess whether sequences surrounding an exon regulated by FMRP are bound by these RBPs, we analyzed published CLIP-seq and RIP-seq datasets for MBNL1 and PTBP1 [28,29].
We detected MBNL1 binding sites in Mapt exon 4, as supported by RIP-seq data. We also observed MBNL1 binding sites downstream of App exon 8, as evidenced by CLIP-seq data, and similarly in Ski exon 2 based on a combination of RIP-seq and CLIP-seq findings (S4). We found PTBP1 binding sites, characterized by the TCTCTC/CTCTCT motif, upstream of Mapt exon 4, both upstream and downstream of App exon 8, and upstream of Ski exon 2. Moreover, PTBP1 binding sites were situated downstream of Tnik exon 21 (S5).

FMRP regulates Mbnl1 RNA translation
To determine whether FMRP might regulate splicing factor expression directly, we first performed RNA co-immunoprecipitation experiments followed by RT-PCR for splicing factor RNAs. RNA [10]. We further found that around 50% of skipped or included exons in N2A cells contain binding sites for MBNL1, while non-target exons contain binding sites at a rate of 36% (Data S1 and Data S2) using RBPmap [30]. IgG and Srsf5 RNA served as a immunoprecipitation controls. All experiments were performed in triplicate. P-values were calculated using one-way ANOVA and mean ± S.D is shown. *p < 0.05; ****p < 0.0001. (B) Western blotting and quantification of splicing factors from control and Fmr1-depleted cells.
(C) Western blotting and quantification of MBNL1 in control and Fmr1-depleted cells following addition of the proteasome inhibitors MG132 or lactacystine for 0-4 or 6 hours. The histograms represent MBNL1 band intensities relative to GAPDH or tubulin.

___________________________________________________________________________
Western blotting of the splicing factors showed that MBNL1 and hnRNPQ were elevated ~1.5-2 fold upon Fmr1-depletion (Fig 4B). Because neither Mbnl1 nor hnRNPQ RNAs are altered by Fmr1 depletion (S3B and H), we infer that these two RNAs are under negative translational control by FMRP. Additional data showing that MBNL1 displays no differential stability in control versus Fmr1-depleted cells incubated with the proteasome inhibitors MG132 or lactacystine further indicate FMRP control of Mbnl1 RNA translation (Fig 4C). MBNL1 and hnRNPQ are each represented by two isoforms; in the case of MBNL1, the slow migrating isoform is reduced when Fmr1 is depleted while the fast migrating form is increased (Fig 4B). For hnRNPQ, the slow migrating isoform is unaffected while the fast migrating isoform is increased upon Fmr1 depletion ( Fig 4B). Neither PTBP1 nor hnRNPF undergo abundance changes in Fmr1-depleted cells ( Fig   4B).

FMRP regulates Mbnl1 RNA auto-splicing and MBNL1 localization
Two of the most frequently alternatively spliced exons of Mbnl1 mRNA are exon 5 and exon 7 (Fig 5A), of which exon 5 skipping arises by autoregulated splicing [31][32][33]. To determine whether alternative Mbnl1 auto-splicing is under FMRP control and involves either of these two exons, we performed RT-PCR with primers that distinguish between these exons. Fig 5B shows that exon 5 is skipped more frequently upon Fmr1 depletion while exon 7 and exon 10 (constitutive exon) skipping is unaffected. Exon 5, which contains a nuclear localization signal (NLS), determines whether MBNL1 is predominantly nuclear or is distributed to both nucleus and cytoplasm [33,34]. To assess whether exon 5 skipping upon Fmr1 depletion alters the nucleus/cytoplasmic ratio of MBNL1, we first performed western blots of protein from cells fractionated into these two compartments. Fig 5C shows that MBNL1 containing the NLS encoded by exon 5 (i.e., the upper band) decreased in the cytoplasm following Fmr1 depletion. Conversely, the NLS-lacking MBNL1 (lower band) increased in the cytoplasm when Fmr1 was depleted. The upper NLS-containing band was decreased in the nucleus after Fmr1 depletion. Immunocytochemical analysis of intact cells also shows that the MBNL1 nucleus/cytoplasmic ratio decreased upon Fmr1 depletion (Fig 5D), which is in concordance with the cell fractionation results. Quantification of the upper and lower MBNL1 bands relative to tubulin or Lamin B1 is indicated.
__________________________________________________________________________ FMRP shuttles to the nucleus [36] where it has been reported to co-localize with Cajal bodies [37], membrane-less structures that frequently coincide with the nucleolus. We detected a low amount of FMRP in the nucleus of N2A cells and considered that it may also associate with splicing factor-rich nuclear speckles [38]. Immunostaining for splicing factor SC35, which detects a few splicing proteins [39], showed abundant nuclear speckles but were not co-localized with FMRP, suggesting that FMRP is unlikely to regulate splicing directly (Fig 5E).
Because we had identified a correlation between elevated SETD2, dys-regulated H3K36me3 chromatin marks, and altered splicing in Fmr1 KO mouse hippocampus [15], we considered this might also occur in FMRP-deficient N2A cells. However, we observed no change in SETD2 levels in these cells, indicating that a changed chromatin landscape and altered splicing in FMRPdeficient cells may not be linked (S6A).

