Human autopsy tissue and peripheral DNA were obtained from the University of Pennsylvania Center for Neurodegenerative Disease Research biorepository [62]. A summary of cases is available in supplemental materials. Lymphoblast cell lines ND16183, ND11836, ND14442 and ND10966 were obtained from the Coriell NINDS Repository (Camden, NJ, USA).

Total RNA was extracted using Trizol (Life Technologies, Carlsbad, CA, USA). RNA was digested with 1 unit of RQ1 DNase (Promega, Madison, WI, USA) per μg followed by ethanol precipitation. RNA was quantified using the Qubit RNA HS Assay kit (Life Technologies). RNA was reverse transcribed to cDNA using the High Capacity RNA to cDNA kit (Life Technologies). qPCR was done with 2× FastStart SYBR Green Master (Roche Applied Science, Indianapolis, IN, USA) to quantify C9orf72 RNA and control housekeeping RNAs using primers as listed in the supplemental materials on the StepOne Plus Real-Time PCR Machine (Life Technologies) using the ΔΔ C_t method. For qPCR measuring C9orf72 intronic RNA, reactions containing RNA without reverse transcriptase did not amplify. DNA from tissue or cells was extracted using the DNeasy Blood and Tissue kit (Qiagen, Valencia, CA, USA).

For quantitative assessment of methylation levels, 100 ng of DNA was digested for 16 h with 2 units of HhaI (New England Biolabs, Ipswich, MA, USA) followed by heat inactivation. qPCR was done with 2× FastStart SYBR Green Master (Roche) using primers amplifying the differentially methylated C9orf72 promoter region (see supplemental materials for primer sequences). The difference in the number of cycles to threshold amplification between digested versus mock-digested DNA was used as a measure of CpG methylation. Mock-digested DNA consisted of either undigested DNA or HaeIII-digested DNA as a negative control restriction enzyme which does not cut within the qPCR amplicon. To determine the linearity of this assay, a methylated DNA standard was generated by in vitro methylating DNA from a non-expanded lymphoblast cell line using M.SssI (New England Biolabs) for 4 h at 37 °C. Methylated DNA was purified by phenol:chloroform:isoamyl alcohol extraction, and different ratios of methylated to unmethylated DNA were subject to the C9orf72 promoter methylation assay.

To determine if methylation occurs in cis or trans, 100 ng of DNA from three C9orf72 promoter hypermethylated repeat-expanded patient cases who were heterozygous for the deletion polymorphism (rs200034037) was digested for 16 h with 2 units of HhaI and HpaII (New England Biolabs) vs. a no enzyme mock digestion followed by heat inactivation. DNA was amplified by PCR using primers flanking rs200034037 and the HhaI and HpaII cut sites within the C9orf72 promoter (see supplemental materials for primer sequences). The PCR product was run on a poly-acrylamide gel and imaged with ethidium bromide or used for Sanger sequencing.

Cerebellar DNA from four C9orf72 expansion carriers and four control cases was bisulfite converted using the EpiTect Bisulfite Kit (Qiagen). Bisulfite-converted DNA was subjected to PCR to amplify both CpG islands. The entire first CpG island was amplified by standard PCR. The second CpG island was divided into two separate reactions. The first half of the 2nd CpG island was amplified by nested PCR, while the second half of the second CpG island was amplified using standard PCR. Primer sequences are available in the supplemental materials. PCR products were run on an agarose gel and gel purified using QIAquick Gel Extraction kit (Qiagen). Amplified DNA was sub-cloned by ligation into the pGEM-T Easy vector (Promega), transformation into competent E. coli and isolation of plasmid DNA from individual bacterial colonies. Five or six individual clones from each case were Sanger sequenced using a T7 promoter sequencing primer.

DNA (1 μg) from post-mortem cerebellar tissue was restriction enzyme digested for 4 h with HpaII (10 U), MspI (10 U), or MspJI (4 U) (New England Biolabs) at 37°C followed by phenol:chloroform:isoamyl alcohol extraction. 100 ng of digested DNA was used for repeat primed PCR as described previously [53]. Fragment length analysis was done using the Genetic Analyzer 3130x (Life Technologies) and Peak Scanner software (Life Technologies). To ensure that methylated DNA could be digested by MspJI under these conditions, 1 μg of DNA from repeat expanded or control cerebellum was in vitro methylated using M.SssI (New England Biolabs) for 4 h at 37 °C followed by phenol:chloroform:isoamyl alcohol extraction. 1 μg of purified DNA was digested with 2 units of MspJI at 37 °C for 4 h and purified using phenol:chloroform:isoamyl alcohol prior to repeat primed PCR.

