Methylomic predictors demonstrate the role of NF-kappa B in old-age mortality and are unrelated to the aging-associated epigenetic drift

Onco a ge 19228 www.impac jou nals.com/onco a ge www.impac jou nals.com/onco a ge / Onco a ge , Vol. 7, No. 15 Me hylomic p edic o s demons a e he ole o NF-κB in old-age mo ali y and a e un ela ed o he aging-associa ed epigene ic d i Juulia Jylhä ä1,2, Lau a Kananen1,2, Jani Rai anen3,4, Saa a Ma ila1,2, Tapio Ne alainen1,2, An i He onen2,3, Ma ja Jylhä2,3 and Mikko Hu me1,2,5 1 Depa men o Mic obiology and Immunology, School o Medicine, Uni e si y o Tampe e, Tampe e, Finland 2 Ge on ology Resea ch Cen e , Uni e si y o Tampe e, Tampe e, Finland 3 School o Heal h Sciences, Uni e si y o Tampe e, Tampe e, Finland 4 UKK Ins i u e o Heal h P omo ion Resea ch, Tampe e, Finland 5 Fimlab Labo a o ies, Tampe e, Finland Co espondence o: Juulia Jylhä ä, email: [email p o ec ed] Keywo ds: me hyla ion, mo ali y, aging, longe i y, Cox model, Ge o a ge Recei ed: No embe 06, 2015 Accep ed: Ma ch 10, 2016 Published: Ma ch 22, 2016 ABSTRACT Changes in he DNA me hyla ion (DNAm) landscape ha e been implica ed in aging and cellula senescence. To un a el he ole o speci ic DNAm pa e ns in la e-li e su i al, we pe o med genome-wide me hyla ion p o iling in nonagena ians (n=111) and de e mined he pe o mance o he me hylomic p edic o s and con en ional isk ma ke s in a longi udinal se ing. The su i al model con aining only he me hylomic ma ke s was supe io in e ms o p edic i e accu acy compa ed wi h he model con aining only he con en ional p edic o s o he model con aining con en ional p edic o s combined wi h he me hylomic ma ke s. A he 2.55-yea ollow-up, we iden i ied 19 mo ali y-associa ed ( alse-disco e y a e <0.5) CpG si es ha mapped o genes unc ionally clus e ing a ound he nuclea ac o kappa B (NF-κB) complex. In e es ingly, none o he mo ali y-associa ed CpG si es o e lapped wi h he es ablished aging-associa ed DNAm si es. Ou esul s a e in line wi h p e ious indings on he ole o NF-κB in con olling animal li e spans and demons a e he ole o his complex in human longe i y. INTRODUCTION The in luen ial ole o genomic ac o s, such as DNA me hyla ion (DNAm) in he cou se o de elopmen , aging and age- ela ed pa hologies is well es ablished. Se e al s udies ha e also ep oducibly demons a ed ha he le el o me hyla ion a speci ic CpG si es changes as a unc ion o age [1-5], hence p o iding a ma ke o ch onological and, po en ially, biological age. An in iguing cha ac e is ic o age- ela ed DNAm signa u es is ha many o he age-associa ed DNAm changes ha e been obse ed o be common in se e al di e en issues, such as whole blood, b ain, lung and ce ix [1, 3, 6]. These obse a ions sugges ha a global mechanism(s) migh be esponsible o age-associa ed modi ica ions in he epigene ic landscape. Ne e heless, s udies wi h monozygo ic wins ha e demons a ed ha he a e o di e gence in me hylomic pa e ns inc eases wi h age [7, 8], sugges ing ha he age- ela ed modi ica ions in DNAm a e also subjec o a ious en i onmen al, s ochas ic and li e s yle- ela ed e ec s. Howe e , he consequences o he aging- accompanied DNAm al e a ions o la e-li e heal h and unc ional abili ies a e la gely unknown. A ecen epigenome-wide associa ion s udy (EWAS) demons a ed ha he associa ion be ween age- ela ed DNAm changes and heal hy aging pheno ypes in indi iduals 32-80 yea s o age is negligible [8]. The esul s o his s udy also e eal ha he DNAm egions associa ed wi h aging pheno ypes a e dis inc om hose associa ed wi h ch onological age. These indings sugges ha he CpG si es in ol ed in heal h- ela ed ou comes in la e li e a e la gely egula ed by si es o he han he es ablished age- ela ed DNAm egions [8]. In addi ion, using an EWAS Onco a ge 19229 www.impac jou nals.com/onco a ge app oach, we ha e ecen ly demons a ed ha he CpG si es ha a e associa ed wi h aging- ela ed in lamma ion, i.e., in lammaging [9] a e la gely di e en om