Alternative splicing of Mbnl1 RNA in FMRP-deficient cells and tissues
We analyzed published datasets to determine whether Mbnl1 exon skipping occurs in the FMRPdeficient tissues. Fig 6A shows that Mbnl1 exon 5 skipping is detected not only in Fmr1-depleted N2A cells, but also in mouse Fmr1 KO peripheral tissues (liver, muscle, testis). Moreover, exon 7, which is important for MBNL1 self-dimerization, is skipped in several peripheral tissues as well as cerebellum. Although the precise function of the dimerization is unclear, exon 7 residues are thought to increase MBNL1 affinity for RNA [40]. Somewhat surprisingly, we did not detect exon 5 skipping in mouse brain, although it and exon 4 were mutually exclusive exons in human Fragile X postmortem brain. For Mbnl2 RNA, we found that exon 5 skipping is increased in Fmr1-depleted N2A cells, but that exon 5 is more included in mouse Fmr1 KO HC. These data show that FMRPregulated alternative splicing of Mbnl1 and Mbnl2 is widespread, but that the exons involved in the splicing events vary according to tissue. It has been reported that specific exons are differentially alternatively spliced in various tissues due to different amounts/activities of splicing factors [41][42][43]. Our investigation confirms that FMRP not only influences the splicing of Mbnl1, but also impacts the splicing of several other RNA-binding protein mRNAs. Moreover, this effect of FMRP on splicing patterns is different across different tissue types. The figure shows the tissue type and frequency where specific exons in Mbnl1 RNA mis-spliced upon FMRP depletion [33,[44][45][46][47].

Effect of FMRP on splicing decisions by MBNL2 and PTBP2
MBNL1 and its paralog MBNL2 have the same binding sites on mRNA [28] as do the PTBP1 and its paralog PTBP2 [29]. Mbnl2 and Ptbp2 RNAs are present in N2A cells, but at lower levels that Mbnl1 (by ~2 fold) and Ptbp2 (by ~8 fold) RNAs, respectively (Fig 7A). Depletion of Fmr1 had no significant effect on Mbnl2 RNA, but interestingly, it increased MBNL2 protein levels and induced MBNL2 auto-splicing of exon 5 (Fig 7B, S4B). Similar to Mbnl1, Mbnl2 exon 5 bears an NLS, and thus determines the nuclear/cytoplasmic localization of this protein [40]. Moreover, Mbnl2 RNA has been reported as a binding target of FMRP [10]. Depletion of Fmr1 caused a reduction of Ptbp2 RNA levels, but had no effect on PTBP2 protein levels (Fig 7B). Depletion of Mbnl2 elicited an increase in Mapt exon 4 inclusion, significantly greater that Mbnl1 depletion (Fig 7C). These data suggest that MBNL1 and MBNL2 may both contribute to FMRP-controlled alternative splicing, but that PTBP2 is likely to have no significant effect. (E) Ptbp1 and Ptbp2 expression as well as Mapt exon 4 inclusion/skipping following Ptbp1 or Ptbp2 depletion. P-values were calculated using one-way ANOVA and mean ± S.D is shown (*p < 0.05, **p <0.01, ***p < 0.001; ****p < 0.0001).

