Mouse tissue dissection

Adult C57BL/6 J male mice were purchased from Jackson Laboratories (strain #000664). Tissues were extracted from 3-month-old, 10-month-old and 18-month-old mice. All dissections were performed consistently in sterile conditions by the same laboratory member. Briefly, prefrontal cortex and dorsal hippocampus were dissected in ice-cold ACSF (in mM: 126 NaCl, 2.5 KCl, 26 NaHCO3, 2 CaCl2, 2 MgCl2, 1.25 NaH2PO4, and 10 glucose). Both the prefrontal cortex and dorsal hippocampus were dissected out from both hemispheres of each mouse, using a brain block and scalpel as described before.71,72 Brain tissues were then immediately “flash frozen” in liquid nitrogen for down-stream applications. In similar fashion the heart and femoral bone and attached musculature were dissected from the animal. The entire heart was dissected, and flash frozen in liquid nitrogen. The quadriceps femoris muscle was dissected from the femur and flash frozen in liquid nitrogen. The femur was then further processed for obtaining bone marrow. Briefly, all tissue was removed from the femur. The distal end was then cut and placed in an eppendorf tube to be centrifuged at 4 °C. The bone marrow was then flash frozen in liquid nitrogen (Please see https://www.jove.com/t/53936/murine-hind-limb-long-bone-dissection-and-bone-marrow-isolation).

Tissue preparation and nuclei isolation for snATAC-seq and snRNA-seqFrontal cortex and dorsal hippocampus

For snATAC-seq on dorsal hippocampus and frontal cortex, tissue was homogenized using mortar and pestle on liquid nitrogen.42 ~20 mg ground tissue was suspended in 1 mL of nuclear permeabilization buffer: 5% BSA, 0.2% IGEPAL CA-630 (Sigma-Aldrich), 1 mM DTT, and 1× EDTA-free protease inhibitor (Roche or Pierce) in PBS). Nuclei were rotated at 4 °C for 5 min before being pelleted again with a swinging-bucket centrifuge (500× g, 5 min, 4 °C; 5920 R, Eppendorf).

For snRNA-seq on hippocampus, ~20 mg snap-frozen and ground hippocampus was suspended in 500 µL of nuclei buffer: 0.1% Triton-X-100 (Sigma-Aldrich, T8787), 1× EDTA free protease inhibitor (Roche or Pierce), 1 mM DTT, and 0.2 U/µL RNase inhibitor (Promega, N211B), 2% BSA (Sigma-Aldrich, SRE0036) in PBS. Sample was incubated on a rotator for 5 min at 4 °C and then pelleted with a swinging bucket centrifuge (500× g, 5 min, 4 °C; 5920 R, Eppendorf). For biological replicates of hippocampus, snATAC-seq and snRNA-seq were performed on aliquots of the same ground tissue. For the other biological replicate, starting tissue from different mice was used for snATAC-seq and snRNA-seq respectively.

For snRNA-seq on frontal cortex, ~20 mg snap-frozen and ground frontal cortex (Biological Replicate 2, aliquot from the same powder used for snATAC-seq) or a whole single snap-frozen frontal cortex were homogenized as described before with modifications.73 Tissue was transferred to a glass dounce and submerged in 1 mL dounce buffer: 0.25 M Sucrose (Sigma), 25 mM KCl, 5 mM MgCl2, Tris-HCl, pH 7.5, 1× EDTA-free protease inhibitor (Roche or Pierce), 1 mM DTT, and 0.2 U/µL RNase inhibitor (Promega, N211B), 2% BSA (Sigma-Aldrich, SRE0036) in PBS. Samples were homogenized using a loose pestle for 5–10 strokes followed by a tight pestle for 15–20 strokes. Suspension was transferred to a pre-chilled 1.5 mL LoBind tube (Eppendorf) through a 30 μM CellTrics filter (Sysmex) and pelleted with a swinging bucket centrifuge (100× g, 10 min, 4 °C; 5920 R, Eppendorf).

