Early-life exercise primes the murine neural epigenome to facilitate gene expression and hippocampal memory consolidation

“Emx1-NuTRAP” mouse allows for simultaneous isolation of nuclear chromatin and translating mRNA from a single population of hippocampal neurons

Our initial goal was to obtain translating mRNA and nuclear DNA from a single population of neurons to couple gene expression changes with alterations in histone modifications resulting from ELE. To do this, we developed a transgenic mouse line (Emx1-NuTRAP) and experimental protocol (SIT protocol28). We crossed male NuTRAP reporter mice29 with female Emx1-Cre32,33 mouse lines to generate Emx1-Cre; NuTRAP transgenic progeny for this study (“Emx1-NuTRAP”; Fig. 1a). Emx1-expressing neurons are predominately excitatory32 and are necessary for hippocampal neurogenesis and motor skill learning34. In the presence of a loxP site-flanked sequence, Emx1-Cre facilitates recombination in approximately 88% of neurons in the neocortex and hippocampus, and in less than 2% of GABAergic inhibitory interneurons in these regions32. Emx1-Cre also targets neural progenitor cells and can be expressed in mature astrocytes in striatum35. Of note, we exclusively used Emx1-Cre female mice in our breeding schemes given that the Cre-recombinase has been reported to be expressed in male germline cells36.

Fig. 1: Generation and validation of the Emx1-NuTRAP mouse as a tool for generating sequencing-grade, neuron-specific DNA and translating mRNA from a single hippocampal homogenate.
figure 1

a Schematic representation of the Emx1-NuTRAP mouse generation and workflow of the Simultaneous INTACT and TRAP (“SIT”) protocol. Created with BioRender.com. b Immunofluorescence imaging of the CA1 region of the hippocampus at ×20 objective and b′ ×60 objective after incubation with an mCherry antibody . c Immunofluorescence imaging of the CA3 region of the hippocampus at ×20 objective and c ×60 objective after incubation with an mCherry antibody . d Immunofluorescence imaging of the DG region of the hippocampus at ×20 objective and d ×60 objective after incubation with an mCherry antibody . Scale bars are set to 50 µm for all images bd. e Flow cytometry for EGFP, Thy1, and S100β (15.3% of cells Thy1 + /EGFP + and 0.13% of cells S100β+/EGFP+). f Heatmap of differentially expressed neuronal and non-neuronal cell type markers from RNA-seq data comparing TRAP-isolated RNA from hippocampus of Emx1-NuTRAP mice vs hippocampal mRNA isolated from wild-type mice. (Abbreviations: NPC neural progenitor cell, Oligo oligodendrocyte).

We first sought to validate neuron-specific expression of the NuTRAP cassette. Immunohistochemichal analysis of Emx1-NuTRAP hippocampal slices show distinct GFP and mCherry expression in CA1 and CA3 pyramidal neurons (Fig. 1b, c) as well as granule cells of the dentate gyrus (DG; Fig. 1d). As Emx1-expressing neural stem cells of the DG can potentially differentiate into astrocytes32, we sought to determine whether a significant number of mature astrocytes were also obtained in the population of isolated cells. Flow cytometry experiments using Thy1 as a neuronal marker and S100β as an astrocytic marker revealed a distinct population of cells double positive for GFP and Thy1, whereas a S100β and GFP double positive cell population was absent (Fig. 1e; see Supplementary Fig. 1, and Supplementary Data 1 for gating). There was a small population of GFP+/Thy1− singlets. We interpret this finding to reflect Emx1 labeling of dividing immature neural progenitors, as Thy1 is not expressed in granule cells of the DG until the cells are 4-5 weeks old37.

