As a first step toward investigating FMR1 reactivation as a therapeutic approach for FXS, we sought to identify epigenetic regulators that promote silencing of FMR1. Toward this end we assembled a small-scale short hairpin RNA (shRNA) library comprising 162 shRNAs directed against 33 well characterized epigenetic regulators that mediate transcriptional repression and that write or erase the epigenetic marks known to be associated with the epigenetically silenced FMR1 promoter (Supplementary Figure 1A). Each shRNA was packaged into lentivirus particles and transduced into an induced pluripotent stem cell (iPSC) line derived from a male patient with FXS (FXS 848-iPS3 cells, hereafter called FXS 848-iPSCs) (Sheridan et al., 2011). Twenty days post-transfection, mRNA was prepared and FMR1 expression analyzed by quantitative RT-PCR (qRT-PCR). We considered a positive result to be at least two unrelated shRNAs directed against the same gene that elicited a statistically significant: (1) increase in FMR1 expression, and (2) decrease in mRNA levels of the target gene, compared to that obtained with a control non-silencing (NS) shRNA. The results obtained in the complete screen of the 162 shRNAs are shown in Figure 1A (and see also Supplementary Figure 1B) and enabled us to identify nine epigenetic regulators of the silenced FMR1 gene: DNMT1, EZH2, RNF2 (also called RING1B), SUV39H1, KDM5C, KDM5D, HDAC5, HDAC10, and SIRT5, whose functions are summarized in Supplementary Figure 1C. For convenience, we refer to the factors that promote FMR1 silencing as FMR1 Silencing Factors (FMR1-SFs). Notably, three of the FMR1- SFs we identified, DNMT1, SUV39H1, and EZH2 have been previously implicated in silencing of the CGG repeat-containing FMR1 gene (Bar-Nur et al., 2012; Bulut-Karslioglu et al., 2012; Kumari and Usdin, 2016; Tabolacci et al., 2016; Kumari et al., 2020; Vershkov et al., 2022).

KDM5C and KDM5D are lysine-specific histone demethylases with 86% amino acid identity and 91% amino acid similarity (Kent-First et al., 1996). It was therefore important to confirm the specificity of the KDM5C and KDM5D shRNAs. Analyses by qRT-PCR (Supplementary Figure 2A) and immunoblot (Supplementary Figure 2B) show that the KDM5C shRNAs efficiently knocked down KDM5C but not KDM5D, whereas the KDM5D shRNAs efficiently knocked down KDM5D but not KDM5C. These results confirm the specificity of the KDM5C and KDM5D shRNAs, and in conjunction with our other results presented above and below demonstrate that both KDM5C and KDM5D contribute to epigenetic silencing of FMR1.

To determine the level of FMR1 reactivation obtained following shRNA-mediated knockdown of an FMR1-SF we analyzed in parallel an iPSC line derived from a normal individual (BJ1-iPS4 cells), hereafter called normal iPSCs (Sheridan et al., 2011). The qRT-PCR results of Figure 1B show that shRNA-mediated knockdown of an FMR1-SF in FXS 848-iPSCs reactivated the epigenetically silenced FMR1 gene to ∼10–20% of normal levels at 20 days following lentivirus transduction, the time at which we found FMR1 reactivation was maximal (Supplementary Figure 3A). We observed a similar level of FMR1 reactivation following knockdown of an FMR1-SF in a TaqMan assay (Supplementary Figure 3B). The immunoblot results of Figure 1C show that knockdown of an FMR1-SF also restored FMRP protein to ∼10–20% of normal levels. Reactivation of epigenetically silenced FMR1 following knockdown of each FMR1-SF was confirmed by qRT-PCR and immunoblotting in a second male FXS iPSC cell line, SC135 cells [FXS SC135-iPSCs; (Brick et al., 2014; Supplementary Figures 4A–D)]. By contrast, in normal iPSCs, FMR1-SF knockdown had no effect on FMR1 expression (Supplementary Figure 4E).

A characteristic feature of epigenetically silenced FMR1 is the presence of DNA hypermethylation (Oberle et al., 1991; Pietrobono et al., 2005; Tabolacci et al., 2008a). The bisulfite sequencing experiment of Figure 1D shows, as expected, that the FMR1 promoter is hypermethylated in FXS 848-iPSCs. Consistent with previous studies (Bar-Nur et al., 2012; Kumari and Usdin, 2014; Tabolacci et al., 2016), treatment of FXS 848-iPSCs with the DNMT inhibitor 5-aza-2′-deoxycytidine (5-aza-dC) led to a substantial decrease in DNA hypermethylation (Figure 1D and Supplementary Figure 5A). Notably, there was a similar decrease in DNA hypermethylation following knockdown of each of the nine FMR1-SFs (Figure 1D and Supplementary Figure 5B). Collectively, these results indicate that the nine FMR-SFs we identified mediate silencing of FMR1 in FXS iPSCs.

