UBC Faculty Research and Publications

An epigenetic clock analysis of race/ethnicity, sex, and coronary heart disease Horvath, Steve; Gurven, Michael; Levine, Morgan E; Trumble, Benjamin C; Kaplan, Hillard; Allayee, Hooman; Ritz, Beate R; Chen, Brian; Lu, Ake T; Rickabaugh, Tammy M; Jamieson, Beth D; Sun, Dianjianyi; Li, Shengxu; Chen, Wei; Quintana-Murci, Lluis; Fagny, Maud; Kobor, Michael S; Tsao, Philip S; Reiner, Alexander P; Edlefsen, Kerstin L; Absher, Devin; Assimes, Themistocles L Aug 11, 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
52383-13059_2016_Article_1030.pdf [ 2.58MB ]
Metadata
JSON: 52383-1.0361800.json
JSON-LD: 52383-1.0361800-ld.json
RDF/XML (Pretty): 52383-1.0361800-rdf.xml
RDF/JSON: 52383-1.0361800-rdf.json
Turtle: 52383-1.0361800-turtle.txt
N-Triples: 52383-1.0361800-rdf-ntriples.txt
Original Record: 52383-1.0361800-source.json
Full Text
52383-1.0361800-fulltext.txt
Citation
52383-1.0361800.ris

Full Text

An epigenetic clock analysis of race/ethnicity, sex,and coronary heart diseaseHorvath et al.Horvath et al. Genome Biology  (2016) 17:171 DOI 10.1186/s13059-016-1030-0RESEARCH Open AccessAn epigenetic clock analyethnicity, sex, and coronar1,Taun1Many demographic and epidemiological studies explore markers of aging also remains an open question. Oneen shown toians [9], des-ation, lowerHorvath et al. Genome Biology  (2016) 17:171 DOI 10.1186/s13059-016-1030-0Los Angeles, CA 90095, USAFull list of author information is available at the end of the articleaverage life expectancies, and higher disease incidence.To date, no studies have employed epigenetic measures1Human Genetics, David Geffen School of Medicine, University of CaliforniaLos Angeles, Los Angeles, CA 90095, USA2Biostatistics, School of Public Health, University of California Los Angeles,lable biomarkers capture only particular aspFor example, African Americans have behave longer telomere lengths than Caucaspite significantly higher levels of inflamm* Correspondence: shorvath@mednet.ucla.edu†Equal contributorsthe effects of chronological age, race/ethnicity, and sex onmortality rates and susceptibility to chronic disease [1–5],but it remains an open research question whether race/ethnicity and sex affect molecular markers of aging dir-ectly. To what extent clinical biomarkers of inflammation,major challenge is the lack of agreement on how to defineand measure biological aging rates [6]. Many biomarkersof aging have been proposed ranging from clinical markers(such as whole-body functional evaluations and gait speed)to molecular markers such as telomere length [7, 8]. Avai-ects of aging.BackgroundAbstractBackground: Epigenetic biomarkers of aging (the “epigenetic clock”) have the potential to address puzzling findingssurrounding mortality rates and incidence of cardio-metabolic disease such as: (1) women consistently exhibiting lowermortality than men despite having higher levels of morbidity; (2) racial/ethnic groups having different mortality rateseven after adjusting for socioeconomic differences; (3) the black/white mortality cross-over effect in late adulthood;and (4) Hispanics in the United States having a longer life expectancy than Caucasians despite having a higher burdenof traditional cardio-metabolic risk factors.Results: We analyzed blood, saliva, and brain samples from seven different racial/ethnic groups. We assessed theintrinsic epigenetic age acceleration of blood (independent of blood cell counts) and the extrinsic epigeneticaging rates of blood (dependent on blood cell counts and tracks the age of the immune system). In blood,Hispanics and Tsimane Amerindians have lower intrinsic but higher extrinsic epigenetic aging rates thanCaucasians. African-Americans have lower extrinsic epigenetic aging rates than Caucasians and Hispanics but nodifferences were found for the intrinsic measure. Men have higher epigenetic aging rates than women in blood,saliva, and brain tissue.Conclusions: Epigenetic aging rates are significantly associated with sex, race/ethnicity, and to a lesser extentwith CHD risk factors, but not with incident CHD outcomes. These results may help elucidate lower than expectedmortality rates observed in Hispanics, older African-Americans, and women.Keywords: DNA methylation, Epigenetic clock, Race, Gender, Aging, Coronary heart disease, Hispanic paradox,Black/white mortality cross-overdyslipidemia, and immune senescence relate to cellularSteve Horvath1,2*† , Michael Gurven3†, Morgan E. LevineHooman Allayee5, Beate R. Ritz6, Brian Chen7, Ake T. Lu1,Dianjianyi Sun9, Shengxu Li9, Wei Chen9, Lluis Quintana-MPhilip S. Tsao13,14, Alexander P. Reiner15, Kerstin L. Edlefse© 2016 The Author(s). Open Access This articInternational License (http://creativecommonsreproduction in any medium, provided you gthe Creative Commons license, and indicate if(http://creativecommons.org/publicdomain/zesis of race/y heart diseaseBenjamin C. Trumble3, Hillard Kaplan4,mmy M. Rickabaugh8, Beth D. Jamieson8,rci10, Maud Fagny11, Michael S. Kobor12,6, Devin Absher17† and Themistocles L. Assimes13†le is distributed under the terms of the Creative Commons Attribution 4.0.org/licenses/by/4.0/), which permits unrestricted use, distribution, andive appropriate credit to the original author(s) and the source, provide a link tochanges were made. The Creative Commons Public Domain Dedication waiverro/1.0/) applies to the data made available in this article, unless otherwise stated.Horvath et al. Genome Biology  (2016) 17:171 Page 2 of 22to estimate and compare molecular aging rates amonggender or racial/ethnic groups.Measures incorporating DNA methylation levels haverecently given rise to a new class of biomarkers thatappear informative of aging given that age has a pro-found effect on DNA methylation levels in most humantissues and cell types [10–18]. Several recent studieshave measured the epigenetic age of tissue samples bycombining the DNA methylation levels of multipledinucleotide markers, known as Cytosine phosphateGuanines or CpGs [19–21]. We recently developed theepigenetic clock (based on 353 CpGs) to measure theage, known as “DNA methylation age” or “epigeneticage,” of assorted human cell types (CD4+ T cells orneurons), tissues, and organs—including blood, brain,breast, kidney, liver, lung [20], and even prenatal brainsamples [22]. The epigenetic clock is an attractive bio-marker of aging because it applies to most humantissues and its accurate measurement of chronologicalage is unprecedented.The following evidence shows that the epigenetic clockcaptures aspects of biological age. First, the epigeneticage of blood has been found to be predictive of all-causemortality even after adjusting for chronological age anda variety of known risk factors [23–25]. Second, theblood of the offspring of Italian semi-supercentenarians(i.e. participants who reached an age of at least 105 years)has a lower epigenetic age than that of age-matchedcontrols [26]. Third, the epigenetic age of blood relatesto frailty [27] and cognitive/physical fitness in the elderly[28]. The utility of the epigenetic clock method has beendemonstrated in applications surrounding obesity [29],Down’s syndrome [30], HIV infection [31], Parkinson’s dis-ease [32], Alzheimer’s disease-related neuropathologies[33], lung cancer [34], and lifetime stress [35]. Here, weapply the epigenetic clock to explore relationships betweenepigenetic age and race/ethnicity, sex, risk factors of coron-ary heart disease (CHD), and the CHD outcome itself.ResultsBlood datasets and racial/ethnic groupsAn overview of our DNA methylation datasets can befound in Table 1. We analyze multiple sources of DNA:mostly blood, saliva, and lymphoblastoid cell lines. Inaddition, brain datasets were used to compare men andwomen (Table 2). We considered the following racial/ethnicgroups (Table 1): 1387 African Ancestry (African Ameri-cans and two groups from Central Africa), 2932 Caucasian(non-Hispanic whites), 657 Hispanic, 127 East Asians(mainly Han Chinese), and 59 Tsimane Amerindians.Accuracy of the epigenetic clockDNAm age, also referred to as epigenetic age, was calcu-lated in human samples profiled with the Illumina Infinium450 K platform using a previously described method [20].As expected, we found DNAm age to have a strong linearrelationship with chronological age in blood and saliva (cor-relations in the range of 0.65–0.93, Figs. 1, 2, 3, 4, and 5)and in lymphoblastoid cell lines (r = 0.59; Additional file 1).Based on a spline regression line, we defined a “universal”measure of epigenetic age acceleration, denoted “AgeAccel.” in our figures, as the difference between the ob-served DNAm age value and the value predicted by a splineregression model in Caucasians. The term “universal” refersto the fact that this measure can be defined in a vast majo-rity of tissues and cell types with the notable exception ofsperm [20]. A positive value of the universal age acceler-ation measure indicates that DNA methylation age ishigher than that predicted from the regression model forCaucasian participants of the same age. Our intrinsic andextrinsic age acceleration measures (see “Methods”) onlyapply to blood data. A measure of intrinsic epigenetic ageacceleration (IEAA) measures cell-intrinsic epigenetic agingeffects that are not confounded by extra-cellular differencesin blood cell counts. The measure of IEAA is an incompletemeasure of the age-related functional decline of the im-mune system because it does not track age-related changesin blood cell composition, such as the decrease of naïveCD8+ T cells and the increase in memory or exhaustedCD8+ T cells [36–38]. The measure of extrinsic epigeneticage acceleration (EEAA) only applies to whole blood andaims to measure epigenetic aging in immune-related com-ponents. It keeps track of both intrinsic epigenetic changesand age-related changes in blood cell composition (see“Methods”). The estimated blood cell counts, which areused in these measures, correlate strongly with correspond-ing flow cytometric measurements from the MACS study(Additional file 2): r = 0.63 for CD8 +T cells, r = 0.77 forCD4+ T, r = 0.67 B cell, r = 0.68 naïve CD8+ T cell, r = 0.86for naïve CD4+ T, and r = 0.49 for exhausted CD8+ Tcells.Hispanics have a lower intrinsic aging rate thanCaucasiansWe find that Hispanics have a consistently lower IEAAcompared to Caucasians (p = 7.1 × 10–10, Fig. 1m). Animportant question is whether the observed differencesin blood can also be observed in other tissues. Using anovel saliva dataset (dataset 4, saliva from PEG) we findthat Hispanics have a lower epigenetic aging rate thanCaucasians (p = 0.042, Fig. 1i). The fact that our findingsin blood can also be validated in saliva is consistent withthe strong correlation between epigenetic age accelerationmeasures of the two sources of DNA (r = 0.70, p = 1.4 ×10–12, Fig. 1n). The lower value of IEAA in Hispanicsunlikely reflects country of birth or of residence (at age35 years) given the robust findings across samples andour detailed analysis in the WHI, where we find thatHispanics born outside US, but living in the US, have aTable 1 Overview of the DNA methylation datasets. The rows correspsource, DNA methylation platform, number of participants, access infoTissue source Array Participants(n)Women(n)African Ancestry,Caucasian, HispanTsimane, East Asi1. Women’s Health Initiative(blood)450 1462 1462 676, 353, 433, 0, 02. Bogalusa (blood) 450 969 547 288, 681, 0, 0, 03. PEG (blood) 450 335 138 0, 289, 46, 0, 04. Saliva from PEG 450 259 113 0, 166, 93, 0, 05. Older Tsimane andothers450 310 150 0, 235, 38, 37, 06. Younger Tsimaneand Caucasians450 46 31 0, 24, 0, 22, 07. East Asians vs.Caucasians (PSPsamples removed)450 312 132 0, 279, 0, 0, 338. African populations 450 256 50 256, 0, 0, 0, 09. Cord blood 27 216 110 92, 70, 0, 0, 010. Male saliva 27 91 0 0, 59, 32, 0, 011. Female saliva 27 42 42 0, 27, 15, 0, 012. Lymphoblastoidcell lines450 237 154 75, 68, 0, 0, 94Table 2 Description of brain datasets for evaluating the effectof gender. Additional details can be found in “Methods”Data Participants(n)Men(%)Age mean ± SE[min, max]BrainregionBrain tissuesamples (n)Study 1 117 41 % 84.0 ± 9.8 [40, 105] CRBLM 112EC 108PFCTX 114STG 117Study 2 142 68 % 48.0 ± 23.2 [16, 96] CRBLM 112FCTX 133PONS 125TCTX 127Study 3 147 63 % 44.3 ± 9.6 [19, 68] CRLM 147Study 4 37 62 % 64.4 ± 17.4 [25, 96] CRBLM 36PFCTX 36Study 5 209 66 % 52.3 ± 29.8 [1, 102] CRBLM 201FCTX 201Study 6 718 37 % 88.5 ± 6.6 [66, 108] DLPFC 718CRBLM cerebellum, DLPFC dorsolateral prefrontal cortex, EC entorhinal cortex,FCTX frontal cortex, PFCTX prefrontal cortex, PONS pons, STG superior temporalgyrus, TCTX temporal cortexHorvath et al. Genome Biology  (2016) 17:171 Page 3 of 22ond to the datasets used in this article. Columns report the tissuermation, and citation and a reference to the use in this textic,an (n)Mean age(years) (range)Available Citation Figure63 (50–80) dbGAP, NHLBI Currentarticle143 (29–51) dbGAP, NHLBI Currentarticle170 (36–91) GSE72775 Currentarticle169 (36–88) GSE78874 Currentarticle166 (35–92) GSE72773 Currentarticle315 (2–35) GSE72777 Currentarticle368 (34–93) GSE53740 Li, 2014 [73] 340 (16–90) EGAS00001001066 Fagny, 2015[42]40 (0–0) GSE27317 Adkins,higher IEAA than Hispanics born and raised in the US(p = 0.025, Additional file 3B).CHD risk factors bear little or no relationship with IEAAWe related our measures of age acceleration to risk fac-tors related to CHD since the latter are significant pre-dictors of mortality. In postmenopausal women from theWomen’s Health Initiative (WHI), we found no evidencethat IEAA is associated with disparities in education,high density lipoprotein (HDL) or low density lipopro-tein (LDL) cholesterol, insulin, glucose, C-reactive pro-tein (CRP), creatinine, alcohol consumption, smoking,diabetes status, or hypertension (see Table 3).Tsimane have a lower intrinsic aging rate than CaucasiansThe Tsimane are an indigenous population (~15,000 in-habitants) of forager-horticulturalists who reside in theremote lowlands of Bolivia. They reside mostly in open-airthatch huts, and actively fish, hunt, and cultivate plantains,rice, and manioc through slash-and-burn horticulture[39]. Tsimane provide a unique contribution to aging re-searchers and epidemiologists because they experiencehigh rates of inflammation due to repeated bacterial, viral,and parasitic infections, yet show minimal risk factors for2011 [44]29 (21–55) GSE34035 Liu, 2010[74]27 (21–55) GSE34035 Liu, 2010[74]34 (5–73) GSE36369 Heyn, 2013[88]Additional file 1aejbfkcglhmdinFig. 