he si es associa ed wi h age [5]. This phenomenon is also obse able in ega d o gene exp ession p o iles and old age mo ali y. We ha e p e iously demons a ed ha he genes exhibi ing aging- ela ed changes in exp ession le els a e p edominan ly di e en om hose ha p edic mo ali y in la e li e [10]. These indings unde sco e he complexi y and unknown na u e o he genomic ac o s ha con ol he human heal h span and la e-li e e en s. Ne e heless, he mo ali y-p edic ing genes in ou p e ious s udy we e ound o be unc ionally connec ed o he nuclea ac o kappa B (NF-κB) complex, which is a cen al media o in immunoin lamma o y esponses and has been ad oca ed as he culp i in aging and cellula senescence ( e iewed in [11]). Abe an ac i a ion o NF- κB has been epo ed in a ious age-associa ed condi ions, such as neu odegene a ion, immunosenescence, in lammaging, sa copenia and os eopo osis ( e iewed in [12-14]), whe eas s udies in ol ing mouse models ha e obse ed ha NF-κB ac i a ion is a key de e minan o accele a ed aging and longe i y [15, 16]. In he mouse models, i was demons a ed ha he hypo halamic ac i a ion o NF-κB is a d i ing o ce o sys emic aging h ough immune-endoc ine connec ions [16]. Li e span egula ion in humans is a mul i ac o ial p ocess, and e y li le is known abou he genomic de e minan s ha con ol la e-li e mo ali y a e he ages o he common kille s, i.e., ca dio ascula e en s and cance , ha e passed. In his s udy, we sough o explo e how he human genome-wide me hylome is associa ed wi h old-age su i al wi hin a sho e (2.55 yea s) and a longe (4 yea s) ollow-up ime. A la ge panel o adi ional (bio)ma ke s and mo ali y isk ac o s was assessed alongside he me hylomic ma ke s o elucida e he ela ionship be ween he aging- ela ed biophysiological changes and epigene ics. RESULTS The cha ac e is ics o he s udy popula ion and dis ibu ion o he a iables in he popula ion wi h me hyla ion da a a ailable (n = 111) a e p esen ed in Table 1. The a iables (i.e., he con en ional ma ke s) exhibi ing signi ican (p < 0.05) uni a ia e and mul i a ia e associa ions a he 2.55 ollow-up a e p esen ed in Supplemen a y Table 1. The p edic o s emaining in he mul i a ia e model, body mass index (BMI) and Mini- Men al S a e Examina ion (MMSE) es sco e, we e used as he model ac o s in he assessmen o he p edic i e accu acy o modeling (see Me hods). The measu e o “epigene ic clock” [17], he DNA me hyla ion age was no p edic i e o mo ali y in ou coho (p = 0.733). In he Cox uni a ia e assessmen , 19,621 and 15,505 CpG si es we e associa ed wi h mo ali y (p < 0.05) in he 2.55-yea and 4-yea ollow-up da a, espec i ely (Supplemen a y Tables 2 and 3). A e B-H co ec ion (FDR < 0.5), 19 CpG si es emained signi ican o he 2.55-yea ollow-up and 7 CpG si es o he 4-yea ollow- up da a (Supplemen a y Tables 2 and 3). The Ingenui y Pa hway Analysis (IPA) -gene a ed ne wo k om he 16 known genes ha bo ing he 19 signi ican CpG si es a he 2.55-yea ollow-up is p esen ed in Figu e 1a. This ne wo k displayed NF-κB as a cen al node and in ol ed 10 o 16 o he genes mapped o he 19 mo ali y- associa ed CpG si es (FDR < 0.5). We also an he IPA ne wo k and pa hway analyses om he genes ha bo ing he 250 op- anking CpG si es acco ding o he 2.55- yea ollow-up da a (si es p esen ed in Supplemen a y Table 2). The highes - anking ne wo k in his analysis also placed NF-κB as a cen al complex (Figu e 1b). The signi ican B-H-co ec ed canonical pa hways om his da a se a e p esen ed in Table 2. A he 4-yea ollow-up, he unc ional implica ions o he me hylomic p edic o s we e a enua ed as no signi ican B-H -co ec ed canonical pa hways we e iden i ied in IPA om he genes ha bo ing he 250 highes - anking CpG si es and no signi ican ne wo k en ichmen was obse ed among he genes ha bo ing he 7 CpG si es (FDR < 0.5). Assessmen o he p edic i e accu acy o he es ed models e ealed ha he Ridge eg ession con aining only he me hylomic ma ke s (Ridge1) pe o med be e han he o he