Epigenetic regulators are typically stably (but reversibly) associated with the promoters and/or genes upon which they act. To determine whether the nine FMR1-SFs are stably associated with the epigenetically silenced FMR1 promoter we performed a chromatin immunoprecipitation (ChIP) experiments. The ChIP experiment of Figure 1E shows that eight of the nine FMR1-SFs are specifically bound to the epigenetically silenced FMR1 promoter in FXS 848-iPSCs and not the transcriptionally active FMR1 promoter in normal iPSCs. The single FMR1-SF that is not associated with the epigenetically silenced FMR1 promoter is SIRT5, which is a mitochondrial protein (Michishita et al., 2005). Thus, although SIRT5 promotes FMR1 silencing, unlike the other FMR1-SFs it functions indirectly.

For several of the FMR1-SFs we identified there are well-described small molecule inhibitors. Figure 2A shows that epigenetically silenced FMR1 could be reactivated by treatment with 5-aza- dC, consistent with the results of previous studies (Bar-Nur et al., 2012; Kumari and Usdin, 2014; Tabolacci et al., 2016; Vershkov et al., 2022). Notably, epigenetically silenced FMR1 was also reactivated following treatment with small molecule inhibitors of EZH2 (EPZ6438, GSK126), SUV39H1 (chaeotocin) and RNF2 (PRT4165). Reactivation was observed by analysis of both FMR1 mRNA (Figure 2B and Supplementary Figure 6) and FMRP protein (Figure 2C) and was again ∼10–20% of normal levels. Similar results were obtained in FXS SC135-iPSCs (Supplementary Figure 7).

We performed several additional experiments with the EZH2 inhibitor EPZ6438 (also called tazemetostat), which recently received FDA approval for the treatment of certain cancers (Straining and Eighmy, 2022).¹ The EPZ6438 titration experiment of Supplementary Figure 8 reveals a very good correlation between the loss of EZH2 enzymatic activity, as evidenced by decreased total H3K27me3, and reactivation of epigenetically silenced FMR1. The time course experiment of Figure 2D shows that the level of FMR1 reactivation increased over 96 h. Withdrawal of EPZ6438 resulted in re-silencing of FMR1, which again occurred over a time course of ∼96 h. The ChIP experiment of Figure 2E shows that the association of DNMT1 with the epigenetically silenced FMR1 promoter was well correlated with the kinetics of FMR1 reactivation following EPZ6438 addition and re-establishment of FMR1 silencing following EPZ6438 withdrawal. Collectively, these results indicate that both silencing and reactivation of FMR1 are reversible.

The experiments described above were performed in undifferentiated FXS iPSCs. We next asked whether inhibition of the same set of nine FMR1-SFs would also reactivate epigenetically silenced FMR1 in FXS neural progenitor cells (NPCs) and post-mitotic neurons, the latter of which is the cell type most relevant to FXS. For these experiments, we used an FXS NPC cell line that was derived from FXS 848-iPSCs [hereafter called FXS 848-NPCs; (Sheridan et al., 2011)]. Knockdown of any one of the nine FMR1-SFs reactivated epigenetically silenced FMR1 in FXS 848-NPCs at both the mRNA (Figure 3A) and protein (Figure 3B) levels. Epigenetically silenced FMR1 was also reactivated in FXS 848-NPCs by small molecule inhibitors of FMR1-SFs including 5- aza-dC, chaetocin, EPZ6438, GSK126, and PRT4165, at both the mRNA (Figure 3C) and protein (Figure 3D) levels.

To derive FXS post-mitotic neurons, FXS 848-iPSCs were transduced with a lentivirus expressing Neurogenin-1 and Neurogenin-2 according to published methods (Busskamp et al., 2014); for convenience we refer to these cells as FXS 848-neurons. Neuronal differentiation was assessed and confirmed by staining with neuronal markers including TUJ1, MAP2, and NeuN (Figure 4A and Supplementary Figure 9A). Glial fibrillary acidic protein (GFAP)-positive glial cells were not detected in the FXS 848-neuronal cultures (Figure 4A). As expected, the FXS 848-neurons were post-mitotic as evidenced by the lack of the mitotic marker phosphorylated histone H3 (Figure 4B). Treatment of FXS 848-neurons with an FMR1-SF shRNA or small molecule FMR1-SF inhibitor reactivated epigenetically silenced FMR1 (Figure 4C). Similar to what we found in FXS 848-iPSCs (Figure 2D), in FXS 848-neurons both reactivation and silencing of FMR1 with the EZH2 inhibitor EPZ6438 was reversible (Figure 4D). As expected, EPZ6438 treatment substantially reduced total H3K27me3 levels in FXS 848-neurons (Supplementary Figure 9B).

