Research ArticleNeuroepigenetics

Long noncoding RNA NEAT1 mediates neuronal histone methylation and age-related memory impairment

See allHide authors and affiliations

Science Signaling  02 Jul 2019:
Vol. 12, Issue 588, eaaw9277
DOI: 10.1126/scisignal.aaw9277

On memory and aging

It is an unfortunate aspect of aging that memory tends to decline. Butler et al. identified one way by which that happens. Using mice, mouse hippocampal neurons, and stem cell–derived human neurons, the authors found that the long noncoding RNA NEAT1—which is more commonly associated with cancer—was present in greater amounts in the aged hippocampus and facilitated histone methylation that suppressed the expression of c-FOS, a critical memory-associated gene. Depleting NEAT1 in old mice improved their performance in memory-associated behavior tests, whereas overexpressing NEAT1 in the hippocampus of younger mice impaired performance. These findings reveal a lncRNA-regulated epigenetic mechanism that contributes to age-associated cognitive impairment.


Histone methylation is critical for the formation and maintenance of long-term memories. Long noncoding RNAs (lncRNAs) are regulators of histone methyltransferases and other chromatin-modifying enzymes (CMEs), thereby epigenetically modifying gene expression. Here, we investigated how the lncRNA NEAT1 may epigenetically contribute to hippocampus-dependent, long-term memory formation using a combination of transcriptomics, RNA-binding protein immunoprecipitation, CRISPR-mediated gene activation (CRISPRa), and behavioral approaches. Knockdown of the lncRNA Neat1 revealed widespread changes in gene transcription, as well as perturbations of histone 3 lysine 9 dimethylation (H3K9me2), a repressive histone modification mark that was increased in the hippocampus of aging rodents. We identified a NEAT1-dependent mechanism of transcriptional repression by H3K9me2 at the c-Fos promoter, corresponding with observed changes in c-Fos mRNA expression. Overexpression of hippocampal NEAT1 using CRISPRa was sufficient to impair memory formation in young adult mice, recapitulating observed memory deficits in old adult mice, whereas knocking down NEAT1 in both young and old adult mice improved behavior test–associated memory. These results suggest that the lncRNA NEAT1 is an epigenetic suppressor of hippocampus-dependent, long-term memory formation.


Although thousands of long noncoding RNAs (lncRNAs) in the human and murine genomes have been characterized, few lncRNAs are as well studied as the human nuclear-enriched abundant transcript 1 (NEAT1). NEAT1 is evolutionarily conserved between rodents and humans, particularly within the 5′ region of the transcript (1). Multiple isoforms of NEAT1 exist in rodents and in humans, with the longer of the major isoforms proving essential for phase separation and induction of nuclear paraspeckle assembly (2, 3), whereas the shorter NEAT1 isoforms do not appear to be a major regulator of paraspeckle formation (4). Several studies have characterized a number of molecular pathways by which NEAT1 regulates the epigenome, including both paraspeckle-dependent sequestration of transcription factors and paraspeckle-independent scaffolding of chromatin-modifying enzymes (CMEs) (57). In addition, NEAT1 itself has been observed to bind numerous genomic loci and to affect the regulation of transcription (8, 9).

Research on the human NEAT1 has been largely focused on its role as an oncogene in various cancers (10), which is mediated largely through its regulation of epigenetic mechanisms. However, the abundance of NEAT1 and its rodent homolog (Neat1) are increased in the brains of aging humans and mice (1113) and are linked to multiple cognitive and neurodegenerative disorders, including schizophrenia (14), Huntington’s disease (15), Parkinson’s disease (16, 17), Alzheimer’s disease (18), and epilepsy (19, 20). Furthermore, evidence from rodent and human samples suggests that NEAT1 may play a role in neuroplasticity (19); however, despite such extensive relevance to physiology and health, the role of NEAT1 in the epigenetic regulation of genes within hippocampal neurons, particularly during long-term memory formation, is not clear. We used RNA sequencing (RNA-seq) in mouse and human tissue, CRISPR-mediated gene activation (CRISPRa) in vivo, and behavioral memory tests in mice to investigate the functional role of NEAT1 in gene expression dynamics and the role that age-related changes in its expression might play in memory deficits seen in older adults.


Expression of the lncRNA NEAT1 is relatively low in human central nervous system tissue

Expression of NEAT1 is abundant in many cultured cell lines including those characterized in the ENCODE project (Fig. 1A) (21). However, we observed that, in contrast to the abundant expression of NEAT1 observed in most tissues, the human central nervous system (CNS) as a whole, as well as specifically the hippocampus therein (Fig. 1, B and C, outlined in red), expresses minimal quantities of NEAT1 (22, 23). Unsupervised hierarchical clustering based on tissue expression of NEAT1 supports this observation, because CNS tissues segregate cleanly when sorted based on NEAT1 transcript expression (Fig. 1D; fig. S1, A and B; and data file S1).

Fig. 1 Restricted expression of lncRNA NEAT1 in human CNS tissues.

(A) University of California Santa Cruz (UCSC) Genome Browser track export showing expression of NEAT1 in seven cell types from ENCODE. H1-hESC, H1-human embryonic stem cell line; HSMM, human skeletal muscle myoblasts; HUVEC, human umbilical vein endothelial cells; NHEK, normal human epidermal keratinocytes; NHLF, normal human lung fibroblasts. (B) Human body plot illustrating the expression of NEAT1 in 53 human tissues from the genotype-tissue expression (GTEx) project; values shown are the median transcripts per million (TPM) values by tissue, and hippocampus are outlined in red. (C) Barplots showing median, upper-quartile, and lower-quartile expression of the NEAT1 gene (ENSG00000245532.4) in 53 human tissues from the GTEx project; hippocampal expression is outlined in red. (D) Hierarchical clustering of NEAT1 based on transcript isoform level expression in 53 human tissues from the GTEx project. Dendrogram scale shows cluster distance. Expression values displayed in the heatmap are the median expression values in TPM for each isoform in each tissue. See data file S1 for complete n values for all tissue types.