Heart and leg muscle tissue

Nuclei were isolated from individual snap-frozen whole heart and leg muscle tissue as described74 using gentleMACS M tubes (Miltenyi) on a gentleMACS Octo dissociator (Miltenyi). Tissue was submerged in magnetic-activated cell sorting (MACS) buffer: 5 mM CaCl2, 2 mM EDTA, 1× protease inhibitor EDTA-free (Roche or Pierce), 300 mM MgAc, 10 mM tris-HCl (pH 8.0), and 0.6 mM DTT (Sigma-Aldrich) and tissue was homogenized using the “Protein_01_01” protocol. Nuclei were pelleted with a swinging-bucket centrifuge (500 rcf, 5 min, 4 °C; 5920 R, Eppendorf) and resuspended in 1 mL of nuclei permeabilization buffer: 5% BSA, 0.2% IGEPAL CA-630 (Sigma-Aldrich), 1 mM DTT, and 1× EDTA-free protease inhibitor (Roche or Pierce) in PBS. Nuclei were rotated at 4 °C for 5 min before being pelleted again with a swinging-bucket centrifuge (500× g, 5 min, 4 °C; 5920 R, Eppendorf).

Bone marrow

Nuclei were isolated from individual snap-frozen bone marrow. 500 μL chilled OMNI buffer:75 10 mM Tris-HCl, pH 7.5, 10 mM NaCl, 3 mM MgCl2, 0.1% IGEPAL CA-630 (Sigma-Aldrich), 0.1% Tween-20, 0.01% Digitonin (Promega) was added to the sample tube and a homogeneous suspension was obtained by gentle pipetting on ice. Suspension was transferred to a pre-chilled 1.5 mL LoBind tube (Eppendorf) through a 30 μM CellTrics filter (Sysmex). Sample tube was rinsed with another 500 μL chilled OMNI buffer and the suspension was transferred to the same LoBind tube through filter. The sample was kept on ice for 5 min and then pelleted with a swinging bucket centrifuge (500 rcf, 5 min, 4 °C; 5920 R, Eppendorf).

snATAC-seq experiments

Combinatorial barcoding snATAC-seq was performed as described previously43,74,76,77 and the protocol for library preparation can be found here: https://www.protocols.io/edit/snatac-seq-library-generation-using-combinatorial-bpwcmpaw.

Pelleted and permeabilized nuclei were resuspended in 500 μL high salt tagmentation buffer (36.3 mM Tris-acetate, pH = 7.8), 72.6 mM potassium-acetate, 11 mM Mg-acetate, 17.6% DMF) and counted using a hemocytometer. Concentration was adjusted to 2000 nuclei/9 μL, and 2000 nuclei were dispensed into each well of one 96-well plate. For tagmentation, 1 μL barcoded Tn5 transposomes43 was added using a BenchSmart™ 96 (Mettler Toledo), mixed five times and incubated for 60 min at 37 °C with shaking (500 rpm). To inhibit the Tn5 reaction, 10 µL of 40 mM EDTA were added to each well with a BenchSmart™ 96 (Mettler Toledo) and the plate was incubated at 37 °C for 15 min with shaking (500 rpm). Next, 20 µL 2× sorting buffer (2% BSA, 2 mM EDTA in PBS) was added using a BenchSmart™ 96 (Mettler Toledo). All wells were combined into a FACS tube and stained with 3 µM Draq7 (Cell Signaling). Using a SH800 (Sony), 20 2n nuclei were sorted per well into eight 96-well plates (total of 768 wells) containing 10.5 µL EB: 25 pmol primer i7, 25 pmol primer i5, 200 ng BSA (Sigma). Preparation of sort-plates and all downstream pipetting steps were performed on a Biomek i7 Automated Workstation (Beckman Coulter). After addition of 1 µL 0.2% SDS, samples were incubated at 55 °C for 7 min with shaking (500 rpm). 1 µL 12.5% Triton-X was added to each well to quench the SDS. Next, 12.5 µL NEBNext High-Fidelity 2× PCR Master Mix (NEB) were added and samples were PCR-amplified using 72 °C 5 min, 98 °C 30 s, (98 °C 10 s, 63 °C 30 s, 72 °C 60 s) × 11 (bone marrow) or 12 cycles, held at 12 °C. After PCR, all wells were combined. Libraries were purified following the MinElute PCR Purification Kit manual (Qiagen) using a vacuum manifold (QIAvac 24 plus, Qiagen). Size selection was performed with SPRI Beads (Beckmann Coulter, 0.55× and 1.5×) followed by another round of SPRI Bead clean-up (Beckmann Coulter, 1.5×). Libraries were quantified using a Qubit fluorimeter (Life technologies) and the nucleosomal pattern was verified using a Tapestation (High Sensitivity D1000, Agilent). The libraries were sequenced on a HiSeq4000, NextSeq500 or NovaSeq6000 sequencer (Illumina) using custom sequencing primers with following read lengths: 50 + 10 + 12 + 50 (Read1 + Index1 + Index2 + Read2).