___________________________________________________________________________
We propose that in N2A cells, Mbnl1 pre-mRNA undergoes alternative splicing such that exon 5containing and exon 5-lacking mRNAs are exported to the cytoplasm where they are bound by FMRP, which limits their translation. The MBNL1 protein that retains the NLS-encoding exon 5 is transported to the nucleus where it could influence alternative splicing of other pre-mRNAs. In Fmr1-deficient cells, exon 5-lacking Mbnl1 RNA is elevated in the cytoplasm relative to exon 5containing RNA, but because there is no FMRP to limit translation in these cells, MBNL1 synthesis is robust, which is particularly the case for those Mbnl1 mRNAs that lack exon 5 NLS. As a consequence, there is reduced MBNL1 transported to the nucleus, which may in turn have adverse effects on RNA splicing relative to normal cells. As presented in this study, MBNL1 is only one of several proteins through which FMRP regulates splicing.

Global transcriptomic changes induced by FMRP and MBNL1
To investigate whether FMRP-mediated splicing is influenced by the absence of nuclear MBNL1, we devised an experiment involving expression of either nuclear or cytoplasmic MBNL1 isoforms.
As a rescue strategy, the depletion of Fmr1 combined with the overexpression of the nuclear MBNL1 isoform (siFmr1+nMBNL1). To identify the distinct changes that could occur from the loss of the nMBNL1, the depletion of Mbnl1 coupled with the overexpression of the cytoplasmic MBNL1 isoform lacking exon 5 (siMbnl1+cMBNL1) (Fig 8A). Using the nMBNL1 and cMBNL1 plasmids, the localization of the expressed RNA in the nucleus and cytoplasm, respectively has been previously elucidated [48]. We validated the knockdown and overexpression of Fmr1 and Mbnl1 through qPCR analysis (Fig 8B and S7A). It is important to note that the depletion of Fmr1 or Mbnl1 did not exert a significant impact on each other. As shown in Fig 8C,  The volcano plots provide evidence of numerous significant alterations in mRNA expression under Fmr1 or Mbnl1 single knockdown conditions (Fig 8E and Fig 8F,

FMRP-mediated SE events are regulated by nuclear MBNL1
To elucidate the role of MBNL1 in mediating splicing events associated with FMRP, we conducted a comprehensive investigation of alternative splicing events across various groups. Our analysis revealed a substantial number of splicing alterations in cells depleted of Fmr1 either Mbnl1, with notable emphasis on changes within the SE and MXE categories (Fig 9A, Fig 9B, and S8A).
We successfully validated that the skipping of Mbnl1 exon 5 occurs upon Fmr1 depletion, and is included upon nMBNL1 overexpression. Conversely, Mbnl1 depletion and cotransfection with cMBNL1 results in cytoplasmic MBNL1 (S8B). Comparing SE under both conditions, we identified an overlap of 1458 RNAs between Fmr1 and Mbnl1 depleted cells, constituting over half of each dataset (Fig 9C). Among these, 1393 RNAs had the same alternative exon between the two groups (S8C). Our correlation analysis based on delta PSI values indicated a robust and statistically significant positive correlation between Fmr1 and Mbnl1 depleted cells, with a correlation coefficient (r) of 0.8382, R-squared (R²) value of 0.7026, and p-value of less than 0.0001 (Fig 9D).
We next examined whether splicing could be rescued by overexpression of nuclear MBNL1, which includes exon 5, in Fmr1-depleted cells. Remarkably, we found that 21% of mis-spliced exons in Fmr1-depleted cells were restored by nuclear MBNL1 (Fig 9E). Specific examples include Thap7 exon 2 and Slc30a4 exon 2, where Fmr1 knockdown disrupted splicing, but nMBNL1 overexpression reversed this effect (Fig 9G). We extended our inquiry to include a comparison of splicing events in Fmr1 depleted cells and cytoplasmic MBNL1 overexpressed cells, which had a reduced level of endogenous Mbnl1. Notably, 37% of mis-spliced exons in Fmr1-depleted cells exhibited congruent splicing patterns with the siFmr1 in the siMbnl1+cMBNL1 group, including Rps24 exon 5 and Atp2a3 exon 21 (Fig 9F and Fig 9H).

