LSD1 modulates stress-evoked transcription of immediate early genes and emotional behavior

Significance

In mammals, different forms of stress, including psychosocial stress, can affect several aspects of health, fostering mood and anxiety disorders in humans. However, a lack of knowledge about the mechanisms underlying the brain physiology of the stress response hinders the development of new therapeutic strategies. We describe the role of the epigenetic enzyme Lysine-Specific Demethylase 1 (LSD1) in the transduction pathway that translates social stress into an altered transcriptional physiology of plasticity genes in hippocampus. In particular, we show that in the brain LSD1 is finely tuned by a mammal-restricted splicing-based mechanism, and its demethylase activity consistently affects anxiety-like behavior in mice. This work addresses a fundamental mechanism, explaining a brain-related phenotype as a result of genome–environment interactions mediated by the epigenome.

Abstract

Behavioral changes in response to stressful stimuli can be controlled via adaptive epigenetic changes in neuronal gene expression. Here we indicate a role for the transcriptional corepressor Lysine-Specific Demethylase 1 (LSD1) and its dominant-negative splicing isoform neuroLSD1, in the modulation of emotional behavior. In mouse hippocampus, we show that LSD1 and neuroLSD1 can interact with transcription factor serum response factor (SRF) and set the chromatin state of SRF-targeted genes early growth response 1 (egr1) and c-fos. Deletion or reduction of neuroLSD1 in mutant mice translates into decreased levels of activating histone marks at egr1 and c-fos promoters, dampening their psychosocial stress-induced transcription and resulting in low anxiety-like behavior. Administration of suberoylanilide hydroxamine to neuroLSD1^KO mice reactivates egr1 and c-fos transcription and restores the behavioral phenotype. These findings indicate that LSD1 is a molecular transducer of stressful stimuli as well as a stress-response modifier. Indeed, LSD1 expression itself is increased acutely at both the transcriptional and splicing levels by psychosocial stress, suggesting that LSD1 is involved in the adaptive response to stress.

Dynamic changes in neuronal chromatin through histone posttranslational modifications affect complex functions such as learning, memory, and emotional behavior (1). Seminal studies have shown that mice experiencing different forms of stress, including psychosocial stress, promote stress-related plasticity through epigenetic changes at specific genes, including brain-derived neurotrophic factor (BDNF) and immediate early genes (IEGs) (2–4). These modifications induce contrasting structural and functional changes in the hippocampus and the amygdala (5), brain areas responsible for the expression of anxiety-like behavior (5–8). A decrease in neural activity in the hippocampus caused by the loss of dendritic arbors and spines is associated with posttraumatic stress disorder and recurrent depressive illness (5). Therefore, an important challenge for molecular psychiatry is a better understanding of the epigenetic regulation of plasticity gene transcription in response to stress (9).

Lysine-Specific Demethylase 1 (LSD1) also known as lysine demethylase 1A (KDM1A) is an epigenetic transcriptional corepressor, tightly associated to Corepressor of REST (CoREST) and histone deacetylase 2 (HDAC2). It removes methyl groups from mono- and di-methylated lysine 4 of histone H3 (H3K4), erasing a histone mark of active transcription (10). In mammals, neurospecific splicing of microexon E8a generates the dominant-negative splicing isoform of LSD1 (neuroLSD1), which is required for the acquisition of proper neurite morphology inherent in neuronal maturation (11). Although conventional LSD1 acts as a constitutive repressor through its H3K4 demethylase activity, neuroLSD1 is unable to repress transcription (11, 12). It has been shown recently that neuroLSD1 lacks dimethylated H3K4 (H3K4me2) demethylase activity, confirming its role as a dominant-negative LSD1 isoform (13). The neuroLSD1-containing complex also can act as transcriptional activator by acquiring H3K9me2 and H4K20 demethylase activity (13, 14). In this context, the finding that LSD1 and neuroLSD1 share a subset of target genes (12–14) suggests that modulation of exon E8a splicing may represent a mechanism for fine-tuning the transcription of selected targets. The generation of a mouse model lacking microexon E8a (neuroLSD1^KO mice) allowed us to characterize neuroLSD1 as an in vivo modulator of hippocampal excitability. We reported that shifts in the LSD1/neuroLSD1 ratio in response to activity contribute dynamically to the control of neuronal excitability by participating in the transcriptional mechanism of homeostatic adaptation (15). In this work we implicate LSD1 and neuroLSD1 in shaping emotional behavior as fine tuners of psychosocial stress-evoked transcription of IEGs. We found that the complete lack (neuroLSD1^KO) or reduction (neuroLSD1 heterozygosity, hereafter neuroLSD1^HET) of neuroLSD1 in the mouse brain causes low anxiety-like behavior. At the molecular level, knocking out neuroLSD1 results in H3K4 hypomethylation and H3 hypoacetylation of early growth response 1 (egr1) and c-fos promoters, decreasing their stress-induced transcription. This impairment hinders the acquisition of stress-related plasticity relevant to anxiety behavior. Moreover, in wild-type animals, LSD1 transcription and alternative splicing are directly modulated by psychosocial stress, indicating that LSD1 and neuroLSD1 not only participate in the transduction of stressful stimuli but also represent stress-response modifiers.

Results

NeuroLSD1-Mutant Mice Display a Low-Anxiety Phenotype.