models; i.e., a model con aining only he con en ional p edic o s, a Ridge eg ession model con aining bo h he con en ional p edic o s (Ridge2) and he me hylomic ma ke s and a model con aining only he me hylomic ma ke s selec ed on he basis o hei signi icance le el in Cox uni a ia e assessmen . Speci ically, he me hylomic ma ke s alone exhibi ed he smalles median de iance om he null model (Supplemen a y Figu e 1), and we e hus used in assessing he inal mo ali y-p edic ing signa u e in he Cox mul i a ia e model o 2.55-yea ollow-up da a. The de iances o he con en ional ma ke s exhibi ed clea ly he smalles a ia ion bu hei median was ne e heless highe han ha o he me hylomic ma ke s in Ridge1. The Ridge eg ession-o ganized 19 me hylomic ma ke s en e ed o he Cox mul i a ia e model a e p esen ed in Supplemen a y Table 4. Inclusion o he me hylomic ma ke s in he inal model was based on selec ion o he model wi h he bes goodness o i (Akaike In o ma ion c i e ion, AIC), which o he selec ed model was 239.0. The inal Cox mul i a ia e model is p esen ed in Table 3 and he dis ibu ions o he be a alues o he se en CpG si es (ba ch e ec -co ec ed) included his mo ali y-p edic ing signa u e a e p esen ed in Supplemen a y Figu e 2. The disc imina i e powe (Ha ell’s C) o his model was 89.9%. The p opo ionali y assump ion in he Onco a ge 19230 www.impac jou nals.com/onco a ge Table 1: Cha ac e is ics o he s udy popula ion (n = 111). Dis ibu ions o he a iables a e p esen ed acco ding o he da a a he 2.55-yea s mo ali y ollow-up. Non-su i o s Su i o s Va iable Mean/Med SEM/IQR/% Mean/Med SEM/IQR/% Women (n/%) 27 75.0 54 72.0 Age (mon hs) 1079.5 0.61 1080.2 0.37 Sys olic blood p essu e (mmHg) 145 4.6 149 3.4 Dias olic blood p essu e (mmHg)* 71.5 13.5 74.0 19.0 Weigh (kg) 61.9 2.2 70.6 1.6 BMI (kg/m2) 24.3 0.75 27.5 0.54 Wais ci cum e ence (cm) 89.6 2.1 95.5 1.4 Hip ci cum e ence (cm)* 98 10.0 102 12.0 MMSE* 23.5 8.0 26.0 4.0 Ba hel index* 95.0 20.0 95 5.0 Handg ip (kg)* 18.0 11.0 20.0 7.0 Able o pe o m chai - ise es (n = yes/%) 19 57.6 59 78.7 Able o pe o m chai -s and es (n = yes/%) 22 71.0 62 82.7 F ail y index (n/%) Non- ail 3 8.3 26 34.7 P e- ail 22 61.1 37 49.3 F ail 11 30.6 12 16.0 CRP le el (ng/ml)* 1.8 3.3 1.9 3.5 IL-1β le el (pg/ml)* 14.2 27.6 19.0 34.0 IL-6 le el (pg/ml)* 4.5 3.3 3.8 3.8 IL-7 le el (pg/ml)* 7.8 5.3 6.4 5.2 IL-10 le el (pg/ml)* 1.8 1.5 1.5 2.6 c -DNA le el (μg/ml)* 0.93 0.19 0.87 0.16 Unme hyla ed c -DNA le el (μg/ml)* 0.75 0.20 0.67 0.15 Plasma m DNA (copy numbe )* 4.30E82.37E83.75E82.09E8 Alu epea c -DNA (GE)* 74.4 50.4 66.8 38.3 DHEAS (μg/ml)* 0.25 0.48 0.25 0.31 Co isol (ng/ml)* 133 54.3 117 68.0 IDO ac i i y (Kyn/T p)* 44.3 25.5 51.8 25.3 An i-CMV an ibody i e * 19.000 8.000 19.000 9000 An i-EBV an ibody i e * 405 315 410 410 DNAm age 76.1 1.04 76.1 0.64 CD3+ cells (%)*a62.0 15.8 57.9 12.0 CD4+ cells (%)b62.9 2.5 63.8 1.6 CD8+ cells (%)b30.6 2.3 28.9 1.5 CD4+/CD8+ cells ( a io)* 2.4 2.3 2.3 2.4 CD4+CD28- cells (%)* 9.2 16.2 9.2 13.0 CD8+CD28- cells (%) 63.3 2.8 63.3 2.1 CD14+ cells (%)*a8.3 5.9 8.1 6.3 *median alues and IQR p esen ed ape cen age o li e-ga ed cells; bpe cen age o o al T lymphocy es (CD3+ cells); cpe cen age o CD4+ cells; dpe cen age o CD8+ cells Abb e ia ions: BMI, body mass index; CD, clus e o di e en ia ion; CMV, cy omegalo i us; CRP, C- eac i e p o ein; c -DNA, cell- ee DNA; DHEAS, dehyd oepiand os e one sul a e; DNAm, DNA me hyla ion; EBV, Eps ein-Ba i us; GE, genomic equi alen ; IDO, indoleamine 2,3-dioxygenase; IL, in e leukin; Kyn, kynu enine; MMSE, Mini-Men al S a e Examina ion; m DNA, mi ochond ial DNA; T p, yp ophan Onco a ge 19231 www.impac jou nals.com/onco a ge Figu e 1: The highes - anking ne wo ks om he 16 known genes ha bo ing he op 19 signi ican (FDR < 0.5) CpG si es a. and om he genes ha bo ing he op 250 CpG si es b. in he 2.55-yea ollow-up (n = 111). Bo h ne wo ks displayed NF-κB as a cen al node and we e en iched o he common e m Hema ological Sys em