What Is Epigenetics? Cell Specialization Is Reversible

Preformationism (17-19th century): each cell contains preformed elements that What is Epigenetics? enlarge during development.

Epigenesis (19th century -): chemical Humunculus reactions among soluble components in Hartsoecker 1695 the cell that execute a complex developmental plan.

Chromosomes are Necessary for Development Cell Specialization is Reversible Before the 20th century Late 20th and early 21st centuries

Skin cell nuclear transfers 99

Central Question: How can a single Original explant removed fertilized egg give rise to a complex Adult frog of \-nu strain Outgrowth of organism with cells of varied as nuclear donor epidermal cells

phenotypes? Parent of 1st transfer Donor cells for recipient eggs Enucleation of nuclear transfer recipient eggs 1st nuclear transfer Cells trypsinized Foot web outgrowth and washed prove frog was 2-nu

Walter Sutton, 1902 Theodor Boveri, 1903 Uncleaved Completely cleaved (70 V) Martially cleaved /c o/\ Columbia University University of Würzburg • Gurdon, Laskey & Reeves 1975 (25%) USA Germany demonstrated that “cell Dissociated cells for serial transfer • Determined that all chromosomes had specialization does not involve I *^i/ KJpZ* Parent of serial ttransfei r any loss, irreversible activation or Enucleation of 1 recipient eggs to be present for proper embryonic recipient eggs Serial nuclear transfer permanent change chromosomal Foot web outgrowth development. prove frog was 2-nu genes required for development” Uncleaved Completely cleaved (40/O Partially cleaved (30/0 (30%)

• Development encoded by irreversible Nuclear transplant tadpole: l-nu diploid from nucleolus and chromosome counts (present in 36% of serial clones) changes in chromosomes? Fig. 2. Plan of serial nuclear transfer experiments, using nuclei from adult skin celJs. Reasons for the various steps are explained in the text (p. 99). The percentages of injected eggs which cleave in different ways are those typically obtained with adult skin cell nuclei. The actual results of the experiments reported here are shown in Tables 1 and 3.

The overall plan of our experiments is illustrated in Fig. 2. Serial transfers were carried out in all cases, and the following comments summarize the reasons for this. A serial nuclear transfer experiment is one in which a donor nucleus is taken from an embryo which has itself resulted from a previous, first-transfer, experiment (Fig. 2). Most of the blastulae obtained from first Surprises in the Genomic Era Epigenetic Landscape Late 20th and early 21st centuries Phenotypic changes occur during development

Number of genes • Human Genome Project 2003 Grape 30,434 • Stem cells start with same Human 22,333 developmental potential • Why so few genes? Chicken 16,736

Fruit fly 14,889 • • How do cell lines know and E. coli 4,149 Can follow different developmental paths remember their function? Influenza 11 Pertea & Salzberg 2010

• Where does individual • Most phenotypic traits are variation come from? inﬂuenced by many genes

• Are there traits/diseases that • Landscape is shaped by cannot be explained by our genotype and environment genetic sequence? Waddington 1957