To test the effect of neuroLSD1 ablation on anxiety-like behavior, we tested neuroLSD1^KO and heterozygous mice (Fig. S1) (15) using three classic approach–avoidance anxiety paradigms (Fig. 1). In the elevated plus maze (EPM), we observed that neuroLSD1^KO mice spent more time in and made more entries into the open arms than their wild-type littermates (Fig. 1A). A trend for increased time and number of entries in the open arms also was observed in neuroLSD1^HET mice. The total number of entries into each arm of the maze was comparable in all genotypes, and no differences in motor function were observed (Fig. S2). In the marble-burying test (MBT), post hoc analysis indicated a significant decrease in the number of buried marbles in both neuroLSD1^KO and heterozygous mice compared with wild-type littermates (Fig. 1B). Both genotypes showed a trend to increased latency to the first burial (Fig. 1B). In the novelty-suppressed feeding (NSF) test, both neuroLSD1^KO and neuroLSD1^HET mice exhibited significantly decreased latency to feed in the center of the arena compared with wild-type littermates. No difference was found in the food intake among genotypes (Fig. 1C). Taken together, the results of behavioral tests consistently show that complete lack (neuroLSD1^KO mice) or reduction (neuroLSD1^HET mice) of neuroLSD1 expression affects emotional behavior, resulting in a low-anxiety phenotype.

Fig. S1.

(A) Schematic representation of the mouse LSD1 gene. (B) The strategy used to remove exon E8a, generating a mutated allele including a neomycin gene cassette. The mutated allele without neomycin is the result of a cre-driven recombination event. The two genotyping primers in the mutated allele without neomycin are indicated by arrows and referred to as genotyping primers in SI Methods. (Adapted from ref. 15.)

Fig. 1.

NeuroLSD1-mutant mice feature a low anxiety-like phenotype. Anxiety-like behavior is evaluated by comparing neuroLSD1^KO and neuroLSD1^HET mice with wild-type littermates. The anxiety trend is shown on the y axis. (A) Time spent in the open arm time (F_2,27 = 4.65, P = 0.018), entries into the open arm (F_2,27 = 6.16, P = 0.0063), and total entries (which did not show any difference among genotypes) evaluated in the EPM. (B) Number of marbles buried (F_2,27 = 10.06, P = 0.0002) and latency to the first burial (F_2,27 = 2.21, P = 0.12) evaluated in the MBT. (C) Latency to feeding (F_2,27 = 9.95, P = 0.0003) and total amount of food intake evaluated during the NSF test (n = 8–10 mice per genotype). Data are presented as mean ± SEM; *P < 0.05, **P < 0.01, ***P < 0.001, one-way ANOVA, Tukey post hoc test.

Fig. S2.

Motor activity does not change among the genotypes. Time course of locomotor activity evaluated for 120 min in terms of number of counts in wild-type, heterozygous, and LSD1-mutant mice (KO). (A, Left) Horizontal counts. (Right) Corresponding calculated area under the curve (AUC). (B, Left) Vertical counts. (Right) Corresponding calculated area under the curve. n = 10 for each group. Results are shown as mean ± SEM.

LSD1 and NeuroLSD1 Are Modulated by Stress and Regulate Stress-Evoked Transcription of IEGs.

To shed light on the molecular mechanism whereby reduction of neuroLSD1 gives rise to the low anxiety-like phenotype, we investigated the role of LSD1 in transducing stressful stimuli into a transcriptional response in the hippocampus. We tested whether stress could modify LSD1 expression and the LSD1/neuroLSD1 splicing ratio in the hippocampus. To address this point, we took advantage of social defeat stress (SDS) paradigm (16), a psychosocial stress protocol that has been well validated with regard to face validity, ability to challenge the anxiety-like phenotype, and proven usefulness in unraveling cellular/molecular mechanisms of affective behaviors. We applied a single session of SDS and measured LSD1 expression and alternative splicing. In the hippocampus of wild-type stressed mice we found an increase of about 30% in total LSD1 protein levels (Fig. 2A) and a twofold increase in total LSD1 mRNA (Fig. 2B). Moreover, analysis of LSD1 splicing by relative quantity fluorescent PCR (rqf-PCR) showed that the inclusion of LSD1 exon E8a was reduced in stressed mice, resulting in the down-regulation of the neuroLSD1 isoform and a relative increase in LSD1 (Fig. 2C). These data indicate that in the mouse hippocampus stress induces an acute increase in LSD1 transcription as well as a change in its splicing pattern. To investigate the role of LSD1 and neuroLSD1 in the molecular pathway of stress transduction that is functionally relevant to the shaping of anxiety-like behavior, we further analyzed psychosocial stress-dependent IEG transcription by comparing the transactivation of egr-1 and c-fos in neuroLSD1^HET mice and wild-type littermates. The expression of these two plasticity genes is known to mediate the translation of specific stimuli to generate inherent neuronal plasticity (4). Upon SDS, we found that Egr1 and C-Fos mRNA were transactivated and protein levels increased only in wild-type hippocampi; in neuroLSD1^HET animals these IEGs did not respond to stress (Fig. 2 D–G). These data indicate that a proper amount of neuroLSD1 is necessary to induce IEG transcription efficiently in response to psychosocial stress and thus suggest that LSD1 and neuroLSD1 have a role as stress-response transducers.

Fig. 2.