Examination of single-cell RNA-seq data from resected human CNS tissue and glioblastoma (24) further suggested that expression of NEAT1 within CNS cells is restricted (low) in neurons, whereas other cell types including astrocytes, oligodendrocytes, and vascular cells express NEAT1 at higher amounts (fig. S1, C and D). This is in contrast to the lncRNA transcript MALAT1 (metastasis-associated lung adenocarcinoma transcript 1), which is transcribed from a region adjacent to the NEAT1 gene locus and which appears to be ubiquitously expressed at high abundance in all CNS cell types (fig. S1E). Given the growing body of literature that has noted increased expression of NEAT1 in the aging brain (12, 13), as well as the established role of NEAT1 as a regulator of epigenetic mechanisms and the recently described role of NEAT1 in cognitive disorders, such as schizophrenia (14), we sought to further investigate the role of NEAT1 on the neuroepigenetic mechanisms of cognition.

NEAT1 suppresses the immediate-early and synaptic plasticity-related gene c-FOS

To investigate the role of NEAT1 at the transcriptomic level, we analyzed a publicly available RNA-seq dataset from induced pluripotent stem cell (iPSC)–derived human neurons. Antisense oligonucleotide (ASO) knockdown of NEAT1 in KCl-treated human neurons revealed an extensive cohort of differentially expressed messenger RNAs (mRNAs). Knockdown alone was not sufficient to perturb the transcriptome in resting iPSC-derived human neurons, as evidenced by an imperfect separation after unsupervised hierarchical clustering before KCl stimulation (Fig. 2A). In contrast, NEAT1 knockdown appeared to potentiate KCl-induced differential expression of many genes (Fig. 2, A and B).

Fig. 2 NEAT1 regulates expression of c-FOS mRNA and the AP-1 complex in iPSC-derived human neurons.

(A) Unsupervised hierarchical clustering transcriptomes from Neat1/Control ASO and KCl-treated iPSC-derived human neurons, based on DESeq2-normalized counts. (B) Venn diagram depicting the total number of differentially expressed genes detected between KCl + Control ASO and KCl + Neat1_ASO groups detected by DESeq2. (C and D) GO term enrichment for differentially expressed genes depicted in (B). All GO terms shown showed statistically significant enrichment [P < 0.05, Benjamini-Hochberg (BH) corrected]. PERK, PKR-like ER kinase; COPII, coat protein complex II. (E and F) Normalized count values for lncRNA NEAT1 and c-FOS mRNA either before (E) or after (F) KCl treatment of iPSC-derived neurons. Count values significantly different between groups (P < 0.05, BH corrected).

To gain some insight into the physiological relevance for the observed NEAT1-mediated changes in gene expression in human neurons, we queried the annotated disease classes from the Genetic Association Database using the Database for Annotation, Visualization and Integrated Discovery (DAVID) tool (25) and observed significant enrichment for three disease classes: cancer, renal, and aging (data file S2). Gene ontology (GO) enrichment was then assessed using a PANTHER (Protein Analysis Through Evolutionary Relationships) Overrepresentation Test, and differentially expressed genes appeared to be nonrandomly distributed among annotated biological processes (Fig. 2C), molecular functions (data file S2), and cellular components (Fig. 2D). Significant GO term enrichment was partially consistent with previous observations of the NEAT1 regulatory axis (26), as we observed significant regulation of GO terms associated with viral gene expression; however, we also observed significant enrichment of GO terms important for hippocampal function, including the transcription factor activator protein 1 (AP-1) complex (GO: 0035976; Fig. 2, C and D).

The human Fos proto-oncogene (FOS, also known as c-FOS), a critical component of the AP-1 transcription factor subunit, appeared to be overexpressed in human neurons after knocking down NEAT1 both in quiescent and KCl-stimulated neurons (Fig. 2, E and F). Because c-Fos has known relevance to hippocampus-dependent memory formation (27), we selected the murine homolog c-Fos as a candidate gene to further study Neat1’s regulatory potential.

NEAT1 is down-regulated by neuronal excitation

Because modeling NEAT1 expression changes in response to in vivo neuronal activity and behavioral experience required the use of mammalian model organisms, we next sought to examine the regulatory capacity of NEAT1 in rodent neurons. For this purpose, we knocked down murine NEAT1 in the mouse Neuro-2a (N2a) cell line using small interfering RNAs (siRNAs). We observed that 24 hours after treatment with Neat1-targeting siRNAs (Fig. 3A), the abundance of the c-Fos mRNA was significantly increased (Fig. 3B).

Fig. 3 NEAT1 represses expression of c-Fos mRNA in murine neuronal cells.