To perform simultaneous INTACT & TRAP (“SIT”)28, hippocampal tissue was dissected from both brain hemispheres, combined, and homogenized in one sample tube (Fig. 1a). We then performed the SIT method on isolated samples by starting the protocol with the beginning steps from the INTACT procedure modified to include cycloheximide. Cycloheximide works rapidly to inhibit protein synthesis and is used for maintaining crosslinks between translating mRNA and ribosomal subunits during purification in the traditional TRAP method. Despite the presence of cycloheximide, nuclear morphology from Emx1-expressing neurons was unchanged, as demonstrated by mCherry positive nuclei from hippocampal neurons remaining successfully bound to magnetic beads after our modified INTACT procedure (Supplementary Fig. 2a). Additionally, a Bioanalyzer was used to determine if there were cycloheximide-induced double-stranded DNA breaks in the nuclear preparations that could potentially interfere with downstream DNA sequencing applications (such as ATAC-, CUT&RUN-, or CUT&Tag-seq). We found no evidence of DNA double strand breaks generating fragments of <1 kb28. Following magnetic purification of biotin-labeled nuclei, the supernatant fraction was removed and taken through TRAP, while the pelleted nuclei were processed through the remaining steps of INTACT (see Methods section and ref. 28). Using qPCR, we found that TRAP-isolated mRNA had significantly reduced expression of Mog and Cd11b compared to total RNA, suggesting depletion of oligodendrocyte and microglial populations in TRAP mRNA, respectively (Supplementary Fig. 2b). Combining bilateral hippocampi from a single mouse yielded TRAP-isolated mRNA of high quality and sufficient concentration for sequencing (RIN > 8 for all samples, average yield RNA = 14.367 ng/ul; Supplementary Data 2)28. INTACT-isolated nuclei were further processed using the cleavage under targets and release using nuclease (CUT&RUN38) method to isolate antibody-specific protein-DNA interactions for downstream DNA sequencing. The resulting DNA libraries were of high quality and concentration when using specific antibodies (H4K8ac: average size = 1238 bp, average concentration  = 126.8 nM; H3K27me3: average size = 1032 bp, average concentration = 146.5 nM; Supplementary Data 2). In contrast, the resulting DNA libraries using the non-specific IgG control had substantially lower concentrations (IgG: average size = 1035 bp, average concentration = 24.2 nM; Supplementary Data 2) further indicating that nuclear DNA from both isolations was of high starting quality.

To confirm our TRAP-isolated hippocampal mRNA came primarily from excitatory neurons, we compared RNA-seq data from hippocampal tissue of wild type mice vs TRAP-isolated RNA-seq (TRAP-seq) data from Emx1-NuTRAP mice to assess for neuronal gene enrichment. We found that TRAP-isolated mRNA had enrichment of several neuronal genes, including Dlg4, Thy1, Eno2, and Syt4, as well as a substantial reduction in astrocytic, microglial, oligodendrocyte, and inhibitory neuronal genes (Fig. 1f; for a heat map with expanded gene list taken from39, see Supplementary Fig. 2c). There was a relative expression increase of glial fibrillary astrocytic protein (GFAP) in our TRAP samples. GFAP can also be expressed in neural stem populations that were present in our whole hippocampus samples, so this finding may reflect the neural stem cell population of the dentate gyrus40,41. Overall, these findings support the Emx1-NuTRAP mouse model as a valid tool for neuron-enriched isolation of sequencing-grade, translating mRNA and nuclear chromatin from a single brain tissue homogenate.

ELE reveals highly comparable translatomic and epigenomic profiles resulting from simultaneous and separate DNA and RNA isolation methods

Previous methods for isolating DNA and mRNA using INTACT and TRAP (respectively) from NuTRAP mouse tissue have taken separate tissue homogenates for each procedure29,42. In this study, we developed an approach to perform SIT on a single tissue homogenate containing bilateral hippocampi28 (the products of SIT are herein referred to as “simultaneous isolations”). To determine if our SIT approach yields comparable results to INTACT and TRAP performed on separate samples, we isolate hippocampal tissue obtained from a separate brain hemisphere for each method and counterbalanced for left versus right (we refer to this protocol as “separate isolations”). A Bioanalyzer was used to determine the amount and quality of the RNA obtained from each type of isolation (Separate isolations: average RNA concentration = 7.805 ng/ul, RIN > 8 for all samples; Simultaneous isolations: average RNA concentration = 14.367 ng/ul, RIN > 8 for all samples; Supplementary Data 2). The average RNA yield from the separate isolations (using a unilateral hippocampus) was approximately equal to half of the average yield of the simultaneous isolations (which combined bilateral hippocampi; Supplementary Data 2). Similarly, the final library concentrations for the separately isolated IgG CUT&RUN-seq libraries were also approximately half the concentration of the simultaneous isolations (average simultaneous: 24.2 nM, average separate: 11 nM; Supplementary Data 2). We interpret this to mean that nuclear DNA was fully intact in the simultaneous isolation because we did not obtain substantially more than double the concentration in the simultaneous vs separate isolations. Unilateral hippocampal homogenates yielded sufficient sequencing concentrations and quality to allow for library preparations from individual mice (Supplementary Data 2).