Hippocampal LSD1 and neuroLSD1 levels are modified in response to psychosocial stress and modulate stress-induced transcription of IEGs. Mice underwent a single session of the SDS test. A schematic representation of the experimental procedure is shown at the top of the figure. (A–C) Total LSD1 protein by Western blot (A), LSD1 total mRNA by qPCR (B), and LSD1/neuroLSD1 relative percentage by rqfRT-PCR (C) in the hippocampi of C57BL/6N wild-type mice. A schematic representation of the four LSD1 splicing isoforms is shown on the right (n = 5 mice per group). *P < 0.05, **P < 0.01, ***P < 0.001, Student t test. (D–G) SDS-induced transcription and protein expression of egr1 and c-fos in in the hippocampi of wild-type and neuroLSD1^HET mice challenged with SDS. (D and E) mRNA analysis of egr1 (treatment: F_1,44 = 3.561, P = 0.0657; genotype: F_1,44 = 6.190, P = 0.0167; treatment × genotype: F_1,44 = 1.339, P = 0.2535) (D) and c-fos (treatment: F_1,44 = 8.865, P = 0.0047; genotype: F_1,44 = 4.586, P = 0.0378; treatment × genotype: F_1,44 = 2.237, P = 0.1419) (E) (n = 10–14 mice per condition). (F and G) Western blot protein analysis of Egr1 (treatment: F_1,13 = 6.542, P = 0.0238; genotype: F_1,13 = 6.612, P = 0.0232; treatment × genotype: F_1,13 = 26,18, P = 0.0002) (F) and C-Fos (treatment: F_1,13 = 8.900, P = 0.0175; genotype: F_1,13 = 8.900, P = 0.0175; treatment × genotype: F_1,13 = 23.69, P = 0.0012) (G) (n = 3 or 4 mice per condition). Results are shown as mean ± SEM; *P < 0.05, **P < 0.01, ***P < 0.001, two-way ANOVA, Bonferroni post hoc test.

LSD1 Is a Serum Response Factor Corepressor of IEGs.

Transcription of egr1 and c-fos has been reported to be induced by SDS in mouse hippocampus (17) and to be related to the behavioral outcome of SDS in terms of stress susceptibility (4). Serum response factor (SRF), together with CREB, is an important transactivator of egr1, c-fos, and other IEGs (18, 19). We tested both SRF and cAMP response element-binding protein (CREB) for their ability to interact with LSD1. Using anti-SRF antibody, we coimmunoprecipitated LSD1 and the molecular partners HDAC2 and CoREST from mouse hippocampus protein extracts under resting conditions (Fig. 3A). Vice versa, using anti-panLSD1 antibody, we coimmunoprecipitated SRF (Fig. 3B). Analysis of the anti-SRF immunoprecipitates from HeLa cells stably expressing HA-LSD1 or HA-neuroLSD1 showed that both isoforms associate with endogenous SRF protein (Fig. 3C). Under the same conditions, we could not demonstrate molecular interaction between LSD1 and CREB (Fig. S4), suggesting that LSD1 modulation of IEGs could be mediated by SRF. Then we analyzed LSD1, HDAC2, and SRF chromatin occupancy at the egr1 and c-fos promoters in wild-type and neuroLSD1^KO mice. Both promoters are known to contain serum-responsive elements (SREs) (20). In particular, five SREs cluster within the egr1 promoter region, whereas a single SRE maps at the c-fos promoter. As shown in Fig. 3 D and E, quantitative ChIP (qChIP) revealed significant enrichment of LSD1 and HDAC2 together with SRF at egr1 and c-fos proximal promoter regions compared with preimmune IgG in both wild-type and KO mouse hippocampi. We did not observe differences between wild-type and neuroLSD1^KO mice in terms of SRF and LSD1 occupancy at each promoter, but we measured a significant increase in HDAC2 enrichment at the level of egr1 in neuroLSD1^KO mice. SRF is constitutively bound to its targets (21). Consistently, we did not observe psychosocial stress-induced modification of the association of SRF with the egr1 and c-fos promoters (Fig. S5).

Fig. 3.

LSD1, CoREST, and HDAC2 are previously unidentified SRF corepressors. (A and B) Hippocampus protein extracts immunoprecipitated with anti-SRF antibody (A) or with anti-LSD1 (B) were separated by SDS/PAGE and detected with the indicated antibodies. In addition to the 67-KDa band in A and B, the SRF antibody detected a 57-KDa band described as a neurospecific splicing isoform (28) (also see Fig. S3). (C) HeLa cells overexpressing HA-LSD1 or HA-neuroLSD1 immunoprecipitated with anti-SRF antibody were separated by SDS/PAGE and were immunodecorated with anti-HA. (D–F) qChIP experiments performed on hippocampal chromatin using anti-SRF, anti-panLSD1, and anti-HDAC2 antibodies together with IgG as mock treatment (dashed lines represent the highest mock treatment of the two genotypes) on egr1 (D) and c-fos (E) promoters and an unrelated control (the egr1 distal genomic region) (F). Data are shown as mean ± SEM; *P = 0.035, Student t test. (G and H) In neurons, LSD1/neuroLSD1 regulation of IEGs’ activity-dependent transcription requires SRF binding to DNA. (G) Reporter assay in primary wild-type and neuroLSD1^HET hippocampal neurons transfected with the pGL3-egr1(−370) construct (treatment: F_1,28 = 34.74, P < 0.0001; genotype: F_1,28 = 2.459, P = 0.1281; treatment × genotype: F_1,28 = 5.384, P = 0.0278) and the mutated version lacking the five SREs, pGL3-egr1(−370m) (treatment × genotype: F_1,20 = 0.2209, P = 0.6435). (H) Reporter assay in primary wild-type and neuroLSD1^HET hippocampal neurons transfected with the pGL3-egr1(−370) construct and the pCGN-HA vector (treatment: F_1,11 = 207.2, P < 0.0001; genotype: F_1,11 = 5.319, P = 0.0416; treatment × genotype: F_1,11 = 7.116, P = 0.0219) or pCGN-HA-neuroLSD1 (treatment × genotype: F_1,11 = 0.6388, P = 0.4411). Reporter activity was assayed in basal conditions and after treatment with BDNF. (n = 3–8 per condition). Results are shown as mean ± SEM; *P < 0.01, two-way ANOVA, Bonferroni post hoc test.

Fig. S4.

LSD1 does not interact with CREB. Mouse hippocampus protein extracts were immunoprecipitated with anti-CREB or preimmune IgG, separated on SDS/PAGE, and immunodecorated with anti-panLSD1 antibody and anti-CREB antibody.

