Distinct Longevity Mechanisms Across and Within

Article
Distinct longevity mechanisms across and within

species and their association with aging
Graphical abstract Authors
Alexander Tyshkovskiy, Siming Ma,
Anastasia V. Shindyapina, ...,
Sergey E. Dmitriev, Richard A. Miller,
Vadim N. Gladyshev
Correspondence
[email protected]
In brief
The analysis of multi-tissue gene
expression signatures associated with
longevity across mammalian species and
their interaction with biomarkers of aging
and signatures of lifespan-extending
interventions reveals universal and
distinct strategies of lifespan regulation
within and across species and provides
tools for the discovery of longevity
interventions in mammals.
Highlights
d Distinct molecular mechanisms control lifespan within and
across species
d Aging effects are reversed by longevity interventions but not

by species longevity
d Regulation of Igf1 and mitochondrial translation are shared

signatures of longevity
d Longevity signatures enable the discovery of

geroprotectors, such as KU0063794
Tyshkovskiy et al., 2023, Cell 186, 1–21

June 22, 2023 ª 2023 Elsevier Inc.
https://doi.org/10.1016/j.cell.2023.05.002 ll
Please cite this article in press as: Tyshkovskiy et al., Distinct longevity mechanisms across and within species and their association with ag-
ing, Cell (2023), https://doi.org/10.1016/j.cell.2023.05.002
ll
Article
Distinct longevity mechanisms across and within
species and their association with aging
Alexander Tyshkovskiy,1,2,10 Siming Ma,1,10 Anastasia V. Shindyapina,1,10 Stanislav Tikhonov,2 Sang-Goo Lee,1
Perinur Bozaykut,1,3 José P. Castro,1,4,5 Andrei Seluanov,6 Nicholas J. Schork,7 Vera Gorbunova,6 Sergey E. Dmitriev,2
Richard A. Miller,8 and Vadim N. Gladyshev1,9,11,*
1Division of Genetics, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA 02115, USA
2Belozersky Institute of Physico-Chemical Biology, Moscow State University, Moscow 119234, Russia
3Department of Molecular Biology and Genetics, Faculty of Engineering and Natural Sciences, Acibadem Mehmet Ali Aydinlar University,
Istanbul 34752, Turkey

4i3S, Instituto de Investigação e Inovação em Saúde, Universidade do Porto, 4200-135 Porto, Portugal
5Aging and Aneuploidy Laboratory, IBMC, Instituto de Biologia Molecular e Celular, Universidade do Porto, 4200-135 Porto, Portugal
6Departments of Biology and Medicine, University of Rochester, Rochester, NY, USA
7Quantitative Medicine and Systems Biology Division, Translational Genomics Research Institute, Phoenix, AZ, USA
8Department of Pathology and Geriatrics Center, University of Michigan, Ann Arbor, MI 48109, USA
9Broad Institute, Cambridge, MA, USA
10These authors contributed equally
11Lead contact
*Correspondence: [email protected]
https://doi.org/10.1016/j.cell.2023.05.002
SUMMARY
Lifespan varies within and across species, but the general principles of its control remain unclear. Here, we
conducted multi-tissue RNA-seq analyses across 41 mammalian species, identifying longevity signatures
and examining their relationship with transcriptomic biomarkers of aging and established lifespan-extending
interventions. An integrative analysis uncovered shared longevity mechanisms within and across species,
including downregulated Igf1 and upregulated mitochondrial translation genes, and unique features, such
as distinct regulation of the innate immune response and cellular respiration. Signatures of long-lived species
were positively correlated with age-related changes and enriched for evolutionarily ancient essential genes,
involved in proteolysis and PI3K-Akt signaling. Conversely, lifespan-extending interventions counteracted
aging patterns and affected younger, mutable genes enriched for energy metabolism. The identified bio-
markers revealed longevity interventions, including KU0063794, which extended mouse lifespan and health-
span. Overall, this study uncovers universal and distinct strategies of lifespan regulation within and across
species and provides tools for discovering longevity interventions.
INTRODUCTION ventions increasing murine lifespan and healthspan are known

today. They include mutations such as growth hormone receptor
Mammals exhibit substantial variability of lifespan both within and knockout (GHRKO),11 drugs such as rapamycin,12 and diets
across species. Generally, there is a strong positive association such as calorie restriction (CR).13–16 In contrast to the inter-spe-
between adult weight (AW) and species longevity,1 exemplified cies data, body size within species is often negatively associated
by Etruscan shrews (Suncus etruscus), weighing 1.8 g with life- with longevity. Life expectancy across dog breeds is inversely
span of 3.2 years, and bowhead whales (Balaena mysticetus), correlated with body mass,17 and various murine dwarf models
weighing >100 tons with a maximum lifespan (ML) of >200 years.2 are often characterized by 30%–60% longer median and ML
However, there are also exceptions to this rule, the so-called compared with wild-type mice,18,19 which appears to be medi-
exceptionally long-lived species, that live significantly longer ated through the decreased activity of growth hormone (GH)
than expected based on their AW, such as the naked mole rat and insulin-like growth factor 1 (IGF-1) signaling.20 Therefore,
(Heterocephalus glaber),3 some microbats, e.g., the Brandt’s some features appear to be differently associated with lifespan
bat (Myotis brandtii),4 and humans. This variability is achieved across and within species. Establishing shared and distinct mo-
on the evolutionary timescale, and multiple mechanisms that lecular mechanisms of longevity, as well as their causative rela-
may contribute to species longevity have been identified.5–10 tionship with aging, is critically important for our understanding
Mammalian lifespan may also be extended within species. of drivers of lifespan regulation and the development of effective
Dozens of genetic, pharmacological, and environmental inter- geroprotectors.
Cell 186, 1–21, June 22, 2023 ª 2023 Elsevier Inc. 1

ll
Article
High-throughput data, including transcriptomics, metabolo- in gene expression, capturing, on average, 84% of the total vari-
mics, and epigenomics, were used to evaluate molecular fea- ance across samples. Within tissues, 78%–95% of the variance
tures of longevity and aging. Aging-related changes in DNA was explained by species, even after controlling for batch effect
methylation21,22 and gene expression23,24 were thoroughly (Figure S1D). Correlation analysis among species revealed a high
examined in individual mammalian species. Several studies similarity of expression profiles in neural tissues, whereas testis
also conducted qualitative meta-analyses of transcriptomic ag- profiles were diverse across species (Figures 1C and S1E),
ing signatures for different tissues and species,25,26 although consistent with the notion that this organ rapidly evolves under
quantitative meta-analyses of age-related gene expression the impact of reproduction-related selection.37,42–44
changes (ECs) have been lacking. Associations between species To characterize organ-specific expression patterns, we iden-
longevity and molecular patterns of mammalian organs or cells tified 6,050 genes that significantly upregulated or downregu-
were also assessed at the level of the metabolome,27 ionome,28 lated (p value, p < 0.01) in one organ relative to the others (Fig-
lipidome,29 and transcriptome.30–32 Finally, by carrying out a ure 1D). In the kidney, genes involved in ion transport and
large-scale meta-analysis, we previously characterized gene sodium transport were overexpressed, consistent with their
expression signatures of lifespan extension induced by various functions in ultrafiltration and selective reabsorption. In the
longevity interventions in mouse tissues.33 liver, a higher expression was detected for genes involved in
Despite the growing number of established geroprotectors steroid metabolism, detoxification, and complement and coag-
and high-throughput data, a comprehensive analysis of universal ulation cascades. Consistent with its energy demand, the heart
and model-specific signatures of longevity has been lacking. Do exhibited an overexpression of genes related to oxidative phos-
existing lifespan-extending interventions in mice induce molecu- phorylation and the tricarboxylic acid (TCA) cycle. The brain
lar mechanisms shared by long-lived mammals, such as the cortex and cerebellum specifically expressed genes involved
naked mole rat, human, and bowhead whale? How do these sig- in synaptic transmission and neuronal differentiation, whereas
natures interplay with age-associated features? What are the testis was characterized by overexpression of genes en-
the common and distinct mechanisms of longevity induced by coding cyclins, centrosomal proteins, and spermatogenesis-
interventions and selected in species during evolution? Can associated proteins, consistent with their roles in sexual repro-
these signatures be utilized to identify lifespan-extending duction. Overall, organ-specific expression patterns were
interventions? congruent with their biological roles, and samples from the
To answer these questions, we (1) performed RNA sequencing same tissues clustered together. The exceptions were chicken
(RNA-seq) and characterized gene expression signatures of and platypus samples, which clustered by species and away
mammalian longevity across 41 species, including the long-lived from the rest of the samples (Figure 1D, bottom), probably
naked mole rat, Brandt’s bat, and bowhead whale; (2) identified due to their significant evolutionary distance from other exam-
species-specific, tissue-specific, and universal transcriptomic ined species.
biomarkers of mammalian aging through a quantitative meta-
analysis of 92 publicly available datasets; (3) investigated com- Transcriptomic signatures of longevity across species
mon and distinct molecular mechanisms of aging and longevity We employed brain, kidney, and liver samples, representing >35
within and across species; and (4) demonstrated that the identi- species, to identify gene expression signatures of long-lived
fied signatures can be used to discover and characterize life- mammals. Using phylogenetic regression (see STAR Methods),
span-regulating interventions in mammals. we identified 100–500 genes that were significantly associated
(p adjusted < 0.05) with animal ML (Figures 2A–2C) and female
RESULTS time to maturity (FTM) (Figure S2A). Expression of 40%–70%
of these genes was significantly associated with species ML
RNA-seq of tissues across mammalian species and FTM after adjustment for body mass (MLres and FTMres,
To investigate gene expression signatures of mammalian life- respectively) (Figures S2B and S2C).
span, we collected RNA-seq data for 48 young adult animals To assess if organ gene expression predicts species ML, we
and aggregated it with publicly available data.4,31,34–41 This utilized the Elastic Net linear regression model and leave-one-
resulted in 371 biological samples covering six tissues (brain, out (LOO) technique, iteratively training the model on all but
kidney, liver, cerebellum, heart, and testis) of 41 mammalian spe- one species and testing it on the remaining species. When
cies from 12 taxonomic orders with a wide range of longevity- applied to the test set, the model captured 78% of the total vari-
associated traits (Figure 1A; Table S1A), including 4 exception- ation in lifespan in the log scale (Figure 2D). The addition of
ally long-lived mammals: naked mole rat, Brandt’s bat, bowhead mammalian AW did not further improve the model, and body
whale, and human. Utilizing a previously developed pipeline30 mass alone captured only 39% of lifespan variation (Figure S2D),
(see STAR Methods), we identified 13,784 one-to-one orthologs suggesting that tissue gene expression significantly outperforms
across examined species (Figures S1A and S1B; Table S2). AW in predicting mammalian longevity. Several genes, including
Principal-component analysis revealed that the samples were Sdc1 and Dtl, were consistently selected by Elastic Net models
segregated predominantly by their organ origin (Figure 1B). The trained on different LOO subsets (Figure 2E). Sdc1 encodes pro-
cerebellum and cerebral cortex samples were clustered teoglycan syndecan-1, whose deficiency induces inflammation
together, whereas the testis samples were distant from the other in mice,45,46 whereas an increased expression of Dtl is associ-
organs (Figure S1C). Analysis of variance (ANOVA) confirmed ated with bladder cancer progression, presumably through the
that organs and species were the primary sources of variation regulation of the AKT/mTOR pathway.47 In agreement, Sdc1
2 Cell 186, 1–21, June 22, 2023

ll
Article
Figure 1. RNA-seq of mammalian tissues

(A) Phylogenetic tree of examined species and their longevity traits. Adult weight (AW), maximum lifespan (ML), and ML residual adjusted for body mass (MLres)
are shown on barplots in logarithmic scale.
(B) Principal-component analysis of mammalian samples. First 3 principal components (PCs) are shown. The percentage of total variance explained by each PC is
indicated in parentheses.
(C) Within-organ variability across mammalian species. Pairwise Spearman correlation coefficients between species expression profiles were calculated for each
organ. The number of available species is indicated in parenthesis.
(D) Expression of genes significantly enriched or depleted in specific organs. Columns represent genes; rows represent biological samples colored by organ, as in
(B). Top pathways enriched for corresponding gene sets are indicated in the text. Ch, chicken; Pl, platypus.
See also Figure S1 and Tables S1 and S2.
and Dtl were upregulated and downregulated in the tissues of ure S2F). In contrast, the total proportion of immune cells was,
long-lived species, respectively (Figure 2E). on average, <3% in every organ. Following adjustment for blood
Comparison of signatures across tissues revealed significant cell abundance across samples, we still detected statistically
pairwise overlaps for both up- and downregulated genes associ- significant associations with ML for >95.9% of signature genes
ated with ML (p adjusted < 2 3 105) (Figure 2F) and MLres (p in every examined tissue, suggesting that modest variation in
adjusted < 0.042) (Figure 2G). Accordingly, longevity-associated blood cell composition is not a defining factor for the observed
ECs were positively correlated across all analyzed organs effects.
(Spearman r > 0.21) (Figure 2H). To examine if common signa- Interestingly, longevity-associated ECs in all organs were also
tures of longevity were driven by universally expressed genes positively correlated with signatures of lifespan in cultured pri-
or blood cell-specific biomarkers, we utilized Tabula Muris sin- mary skin fibroblasts (Figure S2G) collected from 16 mammalian
gle-cell atlas.24 The identified signatures of ML in every organ species.30 Although a similarity between longevity biomarkers in
were, on average, expressed in >49% of cell types (Figure S2E). organs may be partially driven by the systemic effect of signaling
Shared biomarkers of longevity across tissues, aggregated with and catalytic molecules circulating in the bloodstream, including
the harmonic mean p value (HMP) method,48 demonstrated even cytokines,50–52 extracellular miRNAs,53 metabolites,54 and en-
more universal expression, being detected in 66% of the individ- zymes,55 these results suggest that numerous conserved molec-
ual cell types, whereas blood cell-specific biomarkers ac- ular mechanisms of lifespan regulation are preserved at the level
counted for <1.5% of ML-associated genes. of cultured cells and are likely to be encoded in the genome.
Cell type deconvolution49 revealed that liver samples were To identify pathways associated with species longevity, we
largely composed of hepatocytes, kidney samples consisted of performed functional gene set enrichment analysis (GSEA) of
epithelial and endothelial cells, and brain samples included neu- the lifespan-associated ECs (Figures 2I and S2H; Table S3A).
rons, oligodendrocytes, astrocytes, and other glial cells (Fig- Shared molecular features of long-lived mammals across
Cell 186, 1–21, June 22, 2023 3

ll
Article
Figure 2. Gene expression signatures of species longevity

(A–C)Representative genes associated with species maximum lifespan. Selected genes include Rpl30 (A), Rpl28 (B), and Cul4b (C). The association between
log10(maximum lifespan) and average normalized log10(expression) is shown for brain (left), liver (middle), and kidney (right). Linear model equation, slope
adjusted p value and R2 estimated with phylogenetic regression are displayed. Data are mean ± SE.
(D) Accuracy of Elastic Net prediction of species maximum lifespan (ML) based on tissue gene expression. Mean absolute error (MAE), R2, and Pearson’s
correlation coefficient were estimated on LOO test set. Each dot represents a single species and is colored by taxonomic group, as in (A)–(C).
(E) Gene expression predictors of maximum lifespan. For each gene, data represent mean weight in Elastic Net models trained on different subsets ± SE. Only
genes with non-zero average weight are included (p adjusted < 0.05).
(F and G) Overlap of genes associated with maximum lifespan unadjusted (F) and adjusted for body mass (MLres) (G) across mammalian tissues. p value was
calculated with Fisher exact tests.
(H) Spearman correlation of longevity-associated gene expression changes. Asterisks reflect statistical significance.
(I) Functional enrichment of lifespan-associated genes. Dotted lines reflect threshold of p adjusted = 0.1. The whole list of enriched functions is in Table S3A.
*p adjusted < 0.05; **p adjusted < 0.01; ***p adjusted < 0.001.
FTM, female time to maturity; FTMres, female time to maturity residual.
See also Figure S2 and Table S3.
multiple organs included the upregulation of genes involved in Gene expression biomarkers of mammalian aging
translation (e.g., Rpl28 encoding a large ribosomal subunit To identify consistent transcriptomic signatures of aging, we
component [Figures 2A and S2B]) and base excision repair aggregated data on the mouse (Mus musculus), rat (Rattus nor-
(e.g., Mpg encoding N-methylpurine DNA glycosylase [Fig- vegicus), and human (Homo sapiens) age-related gene ECs from
ure 2B]), and downregulation of genes involved in ubiquitin- 92 publicly available datasets (Table S1B) (see STAR Methods).
mediated proteolysis (e.g., Cul4b encoding a cullin protein family Aging-associated ECs were positively correlated across data-
member [Figures 2C and S2C]), TCA cycle, and insulin process- sets, especially after restricting the analyses to the top statisti-
ing (Figures 2I and S2H). cally significant genes for each pair of datasets (Figure S3A).
4 Cell 186, 1–21, June 22, 2023

ll
Article
Figure 3. Transcriptomic signatures of mammalian aging

(A) Spearman correlation of age-related expression changes (ECs) across datasets estimated using a global aging signature. Red frame highlights aggregated
(global) age-related ECs.
(B and C) Spearman correlation of aging-associated ECs between tissues (B) and species (C). Text and asterisks represent correlation coefficient and adjusted p
value, respectively.
(D and E) Overlap of genes with significant age-related ECs across tissues (D) and species (E). p value was calculated with Fisher exact tests.
(F) Vsig4 (upper) and Nrep (lower) age-related ECs. Each dot represents normalized average gene EC calculated from a single dataset ± SE. Red dotted line and
shaded area represent weighted mean EC estimated using mixed-effect model and 95% confidence interval, respectively.
(G) Contribution of tissue, species, and source to pairwise Spearman correlation of age-related ECs. Bar color and asterisks reflect the statistical significance of
factor contribution, assessed with multiple linear regression.
(H) Normalized age-associated ECs across tissues and species. Only genes significantly associated with aging in at least one signature (p adjusted < 0.05)
are shown.
(I) Functional enrichment (GSEA) of aging signatures. Only functions significantly enriched by at least one signature are shown (p adjusted < 0.1). The whole list of
enriched functions is in Table S3B.
^p adjusted < 0.1; *p adjusted < 0.05; **p adjusted < 0.01; ***p adjusted < 0.001.
WBC, white blood cells; SCAT, subcutaneous adipose tissue; BAT, brown adipose tissue; MAT, marrow adipose tissue; muscle, skeletal muscle.
See also Figures S3 and S4 and Tables S1 and S3.
We observed a higher similarity for age-related ECs from the identified 200–900 statistically significant gene expression sig-
same tissue or species (Figures 3G and S3B), although a positive natures of aging (p adjusted < 0.05) for individual tissues (liver,
correlation was detected even for ECs from different tissues, brain, and skeletal muscle) and species (human, mouse, and
species, and sources (p < 1025), suggesting that there are rat). We also discovered 13 universal aging-associated genes
also universal molecular mechanisms of mammalian aging. shared across 3 species and 17 tissues. As expected, age-
Following the normalization of age-related ECs via multiple related ECs exhibited a stronger positive correlation across da-
Deming regression (Figure S4) and quantitative meta-analysis tasets when estimated using significant signature genes
based on a linear mixed-effect model (see STAR Methods), we (Figures 3A and S3C). To verify the introduced algorithm, we
Cell 186, 1–21, June 22, 2023 5

ll
Article
conducted 4-fold cross-validation, identifying significant age- aging were negatively correlated with several signatures of inter-
associated biomarkers in a training set and applying them ventions, including CR, rapamycin, and biomarkers of maximum
to calculate the correlation between ECs from independent and median lifespan (median Spearman r = 0.11) (Figures 4A
test datasets. The utilized normalization and meta-analysis and 4B). This dependence may be partly driven by the decreased
approach resulted in higher correlations across test ECs (p biological age of intervention-treated animals, as previously
adjusted = 1015) (Figure S3D), supporting the generalizability demonstrated at the level of DNA methylation.66–68 In contrast,
of the discovered molecular features of aging. features of long-lived species strikingly showed a significant
Aging signatures of all examined tissues and species were positive association with aging across different tissues (median
positively correlated with each other (Figures 3B and 3C). Spearman r = 0.16). Since animals used to identify signatures
Some of them also exhibited significant pairwise overlaps be- of species longevity had a similar biological age (young adults),
tween aging-associated genes (Figures 3D and 3E). Similar to this finding can hardly be explained by selection bias. Rather,
biomarkers of species longevity, age-related transcriptomic it seems to indicate that not all age-related changes are harmful
changes were mainly driven by widely expressed genes, de- and that adaptive (compensatory) molecular mechanisms regu-
tected, on average, in >48% of the cell types (Figure S3E). Blood lated by gene expression are also involved,69 being at the same
cell-specific biomarkers accounted for <5% of the aging-associ- time selected in long-lived species. Interestingly, transcriptomic
ated genes, and their removal did not affect the significant pos- patterns of longevity across species and lifespan extension
itive correlation between age-related ECs of individual tissues within species did not show a strong correlation, exhibiting
and species (Spearman r > 0.1; p adjusted < 105), suggesting both shared and distinct features (Figures 4A and 4B).
that molecular mechanisms of aging are generally conserved To establish if the observed global patterns were reproduced
across various cell types. in evolutionarily distinct taxons of life, we performed a similar
Among top genes associated with aging in most tissues and analysis in the budding yeast model, utilizing (1) gene expression
species, we identified Vsig4 and Nrep, up- and downregulated biomarkers of replicative lifespan (RLS) across 40 naturally
with age, respectively (global signature p adjusted < 0.05) (Fig- evolved strains of Saccharomyces cerevisiae,70 (2) signatures
ure 3F). Vsig4 encodes an immune checkpoint protein, whose of RLS in lab yeast strains affected by 1,376 single-gene dele-
expression is positively correlated with the physiological frailty tions,71,72 and (3) transcriptomic biomarkers of yeast replicative
index in male mice,56 and the progression of cancer57–59 and aging.73 Interestingly, the interplay between yeast signatures of
several inflammatory diseases60 in humans. Downregulation aging, lifespan-extending deletions, and longevity across
of Nrep was found to produce learning and memory defects,61 evolved strains was consistent with the patterns observed in
blood pressure abnormalities,62 and obesity.63 Based on our mammals (Figure S5A). Gene expression signatures of yeast ag-
meta-analysis, these genes may be considered universal dele- ing were positively correlated with those of long-lived natural
terious signatures of aging, shared by multiple species and strains and negatively with those of lifespan-extending muta-
tissues. tions, suggesting that divergence of molecular mechanisms of
Individual tissues and species demonstrated similar transcrip- longevity in response to simple interventions and selection at
tomic age-related changes both at the level of individual genes great evolutionary distances may be a universal pattern shared
(Figure 3H) and enriched functions (Figure 3I). Functional GSEA across distinct branches of life.
revealed a number of shared aging mechanisms across organs
and species (Table S3B), including the upregulation of pathways Common and distinct biomarkers of longevity and aging
associated with established hallmarks of aging, such as biosyn- To reveal shared and distinct molecular features of mammalian
thesis of reactive oxygen species, senescence-associated longevity and aging, we aggregated individual transcriptomic
secretory phenotype (SASP), and inflammation.64,65 In contrast, signatures within every model using HMP,48 obtaining >450
genes involved in energy metabolism and mitochondrial transla- statistically significant genes in each case. Consistent with the
tion were significantly downregulated with age, according to correlation analysis, we observed a significant overlap between
multiple signatures (Figure 3I). co-regulated aggregated signatures of aging and long-lived
species (p = 1.8 3 1020), whereas aggregated biomarkers of
Global interplay between signatures of longevity lifespan-extending interventions were enriched for genes coun-
and aging teracting aging (p = 7.5 3 106) (Figure 4C).
To examine the relationship between molecular mechanisms of Interestingly, 9 genes were significantly co-regulated with age
longevity and aging, we performed a correlation analysis of the and in both longevity models (Table S4). They included Igf1, a
identified signatures along with mouse gene expression patterns crucial component of insulin and IGF-1 signaling.20,74 Reduced
of lifespan-extending interventions discovered previously33 (Fig- activity of IGF-1 extends lifespan in various species ranging
ure 4A). The latter included genes differentially expressed in from yeasts to mice.74 IGF-1 plasma level in mice is decreased
response to individual interventions (CR, rapamycin, and muta- by established lifespan-extending interventions, including
tions associated with GH deficiency), shared biomarkers of GHRKO75 and CR,76 and declines with age,77 providing an
longevity interventions, and ECs associated with their effect on example of a healthspan-promoting aging feature. Based on
murine maximum and median lifespans. Transcriptomic signa- our data, decreased expression of Igf1 is a universal biomarker
tures were generally consistent within each examined model: of aging and longevity (Figure 4D, upper left), shared by life-
aging, species longevity, and lifespan-extending interventions span-extending interventions and long-lived mammals, even af-
(median Spearman r > 0.55 within each model). Biomarkers of ter adjustment for AW (p < 0.012 for liver signatures).
6 Cell 186, 1–21, June 22, 2023

ll
Article
Figure 4. Interplay between transcriptomic signatures of aging and longevity