(A) Reverse transcription quantitative polymerase chain reaction (RT-qPCR) analysis of the abundance of Neat1 transcript in control or Neat1-targeted siRNA-treated murine N2a cells. Data are n = 5 to 6 independent cultures of N2a cells. ***P < 0.0005 by Student’s t test. (B) RT-qPCR analysis of the abundance of c-Fos mRNA in control or Neat1-targeted siRNA-treated murine N2a cells. Data are cultured cells from n = 3 to 4 experiments. *P < 0.05, Student’s t test. (C and D) RT-qPCR analysis of Neat1 transcript (C) and c-Fos mRNA (D) abundance in dorsal Cornu Ammonis 1 (dCA1) from naïve or contextual fear conditioning (CFC)–trained mice. Data are from n = 8 to 9 mice. *P < 0.05, Student’s t test. (E and F) RT-qPCR analysis of Neat1 transcript (E) and c-Fos mRNA (F) abundance in CFC-trained mice. Data are cultured primary neurons from n = 4 to 7 experiments. *P < 0.05 and ***P < 0.0005, Student’s t test. GAPDH, glyceraldehyde 3-phosphate dehydrogenase; Pyr, pyramidal neurons.

Although our observations of the c-Fos transcript expression in murine neurons recapitulated observations from human neurons, we observed that mRNA expression of the associated immediate-early genes Egr1 and Btg2 and others was not increased in mouse (fig. S2, A and B) as they were in human neurons (data file S2), suggesting that there are species-specific regulatory differences in the NEAT1 regulatory axis.

Given that mice lacking c-Fos expression in the CNS show a specific loss of hippocampus-dependent spatial and associative learning tasks (27), we next sought to investigate the relevance of Neat1 expression during memory consolidation after a hippocampus-dependent learning task. One hour after training in CFC, we observed a significant reduction in the expression of Neat1 in the dorsal hippocampus coinciding with previously reported increases in expression of the c-Fos mRNA (Fig. 3, C and D). Because baseline expression of Neat1 in neurons is expected to be quite restricted compared to other cell types, we next stimulated neurons with KCl to ascertain the effect of activity on the expression of Neat1. Consistent with recent reports (19), we observed that KCl stimulation induced a rapid reduction in Neat1 expression in both N2a cells (fig. S2E) and primary hippocampal pyramidal neurons (Fig. 3, E and F), consistent with the effects of CFC in vivo.

NEAT1 regulates H3K9me2 globally and controls both H3K9me2 of the c-Fos promoter and c-Fos expression

We next sought to investigate the c-Fos–relevant mechanisms of NEAT1-orchestrated transcriptional control. To accomplish this, we used publicly available data assaying NEAT1 binding to chromatin with capture hybridization analysis of RNA targets and high-throughput sequencing (CHART-seq) in human Michigan Cancer Foundation 7 (MCF7) cells (8). After mapping NEAT1-bound peaks to the nearest transcription start sites, we observed that only a small subset of genes directly bound by NEAT1 is differentially expressed either after NEAT1 knockdown or in the context of neuronal activation. However, we observed significant enrichment of NEAT1 binding near genes associated with histone methyltransferase activity, including the histone H3 lysine 9 (H3K9) dimethyltransferase EHMT1 (euchromatic histone lysine methyltransferase 1) [also known as G9a-like protein (GLP)] (Fig. 4, A and B; fig S5; and data file S3) (28).

Fig. 4 NEAT1 modulates neuronal H3K9me2.

(A) NEAT1 CHART-seq peaks were mapped to the nearest gene transcription start site, and functional enrichment was assessed using ChIP-Enrich, with histone methyltransferase activity being noted as a significantly enriched GO term (*P < 0.05 for all terms shown, BH corrected). (B) UCSC Genome Browser plot showing NEAT1-binding peaks overlapping the human EHMT1 gene. (C to G) Analysis of Western blots assessing histone modifications H3K27me3 (C), H3K4me3 (D), H3ac (E), and H3K9me2 (F) globally and ChIP-qPCR assays at the H3K9me2 at the c-Fos gene promoter (G) in N2a cells after siRNA-mediated knockdown of NEAT1. Data are cultured cells from at least three independent experiments. *P < 0.05, Student’s t test [(F) P = 0.0456 and (G) P = 0.0472]. (H) RNA-binding protein immunoprecipitation followed by RT-qPCR to assess for Ehmt2/Neat1 interaction in murine N2a cells. Data are means of two independent experiments, each represented by a data point. IgG, immunoglobulin G.

The c-Fos locus has previously been observed to be methylated by the EHMT1/2 complex in the context of hippocampus-dependent memory formation (29). Therefore, we next sought to investigate the role of NEAT1 in the regulation of histone methylation and histone 3 lysine 9 dimethylation (H3K9me2) specifically. After knockdown of NEAT1 in neuronal cells, we observed that H3K9me2 is reduced at a global scale, whereas we did not observe such changes in the expression of several histone marks previously associated with NEAT1 in other cell types (Fig. 4, C to F).

To assess the functional relevance of the NEAT1-H3K9me2 regulatory axis on the expression of c-Fos mRNA, we performed chromatin immunoprecipitation assays followed by qPCR (ChIP-qPCR) at the c-Fos promoter. We observed that after NEAT1 knockdown with siRNAs, H3K9me2 at the c-Fos promoter was significantly depleted (Fig. 4G), consistent with observed changes in gene expression (Fig. 3B), whereas H3K9me2 within the c-Fos gene body was not significantly changed (fig. S2F).

To ascertain whether NEAT1 physically associates with the H3K9me2 methyltransferase complex in neurons, we performed RNA binding protein immunoprecipitation (RIP assays) against the EHMT2 subunit of the obligatory EHMT1/2 heterodimer (Fig. 4H) (2931). Consistent with published results (6), we observed an interaction between NEAT1 and EHMT2, which suggests that NEAT1 may bind both protein components of EHMT1/2 complex, as well as the Ehmt1 locus (Fig. 4, A and B).

