Extra-coding RNAs regulate neuronal DNA methylation dynamics

Abstract

Epigenetic mechanisms such as DNA methylation are essential regulators of the function and information storage capacity of neurons. DNA methylation is highly dynamic in the developing and adult brain, and is actively regulated by neuronal activity and behavioural experiences. However, it is presently unclear how methylation status at individual genes is targeted for modification. Here, we report that extra-coding RNAs (ecRNAs) interact with DNA methyltransferases and regulate neuronal DNA methylation. Expression of ecRNA species is associated with gene promoter hypomethylation, is altered by neuronal activity, and is overrepresented at genes involved in neuronal function. Knockdown of the Fos ecRNA locus results in gene hypermethylation and mRNA silencing, and hippocampal expression of Fos ecRNA is required for long-term fear memory formation in rats. These results suggest that ecRNAs are fundamental regulators of DNA methylation patterns in neuronal systems, and reveal a promising avenue for therapeutic targeting in neuropsychiatric disease states.

Figure 1. Genome-wide identification and quantification of ecRNAs from neuronal systems.

(a) RNA-seq workflow identifies both polyadenylated and non-polyadenylated transcripts from the same neuronal tissue. (b) Comparison of PolyA+ and PolyA− sequencing from representative gene loci reveals PolyA− transcripts arising from intronic and post-TESs. (c) Genome wide, extra-coding transcripts were characterized by averaging PolyA− reads that mapped to 5′ (pre-TSS), intronic or 3′ (post-TES) of a given gene. (d) Rank plot of ecRNA index at 17,719 rat genes. (e) mRNA expression (PolyA+ RNA-seq) ranked by ecRNA index reveals correlation between ecRNA and mRNA expression. (f) Division of ecRNA into discrete quartiles reveals general profile and expression of PolyA− RNA transcripts. Data are aligned to transcription start sites (TSS) and TESs. Heatmap shows PolyA− transcription from all genes. (g) MBD-seq reveals metagenomic DNA methylation profiles, including hypomethylation at TSS and hypermethylation at TES. ecRNA transcription is associated with hypomethylated promoters across the genome. (h,i) Genome wide, ecRNA levels are positively correlated with mRNA transcription ((h) one-way ANOVA, F_(3,17715)=612.5, P<0.0001) and negatively correlated with promoter DNA methylation ((i) one-way ANOVA, F_(3,17715)=73.27, P<0.0001). (j) Percentile-percentile density scatterplot of ecRNA and promoter DNA methylation. Data in h and i are presented as mean±s.e.m.

Figure 2. Regulation of mRNA and ecRNA by neuronal activity.

(a) PolyA+ RNA-seq following 1 h neuronal depolarization (25 mM KCl) or inactivation (1 μM TTX) reveals altered mRNA expression at a small subset of genes. Top, heatmap of KCl-altered transcripts (each column=1 biological replicate; 2 replicates per treatment). Bottom, Venn diagram of overlap between transcripts altered by KCl and TTX. (b) Corresponding heatmaps from PolyA− RNA-seq reveal relationship between activity-related mRNA and ecRNA changes. PolyA− RNA transcription from 5′, intronic and 3′ sites all correlated significantly with mRNA changes following neuronal depolarization with KCl (linear regression, P<0.0001 for each comparison). (c,d) Representative examples of activity-induced increases (Fos gene, (c)) and decreases (Jrk gene, (d)) in 3′ ecRNA levels. RNA-seq reads shown individually; read depth truncated to highlight ecRNA changes. (e,f) Validation of RNA-seq results with RT-qPCR (n=6 biological replicates) confirms changes in mRNA and ecRNA levels. Data are expressed as mean±s.e.m. Individual comparisons made with Student's t-test versus vehicle, *P<0.05, **P<0.01, ***P<0.001 and ****P<0.0001.

Figure 3. High ecRNA expression is inversely related to promoter DNA methylation at genes involved in neuronal function and brain disease.