(A) Denoised Spearman correlation of signatures of aging (red), lifespan-extending interventions (green), and species longevity (blue). Asterisks reflect statistical
significance of each pairwise correlation.
(B) Normalized expression changes (ECs) of genes associated with aging and longevity. Only genes significantly associated with at least one trait are shown
(p adjusted < 0.05). x axis represents individual signatures.
(C) Overlap of aging- and longevity-associated genes. Number of aggregated signature genes and deviation from the expected random overlap are indicated with
text and color, respectively. p value was estimated with Pearson’s chi-square tests.
(D) Association of Igf1, Mrps15, Ndufa9, Lgals1, Rela, and C1qb expressions with aging and longevity. Statistical significance of each association denoted with
asterisks was estimated with phylogenetic regression (for species longevity signatures) and mixed-effect model (for other signatures). Data are mean normalized
ECs ± SE.
(E) Functional enrichment of gene signatures shared by several models. Presented functions were annotated by Kyoto Encyclopedia of Genes and Genomes
(KEGG). Missing data are shown in gray.
(F) Number of aggregated, shared, and distinct signature genes associated with aging and longevity. For each pair of models, the difference between the number
of co-regulated and distinct genes was assessed using two-sample proportion tests.
CR, calorie restriction; GH, growth hormone; Common, shared signatures of longevity interventions; Median/Max lifespan, signatures of intervention effect on
median/maximum mouse lifespan; ML, maximum lifespan; MLres, maximum lifespan residual; Longevity, shared biomarkers of lifespan-extending interventions
and long-lived species.
See also Figure S5 and Table S4.
In contrast, Lgals1 demonstrated a positive association with Gal-1 was also decreased in the cerebrospinal fluid of patients
all examined models (Figure 4D, upper right). Its product, with Parkinson’s disease.82 Therefore, its upregulation with
Galectin-1 (Gal-1), regulates cell proliferation and has a promi- age, especially in the brain (p adjusted = 0.004), and in the
nent anti-inflammatory and proangiogenic activity78,79 that is longevity models may provide a beneficial effect, protecting
protective during recovery from cardiovascular diseases.80 against aging-related chronic inflammation.
Mice lacking Gal-1 demonstrated cardiac inflammation and Utilizing Fisher’s combined probability test,83 we identified a
increased susceptibility to chronic inflammatory diseases.81 comprehensive list of common and distinct molecular biomarkers
Cell 186, 1–21, June 22, 2023 7

ll
Article
of longevity and aging (see STAR Methods). Consistent with pre- NAD+ concentration in murine liver was substantially increased
vious results, we detected more distinct than shared features of following longevity interventions (Figure 5F). Lifespan extension
aging and lifespan-extending interventions (p adjusted = was also accompanied by upregulation of genes involved in
0.0045), whereas more co-regulated signatures were observed NAD+ biosynthesis, including Naprt, Nampt, Nmnat3, and
for aging and long-lived species (p adjusted = 7.4 3 1048) and Nadsyn1, encoding NAD+ synthetase 1 (p adjusted < 0.008 for
for long-lived species and interventions (p adjusted = 0.012) (Fig- association with median and maximum mouse lifespan)
ure 4F). Shared signatures of species longevity and aging included (Figures 5I and S5C). In contrast, we did not observe a significant
genes involved in oxidative phosphorylation (e.g., Ndufa9), association between NAD+ levels and ML across species. How-
apoptosis (e.g., Rela), and complement and coagulation cas- ever, there was a slight accumulation of its precursor, quinolinic
cades (e.g., C1qb) (Figures 4D and 4E). Interestingly, the expres- acid, in long-lived species (p adjusted = 0.09 for multi-tissue ML
sion of these genes was changed in the opposite direction by signature), along with the upregulation of Nadsyn1 in the brain,
lifespan-extending interventions. In contrast, some signatures even after the adjustment for AW (p < 0.007) (Figures 5I and
were shared by both longevity models but not aging, including S5C). Therefore, NAD+ biosynthesis pathway may be involved
several mitochondrial ribosomal protein genes (e.g., Mrps15) in lifespan regulation, although its role appears to be more prom-
(Figures 4D and 4E). Finally, >200 gene expression biomarkers inent at the intraspecies level (Figure 5B).
were shared across aging and both longevity models (Figure 4F), Genes involved in NAD+ level regulation were also associated
further indicating that age-related changes may include not only with aging and longevity in yeast. The deletion of genes encoding
harmful but also beneficial compensatory effects. mitochondrial NAD+ transporters extended the chronological
lifespan of S. cerevisiae, whereas their overexpression resulted
Longevity-associated metabolic pathways in a shorter lifespan.91 We observed the downregulation of mito-
To uncover specific metabolic mechanisms of longevity chondrial NAD+ carrier NDT1 in response to RLS-extending de-
(Figures 5A–5C), we expanded gene expression signatures with letions and across long-lived natural strains, and its upregulation
the data on metabolite profiling.27,33 Using a phylogenetic regres- in aged yeast cells (Figure S5D). Besides, NPT1, involved in
sion pipeline, we discovered metabolite biomarkers of ML across NAD+ biosynthesis, was highly expressed in yeasts subjected
26 mammalian species in various tissues, including brain, liver, to RLS-extending interventions (p adjusted < 6.7 3 103).
kidney, and heart. By comparing metabolite profiles of wild-type Finally, metabolites negatively associated with mouse lifespan
mice and mice subjected to longevity interventions (acarbose, ra- included L-cystathionine (Figure 5G), an intermediate product of
pamycin, GHRKO, and Snell dwarf mice), we also identified methionine degradation. Depletion of this molecule was accom-
metabolite signatures of lifespan extension within species. panied by significant upregulation of genes responsible for its
Comparison of the identified metabolites revealed shared conversion into glutathione,92 including Cth, Gclc, and Gss (me-
biomarkers of longevity, including adenosine, xanthosine, and dian p < 0.007 for association with median and maximum mouse
succinic acid, whereas several others (e.g., uric acid) showed lifespan) (Figure 5J). This pathway also leads to the production of
a distinct effect (Figure S5B). The metabolite signatures H2S, which accumulates in mice subjected to CR76 and by itself
(Figures 5D–5G) were strongly supported by the corresponding extends the lifespan of roundworms.93 Our results support the
transcriptomic signatures (Figures 5H–5J), allowing us to role of this pathway in longevity regulation, as it is induced by life-
characterize metabolic pathways with consistent multi-omics span-extending interventions both at metabolite and gene
association. expression levels (Figure 5C). Surprisingly, genes involved in
Uric acid exhibited one of the strongest associations with life- cystathionine degradation were downregulated in long-lived
span (Figure S5B). Its concentration in various tissues was higher species (median p < 0.004 for liver ML signature) (Figure 5J),
in long-lived species but was reduced by lifespan-extending in- providing another example of distinct longevity-associated fea-
terventions in mice (Figure 5D), whereas its direct metabolite tures within and across species.
allantoin showed the opposite behavior (Figure 5E). In mammals,
urate is converted to allantoin by urate oxidase (uricase).84 Cellular processes mediating longevity and aging
Accordingly, the expression of uricase gene Uox in the liver Functional enrichment analysis of individual transcriptomic
was negatively correlated with ML across species (Figure 5H, signatures (Figure 6A; Table S3) and significant co-regulated
upper). In part, this is related to the pseudogenization of the uri- and distinct biomarkers across various models (Figure 6B;
case gene in hominoids,85–87 although its expression was also Table S5A) highlighted cellular processes that mediate longevity
low in other long-lived mammals, including naked mole rats regulation and aging. Resembling results obtained on individual
and Damaraland mole rats. In contrast, Uox was upregulated genes, age-associated functional changes were positively
by lifespan-extending interventions in mice (Figure 5H, lower). correlated with patterns of species longevity and negatively
Therefore, long-lived mammals seem to accumulate uric acid correlated with features of lifespan-extending interventions
through the reduced activity of Uox, whereas the opposite hap- (Figure S6A).
pens in mice subjected to longevity interventions (Figure 5A). Upregulation of ribosome protein genes was among the most
Another metabolite regulated by lifespan-extending interven- consistently shared functional signatures of aging and both
tions is nicotinamide adenine dinucleotide (NAD+). NAD+ levels longevity models (Figures 6A and 6B). However, although genes
decline with age and affect SIRT1 function, whereas supplemen- encoding cytosolic ribosomal proteins demonstrated a positive
tation with nicotinamide mononucleotide (NMN) or nicotinamide association with all examined traits, mitochondrial ribosomal
riboside (NR) can improve healthspan in mouse models.88–90 protein genes were upregulated in both longevity models but
8 Cell 186, 1–21, June 22, 2023

ll
Article
Figure 5. Molecular pathways associated with mammalian longevity

(A–C) Pathways associated with lifespan-extending interventions and species longevity. Metabolites and genes are shown in rectangles and Italic, respectively.
Arrows represent significant positive (up) and negative (down) associations with longevity within (green) and across (blue) species.
(D–G)Association of uric acid (D), allantoin (E), NAD+ (F) and cystathionine (G) concentration with longevity within (green) and across (blue) species. The vertical
axis reflects metabolite normalized logFC in murine liver in response to individual lifespan-extending interventions and aggregated across interventions
(Common); slope of association with mouse median (Lifespan.median) and maximum lifespan (Lifespan.max); and slope of association with species maximum
lifespan (ML) and maximum lifespan residual (MLres) in individual organs and aggregated across organs (Pooled). Error bars are ±SE.
(H–J) Association of Uox (H), Nadsyn1 (I), and Cth (J) expressions with lifespan extension induced by interventions (survival curve icon) and across species
(mouse/whale icon). Examined organs are reflected with liver and brain icons. The black line, slope p value and R2 indicate the model fitted with phylogenetic (for
species) or linear (for interventions) regression. Data are mean ± SE.
CR, calorie restriction; EOD, every-other-day feeding; oe, overexpression; M, male; F, female; NA, nicotinic acid; NaMN, nicotinic acid mononucleotide; NaAD,
nicotinic acid adenine dinucleotide; NAD, nicotinamide adenine dinucleotide; NMN, nicotinamide mononucleotide; NAM, nicotinamide; SAM, S-sdenosyl
methionine; GSH, glutathione.
See also Figure S5.
downregulated with age (Figures 4D, 4E, and 6A). Mitochondrial drial function may be an essential conserved mechanism of life-
translation was one of the most significant pathways enriched for span regulation.
gene expression signatures that separated longevity and aging Energy metabolism, including the TCA cycle and oxidative
(p adjusted = 5.7 3 106) (Figure 6B), suggesting that mitochon- phosphorylation, was significantly co-downregulated with age
Cell 186, 1–21, June 22, 2023 9

ll
Article
Figure 6. Common and distinct transcriptomic signatures of longevity

(A) Functional enrichment (GSEA) of longevity and aging-associated signatures. Only functions significantly enriched by at least one signature (p adjusted < 0.1)
are presented. The whole list of enriched functions is in Table S3. NES, normalized enrichment score.
(B) Functional enrichment of aggregated, shared, and distinct signatures of longevity and aging. Proportion of pathway-associated genes and statistical sig-
nificance estimated with Fisher exact test are reflected by bubble size and color, respectively. The whole list of functions is in Table S5A.
(C) Partial correlation network of gene expression signatures of longevity and aging. Sign of partial correlation coefficient (PCC) is indicated by color. AW, adult
weight; Ints, signature of interventions effect on maximum lifespan; AA Met, metabolism of amino acids and derivatives; adaptive immune, adaptive immune
system; complement, complement cascade; FA Met, fatty acid metabolism; innate immune, innate immune system; Interf g, interferon gamma signaling; Mt
trans, mitochondrial translation; Oxid Phosph, oxidative phosphorylation; Ribo, ribosome.
(D) Effect of cardamonin (left), clofilium tysolate (middle), and deguelin (right) on survival of fibroblasts from species with different lifespans following paraquat-
induced oxidative stress. y axis displays log-ratio of the number of survived fibroblasts treated and untreated with the compound. x axis shows maximum lifespan
unadjusted (top) and adjusted (bottom) for adult weight. Slope p value was assessed with mixed-effect linear model. Data are mean ± SE. n = 3–6 per treated and
control group for every species/strain.
(E) Evolutionary features of longevity and aging-associated aggregated signature genes. Data are mean ± SE. Statistical significance of enrichment denoted with
asterisks was assessed with Fisher’s exact test.
(F) Common (bold), distinct, and unique transcriptomic signatures of longevity across and within species.
ML, maximum lifespan; MLres, maximum lifespan residual.
See also Figure S6 and Tables S3 and S5.
10 Cell 186, 1–21, June 22, 2023

ll
Article
and in long-lived species but upregulated by lifespan-extending strains), rat, naked mole rat, western long-eared myotis, Yuma
interventions in mice (Figures 4D, 4E, 6A, and 6B). In contrast, myotis, and human to oxidative stress using paraquat (dose
genes associated with some branches of an innate immune close to LD50). Prior to stress induction, we treated cells with
response, including complement and coagulation cascades several compounds that promote mouse cell survival under
and tumor necrosis factor (TNF) signaling, were co-upregulated these conditions100: cardamonin (CD), an inhibitor of NF-kB101
in aged animals and long-lived species, but not in response to and an activator of NRF2100,102; clofilium tysolate (CT), a K+
lifespan-extending interventions (Figures 4D, 4E, 6A, and 6B). channel blocker that facilitates mitochondrial DNA replication
This association was also shared by Rela, a component of the and restores mitochondrial mass in mutants with defective
NF-kB complex, a crucial regulator of inflammation and polymerase g103,104; and deguelin (DG), a rotenoid that
apoptosis.94 inhibits PI3K/Akt signaling105 and upregulates mitochondrial
Inflammation and NF-kB activity facilitate the development of translation genes.106 We assessed if these compounds would
numerous age-related diseases,94–97 and Rela targets are upre- also improve viability of cells obtained from long-lived species.
gulated with age in several murine organs.98 Based on our data, As expected, all three compounds improved the average survival
Rela expression is a universal biomarker of aging upregulated of mouse and rat fibroblasts subjected to oxidative stress (p
across different tissues, whereas its downregulation is associ- adjusted < 0.011) (Figure 6D). However, the effect of CD was
ated with the extension of mouse lifespan (Figure 4D). Interest- significantly reduced in fibroblasts of long-lived species,
ingly, Rela and other genes encoding NF-kB components were whereas CT and DG induced similar cell survival independent
not downregulated in long-lived mammalian species, instead of species lifespan (Figure 6D, upper). The dependence of CD
showing a trend toward positive association with species stress resistance effect on ML was even stronger after adjust-
longevity (Figure 4D). ment for animal AW (Figure 6D, bottom) and baseline level of
To examine the relationship between the identified mecha- cell survival under control conditions (Figure S6B), whereas the
nisms, we computed a sparse partial correlation network for response to CT and DG did not depend on species ML after
features associated with longevity and aging (Figure 6C). Spe- any of these adjustments. Therefore, the compound that tar-
cies AW was positively associated with the expression of genes geted molecular mechanisms of mouse lifespan extension (NF-
involved in DNA repair and innate immune system response, kB and NRF2 signaling) but not signatures of long-lived species
pointing to the role of these pathways in cancer prevention, was able to improve stress resistance of cells from short-lived
in agreement with Peto’s paradox.99 Even after the adjustment species only. In contrast, compounds that targeted shared
for AW and other features, species longevity was positively mechanisms of longevity (mitochondrial function and insulin
correlated with the expression of genes associated with signaling) had a comparable beneficial effect on cell survival in-
complement cascade, DNA repair, translation, and aging and dependent of mammalian lifespan.
negatively correlated with Igf1 expression and regulation of To make the identified transcriptomic signatures available to
proteolysis and fatty acid metabolism. Complement cascade the research community, we developed an interactive database,
and proteolysis were strongly connected to aging, partially ex- mSALT (mammalian Signatures of Aging and Longevity Traits;
plaining its unexpected positive correlation with signatures of http://gladyshevlab.org/mSALT/). For every gene, mSALT pro-
long-lived species. In contrast, intervention-induced mouse vides (1) a general association of its expression with various
lifespan extension was associated with the upregulation of models of mammalian aging and longevity (Figure 4D); (2) the
metabolic pathways (oxidative phosphorylation, fatty acid relationship between its expression and longevity across
metabolism, and amino acid metabolism), translation and mammalian species (Figures 2A–2C); (3) the relationship be-
DNA repair, and downregulation of aging biomarkers, even af- tween its expression and effect of various interventions on
ter adjustment for all the examined features. Overall, longevity mouse lifespan (Figures 5H–5J); and (4) its age-related EC
within and across species seems to be driven by multiple inter- across various tissues and species (Figure 3F).
connected mechanisms, including age-related and indepen-
dent regulators of lifespan. Evolutionary features of longevity-associated genes
We hypothesized that molecular signatures of lifespan-ex- To examine evolutionary features of longevity and aging bio-
tending interventions in mice that are not shared by long-lived markers, we tested their enrichment for several characteristics,
species, such as the inhibition of complement cascade and including involvement in the basic cellular maintenance, evolu-
NF-kB pathway, may reflect marginally effective ways to pro- tionary age, mutation intolerance, and haploinsufficiency44
mote longevity by regulating the response to the already accu- (see STAR Methods). For every trait, we contrasted proportions
mulated damage. Since they have not been selected during of signature genes with opposite features (e.g., mutation
the evolution of long-lived mammals, such approaches can be tolerant/intolerant, evolutionary old/young, etc.), using randomly
less effective in these organisms compared with short-lived spe- selected non-signature genes with a similar average expression
cies. In contrast, common signatures of longevity, including the level as a background (Figure S6C).
upregulation of mitochondrial function and downregulation of Significant overrepresentation of housekeeping genes was
Igf1, may represent core mechanisms that affect damage accu- observed for aging biomarkers (odds ratio [OR] = 1.19,
mulation itself and can be effectively targeted in both short- and p adjusted = 0.038) (Figure 6E), indicating that aging is deeply
long-lived species. connected with the reduction of core cellular homeostasis. Sig-
To test this hypothesis, we subjected fibroblasts of mouse natures of species longevity were enriched for evolutionary
(inbred C57BL/6J and genetically heterogeneous UM-HET3 ancient genes, originated before the emergence of bony
Cell 186, 1–21, June 22, 2023 11

ll
Article
vertebrates (OR = 1.5, p adjusted = 0.02), whereas transcrip- Other examined models induced ECs negatively associated
tomic biomarkers of lifespan-extending interventions were with longevity interventions but positively associated with aging
mostly evolutionary young genes originated in tetrapods (OR = (Figure 7A, upper). Interestingly, some of them were also nega-
1.23, p adjusted = 0.003) (Figure 6E). tively correlated with aggregated signatures of long-lived spe-
Gene essentiality was assessed by mutation intolerance, i.e., cies (e.g., Dicer1 knockout), whereas others had no or even pos-
probability of being intolerant to gene loss-of-function mutation, itive association (e.g., Rap1 knockout and p21 overexpression).
and haploinsufficiency, i.e., sensitivity to a gene copy-number The identified models indeed exhibited age-related pathological
reduction. According to both metrics, signatures of long-lived spe- phenotypes. Thus, high-fat diet produced hepatocellular dam-
cies were enriched for essential genes (OR > 2, p adjusted < 104) age, fibrosis, and reduced mitochondrial density in the liver
(Figure 6E). Biomarkers of lifespan-extending interventions and markedly affected lifespan of male C57BL/6J mice.111
showed the opposite pattern, being depleted for such genes IKK2 activation in the liver generated severe chronic inflamma-
(OR < 0.77, p adjusted < 0.004), underlying a significant difference tion, also leading to organ fibrosis.115 Rap1 knockout induced
in routes of lifespan regulation across and within species (Fig- glucose intolerance, liver steatosis, and excess fat accumulation
ure 6F). Essential evolutionarily old genes associated with species associated with obesity.112 Hepatocyte-specific deficiency of
longevity were enriched for PI3K-Akt, cytokine, TNF and MAPK Dicer1, an enzyme involved in miRNA processing, produced pro-
signaling, ubiquitin-mediated proteolysis, and cell cycle (Fig- gressive cellular damage, apoptosis, and inflammation in 2- to
ure S6D; Table S5B), whereas evolutionarily young, mutation- 4-month-old mice,114,120 whereas its knockout in the skeletal
tolerant signatures of lifespan-extending interventions were muscle resulted in decreased muscle mass and abnormal myo-
involved in metabolic pathways (fatty acid metabolism, respiratory fiber morphology.121 Finally, overexpression of tumor suppres-
electron transport, and peroxisome proliferator-activated recep- sor p21 induced cellular senescence in human cells122,123 and
tor [PPAR] signaling), and complement and coagulation cascades contributed to the development of muscle atrophy.113 Thus,
(Figure S6E; Table S5C). aggregated signatures of lifespan-extending interventions and
aging reasonably discriminate between potentially detrimental
Longevity signatures reveal lifespan-regulating and beneficial interventions based on their effect on murine
interventions gene expression, whereas positive associations with biomarkers
To discover genetic, environmental, and pharmacological inter- of long-lived species do not always result in the improvement of
ventions, whose effect on mammalian gene expression resem- healthspan in short-lived mammals, as demonstrated by p21
bles transcriptomic signatures of aging and longevity, we utilized overexpression and Rap1 deficiency models.
GeneQuery, a tool that searches for public datasets with similar Based on CMap predictions, we selected 3 chemical com-
ECs, and Connectivity Map (CMap), a resource containing tran- pounds that induced pro-longevity ECs according to signatures
scriptomic profiles of human cells subjected to thousands of of lifespan-extending interventions and long-lived species,
chemical compounds.107,108 including PI3K inhibitor GDC-0941,124 PKCb and PI3K/Akt
Using GeneQuery, we identified 7 interventions in mice that pathway inhibitor enzastaurin,125 and MEK and TNF-a inhibitor
induced ECs significantly associated with aggregated signatures AS-703026.126,127 We applied these compounds orally to UM-
of longevity and aging, including a hepatocyte-specific condi- HET3 male mice for 1 month and performed RNA-seq on
tional knockout of Keap1,109 chronic hypoxia,110 high-fat their liver and kidney samples (Table S1C). Additionally, we
diet,111 deficiency of telomere-binding protein Rap1,112 overex- expanded these data with transcriptomic profiles of mice
pression of p21 in skeletal muscle,113 deficiency of Dicer1 in he- subjected to 3 drugs identified in our previous study33: mTOR
patocytes,114 and constitutive expression of NF-kB activator inhibitors KU0063794128 and AZD8055,129 and antioxidant as-
IKK2 in the liver115 (Table S6A). corbyl-palmitate.130
Utilizing a GSEA-based approach (see STAR Methods), we By comparing expression profiles of treated mice with age-
found that ECs generated by Keap1 knockout and chronic hyp- matched control samples, we identified compound-induced
oxia in mouse liver were positively associated with aggregated ECs in murine organs and examined their association with bio-
and multiple individual signatures of lifespan-extending interven- markers of aging and longevity (Table S6B). Consistent with
tions (CR, GH deficiency, etc.) (Figure 7A, upper). Accordingly, CMap predictions, all selected compounds generated changes
hypoxia response was protective against mitochondrial dysfunc- positively associated with an aggregated signature of at least
tion associated with multiple aging-related diseases,116,117 and one longevity model (Figure 7A, lower). Moreover, pro-longevity
chronic hypoxia extended healthspan and lifespan of mice with effects of KU0063794, ascorbyl-palmitate, AZD8055, and GDC-
genetic mitochondrial disease produced by Ndufs4 knockout.116 0941 were supported simultaneously by aggregated biomarkers
KEAP1, an inhibitor of acute stress regulator NRF2, also signifi- of lifespan-extending interventions (in the kidney and the liver;
cantly affects longevity. Its loss-of-function mutations extended p adjusted < 0.004) and long-lived species (in kidney;
the median lifespan of Drosophila melanogaster males,118 p adjusted < 6 3 104), along with multiple individual signatures
whereas the overexpression of the NRF2 ortholog (SKN-1) (GH deficiency, CR, maximum and median lifespan, etc.).
increased the average lifespan of roundworms.119 Based on To test if compounds that induce longevity-associated
our data, Keap1 knockout and chronic hypoxia generate ECs ECs extend murine lifespan and healthspan, we subjected
associated with intraspecies longevity and do not significantly 25-month-old C57BL/6 male mice to a diet containing a top hit
perturb aggregated biomarkers of aging or long-lived species, from our analysis, KU0063794 (Figure 7B). KU0063794 at 10
being good candidates for lifespan extension in healthy mice. ppm extended the remaining median and ML of old mice by
12 Cell 186, 1–21, June 22, 2023

ll
Article
Figure 7. Transcriptomic longevity signatures reveal lifespan-regulating interventions