Southern blot hybridization was performed as previously described [16]. Briefly, 5–10 μg of genomic DNA was digested with EcoRI and HindIII, denatured at 95 °C for 5 min, and run on a 0.8 % agarose gel at 100 V for 4 h. DNA was transferred to a positively charged nylon membrane (GE Life Sciences, Pittsburg, PA, USA) and crosslinked. A 576-bp digoxigenin (DIG)-labeled probe was amplified using PCR DIG Probe Synthesis Kit. Primers are listed in supplemental materials. The blot was hybridized for 16 h at 49 °C, washed in 2× SSC with 0.1 % SDS at room temperature, and then in 0.1× SSC, 0.1 % SDS for 45 min at 71 °C twice. Anti-digoxigenin antibody (1:10,000, Roche) was used to detect the probe, which was visualized with CSPD (Roche). Blots were visualized with an LAS-3000 Luminescent Image Analyzer (Fujifilm).

Lymphoblast cell lines derived from either a repeat expansion carrier (ND14442) or a non-expanded ALS patient (ND16183) were grown in RPMI 1640 with 15 % fetal bovine serum and 2 mM L-glutamine in 37 °C with 5 % CO₂. Cells were seeded at a concentration of 300,000 cells/ml in 10 ml in a T25 flask, and 0.5 μM 5-aza-dC (Sigma-Aldrich, St. Louis, MO, USA) was spiked daily into lymphoblast cells for 3 days followed by a 3-day recovery period without 5-aza-dC. RNA and DNA were extracted as described above. To assess the toxicity, untreated or 5-aza-dC treated cells were seeded at 300,000 cells/ml and treated 10 μM sodium arsenite or 100 μM chloroquine for 24 h in RPMI 1640 with 5 % FBS and 2 mM L-glutamine. After 24 h, cells and media were collected and measured in triplicate using a lactate dehydrogenase assay (Clontech, Mountain View, CA, USA) to calculate the percent LDH release into the media relative to total LDH from untreated cell pellets.

Lymphoblast cell lines were fixed in 4 % paraformalde-hyde (PFA) in RNase-free PBS for 10 min on ice followed by permeabilization with 0.2 % Triton X-100 in PBS for 10 min on ice. Cells were prehybridized with 40 % formamide (Fisher Scientific)/1× SSC at 37 °C for 10 min. A locked nucleic acid (LNA) probe complementary to the hexanucleotide repeat (FAM-CCCCGGCCCCGGCCCC, batch #613510, Exiqon, Woburn, MA, USA) was denatured at 85 °C for 75 s prior to incubation with cells in hybridization buffer [40 % formamide, 1× SSC, 50 mM sodium phosphate, pH = 7, 10 % RNase-free dextran sulfate (Sigma-Aldrich) with 40 nM LNA probe] at 66 °C for 2 h. Cells were washed once with 0.1 % Tween-20 in 2× SSC at room temperature for 5 min, and three times in 0.1× SSC at 65 °C for 10 min each. Cells were then resuspended in Prolong Gold Antifade Reagent with DAPI (Life Technologies), coverslipped onto glass slides, and imaged on a Leica SPE confocal microscope (Leica Microsystems, Buffalo Grove, IL, USA). RNA foci numbers were scored in 200 cells for each cell line blinded to cell type using a 60× objective.

