Midlife gene expressions identify modulators ofaging through dietary interventionsBing Zhoua,b,c,1, Liu Yangd,1, Shoufeng Lid, Jialiang Huanga,b,c, Haiyang Chene, Lei Houa,b,c, Jinbo Wangc,e,Christopher D. Greena, Zhen Yanf, Xun Huange, Matt Kaeberleing, Li Zhuh, Huasheng Xiaoh, Yong Liud,2,and Jing-Dong J. Hana,2

aChinese Academy of Sciences Key Laboratory for Computational Biology, Chinese Academy of Sciences—Max Planck Partner Institute for ComputationalBiology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai 200031, China; bCenter for Molecular Systems Biology, Institute ofGenetics and Developmental Biology, Chinese Academy of Sciences, Beijing 100101, China; cGraduate University, Chinese Academy of Sciences, Beijing100049, China; dKey Laboratory of Nutrition and Metabolism, Institute for Nutritional Sciences, Shanghai Institutes for Biological Sciences, Chinese Academyof Sciences, Shanghai 200031, China; eState Key Laboratory of Molecular Developmental Biology, Institute of Genetics and Developmental Biology, ChineseAcademy of Sciences, Beijing 100101, China; fDepartments of Medicine-Cardiovascular Medicine and Pharmacology and Center for Skeletal Muscle Research,Robert M. Berne Cardiovascular Research Center, University of Virginia, Charlottesville, VA 22908; gDepartment of Pathology, University of Washington,Seattle, WA 98195; and hKey Laboratory of Systems Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai 200031, China

Edited by Valter D. Longo, University of Southern California, Los Angeles, CA, and accepted by the Editorial Board March 20, 2012 (received for reviewNovember 23, 2011)

Dietary interventions are effective ways to extend or shortenlifespan. By examining midlife hepatic gene expressions in miceunder different dietary conditions, which resulted in differentlifespans and aging-related phenotypes, we were able to identifygenes and pathways that modulate the aging process. We foundthat pathways transcriptionally correlated with diet-modulatedlifespan and physiological changes were enriched for lifespan-modifying genes. Intriguingly, mitochondrial gene expression cor-related with lifespan and anticorrelated with aging-related path-ological changes, whereas peroxisomal gene expression showedan opposite trend. Both organelles produce reactive oxygen species,a proposed causative factor of aging. This finding implicates acontribution of peroxisome to aging. Consistent with this hypoth-esis, lowering the expression levels of peroxisome proliferationgenes decreased the cellular peroxide levels and extended thelifespan of Drosophila melanogaster and Caenorhabditis elegans.These findings show that transcriptional changes resulting fromdietary interventions can effectively reflect causal factors in agingand identify previously unknown or under-appreciated longevitypathways, such as the peroxisome pathway.

systems biology | diet-induced obesity

Aging-related gene expressions have been examined by micro-array analysis for various human and mouse tissues (1–5),

fruit flies (6), and worms (7, 8). These studies have revealedhundreds to thousands of genes and numerous biological func-tions that change as an organism ages. Some of the changes aresimilar across different species. Expressions of genes involved instress response and inflammation consistently increase in ani-mals, and those expressions involved in tissue-specific functionsgradually decrease, reflecting functional decline of tissues ororgans (9). However, most of these changes reflect the conse-quence of aging (some can serve as biomarkers of aging) ratherthan the cause or regulatory factors of aging. For instance, thekey aging regulatory genes identified by genetic approaches arerarely identifiable merely by the expression level changes duringaging (10). However, interventions of the aging process by ge-netic, dietary, or reproductive measures can effectively modulatelifespan and aging (11, 12).Caloric restriction (CR) is the best-studied intervention for

modulating aging and has been reported to prolong both meanand maximal lifespans in most organisms examined (11, 12). Incontrast, feeding mice a high-fat, high-calorie diet results in age-related obesity, cardiovascular diseases, and other metabolicdisorders, and it shortens lifespan (13–15). Exercise, however,can increase energy expenditure, reduce body weight, and pre-vent some age-related functional declines (16). It is, therefore,

not surprising that nutrient- and energy-sensing pathways havebeen identified by genetic approaches to be key regulators oflifespan and aging (12, 17).Similar to the powerful genetic manipulations, manipulations

of energy input or expenditure also provide a means to modifythe phenotypes and course of aging. Analogous to the concept ofquantitative trait loci (QTL), the incremental changes in lifespanand aging-related phenotypes under various dietary/environ-mental conditions from low- to high-energy input could beexplained by gene expression changes of lifespan/aging regu-lators that are responsive to energy input changes. Unlike ge-netic manipulations that only examine one (or at most, a few)loci at a time, the incremental changes of gene expression thatare correlated with lifespan and other aging-related phenotypesunder different energy regimens may capture many regulatorychanges at the same time. Because of the heterogeneous geneticbackground of QTL analysis, this method often requires hun-dreds of samples to infer putative causal variants from naturallyoccurring variants. We expect that the sample size required forenvironmental manipulations is much smaller than the samplesize required for QTL because of the homogeneous geneticbackground of the animals, the standardization of treatments,and the averaging effect within group to minimize individualvariations. Under this assumption, we subjected mice to six dif-ferent diet/energy regimens (30 mice per group) that led to thefollowing order of increasing lifespan: high-fat diet (HF) fed adlibitum high-fat diet combined with voluntary exercise (HF+Ex),low-fat diet (LF) fed ad libitum low-fat diet with voluntary Ex(LF+Ex), high-fat diet combined with 70% CR (HF+CR), andlow-fat diet with 70% CR (LF+CR). In addition to lifespan,health span parameters were assessed for each cohort, including

Author contributions: Z.Y., Y.L., and J.-D.J.H. designed research; B.Z., L.Y., S.L., J.H., H.C.,L.H., J.W., L.Z., and H.X. performed research; B.Z., L.Y., S.L., C.D.G., X.H., M.K., Y.L., andJ.-D.J.H. analyzed the data; and B.Z., L.Y., S.L., C.D.G., X.H., M.K., Y.L., and J.-D.J.H. wrotethe paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. V.D.L. is a guest editor invited by the EditorialBoard.

Data deposition: The microarray data reported in this paper have been deposited in theGene Expression Omnibus (GEO) database, www.ncbi.nlm.nih.gov/geo (accession nos.GSE36836 and GSE36838).1B.Z. and L.Y. contributed equally to this work.2To whom correspondence may be addressed. E-mail: liuy@sibs.ac.cn or jdhan@genetics.ac.cn.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1119304109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1119304109 PNAS Early Edition | 1 of 9

SYST

EMSBIOLO

GY

PNASPL

US

liver and metabolic functions, and gene expression profiles wereobtained at midlife before the increase in mortality.In this context, we first asked if the six different dietary groups

give rise to different lifespans according to their energy input andoutput levels; if so, we asked whether we can predict the lifespandifferences across these groups from the midlife liver phenotypesand hepatic gene expressions and finally, whether the genes orpathways that predict the lifespan differences are regulators oflifespan. The fact that all of the intervention experiments werecarried out in parallel rather than in different laboratories withvariable and/or noncomparable conditions enabled us to conductan integrative analysis that was free of system variations in thedata; we searched for common target genes of different dietaryinterventions that contribute to the consequent lifespan differ-ences through changes in their gene expression levels. Our resultsindicate that midlife liver gene expressions showing positive ornegative correlation with mean lifespan across the six groupsindeed identified not only many genes previously implicated inaging but at least one previously unknown or under-appreciatedaging-related pathway that implicates peroxisomal biogenesis asa key determinant of longevity.

ResultsLifespan and Metabolic Phenotypes of Mice Under Different Inter-vention Regimens. To arrive at six groups with different energy-input and -expenditure states, we maintained male C57BL/6Jmice (n= 30 per group) on either a regular LF diet or a HF diet;the food was fed ad libitum or calorically restricted to 70% of theamount of food offered daily to the ad libitum group (CR) or thegroup subjected to voluntary wheel-running Ex. Kaplan–Meiersurvival curves (Fig. 1A, Left) showed that lifelong feeding of HFdiet markedly reduced the lifespan of mice (by ∼20% for boththe mean and maximal lifespan) (Fig. 1A, Right), and CR exertedmore striking effects than Ex (its effects were monitored andrecorded as shown in SI Appendix, Fig. S1A) on extending thelifespan of animals fed either LF or HF diets. Overall, the orderof the mean and maximal lifespan (weeks) under the six regi-mens was as follows: HF (mean = 101, 118.8 ± 1.5 maximum) <HF+Ex (mean = 114, 139.7 ± 1.9 maximum) < LF (mean =127, 148.7 ± 3.1 maximum) < LF+Ex (mean = 131, 159.6 ± 3.7maximum) < HF+CR (mean = 137, 161.9 ± 1.5 maximum) <LF+CR (153 mean, 185.5 ± 1.6 maximum) (Fig. 1A). While CRunder LF diet feeding (LF+CR) increased the mean and max-imum lifespan by ∼20% and ∼25%, respectively, relative to theLF group, restricted HF intake (HF+CR) extended boththe mean and maximum lifespan by ∼36% compared with theHF group.Physiological analyses revealed the expected improving effects

of CR and Ex on animals’ body weight and whole-body fatcontent throughout their lifetime (Fig. 1 B and C). Notably,glucose tolerance tests indicated a close association of their ad-iposity levels with glucose intolerance when examined at 21 and30 wk of age across the six intervention groups; however, onlyanimals in the LF+CR group exhibited significant improvementof glucose tolerance after 60 wk of age (SI Appendix, Fig. S1B).When assessed at the midlife ages of 60–62 wk, CR not onlymarkedly prevented HF diet-induced obesity (Fig. 1 B and C)but also protected against obesity-associated hyperleptinemia,hyperinsulinemia, and dyslipidemia in the HF+CR mice, whichcontrasted the moderate effects of Ex (SI Appendix, Table S1).The effects of dietary interventions on the lifespan and meta-

bolic syndromes also agreed with the aging phenotypes of animalsunder these conditions, which were shown by aging-associatedbiomarkers including hair regrowth, rotarod, and tightrope assaysperformed in mice at the ages of 85–90 wk (SI Appendix, Fig. S2A–C). Thus, dietary modifications, particularly CR, most dra-matically affected aging rate and lifespan in this model.