De elopmen and Func ion. The g een colo o he molecule indica es ha hypome hyla ion o a CpG si e in he gene was associa ed wi h inc eased mo ali y, and he ed colo indica es ha hype me hyla ion o a CpG si e in he gene was associa ed wi h inc eased mo ali y. The ne wo ks we e gene a ed h ough he use o QIAGEN’s Ingenui y Pa hway Analysis (IPA®,QIAGEN Redwood Ci y, www.qiagen.com/ingenui y). Onco a ge 19232 www.impac jou nals.com/onco a ge Table 2: Canonical pa hways cons uc ed om he genes ha bo ing he op 250 CpG si es associa ed wi h mo ali y a he 2.55-yea ollow-up. Ingenui y Canonical Pa hways -log(p)* Ra io Molecules Ch onic Myeloid Leukemia Signaling 1.91 7.61E-02 TGFBR2, GAB2, HDAC4, SMAD3, PIK3R2, NFKB1, ATM Ge m Cell-Se oli Cell Junc ion Signaling 1.58 5.13E-02 TGFBR2, MAP3K14, MAP3K10, ACTA2, KEAP1, ITGA2, PIK3R2, ATM Role o NFAT in Ca diac Hype ophy 1.58 4.55E-02 IL6ST, TGFBR2, HDAC4, ITPR3, IGF1R, SLC8A3, PIK3R2, ATM Cell Cycle: G1/S Checkpoin Regula ion 1.58 7,94E-02 CCND2, HDAC4, CCND3, SMAD3, ATM Regula ion o he Epi helial-Mesenchymal T ansi ion Pa hway 1.58 4.40E-02 MAML1, TGFBR2, FZD3, SMAD3, PIK3R2, NFKB1, SMURF1, ATM iCOS-iCOSL Signaling in T Helpe Cells 1.58 5.83E-02 GAB2, CD28, ITPR3, PIK3R2, NFKB1, ATM Rac Signaling 1.58 5,83E-02 CYFIP2, ITGA2, PIK3R2, NFKB1, ATM, ANK1 NF-κB Ac i a ion by Vi uses 1.58 6.85E-02 MAP3K14, ITGA2, PIK3R2, NFKB1, ATM Hepa ic Fib osis/Hepa ic S ella e Cell Ac i a ion 1.58 4.08E-02 TGFBR2, TNFSF4, ACTA2, MYH14, SMAD3, IGF1R, NFKB1, FAS GADD45 Signaling 1.58 1.58E-01 CCND2, CCND3, ATM PKCθ Signaling in T Lymphocy es 1.57 5.31E-02 MAP3K14, MAP3K10, CD28, PIK3R2, NFKB1, ATM Molecula Mechanisms o Cance 1.57 3.06E-02 TGFBR2, GAB2, CCND2, CCND3, FZD3, SMAD3, ITGA2, PIK3R2, NFKB1, FAS, ATM CNTF Signaling 1.46 8.16E-02 IL6ST, CNTF, PIK3R2, ATM B Cell Recep o Signaling 1.44 4.09E-02 GAB2, MAP3K14, MAP3K10, PAG1, PIK3R2, NFKB1, ATM RANK Signaling in Os eoclas s 1.44 5.81E-02 MAP3K14, MAP3K10, PIK3R2, NFKB1, ATM Vi us En y ia Endocy ic Pa hways 1.44 5.62E-02 ITSN1, ACTA2, ITGA2, PIK3R2, ATM C oss alk be ween Dend i ic Cells and Na u al Kille Cells 1.44 5.62E-02 CD28, ACTA2, KLRD1, NFKB1, FAS Lympho oxin β Recep o Signaling 1.44 7.41E-02 MAP3K14, PIK3R2, NFKB1, ATM Dea h Recep o Signaling 1.44 5.49E-02 MAP3K14, ACTA2, PARP12, NFKB1, FAS Colo ec al Cance Me as asis Signaling 1.43 3.46E-02 IL6ST, TGFBR2, FZD3, SMAD3, PIK3R2, NFKB1, PTGER4, ATM Myc Media ed Apop osis Signaling 1.40 6.90E-02 IGF1R, PIK3R2, FAS, ATM T Cell Recep o Signaling 1.40 5,21E-02 CD28, PAG1, PIK3R2, NFKB1, ATM Es ogen-Dependen B eas Cance Signaling 1.30 6.45E-02 IGF1R, PIK3R2, NFKB1, ATM CD40 Signaling 1.30 6.25E-02 MAP3K14, PIK3R2, NFKB1, ATM Onco a ge 19233 www.impac jou nals.com/onco a ge Cox Reg ession model was es ed using he global es by calcula ing he scaled Schoen eld esiduals o each o he independen p edic o s in he inal Cox model. S a is ically signi ican dependence o mo ali y on ime was no obse ed (p = 0.280) indica ing ha he p opo ionali y assump ion was no iola ed. Due o he small numbe o mo ali y-associa ed CpG si es in he me hylomic da a a he 4-yea ollow- up, no compa ison o he p edic ion accu acies o he di e en modeling op ions o assessmen o he inal mo ali y-p edic ing signa u e was pe o med o he 4-yea mo ali y da a. Co ela ion analysis be ween he me hyla ion le els in he mo ali y-associa ed CpG si es and he co esponding gene p oduc (s) e ealed a signi ican co ela ion be ween h ee CpG si e/ ansc ip pai s. In e se co ela ions we e obse ed be ween he cg03348466 (CRTC3) and CRTC3 mRNA le el and be ween cg04182483 (RGS10) and he RGS10 mRNA le el. A di ec co ela ion was obse ed be ween cg22794214 (HIVEP3) and HIVEP3 mRNA le el. All he co ela ions a e p esen ed in Supplemen a y Table 5. Analysis o he genomic loca ions o he op 19 CpG si es (FDR < 0.5, in Supplemen a y Table 2) o ansc ip ion ac o (TF) binding si es and o he genomic egula o y elemen s e ealed ha a majo i y o