Content uploaded by Anne Brunet
Author content
All content in this area was uploaded by Anne Brunet on Dec 28, 2020
Content may be subject to copyright.
Available via license: CC BY-NC 4.0
Content may be subject to copyright.
Remodeling of epigenome and transcriptome
landscapes with aging in mice reveals widespread
induction of inflammatory responses
Bérénice A. Benayoun,
1,5,6,7
Elizabeth A. Pollina,
1,8
Param Priya Singh,
1
Salah Mahmoudi,
1
Itamar Harel,
1,9
Kerriann M. Casey,
2
Ben W. Dulken,
1
Anshul Kundaje,
1,3
and Anne Brunet
1,4
1
Department of Genetics,
2
Department of Comparative Medicine, Stanford University School of Medicine, Stanford, California
94305, USA;
3
Department of Computer Science, Stanford University, Stanford, California 94305, USA;
4
Paul F. Glenn Laboratories
for the Biology of Aging, Stanford University, Stanford, California 94305, USA
Aging is accompanied by the functional decline of tissues. However, a systematic study of epigenomic and transcriptomic
changes across tissues during aging is missing. Here, we generated chromatin maps and transcriptomes from four tissues and
one cell type from young, middle-aged, and old mice—yielding 143 high-quality data sets. We focused on chromatin marks
linked to gene expression regulation and cell identity: histone H3 trimethylation at lysine 4 (H3K4me3), a mark enriched at
promoters, and histone H3 acetylation at lysine 27 (H3K27ac), a mark enriched at active enhancers. Epigenomic and tran-
scriptomic landscapes could easily distinguish between ages, and machine-learning analysis showed that specific epigenomic
states could predict transcriptional changes during aging. Analysis of data sets from all tissues identified recurrent age-
related chromatin and transcriptional changes in key processes, including the up-regulation of immune system response
pathways such as the interferon response. The up-regulation of the interferon response pathway with age was accompanied
by increased transcription and chromatin remodeling at specific endogenous retroviral sequences. Pathways misregulated
during mouse aging across tissues, notably innate immune pathways, were also misregulated with aging in other vertebrate
species—African turquoise killifish, rat, and humans—indicating common signatures of age across species. To date, our
data set represents the largest multitissue epigenomic and transcriptomic data set for vertebrate aging. This resource iden-
tifies chromatin and transcriptional states that arecharacteristic of young tissues, which could be leveraged to restore aspects
of youthful functionality to old tissues.
[Supplemental material is available for this article.]
The functional decline of organs and tissues is a hallmark of aging,
and it is accompanied by changes in gene expression and chroma-
tin modifications across cell types and tissues (Benayoun et al.
2015; Booth and Brunet 2016; Pal and Tyler 2016; Sen et al.
2016). Aging is the primary risk factor for a variety of chronic diseas-
es, including neurodegeneration, cardiovascular disease, and can-
cer. Several conserved pathways are misregulated during aging,
defining hallmarks or pillars of aging (López-Otín et al. 2013;
Kennedy et al. 2014). One such hallmark is the accumulation of epi-
genetic alterations, defined here as changes to gene regulation by
chromatin modifications. Perturbation in chromatin-modifying
enzymes can extend lifespan in invertebrate models (Benayoun
et al. 2015; Pal and Tyler 2016; Sen et al. 2016), suggesting that
loss of chromatin homeostasis drives aspects of aging. As chroma-
tin marks are relativelystable and can even persist through cell divi-
sion (Kouskouti and Talianidis 2005), sustained alterations to the
chromatin landscape may mediate the propagationof age-associat-
ed functional decline.
Age-dependent changes in chromatin marks (e.g., DNA meth-
ylation, histone modifications) have been observed in multiple
species and tissues (Benayoun et al. 2015; Booth and Brunet
2016; Pal and Tyler 2016; Sen et al. 2016). However, most of this
knowledge has relied on DNA methylation or global assessments
of histone modification changes (e.g., mass spectrometry, western
blot) rather than locus-specific evaluation (e.g., ChIP-seq) (Horvath
2013; Benayoun et al. 2015; Wagner 2017; Cheung et al. 2018).
Several genome-wide studies have interrogated locus-specific
changes in histone modifications and chromatin states, as well as
changes in gene expression in several cell and tissue types with
mammalian aging (e.g., tissue stem cells, liver cells, pancreatic
beta cells, neurons, and T cells) (Rodwell et al. 2004; Cheung
et al. 2010; Liu et al. 2013; Shulha et al. 2013; Bochkis et al. 2014;
Sun et al. 2014; Avrahami et al. 2015; White et al. 2015; Zheng
et al. 2015; Moskowitz et al. 2017; Stegeman and Weake 2017;
Ucar et al. 2017; Nativio et al. 2018). While these studies have
Present addresses:
5
Leonard Davis School of Gerontology, University
of Southern California, Los Angeles, CA 90089, USA;
6
USC Norris Com-
prehensive Cancer Center, Los Angeles, CA 90089, USA;
7
USC Stem
Cell Initiative, Los Angeles, CA 90089, USA;
8
Harvard Medical School,
Boston, MA 02115, USA;
9
Department of Genetics, Silberman Insti-
tute of Life Sciences, The Hebrew University of Jerusalem, Givat
Ram, Jerusalem, 91904, Israel
Corresponding authors: berenice.benayoun@usc.edu,
anne.brunet@stanford.edu
Article published online before print. Article, supplemental material, and publi-
cation date are at http://www.genome.org/cgi/doi/10.1101/gr.240093.118.
© 2019 Benayoun et al. This article is distributed exclusively by Cold Spring
Harbor Laboratory Press for the first six months after the full-issue publication
date (see http://genome.cshlp.org/site/misc/terms.xhtml). After six months, it
is available under a Creative Commons License (Attribution-NonCommercial
4.0 International), as described at http://creativecommons.org/licenses/by-
nc/4.0/.
Resource
29:1–13 Published by Cold Spring Harbor Laboratory Press; ISSN 1088-9051/19; www.genome.org Genome Research 1
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
provided important insights into genome-wide chromatin and
transcriptome remodeling with age, they have remained restricted
to specific cell types and/or a single histone mark. Thus, whether
general rules and patterns govern age-related chromatin and tran-
scriptional changes with age—and how they are linked—remains
largely unknown.
Results
A genome-wide epigenomic and transcriptomic landscape of four
tissues and one primary cell type during mouse aging
To understand how chromatin and transcriptional profiles change
during aging, we collected several tissues and cells from C57BL/6N
male mice at three different time points: youth (3 mo), middle age
(12 mo), and old age (29 mo). We focused on a subset of tissues—
heart, liver, cerebellum, and olfactory bulb—that are known to dis-
play age-related functional decline and are clearly anatomically de-
fined. We also derived primary cultures of neural stem cells (NSCs)
from these young, middle-aged, and old mice. For each tissue or cell
culture from all three ages, we generated transcriptomic maps (i.e.,
RNA-seq) and epigenomic maps (i.e., ChIP-seq of total Histone 3
[H3] for normalization, trimethylation of Histone 3 at lysine 4
[H3K4me3], and acetylation of Histone 3 at lysine 27 [H3K27ac]),
yielding 143 data sets (Fig. 1A,B; Supplemental Fig. S1A,B;
Supplemental Table S1A). We chose H3K4me3 and H3K27ac
because both marks are associated with ‘active chromatin’and
because spread of these marks is linked to cell identity and specific
transcriptional states.Indeed, H3K4me3 and H3K27ac are preferen-
tially enriched at active promoters and activeenhancers, respective-
ly (Heintzman et al. 2007). Broad H3K4me3 domains that spread
beyond the promoter region are known to mark genes that are im-
portant for cell identity and function (Bernstein et al. 2006;
Benayoun et al. 2014; Chen et al. 2015) and exhibit increased tran-
scriptional levels (Chen et al. 2015) and consistency (Benayoun
et al. 2014). Large clusters of H3K27ac-enriched enhancers that
spread beyond a simple enhancer region are known as ‘super-en-
hancers’(Hnisz et al. 2013) or ‘stretch-enhancers’(Parker et al.
2013), and they mark enhancers of genes that are cell- or tissue-
specific and highly transcribed in that specific cell or tissue.
Importantly, ChIP-seq data sets for the H3K4me3 and H3K27ac his-
tone mark were normalized to paired total Histone H3 ChIP-seq
data to account for potential changes in the local nucleosome land-
scape. Quality metrics and correlation assessment indicated that
most samples were of good quality and well-correlated acrosstissues
and ages (Supplemental Table S1B–E; see Supplemental Material).
To visualize the similarity of our genomic samples, we used
multidimensional scaling (MDS) (Chen and Meltzer 2005). MDS
analysis on RNA, H3K4me3 intensity (i.e., read density of
H3K4me3 ChIP-seq per base pair normalized to H3 ChIP-seq den-
sity), H3K4me3 breadth (i.e., genomic spread of the H3K4me3
peaks), H3K27ac intensity (i.e., read density of H3K27ac ChIP-seq
per base pair normalized to H3 ChIP-seq density), or H3K27ac su-
per-enhancers (i.e., intensity of H3K27ac ChIP-seq at enhancer
clusters) (Hnisz et al. 2013) revealed that, as expected, the main
source of sample separation is the nature of the tissue, regardless
of age (Fig. 1C–F; Supplemental Fig. S1D,E). Principal component
analysis (PCA), another method to visualize similarity of genomic
samples (Ringnér 2008), yielded similar results (Supplemental
Fig. S1F,G). Thus, we used MDS analysis for subsequent analyses.