To determine if normalized TRAP-seq counts were similar between the two isolation methods, we performed a Spearman’s correlation between the datasets from sedentary mice. The two TRAP-seq datasets were found to be highly correlated, with R values >0.5 and p value < 2.2 × 10−16 (R = 1, p < 2.2 × 10−16; Fig. 2a). Next, CUT&RUN-seq was used to identify genomic regions interacting with two histone post-translational modifications (PTMs) of interest in the exercise and hippocampal memory fields: H4K8ac, a permissive histone mark associated with active transcription, or H3K27me3, a generally repressive histone PTM. We again applied Spearman’s correlation to understand whether the simultaneous vs separate INTACT isolation methods could influence CUT&RUN-seq peak distribution. We compared normalized count data for CUT&RUN-seq peaks across a representative chromosome (chromosome 2). We binned 100 bp increments along the entire chromosome from the simultaneous and separate isolations using datasets generated from sedentary mice in our study. Normalized sequencing counts, which reflected reads assigned to binned genomic regions along chromosome 2, were highly similar between separate vs simultaneous conditions (H4K8ac: R = 0.65, p < 2.2 × 10−16; H3K27me3: R = 0.81, p < 2.2 × 10−16; Fig. 2b, c). Taken together, these experiments suggest that translating mRNA and nuclear DNA isolated from hippocampal homogenates using either simultaneous or separate isolation procedures are comparable in terms of quality, concentration, functional characterization, and normalized sequencing reads.

Fig. 2: Comparing simultaneous (SIT) and separate DNA and RNA isolation procedures from Emx1-NuTRAP mice that underwent ELE.
figure 2

For ae, n = 2 mice for simultaneous isolations and n = 3 mice for separate isolations. a Spearman’s correlation for RNA expression between separately isolated RNA and simultaneously isolated RNA (R = 1, p < 2.2 × 10−16). b Spearman’s correlation between separate and simultaneously isolated DNA from CUT&RUN-seq for H4K8ac with 100 bp chromosome 2 position bins (R = 0.65, p < 2.2 × 10−16). c Spearman’s correlation between separate and simultaneously isolated DNA from CUT&RUN-seq for H3K27me3 with 100 bp chromosome 2 position bins (R = 0.81, p < 2.2 × 10−16). d Gene length versus fold change distribution plot for simultaneously isolated RNA. e Gene length versus fold change distribution plot for separately isolated RNA. f Genes upregulated in TRAP-seq after ELE in either simultaneous (n = 2 sedentary mice and n = 3 ELE mice) or separate isolations (n = 3 mice per group). g Genes downregulated in TRAP-seq after ELE in either simultaneous or separate isolations. h, i Panther Gene Ontology: Biological Processes Venn diagram for genes upregulated (h) or downregulated (i) after ELE in either separate or simultaneous isolations. j, k Panther Gene Ontology: Biological Processes bar graph of gene distributions for genes upregulated (j) and downregulated (k) after ELE in either separate or simultaneous isolations.

We next wanted to determine if the different isolation methods could bias resulting gene expression on the basis of gene length. The differentially expressed genes (DEGs) were analyzed using DESeq2 and Student’s t-tests to determine statistically significant differences in RNA-seq and detect potential biologically relevant log-fold changes. When plotting gene length against log fold change of gene expression, we see a similar distribution pattern of ELE-induced hippocampal DEGs between the simultaneous and separate methods (Fig. 2d, e). We then compared the biological categories of the ELE-induced DEGs obtained using either the separate or the simultaneous isolation methods using Panther Gene Ontology (GO). Although many of the genes did not overlap (Fig. 2f, g and Supplementary Data 3), most GO terms did (14/20 GO terms for upregulated genes and 15/17 for downregulated genes; Fig. 2h, i and Supplementary Data 3). Furthermore, there was high similarity between the percentage of genes found in each of the GO categories (Fig. 2j, k). These results demonstrate that the simultaneous approach for isolating and sequencing mRNA and nuclear chromatin can be performed using a single hippocampal homogenate and can generate comparable results to more traditional methods using separate cell samples to pair different types of sequencing data.