(a) Hierarchical clustering of percentile-normalized ecRNA, DNA mC and mRNA highlights major gene clusters. Clustering was performed using percentile scores instead of raw data to equalize data range. (b) Overlap between distinct gene clusters and activity-responsive (Veh versus KCl genes from Fig. 2a), genes harbouring mutations associated with autism spectrum-disorders (ASD risk genes), genes implicated in Alzheimer's disease risk (AD risk genes) and genes altered in an animal model of Alzhiemer's disease (AD mouse model ΔmRNA genes; downregulated (−) and upregulated (+)). Each circle represents a single gene. (c–f) Distribution of ecRNA, DNA mC and mRNA scores (top panels) and gene ontology analysis (bottom panels) of selected clusters. Gene ontology groups are ranked by corrected P value. Cluster 3b (e) is enriched for genes involved in neurogenesis, neuronal projection and neuronal development. This cluster represents 19.8% of genes investigated but contains 34% of activity-regulated genes (odds ratio=1.69; P=0.0079), 40% of the genes implicated in ASD (odds ratio=2.006; P=0.0003) and 39% of genes that are downregulated in a mouse AD model (odds ratio=1.966; P<0.0001).

Figure 4. Fos ecRNA is differentially responsive to neuronal activation and undergoes unique biogenesis.

(a) RT-qPCR template locations used to distinguish between Fos mRNA and ecRNA. (b) Modulation of Fos mRNA by KCl, AMPA, NMDA and TTX reveals overall timecourse and activity-dependence of Fos gene transcription. Fos mRNA is upregulated as soon as 45 min after KCl treatment, and peaks at 4 h following stimulation. AMPA and NMDA treatment (1 h) produced dose-dependent increases in Fos mRNA, whereas neuronal silencing with TTX decreased Fos mRNA. (c) In contrast, Fos ecRNA is induced within 30 min of neuronal depolarization with KCl, and peaks at 45 min following KCl treatment. Neuronal stimulation with KCl, AMPA and NMDA induced much lower levels of ecRNA transcription (as compared with mRNA), and neuronal inactivation with TTX did not alter Fos ecRNA transcription. (n=3–6 per group for KCl experiments, 3 per group for AMPA and NMDA experiments, and 6 per group for TTX experiments). (d,e) 4-h pretreatment with the RNAPII-dependent transcription inhibitor 5, 6-dichloro-1-beta-D-ribofuranosylbenzimidazole (DRB) had no effect on Fos ecRNA but blocked mRNA induction after 1-hour KCl treatment, whereas pretreatment with the RNAPIII inhibitor ML-60218 had no effect on Fos mRNA but decreased ecRNA induction after KCl treatment (n=6–12 per group; mRNA one-way ANOVA, F_(3,32)=78.86, P<0.0001; ecRNA one-way ANOVA, F_(3,32)=66.55, P<0.0001; Tukey's post hoc test for individual comparisons). (f) Left, MeDIP experimental design. Right, DNA methylation decreases 24 h after KCl treatment in the enhancer, promoter and gene body of the Fos locus (n=8, unpaired Student's t-test; t₁₆>2.733 and P<0.015 for each comparison). All data are expressed as mean±s.e.m. Individual comparisons, *P<0.05, **P<0.01, ***P<0.001 and ****P<0.0001.

Figure 5. Fos ecRNA interacts with DNA methyltransferases and blocks DNA methylation.

(a) Immunostaining reveals nuclear localization of DNMT1 and DNMT3a in neuronal cultures. Cell nuclei are stained with 4,6-diamidino-2-phenylindole (DAPI), and neurons are marked by MAP2 (microtubule-associated protein 2). Scale bar, 50 μm. (b) Fos ecRNA/mRNA comparison in total neuronal lysate and nuclear fraction (separated during RNA-IP; n=4 per group; unpaired Student's t-test, t₆=6.301, P=0.007). (c) Fos ecRNA, but not mRNA, immunoprecipitates with anti-DNMT1 or DNMT3a antibodies but not control IgG (n=4–6 per group; ecRNA one-way ANOVA, F_(2,15)=20.53, P<0.0001; Tukey's post hoc test for individual comparisons). (d) Locations of synthetic RNA and DNA oligonucleotides used in mobility shift assays. (e) Electrophoretic mobility shift assay reveals only slight binding of ecRNA probes (1 nM) to recombinant DNMT1 (0.2 μM). RNA/DNMT complexes are evident as low-mobility band on native PAGE gel following incubation with DNMT protein. (f) Synthetic ecRNA probes (1 nM) bind recombinant DNMT3a/DNMT3l protein (0.2 μM). (g) Complete binding of synthetic ecRNA probes (1 nM) to truncated recombinant DNMT3a protein (0.2 μM) containing only catalytic domain (DNMT3a-CD). (h) Incubation of ecRNA-1 or ecDNA-1 probes (1 nM) with escalating concentrations of DNMT3a-CD (0.002–0.2 μM). (i) DNMT3a-CD binds equally to RNA and double-stranded DNA with the same primary sequence. Binding affinity (Kd values; derived from non-linear, one-site regression analysis of complete concentration curve) for RNA and dsDNA was not significantly different (n=2 replicates; comparison of Kd, F_(1,22)=0.47, P=0.49). (j) Competition assay between ecRNA-1 (1 nM) and unlabelled ecDNA-1 probes (1–100 nM) shows intact RNA/DNMT3a-CD complexes even with 10-fold higher concentrations of DNA. (k) Quantification of dsDNA competition assay. (l) Co-incubation of DNMT3a/3l protein and the methyl donor SAM results in cytosine methylation at dsDNA (3.42 nM) from the Fos promoter. Methylation was significantly inhibited by addition of ecRNA (5.16 μM). For (l) n=4 per group, one-way ANOVA, F_(4,15)=40.25, P<0.0001; Tukey's post hoc test for individual comparisons. All data are expressed as mean±s.e.m. Individual comparisons, *P<0.05, **P<0.01, ***P<0.001 and ****P<0.0001.