snATAC-seq data alignment

Paired-end sequencing reads were demultiplexed allowing up to two mismatched to all possible barcode combinations. Reads were aligned to mm10 reference genome using bowtie278 with default parameters and cell barcodes were added as a BX tag in the bam file. Only primary alignments were kept. Then we removed duplicated read pairs with Picard.79 Only proper read pairs with insert size less than 2000 were kept for further analysis.

TSS enrichment calculation

Enrichment of ATAC-seq accessibility at Transcription Start Sites (TSSs) was used to assess data quality. The method for calculating enrichment at TSS was previously described here.80 Briefly, Tn5 corrected insertions (reads aligned to the positive strand were shifted +4 bp and reads aligned to the negative strand were shifted –5 bp) were aggregated ±2000 bp relative (TSS strand-corrected) to each unique TSS genome-wide. TSS positions were obtained from the GENCODE database vM16. Then this profile was normalized to the mean accessibility ±1900–2000 bp from the TSS and smoothed every 11 bp. The max of the smoothed profile was taken as the TSS enrichment.

Clustering and cell type annotation

We used snapATAC package43 to perform read counting and cell clustering for both all-tissue clustering and tissue-level clustering. First, we removed nuclei with less than 500 fragments or TSS enrichment < 10 for all tissues (except for heart and leg muscle we used TSS enrichment cut-off of 7 to keep more usable cells). Second, we calculated a cell-by-bin matrix at 5000-bp resolution for every sample independently, binarized the matrices and subsequently merged them for each clustering task. Third, we filtered out any bins overlapping with ENCODE blacklist (mm10, http://mitra.stanford.edu/kundaje/akundaje/release/blacklists/mm10-mouse/mm10.blacklist.bed.gz). Fourth, we normalized the read coverage of all bins with log10 (count +1) and Z-score transformation, and only removed bins with absolute Z-scores higher than 2.

After these filtering steps, we calculated Jaccard Index and performed dimensional reduction using the runDiffusionMaps function on similarity matrices. The memory usage of the matrices scales quadratically with the number of nuclei. Therefore, we sampled a subset of 40,000 “landmark” nuclei to compute the matrices and then extended to the rest of the cells when the total number of nuclei exceeded 40,000 (this occurs in the clustering of all tissues, FC, LM and BM). After dimensional reduction, we selected top 20 eigenvectors based on the variance explained by each eigenvector. And then we computed 20 nearest neighbors for each nucleus and applied Leiden algorithm81 to define clusters. Cell clusters were annotated with 1–3 marker genes from previous publications.82,83,84,85 Unknown clusters dominated by low-quality cells (with low TSS enrichment scores) or doublet cells (with two marker genes and high read counts) were identified and removed.