To rule out the possibility that the FMR1 reactivation we observed was from contaminating dividing cells, FXS 848-neuronal cultures were treated with a potent inhibitor of DNA synthesis, cytosine arabinoside (Ara-C), which kills proliferating cells (Henriquez et al., 2013). We found that Ara-C treatment had no effect on the ability of FMR1-SF shRNAs and small molecule FMR1- SF inhibitors to reactivate epigenetically silenced FMR1 (Supplementary Figure 9C), indicating that the observed FMR1 reactivation was indeed from post-mitotic cells.

We found it was technically challenging to obtain a sufficient number of iPSC-derived neurons for quantifying FMRP levels by immunoblot analysis. Therefore, as in previous studies (Bar-Nur et al., 2012; Doers et al., 2014), to monitor FMRP levels in neurons, we performed immunocytochemistry (ICC). Figure 4E shows the ICC results following knockdown of a representative FMR1-SF, EZH2 (left), the FMRP level from shEZH2 is significantly higher than shNS but much weaker than normal neuron, and a quantitative summary of the results of all FMR1- SF shRNAs and small molecule FMR1-SF inhibitors (right), which confirm FMR1 reactivation at the protein level.

We confirmed these results in a second FXS neuronal cell line, FXS SC135-neurons, which were derived by transducing FXS SC135-iPSCs with a lentivirus expressing Neurogenin 1 and Neurogenin 2 as described above. Neuronal differentiation was confirmed by staining with neuronal markers including TUJ1 and NeuN (Supplementary Figure 10A), and the FXS SC135-neurons were post- mitotic as evidenced by the lack of phosphorylated histone H3 (Supplementary Figure 10B). Treatment of FXS SC135-neurons with an FMR1-SF shRNA or small molecule FMR1-SF inhibitor reactivated epigenetically silenced FMR1 at both the mRNA (Supplementary Figure 10C) and protein (Supplementary Figure 10D) levels, which was unaffected by pre-treatment with Ara-C (Supplementary Figure 10E).

Finally, we sought to confirm our key findings using an isogenic pair of FXS cell lines in which the CGG repeats are either intact (CGG-intact) or have been excised using CRISPR/Cas9- mediated deletion (CGG-excised) (Xie et al., 2016). For these experiments, we monitored FMR1 reactivation following knockdown or inhibition of a representative FMR1-SF, EZH2. Consistent with our results in other FXS cell lines, treatment of FXS (CGG-intact) iPSCs and neurons with EZH2 shRNAs or small molecule EZH2 inhibitors restored FMR1 mRNA or FMRP protein to ∼10–20% of the levels observed in FXS (CGG-excised) iPSCs and neurons (Supplementary Figures 11A–D).

We next performed a series of experiments to determine whether the level of FMR1 reactivation we obtained was sufficient to normalize characteristic molecular abnormalities of FXS neurons. The transcription factor REST is a master negative regulator of neurogenesis that controls the pool size and timing of differentiation of various neural lineages (Schoenherr and Anderson, 1995; Chen et al., 1998; Covey et al., 2012; Satoh et al., 2013). REST is expressed in embryonic stem cells, NPCs, and non-neuronal cells, where it suppresses neuron-specific genes, and is not expressed in differentiated neurons (Ballas et al., 2005). FMRP helps maintain the levels of a neural-specific miRNA (hsa-mir-382) (Halevy et al., 2015), which is a repressor of REST translation. Thus, in the absence of FMRP, the levels of hsa-mir-382 are decreased, preventing the differentiation- dependent downregulation of REST. The resulting higher levels of REST in FXS neurons lead to the suppression of axonal guidance and other genes important for neural development.

Consistent with previous studies, qRT-PCR analysis revealed that compared to their normal counterparts, FXS 848-neurons contain increased levels of REST (Figure 5A) and decreased levels of the REST target axonal guidance genes DCC, ROBO3, and SLIT1 (Figure 5B). Treatment of FXS 848-neurons with an FMR1-SF shRNA or small molecule FMR1-SF inhibitor substantially decreased expression of REST (Figure 5A) and increased expression of DCC, ROBO3, and SLIT1 (Figure 5B). Similar results were obtained with FXS SC135-neurons (Supplementary Figure 12).