1 Intrinsic epigenetic age acceleration in Caucasians and Hispanics. a-d DNA methylation age (y-axis) versus chronological age (x-axis) in (a)Women’s Health Initiative, (b) blood data from PEG, (c) dataset 5, (d) saliva data from PEG. Dots corresponds to participants and are colored by ethnicgroup (gray = Caucasian, blue =Hispanic). The gray line depicts a spline regression line through Caucasians. We define two measures of age accelerationbased on DNAm age. e-g The bar plots relate the universal measure of epigenetic age acceleration to race/ethnicity, which is defined as residual to thespline regression line through Caucasians, i.e. the vertical distance of a point from the line. By definition, the mean age acceleration in Caucasians is zero.h, m Results after combining the three blood datasets using Stouffer’s meta-analysis method. i Age acceleration residual versus ethnicity in the saliva datafrom PEG. j-m The y-axis reports the mean value of IEAA, which is defined as residual from a multivariate regression model that regresses DNAm age onage and several measures of blood cell counts. Each bar plot reports 1 standard error and the p value from a group comparison test (ANOVA). n Ageacceleration in blood versus age acceleration in saliva for the subset of PEG participants for whom both data were availableadgbehcfiFig. 2 Intrinsic epigenetic age acceleration in Tsimane, Hispanics, East Asians, and Caucasians. a-c DNA methylation age (y-axis) versus chronologicalage (x-axis) in (a) dataset 5, (b) dataset 6, (c) dataset 7. Dots corresponds to participants and are colored by race/ethnicity (green= African American,gray= Caucasian, blue= Hispanic, red = Tsimane, orange= East Asians). The gray line depicts a spline regression line through Caucasians. We define twomeasures of age acceleration based on DNAm age. d-f The bar plots relate the universal measure of epigenetic age acceleration to race/ethnicity,which is defined as residual to the spline regression line through Caucasians, i.e. the vertical distance of a point from the line. g-i The y-axis reports themean value of IEAA, which is defined as residual from a multivariate regression model that regresses DNAm age on age and several measures ofblood cell counts. Each bar plot reports 1 standard error and the p value from a group comparison test (ANOVA)Horvath et al. Genome Biology  (2016) 17:171 Page 4 of 22estcipHorvath et al. Genome Biology  (2016) 17:171 Page 5 of 22acfbdgFig. 3 Intrinsic epigenetic age acceleration versus African or European Anc(a) Women’s Health Initiative, (b) Bogalusa study. Dots corresponds to partiheart disease or type 2 diabetes as they age; they have min-imal hypertension and obesity, low LDL cholesterol andno evidence of peripheral arterial disease [39–41]. SinceHispanics share genetic ancestry with peoples indigenousto the Americas, we hypothesized that a slower intrinsicaging rate might also be observable by analyzing Tsimaneblood samples [39]. Among participants who are older than35 years, Tsimane have the lowest intrinsic age acceleration(Fig. 2d, g). While Tsimane have a significantly lower IEAAthan Caucasians after the age of 35 years (p = 0.0061), nosignificant difference could be observed in younger partici-pants (Fig. 2e, h). In this analysis, the threshold of 35 yearswas chosen so that a sufficient number of young partici-pants would be included in dataset 6. We found no signifi-cant difference in IEAA between older Hispanics andTsimane, which might reflect the relatively low group sizesof n = 37 Tsimane versus n = 38 Hispanics.IEAA is not associated with CHD in the WHIBased on our findings above showing little or no rela-tionship between IEAA and CVD risk factors at baseline,we hypothesized that IEAA would not predict future on-set of CHD. A multivariate logistic regression modelshows that IEAA is not significantly associated with anincreased risk of incident CHD (Table 4). However, asCaucasian). The gray line depicts a spline regression line through Caucasians. WThe bar plots relate the universal measure of epigenetic age acceleration to rathrough Caucasians. e, h Results after combining the two blood datasets usinof IEAA, which is defined as residual from a multivariate regression model thaEach bar plot reports 1 standard error and the p value from a group comparisehry. a-c DNA methylation age (y-axis) versus chronological age (x-axis) inants and are colored by race/ethnicity (green= African Ancestry, gray =expected, current smoking, prior history of diabetes,hypertension, high insulin and glucose levels, and lowerHDL predicted an increased risk of CHD (Table 4).Hispanics and Tsimane have a higher EEAA thanCaucasiansAccording to our measure of EEAA, Hispanics have a sig-nificantly older extrinsic epigenetic age than Caucasians(meta-analysis p = 0.00012, Fig. 4a–d) and fewer naïveCD4+ T cells, based on cytometric data from the WHILLS, the MACS study, and imputed blood cell counts(Fig. 4f–j, Additional file 2H, I). This pattern of fewernaïve CD4+ T cells is even more pronounced for Tsimane(Fig. 4m, n), who experience repeated acute infections andelevated, often chronic, inflammatory loads.Epigenetic age analysis of East AsiansBecause ancient Native American populations sharecommon ancestral lineages with East Asians, we exam-ined whether East Asians also differ from Caucasiansin terms of epigenetic aging rates. We found no signifi-cant difference between Caucasians and East Asians interms of IEAA (Fig. 2i), EEAA (Fig. 4o), or naïve CD4+T cells (Fig. 4p). Similarly, we found no difference inlymphoblastoid cell lines (Additional file 1). However,e define two measures of age acceleration based on DNAm age. c, dce/ethnicity, which is defined as residual to the spline regression lineg Stouffer’s meta-analysis method. f, g The y-axis reports the mean valuet regresses DNAm age on age and several measures of blood cell counts.on test (ANOVA)adgbehcfiFig. 5 Analysis of African rainforest hunter-gatherers and farmers. a DNAm age versus age using 256 blood samples from [42]. The points are coloredas follows: magenta = AGR (urban setting), turquoise = AGR (forest), brown= RHG (forest). Group status versus (b) universal age acceleration, (d) intrinsicage acceleration, (f) extrinsic age acceleration. Habitat versus (c) universal age acceleration, (e) intrinsic age acceleration, (g) extrinsic age acceleration.(h, i) are analogous to (a, b) but the y-axis is based on a DNAm age estimate that excluded CpG that were located near SNPs. In this robustnessanalysis, we removed CpG probes containing genetic variants at a frequency higher than 1 % in the populations studiedagmsbhntcioudjpvekqwflrxFig. 4 Extrinsic epigenetic age acceleration and blood cell counts across groups. EEAA versus race/ethnicity in (a, q) Women’s Health Initiative,(b) blood data from PEG, (c, k) dataset 5, (l) dataset 6, (o) dataset 7, (r) Bogalusa study. Flow cytometric, age adjusted estimates (e, t) naïve CD8+T and (j, x) naïve CD4+ T cell counts in the WHI LLS. Age adjusted estimates of naïve CD4 + T cells based on DNA methylation data from (f, u)Women’s Health Initiative, (g) blood data from PEG, (h, m) dataset 5, (n) dataset 6, (p) dataset 7, (v) Bogalusa study. (d, i, s, w) Meta-analysisacross the respective datasets based on Stouffer’s methodHorvath et al. Genome Biology  (2016) 17:171 Page 6 of 22ationesicat(0(0.2(0(001(0(0(0(0(0(0(0(0.4(1Horvath et al. Genome Biology  (2016) 17:171 Page 7 of 22Table 3 Multivariate model that regresses epigenetic age accelerp values from regressing measures of intrinsic and extrinsic epigeMultivariate linear regression IntrinEstimRace/ethnicity Hispanic vs. African American –0.94White vs. African American 0.71HDL-cholesterol 0.006Triglyceride 0.003Insulin 0 (0.0Glucose 0.003CRP 0.023Creatinine 0.703BMI 0.035Education High school (HS) vs. no HS 0.357Some college vs. no HS 0.469College vs. no HS 0.486Grad school vs. no HS 0.36Alcohol Past drinker vs. Never 1.668these comparative analyses are limited by the relativelysmall number of samples and should be repeated inlarger datasets.Which risk factors for cardiometabolic disease areassociated with EEAA?Our multivariate model analysis in the WHI (Table 3)shows that EEAA tracks better than IEAA with riskfactors for cardiometabolic disease; EEAA was positivelyassociated (higher) with: triglyceride levels (multivariatemodel p = 0.04), CRP (p = 0.023), and creatinine (p = 0.008).EEAA was negatively associated (lower) with higher levelsof education in all ethnic groups (p from 2.0 × 10–8 to 0.05,Additional file 4I–L). For each racial/ethnic group, wefind that women who did not finish high schoolexhibit the highest levels of EEAA (leftmost bar inAdditional file 4J–L).Epigenetic aging rates of African AmericansIn the following, we compare African Americans withEuropean Americans in terms of IEAA and EEAA. Com-parisons of African Americans with Caucasians in termsof IEAA yield contradictory findings across datasets thatLight drinker vs. Never –0.101Moderate vs. Never –0.416Heavy vs. Never –0.354Smoking Former vs. Current –0.573Never vs. Current –0.376Diabetes 0.216 (0Hypertension 0.364 (0R-squared 0.029n on participant characteristics in the WHI. Coefficients andtic age acceleration on participant characteristics from dataset 1EAA Extrinsic EAAe (SE) p Estimate (SE) p.35) 0.007 3.363 (0.439) <10–1595) 0.016 1.94 (0.37) 1.6 × 10–7.01) 0.558 –0.003 (0.013) 0.799.002) 0.059 0.004 (0.002) 0.04) 0.664 0.001 (0.001) 0.337.004) 0.486 0.007 (0.005) 0.112.018) 0.215 0.052 (0.023) 0.023.594) 0.237 1.985 (0.745) 0.008.021) 0.103 0.045 (0.027) 0.093.426) 0.403 –0.784 (0.534) 0.142.381) 0.219 –1.171 (0.478) 0.014.519) 0.349 –2.253 (0.65) 0.00124) 0.396 –1.648 (0.531) 0.002.1) 0.13 –0.598 (1.379) 0.665differ in age range: African American women haveslightly lower IEAA than Caucasian women in the WHI(p = 0.017 Fig. 3f ), but no significant difference can beobserved for the younger participants of the Bogalusa study(Fig. 3g). Indeed, participants in the WHI (aged between50 and 80 years) were older than those of the Bogalusastudy (aged between 29 and 51 years). This failure to detecta significant racial/ethnic difference in IEAA in youngerparticipants is consistent with our results from the com-parison of younger Tsimane and Caucasians (Fig. 2h). Amultivariate model analysis based on the Bogalusa study(comprising African Americans and Caucasians) confirmsthat IEAA does not differ between middle-aged AfricanAmericans and Caucasians but IEAA is higher among men(p = 0.025) and has a marginally significant association withhypertension (p = 0.064, Table 5). When relating individualvariables to IEAA, we find significant associations forhypertension (p = 0.00035, Additional file 5D–F) but notfor type II diabetes status or educational level.Our findings for EEAA are highly consistent across thetwo studies and age groups: African Americans have lowerEEAA than Caucasians in the WHI and in the Bogalusastudy (p = 7.2 × 10–7, Fig. 4q, r, s). Our flow cytometric(0.536) 0.85 –0.751 (0.672) 0.264(0.748) 0.578 –0.401 (0.937) 0.669(0.88) 0.687 –0.833 (1.103) 0.45(1.039) 0.581 –0.104 (1.302) 0.936(1.039) 0.718 –0.122 (1.303) 0.925.43) 0.616 –0.061 (0.539) 0.909.241) 0.131 0.262 (0.302) 0.3860.069Table 4 Logistic model that regresses CHD status on epigenetic age acceleration and participant characteristics in the WHI. Coefficients,Wald Z statistics, and corresponding p values resulting from regressing CHD status on measures of epigenetic age acceleration andvarious participant characteristics. The results for the measure of IEAA and EEAA can be found in columns 2 and 3, respectivelyLogistic model. Outcome CHD case status Intrinsic EAA Extrinsic EAACovariates Estimate (SE) Z p Estimate (SE) Z pEpig. Age Accel –0.017 (0.01) –1.72 0.085 –0.006 (0.008) –0.74 0.458Age 0.027 (0.008) 3.44 0.001 0.028 (0.008) 3.52 4.3 × 10-4Race/ethnicity Hispanic vs. African American 0.083 (0.152) 0.55 0.584 0.118 (0.153) 0.77 0.443White vs. African American 0.141 (0.135) 1.04 0.298 0.135 (0.135) 1.00 0.319HDL-cholesterol –0.02 (0.005) –4.29 1.8 × 10–5 –0.02 (0.005) –4.33 1.5 × 10-5Triglyceride 0.001 (0.001) 1.43 0.153 0.001 (0.001) 1.38 0.169Insulin 0.002 (0.001) 2.26 0.024 0.002 (0.001) 2.25 0.024Glucose 0.005 (0.002) 2.64 0.008 0.005 (0.002) 2.64 0.008CRP 0.013 (0.008) 1.61 0.107 0.013 (0.008) 1.61 0.108Creatinine 0.518 (0.281) 1.84 0.065 0.515 (0.281) 1.84 0.067BMI –0.011 (0.01) –1.19 0.235 –0.012 (0.01) –1.22 0.223Education High school (HS) vs. no HS –0.058 (0.183) -0.32 0.753 –0.067 (0.183) –0.37 0.715Some College vs. no HS 0.008 (0.164) 0.05 0.96 –0.004 (0.165) –0.03 0.979College vs. no HS –0.198 (0.223) –0.89 0.373 –0.219 (0.223) –0.98 0.327Grad school vs. no HS –0.237 (0.183) –1.29 0.196 –0.251 (0.183) –1.37 0.171Alcohol Past drinker vs. Never –0.6 (0.514) –1.17 0.243 –0.641 (0.513) –1.25 0.212Light drinker vs. Never –0.34 (0.233) –1.46 0.145 –0.343 (0.233) –1.47 0.141Moderate vs. Never –0.1 (0.32) –0.31 0.754 –0.096 (0.32) –0.30 0.764Heavy vs. Never –0.34 (0.381) –0.89 0.373 –0.337 (0.381) –0.88 0.377Smoking Former vs. Current –0.997 (0.467) –2.13 0.033 –0.989 (0.467) –2.12 0.034Never vs. Current –1.321 (0.468) –2.82 0.005 –1.317 (0.468) –2.81 0.005Diabetes 0.706 (0.196) 3.61 3.0 × 10-4 0.699 (0.196) 3.58 3.4 × 10-4Hypertension 0.565 (0.103) 5.46 4.8 × 10-8 0.559 (0.103) 5.41 6.3 × 10-8Table 5 Multivariate model that regresses epigenetic age acceleration on participant characteristics in the Bogalusa study. Coefficientsand p values from regressing measures of intrinsic and extrinsic epigenetic age acceleration on participant characteristics from dataset 2Multivariate linear regression Intrinsic EAA Extrinsic EAAEstimate (SE) Z p Estimate (SE) Z pRace Caucasian vs. African American –0.013 (0.316) –0.04 0.97 0.843 (0.316) 2.67 0.0076Gender Female vs. Male –0.622 (0.278) –2.24 0.025 –0.718 (0.277) –2.60 0.0093Education Grade 8–9 vs. < Grade 8 1.583 (1.468) 1.08 0.28 2.177 (1.465) 1.49 0.14Grade 10–12 vs. < Grade 8 1.285 (1.27) 1.01 0.31 2.267 (1.267) 1.79 0.074Vocat/Tech vs. < Grade 8 0.307 (1.299) 0.24 0.81 1.921 (1.295) 1.48 0.14College vs. < Grade 8 0.85 (1.281) 0.66 0.51 2.375 (1.277) 1.86 0.062Graduate vs. < Grade 8 0.147 (1.336) 0.11 0.91 1.53 (1.332) 1.15 0.25Diabetes (II) 0.173 (0.485) 0.36 0.72 0.012 (0.483) 0.03 0.98Hypertension 0.539 (0.291) 1.86 0.064 1.247 (0.29) 4.30 1.7 × 10-5R-squared 0.025 0.043Horvath et al. Genome Biology  (2016) 17:171 Page 8 of 22Horvath et al. Genome Biology  (2016) 17:171 Page 9 of 22data from the WHI LLS show that African Americanwomen exhibit a higher abundance of naïve CD8+ T cellsthan Caucasian women (p = 1.7 × 10–9, Fig. 4t).In multivariate regression analyses of EEAA, we find thatAfrican Americans have indications of a significantly youn-ger immune system age than Caucasians (p= 0.0076) aftercontrolling for gender, educational level, diabetes status, andhypertension. In the Bogalusa study, we find three significantpredictors of EEAA: race/ethnicity, hypertension, and gen-der (p= 0.0093, Table 5). A marginal analysis in the Bogalusastudy identifies a significant association between EEAA andhypertension (p = 8.0 × 10–5, Additional file 5G–I), type IIdiabetes status in Caucasians (p = 0.0085, Additional file6H), but not in African Americans (Additional file 6I).Contrary to our findings in the WHI, no significant associ-ation can be observed between EEAA and educationallevel (Additional file 7).African rainforest hunter-gatherers and farmersTo evaluate the effect of subsistence ecology and environ-ment on epigenetic aging rates, we analyzed 256 bloodsamples from two different groups in Central Africa: rain-forest hunter-gatherers (RHGs, traditionally known as “pyg-mies,” sampled from Baka and Batwa populations) andAfrican populations that have adopted an agrarian lifestyle(AGRs, traditionally known as “Bantus,” sampled from theNzebi, Fang, Bakiga, and Nzime populations) over the last5000 years [42]. The ancestors of the RHGs and AGRsdiverged ~60,000 years ago. These groups have historicallyoccupied separate ecological habitats—the ancestors ofRHGs in the equatorial rainforest while those of AGRs indrier, more open space savannahs and grasslands. ManyRHG groups still live in the rainforest as mobile bands,whereas AGR populations now occupy primarily rural orurban deforested areas, though some AGR groups havesettled in the rainforest over the last millennia.We considered three groups: (1) RHG (n = 102); (2)AGR living in the forest (n = 60); and (3) AGR living in anurban setting (n = 94). The forest habitat was significantlyassociated with an increase in AgeAccel (p = 2.4 × 10–8,Fig. 5c) and EEAA (p = 5.9 × 10–11, Fig. 5g), but no differ-ence was found for IEAA (p = 0.11, Fig. 5e). Further, nosignificant difference could be observed between AGRand RHG when focusing on participants living in the rain-forest, suggesting greater importance of environment overgenetic differences. These results are not affected by dif-ferences in genetic variants between RHG and AGR ascan be seen from a robustness analysis where we removedCpG probes containing genetic variants at a frequencyhigher than 1 % in the populations studied (Fig. 5h, i).Sex effects in blood and salivaWe explored whether differences exist between men andwomen in epigenetic aging rates. According to measuresof IEAA, men are older than women in two racial/ethnicgroups: African Americans (Additional file 8A, B) andCaucasians (Additional file 9A, B, N, Z).Overall, men have higher IEAA and EEAA thanwomen even when controlling for education, diabetes,and hypertension (Table 5). Using saliva data fromPEG, we find that Hispanic men age faster thanHispanic women (p = 0.021, Fig. 6j). According toEEAA, Caucasian men are epigenetically older thanCaucasian women (Additional file 9C, O, ZA), but wedo not observe a significant difference in othergroups such as African Americans (Additional file8C) or central African populations (Fig. 6p, q). Theresults for EEAA are also consistent with significantsex differences in blood cell counts suggesting morerapid immunosenescence in men. Men have fewernaïve CD4+ T cells than women in three racial/ethnicgroups: Caucasians (p = 0.0015 in the Bogalusa study,p = 0.051 in PEG, p = 4.2 × 10–5 in dataset 5); Tsimane(p = 0.0088 in older Tsimane); and African Americans(p = 0.011 in the Bogalusa study).Sex effects in brain tissueWe analyzed the effect of sex on the universal measureof age acceleration (Age Accel.) in six independent braindatasets (Table 2 and “Methods”). In total, we analyzed2287 brain samples from 1370 participants. In our ana-lysis, we distinguished the cerebellum from other brainregions because it is known to age more slowly thanother brain regions according to the epigenetic clock[43]. While sex did not have a significant effect on theepigenetic age of the cerebellum (Fig. 7a), we found thatother brain regions from men exhibit a significantly higherage acceleration than those from women (Fig. 7b, meta-analysis p = 3.1 × 10–5).Studies of young participantsSo far, our results have largely pertained to partici-pants who are middle-aged or older (Table 1, column6) as we only had access to two datasets involvingnewborns, infants, children, adolescents, and/or youngadults. In dataset 6 (which involved participants be-tween the ages of 2 and 35 years), we did not observea significant difference epigenetic aging rates betweenCaucasians and Tsimane. In cord blood samples [44],we found no significant difference in the epigeneticages of cord blood samples between African Americanand Caucasian newborns (p = 0.23).Robustness analysis in the WHIThe epigenetic clock involves 47 CpGs whose broadlydefined neighborhood includes a single nucleotidepolymorphism (SNP) marker according to the probeannotation file from the Illumina 450 K array. Thus, geneticHorvath et al. Genome Biology  (2016) 17:171 Page 10 of 22afkbglchmdifferences coupled with differences in hybridization effi-ciency could give rise to spurious differences between dif-ferent racial/ethnic groups.We addressed this concern in multiple ways. First, were-analyzed the WHI data by removing the 47 CpGs(out of 353 epigenetic clock CpGs) from the analysis.The epigenetic clock software imputes the 47 missingCpGs using a constant value (the mean value observedin the original training set). Using the resulting modifiedepigenetic clock, we validate our findings of racial/ethnicdifferences in terms of IEAA and EEAA (Additional file8A–C). However, this type of robustness analysis is limitedbecause the removal of a subset of DNA methylationprobes, potentially influenced by proximal genetic variation,is not as good a control as directly having matched geneticdata. Second, we used a completely independent epigeneticbiomarker based on a published signature of age-relatedCpGs from Teschendorff et al. [13]. Again, these resultscorroborate our findings (Additional file 8D, E). Third, wevalidated our findings using the original blood-based agingmeasure by Hannum [19] (Additional file 8F, G). Fourth,p q rFig. 6 Sex effect on epigenetic age acceleration in blood and saliva. Panelsand extrinsic epigenetic age acceleration, respectively. Results are reportedall blood studies can be found in panels (i) IEAA, (s) EEAA. Each bar plot redinejowe highlight that both the Horvath and Hannum ageestimators were developed based on training data frommixed populations. The training data underlying theHorvath clock involved four racial/ethnic groups (mainlyCaucasians, Hispanics, African Americans, and to a lesserextent East Asians). The Hannum clock was trained onCaucasians and Hispanics. While race/ethnicity can lead toa significant offset between DNAm age and chronologicalage (which is interpreted as age acceleration), these twovariables are highly correlated in all racial/ethnic groups.DiscussionOur main findings are that: (1) Hispanics and Tsimanehave a lower intrinsic but a higher extrinsic aging ratethan Caucasians; (2) African Americans have a lowerextrinsic epigenetic aging rate than Caucasians andHispanics; (3) levels of education are associated with adecreased level of EEAA in each race/ethnic group(Additional file 4); (4) neither intrinsic nor extrinsic agingrates of blood tissue are predictive of incident CHD in theWHI even though EEAA is weakly associated with severalsof the first two rows (a-j) and last two rows (k-s) relate sex to intrinsicfor blood tissue in all but one panel (j). The combined results acrossports 1 standard error and a Kruskal–Wallis testHorvath et al. Genome Biology  (2016) 17:171 Page 11 of 22cardiometabolic risk factors of CHD (such as hyper-tension, triglycerides, and CRP); (5) men exhibithigher epigenetic aging rates than women in blood,saliva, and brain samples, and (6) the rain forest habi-tat is significantly associated with extrinsic ageacceleration but not with intrinsic age acceleration inAfrican populations. Although precise understandingof the significance of epigenetic aging measures awaitsfurther elaboration, our principal findings may provideadditional context towards resolving several controver-sial, epidemiological paradoxes, including the Hispanicparadox, black–white mortality cross-over, the Tsimaneinflammation paradox, and the sex morbidity–mortalityparadox.Fig. 7 Effect of sex on the epigenetic age of brain tissue. Each panel depicin a forest plot shows the mean difference in epigenetic age between men anestimates from the respective studies into a single estimate, we applied a fixethe metafor R package [89]. a Gender did not have a significant effect on thethan other brain regions according to the epigenetic clock [43]. b When excludexhibit a significantly higher age acceleration than female brain regions (meansignificant even after adjusting for intra-subject correlations using a linear mixeHispanic paradoxThe lower level of IEAA in Hispanics echo the findingthat Hispanics in the US have a lower overall risk ofmortality than Caucasians despite having a disadvan-taged risk profile [45–48]. Our findings stratified bycountry of birth suggest that the lower intrinsic agingrate of Hispanics does not reflect biases arisingthrough immigration such as a “healthy immigrant ef-fect” (Additional file 3). Our finding regarding higherlevels of EEAA in Hispanics parallels the findings thatHispanics have higher levels of metabolic/inflamma-tory risk profiles [49] and that Hispanics have a lowerrelative CD4+ T cell percentage than Caucasians [50].Several articles have explored the question of why thets a forest plot resulting from the meta-analysis of sex effects. Each rowd women and a 95 % confidence interval. To combine the coefficientd-effects model weighted by inverse variance, which is implemented inepigenetic age of the cerebellum, which is known to age more slowlying cerebellar samples from the analysis, we find that male brain regionsdifference = 0.82, meta-analysis p = 3.1 × 10–5). The difference remainsd effects model (mean difference = 0.77, p = 0.0034)Horvath et al. Genome Biology  (2016) 17:171 Page 12 of 22immune system of Hispanics might differ from that ofCaucasians [51–53].Black–white mortality cross-overIn the US, the black–white mortality cross-over refers tothe reported pattern of lower mortality after the age of85 years among black men and women, compared towhites, despite their higher observed mortality rates atyounger ages [54–57]. Although we find