(A) Association of murine gene expression response to genetic, environmental (upper) and pharmacological (lower) interventions with signatures of aging (red),
long-lived species (blue) and lifespan-extending interventions (green). Aggregated (left) and individual (right) signatures were utilized. Statistical significance
denoted with asterisks was estimated with GSEA-based test. The output of the association analysis is in Table S6.^p adjusted < 0.1; *p adjusted < 0.05; **p
adjusted < 0.01; ***p adjusted < 0.001.
(B) Design of the study testing the effect of KU0063794 on mouse survival and healthspan.
(C) Survival curves of male C57BL/6 mice subjected to KU0063794 (10 ppm) at 25 months old. p value was calculated with log-rank tests.
(D) Body weight of 30-month-old control and KU0063794-treated male C57BL/6 mice.
(E) Median time to finish measured in 30-month-old control and KU0063794-treated male C57BL/6 mice.
(F) Frailty index score of 30-month-old control and KU0063794-treated male C57BL/6 mice.
(G) Percentage of immune cell types in spleens of 27-month-old control and KU0063794-treated male C57BL/6 mice.
(H) Glucose tolerance in 24-month-old control and KU0063794-treated C57BL/6 mice. n = 10 for each group.
(I) Area under the curve (AUC) of glucose level shown in (H).
p values on (D)–(I) were calculated with one-tailed Wilcoxon rank-sum tests.
NES, normalized enrichment score; CR, calorie restriction; GH, growth hormone; ML, maximum lifespan; MLres, maximum lifespan residual; KO, knockout; OE,
overexpression; WAT, white adipose tissue.
See also Figure S7 and Tables S1, S6, and S7.
32.6% and 10.9%, respectively (log-rank test p = 0.038) (Fig- ment and observed a slightly reduced percentage of T cells
ure 7C), with no effect on animal body weight (Figure 7D). relative to the total number of CD45+ cells (Figure 7G).
KU0063794 also improved mouse gait speed measured at KU0063794 also affected the proportion of follicular B cells,
30 months (Figure 7E). The frailty index of mice before and but not other B cell populations (Figure S7C). Age-associated
5 months after the treatment initiation showed no difference be- clonal B cells (ACBCs), which produce B cell lymphoma in
tween the control and experimental groups prior to drug aged mice,132 were unaffected by KU0063794, suggesting that
supplementation (Figure S7A); however, mice subjected to its lifespan-extending effect is not driven by the delay of B cell
KU0063794 were significantly less frail following the treatment lymphoma.
(Figure 7F). Detailed analyses also revealed a KU0063794- Since chronic rapamycin treatment also leads to glucose intol-
induced improvement of coat and eye-related features erance in mice,131 we performed a glucose tolerance test on a
(Figure S7B). separate cohort of 24-month-old male mice subjected to
Chronic treatment with another mTOR inhibitor, rapamycin, KU0063794 for 2 months. KU0063794 did not affect glucose
decreases the percentage of T cells in secondary lymphoid or- tolerance in old mice as there was no difference in glucose clear-
gans.131 We tested if KU0063794 affected immune cell propor- ance dynamics between the control and treated groups
tion in the spleen of 27-month-old mice after 5 months of treat- (Figures 7H and 7I).
Cell 186, 1–21, June 22, 2023 13

ll
Article
Necropsy analyses of mice from the lifespan cohort that features of mouse lifespan extension, suggesting that the bene-
died from natural causes or were euthanized due to the ficial effect of established longevity interventions may be associ-
moribund state did not reveal significant differences in ated with the deceleration of damage accumulation resulting in a
incidences of individual pathologies between control and younger biological age. At the same time, biomarkers of long-
treated groups (Table S7). Larger follow-up studies may pro- lived species were positively correlated with age-related
vide more data on specific healthspan-promoting effects of changes. Genes responsible for this trend were involved in en-
KU0063794. ergy metabolism (e.g., oxidative phosphorylation) and certain
branches of the innate immune response (e.g., complement
DISCUSSION cascade). This finding points to the existence of various strate-
gies to achieve longevity (Figure 6F), some of which may be
In this work, we characterized gene expression signatures of marginally efficient for short-lived species but do not provide a
mammalian lifespan in 3 organs. Many patterns of species long-term benefit on an evolutionary timescale. In agreement
longevity were conserved across tissues, including up- and with this hypothesis, CD, an inhibitor of the NF-kB complex,
downregulation of genes associated with DNA repair and protein improved the survival of fibroblasts from short-lived species un-
degradation, respectively. Interestingly, they were also observed der oxidative stress but was significantly less effective in cells
in primary fibroblasts,30 indicating that shared molecular mech- from long-lived mammals.
anisms of longevity are presumably driven not only by the sys- A positive association between biomarkers of aging and spe-
temic effect of circulating signaling molecules (e.g., IGF-1) but cies longevity can also be driven by the multifaceted nature of
also by genetically encoded features (e.g., mutations in regulato- age-related processes, incorporating both deleterious and adap-
ry regions). Activation, rather than suppression, of proteasome tive effects, a pattern also observed for epigenetic changes in hu-
or autophagy extends the lifespan of model organisms, including mans.69 This hypothesis is further supported by the existence of
roundworms,133,134 fruit flies,135 and mice.136 Downregulation of signatures shared by aging and both longevity models, such as
proteolysis in long-lived species may be caused by higher upregulation of ribosomal protein genes and downregulation of
accuracy of protein synthesis, lower level of damage, or better Igf1. The common adaptive effects can reflect the slowdown of
maintenance of the translation machinery.137 Indeed, protein damage accumulation achieved through the reduction of protein
turnover rate is negatively correlated with species lifespan in turnover rate and translation, which are associated with
mammals.138 increased lifespan across species,138 and occur during aging142
The identified transcriptomic signatures were concordant with and in response to lifespan-extending interventions.143 On the
the metabolite data. Thus, urate concentration was positively other hand, shared biomarkers of longevity that display opposite
correlated with species longevity, and the uricase gene expres- association with aging, such as upregulated mitochondrial trans-
sion was downregulated in long-lived mammals. Interestingly, lation, may represent robust determinants of lifespan regulation
the opposite effect was produced by lifespan-extending inter- through counteraction of harmful age-related changes driving
ventions in mice, demonstrating that different molecular mecha- the physiological deterioration. Accordingly, compounds that
nisms may be associated with longevity on short-term and targeted these mechanisms provided similar improvement in
evolutionary timescales. The double-edged sword effect of urate cell survival to fibroblasts from short- and long-lived mammals.
on healthspan is supported by its role both as an antioxidant and Another notable difference between molecular signatures of
activator of the inflammasome, contributing to metabolic disor- longevity within and across species was the evolutionary age
ders.139 Similarly, activation of genes involved in methionine and essentiality of the corresponding genes. Biomarkers of
metabolism was a specific feature of lifespan extension in long-lived species were enriched for evolutionarily ancient, mu-
mice, whereas the upregulation of NAD+ biosynthesis induced tation-intolerant genes, whereas lifespan-extending interven-
by longevity interventions was only partially shared by long-lived tions mainly affected younger genes tolerant to mutations and
mammals. copy-number variation. Overall, the data suggest that longevity
Mammalian age-related transcriptomic changes were also on the evolutionary timescale is achieved by fine-tuning of funda-
generally similar across different organs and species. Many of mental machinery deep-rooted in natural history that affects
them were related to the established hallmarks of aging, primary damage emergence and rate of its accumulation (i.e.,
including mitochondrial dysfunction, senescence, and inflam- upregulation of DNA repair and innate immune response, and
mation.64,65 Thus, detrimental molecular processes leading to downregulation of the IGF-1 and PI3K-Akt pathway).9 On the
the loss of physiological integrity appear to be partly modulated other hand, most established lifespan-extending interventions
at the level of gene expression. Secretory signaling molecules seem to operate through the modulation of less conserved com-
associated with the accumulated damage, such as inflammatory ponents associated with metabolic remodeling (i.e., upregula-
cytokines, may be partially responsible for the induction of com- tion of oxidative phosphorylation and fatty acid and amino acid
mon aging-related changes across tissues, supported by the metabolisms) and deceleration of secondary detrimental pro-
systemic effect of heterochronic parabiosis and plasma dilution cesses induced in response to already accumulated damage
on mouse organs.140,141 (i.e., downregulation of NF-kB, and complement and coagulation
Comparison of the identified transcriptomic signatures of cascades). These two mechanisms may represent complemen-
mammalian aging, species longevity, and lifespan-extending in- tary ways to deal with molecular damage and provide an instru-
terventions in mice revealed a complex interplay between these ment to search for efficient combinatorial therapies, targeting
traits. Overall, aging signatures were negatively correlated with both longevity strategies simultaneously.
14 Cell 186, 1–21, June 22, 2023

ll
Article
By employing the discovered signatures of longevity and ag- B Glucose tolerance test
ing, we identified several candidates for mouse healthspan and B Flow cytometry
lifespan extension, including chronic hypoxia, hepatocyte-spe- B Necropsy analysis
cific Keap1 knockout, KU0063794, AZD8055, GDC-0941, and d QUANTIFICATION AND STATISTICAL ANALYSIS
ascorbyl-palmitate. Interestingly, some of them were supported B RNA-seq data processing
by both models of longevity, suggesting that they may be simi- B Ortholog quality of mammalian species samples
larly effective in short- and long-lived mammalian species. In B Life history data of the species
agreement with the prediction, one of the top candidates re- B Mammalian RNA-seq data analysis
vealed by signature-based screening, KU0063794, significantly B Aggregation of aging data for meta-analysis
extended lifespan and healthspan of old C57BL/6 mice. Remark- B Aging gene expression signatures
ably, its short-term effect on the transcriptomic profile was not B Aggregated biomarkers of aging and longevity
negatively associated with aging, suggesting that it does not B Expression of signatures across cell types
seem to be a necessary condition to achieve a longer lifespan. B Comparison of longevity signatures and traits
Future comprehensive studies may shed light on the differences B Yeast signatures of longevity and aging
in the mechanisms of longevity interventions operating via anti- B Functional enrichment analysis of signatures
aging or aging-independent routes. Thus, transcriptomic B Metabolomics analysis
signatures of longevity may be used to identify lifespan- and B Partial correlation network
healthspan-extending interventions based on their gene expres- B Effect of compounds on fibroblast survival
sion profiles, thereby facilitating the discovery of novel B Enrichment of trait biomarkers by gene sets
geroprotectors. B Prediction of longevity interventions
d ADDITIONAL RESOURCES
Limitations of the study
Some variation in biological age between examined species may ACKNOWLEDGMENTS
be present due to the unknown precise age and exact health sta-
We thank Iaroslava Pavlova for valuable suggestions and assistance with data
tus of young adult animals collected in the wild. Future expansion
analysis and visualization. This study was supported by NIA grants to V.N.G.,
of this dataset with samples of different ages for each species V.G., R.A.M., and N.J.S. A.T. and S.E.D. were members of Interdisciplinary
would be of high value for the identification of age-adjusted Scientific and Educational School of Moscow University ‘‘Molecular Technol-
longevity signatures and analysis of differences in molecular ogies of the Living Systems and Synthetic Biology.’’
mechanisms of aging across species. Besides, signatures of es-
tablished lifespan-extending interventions identified for mouse AUTHOR CONTRIBUTIONS
models may not entirely translate to other mammals. The collec-
tion of similar data from other species would shed light on the A.T. and V.N.G. conceived and designed this research; A.T., S.M., A.V.S., S.T.,
S.-G.L., and J.P.C. performed research and data analysis; P.B., A.S., N.J.S.,
universality and differences in mechanisms of CR, GH defi-
V.G., R.A.M., and V.N.G. contributed new reagents/sample/analytic tools;
ciency, and other interventions. Finally, the shared and distinct A.T., S.M., A.V.S., A.S., N.J.S., V.G., S.E.D., R.A.M., and V.N.G. interpreted
longevity signatures were analyzed based on transcriptomic the data; V.N.G. supervised the study; A.T. and V.N.G. wrote the manuscript
and metabolomic data; however, multi-omics approaches may with contributions from all other authors. All authors read the final version.
yield further insights into mechanisms of lifespan regulation.
DECLARATION OF INTERESTS
STAR+METHODS
A patent application (no. 17/625,425) is pending that includes some of the data
described in this article.
Detailed methods are provided in the online version of this paper
and include the following:
INCLUSION AND DIVERSITY
d KEY RESOURCES TABLE
We support inclusive, diverse, and equitable conduct of research.
d RESOURCE AVAILABILITY
B Lead contact
Received: December 3, 2021
B Materials availability Revised: December 29, 2022
B Data and code availability Accepted: May 2, 2023
d EXPERIMENTAL MODEL AND STUDY PARTICIPANT Published: June 1, 2023
DETAILS
B Mammalian species sample collection REFERENCES
B Cell lines and culture
1. Sacher, G.A. (1959). Relation of lifespan to brain weight and body weight in
B Animals and predicted compounds
mammals. In Ciba Foundation Symposium - The Lifespan of Animals (Collo-
B Animals and KU0063794 treatment quia on Ageing), G.E.W. Wolstenholme and M. O’Conner, eds. (John Wiley
d METHOD DETAILS Sons, Ltd), pp. 115–141. https://doi.org/10.1002/9780470715253.ch9.
B RNA-seq profiling of mammalian tissues 2. de Magalhães, J.P., Costa, J., and Toussaint, O. (2005). HAGR: the hu-
B Stress resistance assay man ageing genomic resources. Nucleic Acids Res. 33, D537–D543.
B Frailty index and gait speed https://doi.org/10.1093/nar/gki017.
Cell 186, 1–21, June 22, 2023 15

ll
Article
3. Buffenstein, R. (2008). Negligible senescence in the longest living rodent, 19. Flurkey, K., Papaconstantinou, J., Miller, R.A., and Harrison, D.E. (2001).
the naked mole-rat: insights from a successfully aging species. J. Comp. Lifespan extension and delayed immune and collagen aging in mutant
Physiol. B 178, 439–445. https://doi.org/10.1007/s00360-007-0237-5. mice with defects in growth hormone production. Proc. Natl. Acad. Sci.
4. Seim, I., Fang, X., Xiong, Z., Lobanov, A.V., Huang, Z., Ma, S., Feng, Y., USA 98, 6736–6741. https://doi.org/10.1073/pnas.111158898.
Turanov, A.A., Zhu, Y., Lenz, T.L., et al. (2013). Genome analysis reveals 20. Junnila, R.K., List, E.O., Berryman, D.E., Murrey, J.W., and Kopchick, J.J.
insights into physiology and longevity of the Brandt’s bat Myotis brandtii. (2013). The GH/IGF-1 axis in ageing and longevity. Nat. Rev. Endocrinol.
Nat. Commun. 4, 2212. https://doi.org/10.1038/ncomms3212. 9, 366–376. https://doi.org/10.1038/nrendo.2013.67.
5. Ma, S., and Gladyshev, V.N. (2017). Molecular signatures of longevity: in- 21. Sziráki, A., Tyshkovskiy, A., and Gladyshev, V.N. (2018). Global remodel-
sights from cross-species comparative studies. Semin. Cell Dev. Biol. ing of the mouse DNA methylome during aging and in response to calorie
70, 190–203. https://doi.org/10.1016/j.semcdb.2017.08.007. restriction. Aging Cell 17, e12738. https://doi.org/10.1111/acel.12738.
6. Zhao, Y., Tyshkovskiy, A., Muñoz-Espı́n, D., Tian, X., Serrano, M., de 22. Meer, M.V., Podolskiy, D.I., Tyshkovskiy, A., and Gladyshev, V.N. (2018).
Magalhaes, J.P., Nevo, E., Gladyshev, V.N., Seluanov, A., and Gorbu- A whole lifespan mouse multi-tissue DNA methylation clock. eLife 7,
nova, V. (2018). Naked mole rats can undergo developmental, onco- 1–16. https://doi.org/10.7554/eLife.40675.
gene-induced and DNA damage-induced cellular senescence. Proc. 23. Schaum, N., Lehallier, B., Hahn, O., Pálovics, R., Hosseinzadeh, S., Lee,
Natl. Acad. Sci. USA 115, 1801–1806. https://doi.org/10.1073/pnas. S.E., Sit, R., Lee, D.P., Losada, P.M., Zardeneta, M.E., et al. (2020).
1721160115. Ageing hallmarks exhibit organ-specific temporal signatures. Nature
7. Harper, J.M., Salmon, A.B., Leiser, S.F., Galecki, A.T., and Miller, R.A. 583, 596–602. https://doi.org/10.1038/s41586-020-2499-y.
(2007). Skin-derived fibroblasts from long-lived species are resistant to
24. Almanzar, N., Antony, J., Baghel, A.S., Bakerman, I., Bansal, I., Barres,
some, but not all, lethal stresses and to the mitochondrial inhibitor rote-
B.A., Beachy, P.A., Berdnik, D., Bilen, B., and Brownfield, D. (2020). A
none. Aging Cell 6, 1–13. https://doi.org/10.1111/j.1474-9726.2006.
single-cell transcriptomic atlas characterizes ageing tissues in the
00255.x.
mouse. Nature 583, 590–595. https://doi.org/10.1038/s41586-020-
8. Whittemore, K., Vera, E., Martı́nez-Nevado, E., Sanpera, C., and Blasco, 2496-1.
M.A. (2019). Telomere shortening rate predicts species life span. Proc.
25. Palmer, D., Fabris, F., Doherty, A., Freitas, A.A., and de Magalhães, J.P.
Natl. Acad. Sci. USA 116, 15122–15127. https://doi.org/10.1073/pnas.
(2021). Ageing transcriptome meta-analysis reveals similarities and dif-
1902452116.
ferences between key mammalian tissues. Aging (Albany NY) 13,
9. Tian, X., Seluanov, A., and Gorbunova, V. (2017). Molecular mechanisms 3313–3341. https://doi.org/10.18632/aging.202648.
determining lifespan in short- and long-lived species. Trends Endocrinol.
26. de Magalhães, J.P., Curado, J., and Church, G.M. (2009). Meta-analysis
Metab. 28, 722–734. https://doi.org/10.1016/j.tem.2017.07.004.
of age-related gene expression profiles identifies common signatures of
10. Cagan, A., Baez-Ortega, A., Brzozowska, N., Abascal, F., Coorens, aging. Bioinformatics 25, 875–881. https://doi.org/10.1093/bioinformat-
T.H.H., Sanders, M.A., Lawson, A.R.J., Harvey, L.M.R., Bhosle, S., ics/btp073.
Jones, D., et al. (2022). Somatic mutation rates scale with lifespan across
27. Ma, S., Yim, S.H., Lee, S.G., Kim, E.B., Lee, S.R., Chang, K.T., Buffen-
mammals. Nature 604, 517–524. https://doi.org/10.1038/s41586-022-
stein, R., Lewis, K.N., Park, T.J., Miller, R.A., et al. (2015). Organization
04618-z.
of the Mammalian Metabolome according to Organ Function, Lineage
11. Zhou, Y., Xu, V., Maheshwari, H.G., He, L., Reed, M., Lozykowski, M., Specialization, and Longevity. Cell Metab. 22, 332–343. https://doi.org/
Okada, S., Cataldo, L., Coschigamo, K., Wagner, T.E., et al. (1997). A 10.1016/j.cmet.2015.07.005.
mammalian model for Laron syndrome produced by targeted disruption
28. Ma, S., Lee, S.G., Kim, E.B., Park, T.J., Seluanov, A., Gorbunova, V., Buf-
of the mouse growth hormone receptor/binding protein gene (the Laron
fenstein, R., Seravalli, J., and Gladyshev, V.N. (2015). Organization of the
mouse). Proc. Natl. Acad. Sci. USA 94, 13215–13220. https://doi.org/
mammalian ionome according to organ origin, lineage specialization, and
10.1073/pnas.94.24.13215.
longevity. Cell Rep. 13, 1319–1326. https://doi.org/10.1016/j.celrep.
12. Harrison, D.E., Strong, R., Sharp, Z.D., Nelson, J.F., Astle, C.M., Flurkey, 2015.10.014.
K., Nadon, N.L., Wilkinson, J.E., Frenkel, K., Carter, C.S., et al. (2009).
29. Bozek, K., Khrameeva, E.E., Reznick, J., Omerbasic, D., Bennett, N.C.,
Rapamycin fed late in life extends lifespan in genetically heterogeneous
Lewin, G.R., Azpurua, J., Gorbunova, V., Seluanov, A., Regnard, P.,
mice. Nature 460, 392–395. https://doi.org/10.1038/nature08221.
et al. (2017). Lipidome determinants of maximal lifespan in mammals.
13. Weindruch, R., Walford, R.L., Fligiel, S., and Guthrie, D. (1986). The retar- Sci. Rep. 7, 5. https://doi.org/10.1038/s41598-017-00037-7.
dation of aging in mice by dietary restriction: longevity, cancer, immunity
30. Ma, S., Upneja, A., Galecki, A., Tsai, Y.M., Burant, C.F., Raskind, S.,
and lifetime energy intake. J. Nutr. 116, 641–654. https://doi.org/10.
Zhang, Q., Zhang, Z.D., Seluanov, A., Gorbunova, V., et al. (2016). Cell
1093/jn/116.4.641.
culture-based profiling across mammals reveals DNA repair and meta-
14. David, J., Van Herrewege, J., and Fouillet, P. (1971). Quantitative under- bolism as determinants of species longevity. eLife 5, 1–25. https://doi.
feeding of drosophila: effects on adult longevity and fecundity. Exp. Ger- org/10.7554/eLife.19130.
ontol. 6, 249–257. https://doi.org/10.1016/0531-5565(71)90037-4.
31. Fushan, A.A., Turanov, A.A., Lee, S.G., Kim, E.B., Lobanov, A.V., Yim,
15. Houthoofd, K., and Vanfleteren, J.R. (2006). The longevity effect of die- S.H., Buffenstein, R., Lee, S.R., Chang, K.T., Rhee, H., et al. (2015).
tary restriction in Caenorhabditis elegans. Exp. Gerontol. 41, 1026– Gene expression defines natural changes in mammalian lifespan. Aging
1031. https://doi.org/10.1016/j.exger.2006.05.007. Cell 14, 352–365. https://doi.org/10.1111/acel.12283.
16. Lin, S., Defossez, P., and Guarentet, L. (2000). Requirement of NAD and 32. Lu, J.Y., Simon, M., Zhao, Y., Ablaeva, J., Corson, N., Choi, Y., Yamada,
SIR2 for life-span extension by calorie restriction in Saccharomyces cer- K.Y.H., Schork, N.J., Hood, W.R., Hill, G.E., et al. (2022). Comparative
evisiae. Science 289, 2126–2128. https://doi.org/10.1126/science.289. transcriptomics reveals circadian and pluripotency networks as two pil-
5487.2126. lars of longevity regulation. Cell Metab. 34, 836–856.e5. https://doi.org/
17. Selman, C., Nussey, D.H., and Monaghan, P. (2013). Ageing: It’s a dog’s 10.1016/j.cmet.2022.04.011.
life. Curr. Biol. 23, R451–R453. https://doi.org/10.1016/j.cub.2013. 33. Tyshkovskiy, A., Bozaykut, P., Borodinova, A.A., Gerashchenko, M.V.,
04.005. Ables, G.P., Garratt, M., Khaitovich, P., Clish, C.B., Miller, R.A., and Gla-
18. Brown-Borg, H.M., Borg, K.E., Meliska, C.J., and Bartke, A. (1996). Dwarf dyshev, V.N. (2019). Identification and application of gene expression
mice and the ageing process. Nature 384, 33. https://doi.org/10.1038/ signatures associated with lifespan extension. Cell Metab. 30, 573–
384033a0. 593.e8. https://doi.org/10.1016/j.cmet.2019.06.018.
16 Cell 186, 1–21, June 22, 2023