A potential concern of the above results is that the normalized expression of REST, DCC, ROBO3, and SLIT1 we observed may be an indirect effect resulting from inhibition of an epigenetic regulator rather than a direct effect of FMR1 reactivation. To address this concern, we first asked whether normalized expression of REST, DCC, ROBO3, and SLIT1 was dependent on FMR1. We found that the decreased expression of REST and increased expression of DCC, ROBO3, and SLIT1 observed following treatment of FXS 848-neurons with an FMR1-SF shRNA or small molecule FMR1-SF inhibitor was abrogated by shRNA-mediated knockdown of FMR1, indicative of FMR1 dependence (Supplementary Figures 13A–C). We also asked whether normalization of REST target gene expression following FMR1 reactivation was due to the decreased levels of REST. We found that the increased expression of DCC, ROBO3, and SLIT1 observed following treatment of FXS 848-neurons with an FMR1-SF shRNA or small molecule FMR1-SF inhibitor was abrogated by ectopic expression of REST (Supplementary Figures 13D, E). Collectively, these results demonstrate that the normalized expression of molecular markers following FMR1-SF inhibition is due to FMR1 reactivation and consequent FMRP-mediated down-regulation of REST.

FMRP has been shown to physically associate with the mRNA encoding DGKK, a kinase that controls the switch between diacylglycerol and phosphatidic acid signaling pathways (Tabet et al., 2016). The absence of FMRP in FMR1 knockout mouse cortical neurons results in decreased levels of DGKK, and shRNA-mediated loss of function of Dgkk is sufficient to cause dendritic spine abnormalities, synaptic plasticity alterations, and behavioral disorders similar to those observed in FMR1 knockout mice (Tabet et al., 2016). Notably, ectopic expression of DGKK rescues the dendritic spine defects of FMR1 knockout neurons (Tabet et al., 2016) and adeno- associated viral vector delivery of DGKK corrects abnormal diacylglycerol and phosphatidic acid homeostasis and behavioral disorders in FMR1 knockout mice (Habbas et al., 2022). Consistent with these published results, immunoblot analysis showed that DGKK was readily detectable in normal but not in FXS 848-neurons (Figure 5C). Treatment of FXS 848-neurons with an FMR1-SF shRNA or small molecule FMR1-SF inhibitor substantially increased DGKK levels. The increased levels of DGKK observed following treatment with an FMR1-SF shRNA or small molecule FMR1-SF inhibitor was abolished upon shRNA-mediated knockdown of FMR1 confirming the FMR1-dependence of this effect (Supplementary Figure 14A). Similar results were obtained with FXS SC135- neurons (Supplementary Figures 14B, C).

In the final set of experiments, we elected to focus on the FMR1-SF EZH2 because EZH2 inhibition is well tolerated in cultured cells, mice and humans, and, as stated above, because small molecule EZH2 inhibitors are clinically well advanced, with EPZ6438 recently receiving FDA approval for the treatment of certain cancers (see text footnote 1; Straining and Eighmy, 2022).

Previous studies have shown that loss of FMRP results in the characteristic electrophysiological abnormality of neuronal hyperexcitability [reviewed in (Contractor et al., 2015)]. Reactivation of FMR1 by targeted demethylation of the CGG repeats has been shown to ameliorate the hyperexcitability of FXS neurons (Liu et al., 2018). We therefore asked whether FMR1 reactivation resulting from inhibition of the FMR1-SF EZH2 could normalize the hyperexcitability of FXS neurons. In these experiments, we measured neuronal hyperexcitability using multielectrode arrays (MEAs). In brief, FXS 848-neurons were treated with either an EZH2 shRNA or the small molecule EZH2 inhibitor EPZ6438 and cultured on MEAs, and the spontaneous firing frequency, a measure of excitability, was monitored over a 48-day time course. Consistent with previous studies (Liu et al., 2018), the MEA results revealed that compared to their normal counterparts, FXS 848-neurons displayed increased firing frequency, indicative of hyperexcitability (Figure 6A). Treatment of FXS 848-neurons with an EZH2 shRNA or EPZ6438 substantially decreased the firing frequency. Thus, the level of FMR1 reactivation obtained by knockdown or pharmacological inhibition of EZH2 is sufficient to correct the hyperexcitability of FXS neurons.

Currently, there is no FXS mouse model to directly analyze the therapeutic benefit of FMR1 reactivation, because incorporation of the CGG repeat expansion in mice fails to recapitulate the DNA hypermethylation and transcriptional silencing of FMR1 that occurs in human cells (Dahlhaus, 2018). As an alternative approach, two groups have reported the development of a mouse engraftment model to analyze FMR1 reactivation in the central nervous system (CNS) (Liu et al., 2018; Vershkov et al., 2019). In this model, FXS NPCs are stereotaxically injected into the brains of mice, and the grafts are subsequently analyzed for human FMR1/FMRP expression.