Discussion
The proteome of the hippocampus, an exceptionally well-studied brain region of Fragile X Syndrome model mice, is largely attributed to altered mRNA translation with perhaps a minor contribution of protein degradation [3,[49][50][51]. This study indicates that mis-regulated alternative splicing may be a contributor to the Fragile X proteome not only in the hippocampus and other brain regions of Fmr1-deficient mice, but in peripheral tissues as well. Our investigation of the mechanism of FMRP-mediated splicing used Fmr1-deficient N2A cells, which was based on the assumption that a single cell type would more likely reveal the involvement of specific factors than a complex mixture of cells such as in the brain. By mapping splicing factor binding sites flanking certain skipped or included exons in 3 mRNAs in Fmr1-depleted cells, we found that four proteins: MBNL1/2, PTBP1, and hnRNPF contribute to alternative splicing mis-regulation, and MBNL1/2 and hnRNPQ, are translationally inhibited by FMRP. Moreover, Mbnl1/2 auto-splicing induced skipping of the NLS-containing exon 5, which is thought to be enhanced by elevated levels of MBNL1/2 protein [40,47], was observed. This event impairs MBNL1/2 nuclear transport, which in turn likely affects downstream splicing decisions. AS events altered by FMRP and MBNL1 demonstrate a robust positive correlation. Moreover, the ectopic expression of MBNL1 isoform containing the NLS within exon 5 reversed approximately one-fifth of the disrupted splicing pattern in Fmr1-depleted cells. Conversely, expression of mainly the cytoplasmic MBNL1 recapitulated a proportion of the splicing changes observed upon Fmr1 depletion. In summary, our collective findings underscore the existence of discrete subsets of nuclear MBNL1-mediated splicing events within the context of Fmr1-regulated splicing.
Mbnl1 exon 5 is also skipped in Fmr1-deficient mouse peripheral tissues as well as in human postmortem Fragile X brain. Exons 6, 7, and 8 are skipped in neural stem cells, and/or liver, muscle, testis, and cerebellum from Fmr1-deficient mice. Thus, FMRP regulation of Mbnl1 splicing is complex and is strongly influenced by cell/tissue-type, which likely contributes to downstream splicing regulation.
The regulation of splicing via MBNL1/2 is only one of several FMRP-dependent mechanisms that mediate RNA processing. PTBP1 and hnRNPF all influence splicing decisions that are downstream of FMRP. For both MBNL1 and hnRNPQ, this involves FMRP-regulated translation of their respective mRNAs. In this sense, FMRP control of splicing is similar to FMRP control of chromatin modifications and transcription; the root cause of the alteration of these molecular events is dys-regulated translation when FMRP is absent [15,19]. We also considered whether FMRP might influence splicing directly. It is a nuclear shuttling protein that at least in mouse testis, binds chromatin and is involved in the DNA damage response [52]. FMRP co-localizes with Cajal bodies in Hela cells, which implies it may modify rRNA biogenesis [37]. We inferred that if FMRP was a direct regulator of splicing, it would co-localize with SC35-containing nuclear splicing/processing bodies or speckles [38]. We did not detect any such co-localization and thus FMRP is unlikely to be a direct modulator of splicing. In addition, we previously reported a correlation between the up-regulation of SETD2, altered H3K36me3 chromatin marks, and RNA splicing mis-regulation in Fmr1-deficient mouse brain [15]. In Fmr1 KO N2A cells, however, we detected no alteration in SETD2 levels, and thus a change in H3K36me3 leading to splicing dysregulation is unlikely. The notable disparity observed between the brain and the cellular model serves to highlight the intricacies of molecular regulation and the intricate manner in which FMRPmediated processes operate. The multifaceted interplay involving FMRP, SETD2, splicing factors, and the dysregulation of splicing emphasizes the need for a more comprehensive investigation into the mechanisms upon the specific cellular context.
In most cases, the dys-regulated inclusion/exclusion of exons in Fmr1-deficient tissues/cells has a mean of ~20%, but with a large distribution. Although the magnitude of such changes is within the range often observed for alternative splicing [53], it is unclear to what extent these splicing changes have biological consequences. However, even modest changes in exon skipping can manifest themselves with changes in biology if a skipped exon is regulatory. For example, an exon encoding a regulatory phosphorylation site in the RNA binding protein CPEB4 is skipped <30% of the time but this skipping is correlated with if not causative for autism [54]. In the Fmr1 KO mouse, we cannot ascribe any single mis-splicing event as contributing to a Fragile X phenotype. Instead, it is more likely that the amalgamation of hundreds of mis-splicing events result in some Fragile X pathophysiology, for example, dys-regulated synaptic transmission or learning and memory [7,55].
Finally, the dys-regulated splicing in Fragile X model mice may represent a point of convergence with other neurodevelopmental disorders [56]. For example, splicing is impaired in autism spectrum disorders [57], Rett Syndrome [58], Pten [59], and others [15]. Whether mis-splicing in these disorders are related mechanistically is unclear, but they may involve several of the same factors (e.g., MBNL1, PTBP1). More intriguing is the prospect that some mis-splicing events link similar behavioral or other physiological impairments among these disorders. This may especially be the case when very small microexons encoding regulatory domains are skipped [60]. Future studies will be necessary determine whether specific mis-splicing events promote pathophysiolgical outcomes.