Disruption of Epigenetic Landscape Can lead to cancer REVIEWS

a Stem cells evidence has confirmed that stochastic DNA methylation alterations in cancer involve large regions of the epigenome190,191. This stochastic epigenetic change does ormal not occur genome-wide. Rather, genome-wide views of development epigenetic variation have shown that large hypomethylated blocks, constituting up to one-third of the genome, contain the most variably methylated regions of the Neoplasia Cancer tumour genome190,192. These domains arise early during Modern Deﬁnition cancer development191,193,194 and contain the most variably expressed genes regulating cancer- relevant functions190. • Moreover, the degree of variation in methylation in early Epigenetics: study of stably heritable phenotypes precursor lesions predicts cancer risk193,194, suggesting a (e.g. gene expression) that result form changes in Dierentiated states causal link between these epigenetic changes and cancer. chromatin, without changes in DNA sequence Hypomethylated blocks in cancer largely correspond to b partially methylated domains in normal cells as well as LADs and LOCKs (FIG. 2). These regions underlie much of the reported variation in methylation at CpG islands, Feinberg 2016 Nature Reviews Genetics Allis et al 2015 pigenetic shores and distant CpG sites, fuelling phenotypic vari- mediators 191,192 e.g., OCT4, SOX2 ation in cancer . In addition, the degree of vari ation in methylation191, as well as the deviation of the variability in gene expression from the normal corresponding tissue, is a predictor of cancer progression195. The combination of ageing and chronic sun exposure — the two leading causes of skin cancer — induces the wide- spread formation of hypomethylated blocks in the epi- dermis at genomic regions that are hypomethylated in pigenetic squamous cell carcinoma and that overlap with colon modiers cancer- specific hypomethy lated blocks196. These same (e.g., DNMT, TET) regions are the very ones that show further alterations in methylation in squamous cell cancers arising within pigenetic modulators (e.g., APC, STAT3) the same skin. Given the overlap of these regions with LADs and LOCKs, these data also indicate that the inter- | Figure 3 Waddington landscape of phenotypic plasticity in development and cancer. play between altered 3D genome organization, stochastic a | The Waddington landscape of development is adapted to compare cellular states of Nature Reviews | Genetics epigenetic change and impaired differentiation mediate different entropy during normal differentiation (left side of image) and in cancer (right side 196 of image). The developmental potential of normal somatic stem cells (grey balls) positioned the effect of environmental damage with photo-ageing . on the top of the hill correlates with high entropy, which is mediated by cellular heterogeneity (different shades of grey). During differentiation, cells are guided towards Network entropy and nuclear structure. Recent work has well-defined cell fates (light blue and brown balls) with lower entropy, paralleled by a described cellular heterogeneity as network entropy — decrease in transcriptional noise and the stabilization of cell states (deepening of the applied as a measure of signalling pathway promiscuity valleys or canalization). Cancer stem cell (CSC) states (yellow ball) arise when epigenetic — and established that the level of network entropy instability interferes with normal differentiation and leads to the erosion of barriers against provides an estimate of developmental potential160,197. dedifferentiation — for example, via the erosion of large organized chromatin K9 In other words, the high entropy of a heterogeneous pluri- modifications and the emergence of hypomethylated blocks. In a similar manner to normal potent stem cell population maintains a diverse range of differentiation, CSCs with higher entropy occupy higher altitudes on the hill than cancer pathways associated with more mature pheno types in a cells (orange and red balls), although the difference is smaller than between normal stem cells and differentiated progeny. Increased transcriptional noise (shallow valleys) and poised state for activation. Consistent with the signalling stochastic switches between diverse cell states (arrows between valleys) are regulated by entropy model of cellular differentiation, the variability the interplay between epigenetic modulators, modifiers and mediators, the deregulated in the expression of signalling factors and developmen- epigenome and fluctuating environmental cues (for example, inflammation, repeated tal regulators has been experimentally linked to the dif- exposure to carcinogens, ageing or an overactive WNT pathway). Finally, cellular ferentiation potential of ESCs198. In a similar manner to heterogeneity (yellow, orange and red balls) within the tumour eventually enables selection normal differentiation, CSCs display a higher entropy mechanisms to drive the growth of the fittest clone. b | Illustration of the role of epigenetic than cancer cells, although the difference is smaller than modifiers, modulators and mediators on the Waddington landscape described in part a. between normal stem cells and differentiated progeny160. Epigenetic modulators (pink hexagon) regulate the activity of epigenetic modifiers Furthermore, CSCs consistently have a lower entropy (green triangles) that induce the ectopic expression of epigenetic mediators. Mediators than their normal counterparts, indicating the presence dynamically alter the contour of the landscape via feedback loops that target epigenetic modifiers such as chromatin modifications (blue circles), lamin proteins (yellow circles) of dominating oncogenic pathways. This is in agreement and chromosomal interactions (new loop on right). The expression of epigenetic mediators with models suggesting that cancers represent hybrid thus produces a shift in the epigenetic landscape, enabling the sampling of aberrant states between aberrantly increased as well as decreased developmental outcomes displaying increased phenotypic plasticity in neoplastic or epigenetic flexibility188 (FIG. 3a). pre-neoplastic cells. APC, adenomatous polyposis coli; DNMT, DNA methyltransferase; Importantly, transitions between cellular states SOX2, sex-determining Y-box 2; STAT3, signal transducer and activator of transcription 3; of different entropy seem to be regulated epigeneti- TET, TET methylcytosine dioxygenase. cally. Using quantitative RNA fluorescence in situ