Frozen cerebellar cortex (100–200 mg) was dounce homogenized in 2 ml of 0.25 M sucrose with TKM (50 mM Tris–HCl, pH 7.5, 25 mM potassium chloride and 5 mM magnesium chloride). Two volumes of 2.3 M sucrose with TKM were mixed with the lysate. Nuclei were pelleted at 10,000×g for 10 min at 4 °C. Nuclei were fixed in 2 % PFA in RNase-free PBS for 10 min at room temperature and quenched with 0.83 M glycine (pH = 7.6). Nuclei were pelleted and resuspended in RNase-free PBS. Nuclei were pre-hybridized with hybridization buffer [40 % formamide, 10 mM ribonucleoside vanadyl complex (New England Biolabs), 1× SSC, 50 mM sodium phosphate, pH = 7, 10 % RNase-free dextran sulfate] for 10 min at 37 °C. The LNA probe complementary to the hexanucleotide repeat was denatured at 100 °C for 10 min in 95 % formamide and then hybridized to nuclei for 12–16 h at 37 °C at 40 nM. Nuclei were washed and imaged as described above for LCLs. RNA foci numbers were scored in at least 100 nuclei for each case using a 60× objective.

Polyclonal anti-(GA)₁₅ and polyclonal anti-(GP)₁₅ antibodies were generated by immunizing rabbits with a purified fusion protein containing his-tagged maltose-binding protein (MBP) with (GA)₁₅ or (GP)₁₅ at the C-terminus. The GA or GP repeats’ fragment was engineered using the primers as described in the supplemental materials (Integrated DNA Technologies). The annealed fragment was ligated into the pDB.His.MBP vector (DNASU Plasmid Repository, AZ, USA) at the NdeI–EcoRI cloning sites. The his-tagged MBP-GA₁₅ and MBP-GP₁₅ fusion proteins were expressed in BL21 bacterial cells upon IPTG induction and purified using Ni–NTA Superflow (Qiagen, CA, USA). (GR)₁₅ antibody was generated by immunizing rabbits with an aminohexanoic acid linked synthetic peptide (C-Ahx-(GR15)) conjugated to KLH and affinity purified using SulfoLinkR-immobilized immunogen peptide (Thermo Scientific). Cerebellar sections were stained with RANT antibodies using standard ABC methods with microwave antigen retrieval [34], and at least 300 granular neurons per case were assessed using a 100× objective blinded to methylation status to determine the percentage of neurons with RANT inclusions.

Frozen cerebellar cortex was homogenized in a series of extraction buffers intermixed with pelleting of insoluble material by ultracentrifugation at 135,000×g for 30 min. The extraction buffers included RIPA buffer (5 ml buffer per gram tissue; 50 mM Tris pH 7.3, 150 mM NaCl, 0.1 % SDS, 0.5 % sodium deoxycholate, 1 % NP-40 with protease inhibitors), myelin floatation buffer (5 ml buffer per gram tissue; 10 mM Tris pH 7.5, 500 mM NaCl, 2 mM EDTA, 1 mM DTT, 30 % sucrose with protease inhibitors), DNAse buffer containing (Promega, 1 unit of RQ DNAse per 10 mg tissue in 1× reaction buffer, incubated at 37° for 30 min) and sarkosyl buffer (5 ml buffer per gram tissue with sonication; 1 % sarkosyl, 10 mM Tris pH 7.5, 500 mM NaCl, 2 mM EDTA, 1 mM DTT, 10 % sucrose with protease inhibitors). The remaining insoluble material was pelleted and sonicated in SDS buffer (1 ml per gram tissue; 2 % SDS, 50 mM Tris, pH 7.6). Five μl of SDS lysates was dot blotted onto nitrocellulose using a Biorad Bio-Dot apparatus and washed with 300 μl of SDS buffer containing β-mercaptoethanol. Dot blots were blocked with 5 % milk, blotted with RANT-specific antibodies and visualized by film using enhanced chemiluminescence.

Two-way ANOVA, one-way ANOVA, t tests, linear regression and Fisher’s exact test were used as described using GraphPad Prism software (GraphPad, San Diego, CA, USA). Post-hoc analyses were Bonferroni corrected. All statistical tests were two-sided.