he si es we e loca ed on ac i e cis- egula o y egions; hey ei he ha bo ed TF binding si es, DNAse I hype sensi i i y egions, and/o we e iden i ied as “P edic ed p omo e egion including ansc ip ion s a si e (TSS)”, “P edic ed enhance (E)” o “P edic ed weak enhance o open ch oma in (WE)”. In addi ion, six CpG si es demons a ed unc ional signi icance as hey we e anno a ed o “P edic ed ansc ibed egion (T)”. The mos abundan TFs we e POLR2A and RELA which bo h had binding si es on ou CpG si e loci. Full da a o his assessmen a e p esen ed in Table 4. DISCUSSION We ha e p e iously demons a ed, using genome- wide gene exp ession da a, ha he NF-κB complex is cen ally in ol ed in con olling human old-age mo ali y [10]. In he p esen s udy, we expanded he examina ion o he genomic ac o s egula ing la e-li e su i al by analyzing he p edic i e abili y o genome-wide me hylomic da a a he 2.55-yea ollow-up. The esul s o his s udy co obo a e he ole o NF-κB in all-cause elde ly mo ali y; he molecula ne wo k cons uc ed om he genes ha bo ing he mo ali y-associa ed CpG si es displayed he NF-κB complex as a cen al media o (Figu e 1). The genes nuclea ac o o kappa ligh polypep ide gene enhance in B-cells 1 (NFKB1) and a axia elangiec asia mu a ed (ATM) we e also iden i ied in he ne wo k. In iguingly, bo h NFKB1 and ATM ha e p e iously been linked wi h accele a ed aging and cellula senescence in s udies wi h gene ically enginee ed mice [15, 18, 19]. These s udies ad oca ed ha NFKB1 and ATM- egula ed abe an NF-κB ac i a ion and he ensuing ch onic sys emic in lamma o y s a e a e he ul ima e HGF Signaling 1.30 4.81E-02 MAP3K14, MAP3K10, ITGA2, PIK3R2, ATM Panc ea ic Adenoca cinoma Signaling 1.30 4.72E-02 TGFBR2, SMAD3, PIK3R2, NFKB1, ATM NGF Signaling 1.30 4.72E-02 MAP3K14, MAP3K10, PIK3R2, NFKB1, ATM T Helpe Cell Di e en ia ion 1.30 5.97E-02 IL6ST, TGFBR2, CD28, IL21R IL-9 Signaling 1.30 8.82E-02 PIK3R2, NFKB1, ATM *Benjamini-Hochbe g-co ec ed p- alue Table 3: The inal mo ali y-p edic ing signa u e a he 2.55-yea ollow-up assessed om he Ridge eg ession -o ganized me hylomic ma ke s. HR (95% CI) S.E. Z p cg08421934 (NA)0.41 (0.26-0.64) 0.10 -3.84 <0.001 cg15770702 (MAP3K14)0.40 (0.27-0.61) 0.08 -4.38 <0.001 cg08596308 (ATP6V1G2; NFKBIL1)0.50 (0.34-0.73) 0.10 -3.60 <0.001 cg23282964 (RIOK1)0.56 (0.37-0.84) 0.12 -2.82 0.005 cg16720947 (PLEC1)0.52 (0.34-0.80) 0.13 -2.94 0.003 cg27027151 (IL21R)2.09 (1.44-3.02) 0.39 3.90 <0.001 cg26843567 (NA)0.68 (0.46-0.99) 0.13 -2.01 0.045 Abb e ia ions: CI, con idence in e al; HR, haza d a io; NA, no a ailable; S.E., s anda d e o Onco a ge 19234 www.impac jou nals.com/onco a ge d i e s o senescence and aging-associa ed de e io a ion [15, 18]. Al hough ou da a do no p o ide a mechanis ic link be ween he hypome hyla ion o hese CpG si es and he isk o mo ali y, we specula e ha he mechanism in ol es an in lamma o y componen by which he genomic ac o s con ol la e-li e mo ali y. Analysis o he 19 mo ali y-associa ed CpG si es (FDR < 0.5) o genomic egula o y elemen s e ealed ha a majo i y o he si es we e loca ed on ac i e cis- egula o y egions (Table 4). Tha is, hey ha bo ed TF binding si es, loca ed on DNAse I hype sensi i i y a eas and/o displayed one o he ollowing p edic ed genomic s a es: Table 4: Assessmen o he 19 mo ali y associa ed CpG si es (FDR<0.5) in he 2.55-yea ollow-up (n = 111) o ansc ip ion ac o binding si es and o he unc ional genomic elemen s using ENCODE da a in he UCSC genome b owse . CpG si e (gene) GRCh37/hg19 coo dina e T ansc ip ion Fac o s Genome s a us DNAse I Hype sensi i i y Clus e cg24859528 (IQSEC1) ch 3:12941421 TNO cg03348466 (CRTC3) ch 15:91104770 CEBPB TYES cg02395768 (ATP5SL) ch 19:41945578 SIN3AK20, POLR2A, SP2, SP1, CHD2, NFYB, PBX3, MAZ, NFIC, GTF2F1, MTA3, TAF1, TBL1XR1, JUND, KDM5B, STAT5A, HDAC1, SAP30, FOS, YY1, PHF8, FOXM1, TBP, CEBPB, REST, TCF12, IRF1, TEAD4, ZBTB7A, GABPA, MEF2A, PML, RELA TSS YES cg15770702 (MAP3K14)ch 17:43384845 