NATURE REVIEWS | GENETICS VOLUME 17 | MAY 2016 | 293

ǟ ƐƎƏƖ !,(++ - 4 +(2'#12 (,(3#"ƥ ++ 1(%'32 1#2#15#"ƥ ǟ ƐƎƏƖ !,(++ - 4 +(2'#12 (,(3#"ƥ ++ 1(%'32 1#2#15#"ƥ ɥ ɥ ɥ ɥ ɥ ɥ ɥ ɥ ɥ ɥ ɥ ɥ ɥ ɥ Epigenetic marks: DNA is not naked DNA Methylation Distinct chromatin conﬁgurations give rise to diverse phenotypes Usually correlated with repression of gene expression

Chromatine remodeling Histone modiﬁcations Histone composition variants • In many animals occurs mostly on cytosines at CpG dinucleotides DNA methyltransferase non-coding RNAs DNA methylation DNA demethylase • Most exons and introns are Cytosine 5-methylcytosine highly DNA methylated (70-80%) in vertebrates SAM = S-adenosylmethionine

Allis et al 2015

Medvedeva et al. BMCMedvedeva Genomics 2013,et al.15 BMC:119 Genomics 2013, 15:119 Page 3 of 12 Page 3 of 12 Modelhttp://www.biomedcentral.com/1471-2164/15/119 of transcriptionhttp://www.biomedcentral.com/1471-2164/15/119 regulation by CpG methylation Binding of transcription initiation complex is inhibited by methylation DNA Methylation Usually associated with silencing of gene expression Transcription initiation complex forms Transcription occurs Unmethylated DNA

Transcription initiation complex cannot form Methylated Transcription is inhibited DNA

Medvedeva et al 2013 BMC Genomics

Pol = RNA Polymerase; TF = Transcription factor

Figure 1 Schematic representation of the interaction between promoter methylation and transcription of the gene. In the absence of DNA methylation, TFs canFigure bind 1 DNASchematic allowing representation RNA polymerase of to the bind interaction and to start between the transcription. promoter Panel methylationa shows the and following transcription scenario: ofthe if gene. In the absence of DNA becomes methylated,DNA TFs methylation, are blocked TFs from can binding bind DNA to DNA allowing and thereforeRNA polymerase RNA polymerase to bind and is unable to start to the bind transcription. and to initiate Panel transcription.a shows the following scenario: if Panel b shows the followingDNA becomes scenario: chromatinmethylated, modifications TFs are blocked reduce from the binding ability ofto TFsDNA to and bind therefore DNA and RNA therefore polymerase RNA polymerase is unable to is bind unable and to to initiate transcription. bind; the repressed conditionPanel b ofshows the chromatin the following is maintained scenario: by chromatin subsequent modifications DNA methylation. reduce thePolII ability is shown of TFs as ato maroon bind DNA pie; and nucleosome therefore is RNA polymerase is unable to shown as a blue cylinder.bind; Plain the (solid) repressed lollipops condition represent of the unmethylated chromatin is(methylated) maintained cytosines. by subsequent TF is shown DNA methylation. as an orange PolII octagon. is shown The as green a maroon pie; nucleosome is hexagon and purple trapezoidshown asare a a blue methyl-binding cylinder. Plain domain (solid) and lollipops Policomb-group represent unmethylated proteins, respectively. (methylated) The brown cytosines. triangle TF is represents shown as an orange octagon. The green an unknown repressor. hexagon and purple trapezoid are a methyl-binding domain and Policomb-group proteins, respectively. The brown triangle represents an unknown repressor.