To assess the contribution of energy balance to lifespan, wequantified the daily energy intake of each group of mice before62 wk of age, which displayed the following order: HF (17.14kcal) > HF+Ex (16.58 kcal) > LF+Ex (12.53 kcal) > LF (12.29kcal) > HF+CR (12.03 kcal) > LF+CR (8.95 kcal) (Fig. 1D).Using the comprehensive laboratory animal monitoring system,we also measured the energy expenditure for mice at 57–58 wkof age. Because of the different body size of mice in each group,we compared the metabolic rate by normalizing to the metabolicsize as reflected by body weight0.75 (18) of each animal. CRdid not significantly affect oxygen consumption in mice fed ei-ther LF or HF (Fig. 1E), whereas voluntary Ex significantlyincreased it in LF-fed mice compared with their correspondingfeeding group. Additionally, CR led to an increase in physical

A

B

C

D E

Fig. 1. Effects of diet and Ex on lifespan and energy state. (A, Left) Kaplan–Meier survival curves (n = 22 per group) showing significant changes inlifespan between intervention pairs of HF/LF (P = 0.9E-6), HF+CR/HF (P =0.87E-9), HF+Ex/HF (P = 0.006), and LF+CR/LF (P = 0.42E-5) by log-rank test.(A, Right) Mean and maximum lifespan of mice in the six interventiongroups as derived from the Kaplan–Meier survival curves. Maximum lifespanwas calculated from the average of the oldest 20% of mice in each group. (B)Body weight was monitored throughout the (Left) indicated ages and(Right) midlife (62-wk age) until the number of mice alive in each group wasless than six. (C) Total body fat content was measured by NMR (n = 8–10 pergroup) at (Left) each indicated age and (Right) midlife (62-wk age) until 92wk of age. (D) Energy intake was calculated based on the daily food con-sumption of mice during the intervention before 62 wk of age (n = 8 pergroup). (E) Oxygen consumption was determined at midlife when mice were57–58 wk of age (n = 5 per group). Data in B–D are presented as means ±SEM. *P < 0.05, **P < 0.01 vs. LF, ++P < 0.01 vs. HF by ANOVAs.

2 of 9 | www.pnas.org/cgi/doi/10.1073/pnas.1119304109 Zhou et al.

activity during the light cycle in the LF-fed group (SI Appendix,Fig. S2D), and it resulted in significantly decreased body tem-perature in both LF- and HF-fed groups at 90 wk of age (SIAppendix, Fig. S2E).

Liver Health at Midlife in Mice Under Different Intervention Regimens.Reasoning that the liver is central to metabolic homeostasis andmay serve as a functional organ indicator of whole-body health,we decided to examine whether liver pathology at midlife underdifferent intervention regimens correlates with the lifespan phe-notype. At around midlife age of 62 wk, mice of the HF group hadenlarged liver, with a twofold increase in liver weight relative tothe LF group, whereas CR more potently antagonized HF-in-duced hepatomegaly than Ex (SI Appendix, Table S1). Next, weevaluated the hepatic function by measuring serum levels of ala-nine aminotransferase (ALT) and aspartate aminotransferase(AST), the well-established markers of liver injury or damage. Inthe HF group of mice, serum ALT or AST levels were markedlyelevated, and in contrast, both were significantly lower in theHF+CR and HF+Ex groups (SI Appendix, Fig. S3A), suggestingreductions of HF-associated hepatocellular damages by CR andEx. Again, CR had a more pronounced effect of reversing thisconsequence of HF feeding than Ex (SI Appendix, Fig. S3A).Consistently, CR completely prevented HF-induced hepatic

overload of triglycerides (TG; i.e., hepatosteatosis) and consid-erably reduced liver cholesterol content in the HF+CR group,while also decreasing liver TG accumulation in the LF-CR group(SI Appendix, Table S1 and Fig. S3B). Because chronic liverdamage induced by nonalcoholic fatty liver can lead to liver fi-brosis, we wondered whether CR or Ex could exert an effect onthis process. Assessment by Sirius Red staining of fibrotic col-lagen deposition showed that ∼12% of the livers from the HFgroup were Sirius Red-positive, representing an almost 10-foldincrease relative to the LF group that exhibited very low levels offibrosis (SI Appendix, Fig. S3C). Both CR and Ex could markedly

reduce HF-induced liver fibrosis from ∼12% to ∼2%, which wascomparable with the level of the LF group of mice (SI Appendix,Fig. S3C).Perturbations of mitochondrial function are known to con-

tribute to metabolic dysfunctions and may play a causal role inthe aging process. In accordance with this notion, mice from theHF group showed an ∼50% reduction in the mitochondrialdensity in hepatocytes relative to the LF group but without ap-parent differences in their size (SI Appendix, Fig. S3D). CR in-creased the mitochondrial density in livers from HF-fed but notLF-fed animals, whereas Ex displayed no detectable effect re-gardless of diets (SI Appendix, Fig. S3D). Notably, both CR andEx increased the mitochondrial size on either diet, with CReliciting a larger effect. Therefore, CR could elevate both thedensity and size of mitochondria in livers of HF-fed mice (SIAppendix, Fig. S3D).

Correlation of Midlife Metabolic Phenotypes with Lifespan. Takentogether, the resultant physical and physiological phenotypes atboth the organismal and liver organ levels were highly coordinatedacross the six intervention regimens, which could be clustered intothree major groups (Fig. 2). The largest group reflected themidlifepathological states of the liver and other associated serum andwhole-body parameters such as serum cholesterol, body weight,and fat content, which positively correlated with energy intake(Fig. 2). The second group, one-half of the size of the aforemen-tioned group, consisted of mitochondria size and density, activityduring the light cycle, and other performance parameters in theaged animals (i.e., tightrope, rotarod, and hair regrowth), whichpositively correlated with lifespan (Fig. 2). The average z scorevalues in these two groups were highly negatively correlated [av-erage Pearson correlation coefficient (PCC) = −0.97]. Finally,loosely clustered with the second group, the third and smallestcluster included energy expenditure and activity during the darkcycle, which did not show a strong correlation with lifespan (Fig. 2).

Fig. 2. Midlife physiological parameters correlated with lifespan. Hierarchical clustering of the diet intervention groups and the indicated physiologicalparameters including body weight, fat content, midlife liver and serum measurements, energy intake and expenditure, and aging-sensitive markers based onthe group mean values as well as mean lifespans.

Zhou et al. PNAS Early Edition | 3 of 9

SYST

EMSBIOLO

GY

PNASPL

US

Midlife Liver Gene Expression Profiles Under Different InterventionRegimens. To understand the molecular basis of the physiologyand lifespan differences under the six intervention regimens, wemeasured the liver transcriptome of three mice from each groupat 62 wk of age using whole-genome microarrays (Materials andMethods). We clustered the profiles of genes differentiallyexpressed between any two groups (Materials and Methods andDataset S1) in the 18 samples using principle component analysis(PCA). Strikingly, the 18 samples were separated into fourclusters that reflected intervention regimens (Fig. 3A), whichwas consistent with hierarchical clustering (SI Appendix, Fig. S4).

The first three principle components (PCs) accounted for 68% ofthe gene expression variations. Generally, PC1 mainly reflected theindividual variation among samples, whereas PC2 accounted for22.7% of the variations and was most correlated with the averagelifespans of the six intervention groups (PCC = 0.98). PC3mostly distinguished the HF+CR group from LF-fed groups. Inparticular, LF+CR samples were the farthest from other sam-ples, indicating that they had a distinctive common expressionpattern (Fig. 3A). Moreover, the HF+CR group was separatedfrom HF-fed groups on PC1 and PC2, and it was more similar tothe LF group than the HF group (Fig. 3A). This finding suggests

A B

C

Fig. 3. Midlife liver gene expression correlations to lifespan are predictive of aging regulators. (A) PCA analysis of the profiles of the differentially expressedgenes (Materials and Methods and Dataset S1) for the indicated 18 samples. (B) Fold enrichment over genome-wide background for aging regulatory genesfrom GenAge database among the top 50, 100, or 200 genes with expressions that are most positively (+) or negatively (−) correlated with lifespan or liverphysiological parameters. Enrichments are shown for genes in pathways significantly positively/negatively correlated with lifespan with (as indicated by thelabels inside parentheses) or without prefiltering for significant correlation with other phenotypes. Significance of correlations was determined by PCCs(P < 0.05). (C) The KEGG pathways with gene expressions that significantly correlated or anticorrelated with the physiological parameters identified by theGSEA analysis.

4 of 9 | www.pnas.org/cgi/doi/10.1073/pnas.1119304109 Zhou et al.

that CR might counteract the effect of HF feeding and shift thetranscriptomic profile of HF mice to the profile of LF mice.Consistent with the relatively weak effect of physical Ex on liverphysiology and lifespan, Ex seemed to exhibit subtle effects onthe transcriptome, especially for LF-fed mice (Fig. 3A). There-fore, the distances between the overall expression profiles of thesix groups did reflect the physiology and lifespan differencesamong these intervention groups.

Differences in Midlife Liver Gene Expressions Are Predictive of Diet-Induced Lifespan Differences. The high correlation between life-span and PC2 (22.7% total variations) indicates that a largeportion of the observed transcriptional differences is associatedwith lifespan/aging differences introduced by the dietary inter-ventions. We, therefore, asked whether incremental changes ofgene expressions in the midlife liver under the six dietary regi-mens might not only reflect the pace of aging but could alsoidentify potential modulators of aging and longevity, which arenot readily identifiable through routine age-dependency analysis.We tested this idea first by determining whether known life-

span-modifying genes were enriched among those genes withexpression changes that are positively or negatively correlatedwith lifespan and other phenotypic differences among the sixdietary regimens. Previously identified lifespan-modifying geneswere obtained from the GenAge database (19), which we refer toas GenAge regulators, and are used as a gold-standard positivegene set. We used PCC to rank the genes’ expression (log2-transformed) correlation with phenotypes or the mean lifespanin each of the six dietary intervention groups. As shown in Fig.3B, genes with expression that showed the highest negativecorrelation with lifespan (measured by PCC between a gene’sexpression level and the mean lifespan across the six groups;Dataset S2) were two- to threefold more enriched for the ge-netically identified lifespan-modifying genes compared with thebackground genome-wide average (approximately twofold inthe top 200 and approximately threefold in the top 100 and top50). These enrichment levels (approximately threefold over thebackground) are much higher than the levels for differentiallyexpressed genes between any single pair of dietary regimens(approximately twofold on average) as determined and rankedby RankProd test percentage of false-positive values (20) (Fig.3B). This finding indicates that expression correlation or anti-correlation with lifespan is a better predictor for aging regulators.Because many of the phenotypes that we measured strongly cor-related with the mean lifespan in the six groups (Fig. 2), it is notsurprising that genes with expression that correlated positively/negatively with liver weight, liver TG, ALT, AST, mitochondriadensity, activity, and energy expenditure in the light cycle also hadhigher than background enrichment for aging regulators (Fig. 3B).Consistently, the combination of correlation to lifespan and otherphenotypes in general did not further increase the enrichment forGenAge lifespan regulators, except for lifespan correlation to-gether with liver weight or fat content (Fig. 3B). Lifespan-modi-fying genes from the mouse genome informatics (MGI) databasewere also overrepresented among genes that either negatively orpositively correlated with lifespan (SI Appendix, Fig. S5). Onedrawback of the MGI-curated gene set is that it includes genesfor which mutation shortens lifespan. Lifespan-shortening muta-tions may not affect aging per se, because nonspecific defects indevelopment or overall health of the animals can also reducelifespan.To determine how our dietary intervention-based tran-

scriptomic approach for predicting aging regulatory genes com-pares with a more traditional approach of using age-dependenttranscriptional changes, we examined the hepatic gene expres-sion changes in LF-fed mice during aging at 4, 8, 13, and 21 mo.At each age point, we pooled liver RNAs from six mice of thesame age to obtain an average microarray profile (Materials and