PML, STAT5A, NFATC1, CEBPB, BCL3, TCF12, EBF1, FOXM1, EP300, RELA, STAT3, NFIC, TBL1XR1, JUND, MEF2A, PAX5, BHLHE40, MEF2C, ATF2, SP1, BATF, RUNX3, IRF4, BCL11A TSS/T YES cg16720947 (PLEC1)ch 8:145048137 n.a. YES cg22794214 (HIVEP3) ch 1:42123463 CTCF WE/R YES cg08596308 (ATP6V1G2; NFKBIL1) ch 6:31516045 CHD1, RBBP5, ZNF274, POLR2A, E2F6, E2F4, KDM5B, MYC, MAX, MAZ TSS YES cg23282964 (RIOK1) ch 6:7417780 TNO cg21200667 (NA) ch 2:30628085 R YES cg08421934 (NA) ch 6:33942413 R NO cg08486432 (ITPR3) ch 6:33598003 T/R YES cg08352439 (VOPP1) ch 7:55637123 POLR2A, POU2F2 TSS YES cg25356639 (FOXP1) ch 3:71349304 R NO cg04395703 (METAP1) ch 4:99982762 TYES cg03171419 (GPR124)ch 8:37700802 POLR2A TYES cg26843567 (NA) ch 12:104846281 R YES cg00291478 (RGS10) ch 10:121301041 RELA, RUNX3, RBBP5 TSS YES cg27027151 (IL21R) ch 16:27461638 POLR2A, MTA3, NFATC1, RELA, BCLAF1, EBF1 E/R YES cg04182483 (RGS10) ch 10:121259610 TNO Abb e ia ions: TSS, P edic ed p omo e egion including ansc ip ion s a si e (TSS); T, P edic ed ansc ibed egion; WE, P edic ed weak enhance o open ch oma in cis- egula o y elemen ; E, P edic ed enhance ; R, P edic ed Rep essed o Low Ac i i y egion Onco a ge 19235 www.impac jou nals.com/onco a ge p omo e egion including ansc ip ion s a si e, enhance o weak enhance /open ch oma in. I is possible ha he associa ion be ween hese si es and longe i y is media ed h ough al e ed binding o TFs o me hyl-binding domain p o eins, o which he la e ec ui ch oma in-modi ying p o eins o achie e a ep essi e ch oma in s a e. Howe e , ou da a do no allow us o de e mine whe he dis up ed egula ion o ch oma in pe missi eness unde lies he inc eased mo ali y isk. In e es ingly, RELA, which is a subuni o he NF- κB complex, was iden i ied o ha e a binding si e on ou o he analyzed 19 CpG si es. This obse a ion u he suppo s he hypo hesis o he unc ional ole o NF-κB in old-age mo ali y. Region o a p edic ed ansc ip ion s a si e was obse ed o cg02395768 (ATP5SL), cg15770702 (MAP3K14), cg08596308 (ATP6V1G2; NFKBIL1), cg08352439 (VOPP1) and cg00291478 (RGS10). Howe e , he me hyla ion le els in hese si es we e no co ela ed wi h gene exp ession (Supplemen a y Table 5). Ins ead, me hyla ion le els o cg03348466 (CRTC3), cg22794214 (HIVEP3) and cg04182483 (RGS10) co ela ed wi h he co esponding ansc ip exp ession le el. The obse a ion ha he co ela ions we e o e all modes is, howe e , in line wi h p e ious indings on minimal co ela ions be ween age-associa ed changes me hyla ion and ansc ip ion [5, 20, 21]. Six si es, including cg03348466 (CRTC3) and cg04182483 (RGS10) esided in p edic ed ansc ibed a ea, and can hence also be conside ed unc ionally signi ican . The po en ial egula o y ole o hese si es (in he gene body egion) may in ol e e.g. al e na i e splicing. Howe e , he exac mechanism connec ing he mo ali y-associa ed changes in me hyla ion o al e na i e splicing equi es u he esea ch. The canonical pa hways cons uc ed om he genes ha bo ing he op 250 mo ali y-associa ed CpG si es a he 2.55-yea ollow-up co e ed a wide a ie y o cellula signaling unc ions among which se e al in lamma ion and immuni y- ela ed p ocesses we e ep esen ed. In e es ingly, pa hways e med NF-κB Ac i a ion by Vi uses, GADD45 Signaling and Cell Cycle: G1/S Checkpoin Regula ion we e also iden i ied. The eme gence o hese pa hways sugges s ha NF-κB migh also be in ol ed la e-li e con ol o cellula g ow h and su i al, DNA epai and apop osis, as hese unc ions a e asc ibed o he induc ion o he NF-κB- GADD45 cascade [22]. In e es ingly, in ou p e ious pape on he ansc ip omic mo ali y p edic o s, we obse ed ha an inc eased exp ession o GADD45B was p edic i e o an inc eased isk o mo ali y in hese nonagena ians [10]. Howe e , as he numbe o mo ali y-associa ed CpG si es was ma kedly educed om he 2.55-yea s ollow-up o he 4-yea s ollow-up, we specula e ha he me hylomic ma ke s migh exhibi a dynamic na u e e en in he ex eme ages. Tha is, a subs an ial pa o he genomic