of their TFBSs [50]. Besides, it was reported that binding Our study shows that a fraction of single CpGs within of some TFs, includingof their Sp1 TFBSs and [50]. CTCF, Besides, is sufficient it was for reportedpromoters that binding exhibits aOur significant study shows negative that correlation a fraction of of single CpGs within maintaining a localof unmethylated some TFs, including state [60-65]. Sp1 Neverthe- and CTCF,their is sufficient methylation for profilespromoters with exhibits the expression a significant profiles negative of correlation of less, this scenario (Figuremaintaining 1b) does a local not unmethylated explain the sensi- state [60-65].neighboring Neverthe- transcriptionaltheir methylation start sites profiles (TSSs) with considered the expression profiles of tivity of certain TFsless, to methylation this scenario of (Figure their TFBSs. 1b) does not explainacross the various sensi- samples.neighboring Moreover, transcriptional we observe start a strong sites (TSSs) considered In this study, wetivity explore of certain the evidence TFs to methylation that supports of theirnegative TFBSs. selection againstacross thevarious presence samples. of such Moreover, cytosines we observe a strong one of these two scenarios.In this Tostudy, achieve we explore this, we the first evidence test within that supportsTFBSs, especiallynegative in their selection core against positions. the Interest- presence of such cytosines whether methylationone of of a these particular two scenarios. cytosine Tocorrelates achieve this,ingly, we we first find test thatwithin repressors TFBSs, are especially more sensitive in their to core the positions. Interest- with transcription.whether This effect methylation may provide of a particular a basis for cytosinepresence correlates of such cytosinesingly, we in find their that binding repressors sites. are more sensitive to the regulation of transcriptionwith transcription. through methylation This effect of may spe- provideThis a basis work for is partpresence of the of FANTOM5 such cytosines project. in their Data binding sites. cific TFBSs. Second,regulation we investigate of transcription whether through some TFs methylationdownloads, of spe- genomicThis tools and work co-published is part of manuscripts the FANTOM5 project. Data are more sensitivecific than TFBSs. others Second, to the presence we investigate of such whetherare collected some TFs at http://fantom.gsc.riken.jp/5/.downloads, genomic tools and co-published manuscripts cytosines in theirare TFBSs more and sensitive what featuresthan others of TFBSs to the presence of such are collected at http://fantom.gsc.riken.jp/5/. can be associatedcytosines with this insensitivity. their TFBSs To this and end, what we features of TFBSs employed ENCODEcan [66] be dataassociated on DNA with methylation this sensitivity. ob- ToResults this end, and we discussion tained by reducedemployed representation ENCODE bisulfite [66] data sequencing on DNA methylationOnly a fraction ob- of cytosinesResults exhibits and discussion significant (RRBS) [67]. RRBStained allows by us toreduced identify representation both methylated bisulfitecorrelation sequencing betweenOnly methylation a fraction and of expressioncytosines exhibits profiles significant and unmethylated(RRBS) cytosines [67]. quantitatively RRBS allows us at to a identify single bothof a methylatedcorrespondingcorrelation TSS between methylation and expression profiles base pair resolutionand in unmethylated the CCGG context cytosines in quantitatively regions It is at well a known single thatof thea corresponding level of cytosine TSS methylation of with high densitiesbase of rarely pair resolution methylated in cytosines, the CCGG usu- contextpromoters in regions is negativelyIt is well correlated known with thatgene the level expression of cytosine methylation of ally co-located withinwith gene high promoters densities of [68]. rarely Tomethylated evaluate [71]; cytosines, the role usu- of methylationpromoters ofis negatively particular correlatedCpGs in the with gene expression genome-wide expressionally co-located across different within gene cell promoters types, we [68].regulation To evaluate of gene[71]; expression the role has of been methylation demonstrated of particular in CpGs in the used FANTOM5genome-wide [69] data obtained expression by cap across analysis different of the cell case types, of ESR1 we [11].regulation The crucial of gene role expression of the location has been of demonstrated in gene expression (CAGE) [70]. FANTOM5 provides methylated regions relative to TSSs is also widely ac- used FANTOM5 [69] data obtained by cap analysis of the case of ESR1 [11]. The crucial role of the location of quantitative estimation of expression in several hun- cepted. The question whether methylation of a particular gene expression (CAGE) [70]. FANTOM5 provides methylated regions relative to TSSs is also widely ac- dreds of different cell types. cytosine may affect expression remains unanswered. quantitative estimation of expression in several hun- cepted. The question whether methylation of a particular dreds of different cell types. cytosine may affect expression remains unanswered. REVIEWS