Detection of age-differential cCREs

For each tissue, cCREs (or ATAC-seq peaks) were called using MACS244 with default parameters. Peaks overlapping with high-signal repetitive regions (specifically, _CCCTAA_n,_TTAGGG_n,GSAT-MM,SYNREP_MM from Repeatmasker annotation) were discarded. Sequencing reads from the cells of the same cell type, age and biological replicate were merged into pseudo-bulk BAM files. Then reads were counted by featureCounts function86 on the cCREs in the corresponding tissue. Age-differential cCREs of each cell type were identified by edgeR46 between 18-month and 3-month datasets with the tagwise dispersion estimator and likelihood ratio test with a P-value cutoff of 0.01. Different P-value cut-off or Benjamini–Hochberg (BH)87 adjusted P-value cutoff were also explored and did not change the main conclusions. cCREs with significant P-value and are more accessible in 18-month sample were denoted as age-up cCREs, while cCREs with significant P-value and are less accessible in 18-month sample were denoted as age-down cCREs.

As a comparison, differential cCRE analysis was also performed with edgeR on all 3 age groups, and MAST47 (1.20.0) using the top 20,000 regions found as differential in edgeR 3 mo vs 18 mo and 10,000 randomly sampled regions. It was impossible to run MAST on all peaks due to time and memory constraints. Age was used as a numerical variable in the linear model generated by MAST with default parameters. The comparison of all three methods is summarized in Supplementary information, Fig. S6.

To ensure fair comparison of the number of age-dependent cCREs for each cell type, we down-sampled each sample from each cell type to 1 million reads. Samples with less than 1 million reads were removed. A total of 32 cell types passed the 1 million reads threshold. Then we performed the differential cCRE analysis as stated above for each cell type. The results are summarized in Supplementary information, Fig. S8.

Motif and gene ontology enrichment analysis

Motif enrichment analysis was performed using HOMER88 for the age-differential cCREs in each cell type, with non-differential cCREs as the background. Enriched gene ontology biological pathways were performed by DAVID89 for age-differential cCREs for each cell type as well.

snRNA-seq experiments

Droplet-based Chromium Single Cell 3′ solution90 (10× Genomics, v3 (hippocampus) and v3.1 chemistry (frontal cortex)) was used to generate snRNA-seq libraries as described.74,76 Isolated nuclei were pelleted at 500 rcf for 5 min (4 °C; R5920, Eppendorf; acceleration/deceleration 3/3). Supernatant was removed and pellet was resuspended in 400 µL of sorting buffer: 1 mM EDTA 0.2 U/µL RNase inhibitor (Promega, N211B), 2% BSA (Sigma-Aldrich, SRE0036) in PBS and stained with DRAQ7 (1:100; Cell Signaling, 7406). 75,000 nuclei were sorted using a SH800 sorter (Sony) into 50 µL of collection buffer consisting of 1 U/µL RNase inhibitor in 5% BSA. Sorted nuclei were pelleted at 1000 rcf for 15 min (4 °C; R5920, Eppendorf; acceleration/deceleration 3/3). Nuclei were resuspended in 35 µL of reaction buffer: 0.2 U/µL RNase inhibitor (Promega, N211B), 2% BSA (Sigma-Aldrich, SRE0036) in PBS and counted on a hemocytometer. 12,000 (hippocampus) or 18,000 nuclei (frontal cortex) were loaded onto a Chromium controller (10× Genomics). Libraries were generated using the Chromium Single Cell 3′ Library Construction Kit v3 (10× Genomics; hippocampus (v3.1): PN-1000268, PN-1000120, PN-1000215; frontal cortex (v3): PN-1000075, PN-1000073, PN-120262) according to manufacturer specifications. cDNA was amplified for 12 PCR cycles. SPRISelect reagent (Beckman Coulter) was sued for size selection and clean-up steps. Library quality control was performed using Qubit dsDNA HS Assay Kit (Thermo-Fischer Scientific) and Tapestation High Sensitivity D1000 (Agilent). Libraries were sequenced using NextSeq500, HiSeq4000 or NovaSeq6000 (Illumina) with these read lengths: 28 + 8 + 91 (Read1 + Index1 + Read2).