NEAT1 knockdown improves hippocampal memory formation and derepresses the epigenetic landscape at the c-Fos promoter in vivo

Having demonstrated that NEAT1 alters the epigenetic landscape and represses neuronal expression of the memory-critical c-Fos gene in cultured cells, we next sought to investigate the functional role of NEAT1 expression on c-Fos promoter methylation and memory formation in vivo. To explore whether Neat1 expression affects hippocampus-dependent memory formation, we knocked down the abundance of NEAT1 in the hippocampal area CA1 by directly infusing Neat1-targeting siRNAs or, as a control, nontargeted siRNAs and assayed long-term memory function by means of CFC, a hippocampus-dependent memory task (Fig. 5A). We observed that 5 days after intra-CA1 injection of Neat1-targeting siRNAs, a time when we observed significantly reduced expression of Neat1 (fig. S3A), mice had no substantial difference in freezing behavior during the training phase of CFC, either before or after delivery of the foot shock (Fig. 5B). However, when returned to the training context 24 hours later, mice that had been injected with Neat1-targeting siRNAs displayed significantly increased freezing behavior relative to mice that had been injected with nontargeting siRNAs (Fig. 5C).

Fig. 5 Neat1 knockdown regulates c-Fos promoter methylation in vivo and improves long-term memory.

(A) Graphic depiction of siRNA infusion into hippocampal area CA1 and single-pairing CFC paradigm. Briefly, 3- to 7-month-old male C57BL/6 mice were trained 5 days after bilateral infusion of siRNAs and tested 24 hours after training. (B) Freezing behavior (Frz) of mice described in (A) as a percent of epoch during training phases of the CFC paradigm. No significant difference was detected for either the “pre-shock” or “post-shock” epochs. Data are means ± SEM from n = 8 mice, assessed by Student’s t test. (C) Freezing behavior of mice described in (A) as a percent of total time during the 5-min test trial. Data are means ± SEM from n = 8 mice. *P = 0.0307, Student’s t test. (D) ChIP-qPCR analysis of H3K9me2 at the c-Fos promoter in dCA1 tissue from animals euthanized 5 days after the conclusion of the behavior experiments described in (A). Data are from dCA1 tissue of 4 to 6 mice. *P = 0.0450, Student’s t test.

To determine whether NEAT1 abundance affects c-Fos promoter methylation in vivo, we euthanized an additional cohort of behaviorally naïve animals 5 days after injection of siRNAs and performed ChIP-qPCR on one hemisphere of dCA1 tissue collected from around the injection site. Consistent with our results in cultured neurons described above (Fig. 4G), we observed that concurrent with NEAT1 knockdown 5 days after infusion with siRNAs, the amount of H3K9me2 at the c-Fos promoter in dorsal area CA1 (dCA1) was significantly reduced (Fig. 5D). Thus, we hypothesized a model in which NEAT1 abundance might be regulating memory formation through epigenetic repression of c-Fos.

Mimicking age-related up-regulation of Neat1 expression impairs, and knocking down Neat1 restores, hippocampus-dependent memory formation

Numerous studies have reported overexpression of NEAT1 in senescing cells, as well as in aging CNS tissues in both humans and mice (11, 32). Upon comparing publicly available hippocampus RNA-seq datasets from 3-month-old (young) mice versus 24-month-old (aged) mice, we observed increased expression of Neat1 in the hippocampus of aged relative to that of young animals (Fig. 6A) and decreased expression of c-Fos (Fig. 6B), both consistent with previously reported results and age-associated hippocampus-dependent memory impairments, respectively (11).

Fig. 6 Neat1 knockdown improves long-term memory in aged animals.

(A and B) DESeq2-generated normalized counts for Neat1 (A) and c-Fos (B) from RNA-seq assay of hippocampi from 3- and 24-month C57/B6 mice. Data are means ± SEM from n = 5 to 6 mice (*P < 0.05, BH adjusted). Neat1 abundance (A) and c-Fos mRNA abundance were (*P < 0.05, BH corrected) (B) was significantly repressed in aged hippocampi relative to the hippocampi of young mice (*P < 0.05, BH corrected). (C) Graphic depiction of siRNA infusion into hippocampal area CA1 and three shock-pairing CFC paradigm. Briefly, 18-month-old male C57/B6 mice were trained 5 days after bilateral infusion of siRNAs and tested 24 hours after training. (D) Freezing behavior of mice is described in (C) as a percent of time during the training phase of the CFC paradigm. No significant difference was detected. Data are means ± SEM from n = 7 to 9 mice, assessed by Student’s t test. (E) Freezing behavior of mice described in (C) as a percent of time during the testing phase of the CFC paradigm. Data are means ± SEM from n = 7 to 9 mice. *P = 0.0039, Student’s t test.

We next tested whether hippocampus-dependent memory formation might be improved in aged mice by knocking down NEAT1. To this end, we knocked down NEAT1 in the hippocampal area CA1 of 18- to 19-month-old mice (fig. S4), an age at which we have previously observed significant up-regulation of H3K9me2 in the aging rat hippocampus (33), by directly infusing Neat1-targeting siRNAs or nontargeted siRNAs. We then assayed long-term memory function using CFC, with three pairings of the shock to the novel context (Fig. 6C). We observed that knockdown of NEAT1 in the dorsal hippocampus of aged mice resulted in significant improvements in freezing after 24 hours but not during training (Fig. 6, D and E), similar to the results we had observed in young mice (described above; Fig. 5C).