Figure 6. Fos ecRNA controls Fos gene methylation.

(a) Anti-sense oligonucleotide (ASO) target locations for Fos mRNA and ecRNA. (b) Fos mRNA ASOs decreased mRNA expression with no significant effect on ecRNA (n=8 per group; one-way ANOVA for Fos mRNA, F_(2,21)=42.96, P<0.0001; Tukey's post hoc test for individual comparisons.). (c) Fos ecRNA ASOs decreased both mRNA and ecRNA (n=8 per group; one-way ANOVAs for Fos mRNA and ecRNA, F_(2,21)=36.43 and 95.48, respectively, P<0.0001; Tukey's post hoc test for individual comparisons). (d) Fos ecRNA knockdown reduces Fos protein quantified by immunoblotting (n=6 per group; one-way ANOVA, F_(2,15)=5.438, P=0.0168). (e) Fos ecRNA knockdown (72 h ASO treatment) resulted in increased enhancer, promoter and gene body methylation as measured by MeDIP (n=4 per group; two-way ANOVA main effect of ASO, F_(1,18)=71.60, P<0.0001, Sidak's post hoc test for multiple comparisons). (f) Bisulfite sequencing following Fos ecRNA knockdown confirmed significant promoter hypermethyation (left, average of all CpG sites; 41–45 individual clones/group, χ² test, z=2.213, P=0.0269), driven largely by specific hypermethylation at two distinct CpG sites (right; methylation status between treatment groups compared with a two-way ANOVA with Sidak's post hoc tests adjusted for multiple comparisons at individual CpG sites). Data are expressed as mean±s.e.m. Individual comparisons, *P<0.05, **P<0.01, ***P<0.001 and ****P<0.0001.

Figure 7. Fos ecRNA is induced by behavioural experience and modulates memory formation.

(a) Contextual fear conditioning is associated with increased Fos mRNA and Fos ecRNA in the CA1 subregion of the hippocampus, as compared with experimentally naïve controls (N=naïve, FC=fear conditioned; n=16 per group; Mann—Whitney U test, U=15 and P<0.0001 for Fos mRNA, U=73 and P=0.038 for Fos ecRNA). Area CA1 was subdissected from total hippocampus 1 h following contextual fear conditioning. (b) Experimental timeline for Fos ecRNA ASO experiments in vivo. (c) Fos ecRNA ASO treatment decreased Fos ecRNA expression in the CA1 of the hippocampus (n=8–9 per group, unpaired Student's t-test, t₁₅=2.173, P=0.0462). (d) Top, contextual fear conditioning design. Bottom, Fos ecRNA ASO treatment impaired long-term memory but did not alter baseline freezing or short-term memory (n=8–9 per group; STM unpaired Student's t-test, t₁₅=1.245, P=0.23; LTM unpaired Student's t-test, t₁₅=2.640, P=0.0186). (e) Open field test. Top, representative traces showing animal location during 30 min test session. Fos ecRNA knockdown did not affect total distance travelled (bottom left; unpaired Student's t-test, t₁₄=1.517, P=0.1514) or time spent in the center of an open field test (bottom right; unpaired Student's t-test, t₁₄=0.005, P=0.99; n=8 per group). All data are expressed as mean±s.e.m. Individual comparisons, *P<0.05 and ****P<0.0001.