ll
Article
34. Kim, E.B., Fang, X., Fushan, A.A., Huang, Z., Lobanov, A.V., Han, L., Mar- 50. Sinha, M., Jang, Y.C., Oh, J., Khong, D., Wu, E.Y., Manohar, R., Miller, C.,
ino, S.M., Sun, X., Turanov, A.A., Yang, P., et al. (2011). Genome Regalado, S.G., Loffredo, F.S., Pancoast, J.R., et al. (2014). Restoring
sequencing reveals insights into physiology and longevity of the naked systemic GDF11 levels reverses age-related dysfunction in mouse skel-
mole rat. Nature 479, 223–227. https://doi.org/10.1038/nature10533. etal muscle. Science 344, 649–652. https://doi.org/10.1126/science.
35. Seim, I., Ma, S., Zhou, X., Gerashchenko, M.V., Lee, S.G., Suydam, R., 1251152.
George, J.C., Bickham, J.W., and Gladyshev, V.N. (2014). The transcrip- 51. Moigneu, C., Abdellaoui, S., Ramos-Brossier, M., Pfaffenseller, B., Wol-
tome of the bowhead whale Balaena mysticetus reveals adaptations of lenhaupt-Aguiar, B., de Azevedo Cardoso, T., Chiche, A., Kuperwasser,
the longest-lived mammal. Aging (Albany NY) 6, 879–899. https://doi. N., Azevedo da Silva, R., Pedrotti Moreira, F., et al. (2023). Systemic
org/10.18632/aging.100699. GDF11 attenuates depression-like phenotype in aged mice via stimula-
36. Fang, X., Seim, I., Huang, Z., Gerashchenko, M.V., Xiong, Z., Turanov, tion of neuronal autophagy. Nat. Aging 3, 213–228. https://doi.org/10.
A.A., Zhu, Y., Lobanov, A.V., Fan, D., Yim, S.H., et al. (2014). Adaptations 1038/s43587-022-00352-3.
to a subterranean environment and longevity revealed by the analysis of 52. Mauer, J., Denson, J.L., and Brüning, J.C. (2015). Versatile functions for
mole rat genomes. Cell Rep. 8, 1354–1364. https://doi.org/10.1016/j.cel- IL-6 in metabolism and cancer. Trends Immunol. 36, 92–101. https://doi.
rep.2014.07.030. org/10.1016/j.it.2014.12.008.
37. Brawand, D., Soumillon, M., Necsulea, A., Julien, P., Csárdi, G., Harrigan, 53. Mori, M.A., Ludwig, R.G., Garcia-Martin, R., Brandão, B.B., and Kahn,
P., Weier, M., Liechti, A., Aximu-Petri, A., Kircher, M., et al. (2011). The C.R. (2019). Extracellular miRNAs: from biomarkers to mediators of phys-
evolution of gene expression levels in mammalian organs. Nature 478, iology and disease. Cell Metab. 30, 656–673. https://doi.org/10.1016/j.
343–348. https://doi.org/10.1038/nature10532. cmet.2019.07.011.
38. Merkin, J., Russell, C., Chen, P., and Burge, C.B. (2012). Evolutionary dy- 54. Sobczak, A.I.S., Blindauer, C.A., and Stewart, A.J. (2019). Changes in
namics of gene and isoform regulation in mammalian tissues. Science plasma free fatty acids associated with type-2 diabetes. Nutrients 11.
338, 1593–1599. https://doi.org/10.1126/science.1228186. https://doi.org/10.3390/nu11092022.
39. Yim, H.S., Cho, Y.S., Guang, X., Kang, S.G., Jeong, J.Y., Cha, S.S., Oh, 55. Yoshida, M., Satoh, A., Lin, J.B., Mills, K.F., Sasaki, Y., Rensing, N.,
H.M., Lee, J.H., Yang, E.C., Kwon, K.K., et al. (2014). Minke whale Wong, M., Apte, R.S., and Imai, S.I. (2019). Extracellular vesicle-con-
genome and aquatic adaptation in cetaceans. Nat. Genet. 46, 88–92. tained eNAMPT delays aging and extends lifespan in mice. Cell Metab.
https://doi.org/10.1038/ng.2835. 30, 329–342.e5. https://doi.org/10.1016/j.cmet.2019.05.015.
40. Fan, Y., Huang, Z.Y., Cao, C.C., Chen, C.S., Chen, Y.X., Fan, D.D., He, J., 56. Hall, B.M., Gleiberman, A.S., Strom, E., Krasnov, P.A., Frescas, D., Vuj-
Hou, H.L., Hu, L., Hu, X.T., et al. (2013). Genome of the Chinese tree cic, S., Leontieva, O.V., Antoch, M.P., Kogan, V., Koman, I.E., et al.
shrew. Nat. Commun. 4, 1426. https://doi.org/10.1038/ncomms2416. (2020). Immune checkpoint protein VSIG4 as a biomarker of aging in mu-
rine adipose tissue. Aging Cell 19, e13219. https://doi.org/10.1111/
41. Qiu, Q., Zhang, G., Ma, T., Qian, W., Wang, J., Ye, Z., Cao, C., Hu, Q.,
acel.13219.
Kim, J., Larkin, D.M., et al. (2012). The yak genome and adaptation to
life at high altitude. Nat. Genet. 44, 946–949. https://doi.org/10.1038/ 57. Byun, J.M., Jeong, D.H., Choi, I.H., Lee, D.S., Kang, M.S., Jung, K.O.,
ng.2343. Jeon, Y.K., Kim, Y.N., Jung, E.J., Lee, K.B., et al. (2017). The significance
of VSIG4 expression in ovarian cancer. Int. J. Gynecol. Cancer 27,
42. Kaessmann, H. (2010). Origins, evolution, and phenotypic impact of new 872–878. https://doi.org/10.1097/IGC.0000000000000979.
genes. Genome Res. 20, 1313–1326. https://doi.org/10.1101/gr.
58. Bianchi-Frias, D., Damodarasamy, M., Hernandez, S.A., Gil Da Costa,
101386.109.
R.M., Vakar-Lopez, F., Coleman, I.M., Reed, M.J., and Nelson, P.S.
43. Khaitovich, P., Enard, W., Lachmann, M., and Pääbo, S. (2006). Evolution (2019). The aged microenvironment influences the tumorigenic potential
of primate gene expression. Nat. Rev. Genet. 7, 693–702. https://doi.org/ of malignant prostate epithelial cells. Mol. Cancer Res. 17, 321–331.
10.1038/nrg1940. https://doi.org/10.1158/1541-7786.MCR-18-0522.
44. Wang, Z.Y., Leushkin, E., Liechti, A., Ovchinnikova, S., Mößinger, K., 59. Xu, T., Jiang, Y., Yan, Y., Wang, H., Lu, C., Xu, H., Li, W., Fu, D., Lu, Y.,
Brüning, T., Rummel, C., Grützner, F., Cardoso-Moreira, M., Janich, P., and Chen, J. (2015). VSIG4 is highly expressed and correlated with
et al. (2020). Transcriptome and translatome co-evolution in mammals. poor prognosis of high-grade glioma patients. Am. J. Transl. Res. 7,
Nature 588, 642–647. https://doi.org/10.1038/s41586-020-2899-z. 1172–1180.
45. Rops, A.L., Götte, M., Baselmans, M.H., Van Den Hoven, M.J., Steenber- 60. Lee, M.Y., Kim, W.J., Kang, Y.J., Jung, Y.M., Kang, Y.M., Suk, K., Park,
gen, E.J., Lensen, J.F., Wijnhoven, T.J., Cevikbas, F., Van Den Heuvel, J.E., Choi, E.M., Choi, B.K., Kwon, B.S., et al. (2006). Z39Ig is expressed
L.P., Van Kuppevelt, T.H., et al. (2007). Syndecan-1 deficiency aggra- on macrophages and may mediate inflammatory reactions in arthritis and
vates anti-glomerular basement membrane nephritis. Kidney Int. 72, atherosclerosis. J. Leukoc. Biol. 80, 922–928. https://doi.org/10.1189/
1204–1215. https://doi.org/10.1038/sj.ki.5002514. jlb.0306160.
46. Vanhoutte, D., Schellings, M.W.M., Götte, M., Swinnen, M., Herias, V., 61. Taylor, G.A., Rodriguiz, R.M., Greene, R.I., Daniell, X., Henry, S.C.,
Wild, M.K., Vestweber, D., Chorianopoulos, E., Cortés, V., Rigotti, A., Crooks, K.R., Kotloski, R., Tessarollo, L., Phillips, L.E., and Wetsel,
et al. (2007). Increased expression of syndecan-1 protects against car- W.C. (2008). Behavioral characterization of P311 knockout mice. Genes
diac dilatation and dysfunction after myocardial infarction. Circulation Brain Behav. 7, 786–795. https://doi.org/10.1111/j.1601-183X.2008.
115, 475–482. https://doi.org/10.1161/CIRCULATIONAHA.106.644609. 00420.x.
47. Luo, Y., He, Z., Liu, W., Zhou, F., Liu, T., and Wang, G. (2022). DTL Is a 62. Badri, K.R., Yue, M., Carretero, O.A., Aramgam, S.L., Cao, J., Sharkady,
Prognostic biomarker and Promotes Bladder Cancer Progression S., Kim, G.H., Taylor, G.A., Byron, K.L., and Schuger, V. (2013). Blood
through Regulating the AKT/mTOR axis. Oxid. Med. Cell. Longev. pressure homeostasis is maintained by a p311-TGF-b axis. J. Clin.
2022, 3369858. https://doi.org/10.1155/2022/3369858. Invest. 123, 4502–4512. https://doi.org/10.1172/JCI69884.
48. Wilson, D.J. (2019). The harmonic mean p-value for combining depen- 63. Nunez, S.K., Young, C., Adebayo, O., Muppuru, K.M., and Badri, K.R.
dent tests. Proc. Natl. Acad. Sci. USA 116, 1195–1200. https://doi.org/ (2019). P311, a novel intrinsically disordered protein, regulates adipocyte
10.1073/pnas.1814092116. development. Biochem. Biophys. Res. Commun. 515, 234–240. https://
49. Chu, T., Wang, Z., Pe’er, D., and Danko, C.G. (2022). Cell type and gene doi.org/10.1016/j.bbrc.2019.05.105.
expression deconvolution with BayesPrism enables Bayesian integrative 64. López-Otı́n, C., Blasco, M.A., Partridge, L., Serrano, M., and Kroemer, G.
analysis across bulk and single-cell RNA sequencing in oncology. Nat. (2013). The hallmarks of aging. Cell 153, 1194–1217. https://doi.org/10.
Cancer 3, 505–517. https://doi.org/10.1038/s43018-022-00356-3. 1016/j.cell.2013.05.039.
Cell 186, 1–21, June 22, 2023 17

ll
Article
65. López-Otı́n, C., Blasco, M.A., Partridge, L., Serrano, M., and Kroemer, G. 80. Seropian, I.M., González, G.E., Maller, S.M., Berrocal, D.H., Abbate, A.,
(2023). Hallmarks of aging: An expanding universe. Cell 186, 243–278. and Rabinovich, G.A. (2018). Galectin-1 as an emerging mediator of car-
https://doi.org/10.1016/j.cell.2022.11.001. diovascular inflammation: Mechanisms and therapeutic opportunities.
66. Petkovich, D.A., Podolskiy, D.I., Lobanov, A.V., Lee, S.G., Miller, R.A., Mediators Inflamm. 2018, 8696543. https://doi.org/10.1155/2018/
and Gladyshev, V.N. (2017). Using DNA methylation profiling to evaluate 8696543.
biological age and longevity interventions. Cell Metab. 25, 954–960.e6. 81. Seropian, I.M., Cerliani, J.P., Toldo, S., Van Tassell, B.W., Ilarregui, J.M.,
https://doi.org/10.1016/j.cmet.2017.03.016. González, G.E., Matoso, M., Salloum, F.N., Melchior, R., Gelpi, R.J., et al.
67. Wang, T., Tsui, B., Kreisberg, J.F., Robertson, N.A., Gross, A.M., Yu, (2013). Galectin-1 controls cardiac inflammation and ventricular remod-
M.K., Carter, H., Brown-Borg, H.M., Adams, P.D., and Ideker, T. eling during acute myocardial infarction. Am. J. Pathol. 182, 29–40.
(2017). Epigenetic aging signatures in mice livers are slowed by https://doi.org/10.1016/j.ajpath.2012.09.022.
dwarfism, calorie restriction and rapamycin treatment. Genome Biol. 82. Marques, T.M., van Rumund, A., Bruinsma, I.B., Wessels, H.J.C.T.,
18, 57. https://doi.org/10.1186/s13059-017-1186-2. Gloerich, J., Esselink, R.A.J., Bloem, B.R., Kuiperij, H.B., and Verbeek,
M.M. (2019). Cerebrospinal fluid galectin-1 levels discriminate patients
68. Cole, J.J., Robertson, N.A., Rather, M.I., Thomson, J.P., McBryan, T.,
with parkinsonism from controls. Mol. Neurobiol. 56, 5067–5074.
Sproul, D., Wang, T., Brock, C., Clark, W., Ideker, T., et al. (2017). Diverse
https://doi.org/10.1007/s12035-018-1426-9.
interventions that extend mouse lifespan suppress shared age-associ-
ated epigenetic changes at critical gene regulatory regions. Genome 83. Yoon, S., Baik, B., Park, T., and Nam, D. (2021). Powerful p-value com-
Biol. 18, 58. https://doi.org/10.1186/s13059-017-1185-3. bination methods to detect incomplete association. Sci. Rep. 11, 6980.
https://doi.org/10.1038/s41598-021-86465-y.
69. Ying, K., Liu, H., Tarkhov, A.E., Lu, A.T., Horvath, S., Kutalik, Z., Shen, X.,
and Gladyshev, V.N. (2022). Causal epigenetic age uncouples damage 84. Ngo, T.C., and Assimos, D.G. (2007). Uric acid nephrolithiasis: recent
and adaptation. Preprint at bioRxiv. https://doi.org/10.1101/2022.10. progress and future directions. Rev. Urol. 9, 17–27.
07.511382. 85. Oda, M., Satta, Y., Takenaka, O., and Takahata, N. (2002). Loss of urate ox-
70. Kaya, A., Phua, C.Z.J., Lee, M., Wang, L., Tyshkovskiy, A., Ma, S., Barre, idase activity in hominoids and its evolutionary implications. Mol. Biol. Evol.
B., Liu, W., Harrison, B.R., Zhao, X., et al. (2021). Evolution of natural 19, 640–653. https://doi.org/10.1093/oxfordjournals.molbev.a004123.
lifespan variation and molecular strategies of extended lifespan in yeast. 86. Wu, X.W., Lee, C.C., Muzny, D.M., and Caskey, C.T. (1989). Urate oxidase:
eLife 10, e64860. https://doi.org/10.7554/eLife.64860. Primary structure and evolutionary implications. Proc. Natl. Acad. Sci. USA
71. Kemmeren, P., Sameith, K., Van De Pasch, L.A.L., Benschop, J.J., Len- 86, 9412–9416. https://doi.org/10.1073/pnas.86.23.9412.
stra, T.L., Margaritis, T., O’Duibhir, E., Apweiler, E., Van Wageningen, S., 87. Wu, X.W., Muzny, D.M., Lee, C.C., and Caskey, C.T. (1992). Two inde-
Ko, C.W., et al. (2014). Large-scale genetic perturbations reveal regulato- pendent mutational events in the loss of urate oxidase during hominoid
ry networks and an abundance of gene-specific repressors. Cell 157, evolution. J. Mol. Evol. 34, 78–84. https://doi.org/10.1007/BF00163854.
740–752. https://doi.org/10.1016/j.cell.2014.02.054. 88. Gomes, A.P., Price, N.L., Ling, A.J.Y., Moslehi, J.J., Montgomery, M.K.,
72. McCormick, M.A., Delaney, J.R., Tsuchiya, M., Kaeberlein, M., and Rajman, L., White, J.P., Teodoro, J.S., Wrann, C.D., Hubbard, B.P., et al.
Kennedy, B.K. (2015). A comprehensive analysis of replicative life- (2013). Declining NAD+ induces a pseudohypoxic state disrupting nu-
span in 4,698 single-gene deletion strains uncovers conserved clear-mitochondrial communication during aging. Cell 155, 1624–1638.
mechanisms of aging. Cell Metab. 22, 895–906. https://doi.org/10. https://doi.org/10.1016/j.cell.2013.11.037.
1177/0333102415576222.Is. 89. Rajman, L., Chwalek, K., and Sinclair, D.A. (2018). Therapeutic potential
73. Janssens, G.E., Meinema, A.C., González, J., Wolters, J.C., Schmidt, A., of NAD-boosting molecules: the in vivo evidence. Cell Metab. 27,
Guryev, V., Bischoff, R., Wit, E.C., Veenhoff, L.M., and Heinemann, M. 529–547. https://doi.org/10.1016/j.cmet.2018.02.011.
(2015). Protein biogenesis machinery is a driver of replicative aging in 90. Yoshino, J., Baur, J.A., and Imai, S.I. (2018). NAD+ intermediates: The
yeast. eLife 4, e08527. https://doi.org/10.7554/eLife.08527. biology and therapeutic potential of NMN and NR. Cell Metab. 27,
74. Fontana, L., Partridge, L., and Longo, V.D. (2010). Extending healthy life 513–528. https://doi.org/10.1016/j.cmet.2017.11.002.
span—from yeast to humans. Science 328, 321–326. https://doi.org/10. 91. Orlandi, I., Stamerra, G., and Vai, M. (2018). Altered expression of mito-
1126/science.1172539. chondrial NAD+ carriers influences yeast chronological lifespan by
75. Coschigano, K.T., Clemmons, D., Bellush, L.L., and Kopchick, J.J. modulating cytosolic and mitochondrial metabolism. Front. Genet. 9,
(2000). Assessment of growth parameters and lifespan of GHR/BP 676. https://doi.org/10.3389/fgene.2018.00676.
gene-disrupted mice. Endocrinology 141, 2608–2613. https://doi.org/ 92. Kabil, O., Vitvitsky, V., Xie, P., and Banerjee, R. (2011). The quantitative
10.1210/endo.141.7.7586. significance of the transsulfuration enzymes for H2S production in murine
76. Mitchell, S.J., Madrigal-Matute, J., Scheibye-Knudsen, M., Fang, E., tissues. Antioxid. Redox Signal. 15, 363–372. https://doi.org/10.1089/
Aon, M., González-Reyes, J.A., Cortassa, S., Kaushik, S., Gonzalez- ars.2010.3781.
Freire, M., Patel, B., et al. (2016). Effects of sex, strain, and energy intake 93. Miller, D.L., and Roth, M.B. (2007). Hydrogen sulfide increases thermo-
on hallmarks of aging in mice. Cell Metab. 23, 1093–1112. https://doi. tolerance and lifespan in Caenorhabditis elegans. Proc. Natl. Acad.
org/10.1016/j.cmet.2016.05.027. Sci. USA 104, 20618–20622. https://doi.org/10.1073/pnas.0710191104.
77. Landin-Wilhelmsen, K., Lundberg, P.A., Lappas, G., and Wilhelmsen, L. 94. Tilstra, J.S., Clauson, C.L., Niedernhofer, L.J., and Robbins, P.D. (2011).
(2004). Insulin-like growth factor I levels in healthy adults. Horm. Res. NF-kB in aging and disease. Aging Dis. 2, 449–465.
62, 8–16. https://doi.org/10.1159/000080753. 95. Leonardi, G.C., Accardi, G., Monastero, R., Nicoletti, F., and Libra, M.
78. Thijssen, V.L.J.L., Postel, R., Brandwijk, R.J.M.G.E., Dings, R.P.M., Nes- (2018). Ageing: From inflammation to cancer. Immun. Ageing 15, 1.
melova, I., Satijn, S., Verhofstad, N., Nakabeppu, Y., Baum, L.G., Bak- https://doi.org/10.1186/s12979-017-0112-5.
kers, J., et al. (2006). Galectin-1 is essential in tumor angiogenesis and 96. Lai, K.S.P., Liu, C.S., Rau, A., Lanctôt, K.L., Köhler, C.A., Pakosh, M.,
is a target for antiangiogenesis therapy. Proc. Natl. Acad. Sci. USA Carvalho, A.F., and Herrmann, N. (2017). Peripheral inflammatory
103, 15975–15980. https://doi.org/10.1073/pnas.0603883103. markers in Alzheimer’s disease: A systematic review and meta-analysis
79. Sundblad, V., Morosi, L.G., Geffner, J.R., and Rabinovich, G.A. (2017). of 175 studies. J. Neurol. Neurosurg. Psychiatry 88, 876–882. https://
Galectin-1: A jack-of-all-trades in the resolution of acute and chronic doi.org/10.1136/jnnp-2017-316201.
inflammation. J. Immunol. 199, 3721–3730. https://doi.org/10.4049/jim- 97. Amdur, R.L., Feldman, H.I., Gupta, J., Yang, W., Kanetsky, P., Shlipak,
munol.1701172. M., Rahman, M., Lash, J.P., Townsend, R.R., Ojo, A., et al. (2016).
18 Cell 186, 1–21, June 22, 2023