We next sought to test the sufficiency of NEAT1 overexpression to impair performance in memory tasks. To this end, we designed a single-guide RNA (sgRNA) targeting Neat1 for overexpression from the endogenous locus (Fig. 7, A and B, and fig. S3, B and C) and delivered a Neat1-targeting CRISPRa system in vivo into dCA1 through in vivo transfection (Fig. 7A). Mice were then trained in CFC with three pairings of the shock to the novel context (Fig. 7C). Animals overexpressing NEAT1 from the endogenous locus (NEAT1-OE) had no significant differences in freezing during the training period, either before or after exposure to the unconditioned stimulus (Fig. 7C); however, when returned to the training context 24 hours after training, NEAT1-OE animals froze significantly less than control animals that received only the sgRNA plasmid (Fig. 7D), suggesting that increased NEAT1 abundance in area CA1 is sufficient to impair hippocampus-dependent memory formation.

Fig. 7 Mimicking age-related Neat1 overexpression using CRISPRa impairs hippocampal memory formation.

(A) Graphic depiction CRISPRa system infusion into hippocampal area CA1, with visualization of hippocampal expression of enhanced green fluorescent protein fluorescent marker and three shock-pairing CFC paradigm. Briefly, male C57BL/6 mice (3 to 7 months old) were trained 21 days after bilateral infusion of either sgRNA plasmid alone or co-delivered with a transcription-activating dCas9-effector protein and tested 24 hours after training. (B) Confirmation of efficacy of CRISPRa system to up-regulate Neat1 expression in murine N2a cells (n = 4, P = 0.0283, Student’s t test). (C) Freezing behavior of mice described in (A) as a percent of time during the training phase of the CFC paradigm. No significant difference detected for either the pre- or post-shock epochs. Data are means ± SEM from n = 18 mice, assessed by Student’s t test. (D) Freezing behavior of mice described in (A) as a percent of time during the testing phase of the CFC paradigm. Data are means ± SEM from n = 18 mice. *P = 0.450, Student’s t test.


Previous studies have observed regulatory roles for the lncRNA NEAT1, including that NEAT1 localizes to chromatin and governs chromatin modification (8, 9). Here, we aimed to resolve this regulatory role of NEAT1 in the context of long-term memory formation. Our data revealed that murine NEAT1 acts as a potent regulator of H3K9me2 both in cultured cells and in vivo. Because of a previously reported observation that NEAT1 interacts directly with EHMT2 (6), which we reproduced here, we cannot yet ascertain whether transcriptional control of EHMT1 or direct interaction with the repressor complex is the rate-limiting factor for H3K9me2 abundance. This intricate multipoint interaction is perhaps illustrative of the intricate systems of regulatory feedback that are thought to control epigenetic mechanisms. Nonetheless, knockdown of NEAT1 was sufficient to perturb this system and to result in both bulk and site-specific changes in H3K9me2, an epigenetic mark that has been observed to play a crucial role in the in neurons.

Previous studies investigating NEAT1 in the context of epilepsy have reported that, in the excitotoxic conditions of this neurodegenerative disorder, activity-dependent down-regulation of NEAT1 expression is impaired (19). Thus, we hypothesized that similar mechanisms might play a role in brain regions that become hyperexcitable during aging. Although our results indicate that NEAT1 up-regulation impairs memory performance, we wonder whether increased abundance of NEAT1 might be involved in neuroprotective feedback in the context of neurodegeneration or excitotoxicity. However, observations from studies on other neurodegenerative diseases would suggest that NEAT1 up-regulation is deleterious to neuronal survival (17). Our observations here also suggest that NEAT1 up-regulation might be sufficient to explain some memory impairments observed in aging, as well as in rodent models of temporal lobe epilepsy and other neurodegenerative diseases.

Investigations as to the epigenetic regulatory role of NEAT1 have resulted in paradoxical observations to the effect that NEAT1 binds to genomic loci and mediates transcription activation (9) but that suppression of Neat1 expression results primarily in increased neuronal gene expression (our results above). We showed here that knockdown of NEAT1 induced widespread down-regulation of neuronal H3K9me2, potentially explaining observed increases in gene expression and further explaining age-related increases in H3K9me2 previously observed in the hippocampus. Moreover, we observed that, in mice, Neat1 expression was correlated with H3K9me2 globally, as well as at the promoter of the aging-repressed memory-related gene c-Fos. Although NEAT1 has been observed to act on numerous epigenetic mechanisms (6, 8, 9, 34, 35), this finding suggests that NEAT1-mediated epigenetic mechanisms may be sufficient to govern cognitive function.

Studies of the neuronal impact of NEAT1 expression have thus far been limited to the context of neurological disorders, and in many cases, to cultured neuronal cells. Our observations suggest that NEAT1 plays a regulatory role in neuronal H3K9me2 both in cultured neurons and in vivo and that increased NEAT1 abundance might play a notable role in the age-related decline of hippocampus-dependent memory formation. In humans, expression of NEAT1 is generally limited to a low amount in the CNS, and overexpression is a common hallmark of several neurological disorders. Although experimental reduction of NEAT1 has been shown to alleviate 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine–induced neurogeneration in human neuronal cell lines and may have therapeutic potential in mouse models of Parkinson’s disease, the impact of age-related changes in expression has remained unexplored until now. Although our experiments were designed to investigate the age-related impact of Neat1, we note that other studies have described neuroinflammation-mediated increases in Neat1 expression (36); hence, the findings described in this manuscript implicate NEAT1 as a potential mechanism by which neuroinflammation might affect memory.