Neuronal gene expression programs resulting from ELE, and histone PTM peak distribution, are biologically comparable between left and right hippocampal hemispheres

Many studies take advantage of the brain’s structural symmetry by using tissue from each hemisphere for separate molecular processing. Prior evidence demonstrates that hippocampal lateralization can influence LTP, hemisphere-specific glutamate receptor density, and performance in certain memory tasks43. We wanted to determine if there were differences in transcriptional programs and biological processes resulting from ELE in left vs right hippocampi. Using the separate isolation approach described above, we compared ELE-induced differential gene expression between left and right hemispheres. We found that although individual gene expression patterns were different between hippocampi originating in the left and right brain hemispheres (Fig. 3a, b and Supplementary Data 4 and 5), the Panther GO: Biological Processes terms were similar, with most categories overlapping (Fig. 3c–f and Supplementary Data 4). Furthermore, applying 2-way ANOVA and likelihood ratio statistical analyses with multiple comparisons find more genes associated with the “exercise” term than “hemisphere” or “interaction” (2-way ANOVA: “exercise” =  1583, “hemisphere” = 808, “interaction” = 869; Likelihood Ratio Test: “exercise” =  565, “hemisphere” = 466, “interaction” = 548; Supplementary Data 5). suggesting a greater impact of ELE over hemisphere on differential gene expression, and the possibility of spatial or hemispheric assignment of genes with functional similarity.

Fig. 3: Functional categorizations of ELE-induced differential gene expression, and distribution of H4K8ac and H3K27me3, are similar regardless of hippocampal hemisphere.
figure 3

a, b Venn diagram of genes upregulated (a) or downregulated (b) by ELE identified between the left and right hemispheres by the separate isolation protocol TRAP-seq (n = 3 mice per group). c, d Venn diagram of Panther Gene Ontology: Biological Process terms for genes upregulated (c) or downregulated (d) by ELE identified between the left and right hemispheres by the separate isolation protocol for performing TRAP-seq. ef Percent of genes fitting into each GO category for the left and right hemisphere for genes upregulated (e) and downregulated (f) by ELE by separate isolation TRAP-seq. g Spearman’s correlation between the left and right hemispheres TRAP-seq data from the separate isolation protocol of sedentary rlog normalized expression (R = 1, p < 2.2 × 10−16). h Spearman’s correlation between hemispheres for CUT&RUN-seq for H4K8ac normalized count data for chromosome 2 in 100 bp bins (R = 0.59, p < 2.2 × 10−16). i Spearman’s correlation between hemispheres for CUT&RUN-seq for H3K27me3 normalized count data for chromosome 2 in 100 bp bins (R = 0.7, p < 2.2 × 10−16).

We next determined if left and right hippocampal hemispheres had significant differences in DEGs and CUT&RUN-seq peak distributions at baseline (without exercise). We performed Spearman’s correlation on the transcript counts from the TRAP-seq data (R = 1, p < 2.2 × 10−16; Fig. 3g), and the CUT&RUN-seq 100 bp binned peak counts along a representative chromosome (chromosome 2) for two histone modifications: H4K8ac (R = 0.59, p < 2.2×10−16; Fig. 3h), and H3K27me3 (R = 0.70, p < 2.2×10−16; Fig. 3i). We found that both the transcript counts and CUT&RUN-seq peaks were highly similar (significance considered as R values >0.5 and p value < 2.2 × 10−16; Fig. 3g–i). Overall, left and right hippocampal hemispheres did not demonstrate significant differences in normalized sequencing counts and peak distributions in the sedentary condition, suggesting that choice of hippocampal hemisphere is a less important factor to consider in obtaining representative data from the translatome and transcriptional regulation via histone modifications.

ELE promotes expression of plasticity-related genes and implicates  transcriptional regulatory pathways involved in hippocampal memory

To identify the gene expression programs induced by ELE in hippocampal neurons, we performed neuron-enriched bulk sequencing on TRAP-isolated mRNA extracted from Emx1-NuTRAP mouse hippocampi using our “SIT” protocol. Using DESeq2 to assess for DEGs (>30% expression increase with a p value < 0.05) our TRAP-seq revealed 297 upregulated and 338 downregulated hippocampal genes resulting from ELE (Fig. 4a and Supplementary Data 3). Many of the genes upregulated after ELE are known to be involved in exercise and/or hippocampal memory mechanisms, including Bdnf44 and Nr4a145. To functionally categorize ELE-induced DEGs, we performed a Panther Gene Ontology (GO) analysis46 focusing on the Molecular Function categorization and separated by upregulated and downregulated genes. Regardless of gene expression directionality, GO term categories with the most genes functionally assigned to them were “binding”, “catalytic activity”, “molecular function regulator”, “transporter activity”, “molecular transducer activity”, and “structural molecule activity” (Fig. 4b). Many of the upregulated genes driving these categories are known to have critical roles in neuronal function (Kcna1, Slc24a4, Stxbp5l, Gabra2, and Camk2n2), neurodevelopment (Artn, Kdm7a, Sox21, Gap43, and Efna5), and hippocampal memory (Bdnf, Nr4a1 and Dusp5).