no differencesin IEAA between African Americans and Caucasians atyounger ages, older African American adults from theBogalusa study had lower IEAA than their Caucasiancounterparts. This finding might reflect selective survivalof more robust individuals or other aspects of health andsystemic risk given its independence from common riskfactors for cardiovascular disease and type II diabetes mel-litus. Our finding regarding the lower EEAA of AfricanAmericans, compared to Caucasians, is consistent with thelonger leukocyte telomere lengths of African Americansrelative to those of Caucasians [3, 9]. Lastly, our flow cyto-metric data show that African Americans have a largernumber of naïve CD8+ T cells than Caucasians (Fig. 4t).Tsimane inflammation paradoxOur results regarding the low intrinsic aging rate inTsimane may help address another paradox (which werefer to as the Tsimane inflammation paradox), whereinhigh levels of inflammation and infection, and low HDLlevels, are not associated with accelerated cardiovascularaging [39]. The finding that Tsimane have decreasedlevels of IEAA has parallels to the following clinical/epi-demiological observations: even older Tsimane show littleevidence of chronic diseases common in high-incomecountries, like diabetes, atherosclerosis, asthma, and otherautoimmune disorders [39]. High levels of physical activityare maintained well into late adulthood [58].The finding that Tsimane have increased levels ofEEAA has parallels to the following observation: a life-time of diverse pathogen stresses, elevated inflammationand extensive immune activation, seems to lead to morerapid depletion of naïve CD4+ T cells and greater expres-sion of exhausted T cells, i.e. more rapid immunosenes-cence [39, 40, 59]. Infectious disease and high chronicinflammatory load contribute to the low life expectancy ofTsimane, 43.5 years at birth during the period 1950–1989,and 54.1 years during 1990–2002 [40, 60].Sex morbidity–mortality paradoxThe sex morbidity–mortality paradox was first describedin the 1970s and refers to the observation that women pos-sess a lower age-adjusted mortality rate compared to mendespite a higher suffering from a higher burden of co-morbid conditions [61, 62]. Most explanations focus ondifferences in lifestyle behaviors or healthcare utilization.However, marked sex differences in health and disabil-ity remain after controlling for differences in work-related behavior, smoking, obesity, and other behaviors[63]. Whereas other explanations attest to sex differ-ences in a variety of biomarkers, our epigenetic agingmarkers show robust and consistent male-biased vul-nerability in multiple tissues (blood, brain, and saliva)in all racial groups. Similar sex differences in blood-based epigenetic aging rates have also been reported inminors and teenagers [64].Strengths and limitationsOur study has several strengths including the analysisof 18 DNA methylation datasets (Tables 1 and 2), largesample sizes (almost 6000 samples), multiple tissues(blood, saliva, brain), access to unique populations(Tsimane Amerindians; rainforest hunter-gatherers andfarmers), two flow cytometric studies, and robustepigenetic biomarkers of aging. Our analysis of race/ethnicity also spanned seven different racial/ethnic groups(African American, Caucasian, Hispanic, Tsimane,East Asian, RHGs, and AGRs from Central Africa).Another strength is that our analysis of race/ethnicityinvolved two sources of DNA: blood and saliva. Lim-itations include the use of some datasets that arecross-sectional as opposed to longitudinal datasetsand the fact that both IEAA and EEAA rely on im-puted blood cell counts based on DNA methylationlevels. Fortunately, the imputed blood cell counts arequite accurate (Additional file 2). Our results re-ported here concerning ethnic/racial differences inblood cell counts are supported both by our twoflow cytometric datasets and by the literature. How-ever, these measured data are not fully reflective ofthe breakdown of blood cell types, representing onlyT and B cells.ConclusionOur exploratory study demonstrates that epigeneticaging rates differ between different racial/ethnic groupsand between men and women. Further, intrinsic epigen-etic aging rates tend to have insignificant associationswith well-studied risk factors of CHD whereas extrinsicaging rates tend to have significant (but weak) associ-ations with several pro-inflammatory risk factors.While racial/ethnic differences have previously beenobserved in DNA methylation levels [44], we are thefirst to directly compare epigenetic aging rates acrossdifferent racial/ethnic groups. Our derived intrinsicand extrinsic epigenetic aging rates in blood offer anindependent glimpse into biological aging that incor-porates genetics and the environment and providespotential insight into a number of epidemiologicalparadoxes. The application of genome-wide DNAm-Horvath et al. Genome Biology  (2016) 17:171 Page 13 of 22based epigenetic analysis to understand race/ethnicand sex disparities in biological aging is novel and of-fers an important perspective that complements exist-ing approaches based on other biomarkers. Future studieswill need to confirm our findings with longitudinal designsand to extend the epigenetic age analysis to other tissuesand organs.MethodsWe differentiate groups according to “race/ethnicity,”mindful about existing controversies over rigid racialdefinitions. Our use of these terms reflects self-identified group membership based on macro-categoriescommonly employed in censuses, human genetics,demography, and epidemiology. The term race/ethnicitythus combines elements of genetic ancestry, populationhistory, and culture.DNA methylation age and epigenetic clockAll of the described epigenetic measures of aging andage acceleration are implemented in our freely availablesoftware. The epigenetic clock is defined as a predictionmethod of age based on the DNAm levels of 353 CpGs.Predicted age, referred to as DNAm age, correlates withchronological age in sorted cell types (CD4+ T cells, mono-cytes, B cells, glial cells, neurons), tissues, and organs, in-cluding: whole blood, brain, breast, kidney, liver, lung, saliva[20]. Mathematical details and software tutorials for theepigenetic clock can be found in the Additional files of [20].An online age calculator can be found at our webpage(https://dnamage.genetics.ucla.edu).Intrinsic versus extrinsic measures of epigenetic ageacceleration in bloodEmpirical studies show that DNAm has a relatively weakcorrelation with various measures of white blood cellcounts [31], which probably reflects the fact that dozensof different tissue and blood cell types were used to de-fine DNAm age. However, we find it useful to explicitlydefine another measure of age acceleration that is com-pletely independent of blood cell counts as described inthe following. We distinguish intrinsic from extrinsicmeasures of epigenetic age acceleration in whole bloodaccording to their relationship with blood cell counts. Ameasure of intrinsic epigenetic age acceleration (IEAA)measures “pure” epigenetic aging effects that are notconfounded by differences in blood cell counts. Ourmeasure of IEAA is defined as the residual resultingfrom a multivariate regression model of DNAm age onchronological age and various blood immune cell counts(naïve CD8+ T cells, exhausted CD8+ T cells, plasma Bcells, CD4+ T cells, natural killer cells, monocytes, andgranulocytes). The measure of IEAA is an incompletemeasure of the age-related functional decline of theimmune system because it does not track age-relatedchanges in blood cell composition, such as the decreaseof naïve CD8+ T cells and the increase in memory orexhausted CD8+ T cells [36–38].We defined a measure of EEAA that only applies towhole blood and aims to measure epigenetic aging inimmune-related components in two steps. First, weformed a weighted average of the epigenetic age measurefrom Hannum et al. [19] and three estimated measuresof blood cells for cell types that are known to changewith age: naïve (CD45RA + CCR7+) cytotoxic T cells;exhausted (CD28-CD45RA-) cytotoxic T cells; andplasma B cells using the approach by Klemera Doubal[65]. Second, we defined the measure of EEAA as the re-sidual resulting from a univariate model that regressedthe weighted average on chronological age. By definition,our measure of EEAA has a positive correlation with theamount of exhausted CD8+ T cells and plasmablast cellsand a negative correlation with the amount of naïveCD8+ T cells. Blood cell counts were estimated based onDNA methylation data. EEAA tracks both age-relatedchanges in blood cell composition and intrinsic epige-netic changes. In most blood datasets, EEAA has a mo-derate correlation (r = 0.5) with IEAA. We note that, bydefinition, none of our three measures of epigenetic ageacceleration are associated with the chronological age ofthe participant at the time of blood draw.Relationship to mortality predictionAlthough the epigenetic clock method was only pub-lished in 2013, there is already a rich body of literaturethat shows that it relates to biological age. Using fourhuman cohort studies, we previously demonstrated thatboth the Horvath and Hannum epigenetic clocks arepredictive of all-cause mortality [23]. Published resultsin Marioni et al. [23] show that DNAm age adjusted forblood cell counts (i.e. IEAA) is prognostic of mortality infour cohort studies. We recently expanded our originalanalysis by analyzing 13 different cohorts (including threeracial/ethnic groups) and by evaluating the prognosticutility of both IEAA and EEAA. All considered measuresof epigenetic age acceleration were predictive of age atdeath in univariate Cox models (pAgeAccel = 1.9 × 10–11,pIEAA = 8.2 × 10–9, pEEAA = 7.5 × 10–43) and multivariateCox models adjusting for risk factors and pre-existingdisease status (pAgeAccel = 5.4 × 10–5, pIEAA = 5.0 × 10–4,pEEAA = 3.4 × 10–19) where the latter adjusted for chrono-logical age, body mass index, education, alcohol, smokingpack years, recreational physical activity, and prior historyof disease (diabetes, cancer, hypertension). These re-sults will be published elsewhere. Further, the offspringof centenarians age more slowly than age matched con-trols according to Age Accel and IEAA [26] whichstrongly suggests that these measures relate to heritableHorvath et al. Genome Biology  (2016) 17:171 Page 14 of 22components of biological age. Two independent re-search groups have shown that epigenetic age acceler-ation predicts mortality [24, 25].Description of the blood datasets listed in Table 1All data presented in this article have been made publiclyavailable as indicated in the column “Available” of Table 1.Dataset 1: Women’s Health Initiative (WHI)Participants included a subsample of participants of theWHI study, a national study that began in 1993 whichenrolled postmenopausal women between the ages of 50and 79 years into either one of two three randomizedclinical trials [66]. None of these women had CHD atbaseline but about half of these women had developedCHD by 2010. Women were selected from one of twoWHI large subcohorts that had previously undergonegenome-wide genotyping as well as profiling for sevencardiovascular disease related biomarkers including totalcholesterol, HDL, LDL, triglycerides, CRP, creatinine,insulin, and glucose through two core WHI ancillarystudies [67]. The first cohort is the WHI SNP HealthAssociation Resource (SHARe) cohort of minorities thatincludes >8000 African American women and >3500Hispanic women. These women were genotyped throughWHI core study M5-SHARe (www.whi.org/researchers/data/WHIStudies/StudySites/M5) and underwent bio-marker profile through WHI Core study W54-SHARe(…data/WHIStudies/StudySites/W54). The second cohortconsists of a combination of European Americans fromthe two Hormonal Therapy trials selected for GWAS andbiomarkers in core studies W58 (…/data /WHIStudies/StudySites/W58) and W63 (…/data/WHIStudies/Study-Sites/W63). From these two cohorts, two sample sets wereformed. The first (sample set 1) is a sample set of 637CHD cases and 631 non-CHD cases as of 30 September2010. The second sample set (sample set 2) is a non-overlapping sample of 432 cases of CHD and 472 non-cases as of 17 September 2012. The ethnic groups differedin terms of the age distribution in the sense that Caucasianwomen tended to be older. Therefore, we randomlyremoved 80 % of the Caucasian women who were olderthan 65 years when it came to the direct comparisonsreported in our figures. This resulted in a total sample sizeof 1462 women, comprising 673 African Americans, 353Caucasians, and 433 Hispanics. There was no significantdifference in age between the three ethnic groups. How-ever, we kept all of the samples in our analysis of clinicalcharacteristics, such as future CHD status and baselinecharacteristics such as education, hypertension, diabetes,and smoking, in order to ensure that sufficient sample sizeswere available for these analyses. Our results are highlyrobust with respect to using the smaller or larger versionsof the datasets. All results are qualitatively the same for thetwo versions of the datasets. We acknowledge a potentialfor selection bias using the above-described samplingscheme in WHI but suspect if such bias is present it isminimal. First, some selection