ll
Article
Inflammation and progression of CKD: The CRIC study. Clin. J. Am. Soc. 112. Yeung, F., Ramı́rez, C.M., Mateos-Gomez, P.A., Pinzaru, A., Ceccarini,
Nephrol. 11, 1546–1556. https://doi.org/10.2215/CJN.13121215. G., Kabir, S., Fernández-Hernando, C., and Sfeir, A. (2013). Nontelomeric
98. Anisimova, A.S., Meerson, M.B., Gerashchenko, M.V., Kulakovskiy, I.V., role for Rap1 in regulating metabolism and protecting against obesity.
Dmitriev, S.E., and Gladyshev, V.N. (2020). Multifaceted deregulation of Cell Rep. 3, 1847–1856. https://doi.org/10.1016/j.celrep.2013.05.032.
gene expression and protein synthesis with age. Proc. Natl. Acad. Sci. 113. Bongers, K.S., Fox, D.K., Kunkel, S.D., Stebounova, L.V., Murry, D.J.,
USA 117, 15581–15590. https://doi.org/10.1073/pnas.2001788117. Pufall, M.A., Ebert, S.M., Dyle, M.C., Bullard, S.A., Dierdorff, J.M., et al.
99. Tollis, M., Boddy, A.M., and Maley, C.C. (2017). Peto’s Paradox: How has (2015). Spermine oxidase maintains basal skeletal muscle gene expres-
evolution solved the problem of cancer prevention? BMC Biol. 15, 60. sion and fiber size and is strongly repressed by conditions that cause
https://doi.org/10.1186/s12915-017-0401-7. skeletal muscle atrophy. Am. J. Physiol. Endocrinol. Metab. 308, E144–
E158. https://doi.org/10.1152/ajpendo.00472.2014.
100. Lombard, D.B., Kohler, W.J., Guo, A.H., Gendron, C., Han, M., Ding, W.,
114. Hand, N.J., Master, Z.R., Lay, J. Le, and Friedman, J.R. (2009). Hepatic
Lyu, Y., Ching, T.T., Wang, F.Y., Chakraborty, T.S., et al. (2020). High-
function is preserved in the absence of mature microRNAs. Hepatology
throughput small molecule screening reveals Nrf2-dependent and -inde-
49, 618–626. https://doi.org/10.1002/hep.22656.
pendent pathways of cellular stress resistance. Sci. Adv. 6, 1–14. https://
doi.org/10.1126/sciadv.aaz7628. 115. Sunami, Y., Leithäuser, F., Gul, S., Fiedler, K., Güldiken, N., Espenlaub,
S., Holzmann, K.H., Hipp, N., Sindrilaru, A., Luedde, T., et al. (2012). He-
101. Ren, G., Sun, A., Deng, C., Zhang, J., Wu, X., Wei, X., Mani, S., Dou, W.,
patic activation of IKK/NFkB signaling induces liver fibrosis via macro-
and Wang, Z. (2015). The anti-inflammatory effect and potential mecha-
phage-mediated chronic inflammation. Hepatology 56, 1117–1128.
nism of cardamonin in DSS-induced colitis. Am. J. Physiol. Gastrointest.
https://doi.org/10.1002/hep.25711.
Liver Physiol. 309, G517–G527. https://doi.org/10.1152/ajpgi.00133.2015.
116. Jain, I.H., Zazzeron, L., Goli, R., Alexa, K., Schatzman-Bone, S., Dhillon,
102. Lewis, K.N., Wason, E., Edrey, Y.H., Kristan, D.M., Nevo, E., and Buffen-
H., Goldberger, O., Peng, J., Shalem, O., Sanjana, N.E., et al. (2016).
stein, R. (2015). Regulation of Nrf2 signaling and longevity in naturally
Hypoxia as a therapy for mitochondrial disease. Science 352, 54–61.
long-lived rodents. Proc. Natl. Acad. Sci. USA 112, 3722–3727. https://
https://doi.org/10.1126/science.aad9642.
doi.org/10.1073/pnas.1417566112.
117. Balaban, R.S., Nemoto, S., and Finkel, T. (2005). Mitochondria, oxidants,
103. Pitayu, L., Baruffini, E., Rodier, C., Rötig, A., Lodi, T., and Delahodde, A. and aging. Cell 120, 483–495. https://doi.org/10.1016/j.cell.2005.02.001.
(2016). Combined use of Saccharomyces cerevisiae, Caenorhabditis ele-
118. Sykiotis, G.P., and Bohmann, D. (2008). Keap1/Nrf2 signaling regulates
gans and patient fibroblasts leads to the identification of clofilium tosylate
oxidative stress tolerance and lifespan in Drosophila. Dev. Cell 14,
as a potential therapeutic chemical against POLG-related diseases.
76–85. https://doi.org/10.1016/j.devcel.2007.12.002.
Hum. Mol. Genet. 25, 715–727. https://doi.org/10.1093/hmg/ddv509.
119. Tullet, J.M.A., Hertweck, M., An, J.H., Baker, J., Hwang, J.Y., Liu, S.,
104. Facchinello, N., Laquatra, C., Locatello, L., Beffagna, G., Brañas Casas,
Oliveira, R.P., Baumeister, R., and Blackwell, T.K. (2008). Direct inhibi-
R., Fornetto, C., Dinarello, A., Martorano, L., Vettori, A., Risato, G., et al.
tion of the longevity-promoting factor SKN-1 by insulin-like signaling in
(2021). Efficient clofilium tosylate-mediated rescue of POLG-related dis-
C. elegans. Cell 132, 1025–1038. https://doi.org/10.1016/j.cell.2008.
ease phenotypes in zebrafish. Cell Death Dis. 12, 100. https://doi.org/10.
01.030.
1038/s41419-020-03359-z.
120. Lu, X.F., Zhou, Y.J., Zhang, L., Ji, H.J., Li, L., Shi, Y.J., and Bu, H. (2015).
105. Xu, H., Li, X., Ding, W., Zeng, X., Kong, H., Wang, H., and Xie, W. (2015).
Loss of Dicer1 impairs hepatocyte survival and leads to chronic inflam-
Deguelin induces the apoptosis of lung cancer cells through regulating a
mation and progenitor cell activation. World J. Gastroenterol. 21,
ROS driven Akt pathway. Cancer Cell Int. 15, 25. https://doi.org/10.1186/
6591–6603. https://doi.org/10.3748/wjg.v21.i21.6591.
s12935-015-0166-4.
121. O’Rourke, J.R., Georges, S.A., Seay, H.R., Tapscott, S.J., McManus,
106. Preston, S., Korhonen, P.K., Mouchiroud, L., Cornaglia, M., Mcgee, S.L., M.T., Goldhamer, D.J., Swanson, M.S., and Harfe, B.D. (2007). Essential
Young, N.D., Davis, R.A., Crawford, S., Nowell, C., Ansell, B.R.E., et al. role for Dicer during skeletal muscle development. Dev. Biol. 311,
(2017). Deguelin exerts potent nematocidal activity via the mitochondrial 359–368. https://doi.org/10.1016/j.ydbio.2007.08.032.
respiratory chain. FASEB J. 31, 4515–4532. https://doi.org/10.1096/fj.
122. McConnell, B.B., Starborg, M., Brookes, S., and Peters, G. (1998). Inhib-
201700288R.
itors of cyclin-dependent kinases induce features of replicative senes-
107. Lamb, J., Crawford, E.D., Peck, D., Modell, J.W., Blat, I.C., Wrobel, M.J., cence in early passage human diploid fibroblasts. Curr. Biol. 8,
Lerner, J., Brunet, J.P., Subramanian, A., Ross, K.N., et al. (2006). The 351–354. https://doi.org/10.1016/s0960-9822(98)70137-x.
connectivity map: using gene-expression signatures to connect small
123. Fang, L., Igarashi, M., Leung, J., Sugrue, M.M., Lee, S.W., and Aaronson,
molecules, genes, and disease. Science 313, 1929–1935. https://doi.
S.A. (1999). p21Waf1/Cip1/Sdi1 induces permanent growth arrest with
org/10.1126/science.1132939.
markers of replicative senescence in human tumor cells lacking func-
108. Subramanian, A., Narayan, R., Corsello, S.M., Peck, D.D., Natoli, T.E., tional p53. Oncogene 18, 2789–2797. https://doi.org/10.1038/sj.onc.
Lu, X., Gould, J., Davis, J.F., Tubelli, A.A., Asiedu, J.K., et al. (2017). A 1202615.
Next Generation Connectivity Map: L1000 platform and the first
124. Folkes, A.J., Ahmadi, K., Alderton, W.K., Alix, S., Baker, S.J., Box, G.,
1,000,000 profiles. Cell 171, 1437–1452.e17. https://doi.org/10.1016/j.
Chuckowree, I.S., Clarke, P.A., Depledge, P., Eccles, S.A., et al. (2008).
cell.2017.10.049.
The identification of 2-(1H-Indazol-4-yl)-6-(4-methanesulfonyl-piperazin-
109. Osburn, W.O., Yates, M.S., Dolan, P.D., Chen, S., Liby, K.T., Sporn, M.B., 1-ylmethyl)-4-morpholin-4-yl-thieno[3,2-d]pyrimidine (GDC0941) as a
Taguchi, K., Yamamoto, M., and Kensler, T.W. (2008). Genetic or phar- potent, selective, orally bioavailable inhibitor of class I PI3 kinase for the
macologic amplification of Nrf2 signaling inhibits acute inflammatory liver treatment of cancer. J. Med. Chem. 51, 5522–5532. https://doi.org/10.
injury in mice. Toxicol. Sci. 104, 218–227. https://doi.org/10.1093/toxsci/ 1021/jm800295d.
kfn079. 125. Graff, J.R., McNulty, A.M., Hanna, K.R., Konicek, B.W., Lynch, R.L.,
110. Baze, M.M., Schlauch, K., and Hayes, J.P. (2010). Gene expression of the Bailey, S.N., Banks, C., Capen, A., Goode, R., Lewis, J.E., et al. (2005).
liver in response to chronic hypoxia. Physiol. Genomics 41, 275–288. The protein kinase Cb-selective inhibitor, enzastaurin. Cancer Res. 65,
https://doi.org/10.1152/physiolgenomics.00075.2009. 7462–7469. https://doi.org/10.1158/0008-5472.CAN-05-0071.
111. Zhou, B., Yang, L., Li, S., Huang, J., Chen, H., Hou, L., Wang, J., Green, 126. Li, P., Wu, Y., Li, M., Qiu, X., Bai, X., and Zhao, X. (2015). AS-703026 in-
C.D., Yan, Z., Huang, X., et al. (2012). Midlife gene expressions identify hibits LPS-induced TNFa production through MEK/ERK dependent and
modulators of aging through dietary interventions. Proc. Natl. Acad. ndependent mechanisms. PLoS One 10, e0137107. https://doi.org/10.
Sci. USA 109, E1201–E1209. https://doi.org/10.1073/pnas.1119304109. 1371/journal.pone.0137107.
Cell 186, 1–21, June 22, 2023 19

ll
Article
127. Park, S.J., Hong, S.W., Moon, J.H., Jin, D.H., Kim, J.S., Lee, C.K., Kim, 141. Mehdipour, M., Mehdipour, T., Skinner, C.M., Wong, N., Liu, C., Chen,
K.P., Hong, Y.S., Choi, E.K., Lee, J.S., et al. (2013). The MEK1/2 inhibitor C.C., Jeon, O.H., Zuo, Y., Conboy, M.J., and Conboy, I.M. (2021). Plasma
AS703026 circumvents resistance to the BRAF inhibitor PLX4032 in hu- dilution improves cognition and attenuates neuroinflammation in old mice.
man malignant melanoma cells. Am. J. Med. Sci. 346, 494–498. https:// GeroScience 43, 1–18. https://doi.org/10.1007/s11357-020-00297-8.
doi.org/10.1097/MAJ.0b013e318298a185. 142. Gerashchenko, M.V., Peterfi, Z., Yim, S.H., and Gladyshev, V.N. (2021).
128. Garcı́a-Martı́nez, J.M., Moran, J., Clarke, R.G., Gray, A., Cosulich, S.C., Translation elongation rate varies among organs and decreases with
Chresta, C.M., and Alessi, D.R. (2009). Ku-0063794 is a specific inhibitor age. Nucleic Acids Res. 49, e9. https://doi.org/10.1093/nar/gkaa1103.
of the mammalian target of rapamycin (mTOR). Biochem. J. 421, 29–42. 143. Tavernarakis, N. (2008). Ageing and the regulation of protein synthesis: a
https://doi.org/10.1042/BJ20090489. balancing act? Trends Cell Biol. 18, 228–235. https://doi.org/10.1016/j.
129. Chresta, C.M., Davies, B.R., Hickson, I., Harding, T., Cosulich, S., Crit- tcb.2008.02.004.
chlow, S.E., Vincent, J.P., Ellston, R., Jones, D., Sini, P., et al. (2010). 144. Tacutu, R., Craig, T., Budovsky, A., Wuttke, D., Lehmann, G., Taranukha,
AZD8055 is a potent, selective, and orally bioavailable ATP-competitive D., Costa, J., Fraifeld, V.E., and De Magalhães, J.P. (2013). Human
mammalian target of rapamycin kinase inhibitor with in vitro and in vivo Ageing Genomic Resources: Integrated databases and tools for the
antitumor activity. Cancer Res. 70, 288–298. https://doi.org/10.1158/ biology and genetics of ageing. Nucleic Acids Res. 41, D1027–D1033.
0008-5472.CAN-09-1751. https://doi.org/10.1093/nar/gks1155.
130. Cort, W.M. (1974). Antioxidant activity of tocopherols, ascorbyl palmi- 145. Lee, S.G., Mikhalchenko, A.E., Yim, S.H., Lobanov, A.V., Park, J.K., Choi,
tate, and ascorbic acid and their mode of action. J. Am. Oil Chem. K.H., Bronson, R.T., Lee, C.K., Park, T.J., and Gladyshev, V.N. (2017).
Soc. 51, 321–325. https://doi.org/10.1007/BF02633006. Naked mole rat induced pluripotent stem cells and their contribution to
131. Arriola Apelo, S.I., Neuman, J.C., Baar, E.L., Syed, F.A., Cummings, N.E., interspecific chimera. Stem Cell Rep. 9, 1706–1720. https://doi.org/10.
Brar, H.K., Pumper, C.P., Kimple, M.E., and Lamming, D.W. (2016). Alter- 1016/j.stemcr.2017.09.013.
native rapamycin treatment regimens mitigate the impact of rapamycin 146. Bolger, A.M., Lohse, M., and Usadel, B. (2014). Trimmomatic: A flexible
on glucose homeostasis and the immune system. Aging Cell 15, trimmer for Illumina sequence data. Bioinformatics 30, 2114–2120.
28–38. https://doi.org/10.1111/acel.12405. https://doi.org/10.1093/bioinformatics/btu170.
132. Shindyapina, A.V., Castro, J.P., Barbieri, A., Strelkova, O.S., Paulo, J.A., 147. Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L.,
Kerepesi, C., Petrashen, A.P., Marriotti, M., Meer, M., Hu, Y., et al. (2021). Gillette, M.A., Paulovich, A., Pomeroy, S.L., Golub, T.R., Lander, E.S.,
Integrative analysis reveals aged clonal B cells, microenvironment and et al. (2005). Gene set enrichment analysis: a knowledge-based approach
c-Myc activation in the origin of age-related lymphoma. Preprint at bio- for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci.
Rxiv. https://doi.org/10.1101/2021.02.23.432500. USA 102, 15545–15550. https://doi.org/10.1073/pnas.0506580102.
133. Chondrogianni, N., Georgila, K., Kourtis, N., Tavernarakis, N., and 148. Kolberg, L., Raudvere, U., Kuzmin, I., Vilo, J., and Peterson, H. (2020).
Gonos, E.S. (2015). 20S proteasome activation promotes life span exten- gprofiler2 – an R package for gene list functional enrichment analysis
sion and resistance to proteotoxicity in Caenorhabditis elegans. FASEB and namespace conversion toolset g:Profiler. F1000Res 9, 709. https://
J. 29, 611–622. https://doi.org/10.1096/fj.14-252189. doi.org/10.12688/f1000research.24956.2.
134. Ghazi, A., Henis-Korenblit, S., and Kenyon, C. (2007). Regulation of Cae- 149. Raudvere, U., Kolberg, L., Kuzmin, I., Arak, T., Adler, P., Peterson, H., and
norhabditis elegans lifespan by a proteasomal E3 ligase complex. Proc. Vilo, J. (2019). G:Profiler: a web server for functional enrichment analysis
Natl. Acad. Sci. USA 104, 5947–5952. https://doi.org/10.1073/pnas. and conversions of gene lists (2019 update). Nucleic Acids Res. 47,
0700638104. W191–W198. https://doi.org/10.1093/nar/gkz369.
135. Simonsen, A., Cumming, R.C., Brech, A., Isakson, P., Schubert, D.R., 150. Chang, W., Cheng, J., Allaire, J., Xie, Y., and McPherson, J. (2016). shiny:
and Finley, K.D. (2008). Promoting basal levels of autophagy in the ner- web application framework for R. R package version 0.14.2. https://
vous system enhances longevity and oxidant resistance in adult CRAN.R-project.org/package=shiny.
Drosophila. Autophagy 4, 176–184. https://doi.org/10.4161/auto.5269.
151. Robinson, M.D., and Oshlack, A. (2010). A scaling normalization method
136. Pyo, J.O., Yoo, S.M., Ahn, H.H., Nah, J., Hong, S.H., Kam, T.I., Jung, S., for differential expression analysis of RNA-seq data. Genome Biol. 11,
and Jung, Y.K. (2013). Overexpression of Atg5 in mice activates auto- R25. https://doi.org/10.1186/gb-2010-11-3-r25.
phagy and extends lifespan. Nat. Commun. 4, 2300. https://doi.org/10.
1038/ncomms3300. 152. Ritchie, M.E., Phipson, B., Wu, D., Hu, Y., Law, C.W., Shi, W., and Smyth,
G.K. (2015). Limma powers differential expression analyses for RNA-
137. Anisimova, A.S., Alexandrov, A.I., Makarova, N.E., Gladyshev, V.N., and sequencing and microarray studies. Nucleic Acids Res. 43, e47.
Dmitriev, S.E. (2018). Protein synthesis and quality control in aging. Aging https://doi.org/10.1093/nar/gkv007.
(Albany NY) 10, 4269–4288. https://doi.org/10.18632/aging.101721.
153. Viechtbauer, W. (2010). Conducting Meta-Analyses in R with the metafor
138. Swovick, K., Firsanov, D., Welle, K.A., Hryhorenko, J.R., Wise, J.P.,J.P., Package. J. Stat. Softw. 36, 1–48.
George, C., Sformo, T.L., Seluanov, A., Gorbunova, V., and Ghaemma-
ghami, S. (2021). Interspecies differences in proteome turnover kinetics 154. Anders, S., and Huber, W. (2010). Differential expression analysis for
are correlated with life spans and energetic demands. Mol. Cell. Prote- sequence count data. Genome Biol. 11, R106. https://doi.org/10.1186/
omics 20, 100041. https://doi.org/10.1074/mcp.RA120.002301. gb-2010-11-10-r106.
139. Kushiyama, A., Nakatsu, Y., Matsunaga, Y., Yamamotoya, T., Mori, K., 155. Dobin, A., Davis, C.A., Schlesinger, F., Drenkow, J., Zaleski, C., Jha, S.,
Ueda, K., Inoue, Y., Sakoda, H., Fujishiro, M., Ono, H., et al. (2016). Batut, P., Chaisson, M., and Gingeras, T.R. (2013). STAR: Ultrafast uni-
Role of uric acid metabolism-related inflammation in the pathogenesis versal RNA-seq aligner. Bioinformatics 29, 15–21. https://doi.org/10.
of metabolic syndrome components such as atherosclerosis and nonal- 1093/bioinformatics/bts635.
coholic steatohepatitis. Mediators Inflamm. 2016, 8603164. https://doi. 156. Liao, Y., Smyth, G.K., and Shi, W. (2014). FeatureCounts: An efficient
org/10.1155/2016/8603164. general purpose program for assigning sequence reads to genomic fea-
140. Zhang, B., Lee, D.E., Trapp, A., Tyshkovskiy, A., Lu, A.T., Bareja, A., Ker- tures. Bioinformatics 30, 923–930. https://doi.org/10.1093/bioinformat-
epesi, C., Katz, L.H., Shindyapina, A.V., Dmitriev, S.E., et al. (2021). ics/btt656.
Multi-omic rejuvenation and lifespan extension upon exposure to youth- 157. Friedman, J., Hastie, T., and Tibshirani, R. (2008). Sparse inverse covari-
ful circulation. Preprint at bioRxiv. https://doi.org/10.1101/2021.11.11. ance estimation with the graphical lasso. Biostatistics 9, 432–441.
468258. https://doi.org/10.1093/biostatistics/kxm045.
20 Cell 186, 1–21, June 22, 2023