Fig. 4: ELE results in novel transcriptomic changes in hippocampal neurons during adolescence.
figure 4

a Volcano plot of differentially expressed genes reaching significance identified by TRAP-seq on ELE versus sedentary (n = 2 sedentary mice and n = 3 ELE mice, absolute value log2 fold change >0.3785 and p-value < 0.05). b Panther Gene Ontology: Molecular Function top terms by most genes assigned. c Top 6 “Upstream Regulators” identified by Ingenuity Pathway Analysis (IPA). d Representative Gene Set Enrichment Analysis: Reactome leading-edge diagrams showing genes upregulated (d) or downregulated (d′) in ELE, and their categories of enrichment. *Abbreviated terms in d: (1) “transport of mature mRNAs derived from intronless transcripts”, d′: (2) “activation of the mRNA upon binding of the cap binding complex and EIFs and subsequent binding to 43S”, (3) srp-dependent cotranslational protein targeting to membrane, and (4) response of eif2ak4 gcn2 to amino acid deficiency.

To evaluate possible transcription factors and upstream regulators implicated by ELE-induced activated gene networks, we applied Qiagen’s Ingenuity Pathway Analysis (IPA) to our TRAP-seq dataset47. We identified significant canonical signaling pathways implicated in ELE effects on hippocampal neuronal function (Supplementary Data 6). The top six IPA-identified “upstream regulators” by significance included CREB1, KMT2A, LEF1, SMAD3, EGR1, and MITF (Fig. 4c and Supplementary Data 6)48,49,50,51,52,53,54,55. The transcription factors KMT2A and LEF1 have not been previously associated with the effects of exercise on hippocampal function. Interestingly, CREB1 (through its associated CBP56) and KMT2A (through its methyltransferase activity49) both have histone modifying properties. LEF1 has been shown to regulate neural precursor proliferation in the hippocampus50. SMAD3 is critical for intermediate progenitor cell survival and negatively regulates serum irisin after exercise51,52. EGR1 is an immediate early gene that recruits TET1 DNA demethylase during neural activation and development53. MITF was the sixth upstream regulator identified and is a transcription factor linked to autophagy mechanisms54. These specific transcription factors were not significantly differentially expressed in our TRAP-seq dataset; however, one possibility is that their activity may not be linked to a change in their own expression but rather modulated by exercise to influence downstream gene expression.

To further understand whether the transcriptional profiles resulting from ELE were enriched for specific a priori assigned molecular functions, we performed Gene Set Enrichment Analysis (GSEA)57. We evaluated the “Reactome” category of functional gene sets followed by a leading-edge analysis to further determine which genes were driving the significant categories (Fig. 4d and Supplementary Data 7). Of the DEGs upregulated after ELE, several interesting categorizations and the genes driving them were revealed. The genes H3c4, H2bc3, H2bc5, and H2bc9 genes, which encode histone family member proteins, were driving the categories: “DNA methylation”, “HDACs deacetylate histones”, and “transcriptional regulation by smRNAs”. “Meiosis” and “meiotic recombination” also emerged in these results which was unusual; however, leading edge analysis showed many of these genes (Atm, Blm, Msh4, and Mnd1) to be generally involved in cell-cycling processes. Exercise is well known to increase adult neurogenesis in hippocampal dentate gyrus and can explain cell cycle gene enrichment in our dataset16. Several nucleoporin complex genes (Nup35/ 42/ 43/ 50/ 58/ 62/ 93/ 107/ 153/ 210) were also identified for their associations with categories such as: “gene silencing by RNA”, “nuclear envelope breakdown”, and “transport of mature mRNAs derived from intronless transcripts”. Nup genes form a variety of nuclear pore complexes that play critical roles in cellular processes including cell-cycle regulation, cellular differentiation, and epigenetic control58.