Methylation of repeat expansions can reduce gene transcription as exemplified by Fragile X syndrome [49]. To determine whether the C9orf72 hexanucleotide repeat expansion is methylated, cerebellar genomic DNA from eight repeat expansion carriers and eight non-expanded controls was digested using three restriction enzymes: (1) MspI that cuts the hexanucleotide repeat irrespective of methylation status, (2) HpaII that cuts the hexanucleotide repeat only if it is not methylated, and (3) MspJI that cuts the hexanucleotide repeat only if it contains methylated CpGs. Digested DNA was then amplified using repeat primed PCR as a qualitative measure of whether the repeat region is methylated. Repeat-primed PCR allows for partial amplification of the expanded hexanucleotide repeat resulting in a characteristic tapering sawtooth pattern, and is used because conventional PCR is unable to amplify repeat expansions due to their size and high GC content [16, 53]. Non-expanded controls demonstrated the pattern expected of unmethylated DNA in which DNA was amplified after mock and MspJI digestion, and was not amplified after HpaII and MspI digestion (Fig. 1a). The same pattern was observed for repeat expansion carriers with amplification of the repeat expansion after MspJI digestion but no amplification after HpaII digestion, indicating that the repeat expansion is not methylated (Fig. 1a). To confirm that MspJI is able to cut the hexanucleotide repeat if it is methylated, DNA from a repeat expansion carrier was in vitro methylated using CpG methyltransferase (M. SssI) and digested with MspJI. Repeat primed PCR analysis showed poor amplification of in vitro methylated DNA relative to unmethylated DNA (Supplemental Figure 1), demonstrating the ability of MspJI to cut methylated GGGGCC repeats.

We then extended our methylation analysis to include the two CpG islands adjacent to the repeat expansion (open boxes in Fig. 1b). A bisulfite cloning screen was performed on cerebellar DNA from four C9orf72 expansion carriers and four non-expanded controls to identify methylated CpG dinucleotides. PCR was used to amplify three regions covering both CpG islands from bisulfite-converted DNA (amplicons A–C, Fig. 1b). Amplified DNA was cloned, and individual clones (≥5 per case) were sequenced. Increased methylation (10–25 % of clones) was observed in the vicinity of the 5′ CpG island upstream of the hexanucleotide repeat region (amplicon A, Fig. 1b). The second CpG island downstream of the repeat expansion was not methylated in both repeat expanded and control cases (amplicons B and C, Fig. 1b). The sequencing results for all clones are provided in Supplemental Figure 1. These results confirm recently published studies demonstrating promoter hypermethylation in some repeat expansion mutation cases [65].

Since methylation was observed only in C9orf72 mutation cases, we hypothesized that promoter methylation occurs in cis to the repeat expansion. To determine whether methylation was monoallelic vs. biallelic, we identified three hypermethylated C9orf72 mutation cases heterozygous for the rs200034037 polymorphism, a dinucleotide deletion with an allele frequency of ~25.6 %. This polymorphism lies upstream of HhaI and HpaII restriction enzyme cut sites within the differentially methylated region (DMR). Both HhaI and HpaII only cut unmethylated DNA. The DMR was amplified after HhaI/HpaII double digestion and subject to PAGE electrophoresis (Fig. 1c). In all three cases, only the major rs200034037 allele was amplified after digestion of unmethylated DNA, in contrast with mock-digested DNA in which both major and minor alleles were observed. These results were confirmed by Sanger sequencing which showed biallelic sequences after the rs200034037 polymorphism in mock-digested DNA. This is in contrast with HhaI/HpaII-digested DNA which showed only the major allele sequence (color traces, bottom of Fig. 1c). Thus, C9orf72 promoter methylation is monoallelic, consistent with promoter hypermethylation in cis relative to the repeat expansion.

To confirm the presence of promoter methylation in a larger cohort of subjects, we developed an alternative method based on the HhaI restriction enzyme recognition site within the DMR (Fig. 2a). HhaI digestion of unmethylated DNA was coupled with quantitative PCR (qPCR) which demonstrated a shift in DNA amplification curves relative to mock-digested DNA (Fig. 2b). This shift was used to calculate the percent DNA that is resistant to HhaI digestion as a measure of methylation status. This method proved to be robust, as standard curves using in vitro methylated DNA showed high linearity from 0 to 100 % methylation (Fig. 2c).

This assay was used to confirm the presence of C9orf72 promoter methylation using DNA from the frontal cortex and cerebellum of repeat-expanded and non-expanded cases. Both the frontal cortex and cerebellum of C9orf72 mutation carriers exhibited higher methylation than controls (p = 0.0009, Fig. 2d). Despite this highly significant result, C9orf72 promoter methylation for many C9orf72 mutation carriers was low and within the normal range, suggesting that C9orf72 promoter methylation is not driving disease pathogenesis.