Methods). Using the Short Time Series Expression Miner soft-ware (21), we identified five expression patterns that were sig-nificantly age-dependent (P < 0.05) (SI Appendix, Fig. S6A).Genes that displayed these significant age-related expressionchanges were not significantly enriched for lifespan-modifyinggenes (Fig. 3B and SI Appendix, Fig. S6A) (P = 0.306, P = 0.438,P = 0.149, P = 0.184, and P = 0.048, respectively). Similar re-sults were obtained using another more comprehensive livergene expression dataset: the AGEMAP dataset (4) (SI Appendix,Fig. S6B). These results are consistent with our previous findingson the human brain and the fruit fly aging gene expressionprofiles (10), suggesting that most of these changes reflect theconsequences of aging rather than causal or regulatory mecha-nisms of aging.

Potential Pathways Contributing to Lifespan Alterations UnderDifferent Dietary Regimens. We next used Gene Set EnrichmentAnalysis (GSEA) (22) to identify pathways with gene expressionsthat are positively or negatively correlated with lifespan of the sixintervention groups (Materials and Methods). Unexpectedly,genes in these pathways were as high as eightfold enriched forGenAge lifespan regulators relative to the genomic background(Fig. 3 B and C), suggesting that the statistical power at thepathway level brings more enrichment for lifespan regulatorscompared with the single-gene level.Our approach of directly measuring lifespan as well as mea-

suring dietary and other metabolic conditions, although ex-tremely tedious and costly (180 mice over a time span of 3 y),allowed us to search for pathways that may directly explainlifespan changes instead of other phenotypic changes. For in-stance, although body weight is highly correlated with lifespan,we were able to identify pathways, such as steroid biosynthesis,biosynthesis of unsaturated fatty acids, primary immunodefi-ciency, primary bile acid biosynthesis, and acute myeloid leuke-mia pathways, etc., that are correlated with lifespan but indepen-dent of body weight (Fig. 3C). Furthermore, we also observedsome lifespan-associated pathways that are independent of mostother lifespan-associated phenotypes. For example, the primaryimmunodeficiency pathway is not associated with any phenotypesexamined other than lifespan, whereas the steroid synthesispathway is only negatively correlated with liver cholesterol level;also, the circadian rhythm pathway is only negatively correlatedwith body weight (Fig. 3C). Steroid hormones, such as preg-nenolone, can act independently or synergistically with insulin/IGF1 signaling to extend lifespan (23). Thus, increased expres-sion of steroid synthesis genes in response to CR under both HFand LF diet feeding conditions and the significant correlationwith lifespan under the six different treatments (Fig. 3C) may beother important components of the dietary effects on longevity.The relationship between circadian rhythm and aging has alsobeen under intense investigation (24).Pathways with gene expressions that were highly positively

correlated with lifespan across the six groups included age-as-sociated neurodegenerative disorders such as Huntington dis-ease, Alzheimer’s disease, and Parkinson disease. Also positivelycorrelated with lifespan were the known aging-related oxidativephosphorylation (25) and ribosome pathways (26). In fact, theribosome pathway showed the strongest positive correlation withlifespan and strongest negative correlation with many physio-logical parameters of liver damage (Fig. 3C).Interestingly, the number of pathways negatively correlated

with lifespan was three times the number of pathways positivelycorrelated with lifespan, although the number of genes positivelyand negatively correlated with lifespan was similar [586 geneswere positively correlated with the mean lifespan (PCC > 0.866or P < 0.05), whereas 429 genes were negatively correlated(PCC < −0.888 or P < 0.05)]. This finding suggests that genesnegatively correlating with lifespan are more functionally diverse

Zhou et al. PNAS Early Edition | 5 of 9

SYST

EMSBIOLO

GY

PNASPL

US

(Fig. 3C and Dataset S1). In addition, several pathways thatnegatively correlated with lifespan were related to inflammationand apoptosis, which have been previously implicated in agingand age-related diseases. Moreover, we also identified peroxi-some proliferator-activated receptor (PPAR) signaling and var-ious metabolic pathways, particularly those pathways involvedin fat metabolism, as negatively correlated with lifespan. Genesinside these pathways frequently displayed coordinated changesthat correlated with lifespan and other phenotypes [for example,within the peroxisome pathway (SI Appendix, Fig. S7), which isthe pathway that most negatively correlated with lifespan].

Validation of Lifespan Regulation by Peroxisome Biogenesis Genes.Fatty acids are activators for the nuclear receptor PPARα, whichpromotes peroxisome proliferation. During PPARα activation,peroxide-generating enzymes for fatty acid β-oxidation are in-duced 10- to 30-fold; the peroxide-catabolizing enzyme catalaseis induced only one- to twofold. Therefore, peroxisome pro-liferation may increase cellular levels of peroxide (27). Amongthe peroxisome genes, those genes involved in peroxisome bio-genesis were most homogeneously anticorrelated with lifespan(Fig. 4A and SI Appendix, Figs. S7 and S8 A and B), suggestingthat peroxisome biogenesis may negatively influence longevity.Consistent with this idea, we observed an overall increase in per-oxisomal gene expression in the aged brain, especially for genesthat participate in peroxisome biosynthesis and proliferation (thePEX genes), based on previously published microarray geneprofiles of 30 human brain samples (2) with (Fig. 4B) or withoutoutliers (SI Appendix, Fig. S9) as described previously (28).To test experimentally whether peroxisome biogenesis nega-

tively affects lifespan, we used two Drosophila melanogastermutants pex1S4868 and pex13KG04339 harboring mutations in thepromoter regions of pex1 and pex13 that resulted in loweredexpression levels of these genes (Fig. 4C, Inset). Compared with

the WT strains, both mutants had lifespan that increased by 16%in male flies and 13% in female flies (Fig. 4C).We also identified the Caenorhabditis elegans orthologs of

peroxisomal proliferation genes, with expression levels (as mea-sured by our microarray analysis) that showed negative correla-tion with lifespan, and tested the effect of RNAi knockdown oftheir expression on lifespan. Indeed, knockdown of four of eightgenes resulted in significantly increased lifespan relative to theempty vector control. In particular, knockdown of prx-13 orF18F11.1, which correspond to the mouse genes showing thestrongest negative correlation with the mean lifespan (Pex13 andPxmp4), resulted in significant lifespan extensions (17% and 15%longer mean lifespan relative to vector control, respectively) (Fig.4D). These results are also supported by a previous genome-widescreen in the enhanced RNAi strain eri-1(mg366) showing thatpostdevelopmental knockdown of prx-5, a C. elegans homolog formouse Pex5, increased the worm’s lifespan (29). Consistent withthis report, we observed an 8% extension of the mean lifespan ofN2 animals after prx-5 knockdown initiated on the first day ofadulthood (Fig. 4D). Interestingly, knockdown of these genesfrom hatching had no effect on lifespan or slightly shortenedlifespan (SI Appendix, Fig. S10A), indicating that these genes areimportant during development but may exert postdevelopmentaldetrimental effects on longevity.

Knockdown of Peroxisomal Genes Leads to Decreased CellularPeroxide Level and Increased Tolerance of Oxidative Stress. Wenext asked whether RNAi knockdown of PEX genes led to re-duction of peroxide in the animals or increased the tolerance ofthe animals to peroxide. Compared with the WT parental flies,both the homozygous pex1 and pex13 mutants (pex1S4868 andpex13KG04339) had reduced levels of hydrogen peroxide (Fig. 5A),which is the major metabolic end product from peroxisomes. Thereduction was stronger in males than females. The pex13mutants

A B

C D

Fig. 4. Peroxisomal pathway’s gene expression correlation with diet-induced lifespan changes is predictive of its regulation of lifespan. (A) The meanlifespan and expression levels of Pex1, Pex5, Pex13, and Pxmp4 among the six groups. (B) Expression level changes of the proxisomal proliferation genes (PEX)during human brain aging. Average and SD of the log2 expression levels of the 11 PEX genes in each of the human brain samples, excluding seven outliers.The outliers were identified as described previously (28). Analysis including the outliers can be found in SI Appendix, Fig. S8B. (C) Survival curves for male andfemale WT (red-eyed cs and white-eyedw1118 strains) and pex1S4868 and pex13KG04339 mutant flies (log-rank test: P = 0.0006 for male and P = 0.0075 for femalepex1 vs. cs; P = 7.59E-5 for male and P = 0.015 for female pex13 vs. cs; P = 1.41E-6 for male and P = 0.0021 for female pex1 vs. w1118; and P = 5.22E-9 for maleand P = 0.0053 for female pex13 vs. w1118). Insets indicate that mutations in pex1 and pex13 promoters reduced their expression levels to 20% of the ex-pression level in WT. (D) Survival curves of C. elegans fed with bacteria containing vector control (EV) or dsRNA constructs directed against daf-16(the negative control) or genes encoding the indicated peroxisomal biogenesis proteins. RNAi knockdown of prx-13, F18F11.1 (PXMP4 homolog), prx-11,and prx-5 significantly extended worm’s lifespan (log rank test: P = 2.3E-7, P = 3.5E-6, P = 0.002, and P = 0.009, respectively). All RNAi constructs wereconfirmed by sequencing.