Of the downregulated pathways identified, translation-associated categories (“translation”, “eukaryotic translation initiation/elongation”, and “SRP-dependent co-translational protein targeting to membrane”) were driven by significant downregulation of ribosomal protein gene families (Rps, Rpl, Rplp) and eukaryotic initiation factor 3 (eIF3) subunits. Over-expression of eIF3 subunits has been linked to neurodegeneration and altered expression of eIF3 has been associated with neurodevelopmental disorders59. Additionally, collagen genes (Col1a1, Col1a2, Col2a1, Col3a1, and Col5a1) were downregulated after ELE, leading to the identification of categories including: “MET activates PTK2 signaling”, “MET promotes cell motility”, and “collagen degradation”. Col1a1 and Col1a2 have been previously identified as putative aging genes that decrease in their expression after chronic exercise in female adult rodents60. By analyzing these enriched pathways and the networks of genes driving them, we were able to identify several expected and unexpected transcriptional programs activated by ELE that could be unique to exercise timing during the juvenile-adolescent period. Ultimately, to determine if gene expression programs identified here are in fact unique to exercise timing, an adult exercised cohort would need to be included for comparison. This would be an important follow-up study to these data.

Neuronal H4K8ac is enriched and H3K27me3 is reduced at a subset of plasticity genes after ELE

The power of the Emx1-NuTRAP mouse model coupled with the “SIT” technical approach is the ability to directly pair differential gene expression with governing epigenetic mechanisms. This is achieved by obtaining both translatomic and epigenomic-sequencing data from the same cell population (Fig. 1a). To investigate ELE-induced changes in histone PTMs across the genome, hippocampal nuclei were obtained from ELE and sedentary mice using the INTACT method followed by CUT&RUN-seq for the modifications H3K27me3 and H4K8ac from nuclear chromatin. We chose these two histone PTMs for their previously described functions and potential roles in exercise and memory mechanisms. H4K8ac is a permissive modification that is enriched at Bdnf after adult exercise, and its presence correlates with improved hippocampal memory22. H3K27me3 is a repressive histone mark and a common control in CUT&RUN-seq studies38,61 and is decreased at the Bdnf promoter region after contextual fear conditioning training61. Peaks were called using SEACR62 and histone PTM enrichment peaks were overlapped with ELE-induced DEGs obtained from our TRAP-seq experiments described above (Fig. 5a and a′, and Supplementary Fig. 3 for Upset plot). We evaluated for the presence of H4K8ac and H3K27me3 peaks in union with upregulated and downregulated mRNA resulting from ELE. 93 upregulated genes had new H4K8ac peaks, while 35 downregulated genes had new H3K27me3 peaks (Fig. 5b and Supplementary Data 8). Notably, of those 93 upregulated genes with H4K8ac peaks, 14 also had new H3K27me3 peaks resulting from ELE. 17 genes had presence of H3K27me3 in both conditions (ELE and sedentary; Fig. 5a). We interpret these results to mean that transcription of these 93 genes was promoted as a result of ELE-induced H4K8ac.

Fig. 5: ELE alters H4K8ac and H3K27me3 occupancy at differentially expressed genes.
figure 5

a Venn diagram of gene peaks identified in CUT&RUN-seq for H4K8ac and H3K27me3 by SEACR (top 1% of peaks FDR < 0.1) overlapped with genes upregulated (a) and downregulated (a′) in RNA by ELE (n = 2 sedentary mice and n = 3 ELE mice). b CUT&RUN-seq SEACR peak calls at genes that overlap with genes differentially expressed by ELE as determined by TRAP-seq. c Stacked bar of the relative distribution of peak calls at various genomic regions identified by CUT&RUN-seq called using SEACR. d Representative track of CUT&RUN-seq data around Prox1 showing differential peak height between ELE and sedentary conditions for H4K8ac (top) and H3K27me3 (bottom).

The loss of either H3K27me3 or H4K8ac after ELE did not directly correlate with directionality of translating mRNA expression changes (Fig. 5b). This result may indicate that ELE-induced differential gene expression is associated with the addition, rather than removal, of these chosen two histone modifications. We found one gene that was downregulated and had loss of H4K8ac (Asphd1), and two genes downregulated with loss of H3K27me3 (Fdxr and Rtl1; Fig. 5b and Supplementary Data 8). Of particular interest, reduced expression of paternally-inherited Rtl1 is associated with improved hippocampal memory performance, whereas overexpression results in memory deficits63,64. Unsurprisingly, the majority of histone PTM peaks newly present as a result of ELE did not correlate with gene expression changes. However, notable upregulated genes with new H4K8ac after ELE include Prox1, Sntb2, and Kptn. PROX1 is involved governing differentiation, maturation, and DG versus CA3 cell identity in intermediate progenitor cells of the hippocampal DG65, while SNTB2 is involved in G protein-coupled receptor cell signaling66. KPTN encodes a protein involved in cytoskeletal cell structure, and its mutation can cause neurodevelopmental disability and seizures67. Interestingly, downregulated genes found to have new H3K27me3 after ELE included Col3a1, Efnb2, Epop, and Myoc. COL3A1 and MYOC are structural proteins involved in the extracellular matrix and the cytoskeleton respectively68,69,70. EFNB2 is a signaling molecule involved in cell migration and its haploinsufficiency can cause neurodevelopmental disability71,72,73. EPOP is a known editor of the chromatin landscape by altering H2Bub and H3K4me3 distributions74,75.