bias is introduced byrestricting our methylation profiling at baseline to womenwith GWAS and biomarker data from baseline as well,given the requirement that these participants must havesigned the WHI supplemental consent for broad sharing ofgenetic data in 2005. However, we believe that selectionbias at this stage is minimized by the inclusion of partici-pants who died between the time of the start of the WHIstudy and the time of supplemental consent in 2005, whichresulted in the exclusion of only ~6–8 % of all WHIparticipants. Nevertheless, participants unable or un-willing to sign consent in 2005 may not represent arandom subset of all participants who survived to 2005.Second, some selection bias may also occur if similargross differences exist in the characteristics of partici-pants who consented to be followed in the two WHIextension studies beginning in 2005 and 2010 com-pared to non-participants at each stage. We believethese selection biases if present have minimal effects onour effect estimates. Data are available from the pagehttps://www.whi.org/researchers/Stories/June%202015%20WHI%20Investigators'%20Datasets%20Released.aspx,see the link https://www.whi.org/researchers/data/Documents/WHI%20Data%20Preparation%20and%20Use.pdf.Dataset 2: BogalusaWe analyzed the blood DNA methylation levels of 968participants (680 Caucasians, 288 African Americans;age range = 28–51.3 years) from the Bogalusa Heartstudy [68] who were examined in Bogalusa, Louisianaduring 2006–2010 for cardiovascular risk factors. Allparticipants in this study gave informed consent at eachexamination. Study protocols were approved by the In-stitutional Review Board (IRB reference no. 12-395283)of the Tulane University Health Sciences Center. DNAwas extracted from 1106 whole blood samples using thePureLink Pro 96 Genomic DNA Kit (LifeTechnology, CA,USA) following the manufacturer’s instructions. The Infi-nium HumanMethylation450 BeadChip (Methy450K) wasused for whole genome DNA methylation analysis.All the samples were processed at the Microarray CoreFacility, University of Texas Southwestern Medical Centerat Dallas, Texas. For DNA methylation analysis, 750 nggenomic DNA from each participant was bisulphiteconverted using the EZ-96 DNA Methylation Kit (ZymoResearch, CA, USA) and the efficiency of the bisulphiteconversion was confirmed by built-in controls on theMethy450K array. The methylation profile of each individ-ual was measured by processing 4 μL of bisulphite-converted DNA, at a concentration of 50 ng/μL, on aMethy450K array. The bisulphite-converted DNA wasHorvath et al. Genome Biology  (2016) 17:171 Page 15 of 22amplified, fragmented, and hybridized to the array. Thearrays were scanned on an Illumina HiScan scanner andthe raw methylation data were extracted using Illumina’sGenome Studio methylation module. Data cleaning proce-dures were undertaken using R package “minfi” [69],generating quality control report, finding sample outliers,cell counts estimation, and annotation accessing. TheR package wateRmelon [70] was used for β-valuenormalization and quality control. For correction ofsystematic technical biases in the 450 K assay, β-valuenormalization was performed by the “dasen” function,in which type I and type II intensities and methylatedand unmethylated intensities will be quantile normal-ized separately after backgrounds equalization of type Iand type II. The R package ChAMP [71] was used forbatch effect analysis and correction with “champ.SVD”and “champ.runCombat” functions. The clinical variablesand participant characteristics are defined in the captionsof the respective Additional files.The are available from https://biolincc.nhlbi.nih.gov/studies/bhs/.Dataset 3: blood from Hispanics and Caucasians of PEGThe Parkinson’s disease, Environment, and Genes (PEG)case-control study aims to identify environmental risk fac-tors (e.g. neurotoxic pesticide exposures) for Parkinson’sdisease.The PEG study is a large population-based study ofParkinson’s disease of mostly rural and township residentsof California’s central valley [72]. Here we only used dis-eased participants from wave 1 (PEG1). Since all partici-pants of dataset 3 had Parkinson’s disease, disease statuscould not confound associations with epigenetic aging.Medication status was not associated with epigenetic ageacceleration. The data are available from Gene ExpressionOmnibus.Dataset 4: saliva samples from PEGThis novel dataset comes from the PEG study (describedabove). Since PD disease status did not relate to epigen-etic age acceleration in these data, we ignored it in theanalysis. However, our findings are unchanged after in-corporating PD status in a multivariate model. Abouthalf of the samples overlapped with those of dataset 3,which is why we could correlate epigenetic age acceler-ation between blood and saliva.Datasets 5 and 6: blood from Tsimane, Hispanics, andCaucasiansDatasets 5 and 6, which were collected and generated inthe same way, only differ in terms of the chronologicalages. All participants in dataset 5 are older than 35 yearswhile those in dataset 6 are younger or equal to 35 years.The dataset involved three different ethnic groups:Tsimane Amerindians, Hispanics living in the US, andCaucasians living in the US. Fasting whole-blood sam-ples were collected from Tsimane via venipuncture infield villages in the vicinity of San Borja, Bolivia as a partof the annual biomedical data collection for a longitu-dinal project on aging during 2004–2009 (TsimaneHealth and Life History Project). Manual complete bloodcounts were conducted using a hemocytometer, erythro-cyte sedimentation rate was calculated following theWestergren method, and hemoglobin was analyzed witha QBC Autoread Plus Dry Hematology System (DruckerDiagnostics, Port Matilda, PA, USA). Specimens werestored in liquid nitrogen until transfer to the US on dryice, where they were stored at –80 °C. All participantsprovided written and informed consent; study protocolsand procedures were approved at the individual, village,and Tsimane government level, as well as by the Univer-sity of California, Santa Barbara and University of NewMexico Institutional Review Boards (IRB Reference num-bers 14-0604 and 07-157, respectively). Specimens wereshipped on dry ice to the University of Southern Californiafor extraction. The same core facility provided blood sam-ples that were collected at the same time and stored in thesame condition as Hispanic participants living in the US.The DNA samples from all participants (Caucasians,Hispanics, Tsimane) were randomized across the Illuminachips to avoid confounding due to chip effects. For ourage prediction analysis, we used background correctedbeta values resulting from Genome Studio.Hispanics for datasets 5 + 6: Participant recruitment:Participation in the BetaGene study was restricted toMexican Americans from families of a proband with ges-tational diabetes mellitus (GDM) diagnosed within theprevious 5 years. Probands were identified from the patientpopulations at Los Angeles County/USC Medical Center,OB/GYN clinics at local hospitals, and the Kaiser Perma-nente health plan membership in Southern California.Probands qualified for participation if they: (1) were ofMexican ancestry (defined as both parents and ≥3/4 ofgrandparents Mexican or of Mexican descent); (2) had aconfirmed diagnosis of GDM within the previous 5 years;(3) had glucose levels associated with poor pancreatic β-cell function and a high risk of diabetes when not pregnant;and (4) had no evidence of β-cell autoimmunity by GAD-65 antibody testing. Recruitment targeted two generalfamily structures using siblings and/or first cousins ofGDM probands, all with fasting glucose levels <126 mg/dl(7 mM): (1) at least two siblings and three first cousinsfrom a single nuclear family; or (2) at least five siblingsavailable for study. Using information from the proband todetermine preliminary eligibility, siblings and first cousinswere invited to participate in screening and, if eligible, de-tailed phenotyping (below) and collection of DNA. Avai-lable parents and connecting uncles and aunts were askedHorvath et al. Genome Biology  (2016) 17:171 Page 16 of 22to provide DNA and had a fasting glucose determination.In addition, women of Mexican ancestry who have gonethrough pregnancy without GDM, as evidenced by aplasma or serum glucose level <120 mg/dl after a 50 g oralglucose screen for GDM, were also collected. Recruitmentcriteria for control probands were similar to that of theGDM probands, but were also selected to be age, BMI, andparity-matched to the GDM probands. Unrelated samplesfor the present methylation analysis were selected ran-domly from all BetaGene participants. The BetaGeneprotocol (HS-06-00045) has been approved by the Institu-tional Review Boards of the USC Keck School of Medicine.Dataset 7: blood from East Asians and CaucasiansHere we downloaded the publicly available DNA methy-lation data from GSE53740 [73]. Since we found thatprogressive supranuclear palsy (PSP) had a significanteffect on epigenetic age acceleration, we removed PSPsamples from the analysis. Further, we focused on com-paring East Asians to Caucasians since other racial/eth-nic groups were represented by fewer than 10 samples.Dataset 8: blood from African populationsWe used blood methylation data from [42]. We studiedperipheral whole-blood DNA from a total of 256 sam-ples (for which the chronological age at the time ofblood draw was available).As detailed in Fagny et al. [42], the samples come fromseven populations located across the Central Africanbelt. These populations can be divided into two maingroups: RHG populations, historically known as “pyg-mies,” who have traditionally relied on the equatorialforest for subsistence and who live close to, or within,the forest; and AGR populations, living either in rural/urban deforested regions or in forested habitats in whichthey practice slash-and-burn agriculture. Informed con-sent was obtained from all participants and from bothparents of any participants under the age of 18 years.Ethical approval for this study was obtained from theinstitutional review boards of Institut Pasteur, France(RBM 2008-06 and 2011-54/IRB/3).Dataset 9: cord blood samples from African Americans andCaucasiansThese 216 cord blood samples from 92 African Americanand 70 Caucasian participants come from a study that de-scribed racial differences in DNA methylation levels [44].Datasets 10 and 11Saliva samples from Caucasians and Hispanics. The datawere generated by splitting the data from [74] by sex,which reflected the use of these data in the developmentof the epigenetic clock software [20]. Note that thesedata were generated on the older Illumina platform(27 K array). Some of the data were used as training datain the development of the epigenetic clock, which mightbias the results. By contrast, the novel saliva data fromPEG (dataset 4) provide an unbiased analysis.Dataset 12: lymphoblastoid cell lines from Han Chinese,African Americans, and CaucasiansWe clustered the samples based on the interarray correl-ation. Since 51 samples were very distinct from theremaining samples, they were removed as potential out-liers. Disease status did not affect the estimates of DNAmage, which is why we ignored it.Description of brain datasetsWe collected brain datasets from six independent stu-dies to assess gender effect on epigenetic age acceler-ation. We focused on Caucasian samples since therewere insufficient numbers of other racial/ethnic groups.Study 1: brain DNA methylation data from a study ofAlzheimer’s disease study from [75], GEO accessionGSE59685. DNA methylation profiles of the cerebellum,entorhinal cortex, prefrontal cortex, and superiortemporal gyrus were available from 117 individuals.We ignored disease status since it was not associatedwith age acceleration.Study 2: brain DNA methylation data from neurologicallynormal participants from [76], GEO accession GSE15745.DNA methylation data of the cerebellum, frontal cortex,pons, and temporal cortex regions from up to 148neurologically normal participants of Europeanancestry [76].Study 3: cerebellar DNA methylation data from [77],GEO GSE38873. DNA methylation data from thecerebellum of 147 participants from a case-controlstudy (121 cases/32 controls) of psychiatric disorders.Since disease status did not affect DNAm age, weignored it.Study 4: prefrontal cortex samples from [78], GEOGSE61431. We analyzed 37 Caucasian participants(European ancestry).Study 5: frontal cortex and cerebellum from neurologicallynormal Caucasian participants from [79]. The DNAmethylation data and corresponding SNP data can befound in dbGAP, http://www.ncbi.nlm.nih.gov/gap(accession: phs000249.v2.p1). We only analyzed 209Caucasian participants who met our stringent qualitycontrol criteria. We excluded several putative outliersfrom the original dataset including three individuals whowere genotyped on a different platform, six participantstube was used to evaluate T lymphocyte subsets: CD45Horvath et al. Genome Biology  (2016) 17:171 Page 17 of 22who were outliers according to a genetic analysis(PC plot), and 13 participants who had the wronggender according to the gender prediction algorithmof the epigenetic clock software.Study 6: dorsolateral prefrontal cortex samples from 718Caucasian participants from