6 of 9 | www.pnas.org/cgi/doi/10.1073/pnas.1119304109 Zhou et al.

showed more substantial reduction in peroxide than pex1mutants for either sex but particularly for males (Fig. 5A). Thisfinding is consistent with the relatively longer lifespan extensionin pex13 males (Fig. 4C). In C. elegans, compared with the vectorcontrol, RNAi knockdown of prx-5, prx-11, prx-13, or F18F11.1decreased the endogenous levels of reactive oxidative species(ROS) (Fig. 5B) and increased the stress tolerance to paraquat(Fig. 5C). In contrast, knockdown of PEX genes did not consistentlyincrease the resistance to heat shock (SI Appendix, Fig. S10B).Taken together, these findings show that reduced expression

of peroxisomal biogenesis genes during adulthood could de-crease cellular ROS levels, enhance resistance to ROS, and in-crease lifespan in both D. melanogaster and C. elegans. Thisfinding suggests that the negative correlation between the ex-pression of peroxisomal biogenesis genes and lifespan in micemay reflect an important, evolutionarily conserved role for per-oxisome biogenesis in longevity control.

DiscussionIn this study, we found that dietary interventions led to con-cordant changes in aging-related physical and physiologicalphenotypes. These changes were reflected by the midlife geneexpression differences under six different dietary regimens.The correlation between incremental changes of hepatic geneexpression and changes of the mean lifespans under theseconditions was predictive of lifespan-modifying/regulating genes.We verified this conclusion in two ways. First, we showed thatknown longevity-modifying genes were enriched among genes andpathways with expression that was correlated or anticorrelatedwith lifespan. Second, we predicted that high expression of per-oximal biogenesis genes might negatively influence lifespan, andwe verified this prediction experimentally in D. melanogaster andC. elegans, two different model organisms that are commonly usedin aging studies. Thus, our dietary perturbation-based method ofgene expression analysis has successfully identified peroxisomalbiogenesis factors as an important class of conserved longevity-modifying genes, providing insight into the molecular mechanismslinking diet, disease, and aging.Mitochondria and peroxisomes are two major organelles

generating ROS. Intriguingly, contrary to peroxisomal genes,mitochondrial genes were up-regulated by CR and down-regu-

lated by HF feeding, which is consistent with the changes in thetotal area of liver mitochondria under these conditions (Fig. 2and SI Appendix, Fig. S3D). Curiously, although SIRT1 andPGC1α (encoded by the Ppargc1a gene) are proposed to mediatethe response to Ex and CR through regulating mitochondrialbiogenesis (30, 31), we did not observe changes in the hepaticexpression of Sirt1 at midlife across the six different dietarygroups as detected by quantitative RT-PCR analysis (SI Appen-dix, Fig. S11); moreover, whereas both CR and Ex led to sig-nificantly increased liver expression of Ppargc1a in LF- but notHF-fed mice (SI Appendix, Fig. S11), these changes did notcorrelate with the lifespan differences. However, CR has beenshown to reduce ROS generation in mitochondria, perhaps bydecreasing proton leak (32). Equally important, reducing ROSfrom peroxisomes by suppressing peroxisome proliferation mightcontribute to lifespan extension by CR, whereas increased per-oxisome proliferation might accelerate aging in response to HFfeeding. Although speculative, this idea is supported by both theexperimental data in worms and flies and the highly negativecorrelation of peroxisome gene expression with lifespan underdifferent dietary regimens in mice. This finding is also supportedby the suppression of many aging-related metabolic phenotypes,including insulin resistance, in PPARα KO mice when fed a HFdiet (33). Intuitively, one might think that the negative correla-tion of PEX genes expression with lifespan would result frominduction of these genes in response to a HF diet, because fattyacids are activators for PPARα; however, in reality, it could bemostly attributed to the strong suppression of these genes underCR (both LF+CR and HF+CR) (Fig. 4A).Mitochondria and peroxisomes are not only linked in meta-

bolic reactions, such as fatty acid metabolism, but also shareessential proteins for biogenesis (34). A number of recent studieshave revealed the crucial role of mitochondrial biogenesis in CR-or diet-responsive regulation of longevity (13–15, 35–37), ROSgeneration, and aging-associated metabolic dysfunction (38–40).Therefore, it is interesting to note that, in our mouse model, thestrong anticorrelation of PEX genes expression with lifespan iscoupled with the strong positive correlation of mitochondrialgene expression, mitochondrial number, and size with lifespan.This finding suggests that the biogenesis of these two organellesmay be competitive and that this competitive relationship may becoupled to modulate lifespan in response to dietary cues.Now that large-scale genome-wide experiments can be rou-

tinely conducted, how to design a systems biology experiment toeffectively capture critical regulatory pathways for aging andlifespan is a pressing issue. In this study, we have shown that,consistent with the fact that dietary intervention is a powerfulapproach to modulating aging and lifespan in nearly all organismsstudied, analyzing gene expression changes in response to dietaryinterventions is an efficient way to identify aging and lifespanmodulators. Using such a design paradigm, additional experi-ments with more dietary conditions and/or measurements at thelevel of individuals may help elucidate the complex networks ofaging and lifespan control under various nutritional conditions.Such an approach also offers the potential to define mechanismsthat underlie individual variations in response to environmental/dietary conditions, and it may provide an avenue to developingpersonalized dietary regimens to optimize health span.

Materials and MethodsSome of the experimental procedures and materials for animal studies andbioinformatic analyses are described in SI Appendix.

Dietary Interventions in Mice. Male C57BL/6J mice at 4 wk of age were pur-chased from Shanghai Animal Co, Ltd. Mice were maintained under a 12-hdark/light cycle (lights on at 6:30 AM) at a temperature of 22 °C ± 3 °C inaccredited animal facilities. All experimental procedures and protocols wereapproved by the Institutional Animal Care and Use Committee of the In-

A

B C

Fig. 5. RNAi knockdown of peroxisome biogenesis genes decreases en-dogenous peroxide level and increases oxidative stress tolerance. (A) Re-duced H2O2 levels in pex1 and pex13 mutant flies. **Significant (P < 0.01)changes compared with the corresponding WT flies (w1118). (B and C) De-creased endogenous ROS levels (B) as measured by DCF-DA (Materials andMethods) and enhanced resistance to paraquat (C) in worms with RNAiknockdown of prx-5, prx-11, prx-13, or F18F11.1. Asterisks indicate significanceof difference compared with the EV control worms (*P < 0.05, **P < 0.01).

Zhou et al. PNAS Early Edition | 7 of 9

SYST

EMSBIOLO

GY

PNASPL

US

stitute for Nutritional Sciences, Chinese Academy of Sciences. Before thestart of the experiment, mice were maintained on an LF diet (Research Diets)for 1 wk. At the age of 5 wk, animals were randomly assigned to one of thesix intervention groups (n = 30 for each group): feeding of an LF diet (10%fat, D12450B; Research Diets) ad libitum or with 30% CR (LF+CR) or volun-tary running Ex (LF+Ex) or feeding of a HF diet (60% fat, D12492l; ResearchDiets) ad libitum or with 30% CR (HF+CR) or voluntary running Ex (HF+Ex).All mice were housed individually during the study. The daily consumptionof food in LF and HF groups was recorded over 1 wk and averaged to de-termine the amount of food for the following week for the LF+CR andHF+CR groups, respectively. After 1 wk acclimation in cage with the lockedrunning wheels, mice in the LF+Ex and HF+Ex groups were allowed freeaccess to a running wheel, and the running distance and time were recordedautomatically by the equipment.

Determination of Lifespan in Mice. After eight mice were randomly selectedfrom each group for analyses of midlife metabolic phenotypes at 62 wk ofage, survival was assessed from the remaining animals (n = 22 per group),which were under close surveillance daily for death recording. Survivalcurves were plotted using the Kaplan–Meier method, and lifespan differ-ences between groups were evaluated by the log-rank test. Maximum life-span was calculated as the mean age of the oldest 20% of mice withineach group.

Statistical Analysis. All of the physiological data for animal interventionstudies were presented as means ± SEM and analyzed by two-way ANOVA.Differences were considered statistically significant when P < 0.05.

Microarray Analysis. The hepatic transcriptional level for threemice from eachintervention group at 62 wk of age was analyzed using Affymetrix MouseGenome 430 2.0 Arrays. The raw data were log2-transformed and normalizedusing the affy package in bioconductor. Differentially expressed genes be-tween different groups were determined by RankProd (41) with proportionof false positive (pfp) < 0.1. PCA and hierarchical clustering based on thedifferentially expressed genes were performed to evaluate the relationshipsof those samples with different interventions.

Aging Regulatory Genes.We compiled a gold-standard set of aging regulatorygenes from three sources: 264 human and mouse lifespan or aging-relatedgenes in GenAge, 319 genes annotated with lifespan or aging (MP:0005372)in MGI, and another 7 genes by literature survey as described (10).

Enrichment of KEGG Pathways. KEGG pathways were downloaded from theKEGG database on October 9, 2009. The significance of enrichment wascalculated using the GSEA program (22).

Peroxisome Functional Categories. Peroxisome functional genes and catego-ries were downloaded from http://www.peroxisomedb.org on January7, 2010.

Drosophila Stocks and Lifespan Analysis. pex1S4868 and pex13KG04339 mutants(42) were obtained from the Bloomington Stock Center and backcrossed sixtimes to remove background differences. Flies were reared on standard cornmeal at 25 °C. Both males and females were collected within 24 h of eclosionand randomly allocated to glass vials at a density of 20 flies per vial and 10vials per genotype (n = 200). Flies were transferred to fresh vials every 3 d,and the number of dead was recorded.

Measurement of H2O2 Level. pex1S4868, pex13KG04339, and w1118

flies (42) werecultured on standard corn meal at 25 °C. Adult males and females were

separated at 1 d after eclosion. At 7–8 d after eclosion, male and female fliesof each genotype were divided into four groups with 10 flies in each group.Flies in each group were then homogenized together, and H2O2 level wasdetermined using a detection kit from Beyotime (S0038) following themanufacturer’s instructions.

C. elegans Strains and RNAi. C. elegans strains were maintained at 20 °C asdescribed (43) unless otherwise noted. The worm strains N2, daf-2(e1370),and daf-16(mu86) were obtained from the Caenorhabditis Genetics Center.RNAi was performed essentially as described previously (10, 44) withAhringer’s RNAi feeding library (45).

C. elegans Lifespan Analysis. The lifespan of C. elegans strains was determinedas described previously (10) with minor modifications. Briefly, worms werecultured at 20 °C for two or more generations in the standard nematodegrowth medium (NGM) plates with E. coli OP50 before the assays. Adultworms at the beginning of egg laying (day 0) were transferred to thestandard RNAi NGM plates with RNAi bacteria. At least 100 gravid wormsper treatment in three plates were moved to fresh plates every 2–3 d.Worms that bagged, exploded, or crawled off the plate were excluded fromthe analysis. The significance of survival curves were calculated by log-ranktest using the Survival package in R (http://www.r-project.org/).