ll
Article
158. Schliep, K.P. (2011). phangorn: Phylogenetic analysis in R. Bioinformat- antioxidants. Free Radic. Biol. Med. 34, 1070–1077. https://doi.org/10.
ics 27, 592–593. https://doi.org/10.1093/bioinformatics/btq706. 1016/S0891-5849(03)00042-X.
159. Seluanov, A., Hine, C., Bozzella, M., Hall, A., Sasahara, T.H.C., Ribeiro, 168. Edgar, R., Domrachev, M., and Lash, A.E. (2002). Gene Expression
A.A.C.M., Catania, K.C., Presgraves, D.C., and Gorbunova, V. (2008). Omnibus: NCBI gene expression and hybridization array data repository.
Distinct tumor suppressor mechanisms evolve in rodent species that Nucleic Acids Res. 30, 207–210. https://doi.org/10.1093/nar/30.1.207.
differ in size and lifespan. Aging Cell 7, 813–823. https://doi.org/10. 169. Hashimshony, T., Senderovich, N., Avital, G., Klochendler, A., de Leeuw,
1111/j.1474-9726.2008.00431.x. Y., Anavy, L., Gennert, D., Li, S., Livak, K.J., Rozenblatt-Rosen, O., et al.
160. Pickering, A.M., Lehr, M., Kohler, W.J., Han, M.L., and Miller, R.A. (2015). (2016). CEL-Seq2: sensitive highly-multiplexed single-cell RNA-Seq.
Fibroblasts from longer-lived species of primates, rodents, bats, carni- Genome Biol. 17, 77. https://doi.org/10.1186/s13059-016-0938-8.
vores, and birds resist protein damage. J. Gerontol. A Biol. Sci. Med. 170. Schultz, M.B., Kane, A.E., Mitchell, S.J., MacArthur, M.R., Warner, E.,
Sci. 70, 791–799. https://doi.org/10.1093/gerona/glu115. Vogel, D.S., Mitchell, J.R., Howlett, S.E., Bonkowski, M.S., and Sinclair,
161. Salphati, L., Wong, H., Belvin, M., Bradford, D., Edgar, K.A., Prior, D.A. (2020). Age and life expectancy clocks based on machine learning
W.W., Sampath, D., and Wallin, J.J. (2010). Pharmacokinetic-pharma- analysis of mouse frailty. Nat. Commun. 11, 4618. https://doi.org/10.
codynamic modeling of tumor growth inhibition and biomarker modula- 1038/s41467-020-18446-0.
tion by the novel phosphatidylinositol 3-kinase inhibitor GDC-0941. 171. Shindyapina, A.V., Cho, Y., Kaya, A., Tyshkovskiy, A., Castro, J.P., Deik,
Drug Metab. Dispos. 38, 1436–1442. https://doi.org/10.1124/dmd. A., Gordevicius, J., Poganik, J.R., Clish, C.B., Horvath, S., et al. (2022).
110.032912. Rapamycin treatment during development extends life span and health
162. Raynaud, F.I., Eccles, S.A., Patel, S., Alix, S., Box, G., Chuckowree, I., span of male mice and Daphnia magna. Sci. Adv. 8, eabo5482. https://
Folkes, A., Gowan, S., De Haven Brandon, A., Di Stefano, F., et al. doi.org/10.1126/sciadv.abo5482.
(2009). Biological properties of potent inhibitors of class I phosphatidyli- 172. Felsenstein, J. (1985). Phylogenies and the comparative method. Am.
nositide 3-kinases: From PI-103 through PI-540, PI-620 to the oral agent Nat. 125, 1–15. https://doi.org/10.1086/284325.
GDC-0941. Mol. Cancer Ther. 8, 1725–1738. https://doi.org/10.1158/
173. Revell, L.J. (2012). phytools: An R package for phylogenetic comparative
1535-7163.MCT-08-1200.
biology (and other things). Methods Ecol. Evol. 3, 217–223. https://doi.
163. Gelardi, T., Caputo, R., Damiano, V., Daniele, G., Pepe, S., Ciardiello, F.,
org/10.1111/j.2041-210X.2011.00169.x.
Lahn, M., Bianco, R., and Tortora, G. (2008). Enzastaurin inhibits tumours
sensitive and resistant to anti-EGFR drugs. Br. J. Cancer 99, 473–480. 174. Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery
https://doi.org/10.1038/sj.bjc.6604493. rate: A practical and powerful approach to multiple testing. J. R. Stat.
Soc. B 57, 289–300.
164. Kim, K., Kong, S.Y., Fulciniti, M., Li, X., Song, W., Nahar, S., Burger, P.,
Rumizen, M.J., Podar, K., Chauhan, D., et al. (2010). Blockade of the 175. Kolesnikov, N., Hastings, E., Keays, M., Melnichuk, O., Tang, Y.A., Wil-
MEK/ERK signaling cascade by AS703026, a novel selective MEK1/2 liams, E., Dylag, M., Kurbatova, N., Brandizi, M., Burdett, T., et al.
inhibitor, induces pleiotropic anti-myeloma activity in vitro and in vivo. (2015). ArrayExpress update–simplifying data submissions. Nucleic
Br. J. Haematol. 149, 537–549. https://doi.org/10.1111/j.1365-2141. Acids Res. 43, D1113–D1116. https://doi.org/10.1093/nar/gku1057.
2010.08127.x. 176. Linnet, K. (1993). Evaluation of regression procedures for methods com-
165. Yongxi, T., Haijun, H., Jiaping, Z., Guoliang, S., and Hongying, P. parison studies. Clin. Chem. 39, 424–432. https://doi.org/10.1093/clin-
(2015). Autophagy inhibition sensitizes KU-0063794-mediated anti- chem/39.3.424.
HepG2 hepatocellular carcinoma cell activity in vitro and in vivo. Bio- 177. Lek, M., Karczewski, K.J., Minikel, E.V., Samocha, K.E., Banks, E., Fen-
chem. Biophys. Res. Commun. 465, 494–500. https://doi.org/10.1016/ nell, T., O’Donnell-Luria, A.H., Ware, J.S., Hill, A.J., Cummings, B.B.,
j.bbrc.2015.08.045. et al. (2016). Analysis of protein-coding genetic variation in 60,706 hu-
166. Garcı́a-Martı́nez, J.M., Wullschleger, S., Preston, G., Guichard, S., mans. Nature 536, 285–291. https://doi.org/10.1038/nature19057.
Fleming, S., Alessi, D.R., and Duce, S.L. (2011). Effect of PI3K- and 178. Shihab, H.A., Rogers, M.F., Campbell, C., and Gaunt, T.R. (2017). HIPred:
mTOR-specific inhibitors on spontaneous B-cell follicular lymphomas An integrative approach to predicting haploinsufficient genes. Bioinfor-
in PTEN/LKB1-deficient mice. Br. J. Cancer 104, 1116–1125. https:// matics 33, 1751–1757. https://doi.org/10.1093/bioinformatics/btx028.
doi.org/10.1038/bjc.2011.83. 179. Shao, Y., Chen, C., Shen, H., He, B.Z., Yu, D., Jiang, S., Zhao, S., Gao, Z.,
167. Veurink, G., Liu, D., Taddei, K., Perry, G., Smith, M.A., Robertson, T.A., Zhu, Z., Chen, X., et al. (2019). GenTree, an integrated resource for
Hone, E., Groth, D.M., Atwood, C.S., and Martins, R.N. (2003). Reduction analyzing the evolution and function of primate-specific coding genes.
of inclusion body pathology in ApoE-deficient mice fed a combination of Genome Res. 29, 682–696. https://doi.org/10.1101/gr.238733.118.
Cell 186, 1–21, June 22, 2023 21

ll
Article
STAR+METHODS
KEY RESOURCES TABLE
REAGENT or RESOURCE SOURCE IDENTIFIER

Antibodies
Anti-mouse CD19 BioLegend Cat#115503; RRID: AB_313638
Anti-mouse/human CD11b BioLegend Cat#101203; RRID: AB_312786
Chemicals, peptides, and recombinant proteins
DMEM, high glucose, GlutaMAX Gibco Cat#10566016
Fetal Bovine Serum R&D Systems Cat#S11150
Bovine Serum Albumin Sigma Cat#A9418
Antibiotic-Antimycotic (100X) Gibco Cat#15240096
DMSO (Dimethyl sulfoxide) Sigma Cat#D2650
Paraquat dichloride hydrate Sigma Cat#36541
Cardamonin Selleck Chemicals Cat#S3942
Clofilium tosylate Targetmol Cat#T14982
Deguelin Selleck Chemicals Cat#S8132
KU-0063794 MedChemExpress Cat#HY-50710
AZD8055 MedChemExpress Cat#HY-10422
Ascorbyl-palmitate MedChemExpress Cat#HY-B0987
GDC-0941 MedChemExpress Cat#HY-50094
Enzastaurin MedChemExpress Cat#HY-10342
AS-703026 MedChemExpress Cat#HY-12042
Critical commercial assays
PureLink RNA Mini Kit Thermo Fisher Scientific Cat#12183020
CellTiter-Glo Luminescent Cell Promega Cat#G7570
Viability Assay
Deposited data
Raw data files and processed data files for This paper GEO: GSE227360
RNA-seq
mSALT database This paper http://gladyshevlab.org/mSALT/
Metabolite profiling data Ma et al.27; Tyshkovskiy et al.33 N/A
Public gene expression data from tissues Fushan et al.31; Brawand et al.37; Kim GEO: GSE43013, GSE30352, GSE30337,
and fibroblasts of mammalian species et al.34; Merkin et al.38; Qiu et al.41; GSE41637, GSE33300, GSE42297, GSE39150,
Seim et al.4; Fan et al.40; Seim et al.35; GSE50726; SRA: PRJNA263931, PRJNA72723
Yim et al.39; Fang et al.36
Public gene expression data on See Table S1B for a list of sources N/A
mammalian aging
Gene expression signatures of lifespan- Tyshkovskiy et al.33 N/A
extending interventions
Yeast gene expression signatures of Kaya et al.70 N/A
longevity
Public gene expression data of yeast Janssens et al.73 N/A
replicative aging
Public gene expression profiles of mouse Osburn et al.109; Baze et al.110; Zhou GEO: GSE15891, GSE11287, GSE11899,
tissues subjected to interventions et al.111; Yeung et al.112; Bongers et al.113; GSE46209, GSE63007, GSE28085, GSE36838
Hand et al.114; Sunami et al.115
(Continued on next page)
e1 Cell 186, 1–21.e1–e12, June 22, 2023

ll
Article
Continued
144
Animal Ageing and Longevity (AnAge) Tacutu et al. https://genomics.senescence.info/species/
Database index.html
Gene sets of housekeeping, essential and Wang et al.44 N/A
evolutionary old genes
Tabula Muris single-cell atlas Almanzar et al.24 GEO: GSE109774
Experimental models: Cell lines
C57Bl/6J mouse fibroblasts (Mus Lee et al.145 N/A
musculus)
UM-HET3 mouse fibroblasts (Mus University of Michigan N/A
musculus)
Rat fibroblasts (Rattus norvegicus) University of Rochester N/A
Naked mole rat fibroblasts (Heterocephalus Lee et al.145 N/A
glaber)
Western long-eared myotis fibroblasts University of Michigan N/A
(Myotis evotis)
Yuma myotis fibroblasts (Myotis University of Michigan N/A
yumanensis)
Human fibroblasts (Homo sapiens) Lee et al.145 N/A
Experimental models: Organisms/strains
Baboon (Papio anubis) Southwest National Primate Research N/A
Center (SNPRC)
Canadian beaver (Castor canadensis) Southwest National Primate Research N/A
Center (SNPRC)
Long-tailed macaque (Macaca fascicularis) Southwest National Primate Research N/A
Center (SNPRC)
Siberian chipmunk (Tamias sibiricus) Fushan et al.31 N/A
American black bear (Ursus americanus) Fushan et al.31 N/A
Sugar glider (Petaurus breviceps) Fushan et al.31 N/A
Greater tube-nosed bat (Murina Fushan et al.31 N/A
leucogaster)
White-footed mouse (Peromyscus Fushan et al.31 N/A
leucopus)
Mouse: C57BL/6J National Institute on Aging Aged Rodent N/A
Colony
Mouse: UM-HET3 University of Michigan Medical School N/A
Software and algorithms
Flow cytometry analysis: FlowJo BD Biosciences N/A
Adaptor removing: Trimmomatic Bolger et al.146 http://www.usadellab.org/cms/index.php?
(version 0.32) page=trimmomatic
Functional enrichment: GSEA Subramanian et al.147 http://software.broadinstitute.org/gsea/
index.jsp
Functional enrichment: gprofiler2 Kolberg et al.148; Raudvere et al.149 https://cran.r-project.org/web/packages/
gprofiler2/index.html
App development: shiny Chang et al.150 https://shiny.rstudio.com/
Programming environment: RStudio https://www.rstudio.com/ N/A
Differential gene expression analysis of Robinson et al.151 https://bioconductor.org/packages/release/ bioc/
RNAseq: edgeR html/edgeR.html
Differential gene expression analysis of Ritchie et al.152 https://bioconductor.org/packages/release/ bioc/
microarrays: limma html/limma.html
Mixed-effect model: metafor Viechtbauer et al.153 http://CRAN.R-project.org/package=metafor
(Continued on next page)
Cell 186, 1–21.e1–e12, June 22, 2023 e2

ll
Article
Continued
154
RNAseq normalization: RLE Anders and Huber https://bioconductor.org/packages/release/ bioc/
html/edgeR.html
Prediction of compounds with similar gene Lamb et al.107; Subramanian et al.147 https://clue.io
expression response: CMap
Identification of similar gene expression https://artyomovlab.wustl.edu/ N/A
profiles: GeneQuery genequery/
Mapping: STAR (version 2.5.2b) Dobin et al.155 https://github.com/alexdobin/STAR
Counting: featureCoutns (version 1.5) Liao et al.156 https://subread.sourceforge.net/
Harmonic mean p value: harmonicmeanp Wilson et al.48 https://cran.r-project.org/web/packages/
harmonicmeanp/index.html
Cell type deconvolution: bayesPrism Chu et al.49 https://github.com/Danko-Lab/BayesPrism
Sparse partial correlation matrix: glasso Friedman et al.157 https://cran.r-project.org/web/packages/
glasso/index.html
Phylogenetic analysis: phangorn Schliep et al.158 https://cran.r-project.org/web/packages/
phangorn/index.html
RESOURCE AVAILABILITY
Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Vadim N.
Gladyshev ([email protected]).
Materials availability
This study did not generate new unique reagents.
Data and code availability

d RNA-seq data reported in this work are available at NCBI GEO data repository under accession number GEO: GSE227360
(SubSeries GSE227358 and GSE227359). Accession numbers for publicly available datasets analyzed in this study are listed
in the key resources table.
d mSALT database can be accessed through the following link: http://gladyshevlab.org/mSALT/. Data analysis code is available
from the authors upon request.
d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.
EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS
Mammalian species sample collection

The 39 organ samples for young adults of baboon (Papio anubis), beaver (Castor canadensis), and long-tailed macaque (Macaca fas-
cicularis) were collected from the Southwest National Primate Research Center (SNPRC). The 9 organ samples for young adults of
chipmunk (Tamias sibiricus), American black bear (Ursus americanus), sugar glider (Petaurus breviceps), tube-nosed bat (Murina leu-
cogaster), and white-footed mouse (Peromyscus leucopus) were collected as previously described.31
Cell lines and culture

Fibroblasts from 6 species were cultured, including Mus musculus (C57BL/6J and UM-HET3 strains), Rattus norvegicus, Heteroce-
phalus glaber, Myotis evotis, Myotis yumanensis and Homo sapiens. Mouse (C57BL/6J strain) and naked mole rat (NMR) fibroblast
cell lines were established as previously described from 3 individual animals per species/strain.145 Primary human fibroblasts were
derived from skin biopsy. The rat fibroblast line was obtained from the University of Rochester collection.159 Primary fibroblast lines
from UM-HET3 mice and two bats (western long-eared myotis and Yuma myotis) were from the University of Michigan collection.160
All fibroblasts were maintained in Dulbecco’s modified Eagle’s medium (DMEM) with Glutamax and supplemented with 10% fetal
bovine serum and 1X antibiotic/antimycotic. All cells were cultured at 37 C and 5% CO2, while NMR cells were cultured at 32 C,
3% O2 and 5% CO2.
e3 Cell 186, 1–21.e1–e12, June 22, 2023

ll
Article
Animals and predicted compounds

Three-month old UM-HET3 mice were subjected to diets containing compounds predicted with the longevity gene expression sig-
natures via Connectivity map (CMap)107,108: GDC-0941 (pictilisib) (50 ppm, as in Salphati et al.161 and Raynaud et al.162), enzastaurin
(75 ppm, as in Graff et al.125 and Gelardi et al.163), AS-703026 (30 ppm, as in Kim et al.164), KU0063794 (10 ppm, as in Yongxi et al.165),
AZD8055 (20 ppm, as in Garcı́a-Martı́nez et al.166) and ascorbyl-palmitate (6.3 ppm, as in Veurink et al.167) for 1 month. Liver and
kidney samples were taken from treated mice along with their untreated sex- and age-matched littermates, which were fed ad libitum.
In all cases, interventions continued until the animals were sacrificed. All animal protocols were approved by the Brigham and
Women’s Hospital and University of Michigan Institutional Animal Care and Use Committees.
All organisms received the same diet (Purina 5LG6) made in the same commercial diet kitchen (TestDiet, Richmond, IN, USA).
Genetically heterogeneous UM-HET3 strain, in which each mouse had unique genetic background but was derived from inbred
grandparents of the same background (C57BL/6J, BALB/cByJ, C3H/HeJ, and DBA/2J), was used in this setting. This cross produces
a set of genetically diverse animals, which minimizes the possibility that the identified signatures represent gene expression patterns
specific to inbred lines. Moreover, this strain was used by ITP to test the lifespan extension potential of the compounds analyzed in
this study. In all cases, interventions continued until the animals were sacrificed.
All mice were kept at a density of 3 males per ventilated cage, in a specific-pathogen free vivarium, with 12:12 light:dark cycle.
Animals were moved to fresh cages every 14 days. Maintenance of specific-pathogen free status was documented quarterly, using
sentinel mice exposed to spent bedding sampled from each experimental cage, and evaluated by a mixture of fecal RT-PCR tests
and serology for anti-viral antibodies. Health was evaluated daily for each mouse.
Animals and KU0063794 treatment

Old C57BL/6J male mice were obtained from the National Institute on Aging Aged Rodent Colony. 22 months old and 25 months old
mice were subjected either to control 5053 diet or to 10 ppm KU0063794 incorporated into 5053 diet and were fed ad libitum. All diets
were irradiated and heat-sealed and stored at 4ºC for 6 months or less. During lifespan analysis mice were checked every weekday.
Mice died from natural causes or were euthanized due to the moribund state (characterized by body condition score of 1.5 or less,
tumor size over 2 cm, reluctance to move, or labor breathing). Animals from cross-sectional cohort were euthanized with CO2
following by cervical dislocation. All experiments using mice were performed in accordance with institutional guidelines for the
use of laboratory animals and were approved by the Brigham and Women’s Hospital and Harvard Medical School Institutional Animal
Care and Use Committees.
METHOD DETAILS
RNA-seq profiling of mammalian tissues

For RNA-seq analysis of mammalian species, 48 samples were collected from different tissues of young males corresponding to 8
species: baboon, beaver, long-tailed macaque, American black bear, chipmunk, sugar glider, tube-nosed bat, and white-footed
mouse (Table S1A). RNA-seq libraries were prepared as previously described.31 Paired-end sequencing with 200bp (for baboon,
beaver, long-tailed macaque) or 100 bp (for bear, chipmunk, sugar glider, tube-nosed bat, white-footed mouse) read length was per-
formed on Illumina HiSeq2000 platform. Additional RNA-seq libraries were obtained from the following NCBI Gene Expression
Omnibus (GEO)168 and Sequence Read Archive (SRA) datasets: GSE43013,31 GSE30352,37 GSE30337,34 GSE41637,38
GSE33300,41 GSE42297,4 GSE39150,40 PRJNA263931,35 PRJNA7272339 and GSE5072636 (Table S1A). Total mammalian dataset
included 371 biological samples from brain (36 species), kidney (36 species), liver (41 species), cerebellum (10 species), heart (13
species), and testis (10 species). As an out-group, we also included 12 chicken (Gallus gallus) samples obtained from the same or-
gans. All analyzed samples were obtained from males, except for vervet and horse where females were used.
For RNA-seq analysis corresponding to drugs predicted with longevity signatures, we used 4 and 8 biological replicates per exper-
imental group for treated and control mice, respectively (Table S1C). RNA was extracted from tissues with PureLink RNA Mini Kit as
described in the protocol and passed to sequencing. RNA-seq libraries were prepared as described in Hashimshony et al.169 and
sequenced with 100 bp read length option on the Illumina HiSeq 2500.
Stress resistance assay

Stress resistance assay was performed as described previously.100 Briefly, on day 1, fibroblasts were suspended at a concentration
of 5,000 cells/100 ml (0.5 3 105 cells/ml) in DMEM plus 0.5% BSA. Cells were then aliquoted into 96-well plates and incubated over-
night at 37 C or 32 C (in the case of NMR cells) in a 5% humidified CO2 or 5% humidified CO2, 3% O2 (for NMR) incubator. On day 2,
0.1 ml of Cardamonin, Clofilium tosylate, Deguelin or DMSO (control) were added, for an initial compound concentration of 8 mM
(Cardamonin), 4 mM (Clofilium tosylate), 2 mM (Deguelin) and the initial DMSO concentration of 0.1%. Plates were then incubated
overnight. On day 3, 5 ml of medium alone (control) or medium plus paraquat were added to the plates. Paraquat concentration
was optimized for each species so that, on average, approximately 50% of fibroblasts survived following this treatment. On day
4, 100 ml of CellTiter-Glo detection reagent was added to each well, and luminescence (RLU) was measured on a SynergyTM HT
Multi-Mode Microplate Reader (BioTek Instruments, Inc). The experiment was reproduced 3-6 times for each compound and con-
trol group for every species and mouse strain.
Cell 186, 1–21.e1–e12, June 22, 2023 e4

ll
Article
Frailty index and gait speed

Frailty index was measured as described in Schultz et al.170 prior to the start of the treatment (in 25-month-old mice) and after
5 months of treatment for control mice and mice subjected to KU0063794 (30-month-old mice). Frailty index score was calculated
per the original protocol. Gait speed was measured as in Shindyapina et al.171 for same groups of mice. In brief, mice were positioned
in the standard mouse cage that had free space on the one end and a food container on the other end. The time of finish was deter-
mined as the time when the mouse nose crossed the finish line. Each mouse was let run four times and the median value was re-
corded. An attempt was considered successful if the mouse crossed the finish line without standing or turning 90 degrees or
more during the run. Ten seconds were recorded if a mouse failed to finish the run in 10 seconds.
Glucose tolerance test