Our results are consistent with the observation that RNA profiles,
H3K4me3, and H3K27ac are associated with cell identity (Hnisz
et al. 2013; Benayoun et al. 2014; Wagner et al. 2016) and indicate
that overall, tissue and cell identities remain quite stable during
aging.
To understand how age impacts the global epigenomic
and transcriptomic landscapes in each tissue or cell type, we per-
formed MDS or PCA on individual tissues/cell types (Fig. 2A–J;
Supplemental Fig. S2A–T). In all tissues and for all features (RNA,
H3K4me3 intensity and breadth, H3K27ac intensity and breadth),
there was a progressive separation based on age, with the young
samples clustering closer to the middle-aged samples and further
away from the old samples (Fig. 2A–J; Supplemental Fig. S2A–T).
For primary NSC cultures, there was also a separation with age for
H3K4me3 intensity, H3K4me3 breadth, and H3K27ac super-en-
hancers (Supplemental Fig. S2F,I,O). However, the transcriptome
and H3K27ac intensity of NSCs did not separate well with respect
to age (Supplemental Fig. S2C,L), possibly because of technical
noise. Our observations are consistent with previous reports of
age-associated epigenetic changes at enhancers in liver tissue and
pancreatic beta cells (Avrahami et al. 2015; Cole et al. 2017).
Thus, genome-wide RNA and features of H3K4me3 and H3K27ac
deposition can distinguish between ages across tissues and cells.
Machine learning reveals that age-related epigenomic changes
can predict transcriptional changes
To understand how age-related changes in the epigenome predict
age-related transcriptional changes, we took advantage of machine
learning (Fig. 3A; Supplemental Fig. S3A). Using four algorithms
(i.e., neural networks [NNETs], support vector machines [SVMs],
gradient boosting [GBM], and random forests [RFs]), we trained
models to discriminate between transcriptional changes with age
—up-regulated, down-regulated, or unchanged gene expression
(Fig. 3A; Supplemental Fig. S3A). As potential predictors for the
models, we used, for each gene, features from chromatin data sets
generated in this study (e.g., H3K4me3 signal at the promoter, or
breadth of H3K4me3), features from mouse ENCODE ChIP-seq
data sets in heart, liver, cerebellum, and olfactory bulb in young
mice (e.g., POLR2A, H3K27me3, etc.) (Supplemental Table S2A;
Shen et al. 2012; Yue et al. 2014), and features from the underlying
DNA sequence (e.g., %CpG in promoter, etc.). The models were
trained with sets of features that were either (1) “dynamic”—
reflecting the age-related changes in the chromatin environment
of genes (e.g., age-related changes in H3K4me3 breadth), (2) “stat-
ic”—describing the youthful state of the chromatin or DNA se-
quence at the gene (e.g., H3K27ac in young tissue), or (3) both
(Fig. 3A; Supplemental Fig. S3A). Because absolute gene expression
levels can influence the ability to call differential gene expression
(Oshlack and Wakefield 2009), we also included, as a feature, the
average expression level of the genes in the young samples.
All machine-learning algorithms assigned genes to the correct
transcriptional change with age 67%–81% of the time on average,
significantly above that of a random classification (50%) (Fig. 3B,C;
Supplemental Fig. S3B,C;Supplemental Table S3A,B). Models de-
rived using tree-based algorithms (GBM and RF) performed slightly
better than other models (69%–81% vs 67%–75%) (Supplemental
Table S3B). The accuracy was similar whether validation was per-
formed within or across tissues (Fig. 3B,C; Supplemental Table
S3B;Supplemental Fig. S3B,C). These observations suggest that
genes that are misregulated with age share common epigenomic
signatures, even if these genes and loci are different across tissues
(see below) (Fig. 4). Models trained with only static ordynamic fea-
tures also had predictive power, but models trained with both types
Benayoun et al.
2 Genome Research
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
of features performed better (Supplemental Table S3B;Supplemen-
tal Fig. S3F–I). Key predictors of age-related transcriptional changes
in all tissues were dynamic features, including changes in the
amount of H3K27ac at enhancers or changes in the breadth of
the H3K4me3 domains during aging (Fig. 3D,E; Supplemental
Fig. S3D,E). Other predictors of age-related transcriptional changes
were static features, describing the young chromatin context (e.g.,
H3K4me3 promoter intensity, H3K4me3 domain breadth) (Fig.
3D,E). While this may result from incomplete accounting for differ-
ences in gene expression (The ENCODE Project Consortium 2012),
the active chromatin context of a gene in youth might predict
changes in expression of that gene with age, perhaps because active
EF
B
A
C
D
Figure 1. A genome-wide epigenomic and transcriptomic landscape in four tissues and one cell type during mouse aging. (A) Experimental setup (see
Supplemental Table S1). (B) Example genome browser region showing tracks of data sets in cerebellum tissue at different ages. (Chr) Chromosome. (C–F)
Multidimensional scaling analysis results across data sets based on RNA expression (C), H3K4me3 peak intensity (D), H3K4me3 peak breadth (E), or
H3K27ac peak intensity at all peaks (F). For RNA-seq data, the input was a matrix of log
2
-transformed DESeq2 1.6.3 normalized counts. For chromatin
marks, the most intense or broadest peak associated with a gene was used when more than one peak was present, and the log
2
-transformed DESeq2
1.6.3 normalized intensity or breadth was used as input.
Genomic landscape with aging in mice
Genome Research 3
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
EF
BA
CD
GH
IJ
Figure 2. Separation of samples across tissues and cell types as a function of age. Multidimensional scaling analysis results across samples derived from
specific tissues, liver and cerebellum, based on RNA expression (A,B), H3K4me3 peak intensity (C,D), H3K4me3 peak breadth (E,F), H3K27ac peak intensity
(all peaks: G,H; super-enhancers only: I,J). For RNA-seq data, the input was log
2
-transformed DESeq2 1.6.3 normalized counts. For chromatin marks, the
most intense or broadest peak associated with a gene was used when more than one peak was present, and the log
2
-transformed normalized intensity or
breadth was used as input.
Benayoun et al.
4 Genome Research
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
loci are more impacted by stresses throughout life. Thus, static and
dynamic chromatin states can both predict age-dependent changes
in transcription.
Immune response pathways are robustly up-regulated during
mouse aging
We asked if some genes are consistently misregulated with age
across tissues (Fig. 4A,B; Supplemental Fig. S4A–E;Supplemental
Fig. S5A–F). We identified 16 genes whose expression was up-reg-
ulated with aging in all tissues (and 0 down-regulated) when
each tissue was assessed separately (FDR < 5%) (Fig. 4A,B). These
genes encode complement and coagulation factors (i.e., C1QA,
C1QC, C4), interferon-response related proteins (i.e., IFI27l1,
IFIT3, IFITM3), and a protein involved in leukocyte trans-endothe-
lial migration (i.e., ITGB2) (Guan et al. 2015). Though the overlap
is small, these results suggest that a common response to aging
across tissues is related to immunity. Genes up-regulated with
age in our data set overlap with genes up-regulated with age in
mouse liver (Bochkis et al. 2014; White et al. 2015) and astrocytes
(Boisvert et al. 2018), and common genes are also involved in the
interferon response and the complement and coagulation cascade
(Supplemental Fig. S6A–I). Analyzing age-related transcriptional
changes combining all tissues and ages, but including tissue of
E
B
A
C
D
Figure 3. Machine-learning analysis reveals that changes in enhancer score and H3K4me3 domain breadth with age can predict transcriptional aging.
(A) Scheme of the three-class machine-learning pipeline. (NNET) Neural network, (SVM) support vector machine, (RF) random forest, (GBM) gradient
boosting machine. (B,C) Balanced classification accuracy over the three classes across tissues for random forest models (B) or gradient boosting machine
models (C). The accuracy of the model trained in a specific tissue on the same tissue (e.g., the liver-trained model on liver data) is measured using held-out
validation data. For cross-tissue validation, the entire data of the tested tissue were used. ‘Random’accuracy illustrates the accuracy of a meaningless model
(∼50%). All tests were more accurate than random. The robustness of the prediction is supported by the fact that samples for RNA and chromatin profiling
were collected from independent mice at two independent times (Supplemental Table S1A). Balanced accuracy across the three classes is reported. (D,E)
Feature importance from random forest models (D; Gini score and mean decrease in accuracy) or gradient boosting machine models (E; Gini score). High
values indicate important predictors. See two-class models in Supplemental Figure S3. Note that two-class models, though containing less biological in-
formation, outperformed three-class models, which is consistent with the increased complexity of a classification problem with the number of classes to
discriminate.