Measurement of Endogenous ROS Level. The endogenous ROS level in C.elegans was measured using 2′,7′-dichlorofluorescin diacetate (DCF-DA,D6883; Sigma) as described previously (46) with modifications. About 1,000adult worms (day 4) were collected in M9 buffer and washed more thanthree times to eliminate bacteria. Then, the worms were washed one time inPBS, transferred to a 1.5-mL tube, and immediately frozen in liquid nitrogen.After thawing at room temperature, the worms were broken up by soni-cation (Misonix). Supernatants were collected after centrifugation (Eppen-dorf F-45-24-11 rotor; 12,000 rpm at 4 °C for 5 min) and transferred to newtubes. The supernatant containing 5 μg protein was incubated with 10 μL100 μM DCF-DA in PBS at room temperature for 10 min before centrifuga-tion. Fluorescence intensity of the supernatant of the mixture was measuredusing an Agilent Stratagene MX3000P at the standard detection range forSYBR Green. Six measurements were performed for each sample at 5-minintervals to examine the DCF-DA fluorescence signal linearity as previouslydescribed (46). The ROS level was determined by the fluorescence intensityat the last time point after subtracting the background fluorescencedetected without the supernatant.

Oxidative Stress Resistance Assays. About 200 adult worms (day 4) per groupfed RNAi bacteria were collected in 300 μL M9 buffer and washed more thanthree times. About 50 worms were transferred to the 24-well plates with 500μL 0.4 M paraquat (Sigma) in M9 per well. Survival was determined after 8 hof treatment.

ACKNOWLEDGMENTS. We thank Prof. Hong Zhang from the NationalInstitute for Biological Sciences and Dangsheng Li from Cell Research forinvaluable suggestions. We also thank Christopher B. Newgard (DukeUniversity) for initial design of the intervention study in mice. This workwas supported by Chinese Ministry of Science and Technology Grants2011CB910900 (to Y.L.), 2011CB504206 (to J.-D.J.H.), and 973 Program2012CB524900; National Natural Science Foundation of China Grants81021002 (to Y.L.), 30988002 (to Y.L.), 30830033 (to Y.L.), 30890033 (toJ.-D.J.H.), and 91019019 (to J.-D.J.H.); and Chinese Academy of SciencesGrants KSCX2-EW-R-09 (to Y.L.), KSCX2-EW-R-02 (to J.-D.J.H.), KSCX2-EW-J-15 (to J.-D.J.H.), and XDA01010303 (to J.-D.J.H.). M.K. is an Ellison MedicalFoundation New Scholar in Aging.

1. Khaitovich P, et al. (2004) Regional patterns of gene expression in human and

chimpanzee brains. Genome Res 14:1462–1473.2. Lu T, et al. (2004) Gene regulation and DNA damage in the ageing human brain.

Nature 429:883–891.3. Somel M, et al. (2009) Transcriptional neoteny in the human brain. Proc Natl Acad Sci

USA 106:5743–5748.4. Zahn JM, et al. (2007) AGEMAP: A gene expression database for aging in mice. PLoS

Genet 3:e201.5. Zahn JM, et al. (2006) Transcriptional profiling of aging in human muscle reveals

a common aging signature. PLoS Genet 2:e115.6. Pletcher SD, et al. (2002) Genome-wide transcript profiles in aging and calorically

restricted Drosophila melanogaster. Curr Biol 12:712–723.7. Lund J, et al. (2002) Transcriptional profile of aging in C. elegans. Curr Biol 12:1566–1573.

8. McCarroll SA, et al. (2004) Comparing genomic expression patterns across species

identifies shared transcriptional profile in aging. Nat Genet 36:197–204.9. Kim SK (2007) Common aging pathways in worms, flies, mice and humans. J Exp Biol

210:1607–1612.10. Xue H, et al. (2007) A modular network model of aging. Mol Syst Biol 3:147.11. Fontana L, Partridge L, Longo VD (2010) Extending healthy life span—from yeast to

humans. Science 328:321–326.12. Kenyon C (2005) The plasticity of aging: Insights from long-lived mutants. Cell 120:

449–460.13. Baur JA, et al. (2006) Resveratrol improves health and survival of mice on a high-

calorie diet. Nature 444:337–342.14. Milne JC, et al. (2007) Small molecule activators of SIRT1 as therapeutics for the

treatment of type 2 diabetes. Nature 450:712–716.

8 of 9 | www.pnas.org/cgi/doi/10.1073/pnas.1119304109 Zhou et al.

15. Pearson KJ, et al. (2008) Resveratrol delays age-related deterioration and mimics

transcriptional aspects of dietary restriction without extending life span. Cell Metab

8:157–168.16. Goodrick CL, Ingram DK, Reynolds MA, Freeman JR, Cider NL (1983) Differential

effects of intermittent feeding and voluntary exercise on body weight and lifespan in

adult rats. J Gerontol 38:36–45.17. Hietakangas V, Cohen SM (2009) Regulation of tissue growth through nutrient

sensing. Annu Rev Genet 43:389–410.18. Kleiber M (1947) Body size and metabolic rate. Physiol Rev 27:511–541.19. de Magalhães JP, Costa J, Toussaint O (2005) HAGR: The Human Ageing Genomic

Resources. Nucleic Acids Res 33:D537–D543.20. Breitling R, Armengaud P, Amtmann A, Herzyk P (2004) Rank products: A simple, yet

powerful, new method to detect differentially regulated genes in replicated

microarray experiments. FEBS Lett 573:83–92.21. Ernst J, Bar-Joseph Z (2006) STEM: A tool for the analysis of short time series gene

expression data. BMC Bioinformatics 7:191.22. Subramanian A, et al. (2005) Gene set enrichment analysis: A knowledge-based

approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA

102:15545–15550.23. Russell SJ, Kahn CR (2007) Endocrine regulation of ageing. Nat Rev Mol Cell Biol 8:

681–691.24. Schibler U (2005) The daily rhythms of genes, cells and organs. Biological clocks and

circadian timing in cells. EMBO Rep 6:S9–S13.25. Lee SS, et al. (2003) A systematic RNAi screen identifies a critical role for mitochondria

in C. elegans longevity. Nat Genet 33:40–48.26. Steffen KK, et al. (2008) Yeast life span extension by depletion of 60s ribosomal

subunits is mediated by Gcn4. Cell 133:292–302.27. Schrader M, Fahimi HD (2004) Mammalian peroxisomes and reactive oxygen species.

Histochem Cell Biol 122:383–393.28. Jin C, et al. (2011) Histone demethylase UTX-1 regulates C. elegans life span by

targeting the insulin/IGF-1 signaling pathway. Cell Metab 14:161–172.29. Curran SP, Ruvkun G (2007) Lifespan regulation by evolutionarily conserved genes

essential for viability. PLoS Genet 3:e56.30. Cantó C, Auwerx J (2009) Caloric restriction, SIRT1 and longevity. Trends Endocrinol

Metab 20:325–331.

31. Gurd BJ (2011) Deacetylation of PGC-1α by SIRT1: Importance for skeletal musclefunction and exercise-induced mitochondrial biogenesis. Appl Physiol Nutr Metab 36:589–597.

32. Guarente LP, Partridge L, Wallace DC (2007) Molecular Biology of Aging (Cold SpringHarbor Laboratory Press, Plainview, NY), pp 409–425.

33. Cha DR, et al. (2007) Peroxisome proliferator-activated receptor-alpha deficiencyprotects aged mice from insulin resistance induced by high-fat diet. Am J Nephrol 27:479–482.

34. Delille HK, Alves R, Schrader M (2009) Biogenesis of peroxisomes and mitochondria:Linked by division. Histochem Cell Biol 131:441–446.

35. D’Antona G, et al. (2010) Branched-chain amino acid supplementation promotessurvival and supports cardiac and skeletal muscle mitochondrial biogenesis in middle-aged mice. Cell Metab 12:362–372.

36. Nisoli E, et al. (2005) Calorie restriction promotes mitochondrial biogenesis byinducing the expression of eNOS. Science 310:314–317.

37. López-Lluch G, et al. (2006) Calorie restriction induces mitochondrial biogenesis andbioenergetic efficiency. Proc Natl Acad Sci USA 103:1768–1773.

38. Hagopian K, et al. (2011) Caloric restriction influences hydrogen peroxide generationin mitochondrial sub-populations frommouse liver. J Bioenerg Biomembr 43:227–236.

39. Hagopian K, et al. (2005) Long-term calorie restriction reduces proton leak andhydrogen peroxide production in liver mitochondria. Am J Physiol Endocrinol Metab288:E674–E684.

40. Civitarese AE, et al. (2007) Calorie restriction increases muscle mitochondrialbiogenesis in healthy humans. PLoS Med 4:e76.

41. Hong F, et al. (2006) RankProd: A bioconductor package for detecting differentiallyexpressed genes in meta-analysis. Bioinformatics 22:2825–2827.

42. Chen H, Liu Z, Huang X (2010) Drosophila models of peroxisomal biogenesis disorder:Peroxins are required for spermatogenesis and very-long-chain fatty acid metabolism.Hum Mol Genet 19:494–505.

43. Brenner S (1974) The genetics of Caenorhabditis elegans. Genetics 77:71–94.44. Kamath RS, Martinez-Campos M, Zipperlen P, Fraser AG, Ahringer J (2001)

Effectiveness of specific RNA-mediated interference through ingested double-stranded RNA in Caenorhabditis elegans. Genome Biol 2:RESEARCH0002.

45. Kamath RS, et al. (2003) Systematic functional analysis of the Caenorhabditis elegansgenome using RNAi. Nature 421:231–237.

46. Lee SJ, Hwang AB, Kenyon C (2010) Inhibition of respiration extends C. elegans lifespan via reactive oxygen species that increase HIF-1 activity. Curr Biol 20:2131–2136.