Animals
Mice were housed under a 12 h light/dark cycle with free access to food and water. Wild-type and

Cytosol and nuclear protein fractionation
Cells were washed with ice-cold PBS, collected by trypsinization, pellets collected by centrifugation, and then resuspended in Triton extraction buffer (TEB, PBS containing 0.5% triton X-100 (v/v), 2 mM phenylmethylsulfonyl fluoride, 0.02% NaN3) and lysed on ice for 10 min.
Following a centrifugation at 12,000 rpm at 4 C, the supernatants were saved for cytoplasmic protein and the pellets were resuspended in nuclear lysis buffer (50 mM Tris-HCl pH 7.4, 120 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% SDS) and lysed by sonication at high power for 8 cycles (15 sec on, 60 sec off) using a Bioruptor (Diagenode). The lysates were collected after centrifugation at 13,000 rpm for 10 min at 4 C and the supernatants were prepared for nuclear protein analysis. Nuclear and cytoplasmic protein concentrations were measured using BCA assays.

RNA-seq
Mouse tissues were powdered in liquid nitrogen with a frozen mortar and pestle. For RNA extraction, TRIzol was added to the tissue powder and homogenized with Dounce tissue homogenizer. The RNA was treated with TurboDNase (Invitrogen, AM2238) to remove genomic DNA contamination. For peripheral tissues and N2A cells, total RNA was extracted and the integrity analyzed by a fragment analyzer. Library preparation and RNA sequencing were

Differential expression and alternative splicing analysis
RNA-seq analysis was performed using DolphinNext pipeline at UMass Chan Medical School [61]. Quality trimming was conducted using Fastqc (v0.11.8) and Trimmomatic (v.0.39). Reads below a minimum quality PHRED score of 15 at the 10nt sliding window were first clipped and the clipped reads shorter than 25nt were trimmed. The trimmed reads were mapped to rRNA by Bowtie2 (v2.3.5) were further filtered out. The cleaned reads were aligned to the mouse reference genome (mm10) with STAR (v1.16.1), and gene expression was quantified by RSEM (v1.3.1).
Differential gene expression was analyzed using DESeq2 (v1.16.1). The FDR adjusted p-value < 0.05 and log2FC > 0.2 or < -0.2 was used as the cut-offs to identify the differentially expressed genes. Alternative splicing events are analyzed using rMATS (v3.0.9) [62] and p-value < 0.05 and |delta PSI|> 0.05 was used as the cut-offs for splicing events. To assess biological function, Gene Ontology (GO) term analysis was conducted using clusterProfiler R package [63,64]. Significant RNA Overlap from WT and Fmr1 KO hippocampus, cortex, and cerebellum was analyzed using DynaVenn [65] using p-value ordered RNA list.