Fig. S5.

SDS does not affect SRF association with target promoter regions. Histograms show SRF enrichment in wild-type mice and neuroLSD1^HET mutants in resting conditions (CTRL) and upon SDS at the level of the egr1 and c-fos proximal promoter regions including the SRE. As a negative control we also measured SRF enrichment at the level of an egr1 distal promoter region lacking SRE (unrelated region). (n = 4 mice per condition). Results are shown as mean ± SEM.

Fig. S3.

Western blot analysis showing that the additional 57-kDa SRF-related band is present only in hippocampal protein extracts and not in a nonneuronal cell line (HeLa). Total protein lysates from HeLa cells (lane1) and mouse hippocampus (lane2) were separated on SDS/PAGE and immunodecorated with anti-SRF antibody and anti-GAPDH antibody.

To determine the role of SRF in LSD1/neuroLSD1 modulation of activity-dependent IEG transcription (Fig. 2 D and E), we investigated whether SRF binding to its target sequences was required for LSD1 and neuroLSD1 to regulate IEG transcription. To this aim, we analyzed the effect of reduced levels of neuroLSD1 on BDNF-induced transactivation of a wild-type egr1 reporter gene (22, 23) or a mutated version lacking the five SRE, using previously described pGL3-egr1(−370) and pGL3-egr1(−370m), respectively (20). The two reporter plasmids were transfected into hippocampal neurons derived from either wild-type or neuroLSD1^HET mice. The egr1 reporter gene was activated by BDNF when transfected in wild-type neurons, but transactivation was less efficient in heterozygous neurons. Remarkably, a clear modulatory effect played by the LSD1/neuroLSD1 ratio on reporter transactivation can be evidenced only when SRF binds to SRE. Indeed, BDNF-mediated transactivation did not occur in either of the two cellular contexts when the egr1 reporter gene carried the SRE mutation pGL3-egr1(−370m) (Fig. 3G). Importantly, BDNF-mediated transactivation of the wild-type egr1 reporter gene was recovered in neurons derived from heterozygous mice when HA-neuroLSD1 was exogenously overexpressed (Fig. 3H). These data support the importance of the correct amount of neuroLSD1, in association with SRF, in eliciting IEG transactivation in response to stimuli instrumental to the acquisition of inherent forms of neuroplasticity.

NeuroLSD1-Mutant Mice Display Altered IEG Chromatin Structure.

SRF is comparably associated with IEG promoters in wild-type and neuroLSD1^KO mice (Fig. 3 D and E). This finding suggests that, rather than being caused by differential SRF association with DNA, the impaired psychosocial stress-induced IEG transcription observed in neuroLSD1-mutant mice could result from altered chromatin structure at IEG promoters. We analyzed chromatin structure in wild-type and neuroLSD1-mutant mice by assessing their level of H3K4 methylation and H3 acetylation (H3K9/K14ac), histone marks erased by LSD1 and HDAC2. We applied qChIP to whole mouse hippocampus and found that in both neuroLSD1^KO and neuroLSD1^HET mice H3K4me2 and H3K9/K14ac were decreased at the egr1 and c-fos proximal promoter regions, including SRE, compared with wild-type littermates (Fig. 4 A and B). No differences were observed among genotypes at an unrelated control region devoid of SRE (Fig. 4C). This specific chromatin compaction primed by LSD1-induced H3K4 hypomethylation, but also based on H3 hypoacetylation, produces minor modifications of IEGs’ basal transcription (CTRL in Fig. 2 D and E) but may account for the defective transactivation of egr1 and c-fos measured in neuroLSD1^HET mice in response to psychosocial stress (SDS in Fig. 2 D and E).

Fig. 4.

NeuroLSD1^KO and neuroLSD1^HET hippocampi display reduced levels of H3K4 dimethylation and H3K9/K14 acetylation at the egr1 and c-fos promoters. Hippocampal chromatin from neuroLSD1^KO or neuroLSD1^HET mice or wild-type littermates was immunoprecipitated with anti-H3K4me2 and anti-H3K9/K14ac antibodies. Enrichments are shown as percentage of input. Shown is qPCR analysis at the level of egr1 (anti-H3K4me2: F_2,23 = 14.07, P = 0.0001; anti-H3K9/14ac: F_2,15 = 1.667, P = 0.2220) (A) or c-fos (anti-H3K4me2: F_2,28 = 19.74, P < 0.0001; anti-H3K9/14ac: F_2,22 = 4.379, P = 0.0251) (B) proximal promoters and the unrelated control (egr1 distal genomic region) (C) (n = 5–13 per condition). Data are shown as mean ± SEM; *P < 0.05; **P < 0.01; ***P < 0.001, one-way ANOVA, Tukey post hoc test; ^§P = 0.0269, Bartlett’s test.

The Low Anxiety-Like Phenotype of NeuroLSD1-Mutant Mice Is Related to Decreased Neuroplasticity.