24-month-old mice subjected KU0063794 for 2 months and age-matched control mice were placed into food-free cages for 16 hours
(usually from 7 pm until 11 am the next day). The next day, mice were weighted, marked, bled from the tail and the fasting blood
glucose level was measured with a glucometer Accu-Chek Perform Nano (Accu-check inform II strips). Filtered 30% glucose in sterile
saline solution (Sigma) was injected i.p. at a final dose of 1 g/kg of body weight. Injections were done at 30-second intervals between
the mice. Glucose level was measured again at 20, 40, 60 and 120 minutes after the injection with a 30-second interval between
the mice.
Flow cytometry
Myeloid cells were gated as CD45+CD3-CD11b+CD19-, B cells as CD45+CD11b-CD3-CD19+, T cells as CD45+CD3+CD11b-CD19-.
mAbs used for staining included: anti-CD19 [6D5], anti-CD11b [M1/70], anti-CD3 [17A2], and anti-CD45 [30-F11] (all from Biolegend).
Dead cells were excluded by DAPI staining. Data was collected on a Cytek DXP11 and analyzed by FlowJo software (BD). Spleens
were gently pressed between microscopy slides to get single-cell suspensions. One ml of cell suspensions were 1) incubated with
14 ml of red blood lysis buffer for 10 min on ice, 2) centrifuged at 4 C, 250 g for 10 min, 3) washed once with 1 ml of FACS buffer (1%
FBS in PBS), 4) stained in 100 ul of AB solution (2 ng/ul of each AB) at 4 C for 20 min protected from light, 5) washed again, 6) filtered
into tubes with cell strainer snap cap (Corning), and 7) analyzed with flow cytometry.
Necropsy analysis
Mice were euthanized with CO2 followed by cervical dislocation. The chest and abdomen were opened, and the body was immersed
into formalin solution and stored at 4 C until further analysis. For necropsy, all organs, including small endocrine organs, were
dissected, trimmed at 5 mm thickness and embedded in paraffin blocks. Paraffin blocks were sectioned at 5 mm and stained with
hematoxylin and eosin. The slides were examined blindly by a pathologist.
QUANTIFICATION AND STATISTICAL ANALYSIS
RNA-seq data processing

For tissue samples from mammalian species, we utilized the pipeline from Ma et al.30 to generate reference sequences and identify
species-specific ortholog gene sets (using the published annotated genomes in Ensembl or NCBI if available; otherwise generating
de novo transcriptome assembly with Trinity). Genome annotation for 29 out of 42 species was available in NCBI or Ensembl, whereas
the other 13 species required de novo assembly of the transcriptomes. The RNA-seq reads were then mapped to the species-spe-
cific ortholog sets using STAR155 and read counting was performed by featureCounts.156 Those ortholog sets with too high counts
(i.e. read counts contributing to >5% of the total counts) or too low counts (i.e. less than 10 counts in >30% of the samples) were
discarded. The library sizes were scaled by trimmed mean of M-values (TMM) method, log10-transformed, and quantile-normal-
ized.151 The final expression set consisted of 13,784 gene orthologs across 42 species, including 41 mammalian species and chicken
(Gallus gallus) used as an out-group. Phylogenetic tree on Figure 1A was constructed based on nucleotide sequences of 13,784 gene
orthologs using neighbor joining method.
For samples corresponding to predicted compounds, quality filtering and adapter removal were performed using Trimmomatic
(version 0.32).146 Processed/cleaned reads were then mapped with STAR (version 2.5.2b)155 and counted via featureCounts (version
1.5).156 To filter out genes with low number of reads, we left only genes with at least 6 reads in at least 66.6% of samples, which re-
sulted in 9,011 detected genes according to Entrez annotation. Filtered data was then passed to RLE normalization.154
Ortholog quality of mammalian species samples

Before proceeding with data analysis of mammalian species RNA-seq samples, we assessed the quality of the orthologs in terms of
the following areas (Figure S1). The coding sequences were checked for completeness of open-reading frame (including start codon,
stop codon, and correct coding frame). Sequence fragments or partially missing sequences (e.g. those species with no published
genomes and had to rely on de novo transcriptome assembly) were filled up using the consensus of related species. In the mamma-
lian transcriptome dataset, 80% of the orthologs did not require filling up or were filled up <10% of sequence length, and there was no
significant bias against those filled up using consensus in terms of standardized expression values (Figure S1B). To assess the quality
of our de novo assembly strategies, we also generated de novo transcriptomes for those species with published genomes (18
e5 Cell 186, 1–21.e1–e12, June 22, 2023

ll
Article
species with genomes available in Ensembl and 11 species with genomes available in NCBI; Table S2). Read alignment rates to the
ortholog sets were consistent across the samples with and without complete genomes (Figure S1A, upper), and the Spearman
correlation coefficients between the read counts based on de novo assembled ortholog set alignment and those based on genome
alignment were >0.95 for most of these species (Figure S1A, lower), suggesting that the de novo assembled ortholog sets accurately
reflected the gene expression counts. Lastly, for each of the 18 species with annotated genomes in Ensembl, we compared our or-
tholog definition with the Ensembl ortholog definition. Ortholog data for 10,000 to 15,000 of our orthologs (per species) could be
found in Ensembl, and 90-99% of them matched our definition (Table S2), suggesting that the results of our pipeline were consistent
with other databases.
Life history data of the species

The adult weight (AW), maximum lifespan (ML) and female time to maturity (FTM) data of the species (or if not available, for a closely
related species) were obtained from the Animal Ageing and Longevity (AnAge) Database.144 In addition, since both ML and FTM in-
crease with AW, we calculated the body mass adjusted residuals (i.e. MLres and FTMres), to represent the ratio between the
observed longevity and the expected longevity based on body mass.27,30 Two allometric equations were used to calculate the re-
siduals. The MLres equation, MLres = ML =ð4:88 3 AW 0:153 Þ, was based directly on the documentation of the AnAge database
(http://genomics.senescence.info/help.html#anage). The FTMres equation, FTMres = FTM =ð78:1 3 AW 0:217 Þ, was based on linear
regression using the FTM and body mass records of 1330 mammalian species in the AnAge database.
Mammalian RNA-seq data analysis

Principal component analysis (PCA) was performed on the standardized expression values and the first three Principal Components
(PCs) were extracted. 6,050 genes significantly enriched or depleted in one organ relative to the others (p value <0.01) were identified
with Wilcoxon rank-sum test. Heatmap showing organ-specific expression patterns was constructed based on standardized expres-
sion values of these genes. Complete linkage hierarchical clustering was performed based on Spearman correlation distance.
ANOVA was used to assess the percentage of variation in the expression of individual genes explained by species and tissues
across all available samples. Within tissues, 82-94% of variance was explained by species (Figure S1D). To account for a potential
batch effect introduced by different data sources aggregated during analysis, we also performed a similar analysis within datasets,
running it on samples from the same data source. Datasets with at least 4 different species and 8 biological replicates available for a
given organ were considered. For each of them, we applied ANOVA for every gene to estimate the percentage of variation in the
expression explained by species. The calculated estimates were pooled together and visualized on boxplots. This analysis resulted
in similar estimates (78-95% of variance explained by species), suggesting that batch effect introduced by various sources of data is
relatively small compared to species-specific differences (Figure S1D).
Spearman correlation between average expression profiles of different species for each tissue was calculated in a pairwise manner
(Figure 1C). To adjust for differences in the number of species and biological replicates available for each organ, a similar analysis was
performed for nine species that were available for all presented organs (bonobo, chicken, chimpanzee, gorilla, human, macaque,
mouse, opossum, and platypus). Besides, the same number of biological replicates for each species was selected across all tissues.
The difference between the median correlation coefficient in the testis and other organs was assessed using Wilcoxon rank-sum test
(Figure S1E).
To identify genes with significant correlation to the longevity traits (ML, MLres, FTM or FTMres), regression was performed using
the generalized least square approach, by incorporating the phylogenetic relationship in the variance-covariance matrix.27,30,172,173
As previously described,27,30 four different trait evolution models (‘null’, ‘Brownian motion’, ’Pagel’s lambda’, and ’Ornstein-Uhlen-
beck’) were tested and the best fit model was selected based on maximum likelihood. A two-step procedure was applied to verify the
robustness of the results. In the first step, the species whose exclusion would lead to most improvement in the slope p value (i.e. a
potential outlier), was identified and removed. The regression p value of this step was reported as ‘sensitive p value’. In the second
step, maximum p value after exclusion of each individual species was identified, reported as ‘robust p value’. The False Discovery
Rate (FDR) correction based on Benjamini-Hochberg (BH) approach was performed to adjust for multiple hypothesis testing.174 To
qualify as a top hit, we required a gene to have adjusted p value < 0.05. To test if genes associated with longevity retained their as-
sociation after adjustment for body mass, we performed the same procedure for identified top hits (adjusted p value < 0.05) using
MLres or FTMres as outcome variables. We qualified gene as significant if corresponding p value < 0.05 after this adjustment.
To account for a potential bias introduced by differences in gene length and sequence variation across species, for every gene in
the ortholog set we calculated its length in all examined species based on the de novo assembled transcriptomes together with the
nucleotide sequence distance from the corresponding Mus musculus sequence. Pairwise sequence distance was estimated based
on Jukes-Cantor model with ‘dist.ml’ function from R package phangorn.158 We then introduced estimated values of gene length and
sequence distance as additional covariates in the regression model, which was used to identify significant gene expression signa-
tures of ML and FTM across mammalian species. For every tissue, we then calculated the proportion of signature genes that retained
a statistically significant association with the longevity traits following this adjustment (p value < 0.05). Adjustment for variation in gene
length and nucleotide sequence across species did not affect the statistically significant association with ML and FTM for more than
96.5% and 93.5% of identified signatures, respectively.
Cell 186, 1–21.e1–e12, June 22, 2023 e6

ll
Article
To build an unbiased model of ML prediction based on tissue gene expression, we applied an Elastic Net linear regression algo-
rithm from scikit-learn library (linear_model.ElasticNet function). We divided species in training and test sets using leave-one-out
(LOO) procedure. Each individual species was consequently used as a test sample, while other species were used for training
and cross-validation (CV). Hyperparameters of the model were trained using 10-fold cross-validation. Trained hyperparameters
included Pearson’s correlation coefficient threshold for feature selection (between 0.15 and 0.75), alpha (between 10-4 and 102)
and l1 ratio (between 0.2 and 1). Hyperparameters associated with the lowest mean absolute error (MAE) on CV set were chosen
during each LOO iteration. The trained model was then applied to the test sample, and predicted value of log10(Maximum Lifespan)
was calculated. Afterwards, predictions for each test sample were pooled together, and the accuracy of the model was assessed
using mean absolute error (MAE), R2 and Pearson’s correlation coefficient. When applied to test set, final model resulted in
R2 = 0.78 and MAE = 0.13, corresponding to 35% difference in ML (Figure 2D). The same approach was used to estimate the ac-
curacy of prediction based on tissue gene expression and adult AW as well as AW alone. AW was included in the model in both linear
and logarithmic scales. Model based only on AW resulted in R2 = 0.39 and MAE = 0.24 (Figure S2D).
Top gene expression predictors of mammalian ML were assessed based on model coefficient distributions across Elastic Net
models trained on different subsets. For every gene, mean model coefficient was calculated together with its standard error. Statis-
tical significance of gene coefficient difference from zero was estimated using t test. p values were adjusted for multiple comparisons
using BH approach. To quality as a robust predictor of maximum lifespan, we required a gene to have adjusted p value < 0.05.
Signatures of maximum lifespan in cultured primary skin fibroblasts collected from 16 mammalian species were obtained from Ma
et al.30 Spearman correlation between gene expression signatures associated with mammalian longevity in different organs and fi-
broblasts was calculated in a pairwise manner. The union of top 500 longevity-associated genes (with the lowest p values) was used
for each pair of signatures. Hierarchical clustering of signatures was performed based on complete linkage and Euclidean distance.
To determine the statistical significance of overlap between signatures associated with ML or MLres in different tissues, we per-
formed Fisher exact test separately for up- and downregulated genes, considering 13,784 genes as a background.
Aggregation of aging data for meta-analysis

To identify signatures associated with mammalian aging, we aggregated gene expression data from the following GEO,
ArrayExpress175 and SRA datasets: GSE9103, GSE123981, GSE3150, GSE6591, GSE74463, GSE53960, GSE66715, GSE11291,
GSE34378, GSE27625, GSE12480, GSE36192, GSE1572, GSE28422, GSE25941, GSE53890, GSE38718, GSE674, GSE17612,
GSE21935, GSE362, GSE132040, E-MTAB-3374, PRJNA281127, PRJNA516151. The datasets included RNA-seq and microarray
samples obtained from different tissues of Mus musculus, Rattus norvegicus and Homo sapiens. In total, our meta-analysis covered
aging-associated changes in 17 different tissues based on 92 datasets from 25 different sources (Figures 3A and S3A; Table S1B).
Prior to aggregating the data into signatures, the following preprocessing protocol was executed for individual datasets. For RNA-
seq data, low-covered genes were filtered out using a soft threshold. Then gene Ensembl IDs were mapped to Entrez IDs. The read
was filtered out if there were zero or multiple Entrez IDs corresponding to the Ensembl ID. In case of multiple Ensembl IDs corre-
sponding to one Entrez ID, the gene coverage was calculated as the sum of the corresponding read coverages. Afterwards, the
expression data was normalized using RLE method,154 log-transformed and scaled.
For microarray data, gene expression was log-transformed to conform to normal distribution if needed. Then samples within every
study were normalized by scaling and quantile normalization. Afterwards, platform IDs were mapped to Entrez ID gene format. If mul-
tiple IDs corresponded to the same Entrez IDs, average log-expression was calculated.
For both types of data, self-consistency was evaluated using PCA, and outlier samples were discarded. Finally, mean and standard
error of aging-associated gene EC for every gene was calculated together with p value using limma.152 Pairwise comparison model
was utilized for datasets with 2 age groups, whereas linear model was used for datasets with multiple ages, resulting in log fold
change (logFC) or slope coefficient as a metric of EC, respectively. Since both these metrics estimate gene expression change
with age – in old animals compared to young ones or per unit of time, respectively – we considered them equally in the subsequent
pipeline (differences in time units were adjusted via normalization later, see ‘‘aging gene expression signatures’’). Obtained p values
were adjusted for multiple hypothesis testing using BH approach. Therefore, for every dataset we estimated mean EC, standard error
of EC, p value and adjusted p value for each expressed gene. Importantly, one study may include multiple datasets if several species
or tissues have been analyzed there. This may be a source of batch effect, which we removed during subsequent steps of the
analysis.
Dependence of pairwise correlation between the gene expression age-related ECs from individual datasets on meta-features
(belonging to the same study, species and tissue) (Figure 3G) was examined using the following model:
r = b0 + b1 3 study + b2 3 tissue + b3 3 species;
where r is a pairwise Spearman correlation coefficient, and study, tissue and species are binary factor variables, equal to 1 if source
ID, tissue and species, respectively, are the same for both datasets in the certain pair, and 0 otherwise. Belonging to the same tissue
had the highest effect on the similarity between aging-associated gene ECs (adding, on average, 0.23 to Spearman r), followed by
dataset- and species-specific effects (each adding, on average, 0.06 to the Spearman r).
e7 Cell 186, 1–21.e1–e12, June 22, 2023

ll
Article
To assess independent effects of tissue and species on similarity across datasets, we tested if they are maintained regardless of
whether the other factor matches. Specifically, we calculated Spearman correlation coefficients for the datasets corresponding to
(i) different tissues and species, (ii) different tissues and the same species, (iii) the same tissue and different species, (iv) the same
tissue and species (Figure S3B). To account for a potential batch effect associated with the source of data, for this analysis we utilized
only the pairs of datasets from independent studies. Differences between the groups were assessed with Wilcoxon rank sum test.
Interestingly, increased similarity between age-related ECs within the same species or tissue was maintained both when the other
factor matched and when it differed, pointing to the existence of independent tissue- and species-specific aging transcriptomic bio-
markers. Statistical significance of positive average correlation between the datasets corresponding to different species, tissues and
studies (group (i)) was assessed with Wilcoxon signed-rank test.
Aging gene expression signatures

Prior to signatures identification, normalization of ECs from individual datasets was performed based on the following algorithm. First,
denoised Pearson correlation of age-related ECs was calculated for each pair of datasets (Figure S3A). For that, we used the union of
top statistically significant aging-associated genes (with the lowest p values) in each dataset within certain pair. The threshold for the
number of genes was chosen so that it maximizes the number of significant pairwise correlations (adjusted p value < 0.05 & absolute
r > 0.1). We identified the threshold of 250 genes to be optimal for noise removal. Second, to bring ECs to the same scale across
individual datasets, we calculated normalization coefficients using multiple Deming regression model. Contrary to the simple linear
regression, Deming regression fits data by minimizing errors on both x and y axes, treating both variables equally. Therefore, it is well
suited for calibration of similar measurements performed in different studies.176 We optimized this method by applying it to all pairs of
aging datasets simultaneously, producing the set of normalization coefficients that would minimize sum of pairwise differences in
ECs across all the data. During this step, we applied multiple Deming regression to the pairs of datasets with significant correlations
(adjusted p value < 0.05 & absolute r > 0.1), considering top 250 significant genes in each case. The cumulative squared loss across
considered pairs of datasets was minimized with R function optim using L-BFGS-B method. Normalization coefficients were allowed
to vary between 0.01 and 100. To establish global minimum of error function, the multiple Deming regression was carried out 10 times
with random initial sets of normalization coefficients, and final coefficients were chosen from the run with the smallest cumulative
regression error. Among these 10 runs, the error minimum was the same for the majority of runs indicating that the global minimum
was achieved for each signature (Figure S4).
Finally, normalized ECs from individual datasets were used to identify robust aging signatures. To account for standard error of
gene ECs and to remove batch effect related to the belonging of several datasets to the same study, we applied linear mixed-effect
model using R package metafor.33,153 As an input, we used both mean and standard error of EC. To mitigate source batch effect and
adjust for non-equal representation of various tissues and species in our data, random terms corresponding to the dataset ID (GEO/
ArrayExpress/SRA ID), tissue and species were introduced in the model. Such approach allowed us to account for the size of the
effect and variance of the estimated EC within each individual dataset, which provides a more sensitive and accurate analysis
compared to previous studies focused on the comparison of lists of differentially expressed genes.
Using this procedure, we obtained aggregated normalized age-related EC and corresponding p value for every gene. We declared
genes to be statistically significant signatures of aging if adjusted p value was less than 0.05. The described algorithm was utilized
separately for 3 species-specific (human, mouse and rat), 3 tissue-specific (liver, brain and muscle) and 1 global (based on all pre-
sented tissues and species) aging signatures (Figure S3C).
To validate that described approach produces aging signatures that can be generalized to the independent data, we performed
4-fold cross-validation. Specifically, we divided the list of datasets into 4 similar subsets so that each independent source of data is
contained only in one subset. Iteratively, we used 3 folds to identify signatures applying the method described above, while the 4th
fold was used as a test set. We estimated pairwise Spearman correlations between gene expression changes from different datasets
in the test fold, using (i) all genes or (ii) significant signature genes (adjusted p value < 0.05) obtained on the training set. To assess the
efficiency of normalization based on multiple Deming regression, we also calculated correlation coefficients using (iii) significant
signature genes identified without the Deming regression step (with simple scaling normalization). Pairwise Spearman correlation
coefficients from test folds were then pooled together and visualized on boxplot (Figure S3D). Differences in average correlation co-
efficients estimated using the described sets of genes were assessed with Wilcoxon signed-rank test and adjusted for multiple com-
parisons using BH approach. On average, correlation between test datasets estimated based on signature genes was significantly
higher compared to those calculated based on the whole transcriptome, and it was further increased with multiple Deming regression
normalization, supporting validity of the utilized approach.
The significance of overlap between different aging signatures was estimated by Fisher exact test separately for up- and down-
regulated genes, considering all genes tested for age association as a background. BH procedure was used to adjust p values
for multiple comparisons. Normalized ECs for visualization of gene expression changes across individual aging datasets and aggre-
ECi
gated signatures (Figures 3F and 3H) were calculated as sdðECÞ , where i denotes a certain gene and sdðECÞ is a standard deviation of
ECs across all genes in the given dataset or signature.
Cell 186, 1–21.e1–e12, June 22, 2023 e8

ll
Article
Aggregated biomarkers of aging and longevity

To identify genes robustly associated with certain trait (aging, lifespan-extending interventions and longevity across species), we
calculated harmonic mean p value (HMP) for every gene across individual signatures corresponding to the trait using R package har-
monicmeanp.48 To account for EC direction, harmonic mean p values were estimated for both positive and negative association as
described in Yoon et al.83 Briefly, individual p values obtained from two-tailed tests were halved and synchronized according to
the effect direction. In other words, every pi was converted to pi/2 if the directions of true effects and tested association coincide
or to 1 pi otherwise. Then harmonic mean p values were calculated for both cases, and the smaller of the two combined p values
was selected. This p value was multiplied by two for the two-tailed meta-analysis p value and adjusted for multiple comparisons using
Benjamini-Hochberg method. We considered a gene as significant in the aggregated signature of the trait if its adjusted HMP was
<0.05 and the rate of signatures where it was changed in the same direction was higher than 80%. Overlap of significant trait signa-
tures was assessed using Pearson’s chi-squared test separately for each pair of traits.
Expression of signatures across cell types

Cell type deconvolution of mammalian organ samples was performed with R package BayesPrism.49 Tabula Muris single-cell atlas
(both 10x and FACS data) corresponding to liver, kidney, brain (myeloid and non-myeloid), limb muscle and heart, was downloaded
from GEO database (GSE109774) and used as a reference for cell type annotation.24 For visualization, different types of epithelial and
endothelial cells were pooled together. B cells, T cells, microglial cells, Kupffer cells, macrophages, natural killer cells and other leu-
kocytes were considered as immune cells. The average proportion of each cell type and standard error was calculated for every or-
gan (Figure S2F). To adjust for subtle variation in blood cell abundance across species, we introduced the proportion of immune cells
estimated for individual samples as a separate covariate into the regression model. For every tissue, we then estimated the percent-
age of previously identified signature genes that retained a statistically significant association with mammalian maximum lifespan (p
value < 0.05) after this adjustment.
A gene was considered to be expressed in a certain cell type if at least 3 counts were detected in at least 5% of cells corresponding
to this type. For every gene, we calculated a proportion of individual cell types where it was expressed. To estimate it, we considered
99 cell types present in the Tabula Muris single-cell atlas across all available tissues. For individual and aggregated signatures of
mammalian lifespan and aging, we visualized the distributions of calculated proportions for the corresponding significant genes
(adjusted p value < 0.05) using boxplots (Figure S2E).
We defined genes as blood cell specific biomarkers, if they were expressed in a higher proportion of blood cell types than non-
blood cell types, and less than in 10% of non-blood cell types, according to the Tabula Muris atlas. This threshold resulted in 830
blood cell specific genes. Blood cell specific biomarkers accounted for fewer than 1.5% of genes significantly associated with
maximum lifespan of mammalian species, and their removal did not affect significant positive correlation between signatures of in-
dividual tissues (Spearman r > 0.21 for all pairwise comparisons).
Comparison of longevity signatures and traits