snRNA data processing

Cellranger90 version 3.0.2 was used to pre-process fastq files from 8 samples (2 replicates for each of 3-month, 10-month, and 18-month). Seurat 3.1.553 was used for subsequent analysis. DoubletFinder 2.0.291 was used to identify and remove doublets from each sample. Seurat’s anchor-based label transfer was used to transfer cluster labels from snRNA to the DH snATAC-seq data. Almost all cells had very high prediction scores, indicating high concordance. Since snRNA-seq and snATAC-seq had different sensitivity for defining cell clusters, we grouped cell clusters in one dataset when all of them were matched to a single cell cluster in the other dataset, to obtain consistent cell type labels for both datasets. Using the transferred labels, we defined 12 matched cell types between the RNA and ATAC data: Ogc, DG, CA1, InhN, Sub_Ent, Asc, CA2/3, Mgc, Opc, Endo, Peri, SMC. These cell-type assignments were subsequently used for gene-cCRE correlation analysis. A pseudo-bulk count table was generated by summing sequencing reads from cells of the same cell type/cluster, age and biological replicate for each gene. Age-differential genes of each cell type/cluster were then identified by edgeR46 between 18-month and 3-month datasets using the likelihood ratio test with an adjusted P-value cutoff of 0.1.

Identification of Gene-cCRE pairs

Cells from the same matched cell types and ages (both snATAC-seq and snRNA-seq) were merged into pseudo bulks, resulting in 36 data points (12 cell type, 3 age groups) for dorsal hippocampus and 48 data points (16 cell types, 3 age groups). For every gene, we computed the weighted Pearson correlation coefficient (WPCC) between the gene transcription levels and the accessibility of any cCRE within 500 kb of the gene TSSs. The number of cells in each cell type is used as the weight, to counter the effect of outlier/extreme values in less abundant cell types. For gene annotations we used GENCODE vM 23 to be consistent with the Cellranger’s annotation. BH adjusted P-value cutoff of 0.05 was used to determine significant gene-cCRE pairs. Gene-cCRE pairs were then used to link age-differential cCREs to age-differential genes.

Gaussian smoothing

We used the R package smoother to perform gaussian smoothing on the number of differential peaks (P-value < 0.001) within each 100 kb region of the genome (smoothing window length of 20). Regions of the genome with a high concentration of differential peaks within a short distance from each other were therefore assigned higher gaussian smoothing scores.

Overlap with histone marks and CTCF-binding sites

We used bedtools intersect -c to overlap all called peaks for each cell type cluster with each of 7 histone ChIP-seq and CTCF ChIP-seq tracks from ENCODE. Fisher’s exact test was used to calculate the enrichment of the overlap of the ChIP-seq called regions with the top 1% of age-associated changing peaks (ordered by P-values calculated using edgeR comparing 3-month and 18-month samples) vs all other (not age-associated) peaks. ENCODE experiment IDs used for overlap analysis are shown in Supplementary information, Table S6. For most experiments, the ENCODE narrowPeak bed files were directly used for overlap analysis with the snATAC-seq data. For LM and HT, histone mark ChIP-seq for 5 time points (E11.5–E15.5) and 7 time points (E11.5–P0) respectively were merged to call peaks. H3K9me3 data from the forebrain were re-aligned to mm10 genome using BWA92 without mapping quality filter (in order not to lose any reads aligning to repetitive elements), and peaks were re-called using SICER93 on both ChIP-seq and input libraries.