We therefore sought to test whether EZH2 inhibition could reactivate epigenetically silenced FMR1 in this mouse engraftment model. Current EZH2 inhibitors do not efficiently cross the blood-brain barrier (Zhang et al., 2015), which limits their potential use as FXS therapeutics. Recently, antisense oligonucleotide (ASO)-based approaches have been developed as effective treatment options for certain CNS disorders (Bennett et al., 2019). Accordingly, we derived two individual EZH2 ASOs that efficaciously reduced EZH2 expression (Supplementary Figure 15A) and decreased H3K27me3 levels in FXS 848-neurons (Supplementary Figure 15B). Treatment of cultured FXS 848-neurons with either of the EZH2 ASOs reactivated FMR1 to ∼20% of normal levels, comparable to EZH2 small molecule inhibitors (Figure 6B). As expected, the EZH2 ASOs also substantially decreased expression of REST (Figure 6C), increased expression of DCC, ROBO3 and SLIT1 (Figure 6D), and increased DGKK levels (Figure 6E). Furthermore, the EZH2 ASOs corrected the hyperexcitability of FXS 848-neurons (Figure 6F). Similar results were obtained in neurons derived from the isogenic FXS (CGG-excised/intact) iPSC lines (Supplementary Figures 15C–G), although the increased expression of DCC, ROBO3, and SLIT1 is not significant may due to cell line difference.

To test whether EZH2 ASOs could reactivate FMR1 in vivo, FXS 848-NPCs were stereotaxically injected into the cerebral lateral ventricles of mice. Three days after transplantation, mice received an intracerebroventricular injection of an EZH2 ASO, and 7 days later, mice were sacrificed and the brain grafts collected and dissected. Analysis by qRT-PCR using primers specific for the human FMR1 transcript (Vershkov et al., 2019) showed that EZH2 ASOs reactivated expression of FMR1 to ∼15% of normal levels (Figure 6G). Furthermore, immunofluorescence staining of mouse brain sections with a human-specific anti-FMRP antibody revealed that EZH2 ASO treatment also resulted in increased FMRP levels (Figure 6H). Collectively, these results demonstrate that EZH2 ASOs can reactivate FMR1 in vivo.

BJ1-iPS4 cells (normal iPSCs), FXS 848-iPS3 cells (FXS 848-iPSCs), and NPCs derived from normal 8330-iPS8 cells (normal NPCs) or FXS 848-iPS3 cells (FXS 848-NPCs) passage 4–6 (Sheridan et al., 2011) were kindly provided by Stephen J. Haggarty (Harvard Medical School). For this study, the FXS 848-iPS3 passage 6–8 were used for shRNA, small molecule treatment and neuron differentiation. In The FXS iPSC line SC135 (FXS SC135-iPSCs) (Brick et al., 2014) was kindly provided by Marius Wernig (Stanford School of Medicine). The isogenic FXS (CGG- excised/intact) iPSC lines (Xie et al., 2016) were kindly provided by Peng Jin (Emory University School of Medicine). Normal and FXS iPSCs and NPCs were authenticated by qRT-PCR analysis to validate the expected FMR1 expression status, and by PCR to confirm the expected CGG repeat length in the FMR1 5′ UTR. Upon receipt, cells were tested for mycoplasma contamination and found to be negative. iPSCs were cultured in mTeSR1 medium (STEMCELL Technologies) on matrigel-coated plates. NPCs were maintained in neural expansion medium as previously described (Sheridan et al., 2011). Neurons were induced from normal iPSCs, FXS 848-iPSCs or FXS SC135 iPSCs by expression of reverse tetracycline transactivator (rTA3), Neurogenin-1 and Neurogenin-2, and cultured in mTeSR1 and neuron growth medium as previously described (Busskamp et al., 2014; Lam et al., 2017).

The 162 shRNAs (listed in Supplementary Table 1) from The RNAi Consortium (TRC) and shRNA^mir (pGIPZ) lentiviral human shRNA libraries (Thermo Fisher Scientific) were obtained through the University of Massachusetts RNAi Core Facility and packaged into lentiviruses. FXS 848-iPSCs were seeded in 12-well plates at a density of 1x10⁴ cells/well and transduced with 200 μl lentivirus (at MOI 5) with 10 μg/ml Polybrene (Qiagen) overnight. Two days later, cells were selected with 1.5 μg/ml puromycin for 3 days. Cells were split at days 12 and 18, and harvested at day 20 for analysis of FMR1 and target gene expression. Based on this analysis, the two most efficacious shRNAs that increased FMR1 expression and decreased mRNA levels of the target gene were selected for each FMR1-SF (indicated in Supplementary Table 1).

iPSCs were seeded at 1x10⁴ cells/well in 12-well plates at 5% confluency, NPCs were seeded at 5x10⁴ cells/well in 12-well plates at 10% confluency, and neurons (at day 4 of the differentiation process) were seeded at 1x10⁵ cells/well in 6-well plates at 40% confluency, and transduced overnight with 200–1000 μl lentivirus (at MOI 5) expressing an FMR1-SF shRNA. Two days later, cells were selected with 1.5 μg/ml puromycin for 3 days. iPSCs and NPCs were split at days 12 and 18 at 60–80% confluency, re-seeded at 10% confluency for iPSC and 20% for NPC, and harvested at day 20 at 20% confluency for iPSCs and 30% confluency for NPCs for subsequent experiments (i.e., qRT-PCR, immunoblotting, bisulfite sequencing, and immunofluorescence). Neurons were harvested at day 12 at 40–50% confluency for subsequent experiments.