Genomic landscape with aging in mice
Genome Research 5
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
E
F
B
A
CD
G
H
Figure 4. Misregulated pathways during aging reveal the activation of an inflammatory innate immune signature. (A,B) Venn diagram for the overlap of
significantly up-regulated (A) or down-regulated (B) genes with aging in each tissue called by DESeq2 1.6.3 at FDR< 5%. (C–F) Functional pathway enrich-
ments (C,E–G) and transcription factor (TF) target enrichments (D) using the minimum hypergeometric test for differential RNA expression (C,D),
H3K4me3 intensity (E), H3K4me3 breadth (F), and H3K27ac intensity (all enhancers) (G). Enriched pathways were plotted if four out of the six tests
(RNA) or three out of the five tests (chromatin marks) were significant (FDR < 5%). (H) Heat maps of expression for all repetitive elements with significant
differential expression with aging (TEtranscripts quantification and DESeq2 1.16.1 statistical test at FDR < 5%). Analysis of repetitive elements using
HOMER, and overlap with TEtranscripts, is reported in Supplemental Table S6A–E.
Benayoun et al.
6 Genome Research
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
origin as a covariate, identified 771 genes up-regulated in a tissue-
independent manner (and 174 genes down-regulated with age;
FDR < 5%) (Supplemental Fig. S4F;Supplemental Table S4). These
genes also include complement and interferon response genes.
Thus, a subset of immune response genes is commonly up-regulat-
ed across tissues during aging.
In contrast, we did not observe any recurrently misregulated
loci across tissues with age at the epigenomic level (H3K4me3 or
H3K27ac marks, FDR < 5%) (Supplemental Fig. S5A–F). The obser-
vation that transcriptomic changes, but not epigenomic changes,
are recurrent with age between tissues could be due to differences
in sensitivity between read-outs or to the fact that the same gene
can be regulated by different regulatory elements in diverse tissues.
Next, we investigated whether specific pathways are recur-
rently misregulated with age across tissues (Fig. 4C–G; Supplemen-
tal Fig. S4G–I). Several pathways were consistently misregulated
with aging, not only at the transcriptome level, but also at the chro-
matin level (Fig. 4C,E–G; Supplemental Fig. S4G–I). Down-regulat-
ed pathways included mitochondrial function (e.g., ‘oxidative
phosphorylation’), consistent with previous work in mouse and
human tissues (Zahn et al. 2006, 2007). Up-regulated pathways
included protein homeostasis (e.g., ‘lysosome’,‘ribosome’) or im-
mune signaling pathways (e.g., ‘inflammatory response’,‘interfer-
on alpha response’) (Fig. 4C,E-G; Supplemental Fig. S4G–I), in line
with previous observations in a number of aging tissues or cells
(Stegeman and Weake 2017), such as choroid plexus (Baruch
et al. 2014), kidney (Rodwell et al. 2004; O’Brown et al. 2015), liver
(Bochkis et al. 2014; White et al. 2015), and astrocytes (Boisvert
et al. 2018). Complement and coagulation-related pathways were
also significantly up-regulated with age across tissues and cell types
(Fig. 4C; Supplemental Fig. S4G), consistent with findings in aging
astrocytes (Boisvert et al. 2018). However,the strongest signal came
from the interferon response pathways, which were significantly
induced across aged tissues at both the transcriptional and chroma-
tin levels (Fig. 4C,E–G; Supplemental Fig. S4G–I). The transcrip-
tional activation of the interferon response was confirmed by
Ingenuity Pathway Analysis (e.g., IFNG, IFNB1, IFNAR) (Supple-
mental Table S5A). This enrichment for immune signaling path-
ways (and other pathways) was also observed when re-analyzing
previously published RNA-seq data sets in aging liver (Bochkis
et al. 2014; White et al. 2015) and astrocytes (Supplemental Fig.
S6J,K; Boisvert et al. 2018). While an age-related inflammatory re-
sponse has been reported at the transcriptional level (Stegeman
and Weake 2017; Boisvert et al. 2018), this is the first time it is ob-
served at both transcriptional and chromatin levels.
Interferon response activation can stem from (1) exogenous
viral infection, (2) reactivation of endogenous transposable ele-
ments (TEs) (De Cecco et al. 2013; Wood and Helfand 2013),
and (3) aberrant cytosolic DNA detection by the cyclic GMP-
AMP synthase (cGAS) pathway (Sun et al. 2013; West et al.
2015). As old and young mice were kept in specific-pathogen-
free facilities and were documented to not have viral infection,
we first queried TE expression in the different tissues during aging
using our RNA-seq data sets. Increased TE activity has been report-
ed with aging in several species (Maxwell et al. 2011; De Cecco
et al. 2013; Wood and Helfand 2013; Van Meter et al. 2014). Using
the TEtranscripts (Jin et al. 2015) and HOMER (Heinz et al. 2010)
repeats pipelines, we identified repeats whose transcription levels
were significantly changed—mostly up-regulated—with aging
(Fig. 4H; Supplemental Table S6A–E). The most significantly up-
regulated repetitive elements belonged to endogenous retrovirus
(ERV) families (Fig. 4H). Consistently, H3K4me3 and H3K27ac in-
tensity at several of these repeat families was remodeled during ag-
ing (Supplemental Fig. S5G,H;Supplemental Table S6F–I).
The interferon signaling pathway up-regulation is also
compatible with the significant up-regulation of the “cytosolic
DNA-sensing pathway,”corresponding to cGAS activation (Sup-
plemental Fig. S4G). The cGAS pathway is up-regulated in senes-
cent cells in response to aberrant cytoplasmic chromatin (Dou
et al. 2017) and deficient mitochondrial DNA—a known conse-
quence of aging (West et al. 2015). Thus, activation of the cGAS
pathway by endogenous DNA may also play a role in the age-relat-
ed increase in the interferon response.
Consistent with functional pathway enrichment results, tar-
get genes of pro-inflammatory transcription factors IRF8 and
TCF3 were significantly up-regulated with aging (Fig. 4D). Similar-
ly, targets of pro-inflammatory transcription factors IRF3, IRF5,
and IRF7 were up-regulated across tissues according to Ingenuity
Pathway Analysis (Supplemental Table S5A). Targets of FOXO fac-
tors were also up-regulated with aging (Fig. 4D). As FOXO factors
are known to be pro-longevity genes (Martins et al. 2016) and to
modulate innate immunity (Seiler et al. 2013), this up-regulation
may result from a compensatory mechanism and could contribute
to the up-regulation of the innate immune response with aging.
MYC targets were also up-regulated (Fig. 4D), consistent with
MYC’s reported pro-aging effects (Hofmann et al. 2015). Finally,
targets of the RNA binding protein TARDBP (also known as TDP-
43) were significantly down-regulated with age across tissues
(Fig. 4D). Mutations in human TARDBP (also known as TDP-43)
are involved in the pathogenesis of amyotrophic lateral sclerosis
and frontotemporal dementia (Scotter et al. 2015), and TDP-43
has been suggested to play a role in retrovirus suppression by
host cells (Ou et al. 1995) and in microglia activation (Zhao et al.
2015). Thus, misregulation of targets of several transcription fac-
tors and RNA binding proteins may be critical for the induction
of innate immune response pathways with age.
Finally, we asked if the transcriptional increase in immune
pathways in tissues with aging could result from the transcriptome
of infiltrated immune cells or from other changes in cellular com-
position (Rodwell et al. 2004; Lumeng et al. 2011; Pinto et al. 2014;
O’Brown et al. 2015; Teschendorff and Zheng 2017). Using
CIBERSORT to perform de-convolution of aging tissue RNA-seq
data sets (Newman et al. 2015), no significant change could be de-
tected in the proportions of predicted inflammatory cell signatures
(Supplemental Fig. S7A–E;Supplemental Table S7A–D). However,
some known immune markers were up-regulated with age
(Supplemental Fig. S7F,G). Thus, a portion of the observed inflam-
matory response might be due to a low amount of infiltrated im-
mune cells in old tissues. Our de-convolution analysis did not
detect significant changes in the proportion of other cell types
in tissue samples (e.g., fibroblasts, astrocytes/neurons, hepato-
cytes, cardiomyocytes, etc.) (Supplemental Table S7D), though
we cannot exclude that actual changes exist below the sensitivity
threshold of the algorithm (Teschendorff and Zheng 2017).
Single-cell RNA-seq will be needed to fully address the impact of
cell composition on transcriptomic changes in aging tissues.
Thus, several factors may contribute to the up-regulation of in-
flammatory responses in old tissues.
Conservation of age-regulated transcriptional trajectories across
vertebrate species
To investigate whether the age-related changes observed across
mouse tissues are conserved in other vertebrate species, we used
Genomic landscape with aging in mice
Genome Research 7
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
publicly available aging transcriptome data sets in rat (Yu et al.
2014), human (The GTEx Consortium 2015), and the naturally
short-lived African turquoise killifish (Baumgart et al. 2014,
2016). We identified rat, human, and killifish orthologs for each
mouse gene that was significantly misregulated with age.