Generation of an Fmr1 CRISPR/Cas9-edited cell line
To construct an Fmr1 KO N2A mouse cell line, an Fmr1 exon 3 DNA oligonucleotide was inserted into pLentiCRISPR (Addgene, 49535) adapted from published methods [12]. Briefly, annealed and phosphorylated oligonucleotides were cloned into a FastDigest BmsBI (Fermentas)-digested vector following the manufacturer's protocol. pLentiCRISPR-mFmr1Exon3 was co-transfected with pMD2.G and psPAX2 into HEK293T cells. The viral particles containing supernatants were collected after 48 h of transfection by filtering through 0.45 µm filters and transduced to N2A cells.
After 3 days of infection, transduced cells were selected with puromycin for 2 weeks. Puromycin resistant cells were seeded in each well of a 96 well plate with a single cell per well. Single cellderived colonies were obtained after several weeks of culture and verified for Fmr1 knockout by Sanger DNA sequencing and western blotting. For the sequencing, genomic DNA was extracted using lysis buffer (10 mM Tris 8.0, 200 mM NaCl, 20 mM EDTA. 0.2% Triton X-100 and 100 µg/ml proteinase K) and the deleted exon region was PCR amplified using primers (sequences noted below). To identify deleted sequences, the PCR products were cloned with a TOPO TA Cloning Kit (ThermoFisher Scientific, 450030) followed by sequencing using T7 primers (Genewiz).

Alternative splicing reporter system
To generate an alternative splicing reporter, total DNA was isolated from N2A cells using the lysis buffer described above. Mapt exon 4 and flanking the intron regions were PCR amplified using Phusion High-Fidelity DNA polymerase and inserted into NheI/BamHI digested pFlareA Plasmid

RNA-Immunoprecipitation (RNA-IP)
N2A cells were transfected with siNT and siFmr1 using Lipofectamine 3000. After 72 h of incubation, the cells were washed with fresh media containing 100 µg/ml cycloheximide (CHX, MilliporeSigma, C4859). After washing with ice-cold PBS-containing CHX, the cells were pelleted and lysed in 1X polysome buffer [20 mM Tris-HCl pH 7.5, 5 mM MgCl2, 100 mM KCl, 1 mM DTT, 100 µg/ml CHX, protease inhibitor cocktails, 1 % Triton X-100 (v/v)] with 10 passages through a 25 G needle to triturate and incubated on ice for 10 min. The lysates were centrifuged at 14,000 x g at 4 C and RNA concentration was measured using Qubit BR RNA Assay Kits (ThermoFisher Scientific, Q10210). For IP, 5 µg of RNA was precleared with 25 µl of Protein G Dynabeads (Invitrogen, 10003D) for 30 min at 4 C. 10% of aliquot of the precleared lysates were saved as an input. 2.5 µg of FMRP antibody (Abcam, ab17722) or IgG (MilliporeSigma, 12-370) was added to the precleared lysates and incubated for 2 h at 4 C. 25 µl of Protein G Dynabeads was added and incubated for 30 min at 4 C and the beads were gently washed with wash buffer [20 mM Tris-HCl, 100 mM KCl, 5 mM MgCl2, 1% Triton X-100 (v/v)] for 3 times. RNA was extracted using TRIzol and 100 ng of RNA were reverse transcribed using Quantitect followed by qPCR using iTaq SYBRgreen (Bio-rad, 1725122).

Protein stability assay
N2A cells were transfected with siNT and siFmr1 using Lipofectamine3000. After 72h of incubation, cells were treated with 10 M MG132 (MilliporeSigma, 474790) or Lactacystine (Tocris, 2267) and harvested at different time points.

Immunocytochemistry
For immunofluorescent staining, 1x10 5 cells were seeded in a Chamber Slide (Nunc Lab-Tek II CC2, 154917) and transfected with 10 pmol siNT and siFmr1. After 48 h, cells were washed and fixed with 4 % formaldehyde solution (ThermoFisher Scientific, AAJ19943K2) for 10 min at RT.
Images were acquired using a Zeiss confocal microscope LSM900.

RBP binding exons
To determine whether sequences surrounding alternative exons are bound by MBNL1, CLIP-seq and RIP-seq data in MBNL Interactome Browser (MIB.amu.edu.pl) [28] were used. MBNL1 binding regions within alternative exons and/or adjacent intron of Mapt, App, Ski and Tnik were investigated. Using the CLIP list [29], alternative exons activated or inhibited by PTBP1 and regulated by FMRP were compared.

Data availability
RNA-seq datasets are available in GEO (accession number GSE207145).