Although changes in gene expression associated with H3K27me3 peaks were less numerous than those associated with H4K8ac, we noticed a striking difference in the distribution of peaks across the genome (Fig. 5c). After ELE, the distribution of H3K27me3 decreases in the promoter region and increases in the distal intergenic and intronic regions (Fig. 5c). To investigate how ELE-induced changes to histone PTM distribution might alter gene expression, we looked at the distribution of peaks around a representative gene, Prox1 (Fig. 5d). Most apparent changes in histone PTM peaks occurred in regions annotated as regulatory regions, rather than the promoter region, which is concordant with our data showing that most histone PTM changes happen outside of the promoter region. Taken together, these results suggest that ELE promotes the new addition of H4K8ac and H3K27me3 PTMs to regulate gene expression. Given that most identified peaks did not correlate with DEGs, they may instead implicate regions of interest that are “primed” for facilitated gene expression following future stimuli, such as learning.

ELE alters H4K8ac and H3K27me3 occupancy at regulatory regions of genes implicated in hippocampal memory consolidation

Prior work from our lab and others has found that both ELE and adult exercise can facilitate hippocampal long-term memory (LTM) formation in mice exposed to a typically subthreshold learning event (3 min of object location memory (OLM) training, which is normally insufficient for LTM formation in sedentary mice)17,22. These findings suggest that ELE may “prime” neuronal function to facilitate hippocampal learning. In this next experiment, we investigate whether ELE-enabled hippocampal memory is associated with altered presence of histone PTMs H4K8ac and H3K27me3 in regulatory regions of genes associated with learning and hippocampal plasticity; the prediction being that alterations of histone PTM occupancy as a result of ELE could promote gene expression programs necessary for facilitating memory consolidation. Wild type animals underwent ELE followed by OLM training on P42 and were sacrificed 60 min later (Fig. 6a). Dorsal hippocampi were dissected, and tissue was homogenized and processed for bulk RNA-seq. These data were compared to the ELE-induced DEGs, H4K8ac, and H3K27me3 occupancy occurring without learning (generated from Emx1-NuTRAP mice TRAP-seq and CUT&RUN-seq). We categorized DEGs based on the experimental group of wild type animals: ELE alone, ELE + 3 min learning stimulus, Sedentary+3 min learning stimulus, and Sedentary+10 min learning stimulus. To discover ELE-induced genes “primed” for regulation during memory consolidation, we focused specifically on cGAME (candidate genes associated with memory after exercise), cGAMES (candidate genes associated with memory after exercise or sedentary), and cGAMS (candidate genes associated with memory after sedentary) conditions (Fig. 6b, c and Supplementary Data 9). We then evaluated cGAME and cGAMES groups of genes for the presence or absence of histone modifications H4K8ac and H3K27me3 after ELE, suggesting addition or removal of these histone marks may promote a permissive (increased gene expression) or repressive (reduced gene expression) chromatin state, respectively (Fig. 6b′–c′ and Supplementary Data 10).

Fig. 6: ELE may prime hippocampal gene expression supporting memory consolidation via alterations in H4k8ac and H3k27me3 occupancy.
figure 6

a Experimental design diagram. Created with BioRender.com. b Venn diagram representing upregulated genes relative to mice without ELE or learning in either ELE, ELE and 3 min OLM training, Sed and 3 min OLM training, and Sed and 10 min OLM training. b′ Pie chart showing the relative percentage of genes that have new H4K8ac presence or H3K27me3 loss as a result of ELE in specific groups identified in the b labeled cGAME and cGAMES. c Same as b but for downregulated genes relative to mice without ELE. c′ Pie chart of downregulated genes in the cGAME or cGAMES groups that have new H3K27me3 or new loss of H4K8ac as a result of ELE. d Z-score based heatmap of genes in the cGAMES and cGAME groups that have new H4K8ac or H3K27me3 presence or loss as a result of ELE. Expression values used to calculate the Z-scores were relative to mice that did not experience OLM training or ELE. e Upstream regulators identified by IPA of the genes represented by d. e′ Upstream regulators identified by IPA of the genes in the cGAME (up and downregulated) group that are unique to that when only those are included in the IPA analysis. e″ Upstream regulators identified by IPA of the genes in the cGAME and cGAMES (up and downregulated) group when the genes in those groups are put through IPA together. (asterisk indicates that the activated or inhibited indicator is registered from the IPAs based on the z-score for that regulator’s expression and not directly indicated based on software prediction).