Zhou et al. PNAS Early Edition | 9 of 9

SYST

EMSBIOLO

GY

PNASPL

US

1

Supporting Information Appendix

Midlife gene expressions identify modulators of aging through dietary interventions

Bing Zhoua,b,c,1, Liu Yangd,1, Shoufeng Lid, Jialiang Huanga,b,c, Haiyang Chene, Lei Houa,b,c, Jinbo Wangc,e, Christopher D. Greena, Zhen Yanf, Xun Huange, Matt Kaeberleing, Li Zhuh, Huasheng Xiaoh, Yong Liud,2 and Jing-Dong J. Hana,2

aChinese Academy of Sciences Key Laboratory for Computational Biology, Chinese Academy of Sciences—Max Planck Partner Institute for Computational Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai 200031, China; bCenter for Molecular Systems Biology, Institute of Genetics and Developmental Biology, Chinese Academy of Sciences, Beijing 100101, China; cGraduate University, Chinese Academy of Sciences, Beijing 100049, China; dKey Laboratory of Nutrition and Metabolism, Institute for Nutritional Sciences, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai 200031, China; eState Key Laboratory of Molecular Developmental Biology, Institute of Genetics and Developmental Biology, Chinese Academy of Sciences, Beijing 100101, China; fDepartments of Medicine-Cardiovascular Medicine and Pharmacology and Center for Skeletal Muscle Research, Robert M. Berne Cardiovascular Research Center, University of Virginia, Charlottesville, VA 22908; gDepartment of Pathology, University of Washington, Seattle, WA 98195; and hKey Laboratory of Systems Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai 200031, China

1B.Z. and L.Y. contributed equally to this work. 2To whom correspondence may be addressed. E-mail: liuy@sibs.ac.cn or jdhan@genetics.ac.cn.

2

Table of Contents Supplemental Materials and Methods .................................................................................................. 2

Tightrope, rotarod and hair re-growth assays .............................................................................. 2 Metabolic and serum measurements of mice ................................................................................ 2 Body temperature .......................................................................................................................... 3 Liver physiological analysis.......................................................................................................... 3 Identification of aging patterns ..................................................................................................... 4 Visualization of the KEGG peroxisome pathway .......................................................................... 4 RT-PCR analyses ........................................................................................................................... 4 Heat stress resistance assays ........................................................................................................ 4 Supplemental References .............................................................................................................. 5

Supplemental Figures ........................................................................................................................... 6 Supplemental Tables .......................................................................................................................... 17

Supplemental Materials and Methods

Tightrope, rotarod and hair re-growth assays

For tightrope test, mice were placed on the middle of a bar of circular section (60 cm long; 3 cm in diameter). The test was considered successful when a mouse passed the tightrope during a period of 60 sec in at least one out of five consecutive trials. For rotarod test, each mouse was pre-trained at a constant speed (3 rpm) until it was able to remain on the apparatus for 60 sec. On the following day, each mouse was given three trials, with a 30-min rest period in between. During the test, the rotarod started at 3 rpm and accelerated to 24 rpm over a period of 9 min, the maximum length of each trial. For hair re-growth analysis, dorsal hair was removed by plucking from a square of approximately 1.5 cm×1.5 cm. Hair re-growth was scored three weeks later blindly by two other investigators from digital photographs and a semi-quantitative assessment was done using an arbitrary scale from one to four, where four represents complete hair regeneration.

Metabolic and serum measurements of mice

Total body fat content was measured by nuclear magnetic resonance (NMR) using the Minispec Mq7.5 (Bruker, Germany). Oxygen consumption and physical activity were measured using the comprehensive laboratory animal monitoring system (CLAMS, Columbus Instruments, Columbus, Ohio, USA) according to the manufacturers’ instructions. Mice were allowed for acclimation to the CLAMS for 16-20 hours, and measurements were conducted for the following 24 hours. Voluntary activity was monitored from the x-axis beam breaks recorded every 15 min.

Glucose tolerance test was performed in mice after overnight fasting. Mice were injected

3

intraperitoneally with glucose at 1g/kg, and blood glucose concentrations were measured at 15, 30, 45, 60, and 120 min after insulin injection using a glucometer (FreeStyle, Alameda, CA).

Blood samples were collected from tail bleeding after overnight fasting in mice at the desired ages. After centrifugation at 12,000 rpm for 30 min to pellet blood cells, serum was transferred to a new tube and stored at -80 °C until further use. The serum levels of insulin and leptin were quantified by the Bioplex Suspension Array System (Bio-Rad) using a mouse cytokine immunoassay panel (Linco Research, St. Charles, MO). Serum triglyceride and cholesterol were respectively determined using the Serum Triglyceride Determination Kit (Sigma, St. Louis, MO) and Amplex Red Cholesterol Assay Kit (Molecular Probes, Eugene, OR).

Body temperature

The rectal body temperature was measured for mice at 90 weeks of age. The measurements were performed between 8:00 a.m. and 10:00 a.m. after an overnight fast using microprobe thermometer (Physitemp Instruments, Clifton, New Jersey, USA).

Liver physiological analysis

For midlife metabolic phenotype analyses, 8 mice at 62 weeks of age were randomly selected from each group and were humanely euthanized for blood and tissue collection. To analyze the liver function, serum levels of alanine transaminase (ALT) and aspartate transaminase (AST) were determined by the Alanine/Aspartate Transaminase Detection Kit (ShenSuoYouFu, Shanghai, China). To measure hepatic triglyceride and cholesterol levels, 40-50 mg of liver tissue were homogenized in 1.5 ml of CHCl3-CH3OH (2:1, v/v), followed by shaking at room temperature for 2 hr. After addition of 0.5 ml of 0.1 M NaCl, the suspension was centrifuged at 3,700 rpm for 10 min at room temperature. The lower organic phase was then transferred and air-dried overnight. The residual liquid was subsequently resuspended in 400 µl of 1% Triton X-100 in absolute ethanol, and the concentrations of triglyceride and cholesterol were determined using the Serum Triglyceride Determination Kit (Sigma, St. Louis, MO) and Amplex Red Cholesterol Assay Kit (Molecular Probes, Eugene, OR), respectively. For histology analysis, liver tissue specimens were fixed in 10% neutral formalin, and paraffin-embedded tissue sections were stained with hematoxylin-eosin (H&E) or picrosirius red (Sigma-Aldrich, St. Louis, MO). Three whole sections (4× magnifications) from each animal were examined by BX61 fluorescence microscope (Olympus, Japan) and images were quantified using NIH ImageJ software (http://rsb.info.nih.gov/ij/). For measurement of the mitochondria, liver tissues were processed as described (1). Briefly, livers that were cut into 1-mm3 sections were subsequently put into the electron microscopy fixative buffer, embedded in spur resin and sectioned. Thin sections were obtained and viewed under transmission electron microscope (H-7650, HITACHI) equipped with a charge-coupled device camera (ER-B; AMT). Mitochondrial number was counted from images at 1200X and normalized to the hepatocyte size. The sizes of hepatocytes and mitochondria were quantified from images at 6000X by NIH ImageJ software.

4

Identification of aging patterns

Liver RNA samples from six mice at 4, 8, 13 or 21 months of age were respectively pooled and subjected to whole-genome microarray analyses following the same protocol as for the 18 mice from the intervention study. The microarray data were likewise log2-transformed and normalized using 'affy' package in bioconductor. The significance of age-related patterns was determined by STEM (2) (P < 0.01).

Visualization of the KEGG peroxisome pathway

KEGG peroxisome pathway was colored using the pathway visualization tool in KEGG (http://www.genome.jp/kegg/tool/color_pathway.html).

RT-PCR analyses

Total mouse liver RNA was isolated using TRIzol reagent (Invitrogen, Carlsbad, CA). After treatment with RNase-free DNase I (Roche Applied Science, Penzberg, Germany) to eliminate possible DNA contamination, first-strand cDNA was synthesized with M-MLV reverse transcriptase and random hexamer primers (Invitrogen). Real-time quantitative PCR was conducted with the SYBR Green PCR system (Applied Biosystems, Foster City, CA), using cyclophilin as an internal control for normalization. The oligonucleotide primers used for each target gene were as follows: mouse Pex1, forward primer 5’-AGCTTGGTGGCACTCATCG-3’ and reverse primer 5’-GTTCCGGATTGGGAGGCT-3’; mouse Pex5, forward primer 5’-AATGCAACTCTTGTATCCCGAG-3’ and reverse primer 5’-GGCGAAAGTTTGACTGTTCAATC-3’; mouse Pex13, forward primer 5’-GATTTTGTTGCCGTGTCTGATG-3’ and reverse primer 5’-TTTGACCGTCAAGACTAGCCAG-3’; mouse Pxmp4, forward primer 5’-TGTCTATGGAGTCAAAATCCGGG-3’ and reverse primer 5’-AGAGTGGATGTACGTGGCTTT-3’; mouse Sirt1, forward primer 5’-TGAGCTGGATGATATGACGC-3’ and reverse primer 5’- GGAAGTCCACCGCAAGG -3’； mouse Ppargc1a, forward primer 5’-TATGGAGTGACATAGAGTGTGCT-3’ and reverse primer 5’- CCACTTCAATCCACCCAGAAAG-3’； mouse cyclophilin, forward primer 5'-ATGGCAAATGCTGGACCAAA-3' and reverse primer 5'-CATGCCTTCTTTCACCTTCCC-3';

Heat stress resistance assays

Heat stress assays in C. elegans were performed as described previously (3) with minor modifications. About 100 adult worms (day 4) per group were transferred to 3 new RNAi NGM plates with RNAi bacteria. The number of dead worms on each plate was scored every 2 hours for the first 8 hours after shifting to 35oC. Survival was determined as described for lifespan analysis.

5

Supplemental References

1. Baur JA, et al. (2006) Resveratrol improves health and survival of mice on a high-calorie diet. Nature 444(7117):337-342.

2. Ernst J & Bar-Joseph Z (2006) STEM: a tool for the analysis of short time series gene expression data. BMC Bioinformatics 7:191.

3. Chen D, Thomas EL, & Kapahi P (2009) HIF-1 modulates dietary restriction-mediated lifespan extension via IRE-1 in Caenorhabditis elegans. PLoS Genet 5(5):e1000486.