To compare mean ECs corresponding to individual gene expression signatures associated with mammalian aging and longevity at
inter- and intra-species level, we utilized denoised Spearman correlation method as described previously. p values were adjusted for
multiple comparisons using BH method. Normalized ECs for barplot visualization were calculated by dividing aggregated gene
expression changes on standard deviation of ECs across all genes within the signature as described previously (Figure 4D). Genes
were selected for visualization if they (i) were significantly associated (adjusted p value < 0.05) with multiple longevity or aging traits
and (ii) were established to be associated with longevity based on other studies or were annotated as a member of gene set demon-
strating significant association with the analyzed traits.
To identify a comprehensive list of shared and distinct molecular biomarkers across examined traits, we utilized Fisher’s combined
probability test from R package metap, an effective and sensitive technique to integrate p values from various tests. Contrary to the
HMP method, this approach requires aggregated tests to be independent, but provides high statistical power for detection of incom-
plete associations.83 To detect common biomarkers for a group of traits, Fisher’s combined probability test was applied to every
gene, using HMP corresponding to each model and adjusting for the direction of EC (similar to the HMP algorithm). To detect distinct
biomarkers, the same algorithm was applied, but sign of EC was inverted for one of the models. In each case, Fisher’s combined
probability test was applied separately for both positive and negative associations, and the minimum of the two resulted p values
was chosen, similar to the HMP algorithm described previously. BH method was used to adjust resulted p values for multiple hypoth-
esis testing. We considered a gene as a significant common or distinct biomarker of the traits if its adjusted Fisher’s combined prob-
ability p value was <0.05 and the rate of signatures where it was changed in the same direction was higher than 80% within each of the
traits included in the analysis. The described algorithm was performed separately for identification of common and distinct bio-
markers within each pair of traits (‘‘Aging & Interventions’’, ‘‘Aging & Species’’, ‘‘Interventions & Species’’) as well as between aging
traits and both longevity traits at once (‘‘Longevity & Aging’’). As a result, we found 300-1,200 significant shared and opposite sig-
natures across various pairs of traits (Figure 4F). Difference between the number of significant common and distinct biomarkers
for each of the analyzed trait sets was evaluated using two-sample proportion test.
e9 Cell 186, 1–21.e1–e12, June 22, 2023

ll
Article
Yeast signatures of longevity and aging

Gene expression signatures of budding yeast longevity across natural strains and deletion mutants were identified in our previous
study.70 Briefly, 40 natural strains of S. cerevisiae were used to evaluate biomarkers of replicative lifespan (RLS) across strains,
while signatures of long-lived deletion strains were calculated based on RLS and gene expression data of 1,376 knockout strains
from McCormick et al.72 and Kemmeren et al.,71 respectively.
To obtain gene signatures of yeast replicative aging, we utilized deconvolved transcriptome data from Janssens et al.73 Genes,
which expression was associated with time in linear and logarithmic scale, were identified using limma.152 Resulted p values were
adjusted for multiple hypothesis testing using BH method. Genes with adjusted p value < 0.05 were considered significant.
Mean slope coefficients corresponding to ECs associated with yeast aging and longevity across natural and deletion strains were
compared using Spearman correlation. p values were adjusted for multiple comparisons using BH method. Normalized ECs for bar-
plot visualization were calculated by dividing gene expression changes on standard deviation as described previously.
Functional enrichment analysis of signatures

For identification of functions enriched by individual aging and longevity signatures, we performed gene set enrichment analysis
(GSEA)147 on a pre-ranked list of genes based on log10(p value) corrected by the sign of regulation, calculated as:
log10 ðpvÞ 3 sgnðecÞ;
where pv and ec are p value and aggregated expression change for certain gene, respectively, estimated with mixed-effect model,
and sgn is signum function (is equal to 1, -1 and 0 if value is positive, negative and equal to 0, respectively). REACTOME, KEGG and
GO BP from Molecular Signature Database (MSigDB) have been used as gene sets for GSEA.147 Adjusted p value cutoff of 0.1 was
used to select statistically significant functions.
To identify functions enriched by aggregated biomarkers of traits as well as common and distinct signatures of various models, we
performed Fisher exact test using R package gprofiler2,148,149 considering all genes tested for association with the corresponding
trait as a background. KEGG, REACTOME and GO BP ontologies were used. We declared functions to be enriched if their BH
adjusted Fisher exact test p value was smaller than 0.1. Gene ratio was estimated for enriched functions as proportion of genes
in the given signature, which are associated with the given pathway. Significance score was calculated as log10(adjusted p value)
corrected by the sign of direction.
Hierarchical clustering of enriched functions for a heatmap was performed based on normalized enrichment scores (NES) and sig-
nificance score for GSEA and Fisher exact test, respectively, using complete linkage and Spearman correlation distance. Enriched
functions that (i) were significantly enriched (adjusted p value < 0.1) by multiple signatures and (ii) represented different aspects of
cellular biology, as assessed manually and based on the overlap of corresponding gene sets, were selected for visualization. The
whole lists of statistically significant enriched functions are available in Tables S3 and S5.
Metabolomics analysis
Metabolite profiling data corresponding to 26 mammalian species and the effect of 5 lifespan-extending interventions on mouse liver
in males and females were based on data from Ma et al.27 and Tyshkovskiy et al.33 Each species in the mammalian dataset was rep-
resented by 1-4 biological replicates per tissue (2.6 samples on average), while each treatment in the longevity intervention dataset
was represented by 5-6 biological replicates per sex. To filter out metabolites with low coverage, only metabolites detected in at least
66.6% of the samples were kept. Afterwards, filtered data were log10-transformed and scaled.
The effect of individual interventions on metabolite concentration was assessed separately for males and females using limma.
Besides, a single model combining the effect of all interventions was built to identify common signatures of lifespan extension. Finally,
to find metabolites associated with the quantitative effect of interventions, a linear model incorporating the effect of certain interven-
tion on median and maximum lifespan (taken from Tyshkovskiy et al.33) was created. In each case, sex and batch were introduced in
the model as separate covariates to account for the possible bias. Metabolites were declared significant if their p values, adjusted by
BH method, were <0.1.
To investigate associations of metabolite levels with ML, MLres, FTM and FTMres across species, phylogenetic regression model
was built separately for each tissue (brain, liver, kidney, and heart), as described previously for gene expression data. To identify sig-
natures of mammalian longevity across tissues, we aggregated mean slopes of metabolite concentration from all the organs together
with their standard errors in a fixed-effect model using R package metafor. Metabolites were declared significant if their p values,
adjusted by BH method, were less than 0.1.
Partial correlation network

To assess interdependence between various molecular mechanisms of longevity and aging, we utilized the mammalian tissue gene
expression dataset and computed a partial correlation network for species maximum lifespan, adult weight, transcriptomic signa-
tures of aging and lifespan-extending interventions as well as the expression of key individual genes (Igf1, Rela, Cth) and functional
gene sets significantly associated with different models of longevity and aging (Figure 6C). For every available tissue and species,
each of these features was calculated. For functional gene sets, the score was estimated as mean normalized expression of all genes
Cell 186, 1–21.e1–e12, June 22, 2023 e10

ll
Article
associated with this function according to KEGG or REACTOME ontology. For the signatures of aging and interventions, the differ-
ence of mean normalized expression between genes positively and negatively associated with this trait was calculated. Biomarkers
of mouse maximum lifespan and the aggregated signature across species and tissues were used as gene sets associated with
longevity interventions and aging, respectively. Sparse partial correlation network of estimated features was computed with R pack-
age glasso, using graphical lasso regularization and model selection based on Extended Bayesian information criterion (EBIC).157 The
resulted Gaussian graphical model was visualized with R package qgraph.
Effect of compounds on fibroblast survival

Percentage of viable cells in each sample was subjected to log transformation. The effect of every compound on cell survival under
oxidative stress conditions was calculated as a difference between mean log percentage of viable cells in experimental group
(treated with certain compound) and mean log percentage of viable cells in control group (treated by DMSO). Standard error was
calculated using t test. To estimate the average effect of the compound on the survival of cells from short-lived species (mice and
rats), we aggregated the corresponding estimates of an average compound effect and SE across these species in a single
random-effect model using R package metafor. An association between the effect of a compound and species longevity was as-
sessed with fixed effect model where ML or MLres was included as an independent variable. Percentage of viable cells in the control
group was not significantly correlated with species lifespan (Spearman r = 0.16, p value = 0.24). To adjust for a potential effect of this
factor on the compound effect, we included mean log percentage of viable cells in control group to the regression model as a co-
variate. This adjustment did not affect statistical significance of investigated associations.
Enrichment of trait biomarkers by gene sets

To identify enrichment of trait biomarkers by housekeeping, essential and evolutionary old genes, we utilized gene sets from Wang
et al.44 Gene essentiality was defined based on two metrics: mutation intolerance (the probability of being intolerant to loss-of-func-
tion mutation)177 obtained from ExAC release 0.3.1 (http://exac.broadinstitute.org/), and haploinsufficiency (sensitivity to a gene
copy number reduction) estimated in Shihab et al.178 Phylogenetic age of the genes was obtained from the GenTree database.179
Evolutionary old and young genes were defined as 1:1 therian orthologs that originated before the emergence of bony vertebrates
and after the emergence of tetrapods, respectively.
To assess if significant trait biomarkers as well as common and opposite signatures of various traits are enriched for a certain
feature described above, we utilized Fisher exact test. For each pair of gene sets (housekeeping / non-housekeeping, haplosufficient
/ haploinsufficient, mutation tolerant / intolerant, evolutionary old / young), we estimated overrepresentation of one gene set over the
other in a subset of signature genes. As a background for each trait, we used genes that were tested for an association with the cor-
responding trait in the signature analysis but did not reach statistical significance. Besides, to account for the fact that some of the
gene sets may be enriched for genes with a higher expression, and highly expressed genes have also a larger chance of being iden-
tified as a signature of a certain trait due to a higher statistical power, we controlled for this potential confounder by randomly select-
ing background genes in such a way that the genes associated and not associated with the trait have a similar distribution of average
expression in mammals (Figure S6C). Specifically, for every trait, we divided the range of average log(expression) of genes across
mammals in 100 uniform windows, and for every window we randomly picked up the same proportion of genes showing and not
showing a significant association with the trait. The resulting subset of non-significant genes was used as a background for Fisher
exact test. Enrichment for certain feature within each pair of gene sets was considered significant if BH adjusted p value was smaller
than 0.1. Odds ratios (OR) were calculated as:

signset1
nonsignset1

signset2
;
nonsignset2
where set1 and set2 correspond to the opposite sets of genes (etc. housekeeping and non-housekeeping genes), while sign and
nonsign correspond to the number of genes associated and not associated with the trait, respectively.
Pathway enrichment analysis for evolutionary old, haploinsufficient and mutation-intolerant signatures of longevity across species
and evolutionary young, haplosufficient and mutation-tolerant signatures of lifespan-extending interventions was performed using
Fisher exact test in a similar way. REACTOME and KEGG ontologies were utilized.
Prediction of longevity interventions

To identify interventions associated with longevity at gene expression level, we employed GSEA-based algorithm developed in our
previous study.33 First, for every individual signature and global trait we specified 1000 genes with the lowest p values and divided
them into up- and downregulated genes. These lists were considered as gene sets. Then we ranked genes related to interventions of
interest based on their p values, as described in functional enrichment section.
To find interventions associated with longevity traits from publicly available sources, we utilized GeneQuery tool (https://
artyomovlab.wustl.edu/genequery/). Datasets of interest were then downloaded from GEO under the following accession numbers:
e11 Cell 186, 1–21.e1–e12, June 22, 2023

ll
Article
GSE15891,110 GSE11287,109 GSE11899,114 GSE46209,112 GSE63007,113 GSE28085115 and GSE36838.111 We preprocessed each
dataset, performed quantile normalization and Entrez ID transformation and applied limma model for calculation of p values, which
were converted to log10(p value) corrected by the sign of regulation, as described earlier.
For compounds predicted via CMap, we calculated p values of gene expression logFC in livers of treated mice compared to control
independently for every drug using edgeR. We then converted them to log10(p value) corrected by the sign of regulation as described
earlier and proceeded to GSEA-based analysis.
We calculated GSEA scores separately for up- and downregulated lists of gene set as described in Lamb et al.107 and defined final
GSEA score as a mean of the two. To calculate statistical significance of obtained GSEA score, we performed permutation test where
we randomly assigned genes to the lists of gene set maintaining their size. To get p value of association between certain intervention
and longevity signature, we calculated the frequency of observed final GSEA score being bigger by absolute value than random final
GSEA scores obtained from 5,000 random permutations. Permutation test p values were further adjusted for multiple hypothesis
testing using BH method. Associations were considered significant if adjusted p value was smaller than 0.1.
ADDITIONAL RESOURCES
Interactive database mSALT based on shiny framework150: http://gladyshevlab.org/mSALT/.
Cell 186, 1–21.e1–e12, June 22, 2023 e12

ll
Article
Supplemental figures
Figure S1. Mammalian RNA-seq data quality assessment, related to Figure 1

(A) Comparison of read alignment to ortholog sets and to genome. Percentage of reads for each species aligned to the ortholog sets are shown (top). For species
with complete genomes, the reads were also aligned to the respective genomes (middle), and the average Spearman correlation coefficient between the read
counts of ortholog set alignment and the read counts of genome alignment was calculated (bottom). Data are mean ± SE. See Table S2 for more details.
(B) The effect of filling up with consensus. The ortholog sets were categorized by their lengths that required filling up (left), and corresponding relative expression
values are shown (right).
(C) Clustering of the samples from different sources. The symbols in the parenthesis indicate the source of sample. Hierarchical clustering was performed using
complete linkage on a Pearson correlation distance matrix.
(D) Percentage of gene expression variation explained by species and replicates in various tissues. Boxplots reflect the percentage of explained variance in gene
expression estimated using ANOVA across datasets (left) and within individual datasets (right).
(E) Gene expression patterns are least conserved in testes. Pairwise Spearman correlation coefficients were calculated for each tissue based on 9 species with all
organs being present. Same number of biological replicates were used across tissues. Median values of correlation coefficients between testis and other tissues
were compared using Wilcoxon rank-sum test.
*p adjusted < 0.05; **p adjusted < 0.01; ***p adjusted < 0.001.
ll
Article
Figure S2. Additional features of gene expression signatures of longevity across mammalian species, related to Figure 2
(A–C) Association of selected genes with female time to maturity (A) and maximum lifespan adjusted for body mass (B and C). Selected genes include Rpl30 (A),
Rpl28 (B), and Cul4b (C). Association between log10(female time to maturity) (A) or log10(maximum lifespan adjusted for body mass) (B and C) and average
normalized log10(expression) is shown for brain (left), liver (middle), and kidney (right). The black line, equation, slope p value, and R2 correspond to the model
fitted with phylogenetic regression. Data are mean ± SE.
(D) Accuracy of Elastic Net species maximum lifespan prediction based on adult animal weight evaluated on a test set. Adult weight was introduced in linear and
logarithmic scale. Each dot represents a single species and is colored by taxonomic group, as in (A)–(C). Mean absolute error (MAE), R2 and Pearson’s correlation
coefficient are shown in text.
(E) Expression of genes associated with mammalian maximum lifespan across cell types. Percentages of cell types with detected expression of significant
individual (brain, liver, and kidney) and aggregated longevity-associated genes (adjusted p value < 0.05) are shown.
(F) Proportion of various cell types in brain, liver, and kidney samples. Immune cells, including Kupffer cells, microglial cells, macrophages, B cells, T cells, natural
killer cells, and other leukocytes, were pooled together. Data are mean proportions across species ± SE.
(G) Spearman correlation between transcriptomic signatures of longevity across species in organs and primary fibroblasts. Spearman correlation coefficients and
adjusted p values are reflected by numbers and asterisks, respectively.
(H) Functional enrichment of mammalian longevity signatures. Only functions significantly enriched by at least one signature are shown (adjusted p value < 0.1).
Statistical significance is reflected with asterisks. The whole list of enriched functions is in Table S3A.
ll
Article
(legend on next page)

ll
Article
Figure S3. Identification of gene expression signatures of aging, related to Figure 3

(A) Denoised Spearman correlation between aging-associated gene expression changes across individual datasets. Correlations are shown separately for all the
datasets (global), species-specific signatures (human, rat, and mouse), and tissue-specific signatures (brain, skeletal muscle, and liver).
(B) Similarity of aging-related gene expression changes across datasets corresponding to the same or different tissues and species. Denoised Spearman
correlation coefficients were calculated for every pair of datasets from independent sources. Statistical difference between 4 subsets of dataset pairs was
assessed with Wilcoxon rank-sum test. *p adjusted < 0.05; **p adjusted < 0.01; ***p adjusted < 0.001.
(C) Spearman correlation of age-associated changes across datasets calculated based on the signatures of aging. Correlations were calculated separately for the
global signature, species-specific signatures (human, rat, and mouse), and tissue-specific signatures (brain, skeletal muscle, and liver).
(D) Spearman correlation of age-associated changes across datasets in test set calculated based on all genes or significant signature genes identified from the
training test. Signatures were calculated without (middle) or with (right) normalization based on multiple Deming regression. BH-adjusted p values calculated with
Wilcoxon signed-rank test are shown in the text.
(E) Expression of aging-associated genes across cell types. Percentages of cell types with detected expression of tissue-specific (brain, liver, and skeletal
muscle), species-specific (human, rat, and mouse), and aggregated signatures of aging (adjusted p value < 0.05) are shown.
ll
Article
Figure S4. Normalization of aging datasets based on multiple Deming regression, related to Figure 3
Each dot represents a single run from a different set of initial parameters. Boxplots represent the distribution of final normalization coefficients for a given dataset,
whereas color of the dots reflects the associated mean squared error (MSE).
ll
Article
Figure S5. Metabolite changes associated with longevity across and within species, related to Figures 4 and 5
(A) Denoised Spearman correlation of S. cerevisiae aging and longevity signatures. Signatures of replicative aging, lifespan-extending deletions, and inter-strain
longevity are shown in red, green, and blue, respectively. Statistical significance of each pairwise correlation is shown with asterisks.
(B) Metabolites associated with the effect of lifespan-extending interventions in mice and longevity across mammalian species. Normalized slopes of association
between metabolite concentration and mouse median lifespan (y axis) or mammalian maximum lifespan adjusted for body mass (y axis) are shown. Metabolites,
whose concentration is significantly associated with longevity according to interventions, species, or both models, are shown in green, blue, and red,
respectively.
(C) Association of murine Nadsyn1 expression change with longevity signatures. Statistical significance of each association denoted with asterisks was estimated
with phylogenetic regression (for species longevity signatures) and mixed-effect model (for other signatures). Data are mean normalized ECs ± SE.
(D) Association of yeast Ndt1 (YIA6) expression change with longevity signatures. Statistical significance of each association denoted with asterisks was esti-
mated with linear regression. Data are mean normalized ECs ± SE.
^p adjusted < 0.1; *p adjusted < 0.05; **p adjusted < 0.01; ***p adjusted < 0.001. EC, expression change; CR, calorie restriction; GH, growth hormone; ML,
maximum lifespan; MLres, maximum lifespan residual.
ll
Article
Figure S6. Common and distinct molecular signatures of longevity across and within species, related to Figure 6
(A) Spearman correlation of gene expression signatures based on functional enrichment (GSEA) scores. Only functions enriched by at least one signature
(adjusted p value < 0.1) were used for the calculation.
(B) Adjusted effect of cardamonin (left), clofilium tysolate (middle), and deguelin (right) on the survival of fibroblasts from species with different lifespans following
paraquat-induced oxidative stress. The effect of compound was calculated as a log ratio of the number of survived fibroblasts with and without treatment.
Here, the effect on survival was adjusted for the baseline survival rate (estimated for fibroblasts treated with DMSO) by incorporating the latter term into the
regression model. Dependence with maximum lifespan unadjusted (top) and adjusted (bottom) for adult weight was examined. Slope adjusted p value is shown in
text. Data are mean ± SE. n = 3–6 per compound and control group for every species and mouse strain.
(C) Average expression of genes associated and not associated with longevity and aging traits before (left) and after (right) filtering. 100 uniform windows were
used to randomly select subsets of signature and non-signature genes with similar distribution of log expression for every trait.
(D) Pathways enriched by evolutionary old, mutation-intolerant, and haploinsufficient genes associated with longevity across mammalian species. Adjusted p
value threshold of 0.1 is shown as a dotted line. The whole list of enriched functions is in Table S5A.
(E) Pathways enriched for evolutionary young, mutation-tolerant, and haplosufficient genes associated with lifespan-extending interventions in mouse. Adjusted
p value threshold of 0.1 is shown as a dotted line. The whole list of enriched functions is in Table S5B.
CR, calorie restriction; GH, growth hormone; ML, maximum lifespan; MLres, maximum lifespan residual; REAC, REACTOME.
ll
Article
Figure S7. KU0063794 effect on frailty features and B cell population in old mice, related to Figure 7
(A) Frailty index score in 25-month-old male C57BL/6 mice that were later assigned to treated and control groups.
(B) Frailty index score features in 30-month-old male C57BL/6 mice from treated and control groups.
(C) Percentage of B cell subpopulations in spleens of 27-month-old male C57BL/6 mice treated for 5 months or age-matched control groups measured by flow
cytometry. ABC, age-associated B cells; ACBCs, age-associated clonal B cells.
p values were calculated with one-tailed Wilcoxon rank-sum tests.

Distinct Longevity Mechanisms Across and Within

Uploaded by

Copyright:

Available Formats

Distinct Longevity Mechanisms Across and Within

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Distinct Longevity Mechanisms Across and Within

Uploaded by

Copyright:

Available Formats

Article

Distinct longevity mechanisms across and within

d Aging effects are reversed by longevity interventions but not

d Regulation of Igf1 and mitochondrial translation are shared

d Longevity signatures enable the discovery of

Tyshkovskiy et al., 2023, Cell 186, 1–21

Istanbul 34752, Turkey

INTRODUCTION ventions increasing murine lifespan and healthspan are known

Cell 186, 1–21, June 22, 2023 ª 2023 Elsevier Inc. 1

2 Cell 186, 1–21, June 22, 2023

Figure 1. RNA-seq of mammalian tissues

Cell 186, 1–21, June 22, 2023 3

Figure 2. Gene expression signatures of species longevity

4 Cell 186, 1–21, June 22, 2023

Figure 3. Transcriptomic signatures of mammalian aging

Cell 186, 1–21, June 22, 2023 5

6 Cell 186, 1–21, June 22, 2023

Figure 4. Interplay between transcriptomic signatures of aging and longevity

Cell 186, 1–21, June 22, 2023 7

8 Cell 186, 1–21, June 22, 2023

Figure 5. Molecular pathways associated with mammalian longevity

Cell 186, 1–21, June 22, 2023 9

Figure 6. Common and distinct transcriptomic signatures of longevity

10 Cell 186, 1–21, June 22, 2023

Cell 186, 1–21, June 22, 2023 11

12 Cell 186, 1–21, June 22, 2023

Figure 7. Transcriptomic longevity signatures reveal lifespan-regulating interventions

Cell 186, 1–21, June 22, 2023 13

14 Cell 186, 1–21, June 22, 2023

Cell 186, 1–21, June 22, 2023 15

16 Cell 186, 1–21, June 22, 2023

Cell 186, 1–21, June 22, 2023 17

18 Cell 186, 1–21, June 22, 2023

Cell 186, 1–21, June 22, 2023 19

20 Cell 186, 1–21, June 22, 2023

Cell 186, 1–21, June 22, 2023 21

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER

e1 Cell 186, 1–21.e1–e12, June 22, 2023

Cell 186, 1–21.e1–e12, June 22, 2023 e2

Data and code availability

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mammalian species sample collection

Cell lines and culture

e3 Cell 186, 1–21.e1–e12, June 22, 2023

Animals and predicted compounds

Animals and KU0063794 treatment

RNA-seq profiling of mammalian tissues

Stress resistance assay

Cell 186, 1–21.e1–e12, June 22, 2023 e4

Frailty index and gait speed

Glucose tolerance test

QUANTIFICATION AND STATISTICAL ANALYSIS

RNA-seq data processing

Ortholog quality of mammalian species samples

e5 Cell 186, 1–21.e1–e12, June 22, 2023

Life history data of the species

Mammalian RNA-seq data analysis