the Religious Order Study(ROS) and the Memory and Aging Project (MAP). TheDNA methylation data are available at the followingwebpage https://www.synapse.org/#!Synapse:syn3168763.We focused on brain samples of Caucasian participantsfrom these two prospective cohort studies of agingthat include brain donation at the time of death [80].Additional details on the DNA methylation data canbe found in [81]. We were not able to evaluate theeffect of race/ethnicity on epigenetic age accelerationsince the dataset contained only 12 Hispanic samples(which did not differ significantly from Caucasians interms of epigenetic age). Further, we found no associationbetween disease status and epigenetic age acceleration,which is why we ignored disease status in our analysis.Preprocessing of Illumina Infinium 450 K arraysIn brief, bisulfite conversion using the Zymo EZ DNAMethylation Kit (ZymoResearch, Orange, CA, USA) aswell as subsequent hybridization of the HumanMethyla-tion450k Bead Chip (Illumina, San Diego, CA, USA), andscanning (iScan, Illumina) were performed according tothe manufacturers’ protocols by applying standard set-tings. DNA methylation levels (β values) were determinedby calculating the ratio of intensities between methylated(signal A) and unmethylated (signal B) sites. Specifically,the β value was calculated from the intensity of the meth-ylated (M corresponding to signal A) and unmethylated(U corresponding to signal B) sites, as the ratio of fluores-cent signals β =Max(M,0)/[Max(M,0) +Max(U,0) + 100].Thus, β values range from 0 (completely unmethylated) to1 (completely methylated) [82]. The epigenetic clocksoftware implements a data normalization step thatrepurposes the BMIQ normalization method fromTeschendorff [83] so that it automatically referenceseach sample to a gold standard based on type II probesas detailed in [20].Estimating blood cell counts based on DNA methylationlevelsWe estimate blood cell proportions using two differentsoftware tools. Houseman’s estimation method [84], whichis based on DNA methylation signatures from purifiedleukocyte samples, was used to estimate the proportions ofcytotoxic (CD8+) T cells, helper (CD4+) T, natural killer, Bcells, and granulocytes. The software does not allow us toidentify the type of granulocytes in blood (neutrophil, eo-sinophil, or basophil) but we note that neutrophils tend to(KO), CD8 (BV), CD45RA (F), CCR7 (PE), CD5 (ECD),CD56 (PC5), CD3 (APC-H7), CD4 (A594), CD28 (APC),CD27 (PC7). A second tube evaluated B lymphocyte sub-sets: CD45 (APC-H7), CD20 (V450), kappa (F), lambda(PE), CD23 (ECD), CD5 (PC5.5), CD19 (BV650), CD38(A594), CD10 (APC), CD27 (PC7), CD3 (APC-A700).Categories of circulating cells were quantified using apredefined population-based gating strategy based onbe the most abundant granulocyte (~60 % of all blood cellscompared with 0.5–2.5 % for eosinophils and basophils).The advanced analysis option of the epigenetic clock soft-ware [20] was used to estimate the percentage of exhaustedCD8+ Tcells (defined as CD28-CD45RA-) and the number(count) of naïve CD8+ T cells (defined as (CD45RA +CCR7+) as described in [31].Flow cytometric data from the Long Life Study of the WHIWhile our DNA methylation data from the WHI wereassessed at baseline, the flow cytometric data were mea-sured 14.6 years after baseline. Between March 2012 andMay 2013, a subset of WHI participants were enrolled inthe Long Life Study (LLS) and additional biospecimens,physiometric, and questionnaire data were collected. Allsurviving Hormone Trial participants followed through2010 and all African American and Hispanic/Latinoparticipants from the SNP Health Association Resource(WHI-SHARe) sub-cohort were included if CVD bio-marker from WHI baseline exam and genome-widegenotyping (GWAS) data were available and if theywere at least 63 years old by 1 January 2012. Womenwho were either unable to provide informed consent(e.g. dementia) or those residing in an institution (e.g.skilled nursing facility) were excluded. Of a total of 14,081eligible WHI participants, 9242 women consented to par-ticipate, 7875 were enrolled, and 7481 underwent success-ful blood draws. Blood was collected at locations acrossthe US using a standardized protocol between March2012 and May 2013 (Examination Management Services,Inc.) Fresh peripheral blood samples were packaged inStyrofoam with cold packs and were sent overnight to acentral testing facility in Seattle.A random sample of 600 residual fresh peripheral bloodspecimens (single tube, following CBC analysis) was trans-ported to the University of Washington Medical Center’s(UWMC’s) flow cytometry laboratory and high-sensitivity,multi-parameter flow cytometry was performed utilizing amodified four-laser, multi-color Becton-Dickinson (BD;San Jose, CA, USA) LSRII flow cytometer. All of the flowcytometry studies were performed within 72 h of samplecollection between June 2012 and February 2013. A singleestablished gating strategies for both T lymphocyte [85]and B lymphocyte [86] subsets.AIDS Cohort Study (MACS). The data were generatedas described in [87]. Briefly, human peripheral bloodAdditional file 7: Educational level versus age acceleration in theHorvath et al. Genome Biology  (2016) 17:171 Page 18 of 22mononuclear cell (PBMC) samples were isolated fromfresh blood samples and either stained for flow cytome-try analysis or used for genomic DNA isolation. DNAwas isolated from 1 × 106 PBMC using Qiagen DNeasyblood and tissue mini spin columns. Quality of DNAsamples was assessed using Nanodrop measurementsand accurate DNA concentrations were measured using aQubit assay kit (Life Technology). Cryopreserved PBMCobtained from the repository were thawed and assayed forviability using trypan blue. The mean viability of the sam-ples was 88 %. Samples were stained for 30 min at 4 °Cwith the following antibody combinations of fluorescentlyconjugated monoclonal antibodies using the manufacturersrecommended amounts for 1 million cells: tube 1: CD57FITC (clone HNK-1), CD28 phycoerythrin (PE, L293),CD3 peridinin chlorophyll protein (PerCP,SK7), CD45RAphycoerythrin cyanine dye Cy7 tandem (PE-Cy7, L48),CCR7 Alexa Fluor 647 (AF647, 150503), CD8 allophyco-cyanin H7- tandem (APC-H7, SK1) and CD4 horizonV450 (V450, RPA-T4); tube 2: HLA-DR FITC (L243),CD38 PE (HB7), CD3 PercP, CD45RO PE-Cy7 (UCHL-1),CD95-APC(DXZ), CD8 APC-H7, and CD4 V450); tube 3:CD38 FITC (HB7), IgD PE (1A6–2), CD3 PerCP, CD10PE-Cy7 (HI10a), CD27 APC (eBioscience, clone 0323, SanDiego, CA), CD19 APC-H7 (SJ25C1) and CD20 V450(L27). Antibodies were purchased from BD Biosciences,San Jose, CA (BD) except as noted. Stained samples werewashed twice with staining buffer and run immediately onan LSR2 cytometer equipped with a UV laser (BD, SanJose, CA, USA) for the detection of 4′,6-diamidino-2-phe-nylindole dihydrochloride (DAPI) which was used as aviability marker at a final concentration of 0.1 ug/mL.Lineage gated isotype controls to measure non-specificbinding were run and used CD3, CD4, and CD8 for T-cellsor CD19 for B-cells. Fluorescence minus one controls(FMO) were also utilized to assist gating and cursor setting.A range of 20,000–100,000 lymphocytes were acquired andanalyzed per sample using the FACSDiva software package(BD, San Jose, CA, USA).Additional filesAdditional file 1: Lymphoblastoid cell lines from Han Chinese, Caucasians,and African Americans. A Gray line corresponds to a natural spline regressionthrough Caucasian samples. Age acceleration was defined as residual withFlow cytometric data from the MACS cohortAs part of Additional file 2, we validated imputed bloodcell counts using flow cytometric data and DNA methy-lation data collected from men of the Multi-Centerrespect to this line. B Marginally significant evidence that African American’sare younger than other ethnic groups. (PDF 33 kb)Additional file 2: Accuracy of imputed blood cell counts. Here we used anindependent dataset, which was not used to develop estimators of blood cellcounts based on DNA methylation data, to evaluate the accuracy of theimputed blood cell counts. For each participant, both flow cytometricmeasures and Illumina Inf450 data were available from 96 participants asdescribed in [88]. A-G The scatter plots depict the predicted abundance ofblood cell count (based on DNA methylation levels) versus the correspondingobserved flow cytometric measurement (y-axis). Each panel reports a robustcorrelation coefficient (biweight midcorrelation) and a corresponding p value.The Houseman method was used to impute (A) CD8+ T cells, (B) CD4+ T, (C)B cells. The epigenetic clock software was used for imputing (D) naïve CD8+ Tcells, (E) naïve CD4 + T cells, (F) plasma blasts, and (G) exhausted CD8+ T cells.H, I Another flow cytometric dataset was used to test for ethnic differences innaïve CD4+ T cells. The y-axis shows the log transformed flow cytometricmeasurement of naïve CD4+ T cells (adjusted for age). Specifically, the y-axisreports the residual resulting from regressing log(naïve CD4+ T cellabundance) on chronological age. H Findings for HIV– participants (198Caucasians versus 34 Hispanics). I Findings for HIV+ participants (101Caucasians, 58 Hispanics). Stouffer’s meta-analysis across the two strata(HIV+ and HIV– strata) shows that Hispanics have significantly fewernaïve CD4+ T cells (Stouffer’s p = 0.030, Stouffer’s Z = (1.75 + 1.31)/sqrt(2)).(PDF 130 kb)Additional file 3: Epigenetic age acceleration in Hispanics versus countryof residence in the WHI. Each column corresponds to different measure ofage acceleration: (A, D) age acceleration residual, (B, E) IEAA (C, F) EEAA.(A-C, first row) results for “country of birth” (x-axis). (D-F, second row) resultsfor “country of residence” at age 35 years, which was defined by combiningtwo variables country of birth and “living in the US at age 35.” The left-mostbar corresponds to Hispanic women who were born outside the US andlived outside the US at age 35 years, the middle bar corresponds to Hispanicwomen who were born outside the US but lived already in the US at theage of 35 years; the right-most bar reports results for women who wereborn in the US and lived in the US at age 35 years. Incidentally, all of thesepostmenopausal Hispanic women lived in the US at the age of the blooddraw. As a caveat, we mention the relatively small group sizes (small graynumbers underneath the bars). (PDF 3 kb)Additional file 4: Educational level versus age acceleration in the WHI.Each row relates educational level (x-axis) to three respective measures ofepigenetic age acceleration: (A-D) Age Accel., (E-H) IEAA, and (I-L) EEAA.The columns correspond to different groups of women from the WHI.The first, second, third, and fourth columns report findings for (A, E, I) allwomen, (B, F, J) Caucasians, (C, G, K) African Americans, and (D, H, L)Hispanics, respectively. Each bar plot reports the mean values, 1 standarderror, and the p value from a non-parametric group comparison test (Kruskal–Wallis). Education was assessed using the form “Demographics and StudyMembership.” We find that education predicts future EEAA. (PDF 6 kb)Additional file 5: Hypertension status versus age acceleration in theBogalusa study. Each row relates hypertension status (x-axis) to threerespective measures of epigenetic age acceleration: (A-C) Age Accel.,(D-F) IEAA, and (G-I) EEAA. The columns correspond to different groups. Thefirst, second, and third columns report findings for (A, D, G) all participants, (B,E, H) Caucasians, (C, F, I) African Americans, respectively. Each bar plot reportsthe mean values, 1 standard error, and the p value from a non-parametricgroup comparison test (Kruskal–Wallis). Hypertension status was defined asmeeting any of the three conditions: (1) blood pressure > =140/90; (2) takingmedication; or (3) having been diagnosed as having hypertension. (PDF 4 kb)Additional file 6: Type II diabetes status versus age acceleration in theBogalusa study. Each row relates type II diabetes status (x-axis) to threerespective measures of epigenetic age acceleration: (A-C) Age Accel.,(D-F) IEAA, and (G-I) EEAA. The columns correspond to different groups.The first, second, and third columns report findings for (A, D, G) allparticipants, (B, E, H) Caucasians, (C, F, I) African Americans, respectively.Each bar plot reports the mean values, 1 standard error, and the p valuefrom a non-parametric group comparison test (Kruskal–Wallis). Type 2diabetes status was defined as fasting glucose > =126 mg/dl or takingdiabetes medication. (PDF 3 kb)Bogalusa study. Each row relates educational level (x-axis) to threerespective measures of epigenetic age acceleration: (A-C) Age Accel.,Horvath et al. Genome Biology  (2016) 17:171 Page 19 of 22(D-F) IEAA, and (G-I) EEAA. The columns correspond to different groups.The first, second, and third columns report findings for (A, D, G) allparticipants, (B, E, H) Caucasians, (C, F, I) African Americans, respectively.Each bar plot reports the mean values, 1 standard error, and the p valuefrom a non-parametric group comparison test (Kruskal–Wallis). Educationwas grouped as follows: group 1 = grades 1–7; group 2 = grades 8–9; group3 = grades 10–12; group 4 = vocational/tech training; group 5 = college;group 6 = postgraduate. (PDF 5 kb)Additional file 8: Robustness analysis with respect to other epigeneticbiomarkers of