NeuroLSD1 genetic ablation affects the psychosocial stress-dependent transcription of egr1 and c-fos through direct LSD1-mediated chromatin compaction, suggesting reduced neuroplasticity. We reasoned that if the low-anxiety behavior of neuroLSD1-mutant mice is related to this peculiar molecular phenotype, reactivation of IEG transcription to reinstate neuroplasticity by systemic HDAC inhibition (24, 25) should restore normal anxiety. We systemically treated mice with a clinically approved HDAC inhibitor (HDACi), suberoylanilide hydroxamic acid (SAHA), 25 mg/kg i.p. for 10 d Fig. S6), to overcome IEGs’ repressive chromatin structure observed in neuroLSD1 mutants. We chose SAHA, which is able to cross the blood–brain barrier (24), for its established ability to increase hippocampal plasticity by promoting the transcription of plasticity-related genes, including egr1 and c-fos (24, 26). We found that in resting conditions SAHA treatment induced sustained levels of egr1 and c-fos transcription in both wild-type and neuroLSD1^HET mice (SAHA-treated CTRL in Fig. S7). These levels were comparable to (or higher than) those observed in wild-type mice exposed to psychosocial stress (SDS in Fig. S7), and c-fos levels were further increased by SDS (SAHA-treated SDS in Fig. S7). Importantly, SAHA treatment did not affect the association of SRF with its chromatin targets (Fig. S8). To analyze the effect of SAHA treatment on the anxiety-like behavior of neuroLSD1^KO and wild-type mice, we used the paradigms described earlier. Remarkably, SAHA significantly increased anxiety-like behavior in neuroLSD1^KO mice as evidenced by decreased time spent in the open arms and the decreased number of entries into the open arm as compared with vehicle-treated neuroLSD1^KO mice evaluated in the EPM (Fig. 5A). Post hoc analysis revealed that in neuroLSD1^KO mice SAHA restored anxiety-like behavior to wild-type levels. Similarly in the MBT the two parameters—reduced number of marbles buried and increased latency to first burial proper of neuroLSD1^KO mice—were restored to wild-type levels by SAHA but not by vehicle treatment (Fig. 5B). In the NSF test, SAHA increased the latency to feeding of neuroLSD1^KO mice compared with vehicle-treated neuroLSD1^KO mice (Fig. 5C). It is important to note that SAHA treatment did not exert a measurable effect on the behavior of wild-type animals, although it induced up-regulation of basal IEG transcription (Fig. S7). In neuroLSD1-mutant mice, poor IEG transcription in response to stimuli results from the LSD1/HDAC2-mediated decrease in positive histone marks at IEG proximal promoters (Fig. 4 A and B). Interestingly, SAHA-treated neuroLSD1^KO mice exhibited a significant increment in H3K9/K14ac and H3K4me2 compared with vehicle-treated mice at both the egr1 and c-fos promoters (Fig. 5 D and E). The observation that the same loci were not affected by SAHA in wild-type mice suggests that the increased transcription of IEGs should be ascribed, at least in part, either to modifications of different histone marks at regulatory regions targeted by SRF/LSD1/HDAC2 or to changes at genomic regions not targeted by SRF, as would be expected, given the broad effect of SAHA on the overall acetylation of both histone H3 and H4 (24). We cannot exclude the possibility that SAHA-mediated increase of neuroplasticity in other anxiety-related brain structures could contribute to the normalization of anxiety in neuroLSD1 mutants.

Fig. S6.

Treatment with SAHA (25 mg/kg) or vehicle (VEH) for 10 d does not affect the body weight of wild-type or neuroLSD1^KO mice. Results are shown as mean ± SEM.

Fig. S7.

Treatment with SAHA restores egr1 and c-fos transcription in neuroLSD1 mutants. The experimental procedure is summarized graphically at the top of the figure. egr1 (A) and c-fos (B) mRNA levels in SAHA-treated wild-type and neuroLSD1^HET mice in resting conditions (CTRL) and exposed to SDS. For egr1, treatment: F_1,20 = 0.2393 P = 0.6300; genotype: F_1,20 = 0.4795, P = 0.4966; treatment × genotype interaction: F_1,20 = 0.7077, P = 0.4102. For c-fos, treatment: F_1,19 = 18.35, P = 0.0004; genotype: F_1,19 = 3.721, P = 0.0688; treatment × genotype interaction: F_1,19 = 3.923, P = 0.0623. As reference, egr1 and c-fos mRNA levels of CTRL and SDS mice that did not receive SAHA are reported also (data are from Fig. 2D in the main text).

Fig. S8.

SAHA treatment does not affect SRF association with target promoter regions. Histograms show SRF enrichment at the level of the egr1 and c-fos proximal promoter regions including the SRE in wild-type mice and neuroLSD1^HET mutants administered vehicle and upon chronic systemic SAHA treatment. Results are shown as mean ± SEM.

Fig. 5.

Treatment with SAHA normalizes the anxiety-like phenotype of neuroLSD1 mutants. (A–C) Anxiety-like behavior in neuroLSD1^KO mice compared with wild-type mice treated with SAHA or vehicle (VEH). The anxiety trend is shown. (A) Entries into the open arm (treatment: F_2,54 = 1.847, P = 0.1676; genotype: F_1,54 = 19.70, P = 0.001; treatment × genotype: F_2,54 = 3.643, P = 0.032) and time spent in the open arm (treatment: F_2,54 = 5.43, P = 0.007; genotype: F_1,54 = 6.979, P = 0.0108; treatment × genotype: F_2,54 = 6.969, P = 0.002) evaluated in the EPM. (B) The number of buried marbles (treatment: F_2,54 = 4.436, P = 0.0165; genotype: F_1,54 = 22.35, P < 0.0001; treatment × genotype: F_2,54 = 4.566, P = 0.0147), latency to first burial (treatment: F_2,54 = 5.45, P = 0.006; genotype: F_1,54 = 2.285, P = 0.1365; treatment × genotype: F_2,54 = 7.834, P = 0.001) evaluated in the MBT. (C) Latency to the first bite (treatment: F_2,54 = 0.08, P = 0.92; genotype: F_1,54 = 10.29, P = 0.0023; treatment × genotype: F_2,54 = 6.26, P = 0.004) evaluated in the NSF. (n = 8–10 mice per genotype for each condition). (D–F) Hippocampal chromatin from neuroLSD1^KO or wild-type mice treated with SAHA or vehicle and immunoprecipitated by anti-H3K4me2 and anti-H3K9/K14ac antibodies. Enrichment at specific loci is shown as percentage of input. (D) egr1 promoter (anti-H3K4me2, treatment: F_1,16 = 2.141, P = 0.1628; genotype: F_1,16 = 1.4405, P = 0.2468; treatment × genotype: F_1,16 = 9.080, P = 0.0082) (anti-H3K9/K14, treatment: F_1,14 = 0.0429, P = 0.8388; genotype: F_1,14 = 2.376, P = 0.1455; treatment × genotype: F_1,14 = 2.723, P = 0.1212; genotype). (E) c-fos promoter (anti-H3K4me2, treatment: F_1,16 = 11.57, P < 0.0036; genotype: F_1,16 = 3.018, P = 0.1015; treatment × genotype: F_1,16 = 2.938, P = 0.1058) (anti-H3K9/K14, treatment F_1,13 = 3.438, P < 0.0865; genotype: F_1,13 = 0,2726, P = 0.6104; treatment × genotype: F_1,13 = 0.3492, P = 0.5647). (F) Unrelated control (egr1 distal genomic region) (n = 3–6 per condition). Datasets referred to as “VEH” in D–F are different from those in Fig. 4 in which naive mice were used. Data are shown as mean ± SEM; *P < 0.05, **P < 0.01, ***P < 0.001, two way ANOVA, Bonferroni post hoc test; ^§P < 0.05, Student t test.