A6

4

2

00 20 40 60 80 100 120

LF+ExHF+Ex

Age (weeks)

Dis

tanc

e (k

m/d

)

BLFLF+ExLF+CRHFHF+ExHF+CR

12 weeks

0 30 60 90 120Time (min)

21 weeks

30 weeks

60 weeks

85 weeks

500

400

300

200

100

0

500

400

300

200

100

0

500

400

300

200

100

0

Glu

cose

(mg/

dl)

20

15

10

5

0

++

12 weeks

20

15

10

5

0

++**

**

++++

30 weeks

20

15

10

5

0

**

60 weeks

20

15

10

5

0

**

85 weeks

GTT

AU

C

20

15

10

5

0

++

**

**

++

21 weeks

GTT

AU

CG

TT A

UC

GTT

AU

CG

TT A

UC

LF+C

RHF+

CRLF

+Ex LF

HF+Ex HF

500

400

300

200

100

0

500

400

300

200

100

0

Glu

cose

(mg/

dl)

Glu

cose

(mg/

dl)

Glu

cose

(mg/

dl)

Glu

cose

(mg/

dl)

6

Supplemental Figures

Figure S1. Exercise monitoring and glucose tolerance of mice. (A) Voluntary running distancewas recorded and the average running distance was plotted biweekly for mice fed LF or HF in theexercise groups (n>6). (B) Glucose tolerance test (GTT) was performed at the indicated ages foreach intervention group (n=8-10 per group). In the left panels, the blood glucose levels are plottedagainst the times after i.p. injection of 1 g/kg glucose. Shown in the right panels are the calculatedareas under the curve (AUC) of blood glucose, presented as mean ± SEM, **P < 0.01 versus LF, and++P < 0.01 versus HF tested by ANOVAs.

+

LF+CR

HF+CRLF+Ex LF

HF+ExHF

Hai

r Reg

ener

ated

Area

(a.u

.)3.0

2.0

1.0

0

DC

B

Late

ncy

to F

all (

s)

600

400

200

0

++

**

+

**

++

LF+CR

HF+CRLF+Ex LF

HF+ExHF

Activ

ity

1400

1200

1000

800

600

400

200

0

**

Light CycleDark Cycle

LF+CR

HF+CRLF+Ex LF

HF+ExHF

Tigh

trope

Suc

cess

(%)

100

80

60

40

20

0

**

++

LF+CR

HF+CRLF+Ex LF

HF+ExHF

A

E

7

38

36

34

32

30

*+

Rec

tal T

empe

ratu

re (o C

)

LF+CR

HF+CRLF+Ex LF

HF+ExHF

Figure S2. Aging-sensitive markers, midlife physical activity and body temperature of mice.(A-C) Aging-sensitive markers. (A) Hair re-growth assay for mice at 90 weeks of age (n=9-12 pergroup). Hair re-growth capacity of the dorsal skin was scored at 21 days after shaving andquantified in arbitrary units (a.u.). (B) Rotarod tests for mice at 85 weeks of age. Shown are themaximal time to fall from the accelerating rotarod, averaged from three trials per mouse (n=11-12per group). (C) Tightrope test for mice at 85 weeks of age. The percentage of mice passing the testis shown for each group (n=8-12 per group). (D) Physical activity during the light and dark cycleswas determined by CLAMS for mice at 57-58 weeks of age (n=5 per group). (E) Rectaltemperatures of mice at 90 weeks of age (n=10 per group). Data are presented as mean ± SEM, *P <0.05, **P < 0.01 versus LF, and +P < 0.05, ++P < 0.01 versus HF tested by ANOVAs.

DensitySize

Mito

chon

dria

Den

sity

/Siz

e(R

elat

ive

to L

F)

2.0

1.5

1.0

0.5

0.0

*

+

**

LF+CRHF+CR

LF+Ex LFHF+Ex HF

D

HF HF+CR HF+Ex

LF LF+CR LF+Ex

**

++++++

Fib

rotic

Are

a(R

elat

ive

to L

F)

12

10

8

6

4

2

0

LF+CRHF+CR

LF+Ex LFHF+Ex HF

A 200

150

100

50

0

Enzy

me

Activ

ity (U

/l)

ALTAST

**

**

++

++++

++

++

++

LF+CRHF+CR

LF+Ex LFHF+Ex HF

HF HF+CR HF+Ex

LF LF+CR LF+ExC

B

HF HF+CR HF+Ex

LF LF+CR LF+Ex

8

Figure S3. Liver functions affected by the diet interventions. Liver functions were measured inmice (n=8 per group) that were sacrificed at 62 weeks of age. (A) Liver function was assessed bythe serum abundance of alanine aminotransferase (ALT) and aspartate aminotransferase (AST). (B)Representative images of liver sections of mice of each indicated group (n=4-5 per group) that weresubjected to hematoxylin and eosin staining. Original magnification: 20X. (C) Evaluation of liverfibrosis by Sirius-Red staining for collagens. Representative images are shown. Originalmagnification, 20X. The extent of fibrosis was quantified as the percentage of Sirius-Red positivearea (n=4 per group). (D) Mitochondrial density and size. Liver sections were analyzed bytransmission electron microscopy. Shown are representative images at a magnification of 6,000.Mitochondrial densities were determined by normalizing the counted number of mitochondria to thearea of each randomly selected cell (n=30 cells for each group). Quantification of mitochondrialsize was performed for 4 mice from each group. Data are shown as mean ± SEM, *P < 0.05, **P <0.01 versus LF, and +P < 0.05, ++P < 0.01 versus HF by ANOVAs.

VC12

VE20

VE4

VE18

VC20

VC27

VR4

VR17

VR27

NC20

NC18

NE12

NE4

NC29

NE10

NR18

NR12

NR29

0.00

0.02

0.04

0.06

0.08

1-PC

C

9

Figure S4. Hierarchical clustering of the samples based on the differentially expressed genes(Methods). The sample prefix ‘VC’ represents HF group, ‘VE’ HF+Ex, ‘VR’ HF+CR, ‘NC’ LF, ‘NE’LF+Ex and ‘NR’ LF+CR, respectively.

GenAge + MGI (Aging)

Aging Liver

AGEMAP

HF+CR vs HF

HF+Ex vs HF

HF vs LF

LF+Ex vs LF

LF+CR vs LF

Life Span(KEGG Pathways)

Life Span(Life Span)

Life Span(Body Weight)

Life Span(Liver Weight)

Life Span(Liver TG)

Life Span(Liver Cholesterol)

Life Span(Liver Fibrosis)

Life Span(Fat Content)

Life Span(ALT)

Life Span(AST)

Life Span(Mitochondria Density)

Life Span(Mitochondria Area)

Life Span(Energy Intake)

Life Span(Energy Exp Dark)

Life Span(Energy Exp Light)

Life Span(Activity Dark)

Life Span(Activity Light)

Body Weight(Life Span)

Liver Weight(Life Span)

Liver TG(Life Span)

Liver Cholesterol(Life Span)

Liver Fibrosis(Life Span)

Fat Content(Life Span)

ALT(Life Span)

AST(Life Span)

Mitochondria Density(Life Span)

Mitochondria Area(Life Span)

Energy Intake(Life Span)

Energy Exp Dark(Life Span)

Energy Exp Light(Life Span)

Activity Dark(Life Span)

Activity Light(Life Span)

0 1 2 3 4 5 6

MGI (Aging)

Aging Liver

AGEMAP

HF+CR vs HF

HF+Ex vs HF

HF vs LF

LF+Ex vs LF

LF+CR vs LF

Life Span(KEGG Pathways)

Life Span(Life Span)

Life Span(Body Weight)

Life Span(Liver Weight)

Life Span(Liver TG)

Life Span(Liver Cholesterol)

Life Span(Liver Fibrosis)

Life Span(Fat Content)

Life Span(ALT)

Life Span(AST)

Life Span(Mitochondria Density)

Life Span(Mitochondria Area)

Life Span(Energy Intake)

Life Span(Energy Exp Dark)

Life Span(Energy Exp Light)

Life Span(Activity Dark)

Life Span(Activity Light)

Body Weight(Life Span)

Liver Weight(Life Span)

Liver TG(Life Span)

Liver Cholesterol(Life Span)

Liver Fibrosis(Life Span)

Fat Content(Life Span)

ALT(Life Span)

AST(Life Span)

Mitochondria Density(Life Span)

Mitochondria Area(Life Span)

Energy Intake(Life Span)

Energy Exp Dark(Life Span)

Energy Exp Light(Life Span)

Activity Dark(Life Span)

Activity Light(Life Span)

0 1 2 3 4 5 6 7

TOP 50 (+)TOP 100 (+)TOP 200 (+)TOP 50 (-)TOP 100 (-)TOP 200 (-)

A B

10

Figure S5. Fold of enrichments over the background of lifespan correlated or anti-correlated genes,significantly differentially expressed genes between any single pair of treatments, lifespan/agingrelated genes in AGEMAP, and our aging liver time-course dataset on (A) MGI aging dataset and(B) combination datasets of GenAge, MGI aging and reviews recruited manually.

-1 0 1 2 3

Normalized Expression Value

4M 8M 13M 21M

Aging Pattern

A

B

B1M

C1M

D1M

E1M

B6M

C6M

D6M

E6M

B16

M

C16

M

D16

M

E16

M

B24

M

C24

M

D24

M

E24

M

Normalized Expression Value

-2 0 2Aging Pattern

11

Figure S6. Hierarchical clustering of the expression profiles of age-related genes based onmicroarray profiles from mice at 4, 8, 13 and 21 months of age (A), or based on AGEMAP liverexpression profiles (B) constituting the significant age-dependent gene expression patternsdetermined by STEM (P < 0.01).

Peroxisome Biogenesis

Membrane Protein Import

peroxisomal membrane protein

pre-peroxisomal vesicles

endoplasmic reticulum

Receptor Recycling

Matrix Protein Import

ROS Metabolism

Peroxisomal division

peroxisomal targeting sequence (PTS)

adenin nucleotidetransporter

matrix proteins

PTS receptors

PEX16PEX3

PEX1

PEX26

PEX5

PEX6

PEX2 PEX10 PEX12 PEX13 PEX14

PEX19 PMP70

PXMP2 MPV17 PXMP4 PEX11

PMP34

PEX7 PEX5PTS2

PTS1

PTS1

PTS2

12

Figure S7. KEGG ‘peroxisome’ pathway with each gene (node) color-coded for its correlation tothe mean lifespan across the six dietary groups according to the color key. Red color representspositive correlation between PEX genes and lifespan, while green color indicates negativecorrelation.