We analyzed ENCODE CAGE-seq data to determine whether there are differences in C9orf72 transcription across diverse cell lineages [54]. C9orf72 expression is highly variable across different cell lines, with both hematopoietic and neuronal cell lines expressing similar levels of C9orf72 in contrast with epithelial, fibroblastic or other embryonic or mesenchymal cell types (Fig. 3a). CAGE-seq data also indicated that of the three annotated C9orf72 mRNAs, variant 2 (V2) represented the major RNA transcript, representing 92.6 ± 1.2 % of all C9orf72 transcripts (Fig. 3b). Based on these analyses, we turned to lymphoblast cell lines (LCLs) to demonstrate that methylation is associated with reduced C9orf72 expression. LCLs from non-expanded (ND16183) versus expanded ALS cases (ND11836, ND10966 and ND14442) were analyzed by Southern blotting using a probe specific to C9orf72 which confirmed the presence of large repeat expansions in the three mutant LCLs (Fig. 3c). Methylation was very low (<1 %) in non-expanded ND16183 cells and was high in two expanded cell lines (ND10966 and ND14442, Fig. 3d). Despite harboring the hexanucleotide repeat expansion, ND11836 cells were not methylated, akin to our results above which showed many carriers exhibit minimal C9orf72 promoter methylation.

C9orf72 mRNA was measured using RT-qPCR which demonstrated that the unmethylated but repeat-expanded cell line, ND11836, showed essentially normal total C9orf72 mRNA levels (Fig. 3e). Despite expressing normal levels of total C9orf72 mRNA, ND11836 cells demonstrated increased usage of the alternative upstream TSS, as shown by the increase in V3 C9orf72 mRNA relative to control cells (Fig. 3g). Since transcription from the upstream TSS results in transcription through intron 1 which contains the hexanucleotide repeat, we speculated that the repeat expansion mutation is associated with accumulation of intron 1-containing RNA. Indeed, using primers 5′ to the hexanucleotide repeat within intron 1, unmethylated expanded cells exhibited over fourfold more intronic C9orf72 RNA relative to control cells (Fig. 3h).

In contrast, expanded cells with C9orf72 promoter hypermethylation exhibited reduced total (Fig. 3e) and V2 (Fig. 3f) mRNA relative to control cells. Promoter methylation was also associated with reduced expression from the alternative TSS, as determined by reduced V3 mRNA expression relative to unmethylated expanded cells (Fig. 3g). The reduction of V3 mRNA expression was coupled to reduced intronic RNA accumulation in methylated expanded cells relative to unmethylated expanded cells (Fig. 3h).

To verify that the C9orf72 mutation causes a shift in TSS usage and to confirm that methylation inhibits C9orf72 RNA levels, the effect of DNA demethylation was tested in hypermethylated expanded LCLs using the DNA methyltransferase inhibitor 5-aza-2′-deoxycytidine (5-aza-dC). Repeat-expanded lymphoblast cells (ND14442) and non-expanded control cells (ND16183) were treated with 5-aza-dC for 3 days followed by a 3-day recovery without 5-aza-dC. 5-aza-dC treatment resulted in a 41.8 % reduction in C9orf72 promoter methylation (p < 0.0001, Fig. 4a). 5-aza-dC treatment increased total (p < 0.05, Fig. 4b) and V3 (p < 0.001, Fig. 4d) C9orf72 mRNA expression in expanded but not control cells, demonstrating that C9orf72 promoter methylation is linked to transcriptional silencing of mutant C9orf72. However, V2 mRNA was not significantly changed in expanded cells upon demethylation (Fig. 4c). Thus, promoter demethylation reverses transcriptional silencing of C9orf72 but not of the major V2 transcript. Rather, demethylation rescued only total and V3 mRNA indicating that the expanded allele is preferentially transcribed from the upstream TSS. This alternative TSS usage translated into a marked increase in intronic C9orf72 accumulation (p < 0.01, Fig. 4e). Thus, methylation inhibits the accumulation of potentially toxic intronic RNA by inhibiting transcription through the repeat expansion.