Discussion

In this work we shed light on the molecular mechanisms underlying gene–environment interactions in the mouse brain. We propose that epigenetic enzyme LSD1 and its dominant negative isoform neuroLSD1 serve as transcriptional modulators shaping the chromatin-encoded memory of stressful experiences. In particular, we suggest that the splicing-regulated dual system LSD1/neuroLSD1 can act as a fine transduction tuner of psychosocial stress by mediating transcriptional modulation of plasticity genes. We discovered that neuroLSD1 mutants’ low-anxiety behavior is related to a defective stress response, i.e., deficient psychosocial stress-induced transactivation of egr1 and c-fos. Stress is known to affect the plasticity and excitability of the hippocampus (27). Together with the amygdala and prefrontal cortex, the hippocampus is part of the complex circuitry involved in mammalian trait anxiety, and its substantial role has been acknowledged (5, 7, 8, 28). We show that in the hippocampus, LSD1/neuroLSD1 repression of plasticity genes can be exerted in association with the transcription factor SRF. However, we cannot exclude the possibility that transcription factors other than SRF could contribute to recruiting LSD1/CoREST/HDAC2 to the IEG promoters. The association of LSD1 and neuroLSD1 with SRF is consistent with its ability to regulate the same targets either as a transcriptional activator in response to stimuli or as a repressor in basal conditions (19). SRF’s repressive function has been described extensively (29); this work contributes to our understanding by characterizing the molecular basis of SRF’s repressive activity. Interestingly, SRF is known to be constitutively bound to the DNA of its target genes both in resting conditions and upon stimulation (21), and its association with IEG promoters is not affected by psychosocial stress. Thus, the difference in the proneness of egr1 and c-fos to transcription observed in wild-type versus neuroLSD1-mutant mice depends mainly on differences in the LSD1/neuroLSD1 ratio, resulting in changes in the H3K4me2 levels, in the two genotypes. When the dominant-negative neuroLSD1 isoform, lacking H3K4 demethylase activity, is low (neuroLSD1^HET) or absent (neuroLSD1^KO), the LSD1 isoform is increased or is the only one expressed. Its repressive function toward egr1 and c-fos transcription is increased aberrantly, establishing a repressive chromatin state that limits the extent to which transactivation of these genes can occur in response to psychosocial stress (Fig. 6). Elevated LSD1 levels not only cause H3K4 hypomethylation but also facilitate HDAC deacetylase activity, as indicated by decreased H3K9/K14ac levels observed at the same promoters. In vitro, interdependence of LSD1 and HDAC activities has been described extensively (30), and our findings further demonstrate this functional interplay in the brain.

Fig. 6.

Graphical model of SRF-LSD1/neuroLSD1–mediated transcriptional modulation of the IEGs. Given the dimeric nature of the corepressor complex, it is conceivable that, depending on the relative amounts of LSD1 and neuroLSD1, both LSD1/LSD1 homodimers and LSD1/neuroLSD1 heterodimers could be present in vivo (11). (A) In wild-type mice, in resting conditions, IEG transcription is repressed but permits activity-induced transcription. (B) Upon stressful stimuli, IEG transcription is fully activated. (C and D) In neuroLSD1-mutant mice, in resting conditions (C), a more condensed chromatin structure at the IEG proximal promoters does not permit stress-induced transcription (D). (E and F) Chronic treatment with the HDAC inhibitor SAHA leads to normalization of anxiety in neuroLSD1 mice.

A central finding of this study is the discovery that in wild-type animals, in physiological conditions, LSD1 is increased and neuroLSD1 is decreased in response to stress. In heterozygous mice, the genetically produced reduction of neuroLSD1 is able per se to impact IEG transactivation negatively in response to stress. Thus, stress-induced neuroLSD1 down-regulation should be sufficient to buffer the stress-driven transcription of plasticity genes in a wild-type animal. This modulation could represent a possible homeostatic mechanism able to mitigate intense or prolonged stress-induced molecular responses and thereby maintain acquired plasticity at an adaptive level, although immediately after the SDS we did not observe a readout on chromatin (Fig. S9). Chronic stress results in long-lasting chromatin modifications that increase the repression of plasticity-related genes (2–4). In particular, egr1 has been reported to be down-regulated in mice stressed by chronic social defeat and in postmortem brains of depressed human subjects, suggesting a pathogenic role for this down-regulation (4, 31). In this regard, it is conceivable that repeated LSD1 up- and neuroLSD1 down-regulation in response to chronic stress could induce H3K4 hypomethylation at the egr1 promoter (Fig. S10). We have shown previously that dendrite shrinkage can be produced in cultured neurons by knocking down neuroLSD1 (11). The decrease of neuroLSD1 in the hippocampus that we observed in response to stress might represent an epigenetic mechanism instrumental in the shrinkage of dendritic arbors and loss of dendritic spines seen in stress-related psychiatric disorders (32).