iPSCs, NPCs and neurons were seeded as described above and 24 h later treated with the following small molecule inhibitors: 5-aza-2′-deoxycytidine (Sigma) at 1 μM and refresh every 24 h to 96 h, chaetocin (Cayman Chemical) at 0.5 μM, EPZ6438 (Cayman Chemical) at 5 μM, GSK126 (APExBIO) at 5 μM, and PRT4165 (Tocris Bioscience) at 5 μM for 96 h. 5-aza-2′-deoxycytidine and chaetocin are toxic to cells, there are 10–30% cell death. iPSCs, NPCs and neurons were harvested at the confluency described above for all subsequent experiments. For the EPZ6438 addition experiments of Figures 2D, 4D, FXS iPSCs or neurons were collected at different time points following incubation with 5 μM EPZ6438. For the EPZ6438 withdrawal experiments, cells were treated with 5 μM EPZ6438 for 96 h, then washed 3 times with PBS, refreshed with mTESR1 medium (for iPSCs) or neuron growth medium (for neurons), and collected at different time points for analysis. For the EPZ6438 titration experiments of Supplementary Figures 2, 8, 9, FXS iPSCs or neurons were treated with varying doses of EPZ6438 (0.001, 0.01, 0.1, 1, or 10 μM) for 96 h. For the cell density dependency experiment of Supplementary Figure 16, FXS iPSCs were seeded at varying densities, and 24 h later treated with 1 μM EPZ6438 for 72 h. For Ara-C treatment, neurons were incubated with 2 μM cytosine arabinoside (Sigma), or water as a control, for 48 h prior to shRNA knockdown or small molecule treatment.

Total RNA was isolated using TRIZOL (Invitrogen) at day 20 or other time points following shRNA lentivirus transduction, puromycin selection or small molecules treatment, or ASO transfection. Reverse transcription was performed using SuperScript II Reverse Transcriptase (Invitrogen) as per the manufacturer’s instructions, followed by quantitative real-time PCR using Platinum SYBR Green qPCR SuperMix-UDG with Rox (Invitrogen). Gene-specific primers are listed in Supplementary Table 2. The Taqman assay was performed using Taqman Fast Advanced Master Mix (Thermo Fisher Scientific) and probes for FMR1 (Hs00924547_m1) and ACTB (Hs99999903_m1), all from Thermo Fisher, according to the manufacturer’s instructions. The Taqman assay for FMR1 was only used in Supplementary Figure 3, for other figures mRNA levels were determined with SYBR Green and primer in Supplementary Table 2 and with hRPL41 as endogenous control. The experiments were performed in biological triplicates, each with three technical triplicates.

Protein extracts were prepared by lysis in a buffer containing 50 mM Tris–HCl (pH 7.4), 0.1% Triton X-100, 5 mM EDTA, 250 mM NaCl, 50 mM NaF, 0.1 mM Na3VO4, and protease inhibitors (Roche). Blots were probed with an antibody recognizing FMRP (Abcepta, AP6879A), DGKK (Abcam, ab111042), α-tubulin (Sigma, F2168), EZH2 (Fortislife, A304-196A), H3K27me3 (Cell Signaling, 9733), KDM5C (Fortislife, A301-034A), or KDM5D (Fortislife, A301-751A). The FMRP and DGKK signals were quantified using Image J software (NIH), and normalized to α- tubulin levels as previously described (Sheng et al., 2011).

Bisulfite modification was carried out using an EpiTect Bisulfite Kit (Qiagen) followed by PCR amplification as previously described (de Esch et al., 2014). Twenty independent clones were initially sequenced from the PCR product within each cell line, of which six representative clones are displayed in Figure 1D. For quantification, clones with strong sequencing signals (12–16 clones) were analyzed and the percent methylation at each CpG was calculated.