Hi-C data processing

To understand the three-dimensional structure of the heterochromatin domains that were reduced during aging, we downloaded Hi-C data from mouse embryonic stem cells.94 Reads were mapped to mm10 genome as previously described95 (https://github.com/ren-lab/hic-pipeline), with a mapping quality filter of 0, to allow interrogation of contacts of the repetitive regions of the genome. First principal components (PC) were computed for the Hi-C matrix. Positive and negative PCs correspond to euchromatin and heterochromatin domains.96

Paired-tag experiments

Paired-tag experiments are carried out as previous described62 with slightly modification. After nuclei isolation with nuclei isolation buffer: 0.2% IGEPAL CA-630 (Sigma #63069), 5% BSA (Sigma #A1595) and 1 mM DTT (Invitrogen #P2325) in PBS (Invitrogen #AM9624), supplemented with 1× Proteinase Inhibitor (Roche #4693132001), 0.5 U/µL SUPERaseIn (Invitrogen #AM2696), and 0.5 U/µL RNase OUT (Invitrogen #10777019), each 300,000 of nuclei were aliquot into the 12 1.5 mL low-bind tubes. Nuclei were spin-down and resuspended in 30 µL MED#1 buffer: 20 mM HEPES (Gibco #15630080), 300 mM NaCl (Invitrogen #AM9760G), 0.5 mM Spermidine (Sigma #S2626), 1× Proteinase Inhibitor, 0.5 U/µL SUPERase In, 0.5 U/µL RNase OUT, 0.01% IGEPAL-CA630, 0.01% Digitonin (Millipore #300410), 2 mM EDTA (Invitrogen #15575020)) and keep on ice. 2 µg of H3K9me3 antibody (Abcam #ab8580) were added into 12 of 200 µL tubes containing 20 µL MED#1 buffer, and pA-Tn5 protein with 12 DNA barcodes were added and incubated at room temperature with gently rotation for 1 h. The 12 tubes of antibody-pA-Tn5 mix were then mixed with each tube of nuclei, respectively and the incubation was carried out in 4 °C with gently rotation overnight. The nuclei were then spun-down and washed two times with MED#2 buffer (20 mM HEPES, 300 mM NaCl, 0.5 mM Spermidine, 1× Proteinase Inhibitor, 0.5 U/µL SUPERase In, 0.5 U/µL RNase OUT, 0.01% IGEPAL-CA630, 0.01% Digitonin) and resuspended in 50 µL MED#2 buffer. Tagmentation reaction were activated out by adding 2 µL of 250 mM MgCl2 (Sigma #63069), carried out in a ThermoMixer set at 37 °C, 550 rpm for 60 min and quenched by adding 16.5 µL of 40 mM EDTA. Nuclei were then spin-down and reverse transcription were carried out with Maxima H minus reverse transcriptase (Thermo #EP0751). Nuclei were then barcoded by ligation-based combinatorial barcoding with T4 DNA Ligase (NEB #M0202L), aliquoted into 2.5–3.5k nuclei sub-libraries and lysed. Library preparation were then carried out as previous described62 and sequenced with read cycles 100 (read1) + 8 (index1) + 8 (index2) + 100 (read2) on a NovaSeq 6000 platform.

Paired-tag data processing

Preprocessing of Paired-tag were carried out with the scripts available from GitHub (https://github.com/cxzhu/Paired-Tag). Briefly, cellular barcodes were extracted from Read2 and assigned to each sample barcodes (12 initial tubes for tagmentation and reverse transcription) and combination of ligated barcodes. Adaptors were trimmed from Read1 and then mapped to the reference genome with bowtie278 (for DNA) and STAR97 (for RNA, with annotation from GENCODE GRCm38.p6). Before generating cell-counts matrices, DNA alignment files were further filtered by removing high-pileup positions (cutoff = 10). Cells with less than 500 unique H3K9me3 loci and 200 unique transcripts were removed from downstream analysis. To remove potential doublets, cells were first clustered with Seurat53 package based on scRNA-seq profiles with resolution = 5, cell groups with both number of DNA and RNA reads per nuclei higher than 5-fold of average reads per nuclei were excluded from further analysis. The remaining cells were again clustered with Seurat package based on scRNA-seq profiles with resolution = 0.5 and annotated based on expression level of marker genes.62

H3K9me3 associated domains (peaks) were called using SICER93 on aggregated H3K9me3 signals from Paired-tag (without input). All default parameters were used, except that window size parameter was set to 5000 and gap size was set to 10000 to detect large peaks. Peaks larger than 100Kb were kept for further downstream analysis (for instance, Fig. 6d).