CpG si es migh be cons an ly emodeled, and du ing 4 yea s, hei me hyla ion le els a e likely o change o an ex en ha hei p edic i e abili y in ou popula ion is educed. The longe ollow-up ime also allows mo e ime o s ochas ic mo ali y de e minan s, such as auma, o ope a e, which may hus weaken he ole o he genomic p edic o s. Al hough he me hylomic ma ke s did no exhibi e y s ong s a is ical signi icances a e FDR-co ec ion and we used a libe al h eshold o including hem in he Ridge eg ession (FDR < 0.5), he me hylomic da a demons a ed good pe o mance in e ms o gene alizabili y and disc imina i e powe . Speci ically, he me hylomic da a alone exhibi ed be e p edic i e accu acy han he con en ional ma ke s alone o in combina ion wi h he me hylomic ma ke s, and he se en CpG si es in he inal Cox model had a disc imina i e powe o 89.9%. In his espec , he me hylomic da a also pe o med be e han he ansc ip omic mo ali y p edic o s in ou p e ious s udy [10]. Ne e heless, we acknowledge ha he majo weaknesses o ou s udy a e a lack o a sepa a e e i ica ion coho and a a he small s udy popula ion. Hence, he esul s mus be conside ed as en a i e and hypo hesis-gene a ing. The s eng h o ou s udy, howe e , is he ac ha all he s udy pa icipan s we e 90 yea s o age a baseline. The e o e ou esul s a e no con ounded by he e ec o ch onological age on DNAm. A ecen s udy by Moo e e al. analyzed genome- wide me hylomic mo ali y p edic o s in indi iduals wi h a wide age ange (30-100 yea s a 9-yea ollow- up, mean mo ali y ollow-up ime 4.4 yea s) [23]. They iden i ied 76 CpG si es whe e he a e o change in DNAm was associa ed wi h mo ali y and 88 ma ke s whe e he yea 9 le el o DNAm was associa ed wi h mo ali y. In e es ingly, hei mo ali y-associa ed DNAm si es also included genes wi h immunoin lamma o y unc ions and a link o NF-κB egula ion. Howe e , no o e lap be ween indi idual mo ali y-associa ed CpG si es we e ound in ou da a se s. These di e ences may a ise due o di e en popula ion cha ac e is ics, such as age ange and he causes o dea h. Hence, u he s udies a e equi ed o es ablish he po en ially age and popula ion-speci ic ela ionships be ween DNAm and mo ali y. When we examined he se en inal signa u e mo ali y-p edic ing CpG si es and hei co esponding genes (Table 3) o o e lap wi h he genes ha bo ing he mos commonly aging-associa ed CpG si es - ELOVL2, FHL2 [2, 21, 24-26], KLF14 [2, 21, 25, 26], SST [2, 25, 26], OTUD7A [2, 24, 26], PENK [2, 21, 24, 25] and EDARADD [2, 6, 24, 26] - we ound no o e lap be ween hese ma ke s. Mo eo e , none o ou op 250 mo ali y- associa ed me hylomic si es (Supplemen a y Table 2) we e among he 525 common age-associa ed CpG si es ha ha e been obse ed in mo e han one s udy (summa ized in [21]). Moo e e al. [23] ha e also in obse ed a simila phenomenon in hei popula ion: e y ew ( < 0,05%) Onco a ge 19236 www.impac jou nals.com/onco a ge o he aging-associa ed CpG si s we e also mo ali y- associa ed. These obse a ions sugges ha aging- associa ed epigene ic d i and he epigene ic con ol o he li e span in old age migh ope a e h ough di e en genomic mechanisms. This hypo hesis is also in line wi h ou p e ious indings on age-associa ed ansc ip s [27], which displayed e y li le simila i y wi h mo ali y- p edic ing ansc ip s [10]. Despi e he inc easing body o da a ha sugges s ha se e al mani es a ions o o ganismal aging and de elopmen a e o epigene ic o igin, he associa ions epo ed hus a on DNAm and aging-pheno ypes a e sca ce and/o he indings ha e been nega i e. Bell e al. (2012) examined he genome-wide associa ions be ween DNAm and 16 age- ela ed pheno ypes and ound ha wo pheno ypes - lung unc ion and low- densi y lipop o ein le els - exhibi ed an associa ion wi h one CpG si e (cg16463460 in WT1 and cg03001305 in STAT5A, espec i ely) and ma e nal longe i y exhibi ed an associa ion wi h wo CpG si es (cg13870866 in TBX20 and cg09259772 in ARL4A) [8]. In ano he EWAS, Ma ioni e al. (2015) de ec ed