ChIP assays were performed as previously described (Gazin et al., 2007) using the following antibodies: DNMT1 (Novus Biologics, AF6110), EZH2 (Fortislife, A304-196A), SUV39H1 (Fortislife, A302-127A), RNF2 (Fortislife, A302-869A), KDM5C (Fortislife, A301-034A) or KDM5D (Fortislife, A301-751A), HDAC5 (Fortislife, 303-464A), and HDAC10 (Sigma, H3413). Briefly twenty millions iPSC cells were harvested and crosslinked with 1% formaldehyde. Extraction and sonication of nuclei were conducted as in Gazin et al. (2007). In the next step, 10 μL of antibody was added to each sample to immunoprecipitate protein DNA complexes. Following reverse-crosslinking and DNA purification, qPCR was performed to quantitate the ChIP products (see Supplementary Table 2 for primers). Samples were quantified as percentage of input, and then normalized to an irrelevant region in the genome (3.2 kb upstream from the transcription start site of GCLC) as described (Serra et al., 2014). Fold enrichment was calculated by setting the IgG control IP sample to a value of 1.

iPSCs, and neurons were fixed with 4% paraformaldehyde in PBS for 10 min, blocked with 10% normal goat serum (Vector Laboratories), and then stained with TUJ1 (Covance), FMRP (Abgent), MAP2 (Cell Signaling), NeuN (Biolegend), GFAP (Abgent), or phosphorylated histone H3 (Abcam) antibodies for 1 h at room temperature. Cells were then rinsed several times with PBS, incubated with Alexa 488- or Alexa 594-conjugated donkey anti-mouse or anti-rabbit secondary antibody (Molecular Probes) and DAPI (Molecular Probes) in the appropriate buffer for 1 h at room temperature. After several more rinses, cells were mounted with Vectashield (Vector Laboratories) and imaged with a Zeiss Imager Z2 microscope equipped with a Zeiss Axiocam digital camera. Images were acquired, and background signal was subtracted, using AxioVision Rel 4.8 software. FMRP or phosphorylated H3 positively stained cells and DAPI- stained nuclei were counted in an automated, blinded fashion using ImageJ. Briefly, the images were converted to grayscale, the lower threshold was set at 20 for FMRP, 82 for phosphorylated H3, and 24 for DAPI, and the Analyze Particle feature was used to obtain the total cell counts. The percentage of cells with positive staining were determined by counting at least 300 cells/nuclei per sample in three separate experiments (for a total of 900 cells/nuclei).

Neurons (at day 4 of the differentiation process) were seeded at 1x10⁵ cells/well in 6-well plates at 40% confluency and transduced overnight with lentivirus (200 μl at MOI > 5) expressing an NS or FMR1-SF shRNA. Cells were selected with 0.5 μg/ml puromycin for 3 days, and then transduced overnight with a lentivirus (400 μl at MOI 5) expressing an NS or FMR1 shRNA (TRCN0000059762), or expressing empty vector (pLIX_403, Addgene Plasmid #41395) or REST (pLIX-REST, Addgene Plasmid #91896). Cells were harvested for analysis by qRT-PCR or immunoblot at day 16 post-differentiation at 50% confluency. For small molecule FMR1-SF inhibitor treatment, neurons (at day 4 of the differentiation process) were first transduced overnight with a lentivirus (200 μl at MOI > 5) expressing an NS or FMR1 shRNA, or expressing empty vector or REST, selected with 0.5 μg/ml puromycin for 2 days, and then incubated with a small molecule inhibitor (1 μM 5-aza-2′-deoxycytidine, 0.5 μM chaetocin, 5 μM EPZ6438, 5 μM GSK126 or 5 μM PRT4165) for 96 h. Cells were harvested for analysis by qRT-PCR or immunoblot at day 11 post-differentiation at 50% confluency.

For shRNA treatment, neurons (at day 4 of the differentiation process) were seeded at 1x10⁵ cells/well in 6-well plates at 40% confluency, transduced overnight with 200 μl lentivirus (at MOI 5) expressing an NS or EZH2 shRNA, and 2 days later selected with 0.5 μg/ml puromycin for 3 days. Following shRNA treatment, 8x10⁴ neurons were overlaid at 90–100% confluency in 24- well MEA plates that had been coated with polyethyleneimine and laminin and seeded 3 days prior with 4x10⁴ rat astrocytes (Thermo Fisher). Thereafter, 80% of the medium was refreshed every 8 days. For small molecule inhibitor treatment, neurons were overlaid on MEA plates and incubated with DMSO or 5 μM EPZ6438 for 4 days. Thereafter, 80% of the medium was refreshed at day 5 without inhibitor, at day 10 with a lower inhibitor concentration (1 μM EPZ6438), and then subsequently every 8 days alternatingly with or without inhibitor. For ASO treatment, neurons were incubated with 1 μM control or EZH2 ASO (see below), and 80% of the medium (with ASO) was refreshed every 8 days.

Starting on the second day of co-culture (at day 8 of the differentiation process), MEA assays were performed every 4 days using a MED64 Presto MEA system (Alpha MED Scientific Inc.) according to the manufacturer’s instructions. Recordings of spontaneous activities were performed using a 3-kHz two-pole Butterworth low pass filter. Experiments were performed in biological triplicate.