Although the experiments described here may explain previous observations of increased H3K9me2 in the aging hippocampus (33), our experiments indicate that increased expression of NEAT1 is not sufficient to explain all of the aging-related neuroepigenetic changes observed in this region. It is likely that many hippocampal lncRNAs have distinct or overlapping roles in the regulation of the neuroepigenetic aging process. Human NEAT1 itself has been observed to associate with multiple CMEs (6, 26, 37, 38). Although we did not detect significant regulation of histone modifications other than H3K9me2 at the global level after knockdown of NEAT1, the absence of such observations does not preclude the existence of biologically or behaviorally meaningful epigenetic regulation of marks not explored in this manuscript or site-specific epigenetic changes that might be uncovered in future studies with a large-scale sequencing approach. In summary, our findings demonstrate that the lncRNA NEAT1 regulates a critical transcriptional pathway for hippocampus-dependent memory in rodent neurons in vitro and in vivo and likewise in iPSC-derived human neurons and suggest that NEAT1 may serve as an endogenous molecular brake on the formation of hippocampus-dependent spatial memories.


Animal housing

Naïve 3- to 7-month-old or 18-month-old C57BL/6 mice were group-housed (two to seven animals per cage) in plastic cages with ad libitum access to food and water and were maintained on a 12-hour light/12-hour dark cycle. All behavioral tests were conducted during the light cycle, and all procedures were approved by The University of Alabama at Birmingham Institutional Animal Care and Use Committee and performed in accordance with the National Institute of Health ethical guidelines.

Cell culture

N2a cells were maintained in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum. After thawing, the cells were passaged a minimum of two times before use in experiments. The cells were kept at 37°C in a 5% CO2 incubator. Dissociated cultures of hippocampal pyramidal cells were obtained from embryonic day 18 rat embryos as described previously (39). Briefly, timed-pregnancy female Sprague-Dawley rats were terminally anesthetized, and embryos were removed from the uterus and then transferred to Hank’s balanced salt solution (HBSS; Gibco) for dissection. Primary rat hippocampal neurons were dissociated using incubation with papain for 20 min at 37°C, rinsed in HBSS, and then resuspended in Neurobasal medium (Gibco) and further mechanically dissociated by passing through a series of progressively smaller fire-polished glass Pasteur pipettes. The resulting suspension was passed through a 70-μm cell strainer and plated on poly-l-lysine–coated 24-well plates (~7.5 × 104 cells per well). Cells were maintained for 2 weeks in Neurobasal medium supplemented with B-27 and GlutaMAX (Thermo Fisher Scientific) at 37°C and 5% CO2. For KCl stimulation, 6.25 μl of 1 M KCl (Sigma-Aldrich) was added to 2-week in vitro cultures, for a final concentration of 12.5 mM KCl.

siRNA delivery

Young (3 to 7 months old) mice were anesthetized with an intraperitoneal injection of ketamine-dexmedetomidine and received bilateral intra-CA1 injections of Lincode SMARTpool siRNAs (Dharmacon), targeting the murine Neat1 (no. R-160022-00-0005) or a negative control (no. D-001320-10-05), conjugated with in vivo jetPEI (Polyplus-transfection), an in vivo transfection reagent, at the stereotaxic coordinates [anterior-posterior (AP), −2.0 mm; medial-lateral (ML), ±1.5 mm; dorsal-ventral (DV), − 1.7 mm] with respect to bregma. Aliquots of siRNA stocks (100 μM) were diluted to a concentration of ~2.5 μM and conjugated with in vivo jetPEI on the day of surgery. Infusions were given over a 10-min period (0.1 μl/min) for a total volume of 1 μl per hemisphere. After a 48-hour recovery period, mice were handled daily for >3 min and trained in CFC at 5 days after surgery. Aged (18 to 19 months old) mice were treated similarly but were anesthetized with vaporized isoflurane (3% induction and 2% maintenance). Mice were euthanized at 10 days after surgery, and dCA1 was harvested from each hemisphere.

CRISPRa delivery

Mice were anesthetized with an intraperitoneal injection of ketamine-dexmedetomidine and received bilateral intra-CA1 injections of an sgRNA expression vector driven by the murine U6 promoter and targeting the murine Neat1 promoter region (Addgene plasmid no. 44248) either alone or in conjunction with an expression vector coding for the S. pyogenes dCas9 fused to two copies of the VP64 transactivator domain (Addgene plasmid no. 59791). Endotoxin-free plasmids were purified using an endotoxin-free plasmid DNA purification kit (Macherey-Nagel) and aliquoted to minimize freeze-thaw cycles. Endotoxin-free plasmid stocks were diluted to a final concentration of ~500 ng/μl in sterile 10% glucose and incubated with in vivo jetPEI for 15 min at room temperature on the day of surgery. The resulting transfection complex was delivered by direct infusion at the stereotaxic coordinates (AP, −2.0 mm; ML, ±1.5 mm; DV, −1.4 mm) with respect to bregma. Infusions were given over a 10-min period (0.1 μl/min) for a total volume of 1 μl (~500 ng of plasmid DNA) per hemisphere.

Contextual fear conditioning

Mice were trained to either a weak or strong CFC paradigm in a novel context, and long-term memory was assessed upon returning the animals to the training context 24 hours after training. The weak CFC paradigm consisted of a 118-s baseline followed by a single-shock (0.5 mA, 2 s) pairing in the novel context, whereas the strong CFC paradigm consisted of a 119-s baseline followed by three-shock pairings (0.5 mA, 1 s) with interleaved rest periods of 59 s each. Twenty-four hours after training, animals were placed back into the training context for 5 min to test retention. Freezing behavior was scored by Med Associates software.