aging in the WHI. A-C Results for the Horvath method when 47out of 353 CpGs were removed from the epigenetic clock (because they arein the vicinity of a SNP). Since none of the remaining clock CpGs are near aSNP, the resulting age acceleration is not trivially related to race/ethnicity. ADNA methylation age versus chronological age. B Ethnicity versus ageacceleration (defined as residual resulting from regressing DNAm age onchronological age). C Intrinsic epigenetic age acceleration versus ethnicity. D,E Alternative epigenetic biomarker of aging based on 589 age-related CpGsfrom Teschendorff [13]. The biomarker was defined using the following steps.First, the DNA methylation levels of each CpGs were standardized (to meanzero and variance 1). Second, a weighted average was formed by multiplyingeach CpG by the T test statistic from the chronological age relationship basedon the table from the original reference. Third, the weighted average wasregressed on chronological age to arrive at a residual. The resulting residual isreferred to as extrinsic measure of age acceleration since it was not adjustedfor blood cell counts. Fourth, the resulting measure was regressed onestimated blood cell counts (analogous to those used for IEAA) in orderto arrive an intrinsic measure of age acceleration. F, G Epigeneticmeasures of age acceleration using the Hannum method 71 CpGs [19].D, F Results for intrinsic measures, i.e. measures of age acceleration thatadjust both for blood cell counts and chronological age. E, G reportsextrinsic measures, i.e. no adjustment for imputed blood cell counts.Each bar plot depicts 1 standard error and reports the results from aKruskal–Wallis test. (PDF 55 kb)Additional file 9: Demographic and physiologic characteristics ofwomen from the WHI. Case-control status refers to CHD. Two designswere used to select samples: case/control and case-cohort. (DOC 48 kb)AcknowledgementsWe would like to acknowledge the following WHI investigators. Program Office(National Heart, Lung, and Blood Institute, Bethesda, MD, USA): Jacques Rossouw,Shari Ludlam, Dale Burwen, Joan McGowan, Leslie Ford, and Nancy Geller.Clinical Coordinating Center (Fred Hutchinson Cancer Research Center,Seattle, WA, USA): Garnet Anderson, Ross Prentice, Andrea LaCroix, andCharles Kooperberg. Investigators and Academic Centers: JoAnn E. Manson(Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA);Barbara V. Howard (MedStar Health Research Institute/Howard University,Washington, DC, USA); Marcia L. Stefanick (Stanford Prevention Research Center,Stanford, CA, USA); Rebecca Jackson (The Ohio State University, Columbus, OH,USA); Cynthia A. Thomson (University of Arizona, Tucson/Phoenix, AZ, USA);Jean Wactawski-Wende (University at Buffalo, Buffalo, NY, USA); MarianLimacher (University of Florida, Gainesville/Jacksonville, FL, USA); RobertWallace (University of Iowa, Iowa City/Davenport, IA, USA); Lewis Kuller(University of Pittsburgh, Pittsburgh, PA, USA); Sally Shumaker (Wake ForestUniversity School of Medicine, Winston-Salem, NC, USA). Women’s HealthInitiative Memory Study (Wake Forest University School of Medicine,Winston-Salem, NC): Sally Shumaker.FundingThis study was supported by NIH/NHLBI 60442456 BAA23 (Assimes, Absher,Horvath), National Institutes of Health NIH/NIA 1U34AG051425-01 (Horvath). TheWHI program is funded by the National Heart, Lung, and Blood Institute, NationalInstitutes of Health, U.S. Department of Health and Human Services throughcontracts HHSN268201100046C, HHSN268201100001C, HHSN268201100002C,HHSN268201100003C, HHSN268201100004C, and HHSN271201100004C. The PEGdata were supported by NIEHS RO1ES10544 (Ritz) and NIEHS R21 ES024356(Horvath, Ritz). Gurven and Trumble were funded by NIH/NIA R01AG024119 andR56AG02411. The Religious Order study and Rush Memory and Aging Project(brain dataset 6) were funded by P30AG10161, R01AG17917, RF1AG15819, andR01AG36042.One of our flow datasets was collected by the Multicenter AIDS Cohort Study(MACS) at UCLA (Principal Investigators, Roger Detels and Otoniel Martinez-Maza),U01-AI35040. The MACS is funded primarily by the National Institute of Allergyand Infectious Diseases (NIAID) with additional co-funding from the NationalCancer Institute (NCI P30 CA016042), the National Institute on Drug Abuse (NIDA5P30 AI028697), the National Institute of Mental Health (NIMH), the NationalInstitute on Aging (NIA Grant 1RO1-AG-030327 by BDJ), and UL1-TR000424(JHU CTSA). The content is solely the responsibility of the authors and does notnecessarily represent the official views of the National Institutes of Health or do-nors to the David Geffen School of Medicine. The funders had no role in studydesign, data collection and analysis, decision to publish, or preparation ofthe manuscript.Availability of data and materialsOur DNA methylation data are publicly available through gene expressionomnibus (GEO) accession numbers: GSE72775, GSE78874, GSE72773, andGSE72777. Further, the WHI and Bogalusa datasets are available through dbGAP(http://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000200.v10.p3 and https://biolincc.nhlbi.nih.gov/studies/bhs/).African Populations: The genotyping data generated in this study have beendeposited in the European Genome-Phenome Archive under accession codesEGAS00001000605, EGAS00001000908 and EGAS00001001066. The DNA methy-lation data generated in this study have been deposited in the EuropeanGenome-Phenome Archive under accession code EGAS00001001066.The GSE numbers for the brain datasets are as follows: GSE59685,GSE15745, GEO GSE38873, and GEO GSE61431. Brain data 5 can be foundat http://www.ncbi.nlm.nih.gov/gap (accession: phs000249.v2.p1) and braindata 6 at https://www.synapse.org/#!Synapse:syn3168763.Authors’ contributionsSH conceived of the study, developed the methods, analyzed the data, andwrote the first draft of the article. MG, BT, HK, and HA contributed the DNAfrom the Tsimane Amerindians and interpreted the findings. ML, BR, and BChelped to interpret the data and edited the article. BR and SH contributedthe PEG DNA methylation data. AL analyzed the brain datasets. DS, SL, andWC contributed the DNA methylation data from the Bogalusa Heart Study.SH, PT, DA, and TA contributed the DNA methylation data from the WHI. KEand AR contributed flow cytometric data from the WHI LLS. BJ and TRcontributed flow data from the MACS. LQM, MF and MSK contributed DNAmdata from African hunter gatherers. All authors helped interpret the data andedited the manuscript. All authors read and approved the final manuscript.Competing interestsThe authors declare that they have no competing interests.Ethics approval and consent to participateThis study was reviewed by the UCLA institutional review board (IRB#13-000671and IRB#14-000061) as well as the University of California Santa Barbara andUniversity of New Mexico Institutional Review Boards (IRB Reference numbers14-0604 and 07-157 respectively).Author details1Human Genetics, David Geffen School of Medicine, University of CaliforniaLos Angeles, Los Angeles, CA 90095, USA. 2Biostatistics, School of PublicHealth, University of California Los Angeles, Los Angeles, CA 90095, USA.3Department of Anthropology, University of California Santa Barbara, SantaBarbara, CA 93106, USA. 4Department of Anthropology, University of NewMexico, Albuquerque, NM 87131, USA. 5Department of Preventive Medicineand Institute for Genetic Medicine, Keck School of Medicine, University ofSouthern California, Los Angeles, CA 90089, USA. 6Department ofEpidemiology, Fielding School of Public Health, University of California LosAngeles, Los Angeles, CA 90095, USA. 7Longitudinal Studies Section,Translational Gerontology Branch, National Institute on Aging, NationalInstitutes of Health, Baltimore, MD 21224, USA. 8Department of Medicine,Division of Hematology/Oncology, AIDS Institute, University of California LosAngeles, Los Angeles, CA, USA. 9Department of Epidemiology, TulaneUniversity, New Orleans, LA 70112, USA. 10Unit of Human EvolutionaryGenetics, Centre National de la Recherche Scientifique, URA3012, URA3012Institut Pasteur, Paris 75015, France. 11Department of Biostatistics, Harvard THChan School of Public Health and Department of Computational Biology andBiostatistics, Dana-Farber Cancer Institute, Boston, MA 02115, USA. 12Centrechromatin domains. Genome Res. 2010;20:434–9.Horvath et al. Genome Biology  (2016) 17:171 Page 20 of 2213. Teschendorff AE, Menon U, Gentry-Maharaj A, Ramus SJ, Weisenberger DJ,Shen H, et al. Age-dependent DNA methylation of genes that are suppressedin stem cells is a hallmark of cancer. Genome Res. 2010;20:440–6.14. Vivithanaporn P, Heo G, Gamble J, Krentz H, Hoke A, Gill M, Leistung C.Neurologic disease burden in treated HIV/AIDS predicts survival. Neurology.2010;75:1150–8.15. Horvath S, Zhang Y, Langfelder P, Kahn R, Boks M, van Eijk K, Ophoff RA.Aging effects on DNA methylation modules in human brain and bloodtissue. Genome Biol. 2012;13:R97.16. Alisch RS, Barwick BG, Chopra P, Myrick LK, Satten GA, Conneely KN, WarrenST. Age-associated DNA methylation in pediatric populations. Genome Res.2012;22:623–32.17. Johansson A, Enroth S, Gyllensten U. Continuous aging of the human DNAfor Molecular Medicine and Therapeutics, Child and Family Research Instituteand Department of Medical Genetics, University of British Columbia,Vancouver, BC V5Z 4H4, Canada. 13Department of Medicine, StanfordUniversity School of Medicine, Stanford, CA 94305, USA. 14VA Palo AltoHealth Care System, Palo Alto, CA, USA. 15Department of Epidemiology, FredHutchinson Cancer Research Center, University of Washington, Seattle, WA98109, USA. 16Department of Laboratory Medicine, University of Washington,Seattle, WA 98195, USA. 17HudsonAlpha Institute for Biotechnology,Huntsville, AL 35806, USA.Received: 6 July 2016 Accepted: 18 July 2016References1. Levine M, Crimmins E. Evidence of accelerated aging among AfricanAmericans and its implications for mortality. Soc Sci Med. 2014;118:27–32.2. Diez Roux AV, Ranjit N, Jenny NS, Shea S, Cushman M, Fitzpatrick A, SeemanT. Race/ethnicity and telomere length in the Multi-Ethnic Study ofAtherosclerosis. Aging Cell. 2009;8:251–7.3. Rewak M, Buka S, Prescott J, De Vivo I, Loucks EB, Kawachi I, Non AL,Kubzansky LD. Race-related health disparities and biological aging: doesrate of telomere shortening differ across blacks and whites? Biol Psychol.2014;99:92–9.4. Crimmins EM, Kim JK, Seeman TE. Poverty and biological risk: the earlier“aging” of the poor. J Gerontol A Biol Sci Med Sci. 2009;64:286–92.5. Arbeev KG, Butov AA, Manton KG, Sannikov IA, Yashin AI. Disability trends ingender and race groups of early retirement ages in the USA. SozPraventivmed. 2004;49:142–51.6. Ingram DK, Nakamura E, Smucny D, Roth GS, Lane MA. Strategy foridentifying biomarkers of aging in long-lived species. Exp Gerontol.2001;36:1025–34.7. Blackburn EH, Gall JG. A tandemly repeated sequence at the termini of theextrachromosomal ribosomal RNA genes in Tetrahymena. J Mol Biol.1978;120:33–53.8. Lin J, Epel E, Cheon J, Kroenke C, Sinclair E, Bigos M, et al. Analyses andcomparisons of telomerase activity and telomere length in human T and Bcells: insights for epidemiology of telomere maintenance. J ImmunolMethods. 2010;352:71–80.9. Hunt SC, Chen W, Gardner JP, Kimura M, Srinivasan SR, Eckfeldt JH,Berenson GS, Aviv A. Leukocyte telomeres are longer in AfricanAmericans than in whites: the National Heart, Lung, and Blood InstituteFamily Heart Study and the Bogalusa Heart Study. Aging Cell. 2008;7:451–8.10. Christensen B, Houseman E, Marsit C, Zheng S, Wrensch M, Wiemels J,Nelson H, Karagas M, Padbury J, Bueno R et al. Aging and environmentalexposures alter tissue-specific DNA methylation dependent upon CpGisland context. PLoS Genet. 2009;5:e1000602.11. Bollati V, Schwartz J, Wright R, Litonjua A, Tarantini L, Suh H, et al. Decline ingenomic DNA methylation through aging in a cohort of elderly subjects.Mech Ageing Dev. 2009;130:234–9.12. Rakyan VK, Down TA, Maslau S, Andrew T, Yang TP, Beyan H, et al. Humanaging-associated DNA hypermethylation occurs preferentially at bivalentmethylome throughout the human lifespan. PLoS One. 2013;8:e67378.18. Day K, Waite L, Thalacker-Mercer A, West A, Bamman M, Brooks J, Myers R,Absher D. Differential DNA methylation with age displays both commonand dynamic features across human tissues that are influenced by CpGlandscape. Genome Biol. 2013;14:R102.19. Hannum G, Guinney J, Zhao L, Zhang L, Hughes G, Sadda S. Genome-widemethylation profiles reveal quantitative views of human aging rates. MolCell. 2013;49:359–67.20. Horvath S. DNA methylation age of human tissues and cell types. GenomeBiol. 2013;14:R115.21. Lin Q, Weidner CI, Costa IG, Marioni RE, Ferreira MRP, Deary IJ. DNA methylationlevels at individual age-associated CpG sites can be indicative for lifeexpectancy. Aging. 2016;8:394–401.22. Spiers H, Hannon E, Schalkwyk LC, Smith R, Wong CC, O’Donovan MC, BrayNJ, Mill J. Methylomic trajectories across human fetal brain development.Genome Res. 2015;25:338–52.23. Marioni R, Shah S, McRae A, Chen B, Colicino E, Harris S, Gibson J, HendersA, Redmond P, Cox S et al. DNA methylation age of blood predicts all-causemortality in later life. Genome Biol. 2015;16:25.24. Christiansen L, Lenart A, Tan Q, Vaupel JW, Aviv A, McGue M, Christensen K.DNA methylation age is associated with mortality in a longitudinal Danishtwin study. Aging Cell. 2016;15:149–54.25. Perna L, Zhang Y, Mons U, Holleczek B, Saum K-U, Brenner H. Epigeneticage acceleration predicts cancer, cardiovascular, and all-cause mortality in aGerman case cohort. Clin Epigenetics. 