Notably, the interferon alpha and gamma response pathways
were significantly misregulated in rat,human, and killifish samples
(Fig. 5A; Supplemental Fig. S8A). In addition, similar aging trajecto-
ries (i.e., up-regulation or down-regulation with age) were observed
for the same genes in similar tissues across vertebrate species (Fig.
5B; Supplemental Fig. S8B). These trajectories were less conserved
in the GTEx human data, perhaps because other factors (e.g.,
B
A
Figure 5. Age-related transcriptional signatures are overall conserved across vertebrate species. (A) Functional enrichments using the minimum hyper-
geometric test for differential RNA expression with aging in mouse, rat, human, and killifish samples. The mouse data are a subset of Figure 4C and are
plotted as a reference. (B) DESeq2 1.6.3 normalized log
2
fold changes per unit of time for genes orthologous to differentially expressed mouse genes
in rat, human, and killifish samples. The mouse data are plotted for comparison. P-values were calculated using a one-sample, one-sided Wilcoxon test
to test the differences between observed fold changes and 0 (i.e., no change with age). Only male data are plotted. Data with the contribution of females
(when available) are in Supplemental Figure S8B.
Benayoun et al.
8 Genome Research
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
environment, diseases) overshadow aging differences in human
tissues. Indeed, when accounting for body mass index and meta-
bolic status in an independent human liver microarray data set
(GSE61260) (Horvath et al. 2014), gene expression trajectories
with aging were more similar between mouse and human
(Supplemental Fig. S8C,D). Collectively, these data indicate that
core signatures of innate immune responses are consistently up-
regulated with aging across vertebrate species.
Discussion
A resource for the study of aging
To understand the effect of aging on genomic regulation and chro-
matin identity with aging, we have generated transcriptomic and
epigenomic maps in young, middle-aged, and old mice from a va-
riety of tissues and cells known to functionally decline with age
(i.e., heart, liver, cerebellum, olfactory bulb, primary NSCs). To
our knowledge, this data set is the largest epigenomic and tran-
scriptomic data set for mammalian aging to date and will serve
as a resource for the aging community. It is one of the rare cases
with a middle-aged point, in addition to young and old time
points, which helps understand epigenomic and transcriptomic
aging as trajectories rather than end-point results. Thanks to the
inclusion of this middle-age time point, we find that progressive
changes accumulate throughout mouse lifespan not only at the
transcriptional level but also at the level of several chromatin fea-
tures. The progressive accumulation of remodeling of histone
modifications with aging is reminiscent of the DNA methylation
clock paradigm (Horvath 2013; Cole et al. 2017; Quach et al.
2017; Stubbs et al. 2017; Wang et al. 2017). The existence of these
progressive changes is compatible with the existence of chromatin
modification clocks. Additional time points and individuals will
be required to build such clocks and compare their performance
to that of the well-established DNA methylation clock. This
data set could also be integrated to future studies with additional
marks. The potential interaction between different molecular
clocks could provide key insights into the regulation of cellular
and organismal aging.
The transcriptomic arm of our data set is consistent with the
wealth of published transcriptional aging data sets (Stegeman and
Weake 2017), at least at the pathways level, with an increase in in-
flammation and a decrease in mitochondria function. Several stud-
ies have started to interrogate genome-wide chromatin remodeling
with vertebrate aging (Cheung et al. 2010; Liu et al. 2013; Shulha
et al. 2013; Bochkis et al. 2014; Sun et al. 2014; Avrahami et al.
2015; Zheng et al. 2015; Cole et al. 2017; Moskowitz et al. 2017;
Ucar et al. 2017), with concomitant changes in transcription.
What is unique to our study is the combination of cross-tissue as-
sessment (i.e., heart, liver, cerebellum, olfactory bulb, primary
NSCs), multiple chromatin feature profiling (i.e., total H3,
H3K4me3, H3K27ac), and the inclusion of three ages, thereby
allowing us to conduct an integrated study of conserved and coor-
dinated genomic misregulation with mammalian aging. This
resource will help identify candidate regulators that affect age-
dependent dysfunction across multiple tissues in vertebrates.
Machine learning as a powerful tool to study
aging epigenomics
By using machine learning, we show that age-related epigenomic
remodeling is predictive of age-related transcriptional changes.
This is consistent with the ‘histone code hypothesis,’whereby
the chromatin context may direct transcriptional outputs
(Jenuwein and Allis 2001), and it supports the idea that the rules
that govern the relationship between the chromatin landscape
and transcriptional outputs are mostly preserved throughout
life. However, these models cannot be used to infer directionality
or causative nature of the flow of information between chromatin
and transcriptional changes, and age-related transcriptional
changes might precede or even guide observed chromatin chang-
es. What machine-learning models do indicate is that changes at
the chromatin and transcription levels with aging are to some ex-
tent coordinated and that the breakdown of gene regulation with
age is a complex process. In this study, we focused on active chro-
matin marks (H3K4me3 and H3K27ac) because of their associa-
tion with cell identity in several organisms and with lifespan in
invertebrates (Benayoun et al. 2015). Machine-learning models
built with these active marks are fairly accurate, in line with the
observation that active chromatin marks are the most predictive
of expression levels by the ENCODE Consortium (The ENCODE
Project Consortium 2012). Nevertheless, changes in constitutive
and/or facultative heterochromatin marks (e.g., DNA methyla-
tion, H3K9me3, H3K27me3) are major events during aging
(Tsurumi and Li 2012; Benayoun et al. 2015). Thus, information
about age-related remodeling of heterochromatin marks may fur-
ther improve the predictive power of machine-learning models.
Key predictors of the relationship between chromatin and tran-
scriptional aging could provide relevant candidates for future
functional studies of aging epigenomics. These models and their
features may also have a broader applicability in other contexts,
including development and disease.
Innate immune pathways are broadly induced during
aging across tissues and species
Our analyses reveal that immune pathways are broadly up-regulat-
ed with aging across tissues and species. This increase in immune
activity in the absence of exogenous pathogens is consistent with
the concept of “inflamm-aging”(Xia et al. 2016). We find that in-
terferon response pathways, both alphaand gamma, are recurrent-
ly and robustly activated with vertebrate aging across tissues.
Although interferon signaling is traditionally associated with the
response to viral infection, the interferon pathway can also be in-
duced in response to mitochondrial DNA stress and cytosolic DNA
detection (Sun et al. 2013; West et al. 2015) and reactivation of en-
dogenous transposable elements (TEs) (De Cecco et al. 2013; Wood
and Helfand 2013). Our analyses find evidence for induction of cy-
tosolic DNA-sensing pathway genes as well as a significant up-reg-
ulation of several families of TEs. Many TEs can retain viral
characteristics, including the ability to replicate, form viral parti-
cles, and trigger host immune responses (Kassiotis and Stoye
2016). Thus, the global increase in innate immunity signals across
tissues with aging may be mediated through the detection of en-
dogenous aberrant DNA or reactivated endogenous retroviral par-
ticles. The increased interferon response could also be due to
infiltrated immune cells. Further studies, especially at the single-
cell level, will be needed to disentangle the relative contribution
of infiltrated immune cells and endogenous detection of aberrant
DNA. Our study indicates that inflammation is a conserved hall-
mark of aging, and it identifies candidate factors that could be in-
volved in this phenomenon. This information should help define
strategies to counter aging and age-related diseases.
Genomic landscape with aging in mice
Genome Research 9
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
Methods
Chromatin preparation, quantification, and immunoprecipitation
We performed ChIP experiments on different tissues and one
cell type from independent animals (Supplemental Table S1A).
ChIP experiments were performed on tissues and cells using a stan-
dard protocol (Webb et al. 2013; Benayoun et al. 2014). For liver,
heart, and cerebellum, chromatin content was measured and
equalized for all ages to enable comparison across samples of a tis-
sue. We used 20 µg of chromatin for the H3 ChIPs, 50 µg for the
H3K4me3 ChIPs, and, respectively, 70 µg (heart) or 100 µg (liver,
cerebellum) for the H3K27ac ChIPs. For the olfactory bulb, chro-
matin from ∼30 mg of tissue was used for immunoprecipitation
with anti-H3 antibody, and 60 mg was used for immunoprecipita-
tion with anti-H3K4me3 and -H3K27ac antibodies. For NSCs, we
used chromatin from ∼250,000 cells for the H3 ChIPs, ∼750,000
cells for the H3K4me3 ChIPs, and ∼1,000,000 for the H3K27ac
ChIPs. Immunoprecipitations were performed with the following
antibodies: 5 µg H3K4me3 antibody (Active Motif #39159, lot
#1609004), 5 µg Histone H3 (Abcam #1791, lot #GR178101-1),
and 7 µg H3K27ac (Active Motif #39133, lot #1613007) (see
Supplemental Material for more details).