The accumulation of repeat-containing intronic RNA has been postulated to promote neurodegeneration. Since promoter hypermethylation inhibited intronic RNA levels, we hypothesized that hypermethylation may protect cells from cellular insults, such as oxidative stressors which lead to stress granule formation or altered proteostasis. Both processes have been implicated in ALS and FTD [36, 37]. To test this hypothesis, expanded and control cells were dem-ethylated with 5-aza-dC and exposed to sodium arsenite or chloroquine, known to induce stress granules or inhibit autophagy, respectively. Toxicity was assessed by measuring lactate dehydrogenase (LDH) release into media. In both control and expanded cell lines, 5-aza-dC resulted in mild toxicity compared to untreated cells consistent with the known toxicity profile of 5-aza-dC (control p < 0.01, mutant p < 0.05, Fig. 4f, g). Importantly, demethylation of control non-expanded cells with 5-aza-dC had no effect on arsenite or chloroquine toxicity, indicating that global hypo-methylation does not result in an altered response to these cellular stressors (Fig. 4f). In contrast, hypomethylation of mutant cells with 5-aza-dC resulted in enhanced sensitivity to both arsenite and chloroquine (arsenite p < 0.05, chloroquine p < 0.001, Fig. 4g). Thus, reactivation of mutant C9orf72 transcription which results in the accumulation of mutant intronic C9orf72 RNA appears to be associated with increased vulnerability to cellular stressors.

Having demonstrated that C9orf72 promoter hypermethylation was associated with lower levels of intronic RNA using RT-qPCR, we hypothesized that hypermethylation is associated with reduced accumulation of RNA foci. LCLs were studied using in situ hybridization with a fluorescently labeled locked nucleic acid (LNA) probe that recognizes the GGGGCC hexanucleotide repeat. RNA foci were observed in expanded cell lines but not in non-expanded cells (Fig. 5a). LCLs were scored for the presence of RNA foci which demonstrated that 12 % of ND11836 cells (expanded and unmethylated) exhibited RNA foci, with several cells exhibiting multiple RNA foci (Fig. 5c). In contrast, expanded and methylated cell lines, ND10966 and ND14442, exhibited significantly fewer RNA foci compared to ND11836 (1.5 % for ND10966, p < 0.0001; 3 % for ND14442, p = 0.00094, Fisher’s exact test, Fig. 5c).

Human brain tissue was also used to test whether C9orf72 hypermethylation is associated with reduced accumulation of RNA foci. Intact nuclei were isolated from post-mortem cerebellum and stained for RNA foci. RNA foci were only observed in repeat-expanded cases and not in non-expanded control cases (Fig. 5b). The percentage of nuclei containing RNA foci was quantified from 12 repeat-expanded cases which revealed an inverse relationship between C9orf72 methylation and the percentage of nuclei with RNA foci (p = 0.0148, Fig. 5d).

RANT aggregates accumulate as a consequence of non-ATG translation of mutant C9orf72 RNA which may contribute to disease pathogenesis [2, 46]. Given that hypermethylation reduced mutant C9orf72 expression, it seemed likely that hypermethylation would be associated with reduced RANT pathology. To understand the relationship between methylation and RANT pathology, cerebellar tissue sections from hypomethylated (<12 % HhaI resistance) and hypermethylated (>12 % HhaI resistance) repeat expansion carriers were stained with antibodies specific for GA (glycine–alanine), GP (glycine–proline), or GR (glycine–arginine) repeats corresponding to the three different sense reading frames of the repeat expansion. The 12 % threshold corresponds to three standard deviations above the mean HhaI resistance values for control, non-expanded cases. Immunohistochemistry showed abundant numbers of inclusions in hypomethylated C9orf72 expansion cases and markedly fewer RANT inclusions in hypermethylated cases (Fig. 6a). Quantification of the percent of cerebellar granule neurons with RANT inclusions demonstrated a significant reduction in RANT pathology burden associated with C9orf72 promoter hypermethylation (Fig. 6b). To confirm this finding, insoluble proteins were extracted from hypermethylated and hypomethylated cerebellum and assayed by dot blot analysis which again demonstrated that hypermethylation was associated with reduced accumulation of GA, GP, and GR proteins (Fig. 6c). These results show a tight relationship between C9orf72 promoter methylation and RANT pathology, and indicate that endogenous transcriptional silencing reduces the downstream pathologic features associated with the hexanucleotide repeat expansion.