In the cGAMES group, there were 145 genes found to be significantly increased in the mice that underwent a threshold learning event (3 min for ELE; 10 min for sedentary mice). Of these 145 genes, 40% of them (58) had new H4K8ac following ELE (Fig. 6b–b′). We consider this group of genes to potentially be “primed,” or readied, by ELE for rapid transcription to support hippocampal memory consolidation at 3 min OLM training, while also being involved in memory consolidation mechanisms in the threshold learning of sedentary mice (10 min OLM training). The cGAME group of genes are those which significantly increased in the ELE mice that underwent a threshold learning event but were not increased in any of the other groups (including the sedentary, threshold learning event/cGAMS group). These genes may be of unique importance for memory consolidation specifically occurring after ELE given their absence in sedentary learners. We found that the cGAME group had significant upregulation of 256 genes, with ~30% (76) of those genes receiving new H4K8ac following ELE alone (Fig. 6b–b′). We also evaluated the cGAMES and cGAME that were significantly downregulated during memory consolidation. We found 304 cGAMES and 361 cGAME genes (Fig. 6c). In both of these groups, a small minority of down-regulated genes had altered H3K27me3 after ELE, and this was mostly gain of H3K27me3 (cGAMES: 6.3%; cGAME: 6.4%; Fig. 6c′).

We next wanted to determine how ELE “priming” directed gene expression during memory consolidation that occurs after exercise (ELE 3 min OLM) relative to sedentary memory consolidation (Sed 10 min OLM). We compared these to the group that experienced the sub-theshold learning event (Sed 3 min OLM). We compared the relative log2 fold changes (LFCs) of the genes in cGAME and cGAMES using z-scores for each group across the genes (Fig. 6d). In the vast majority of these primed genes, the ELE drove gene expression during 3 min of OLM in the same direction as 10 min of OLM training in sedentary animals (Fig. 6d). In some cases, the gene expression changes were more exaggerated in the same direction as a sedentary threshold event (Fig. 6d). We interpret this to indicate that ELE enables a 3 min consolidation event to be threshold (when 10 min is required for a sedentary mouse) by priming these genes for altered gene expression in the same direction as a sedentary threshold event.

We next identified upstream regulators of cGAME and cGAMES genes with ELE-induced histone modifications, as these mediators are likely involved in long-term memory formation of a subthreshold learning event following ELE. The cGAMES and cGAME lists were evaluated using Qiagen’s Ingenuity Pathway Analysis (IPA). First, candidate regulators were identified for the cGAME and cGAMES groups that have histone PTM changes as a result of ELE that we measured (Fig. 6e and Supplementary Data 11). Inhibition of MECP2 may indicate a reduction in DNA methylation. CREB1 is required for LTM48,76. SNCA may play a role in neuroplasticity by modulating synaptic vesicle transport77.

Next, we wanted to identify gene networks that might be at play that are not just regulated by the histone PTMs we selected. Many more histone modifications exist and likely play a critical role in regulating gene expression during consolidation in addition to those we selected. We investigated the genes in cGAME and cGAMES using IPA and identified a set of likely upstream regulators for these genes taken together (Fig. 6e” and Supplementary Data 11). Once again MECP2 inhibition and CREB1 activation appear to be indicated validating their presence in the previous IPA. Notably new to this group are inhibition of FMR1, KMT2D, and TCF20, along with activation of FASN, CTNNB1, and MKNK1. Next, we interrogated what networks might be unique to exercise-enabled consolidation by running the cGAME list through IPA analysis (Fig. 6e′ and Supplementary Data 11). Unique to this group, RTN4, an inhibitor of neurite outgrowth78, was predicted to trend towards inhibition. GRIN3A, a glutamatergic NMDA receptor important to synaptic development and refinement79, was predicted activated in this group. These findings suggest that ELE may enable consolidation by increasing neurite and synapse growth and development. TCF7L2 is also predicted to be activated. Given that this transcription factor is involved in Wnt signaling, and neurogenesis80, and activated CTNNB1 is also predicted to be associated with threshold consolidation, altered Wnt signaling may be critical for ELE-enabled hippocampal memory consolidation.