Next-generation sequencing ChIP library generation
For olfactory bulb libraries and the first set of NSCs libraries, librar-
ies were generated with the Illumina TruSeq kit (IP-202-1012) ac-
cording to the manufacturer’s instructions. Briefly, repaired
and adapter-ligated DNA was size-selected in the range of 250–
400 bp and PCR-amplified for 16 (H3), 17 (H3K4me3), and 18–
19 (H3K27ac) cycles. After the TruSeq kit became backordered,
we generated libraries for the liver, heart, cerebellum, and the sec-
ond set of H3 and H3K4me3 NSCs libraries using the NEBNext
DNA library prep kit (E6040L). Repaired and adapter-ligated DNA
was size-selected in the range of 250–400 bp using agarose gel elec-
trophoresis and PCR-amplified for 14 (H3), 16–17 (H3K4me3), or
17–18 (H3K27ac) cycles. Library quality was assessed using the
Agilent 2100 Bioanalyzer (Agilent Technologies). Single-end 101-
bp reads were generated on Illumina HiSeq 2000 machines at the
Stanford Genome Center.
ChIP-seq data processing
ChIP-seq data sets were processed using a standard data processing
pipeline, including quality trimming, mapping to the mm9 assem-
bly, and duplicate removal. Significant peaks were called using
MACS2 2.0.8 (Zhang et al. 2008) with the ‐‐broad option to enable
detection of wider enrichment regions (see Supplemental Material
for more details).
Statement on the use of the mm9 assembly
We used the GRCm37 (mm9) assembly to map all sequencing
reads from mouse origin in this study (RNA-seq and ChIP-seq)
because many programs did not yet support the mm10 build
when we started our study. Because our study compares samples
across different ages and does not perform absolute analyses, re-
aligning the reads to GRCm38 (mm10) should not significantly af-
fect our conclusions.
H3K4me3 breadth remodeling analysis
To compare changes in the breadth of H3K4me3 domains, we im-
proved upon our pipeline to computationally adjust samples such
that the signal-to-noise ratio across all peaks is equalized between
samples (Benayoun et al. 2014). We created a reference peakset
for all comparative analyses using pooled QC reads from all ages
and replicates (hereafter referred to as ‘metapeaks’). To match the
signal-to-noise ratios across all aging samples, we down-sampled
reads separately in each H3K4me3 ChIP-seq biological sample to
match the coverage histogram across all samples over the meta-
peaks intervals, similar to Benayoun et al. (2014). This procedure
matches the “height”of the peaks from the peak caller’s point of
view. The appropriate down-sampling rate that allows the coverage
histogram of higher sensitivity H3K4me3 ChIP-seq samples to be
equal or lower than that of the lowest sensitivity H3K4me3 ChIP-
seq sample was determined by minimizing the P-value of the
Kolmogorov-Smirnov test (comparison to the sample with lowest
H3K4me3 ChIP-seq sensitivity). To limit the effect of variations
in input sample depth, we also matched the effective depth of H3
input samples to that of the lowest available sample. Final
H3K4me3 domain breadth calls per samples were performed by us-
ing MACS2 2.08 with the same parameters as above. IntersectBed
(BEDTools 2.16.1) (Quinlan and Hall 2010) was used to estimate
the length coverage of the sample peaks over the reference meta-
peaks. This pipeline increases the likelihood that called gains/losses
of breadth result from a change in breadth of the enriched region
and not simply from an underlying difference in H3K4me3 inten-
sity. Differential breadth was estimated using R (R Core Team 2018)
and the DESeq2 R package (DESeq2 1.6.3) (Love et al. 2014).
Super-enhancer calling
Super-enhancers were called as outlined in Hnisz et al. (2013).
Briefly, MACS2 H3K27ac peaks were stitched together if within
12.5 kb of one another (Hnisz et al. 2013), using mergeBed from
BEDTools 2.16.0. Reads mapping within these peaks were counted
using intersectBed from BEDTools 2.16.0, and the ROSE algorithm
(Hnisz et al. 2013) was used to determine the H3K27ac intensity
inflexion point determining typical versus super-enhancers.
H3K4me3 and H3K27ac intensity remodeling analysis
Similar to the above, we created reference peak sets (i.e., “meta-
peaks”) for all comparative analyses using pooled QC reads from
all ages and replicates. Intensity signals for histone H3 modifica-
tions normalized to the local H3 occupancy were obtained using
the “DiffBind”R package (DiffBind 1.12.3) (Ross-Innes et al.
2012). Normalized intensities were then used to estimate differen-
tial intensities as a function of age using the DESeq2 R package
(DESeq2 1.6.3) (Love et al. 2014).
Cell and tissue isolation for RNA purification
For RNA isolation, we used a new cohort of aging male C57BL/6N
mice (same ages as the ChIP-seq cohort), and RNA-seq data sets
were generated at a later time than the ChIP-seq data sets
(Supplemental Table S1A). For RNA extraction on tissues: Olfacto-
ry bulbs were microdissected from 3-mo-, 12-mo-, and 29-mo-old
C57BL/6N male mice and weighed, and tissues from two indepen-
dent mice of the same age were pooled per biological replicate. Cer-
ebellum samples were dissected, weighed, and samples from two
mice of the same age were pooled per biological replicate. For the
liver, the leftmost part of upper left lobe of the liver was dissected
and weighed from an individual mouse and was used for a single
biological replicate. For the heart, following removal of blood,
the bottommost part of heart ventricles from an individual mouse
was dissected, weighed, and used as a single biological replicate. All
tissue samples were flash-frozen in liquid nitrogen until further
processing. Tissues were resuspended in 600 µL of RLT buffer
(RNeasy Plus Mini kit, Qiagen) supplemented with 2-mercaptoe-
thanol, then homogenized on Lysing Matrix D 2-mL tubes (MP
Benayoun et al.
10 Genome Research
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
Biomedicals) on a FastPrep-24 machine (MP Biomedicals) with a
speed setting of 6. Heart tissue was homogenized for 4 × 30 sec,
and all other tissues were homogenized for 40 sec. Subsequent
RNA extraction was performed using the RNeasy Plus Mini kit
(Qiagen) following the manufacturer’s instructions. Primary NSC
neurospheres (passages 2–3) were dissociated 16–18 h prior to col-
lection and seeded in 12-well plates. Cells were spun down and col-
lected in RLT buffer supplemented with 2-mercaptoethanol and
processed as above.
RNA-seq library preparation
For RNA-seq library preparation, 1μg of total RNA was combined
with 2 µL of a 1:100 dilution of ERCC RNA Spike-In Mix
(Thermo Fisher Scientific). The resulting mix was subjected to
rRNA depletion using the NEBNext rRNA Depletion kit (NEB) ac-
cording to the manufacturer’s protocol. Strand-specific libraries
were constructed using the SMARTer Stranded RNA-Seq kit
(Clontech) according to the manufacturer’s protocol. Paired-end
75-bp reads were generated on the Illumina NextSeq 500 platform.
RNA-seq analysis pipeline
Paired-end 75-bp reads were trimmed using Trim Galore! 0.3.1
(github.com/FelixKrueger/TrimGalore) to retain high-quality bas-
es with phred score >15 and a remaining length >35 bp. Read pairs
were mapped to the UCSC mm9 genome build using STAR 2.4.0j
(Dobin et al. 2013). Read counts were assigned to genes using sub-
read 1.4.5-p1 (Liao et al. 2014) and were imported into R to esti-
mate differential gene expression as a function of age using the
DESeq2 R package (DESeq2 1.6.3) (Love et al. 2014). Because no
overt variation of ERCC spike-in levels were observed from sample
to sample within a tissue/cell type, and because their use can in-
crease technical noise, ERCC reads were not used after initial qual-
ity-checking.
To map repetitive element expression, we used the TEtran-
scripts 1.5.1 software (Jin et al. 2015), with mm9 RepeatMasker
data (Smit 1996–2005) downloaded from the UCSC Table Browser.
Read counts were imported into R to estimate differential gene ex-
pression as a function of age using the DESeq2 R package (DESeq2
1.6.3) (Love et al. 2014). We also used the “analyzeRepeats.pl”
functionality of the HOMER software (Heinz et al. 2010). In that
case, read counts were imported into R (R Core Team 2018) to es-
timate differential gene expression as a function of age using the
DESeq2 R package (DESeq2 1.16.1) (Love et al. 2014). A more re-
cent version of DESeq2 was used for this because these analyses
were run at a later time than the rest of the study. Because there
are no major changes between these versions, the overall results
should not be significantly affected.
Machine-learning analysis
Machine-learning models were built for each tissue, but not in
NSCs since no gene was found to be significantly misregulated
by RNA-seq in these cells. We built classification models in each
tissue independently using four different classification algorithms
as implemented through R package ‘caret’(caret 6.0-80). Classifica-
tion algorithms for neural nets (NNET; “pcaNNet”) are directly im-
plemented in the caret 6.0-80 package. Auxiliary R packages were
used with caret to implement random forests (‘randomForest’
4.6-14), gradient boosting (‘gbm’2.1.3) and radial support vector
machines (kernlab 0.9-27). These package versions for machine
learning are used throughout our machine-learning analyses. Us-
ing 10-fold cross validation, caret optimized model parameters
on the training data. Accuracies, sensitivities, and specificities for
all classifiers in their cognate tissue were estimated using a test
set of randomly held-out 1/3 of the data (not used for training) ob-
tained using the “createDataPartition”function (Supplemental Ta-
ble S3). Feature importance estimation was only done for RFs and
GBMs, as other algorithms do not allow for it. The Gini score for
feature importance was computed by caret 6.0-80 for each feature
in the GBM and RF models, and the maximum in each model was
scaled to ‘100’for ease of visualization. For each gene in each tis-
sue, we extracted two types of ‘features’: (1) dynamic features,
which reflect changes to the chromatin landscape with age; and
(2) static features, which reflect the state of the chromatin and
transcriptional landscape in young animals. Models were built
with (1) all features, (2) static features only, and (3) dynamic fea-
tures only. Details of feature extraction are reported in the Supple-
mental Material.