Collection of whole-area CA1

One hour after training, the whole brain was removed by gross dissection and placed in oxygenated (95%/5% O2/CO2) ice-cold cutting solution (110 mM sucrose, 60 mM NaCl, 3 mM KCl, 1.25 mM NaH2PO4, 28 mM NaHCO3, 0.5 mM CaCl2, 7 mM MgCl2, 5 mM glucose, and 0.6 mM ascorbate). The CA1 region of the hippocampus was then microdissected from each hemisphere and flash-frozen on dry ice.

Collection of dCA1

Animals were euthanized by cervical dislocation after overdosing with isoflurane at experiment-specific time points, and the whole brain was rapidly removed and immediately frozen on dry ice. The CA1 region of the dorsal hippocampus was then dissected out with the aid of a mouse brain matrix (Harvard Apparatus) to collect the area of CA1 targeted by siRNA or CRISPRa infusions. All tissues were stored at −80°C before processing.

Western blotting

Normalized proteins (2 to 10 μg) were separated by electrophoresis on either 10 or 20% polyacrylamide gels and transferred onto an Immobilon-FL PVDF membrane using a turbo transfer system (Bio-Rad). Membranes were blocked in LiCor blocking buffer and probed with primary antibodies for histone H3 (1:1000; Abcam, no. ab1791), H3K9me2 (1:1000; Millipore, no. 07-441), H3K27me3 (1:1000; Millipore, no. 07-449), and H3K4me3 (1:1000; Millipore, no. 04-745). Secondary goat anti-rabbit 700CW antibody (1:20,000; LiCor Biosciences) was used for detection of proteins using the LiCor Odyssey system. All Western blot quantifications were done using Image Studio Lite software (LiCor).

Reverse transcription qPCR

RNA was extracted from isolated CA1 or cultured cells using TRIzol reagent according to the manufacturer’s recommended protocol (Fisher Scientific). RNA yield was quantified spectrophotometrically (NanoDrop 2000c), and ~200 ng of RNA was deoxyribonuclease (DNAse)–treated (amplification grade DNAse I, Sigma-Aldrich), converted to complementary DNA (cDNA; iScript cDNA Synthesis Kit, Bio-Rad), and PCR-amplified on the CFX 1000 real-time PCR system (Bio-Rad), with primer annealing temperatures of 60°C. Full descriptions of primers used are in the Supplementary Materials (data file S4). All data were analyzed using the ΔΔCt method (40).

Chromatin immunoprecipitation

ChIP was performed as described previously (33, 41). Briefly, samples were fixed in phosphate-buffered saline with 1% formaldehyde for 10 min at room temperature, chromatin was sheared using a Bioruptor XL on high power, and lysates were cleared by centrifugation and diluted in tris-EDTA buffer and RIPA buffer. Extracts were mixed with MagnaChIP protein A/G beads, and immunoprecipitations were carried out at 4°C overnight with 5 μg of primary antibody (anti-H3K9me2, Cell Signaling, no. D85B4; rabbit IgG, Abcam, no. ab37415) or no antibody (control). Immune complexes were sequentially washed with low-salt buffer [20 mM tris (pH 8.0), 0.1% SDS, 1% Triton X-100, 2 mM EDTA, and 150 mM NaCl], high-salt buffer [20 mM tris (pH 8.1), 0.1% SDS, 1% Triton X-100, 500 mM NaCl, and 1 mM EDTA], LiCl immune complex buffer [0.25 M LiCl, 10 mM tris (pH 8.1), 1% deoxycholic acid, 1% IGEPAL-CA630, 500 mM NaCl, and 2 mM EDTA], and TE buffer and eluted into 1× TE containing 1% SDS. Protein-DNA cross-links were reversed by heating at 65°C overnight. After proteinase K digestion (100 μg; 2 hours at 37°C), DNA was purified by phenol/chloroform/isoamyl alcohol extraction and ethanol precipitation. Immunoprecipitated DNA was quantified by spectrophotometry (NanoDrop 2000c), and ~15 ng of DNA from each sample was assayed using RT-qPCR using primers specific to mouse genes of interest. Full descriptions of primers used are in the Supplementary Materials (data file S4).

RNA-binding protein immunoprecipitation

RIP was performed as described previously (42). Briefly, ~5 μg of primary antibody against Ehmt2 (Abcam, no. ab40542), Ezh2 (Abcam, no. ab3748), or normal rabbit immunoglobulin G (Cell Signaling) were conjugated with 25 μl of MagnaChIP protein A/G beads (EMD Millipore). Freshly harvested nuclear pellets from at least 106 N2a cells were sheared by Dounce homogenization (15 to 20 strokes) in RIP buffer [150 mM KCl, 25 mM tris (pH 7.4), 5 mM EDTA, 0.5 mM dithiothreitol, 0.5% NP-40, 1× protease inhibitor cocktail (Sigma-Aldrich), and SUPERASin (100 U/ml; Ambion)], cleared using centrifugation at 13,000 rpm to remove nuclear membrane and debris, and split into fractions for immunoprecipitation. Sheared nuclear extracts were mixed with antibody-conjugated MagnaChIP protein A/G beads, and immunoprecipitations were carried out at 4°C for 4 hours. Beads were then immobilized on a magnetic tube rack, and immune complexes were sequentially washed three times with RIP buffer. Beads were then resuspended in 1 ml of TRIzol (Thermo Fisher Scientific), and coprecipitated RNAs were isolated according to the manufacturer’s recommended protocol. RT-qPCR for Neat1 was then performed as described above.