HF+

CR

vs

HF

HF+

CR

vs

LFLF

+CR

vs

LFLF

+Ex

vs L

FH

F+Ex

vs

HF

HF

vs L

FH

F+Ex

vs

LF

Life

Spa

n

PXMP 2/4 family proteins_Mpv17Peroxisome docking_Pex14Gluthatione peroxidase/Thioredoxin_Gstk1D_amino acid degradation_DdoUnsaturated fatty acid beta_oxidation_Ech1Branched chain fatty acid beta_oxidation_Hsd17b4Di_ trihydroxycholestanoic acid beta_oxidation_Hsd17b4Long_chain dicarboxylic acids oxidation_Hsd17b4Straight chain fatty acids beta_oxidation_Hsd17b4Unsaturated fatty acid beta_oxidation_Hsd17b4Peroxisome organization_RhoaBranched chain fatty acid beta_oxidation_Acox1Di_ trihydroxycholestanoic acid beta_oxidation_Acox1Long_chain dicarboxylic acids oxidation_Acox1Straight chain fatty acids beta_oxidation_Acox1Unsaturated fatty acid beta_oxidation_Acox1Unsaturated fatty acid beta_oxidation_Decr2Unknown_Mosc2Lipid metabolism_HmgclPeroxisomal AAA_ATPases_Pex1Branched chain fatty acid beta_oxidation_Acox2Di_ trihydroxycholestanoic acid beta_oxidation_Acox2Long_chain dicarboxylic acids oxidation_Acox2Straight chain fatty acids beta_oxidation_Acox2Unsaturated fatty acid beta_oxidation_Acox2ZN RING proteins_Pex12Peroxisome division_proliferation_Pex11aNicotinate and nicotinamide metabolism_Nudt12Unknown_SerhlPeroxisome membrane assembly_Pex16Retinoid metabolism_Dhrs4Straight chain fatty acids oxidation_CrotAlpha_oxidation_Slc27a2Long/very long fatty acids activation_Slc27a2ABC transporters_Abcd4Di_ trihydroxycholestanoic acid oxidation/Bile acid synthesis_Abcd4Fatty acid synthesis/PUFAS synthesis_Abcd4Straight chain fatty acids oxidation_Abcd4Superoxide dismutase_Sod2Peroxisome division_proliferation_Pex11cZN RING proteins_Pxmp3ABC transporters_Abcd1Di_ trihydroxycholestanoic acid oxidation/Bile acid synthesis_Abcd1Fatty acid synthesis/PUFAS synthesis_Abcd1Straight chain fatty acids oxidation_Abcd1Etherlipid and plasmalogens synthesis_Far1Peroxisome matrix protein import_Pex26PXMP 2/4 family proteins_Pxmp4Regulation of acyl_CoA/CoA ratio_Acot8Straight chain fatty acids oxidation_Acot8Peroxisome membrane assembly_Pex3Di_ trihydroxycholestanoic acid oxidation/Bile acid synthesis_Acnat2Peroxisome docking_Pex13Peroxisome division_proliferation_Pex11bDi_ trihydroxycholestanoic acid oxidation/Bile acid synthesis_BaatPXMP 2/4 family proteins_Pxmp2Etherlipid and plasmalogens synthesis_Far2Peroxisomal AAA_ATPases_Pex6Catalase_CatEpoxide/Isochorismatase hydrolases_Ephx2Peroxisome membrane assembly_Pex19Long/very long fatty acids activation_Acsl1Peroxisome targeting sequence binding_Pex5Proteases_Lonp2Unsaturated fatty acid beta_oxidation_PeciAlpha_oxidation_PhyhBranched chain fatty acid oxidation_AmacrDi_ trihydroxycholestanoic acid oxidation/Bile acid synthesis_AmacrRegulation of acyl_CoA/CoA ratio_CratL_Lysine metabolism_PipoxABC transporters_Abcd2Di_ trihydroxycholestanoic acid oxidation/Bile acid synthesis_Abcd2Fatty acid synthesis/PUFAS synthesis_Abcd2Straight chain fatty acids oxidation_Abcd2Regulation of acyl_CoA/CoA ratio_Acot4Straight chain fatty acids oxidation_Acot4Regulation of acyl_CoA/CoA ratio_Acot6Straight chain fatty acids oxidation_Acot6Regulation of acyl_CoA/CoA ratio_Acot3Straight chain fatty acids oxidation_Acot3Unknown_Acbd5ABC transporters_Abcd3Di_ trihydroxycholestanoic acid oxidation/Bile acid synthesis_Abcd3Fatty acid synthesis/PUFAS synthesis_Abcd3Straight chain fatty acids oxidation_Abcd3Purines and pyrimidines_XdhUnknown_Trim37Peroxisome targeting sequence binding_Pex7Unknown_Zadh2D_amino acid degradation_DaoTransaminases_AgxtRegulation of acyl_CoA/CoA ratio_Nudt7Fatty acid chain elongation_PecrLong/very long fatty acids activation_Acsl6Alpha_oxidation_Hacl1Long_chain dicarboxylic acids oxidation_EhhadhPeroxiredoxin_Prdx5Unknown_Tmem135Proteases_IdeRegulation of acyl_CoA/CoA ratio_Slc22a21Retinoid metabolism_Dhrs2Peroxisome organization_Cldn15Superoxide dismutase_Sod1Unknown_Fndc5Peroxisome organization_Fis1Unsaturated fatty acid beta_oxidation_1810022C23RikEtherlipid and plasmalogens synthesis_AgpsIsoprenoid metabolism_Idi1Peroxiredoxin_Prdx1PXMP 34 family proteins_Slc25a17Alpha_oxidation_Idh1Unsaturated fatty acid beta_oxidation_Idh1Etherlipid and plasmalogens synthesis_GnpatAntiinflammatory_antimicrobial_Nos2Peroxisome organization_Dnm1lBranched chain fatty acid beta_oxidation_Scp2Di_ trihydroxycholestanoic acid beta_oxidation_Scp2Long_chain dicarboxylic acids oxidation_Scp2Straight chain fatty acids beta_oxidation_Scp2Unsaturated fatty acid beta_oxidation_Scp2Regulation of acyl_CoA/CoA ratio_Nudt19Branched chain fatty acid beta_oxidation_Acox3Di_ trihydroxycholestanoic acid beta_oxidation_Acox3Long_chain dicarboxylic acids oxidation_Acox3Straight chain fatty acids beta_oxidation_Acox3Unsaturated fatty acid beta_oxidation_Acox3Long/very long fatty acids activation_Acsl4Purines and pyrimidines_UoxIsoprenoid metabolism_PmvkLong/very long fatty acids activation_Acsl5Peroxisome targeting sequence binding_Pex5l

Log2 Fold Change

-2 -1 0 1 2

A B

******

*

**

**

***

+++

+

++++

++

1.5

1.0

0.5

0Pex1 Pex5 Pex13 Pxmp4

LF+CRHF+CRLF+ExLFHF+ExHF

Nor

mal

ized

gen

e ex

pres

sion

(Rel

ativ

e to

LF)

13

Figure S8. Gene-wise analysis of peroxisomal function categories. (A) With each peroxisomalfunction category, the correlation (PCC) of each gene’s expression level with lifespan across the sixdifferent treatment groups and the fold changes between seven comparisons as indicated arevisualized in the heatmap according to the color keys. (B) The midlife expression levels of theindicated four hepatic PEX genes, which showed negative correlation with the lifespan of micethrough microarray analysis, were confirmed by real-time quantitative PCR. Data are presented asmean ± SEM (n=8/group). *P <0.05, **P<0.01 versus LF, and +P < 0.05, ++P < 0.01 versus HF byANOVAs.

M26a

M26b

F27

M29

F30

F36

M37

F38

M40

M42

M45

F48

M52

M53

F56

F61

M66

M70

F71

M73

M77

M80

F81

M85

F87

M90

F90

M91

M95

F106

0.8

0.9

1

1.1YoungOld

P=0.0593{N

orm

aliz

ed E

xpre

ssio

n Le

vel

Age (years)

14

Figure S9. Expression level changes of the proxisomal proliferation genes (PEX) duringhuman brain aging. Average and standard deviation of the log2 expression levels of the 11 PEXgenes in each of the 30 sample are shown by the height of the bar and whisker, respectively.

Days

Surv

ival

Rat

e

1.0

0.8

0.6

0.4

0.2

00 5 10 15 20 25 30 35

EVdaf-16 (RNAi)prx-11 (RNAi)prx-13 (RNAi)prx-5 (RNAi)F18F11.1 (RNAi)

Hours

Surv

ival

Rat

e

0 5 10 15 20

1.0

0.8

0.6

0.4

0.2

0

EVdaf-16 (RNAi)prx-11 (RNAi)prx-13 (RNAi)prx-5 (RNAi)F18F11.1 (RNAi)

A

B

15

Figure S10. Survival curves of the N2 stain of worms with RNAi knockdown of prx genes startingfrom L1 (A) or in response to heat stress at 35oC (B).

2.5

2.0

1.5

1.0

0.5

0.0

LF+CRHF+CRLF+ExLFHF+ExHF

Sirt1 Ppargc1a

****

Nor

mal

ized

gen

e ex

pres

sion

(Rel

ativ

e to

LF)

16

Figure S11. Hepatic expression levels of Sirt1 and Ppargc1a in the six intervention groups ofmice. Real-time quantitative RT-PCR was performed, and data are presented as mean ± SEM(n=8/group). **P<0.01 versus LF by ANOVAs.

17

Supplemental Tables

Table S1. Metabolic biomarkers in mice of the six intervention groups at 60~62 weeks of age.

Parameter LF+CR HF+CR LF+Ex LF HF+Ex HF

Serum Triglycerides (mg/dl) 14.53 (3.63)

13.23 (5.07)

14.88 (2.84)

18.85 (5.58)

24.48 (5.64)

27.425 (2.16)

Cholesterol (mg/dl)

64.18 (7.72)##**

140.37 (17.40)##

106.32 (6.82)##

121.73 (8.18)##

200.17 (14.64)#

243.97 (17.10)**

insulin (ng/ml)

0.48 (0.03)##

0.89 (0.16)##

0.65 (0.06)##

0.52 (0.08)##

3.80 (0.77)

3.46 (0.56)**

leptin (ng/ml)

0.83 (0.14)## **

10.04 (1.04)##

2.12 (0.09)##

5.14 (0.38)##

53.6 (12.17)

72.09 (5.53)**

GTT AUC

2.79 (0.79)**

9.99 (1.61)

8.09 (1.67)

11.41 (1.79)

11.93 (2.28)

12.06 (1.85)

Liver Liver weight (g) 1.21 (0.04)##

1.26 (0.05)##

1.26 (0.06)##

1.37 (0.05)##

2.43 (0.26)

2.98 (0.17)**

Triglycerides (mg/dl)

14.79 (1.90)##**

27.50 (2.16)##

29.30 (3.84)##

30.80 (3.19)##

65.84 (7.08)

71.36 (5.43)**

Cholesterol (mg/dl)

2.18 (0.23)

2.26 (0.26)

2.77 (0.31)

3.15 (0.31)

3.40 (0.35)

3.53 (0.32)

Values shown are mean (± SEM). **P<0.01 versus LF, # P<0.05 and ## P<0.01 versus HF by two-way ANOVA.