2016;8:1–7.26. Horvath S, Pirazzini C, Bacalini MG, Gentilini D, Di Blasio AM, Delledonne M,Mari D, Arosio B, Monti D, Passarino G, et al. Decreased epigenetic age ofPBMCs from Italian semi-supercentenarians and their offspring. Aging(Albany NY). 2015;7:1159–70.27. Breitling LP, Saum KU, Perna L, Schöttker B, Holleczek B, Brenner H. Frailty isassociated with the epigenetic clock but not with telomere length in aGerman cohort. Clin Epigenetics. 2016;8:21.28. Marioni RE, Shah S, McRae AF, Ritchie SJ, Muniz-Terrera G, Harris SE. Theepigenetic clock is correlated with physical and cognitive fitness in theLothian Birth Cohort 1936. Int J Epidemiol. 1936;2015:44.29. Horvath S, Erhart W, Brosch M, Ammerpohl O, von Schönfels W, AhrensM, Heits N, Bell JT, Tsai P-C, Spector TD et al. Obesity acceleratesepigenetic aging of human liver. Proc Natl Acad Sci U S A. 2014;111:15538–43.30. Horvath S, Garagnani P, Bacalini M, Pirazzini C, Salvioli S, Gentilini D,DiBlasio A, Giuliani C, Tung S, Vinters H, Franceschi C. Acceleratedepigenetic aging in Down syndrome. Aging Cell. 2015;14:491–5.31. Horvath S, Levine AJ. HIV-1 infection accelerates age according to theepigenetic clock. J Infect Dis. 2015;212:1563–73.32. Horvath S, Ritz BR. Increased epigenetic age and granulocyte counts in theblood of Parkinson’s disease patients. Aging (Albany NY). 2015;7:1130–42.33. Levine M, Lu A, Bennett D, Horvath S. Epigenetic age of the pre-frontal cortex is associated with neuritic plaques, amyloid load, andAlzheimer’s disease related cognitive functioning. Aging (Albany NY).2015;7:1198–211.34. Levine ME, Hosgood HD, Chen B, Absher D, Assimes T, Horvath S. DNAmethylation age of blood predicts future onset of lung cancer in thewomen’s health initiative. Aging (Albany NY). 2015;7:690–700.35. Zannas A, Arloth J, Carrillo-Roa T, Iurato S, Roh S, Ressler K, Nemeroff C,Smith A, Bradley B, Heim C, et al. Lifetime stress accelerates epigeneticaging in an urban, African American cohort: relevance of glucocorticoidsignaling. Genome Biol. 2015;16:266.36. Fagnoni FF, Vescovini R, Mazzola M, Bologna G, Nigro E, Lavagetto G,Franceschi C, Passeri M, Sansoni P. Expansion of cytotoxic CD8+ CD28- Tcells in healthy ageing people, including centenarians. Immunology.1996;88:501–7.37. Fagnoni FF, Vescovini R, Passeri G, Bologna G, Pedrazzoni M, Lavagetto G,Casti A, Franceschi C, Passeri M, Sansoni P. Shortage of circulating naiveCD8+ T cells provides new insights on immunodeficiency in aging. Blood.2000;95:2860–8.38. Gruver AL, Hudson LL, Sempowski GD. Immunosenescence of ageing. J Pathol.2007;211:144–56.39. Gurven M, Kaplan H, Winking J, Eid Rodriguez D, Vasunilashorn S, Kim JK,Finch C, Crimmins E. Inflammation and infection do not promote arterialaging and cardiovascular disease risk factors among lean horticulturalists.PLoS One. 2009;4:e6590.40. Gurven M, Kaplan H, Winking J, Finch C, Crimmins EM. Aging andinflammation in two epidemiological worlds. J Gerontol A Biol Sci MedSci. 2008;63:196–9.Horvath et al. Genome Biology  (2016) 17:171 Page 21 of 2241. Gurven M, Blackwell AD, Rodriguez DE, Stieglitz J, Kaplan H. Does bloodpressure inevitably rise with age?: longitudinal evidence among forager-horticulturalists. Hypertension. 2012;60:25–33.42. Fagny M, Patin E, MacIsaac JL, Rotival M, Flutre T, Jones MJ, Siddle KJ,Quach H, Harmant C, McEwen LM, et al. The epigenomic landscape ofAfrican rainforest hunter-gatherers and farmers. Nat Commun. 2015;6:10047.43. Horvath S, Mah V, Lu AT, Woo JS, Choi OW, Jasinska AJ, Riancho JA, Tung S,Coles NS, Braun J et al. The cerebellum ages slowly according to theepigenetic clock. Aging (Albany NY). 2015;7:294–306.44. Adkins RM, Krushkal J, Tylavsky FA, Thomas F. Racial differences in gene-specific DNA methylation levels are present at birth. Birth Defects Res AClin Mol Teratol. 2011;91:728–36.45. Markides KS, Coreil J. The health of Hispanics in the southwestern UnitedStates: an epidemiologic paradox. Public Health Rep. 1986;101:253–65.46. Sorlie PD, Backlund E, Johnson NJ, Rogot E. Mortality by Hispanic status inthe United States. JAMA. 1993;270:2464–8.47. Arias E, Eschbach K, Schauman WS, Backlund EL, Sorlie PD. The Hispanicmortality advantage and ethnic misclassification on US death certificates.Am J Public Health. 2010;100 Suppl 1:S171–7.48. Ruiz JM, Steffen P, Smith TB. Hispanic mortality paradox: a systematicreview and meta-analysis of the longitudinal literature. Am J PublicHealth. 2013;103:e52–60.49. Crimmins EM, Kim JK, Alley DE, Karlamangla A, Seeman T. Hispanic paradoxin biological risk profiles. Am J Public Health. 2007;97:1305–10.50. Tracy RP, Doyle MF, Olson NC, Huber SA, Jenny NS, Sallam R, Psaty BM,Kronmal RA. T-helper type 1 bias in healthy people is associated withcytomegalovirus serology and atherosclerosis: the Multi-Ethnic Study ofAtherosclerosis. J Am Heart Assoc. 2013;2:e000117.51. Contreras G, Alaez C, Murguia A, Garcia D, Flores H, Gorodezky C. Distributionof the killer cell immunoglobulin-like receptors in Mexican Mestizos. TissueAntigens. 2007;69 Suppl 1:125–9.52. Dowd JB, Aiello AE, Alley DE. Socioeconomic disparities in the seroprevalenceof cytomegalovirus infection in the US population: NHANES III. EpidemiolInfect. 2009;137:58–65.53. Dowd JB, Zajacova A, Aiello A. Early origins of health disparities: Burden ofinfection, health, and socioeconomic status in U.S. children. Soc Sci Med.2009;68:699–707.54. Johnson NE. The racial crossover in comorbidity, disability, and mortality.Demography. 2000;37:267–83.55. Corti MC, Guralnik JM, Ferrucci L, Izmirlian G, Leveille SG, Pahor M, CohenHJ, Pieper C, Havlik RJ. Evidence for a black-white crossover in all-cause andcoronary heart disease mortality in an older population: the North CarolinaEPESE. Am J Public Health. 1999;89:308–14.56. Yao L, Robert SA. Examining the racial crossover in mortality betweenAfrican American and white older adults: a multilevel survival analysis ofrace, individual socioeconomic status, and neighborhood socioeconomiccontext. J Aging Res. 2011;2011:8.57. Fenelon A. An examination of black/white differences in the rate of age-related mortality increase. Demogr Res. 2013;29:441–72.58. Gurven M, Jaeggi AV, Kaplan H, Cummings D. Physical activity andmodernization among Bolivian Amerindians. PLoS One. 2013;8:e55679.59. Blackwell AD, Trumble BC, Maldonado Suarez I, Stieglitz J, Beheim B,Snodgrass JJ, et al. Immune function in Amazonian horticulturalists. AnnHum Biol. 2016;43:382–96.60. Gurven M, Kaplan H, Supa AZ. Mortality experience of Tsimane Amerindians ofBolivia: regional variation and temporal trends. Am J Hum Biol. 2007;19:376–98.61. Case A, Paxson C. Sex differences in morbidity and mortality. Demography.2005;42:189–214.62. Oksuzyan A, Juel K, Vaupel JW, Christensen K. Men: good health and highmortality. Sex differences in health and aging. Aging Clin Exp Res. 2008;20:91–102.63. Crimmins EM, Kim JK, Langa KM, Weir DR. Assessment of cognition usingsurveys and neuropsychological assessment: the Health and RetirementStudy and the Aging, Demographics, and Memory Study. J Gerontol BPsychol Sci Soc Sci. 2011;66 Suppl 1:i162–71.64. Simpkin AJ, Hemani G, Suderman M, Gaunt TR, Lyttleton O, McArdle WL,et al. Prenatal and early life influences on epigenetic age in children: astudy of mother-offspring pairs from two cohort studies. Hum Mol Genet.2016;25:191–201.65. Klemera P, Doubal S. A new approach to the concept and computation ofbiological age. Mech Ageing Dev. 2006;127:240–8.66. [No authors listed]. Design of the Women’s Health Initiative clinical trial andobservational study. The Women’s Health Initiative Study Group. ControlClin Trials. 1998;19:61–109.67. Curb JD, McTiernan A, Heckbert SR, Kooperberg C, Stanford J, Nevitt M,Johnson KC, Proulx-Burns L, Pastore L, Criqui M, Daugherty S. Outcomesascertainment and adjudication methods in the Women’s Health Initiative.Ann Epidemiol. 2003;13:S122–8.68. Freedman DS, Newman 3rd WP, Tracy RE, Voors AE, Srinivasan SR,Webber LS, Restrepo C, Strong JP, Berenson GS. Black-white differences inaortic fatty streaks in adolescence and early adulthood: the Bogalusa HeartStudy. Circulation. 1988;77:856–64.69. Aryee MJ, Jaffe AE, Corrada-Bravo H, Ladd-Acosta C, Feinberg AP, HansenKD, Irizarry RA. Minfi: a flexible and comprehensive Bioconductor packagefor the analysis of Infinium DNA methylation microarrays. Bioinformatics.2014;30:1363–9.70. Pidsley R, Y Wong CC, Volta M, Lunnon K, Mill J, Schalkwyk LC. A data-drivenapproach to preprocessing Illumina 450K methylation array data. BMC Genomics.2013;14:293.71. Morris TJ, Butcher LM, Feber A, Teschendorff AE, Chakravarthy AR, WojdaczTK, Beck S. ChAMP: 450 k Chip Analysis Methylation Pipeline. Bioinformatics.2014;30:428–30.72. Costello S, Cockburn M, Bronstein J, Zhang X, Ritz B. Parkinson’s disease andresidential exposure to maneb and paraquat from agricultural applicationsin the Central Valley of California. Am J Epidemiol. 2009;169:919–26.73. Li Y, Chen JA, Sears RL, Gao F, Klein ED, Karydas A, Geschwind MD,Rosen HJ, Boxer AL, Guo W, et al. An epigenetic signature in peripheralblood associated with the haplotype on 17q21.31, a risk factor forneurodegenerative tauopathy. PLoS Genet. 2014;10:e1004211.74. Liu J, Morgan M, Hutchison K, Calhoun VD. A study of the influence of sexon genome wide methylation. PLoS One. 2010;5:e10028.75. Lunnon K, Smith R, Hannon E, De Jager PL, Srivastava G, Volta M,Troakes C, Al-Sarraj S, Burrage J, Macdonald R, et al. Methylomicprofiling implicates cortical deregulation of ANK1 in Alzheimer’s disease.Nat Neurosci. 2014;17:1164–70.76. Gibbs JR, van der Brug MP, Hernandez DG, Traynor BJ, Nalls MA, Lai S-L,Arepalli S, Dillman A, Rafferty IP, Troncoso J, et al. Abundant quantitativetrait loci exist for DNA methylation and gene expression in human brain.PLoS Genet. 2010;6:e1000952.77. Zhang D, Cheng L, Badner JA, Chen C, Chen Q, Luo W, Craig DW,Redman M, Gershon ES, Liu C. Genetic control of individual differences ingene-specific methylation in human brain. Am J Hum Genet. 2010;86:411–9.78. Pidsley R, Viana J, Hannon E, Spiers H, Troakes C, Al-Saraj S, Mechawar N,Turecki G, Schalkwyk L, Bray N, Mill J. Methylomic profiling of human braintissue supports a neurodevelopmental origin for schizophrenia. GenomeBiol. 2014;15:483.79. Hernandez D, Nalls M, Gibbs J, Arepalli S, van der Brug M, Chong S, et al.Distinct DNA methylation changes highly correlated with chronological agein the human brain. Hum Mol Genet. 2011;20:1164–72.80. Bennett DA, Schneider JA, Arvanitakis Z, Wilson RS. Overview andfindings from the religious orders study. Curr Alzheimer Res. 2012;9:628–45.81. De Jager PL, Srivastava G, Lunnon K, Burgess J, Schalkwyk LC, Yu L,Eaton ML, Keenan BT, Ernst J, McCabe C, et al. Alzheimer’s disease: earlyalterations in brain DNA methylation at ANK1, BIN1, RHBDF2 and other loci.Nat Neurosci. 2014;17:1156–63.82. Dunning M, Barbosa-Morais N, Lynch A, Tavare S, Ritchie M. Statistical issuesin the analysis of Illumina data. BMC Bioinformatics. 2008;9:85.83. Teschendorff AE, Marabita F, Lechner M, Bartlett T, Tegner J, Gomez-CabreroD, Beck S. A beta-mixture quantile normalization method for correctingprobe design bias in Illumina Infinium 450 k DNA methylation data.Bioinformatics. 2013;29:189–96.84. Houseman E, Accomando W, Koestler D, Christensen B, Marsit C, Nelson H,Wiencke J, Kelsey K. DNA methylation arrays as surrogate measures of cellmixture distribution. BMC Bioinformatics. 2012;13:86.85. Appay V, van Lier RA, Sallusto F, Roederer M. Phenotype and function ofhuman T lymphocyte subsets: consensus and issues. Cytometry A. 2008;73:975–83.86. Perez-Andres M, Paiva B, Nieto WG, Caraux A, Schmitz A, Almeida J, Vogt RF,Jr., Marti GE, Rawstron AC, Van Zelm MC, et al. Human peripheral bloodB-cell compartments: a crossroad in B-cell traffic. Cytometry B Clin Cytom.2010;78 Suppl 1:S47–60.87. Rickabaugh TM, Baxter RM, Sehl M, Sinsheimer JS, Hultin PM, Hultin LE,Quach A, Martínez-Maza O, Horvath S, Vilain E, Jamieson BD. Acceleration ofage-associated methylation patterns in HIV-1-infected adults. PLoS One.2015;10(3):e0119201. doi:10.1371/journal.pone.0119201. eCollection 2015.88. Heyn H, Moran S, Hernando-Herraez I, Sayols S, Gomez A, Sandoval J, MonkD, Hata K, Marques Bonet T, Wang L, Esteller M. DNA methylationcontributes to natural human variation. Genome Res. 2013;23(9):1363–72.89. Viechtbauer W. Conducting meta-analyses in R with the metafor Package.J Stat Software. 2010;36:1–48.•  We accept pre-submission inquiries •  Our selector tool helps you to find the most relevant journal•  We provide round the clock customer support •  Convenient online submission•  Thorough peer review•  Inclusion in PubMed and all major indexing services Submit your next manuscript to BioMed Central and we will help you at every step:Horvath et al. Genome Biology  (2016) 17:171 Page 22 of 22•  Maximum visibility for your researchSubmit your manuscript atwww.biomedcentral.com/submit

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.52383.1-0361800/manifest

Comment

Related Items