Statistical analyses

Data from all experiments were analyzed using analysis of variance (ANOVA) with Fisher least significant difference (LSD) post hoc test or with Student’s t test unless otherwise noted in the figure legend. Values were reported in the text, and error bars are the means ± SEM unless otherwise noted. All datasets were screened for outliers before analysis using Grubb’s test (α = 0.05), and outliers were subsequently excluded. Statistical tests were performed in R or Prism 7 (GraphPad). Nonparametric tests were used where appropriate, and tests were two-tailed unless otherwise noted. For all experiments, n indicates the number of biological replicates. For cell culture experiments, this indicates the number of independently growing flasks or wells. For experiments involving animal behavior, this indicates the number of animals used. For experiments involving tissue collection from animals, this indicates the number of animals that we collected the tissue from.

Human tissue expression data and analysis

Data from the GTEx Analysis Release V7 [database of genotypes and phenotypes (dbGaP) accession no. phs000424.v7.p2] were obtained using the GTEx portal web tool. Expression values plotted are in TPM, using the GENCODE-annotated transcript for isoforms or a gene-level model based on the GENCODE model with isoforms collapsed to single genes. Isoform expression values were hierarchically clustered using Euclidean distance and average linkage; dendrogram scale shows cluster distance. Body plot was generated in R from median TPM using the gganatogram package (43, 44).

Analysis of bulk RNA-seq and ChIP-seq data

Single- or paired-end RNA-seq data were imported into the public Galaxy server at directly from the European Nucleotide Archive (study accession numbers PRJEB9006 and PRJNA262674) in FASTQ format and run through a standardized workflow consisting of quality trimming using TrimGalore! (45) (Galaxy version 0.4.2), read alignment using hierarchical indexing for spliced alignment of transcripts (HISAT) (46) (Galaxy version 2.0.3), and feature counting using featureCounts (Galaxy version 1.4.6.p5). Individual count files were grouped by treatment, and differential expression testing was performed using DESeq2 (47) (Galaxy version 2.11.39). All reference genomes and annotations were obtained from GENCODE releases current at the time of analysis, including the Genome Reference Consortium Mouse Build 38 patch release 5 (GRCm38.p5) and evidence-based annotation of the mouse genome (GRCm38), version M16 (Ensembl 91), human build GRCh38, and the human annotation Release 25 (GRCh38.p7). GO enrichment was assessed using a PANTHER Overrepresentation Test web tool provided by the Gene Ontology Consortium (48, 49) (release date, 28 November 2017). DAVID functional annotation was used to assess gene set enrichment for GAD_DISEASE_CLASS using default settings (DAVID 6.8).

CHART-seq data were accessed using the National Institutes of Health (NIH) SRA Toolkit from accession PRJNA252626 and analyzed using similar read quality control and alignment tools as described above. CHART-seq peaks were called using the MACS2 algorithm (50, 51). Overlapping peaks were combined into a single peak, as recommended for input into ChIP-Enrich package. Using the ChIP-Enrich R package (28) (version 2.4.0), CHART-seq peaks from MACS2 were assigned to the nearest transcription start site, and GO enrichment was assessed for Biological Processes and Molecular Functions.

scRNA-seq analysis

Data were obtained from the European Bioinformatics Institute’s Single-cell Expression Atlas. T-distributed Stochastic Neighbor Embedding (t-SNE) plots were constructed using TPM values from the transcriptomes of 3589 single cells biopsied from four patients with glioblastoma (24). Unbiased clusters were generated using a t-SNE perplexity of 10; plots were colored according to biased inferred cell type, as reported by the authors of the dataset. Biopsied tissue included cells from the tumor core and peripheral tissue; however, all cells inferred to be neurons were collected from noncancerous tissue adjacent to the glioblastoma.


Fig. S1. NEAT1 expression is uniquely reduced in the human CNS, and baseline expression is low in human neurons relative to other cell types.

Fig. S2. Neuronal regulation of immediate early genes after NEAT1 knockdown.

Fig. S3. Validation of NEAT1 abundance manipulation using RNA interference and CRISPRa.

Fig. S4. Age-related increase of H3K9me2 in dCA1.

Fig. S5. Quality control plots from ChIP-Enrich.

Data file S1. GTex tissue data.

Data file S2. iPSC differentially expressed genes.

Data file S3. ChIP-Enrich GO results.

Data file S4. Primers and oligos.


Acknowledgments: We would like to thank P. Li at the UAB for providing statistical review. We would also like to thank A. Arrant, A. Tran, and A. Hjelmeland at UAB for logistic and troubleshooting support. Funding: This work was supported in part by NIH grant MH097909 (to F.D.L.), the Evelyn F. McKnight Brain Institute at UAB, and the UAB Neuroscience Behavior Assessment Core P30 NS47466, and A.A.B. was supported by an NIH Institutional Training grant (NS061788). Author contributions: A.A.B. and F.D.L. designed experiments and interpreted the results. A.A.B., D.R.J., and S.K. carried out the experiments and analyzed collected data. All authors wrote the manuscript. A.A.B. and F.D.L. generated figures and revised the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: pLV hUbC-VP64 dCas9 VP64-T2A-GFP was a gift from C. Gersbach [Addgene plasmid no. 59791 (, RRID:Addgene_59791]. pgRNA-humanized was a gift from S. Qi [Addgene plasmid no. 44248 (, RRID:Addgene 44248]. RNA-seq data are available at BioProject study accession numbers PRJEB9006 for iPSC data and PRJNA262674 for aging hippocampus RNA-seq. CHART-seq data are available at study accession number PRJNA252626. All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.
View Abstract

Stay Connected to Science Signaling

Navigate This Article