Fig. S9.

SDS-induced modification of H3K4me2 and H3K9/14ac. Hippocampal chromatin from C57BL/6 wild-type mice was immunoprecipitated by anti-H3K4me2 and anti-H3K9/14ac antibodies or IgG (not shown). Enrichments are shown as percentage of the input. qPCR analysis at the level of egr1 or c-fos or an unrelated regulatory region. (n = 8–10 mice per condition). Results are shown as mean ± SEM. *P < 0.01, Student t test. We did not detect a parallel H3K4 hypomethylation in the same time window in which we measured neuroLSD1 stress-induced down-regulation. This lack of immediate modification of the histone marks could be explained by the different kinetics of chromatin-based adaptations: A longer time interval after the cease of stressful conditions might be required to appreciate the effect of LSD1/neuroLSD1 stress on chromatin compaction in the IEG promoter.

Fig. S10.

A working model graphically showing the pathogenic perspective of chronic stress-induced neuroLSD1 down-regulation. (A) In normal conditions, the balance between the two LSD1 isoforms guarantees a chromatin environment that permits experience-induced transcription of plasticity-related genes. (B) Upon exposure to chronic stress, repeated neuroLSD1 down-regulation could translate into a chromatin compaction that hinders activity-evoked egr1 and c-fos transactivation. This molecular adaptation/maladaptation could contribute to the atrophy of apical hippocampal dendrites observed in stress-related psychiatric disorders.

SAHA treatment successfully restored IEG transcription and at the same time led to a phenotypical normalization of neuroLSD1 mutants’ emotional behavior, suggesting that the low anxiety-like phenotype is related to reduced neuroplasticity. As expected, we also observed an SAHA-mediated increase in H3K9/K14ac and H3K4me2, histone marks specifically affected in neuroLSD1 mutants. However, because of the broad SAHA-induced chromatin remodeling (24), other effects on histone and nonhistone proteins are likely also. All our results are consistent with the idea that pharmacological targeting of LSD1 demethylase activity through specific inhibitors or by direct neurospecific splicing modulation, in combination with or as an alternative to HDACi, could represent a strategy for treating mood and anxiety disorders, including depression (33, 34). HDACi have been proposed to exert antidepressant and anxiolytic effects (34). Therefore, the SAHA-mediated restoration of anxiety in neuroLSD1-mutant mice to wild-type levels highlights an apparently counterintuitive effect of SAHA on emotional behavior. However, similar bidirectional effects of HDACi have been described in regard to memory. Indeed, SAHA and other HDACi are able to improve memory by increasing synaptic plasticity in the hippocampus (24), but they also can attenuate remote fear memories by the same molecular mechanism (25). In both these contexts, it has been suggested that HDACi act by reactivating plasticity-related genes, priming the hippocampal transcriptome to reinstate neuroplasticity (25). In conclusion we suggest that specific amounts of neuroLSD1 are necessary to couple stress exposure with its molecular response, i.e., in the expression of plasticity genes. It is tempting to speculate that neuroLSD1 could enhance mammals’ ability to survive in a complex and ever-changing environment by finely tuning the molecular machinery that memorizes harmful events through stress-related plasticity.

Methods

All procedures involving animals were carried out in accordance with the European regulations (2010/63/UE) and Italian Council on Animal Care Guidelines (Legislative Decree no. 26, 2014) and were approved by the Italian Ministry of Health (no. 234/2012B and 275/2015) and by University of Milan IRB no. 27/2014. See SI Methods for further details of SDS procedures (16), the anxiety test (35–37), gene-expression analysis through quantitative PCR (qPCR) and rqf-PCR (11), ChIP protocols (15), primers, and antibodies.

SI Methods

Protein Extraction and Western Blotting.

Tissues or cultured cells were homogenized essentially as described (11). The antibodies used are listed below.

ChIP.

A detailed description of the ChIP methodology is given in ref. 15. The antibodies and primers used are listed below.

Total RNA Extraction, qPCR Analysis, and rqf-PCR.

Total RNA was isolated from hippocampal extracts using TRIzol reagent (Sigma-Aldrich), and the purified RNA was treated with the RNase-free DNase set (Qiagen) to remove any residual DNA. qRT-PCR analysis was performed as previously described (11). The relative expression of the investigated genes was quantified after normalization against ribosomal protein SA (RPSA). Exon E8a splicing inclusion was measured by rqf-PCR as described (11). Primers are listed below.

Hippocampal Neuronal Cultures.

Primary hippocampal neurons were prepared from embryonic day 18.5 mouse brains (12) and were placed on coverslips coated with poly-d-lysine at a density of 75,000 neurons per well. At day in vitro (DIV) 12 the cultures were transfected using the calcium phosphate method and were analyzed at DIV15. Human BDNF (Miltenyi Biotech GmbH) was administered at 10 ng/mL for 8 h.

Reporter Gene Assays.

pGL3-egr1 and pGL3-egr1–mutant reporter plasmids (20) were used. Control experiments were carried out using equivalent molar amounts of pGL3-basic reporter. DNA was buffered by pUC8. The luciferase reporter activity was determined using the Dual-Luciferase reporter assay system (Promega) according to the manufacturer’s instructions. The activity of each construct was expressed normalized to pGL3 basic-related signals (both expressed as Luciferase/Renilla luminescent units).

Experimental Animals.