Data access
ChIP-seq and RNA-seq data generated in this study havebeen sub-
mitted to the NCBI BioProject database (https://www.ncbi.nlm
.nih.gov/bioproject/) under accession number PRJNA281127. All
code for this study is available in Supplemental Code and at the
Benayoun Laboratory GitHub repository (https://github.com/
BenayounLaboratory/Mouse_Aging_Epigenomics_2018). The co-
ordinates of significantly changed chromatin peaks can also be
found in this repository.
Acknowledgments
We thank Aaron Newman and Ash Alizadeh for advice on the use
of CIBERSORT for RNA-seq de-convolution and Art Owen for ad-
vice on statistical analyses. We thank Ashley Webb for assistance
in tissue dissection for ChIP-seq and RNA-seq and Katja
Hebestreit for advice on ChIP-seq and RNA-seq analyses. We thank
Lauren Booth, Kévin Contrepois, Aaron Daugherty, C. David Lee,
Dena Leeman, John Tower, Marc Vermulst, and Robin Yeo for
feedback on analyses and manuscript. We thank Matthew
Buckley, Brittany Demmitt, Andrew McKay, and Robin Yeo for in-
dependent code-checking. Illumina HiSeq 2000 sequencing was
performed at the Stanford Genome Sequencing Service Center,
and Illumina NextSeq 500 sequencing was performed at the
Stanford Functional Genomics Facility, supported in part by
National Institutes of Health (NIH) P30 CA124435 through the
use of the Genetics Bioinformatics Service Center (Stanford
Cancer Institute Shared Resource). Support for this work was pro-
vided by NIH DP1 AG044848 (A.B.), NIH P01 AG036695 (A.B.
and A.K.), a generous gift from Tim and Michele Barakett (A.B.),
NIH R00 AG049934 (B.A.B.), the Hanson-Thorell family fellow-
ship (B.A.B.), and NIH F31 AG043232 (E.A.P.).
Author contributions: B.A.B., E.A.P., and A.B. designed the
study. B.A.B. and E.A.P. generated the ChIP-seq and RNA-seq
data sets for this study. B.A.B. processed the data and performed
the analyses. P.P.S. mapped and quantified the killifish RNA-seq
data sets and generated homology tables between killifish and
mouse genes. I.H. and B.W.D. helped with data set generation,
and I.H. processed tissues for histological analysis. P.P.S., S.M.,
and B.W.D. helped with independent code-checking. S.M. per-
formed Ingenuity Pathway Analysis. K.M.C. performed the histo-
pathology analysis. A.K. advised on data processing pipelines
and on machine learning. B.A.B. and A.B. wrote the paper. All au-
thors edited and commented on the manuscript.
References
Avrahami D, Li C, Zhang J, Schug J, Avrahami R, Rao S, Stadler MB, Burger L,
Schübeler D, Glaser B, et al. 2015. Aging-dependent demethylation of
Genomic landscape with aging in mice
Genome Research 11
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
regulatory elements correlates with chromatin state and improved βcell
function. Cell Metab 22: 619–632. doi:10.1016/j.cmet.2015.07.025
Baruch K, Deczkowska A, David E, Castellano JM, Miller O, Kertser A,
Berkutzki T, Barnett-Itzhaki Z, Bezalel D, Wyss-Coray T, et al. 2014.
Aging. Aging-induced type I interferon response at the choroid plexus
negatively affects brain function. Science 346: 89–93. doi:10.1126/sci
ence.1252945
Baumgart M, Groth M, Priebe S, Savino A, Testa G, Dix A, Ripa R, Spallotta F,
Gaetano C, Ori M, et al. 2014. RNA-seq of the aging brain in the short-
lived fish N. furzeri –conserved pathways and novel genes associated
with neurogenesis. Aging Cell 13: 965–974. doi:10.1111/acel.12257
Baumgart M, Priebe S, Groth M, Hartmann N, Menzel U, Pandolfini L, Koch
P, Felder M, Ristow M, Englert C, et al. 2016. Longitudinal RNA-seq anal-
ysis of vertebrate aging identifies mitochondrial complex I as a small-
molecule-sensitive modifier of lifespan. Cell Syst 2: 122–132. doi:10
.1016/j.cels.2016.01.014
Benayoun BA, Pollina EA, Ucar D, Mahmoudi S, Karra K, Wong ED,
Devarajan K, Daugherty AC, Kundaje AB, Mancini E, et al. 2014.
H3K4me3 breadth is linked to cell identity and transcriptional consis-
tency. Cell 158: 673–688. doi:10.1016/j.cell.2014.06.027
Benayoun BA, Pollina EA, Brunet A. 2015. Epigenetic regulation of ageing:
linking environmental inputs to genomic stability. Nat Rev Mol Cell
Biol 16: 593–610. doi:10.1038/nrm4048
Bernstein BE, Mikkelsen TS, Xie X, Kamal M, Huebert DJ, Cuff J, Fry B,
Meissner A, Wernig M, Plath K, et al. 2006. A bivalent chromatin struc-
ture marks key developmental genes in embryonic stem cells. Cell 125:
315–326. doi:10.1016/j.cell.2006.02.041
Bochkis IM, Przybylski D, Chen J, Regev A. 2014. Changes in nucleosome
occupancy associated with metabolic alterations in aged mammalian
liver. Cell Rep 9: 996–1006. doi:10.1016/j.celrep.2014.09.048
Boisvert MM, Erikson GA, Shokhirev MN, Allen NJ. 2018. The aging astro-
cyte transcriptome from multiple regions of the mouse brain. Cell Rep
22: 269–285. doi:10.1016/j.celrep.2017.12.039
Booth LN, Brunet A. 2016. The aging epigenome. Mol Cell 62: 728–744.
doi:10.1016/j.molcel.2016.05.013
Chen Y, Meltzer PS. 2005. Gene expression analysis via multidimensional
scaling. Curr Protoc Bioinformatics Chapter 7: Unit 7 11. doi:10.1002/
0471250953.bi0711s10
Chen K, Chen Z, Wu D, Zhang L, Lin X, Su J, Rodriguez B, Xi Y, Xia Z, Chen
X, et al. 2015. Broad H3K4me3 is associated with increased transcription
elongation and enhancer activity at tumor-suppressor genes. Nat Genet
47: 1149–1157. doi:10.1038/ng.3385
Cheung I, Shulha HP, Jiang Y, Matevossian A, Wang J, Weng Z, Akbarian S.
2010. Developmental regulation and individual differences of neuronal
H3K4me3 epigenomes in the prefrontal cortex. Proc Natl Acad Sci 107:
8824–8829. doi:10.1073/pnas.1001702107
Cheung P, Vallania F, Warsinske HC, Donato M, Schaffert S, Chang SE,
Dvorak M, Dekker CL, Davis MM, Utz PJ, et al. 2018. Single-cell chroma-
tin modification profiling reveals increased epigenetic variations with
aging. Cell 173: 1385–1397.e1314. doi:10.1016/j.cell.2018.03.079
Cole JJ, Robertson NA, Rather MI, Thomson JP, McBryan T, Sproul D, Wang
T, Brock C, Clark W, Ideker T, et al. 2017. Diverse interventions that ex-
tend mouse lifespan suppress shared age-associated epigenetic changes
at critical gene regulatory regions. Genome Biol 18: 58. doi:10.1186/
s13059-017-1185-3
De Cecco M, Criscione SW, Peterson AL, Neretti N, Sedivy JM, Kreiling JA.
2013. Transposable elements become active and mobile in the genomes
of aging mammalian somatic tissues. Aging (Albany NY) 5: 867–883.
doi:10.18632/aging.100621
Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P,
Chaisson M, Gingeras TR. 2013. STAR: ultrafast universal RNA-seq align-
er. Bioinformatics 29: 15–21. doi:10.1093/bioinformatics/bts635
Dou Z, Ghosh K, Vizioli MG, Zhu J, Sen P, Wangensteen KJ, Simithy J, Lan Y,
Lin Y, Zhou Z, et al. 2017. Cytoplasmic chromatin triggers inflamma-
tion in senescence and cancer. Nature 550: 402–406. doi:10.1038/
nature24050
The ENCODE Project Consortium. 2012. An integrated encyclopedia of
DNA elements in the human genome. Nature 489: 57–74. doi:10
.1038/nature11247
The GTEx Consortium. 2015. Human genomics. The Genotype-Tissue
Expression (GTEx) pilot analysis: multitissue gene regulation in hu-
mans. Science 348: 648–660. doi:10.1126/science.1262110
Guan S, Tan SM, Li Y, Torres J, Uzel G, Xiang L, Law SK. 2015.