no indi idual CpG si es associa ed wi h physical o cogni i e i ness in an elde ly popula ion [28]. Howe e , hey did ind a c oss-sec ional associa ion be ween a measu e o DNAm age - he epigene ic clock based on he Ho a h p edic o [17] -, and physical and cogni i e i ness ye he DNAm age was no p edic i e o a longi udinal chance in he i ness measu es [28]. The DNAm age has also been ecen ly demons a ed o p edic all-cause mo ali y in ou di e en coho s o elde ly indi iduals [29] and in Danish wins [30]. Howe e , he DNAm age was no p edic i e o mo ali y in ou s udy. One eason o he nega i e inding migh be ha indi iduals in ou coho we e all e y old a baseline (90 yea s), and dea h a his age likely has di e en unde pinnings han a younge old ages and when assessed in coho s wi h wide age spec a. In conclusion, he esul s o his s udy suppo he genomic-le el ole o NF-κB a he e y end o he human li e span. We hypo hesize ha ou indings could ela e o he ecen obse a ion o a p og amma ic ole o hypo halamic NF-κB and IκB kinase-β ac i a ion in he con ol o he li e span in expe imen al mouse models [16]. Adhe ing o he conclusion o his mouse s udy ha he decisi e ole o hypo halamic NF-κB is exe ed sys emically le el h ough immune-neu oendoc ine c oss alk [16], we sugges ha ou indings on immune cells migh ep esen he pe iphe al co espondence o hypo halamic NF-κB ac i a ion. Howe e , es ablishing he sys emic-le el e en s ha connec NF-κB unc ion o all cause-mo ali y in aged humans will equi e u he esea ch. MATERIALS AND METHODS S udy popula ion The s udy popula ion consis ed o nonagena ian subjec s pa icipa ing o he Vi ali y 90+ s udy, which is an ongoing, p ospec i e popula ion-based s udy on indi iduals aged 90 yea s and abo e and who eside in he ci y o Tampe e, Finland. The Vi ali y 90+ s udy was ini ia ed in 1995, and since hen se e al nonagena ian coho s ha e been ec ui ed and examined o biological, clinical, demog aphic and social measu es. Mo ali y a es ha e been analyzed longi udinally using comple e ollow-ups. The ec ui men p o ocol and cha ac e iza ion o he subjec s in he cu en s udy has been p e iously desc ibed [10]. The da a in his s udy conce n indi iduals bo n in 1920 and ec ui ed in 2010 o sample collec ion. Genome-wide me hyla ion da a and he ull co a ia e da a including cell ype p opo ions we e a ailable o 111 subjec s (n = 81 women and n = 30 men). The all-cause mo ali y da a we e collec ed om he Popula ion Regis e Cen e . As we wan ed o assess bo h sho e and longe - e m su i al p edic o s o his coho , he mo ali y da a was collec ed in wo di e en ime poin s. The i s da a collec ion was pe o med on 31s o Janua y in 2013 co esponding o a 2.55-yea median ollow-up and he second one was on 31 s o May in 2014 co esponding o a 4-yea ollow-up. The mo ali y a e a he 2.55-yea ollow-up was 32.4% (36/111) and 47.7% (53/111) a he 4-yea ollow-up. All he pa icipan s ga e hei w i en in o med consen . The s udy was conduc ed ollowing he guidelines o he Decla a ion o Helsinki, and he s udy p o ocol was app o ed by he e hics commi ee o he ci y o Tampe e. Sample collec ion and p ocessing Venous blood samples we e collec ed in EDTA- con aining ubes by a ained home- isi ing medical s uden be ween 8 am and 12 am. Plasma was sepa a ed and s o ed a -70°C. Genomic DNA and o al RNA we e ex ac ed om PBMCs in which he blood samples we e subjec ed o leucocy e sepa a ion using a Ficoll-Paque densi y g adien (Ficoll-Paque™ P emium, ca . no. 17- 5442-03, GE Heal hca e Bio-Sciences AB, Uppsala, Sweden). The PBMC laye was collec ed, and he cells alloca ed o RNA ex ac ion we e suspended in 150 µl o RNAla e solu ion (Ambion Inc., Aus in, TX, USA). Cells ha we e alloca ed o FACS analysis and DNA ex ac ion we e suspended in 1 ml o a eezing solu ion (5/8 FBS, 2/8 RPMI-160 medium, 1/8 DMSO; FBS ca . no. F7524, Sigma-Ald ich, MO, USA; RPMI: ca . no. R0883, Sigma- Ald ich, MO, USA; DMSO: ca . no. 1.02931.0500, VWR, Espoo, Finland).