Conserved regions between the human and mouse EZH2 genes were identified by bioinformatic analysis, and a series of 10 locked nucleic acid (LNA) ASOs targeting these conserved regions were designed using the LNCASO web server.² The LNA ASOs were tested for efficacy in 293T cells by monitoring expression of EZH2 by qRT-PCR. Derivatives of the most efficacious EZH2 ASO were re-synthesized with 2′-O-methoxyethyl-RNA (MOE) modification and have the following sequences: EZH2 ASO1, &G*&T&C&T& A*C*A*T*G*T*T*T*T*&G&G&T&C*&C, and EZH2 ASO2, &T* &G&T&C&T*A*C*A*T*G*T*T*T*T*&G&G&T&C*&C (where “&” represents MOE modification, and “*” represents phosphorothioate linkage). A non-targeting control ASO was also synthesized with the sequence &C*&C*&T*&A*&T*A*G*G* A*C*T*A*T*C*C*&A*&G*&G*&A*&A. ASOs were synthesized as MOE gapmers (i.e., 5 MOE nucleotides, 8–10 DNA nucleotides, 5 MOE nucleotides) using standard phosphoramidite methods on a Dr. Oligo 48 synthesizer (Biolytic). Phosphoramidites and other standard reagents were purchased from ChemGenes. Coupling times for MOE nucleotides were extended to 2 min. Oligonucleotides were cleaved and deprotected in concentrated aqueous ammonia at 55°C for 16 h. ASOs were characterized by LC-MS analysis using an Agilent Q-TOF LC-MS instrument and were desalted using Amicon ultrafiltration columns (3-kDa cutoff). For ASO treatment, neurons were incubated with 1 μM control or EZH2 ASO for 3 weeks, and 80% of the medium (with ASO) was refreshed every 7 days.

All mouse studies were performed in accordance with the Guide for the Care and Use of Laboratory Animals from NIH, and protocols (A2060) approved by the UMMS Institutional Animal Care and Use Committee (IACUC). Mice were randomly allocated to each group. No blinding was done as animal groups were identified by tagging and labeling the cages with the cells/ASOs injected.

C57BL/6J mice (aged 6–9 months, n = 4 per group) were anesthetized and placed in a rodent stereotaxic frame (Stoelting), and intracerebroventricular (ICV) injections were performed as previously described (Vershkov et al., 2019). Briefly, the skull was exposed by a small longitudinal incision (<1 cm) along the midline, the periosteum was removed from the surgical area, and small burr holes (<1 mm) were made in each hemisphere using a high-speed drill at the stereotaxic coordinates for ICV injection (X: ±1 mm; Y: −.4 mm; Z: −1.6 mm). NPCs (5x10⁴ in 5 μl volume) were injected slowly into each site using an UltramicroPump (World Precision Instruments) to drive a Hamilton Syringe attached to a 31-gauge steel needle (Hamilton). Staples were used to close the incision and mice were allowed to recover. Three days after NPC transplantation, the initial incision was reopened and 30 pM control or EZH2 ASO was injected into the same site. Seven days after the treatment, mice were sacrificed and the subventricular region was manually dissected for RNA and immunohistochemistry analysis. For RNA analysis, the tissue was mechanically disrupted, and RNA was extracted using a NucleoSpin RNA Plus Kit (Macherey Nagel) and analyzed for FMR1 expression using primers specific for the human FMR1 gene (Vershkov et al., 2019). For immunohistochemical staining, brain tissues were fixed in paraformaldehyde, embedded in paraffin and sectioned at a thickness of 10 μm. The immunolabeling was performed by the Morphology Core at the University of Massachusetts Chan Medical School. Briefly, following deparaffinization (60°C for 30 min, xylene for 10 min, 100% ethanol 20 dips, 95% ethanol 10 dips, 75% ethanol 10 dips, H₂O 1 min), brain sections were processed for antigen retrieval with citrate acid (pH 6.0), blocked with 5% milk in TBS (wt/vol, pH 7.4) at 23–25°C and incubated at 4°C overnight in TBS containing 5% milk with a mouse antibody to human FMRP (AbFrontier, YF-MA10356, 1:500) or a rabbit monoclonal antibody to human mitochondria (Millipore, 1:200). The next day, sections were washed in PBS (pH 7.4) three times, 5 min each, and incubated in secondary antibodies (1: 400) for 1 h at 23–25°C. Sections were also stained with DAPI (1:10,000; Molecular Probes). All images were obtained using a Zeiss AXIO Imager Z2 microscope.

All quantitative data were collected from experiments performed at least three independent times. The results of three technical or biological replicates are shown and expressed as mean ± SD. Differences between groups were assayed using One-way ANOVA with a Dunnett test using Microsoft Excel. Significant differences were considered when P < 0.05. Data normalization, if applicable, is described in the relevant figure legend.