Characterization of single amino acid substitutions in the β2 integrin
subunit of patients with leukocyte adhesion deficiency (LAD)-1. Blood
Cells Mol Dis 54: 177–182. doi:10.1016/j.bcmd.2014.11.005
Heintzman ND, Stuart RK, Hon G, Fu Y, Ching CW, Hawkins RD, Barrera
LO, Van Calcar S, Qu C, Ching KA, et al. 2007. Distinct and predictive
chromatin signatures of transcriptional promoters and enhancers in
the human genome. Nat Genet 39: 311–318. doi:10.1038/ng1966
Heinz S, Benner C, Spann N, Bertolino E, Lin YC, Laslo P, Cheng JX, Murre
C, Singh H, Glass CK. 2010. Simple combinations of lineage-determin-
ing transcription factors prime cis-regulatory elements required for mac-
rophage and B cell identities. Mol Cell 38: 576–589. doi:10.1016/j
.molcel.2010.05.004
Hnisz D, Abraham BJ, Lee TI, Lau A, Saint-André V, Sigova AA, Hoke HA,
Young RA. 2013. Super-enhancers in the control of cell identity and dis-
ease. Cell 155: 934–947. doi:10.1016/j.cell.2013.09.053
Hofmann JW, Zhao X, De Cecco M, Peterson AL, Pagliaroli L, Manivannan J,
Hubbard GB, Ikeno Y, Zhang Y, Feng B, et al. 2015. Reduced expression
of MYC increases longevity and enhances healthspan. Cell 160: 477–
488. doi:10.1016/j.cell.2014.12.016
Horvath S. 2013. DNA methylation age of human tissues and cell types.
Genome Biol 14: R115. doi:10.1186/gb-2013-14-10-r115
Horvath S, Erhart W, Brosch M, Ammerpohl O, von Schonfels W, Ahrens M,
Heits N, Bell JT, Tsai PC, Spector TD, et al. 2014. Obesity accelerates epi-
genetic aging of human liver. Proc Natl Acad Sci 111: 15538–15543.
doi:10.1073/pnas.1412759111
Jenuwein T, Allis CD. 2001. Translating the histone code. Science 293: 1074–
1080. doi:10.1126/science.1063127
Jin Y, Tam OH, Paniagua E, Hammell M. 2015. TEtranscripts: a package for
including transposable elements in differential expression analysis of
RNA-seq data sets. Bioinformatics 31: 3593–3599. doi:10.1093/bioinfor
matics/btv422
Kassiotis G, Stoye JP. 2016. Immune responses to endogenous retroele-
ments: taking the bad with the good. Nat Rev Immunol 16: 207–219.
doi:10.1038/nri.2016.27
Kennedy BK, Berger SL, Brunet A, Campisi J, Cuervo AM, Epel ES, Franceschi
C, Lithgow GJ, Morimoto RI, Pessin JE, et al. 2014. Geroscience: linking
aging to chronic disease. Cell 159: 709–713. doi:10.1016/j.cell.2014.10
.039
Kouskouti A, Talianidis I. 2005. Histone modifications defining active genes
persist after transcriptional and mitotic inactivation. EMBO J 24: 347–
357. doi:10.1038/sj.emboj.7600516
Liao Y, Smyth GK, Shi W. 2014. featureCounts: an efficient general purpose
program for assigning sequence reads to genomic features.
Bioinformatics 30: 923–930. doi:10.1093/bioinformatics/btt656
Liu L, Cheung TH, Charville GW, Hurgo BMC, Leavitt T, Shih J, Brunet A,
Rando TA. 2013. Chromatin modifications as determinants of muscle
stem cell quiescence and chronological aging. Cell Rep 4: 189–204.
doi:10.1016/j.celrep.2013.05.043
López-Otín C, Blasco MA, Partridge L, Serrano M, Kroemer G. 2013. The
hallmarks of aging. Cell 153: 1194–1217. doi:10.1016/j.cell.2013.05
.039
Love MI, Huber W, Anders S. 2014. Moderated estimation of fold change
and dispersion for RNA-seq data with DESeq2. Genome Biol 15: 550.
doi:10.1186/s13059-014-0550-8
Lumeng CN, Liu J, Geletka L, Delaney C, Delp roposto J, Desai A, Oatmen K,
Martinez-Santibanez G, Julius A, Garg S, et al. 2011. Aging is associated
with an increase in T cells and inflammatory macrophages in visceral
adipose tissue. J Immunol 187: 6208–6216. doi:10.4049/jimmunol
.1102188
Martins R, Lithgow GJ, Link W. 2016. Long live FOXO: unraveling the role
of FOXO proteins in aging and longevity. Agin g Cell 15: 196–207. doi:10
.1111/acel.12427
Maxwell PH, Burhans WC, Curcio MJ. 2011. Retrotransposition is associat-
ed with genome instability during chronological aging. Proc Natl Acad
Sci 108: 20376–20381. doi:10.1073/pnas.1100271108
Moskowitz DM, Zhang DW, Hu B, Le Saux S, Yanes RE, Ye Z, Buenrostro JD,
Weyand CM, Greenleaf WJ, Goronzy JJ. 2017. Epigenomics of human
CD8 T cell differentiation and aging. Sci Immunol 2: eaag0192. doi:10
.1126/sciimmunol.aag0192
Nativio R, Donahue G, Berson A, Lan Y, Amlie-Wolf A, Tuzer F, Toledo JB,
Gosai SJ, Gregory BD, Torres C, et al. 2018. Dysregulation of the epige-
netic landscape of normal aging in Alzheimer’s disease. Nat Neurosci
21: 497–505. doi:10.1038/s41593-018-0101-9
Newman AM, Liu CL, Green MR, Gentles AJ, Feng W, Xu Y, Hoang CD,
Diehn M, Alizadeh AA. 2015. Robust enumeration of cell subsets from
tissue expression profiles. Nat Methods 12: 453–457. doi:10.1038/
nmeth.3337
O’Brown ZK, Van Nostrand EL, Higgins JP, Kim SK. 2015. The inflammatory
transcription factors NFκB, STAT1 and STAT3 drive age-associated tran-
scriptional changes in the human kidney. PLoS Genet 11: e1005734.
doi:10.1371/journal.pgen.1005734
Oshlack A, Wakefield MJ. 2009. Transcript length bias in RNA-seq data con-
founds systems biology. Biol Direct 4: 14. doi:10.1186/1745-6150-4-14
Ou SH, Wu F, Harrich D, García-Martínez LF, Gaynor RB. 1995. Cloning and
characterization of a novel cellular protein, TDP-43, that binds to hu-
man immunodeficiency virus type 1 TAR DNA sequence motifs. J
Virol 69: 3584–3596.
Benayoun et al.
12 Genome Research
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
Pal S, Tyler JK. 2016. Epigenetics and aging. Sci Adv 2: e1600584. doi:10
.1126/sciadv.1600584
Parker SC, Stitzel ML, Taylor DL, Orozco JM, Erdos MR, Akiyama JA, van
Bueren KL, Chines PS, Narisu N, Program NCS, et al. 2013. Chromatin
stretch enhancer states drive cell-specific gene regulation and harbor
human disease risk variants. Proc Natl Acad Sci 110: 17921–17926.
doi:10.1073/pnas.1317023110
Pinto AR, Godwin JW, Chandran A, Hersey L, Ilinykh A, Debuque R, Wang
L, Rosenthal NA. 2014. Age-related changes in tissue macrophages pre-
cede cardiac functional impairment. Aging (Albany NY) 6: 399–413.
doi:10.18632/aging.100669
Quach A, Levine ME, Tanaka T, Lu AT, Chen BH, Ferrucci L, Ritz B,
Bandinelli S, Neuhouser ML, Beasley JM, et al. 2017. Epigenetic clock
analysis of diet, exercise, education, and lifestyle factors. Aging (Albany
NY) 9: 419–446. doi:10.18632/aging.101168
Quinlan AR, Hall IM. 2010. BEDTools: a flexible suite of utilities for compar-
ing genomic features. Bioinformatics 26: 841–842. doi:10.1093/bioinfor
matics/btq033
R Core Team. 2018. R: a language and environment for statistical computing.R
Foundation for Statistical Computing, Vienna. https://www.R-project
.org/.
Ringnér M. 2008. What is principal component analysis? Nat Biotechnol 26:
303–304. doi:10.1038/nbt0308-303
Rodwell GE, Sonu R, Zahn JM, Lund J, Wilhelmy J, Wang L, Xiao W,
Mindrinos M, Crane E, Segal E, et al. 2004. A transcriptional profile of
aging in the human kidney. PLoS Biol 2: e427. doi:10.1371/journal
.pbio.0020427
Ross-Innes CS, Stark R, Teschendorff AE, Holmes KA, Ali HR, Dunning MJ,
Brown GD, Gojis O, Ellis IO, Green AR, et al. 2012. Differential oestro-
gen receptor binding is associated with clinical outcome in breast can-
cer. Nature 481: 389–393. doi:10.1038/nature10730
Scotter EL, Chen HJ, Shaw CE. 2015. TDP-43 proteinopathy and ALS: in-
sights into disease mechanisms and therapeutic targets.