Ten-week-old male C57BL/6N mice and neuroLSD1 knock-out (neuroLSD1^KO), heterozygous (neuroLSD1^HET), or wild-type littermates mice on the C57BL/6N genetic background were used (15) upon cre-mediated removal of the neomycin cassette (Fig. S1). Mice were housed individually throughout the testing period with free access to food and water at controlled temperature (20–22 °C) with a 12-h light/dark cycle (lights on at 7:00 AM). SAHA (Vorinostat), obtained from Sigma-Aldrich, was administered i.p. at a dose of 25 mg/kg for 10 d. All experimental procedures involving the use of rodents followed the guidelines established by the Italian Council on Animal Care (Legislative Decree no. 26, March 2014) and European regulations (2010/63/UE) and were approved by Italian Ministry of Health (no. 234/2012B and 275/2015). Every effort was made to minimize the number of mice used and their suffering. All current rules and regulations regarding animal treatment were followed strictly.

Behavioral Testing.

The EPM test was conducted as described by Godar et al. (36). The number of entries into open and closed arms and the time spent in the open arms were recorded and expressed as percentages (open entries/total entries × 100 and open time/total time × 100). The total arm entries were analyzed as measures of nonspecific changes in locomotor activity.

The MBT was performed using a method similar to that of Njung’e et al. (35). Twenty glass marbles (15-mm diameter) in 5 cm of sawdust were spaced evenly in a 4 × 5 grid on the surface of the bedding. During the testing phase each mouse was placed in the cage and allowed to explore for 15 min. A marble was scored as buried if more than two-thirds of it was covered with sawdust.

In the NSF test animals were food-deprived for 24 h before the test, which was performed according to Mosienko et al. (37). The latency to the first eating episode (the time between the introduction of the mouse into the arena with a food pellet in the middle and the first feeding event) was used as an index of induced anxiety-like behavior. The amount of food consumed in the home cage during the subsequent 5-min free-feeding provided a measure of appetitive drive.

Single Session of Psychosocial Defeat Stress.

A modified protocol of the SDS test (16) was used. Briefly, CD1 retired breeder mice were used to defeat 10-wk-old neuroLSD1^HET and wild-type littermates in a single session of SDS. Mice were exposed to a novel CD1 aggressor for 5 min. After the 5 min of physical contact, the mice were separated by a perforated Plexiglas divider, and the experimental mouse was allowed sensory interaction with the CD1 aggressor for 7 h. Control mice were housed (two per cage) in the opposite sides of Plexiglas divider in cages identical to those used for the SDS test. Seven hours after physical contact, the control and stressed mice were used for molecular analysis.

Data Analysis and Statistical Methods.

The number of animals was reduced to the minimum needed to reach statistically significant results. Data are shown as means ± SEM. Comparisons of two groups were analyzed by unpaired Student’s t test. For comparison of more than two groups with comparable variances, one-way ANOVA or two-way ANOVA were performed and then followed by Bonferroni post hoc comparisons to evaluate pairwise group differences. The analyses were performed with PRISM 6.0 software (GraphPad).

Antibodies Used for Western Blot or Immunoprecipitation Experiments.

The following antibodies were used for Western blot or immunoprecipitation experiments: anti-LSD1 (C69G12; Cell Signaling Technology); anti-SRF (D71A9; Cell Signaling Technology); anti-CoREST (07-455; Merck Millipore); anti-HDAC2 (ab7029; Abcam); anti-GAPDH (NB300-320; Novus Biological); anti-α/β tubulin (2148; Cell Signaling Technology); and anti-HA (sc-7392; Santa Cruz Biotechnology).

Antibodies Used for ChIP.

Antibodies used for ChIP were anti-LSD1 (ab17721; Abcam); anti-SRF (D71A9; Cell Signaling Technology); anti-H3K4me2 (07-030; Merck Millipore); and anti-H3K9/K14ac (06-599; Merck Millipore).

Primers.

Primers for mouse genotyping.

The following primers were used for mouse genotyping:

• P3 forward: ACGCGTCGACTCTTCAGTGCTTTCTCACTCCCA (3)

• P6 reverse: ATAGTTTAGCGGCCGCCCTCTATTTTCTGAGCAGCC (3)

The wild-type allele amplicon length was 589 bp. The deleted allele amplicon length was 611 bp.

Primers for ChIP.

The following primers were used for ChIP:

• mEgr1 forward: 5′-GCTTTCCAGGAGCCTGAGCG-3′

• mEgr1 reverse: 5′-CAAGGGTCTGGAACAGCACGG-3′

• m_cFos forward: 5′-TCAGAGTTGGCTGCAGCCGGC-3′

• m_cFos reverse: 5′-GCGTGTAGGATTTCGGAGATG-3′

• mCtrl forward: 5′-GGCCTGGGTCACAGACCTAC-3′

• mCtrl reverse: 5′-CTTCTGTGTACCCAGCACCTG-3′

Primers for mRNA quantification.

The following primers were used for mRNA quantification:

• mEgr1 forward:5′-CCTTCAATCCTCAAGGGGAGC-3′

• mEgr1 reverse:5′-AACCGAGTCGTTTGGCTGGGA-3′

• m_cFos forward:5′-CTGCAGCCAAGTGCCGGAATC-3′

• m_cFos reverse:5′-GGCAATCTCAGTCTGCAACGC-3′

• mLSD1 forward:5′-GCCTCAGCAGACACAGAAGG-3′

• mLSD1 reverse: 5′-TGTTGTAAGGCGCTTCCAGC-3′

Primers for isoform quantification.

The following primers were used for isoform quantification:

• mLSD1 forward: 5′-CTACCATTTCATCTTTTTCTTTTGG-3′

• mLSD1 reverse: [6FAM] 5′-AGTGAGCCGGAAGAGCCGTCTG-3′