Quantifying transposable elements (TEs) expression

scTE98 version 1.0 was used to build a genome index for the alignment of reads to genes (gencode vM21) and TEs (rmsk mm10) using scTE_build. The scTE command was used to map reads from the unfiltered BAM files generated by Cellranger to genes and transposable element families, generating a cell by feature read count matrix. Cells with fewer than 100 genes expressed were excluded using “min_genes 100”. A pseudo-bulk count table for both genes and TEs was generated by summing reads from cells of the same cell type, age and biological replicate for each feature. Age-differential genes and TEs for each cell type were then identified by edgeR46 between 18-month and 3-month datasets using the likelihood ratio test. The same strategy was applied on snATAC-seq and Paired-Tag data to quantify the chromatin accessibility and H3K9me3 signal on TEs.

Immunofluorescence staining

Eight mice were used for immunostaining experiments; the 3-month-old and 18-month-old groups included 2 male and 2 female mice each. Mice were perfused intracardially with 4% paraformaldehyde in PBS. After an overnight post-fixation in the same fixative at 4 °C, brain tissues were cut into 50 μm sections with a microtome. Brain sections were blocked with 0.3% Triton X-100 and donkey serum in PBS for 1 h at room temperature and then incubated with H3K9me3 (1:500, Abcam, ab 8898) or Lamin B1 (1:500, Abcam, Cat#229025) or L1-ORF-1p (1:200, Abcam, ab 216324) and CaMKIIα (1:300, ThermoFisher Scientific, MA1–048) primary antibody overnight at 4 °C. Next, brain sections were incubated with Alexa Fluor546-conjugated goat anti-rabbit (Invitrogen, 1:500, A-11035) or Alexa Fluor488-conjugated goat anti-mouse secondary antibodies (Invitrogen, 1:500, A-11029) for 1 h at room temperature, washed in PBS, and mounted in Vectashield containing DAPI (Vector Labs Cat#H-1500).

Image data acquisition and quantitative fluorescence intensity analysis

After immunostaining, the sections were examined, and low- and high-power images were acquired by using a confocal microscope (FV3000, Olympus Microscopy, Japan). The slides were imaged with a 10× or 60× objective with identical settings for all matched images. Image maximum projections, z-stacking of sections, and cell fluorescence intensity measurements were performed by using the Fiji-ImageJ software analysis tools. We measured ~200 excitatory cells and ~100 other cells from young and aged frontal brain sections, respectively, for H3K9me3 staining. We measured ~150 excitatory cells and ~100 other cells from young and aged frontal brain sections, respectively, for Lamin B1 staining. We measured ~200 excitatory cells and ~150 other cells from young and aged frontal brain sections, respectively, for L1-ORF-1p staining. The corrected total cell fluorescence (CTCF) in an arbitrary unit (a.u.) was used for data reporting and statistical analysis.

The Linear Mixed-Effect Model (LME) has been widely used to analyze correlated data. The main idea of LME (“fitlme” in MATLAB) is to take the inherent correlations in correlated data, such as the neurons from the same mouse, into consideration when conducting statistical modeling and hypothesis testing.99 The LME test includes paired t-test and repeated-measures ANOVA as two special cases. The importance of LME and its more generalized versions has been increasingly recognized in recent studies involving large cell sample data collected from a relatively small number of animals. In this study, we used LME for data analysis shown in Fig. 7, in which measurements of staining intensity are presented based on hundreds of cells from 8 mice. We fitted an LME by using age for a fixed effect and mouse group for a random effect.

General data processing and plots

Most of the described data-processing steps (statistical tests, clustering, plotting, and so on) were performed in Python 3.4.5 (www.python.org) and the statistical computing environment R 3.4.3 (www.r-project.org). Box plots were made with ggplot2 (https://cran.r-project.org/web/packages/ggplot2). The elements of the box plots are: center line, median; box limits, upper and lower quartiles; whiskers, 1.5× the interquartile range; points, outliers.