Neurotherapeutics 12: 352–363. doi:10.1007/s13311-015-0338-x
Seiler F, Hellberg J, Lepper PM, Kamyschnikow A, Herr C, Bischoff M, Langer
F, Schafers HJ, Lammert F, Menger MD, et al. 2013. FOXO transcription
factors regulate innate immune mechanisms in respiratory epithelial
cells. J Immunol 190: 1603–1613. doi:10.4049/jimmunol.1200596
Sen P, Shah PP, Nativio R, Berger SL. 2016. Epigenetic mechanisms of lon-
gevity and aging. Cell 166: 822–839. doi:10.1016/j.cell.2016.07.050
Shen Y, Yue F, McCleary DF, Ye Z, Edsall L, Kuan S, Wagner U, Dixon J, Lee L,
Lobanenkov VV, et al. 2012. A map of the cis-regulatory sequences in
the mouse genome. Nature 488: 116–120. doi:10.1038/nature11243
Shulha HP, Cheung I, Guo Y, Akbarian S, Weng Z. 2013. Coordinated cell
type–specific epigenetic remodeling in prefrontal cortex begins before
birth and continues into early adulthood. PLoS Genet 9: e1003433.
doi:10.1371/journal.pgen.1003433
Smit A. 1996–2005. RepeatMasker. http://www.repeatmasker.org.
Stegeman R, Weake VM. 2017. Transcriptional signatures of aging. J Mol Biol
429: 2427–2437. doi:10.1016/j.jmb.2017.06.019
Stubbs TM, Bonder MJ, Stark AK, Krueger F, Team BIAC, von Meyenn F,
Stegle O, Reik W. 2017. Multi-tissue DNA methylation age predictor
in mouse. Genome Biol 18: 68. doi:10.1186/s13059-017-1203-5
Sun L, Wu J, Du F, Chen X, Chen ZJ. 2013. Cyclic GMP-AMP synthase is a
cytosolic DNA sensor that activates the type I interferon pathway.
Science 339: 786–791. doi:10.1126/science.1232458
Sun D, Luo M, Jeong M, Rodriguez B, Xia Z, Hannah R, Wang H, Le T, Faull
KF, Chen R. 2014. Epigenomic profiling of young and aged HSCs reveals
concerted changes during aging that reinforce self-renewal. Cell Stem
Cell 14: 673–688. doi:10.1016/j.stem.2014.03.002
Teschendorff AE, Zheng SC. 2017. Cell-type deconvolution in epigenome-
wide association studies: a review and recommendations. Epigenomics
9: 757–768. doi:10.2217/epi-2016-0153
Tsurumi A, Li WX. 2012. Global heterochromatin loss: a unifying theory of
aging? Epigenetics 7: 680–688. doi:10.4161/epi.20540
Ucar D, Márquez EJ, Chung CH, Marches R, Rossi RJ, Uyar A, Wu TC, George
J, Stitzel ML, Palucka AK, et al. 2017. The chromatin accessibility signa-
ture of human immune aging stems from CD8
+
T cells. J Exp Med 214:
3123–3144. doi:10.1084/jem.20170416
Van Meter M, Kashyap M, Rezazadeh S, Geneva AJ, Morello TD, Seluanov A,
Gorbunova V. 2014. SIRT6 represses LINE1 retrotransposons by ribosy-
lating KAP1 but this repression fails with stress and age. Nat Commun 5:
5011. doi:10.1038/ncomms6011
Wagner W. 2017. Epigenetic aging clocks in mice and men. Genome Biol 18:
107. doi:10.1186/s13059-017-1245-8
Wagner A, Regev A, Yosef N. 2016. Revealing the vectors of cellular identity
with single-cell genomics. Nat Biotechnol 34: 1145–1160. doi:10.1038/
nbt.3711
Wang T, Tsui B, Kreisberg JF, Robertson NA, Gross AM, Yu MK, Carter H,
Brown-Borg HM, Adams PD, Ideker T. 2017. Epigenetic aging signatures
in mice livers are slowed by dwarfism, calorie restriction and rapamycin
treatment. Genome Biol 18: 57. doi:10.1186/s13059-017-1186-2
Webb AE, Pollina EA, Vierbuchen T, Urbán N, Ucar D, Leeman DS,
Martynoga B, Sewak M, Rando TA, Guillemot F, et al. 2013. FOXO3
shares common targets with ASCL1 genome-wide and inhibits ASCL1-
dependent neurogenesis. Cell Rep 4: 477–491. doi:10.1016/j.celrep
.2013.06.035
West AP, Khoury-Hanold W, Staron M, Tal MC, Pineda CM, Lang SM,
Bestwick M, Duguay BA, Raimundo N, MacDuff DA, et al. 2015.
Mitochondrial DNA stress primes the antiviral innate immune re-
sponse. Nature 520: 553–557. doi:10.1038/nature14156
White RR, Milholland B, MacRae SL, Lin M, Zheng D, Vijg J. 2015.
Comprehensive transcriptional landscape of aging mouse liver. BMC
Genomics 16: 899. doi:10.1186/s12864-015-2061-8
Wood JG, Helfand SL. 2013. Chromatin structure and transposable ele-
ments in organismal aging. Front Genet 4: 274. doi:10.3389/fgene
.2013.00274
Xia S, Zhang X, Zheng S, Khanabdali R, Kalionis B, Wu J, Wan W, Tai X.
2016. An update on inflamm-aging: mechanisms, prevention, and treat-
ment. J Immunol Res 2016: 8426874. doi:10.1155/2016/8426874
Yu Y, Fuscoe JC, Zhao C, Guo C, Jia M, Qing T, Bannon DI, Lancashire L, Bao
W, Du T, et al. 2014. A rat RNA-Seq transcriptomic BodyMap across 11
organs and 4 developmental stages. Nat Commun 5: 3230. doi:10
.1038/ncomms4230
Yue F, Cheng Y, Breschi A, Vierstra J, Wu W, Ryba T, Sandstrom R, Ma Z,
Davis C, Pope BD, et al. 2014. A comparative encyclopedia of DNA ele-
ments in the mouse genome. Nature 515: 355–364. doi:10.1038/
nature13992
Zahn JM, Sonu R, Vogel H, Crane E, Mazan-Mamczarz K, Rabkin R, Davis
RW, Becker KG, Owen AB, Kim SK. 2006. Transcriptional profiling of ag-
ing in human muscle reveals a common aging signature. PLoS Genet 2:
e115. doi:10.1371/journal.pgen.0020115.eor
Zahn JM, Poosala S, Owen AB, Ingram DK, Lustig A, Carter A, Weeraratna
AT, Taub DD, Gorospe M, Mazan-Mamczarz K, et al. 2007. AGEMAP: a
gene expression database for aging in mice. PLoS Genet 3: e201. doi:10
.1371/journal.pgen.0030201
Zhang Y, Liu T, Meyer CA,Eeckhou te J, Johnson DS, Bernstein BE,Nusbaum
C, Myers RM, Brown M, Li W, et al. 2008. Model-based Analysis of ChIP-
Seq (MACS). Genome Biol 9: R137. doi:10.1186/gb-2008-9-9-r137
Zhao W, Beers DR, Bell S, Wang J, Wen S, Baloh RH, Appel SH. 2015. TDP-43
activates microglia through NF-κB and NLRP3 inflammasome. Exp
Neurol 273: 24–35. doi:10.1016/j.expneurol.2015.07.019
Zheng X, Yue S, Chen H, Weber B, Jia J, Zheng Y. 2015. Low-cell-number
epigenome profiling aids the study of lens aging and hematopoiesis.
Cell Rep 13: 1505–1518. doi:10.1016/j.celrep.2015.10.004
Received May 31, 2018; accepted in revised form January 25, 2019.
Genomic landscape with aging in mice
Genome Research 13
www.genome.org
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from
10.1101/gr.240093.118Access the most recent version at doi:
published online March 11, 2019Genome Res.
Bérénice A. Benayoun, Elizabeth A. Pollina, Param Priya Singh, et al.
responses
aging in mice reveals widespread induction of inflammatory
Remodeling of epigenome and transcriptome landscapes with
Material
Supplemental http://genome.cshlp.org/content/suppl/2019/03/11/gr.240093.118.DC1
P<P
Published online March 11, 2019 in advance of the print journal.
License
Commons
Creative
.http://creativecommons.org/licenses/by-nc/4.0/described at
a Creative Commons License (Attribution-NonCommercial 4.0 International), as
). After six months, it is available underhttp://genome.cshlp.org/site/misc/terms.xhtml
first six months after the full-issue publication date (see
This article is distributed exclusively by Cold Spring Harbor Laboratory Press for the
Service
Email Alerting
click here.top right corner of the article or
Receive free email alerts when new articles cite this article - sign up in the box at the
http://genome.cshlp.org/subscriptions
go to: Genome Research To subscribe to
© 2019 Benayoun et al.; Published by Cold Spring Harbor Laboratory Press
Cold Spring Harbor Laboratory Press on March 13, 2019 - Published by genome.cshlp.orgDownloaded from