Establishm ent. After erasure, de novo methylation begins in both germ lines at late fetal stages, and continues after Allele 1 CpG birth41,50 (FIGS 3,4).Oocytes are in meiotic arrest and methylation occurs during their growth47,whereas during spermatogenesis, methylation occurs before meio- sis44,45.Nuclear transplantation experiments indicate that this DNA methylation coincides roughly with the acqui- sition of functional imprints both for autosomal genes and for X chromosome imprinting, at least in oocytes47,51.It is not yet clear which enzymes are respon- Allele 2 CpG sible for de novo methylation in germ cells (BOX 2). Dnmt1 (DNA methyltransferase 1) and its germ-cell- specific isoforms are candidates52,but it is also possible that Dnmt3a or Dnmt3b,which are required for de novo methylation in postimplantation embryos53,carry out REVIEWSthis function in germ cells.It is also unclear how Dnmts specifically target DMRs in either female or male germ Methylation Establishm ent. After erasure, de novo methylation begins cells. DMRs in imprinted genes might be specificallyin both germ tar- lines at late fetal stages, and continues after Allele 1 CpG Acetylation CpG 41,50 geted for de novo methylation in one of the germbirth lines.It41,50 (FIGS 3,4).Oocytes are in meiotic arrest and Transcription complex methylation occurs during their growth47,whereas dur- is equally possible that there is general de novoingmethyla- spermatogenesis, methylation occurs before meio- sis44,45.Nuclear transplantation experiments indicate that Figure 2 | Characteristics of imprinted genes. The figure tion in both germ lines and that DMRs are specificallythis DNA methylation coincides roughly with the acqui- shows a schematic pair of imprinted alleles. Hallmarks of sition of functional imprints both for autosomal genes protected from methylation in one germ line butand fornot X chromosomein imprinting, at least in imprinted genes such as CpG islands and repeats (arrows) the other.In either case,this would require factorsoocytes47,51 that.It is not yet clear which enzymes are respon- are indicated. The enlarged region below the chromosomes Allele 2 CpG sible for de novo methylation in germ cells (BOX 2). recognize DMRs and that are germline-specific.Dnmt1 (DNA The methyltransferase 1) and its germ-cell- highlights the allele-specific epigenetic changes, such as specific isoforms are candidates52,but it is also possible nucleosomal condensation through deacetylation, and existence of such factors is supported by the observationthat Dnmt3a or Dnmt3b,which are required for de novo methylation in postimplantation embryos53,carry out methylation (allele 1) and opening of the chromatin by that deficiency of Dnmt1 causes loss of imprintsthis function post- in germ cells.It is also unclear how Dnmts specifically target DMRs in either female or male germ acetylation and demethylation (allele 2). The transcriptional zygotically, andMethylation the imprints cannot be restoredcells. DMRs inby imprinted genes might be specifically tar- Acetylation Acetylation geted for de novo methylation in one of the germ lines.It competence of allele 2 is indicated by the binding of a Transcription complex 54 Dnmt1, Dnmt3a or Dnmt3b . is equally possible that there is general de novo methyla- transcription complex. DMRs areFigure generally 2 | Characteristics CpG of imprinted rich genes. andThe figure oftention fulfil in both the germ lines and that DMRs are specifically shows a schematic pair of imprinted alleles. Hallmarks of protected from methylation in one germ line but not in imprinted genes such as CpG islands and repeats (arrows) imprinted genes such as CpG islands and repeats (arrows) the other.In either case,this would require factors that criteria for CpGare indicated. islands The enlarged (see region below). below the chromosomes However, autoso- highlights the allele-specific epigenetic changes, such as recognize DMRs and that are germline-specific. The ing new insights about the role of methylation mal CpG islandsnucleosomal do condensation not become through deacetylation, methylated and existence de novo of such .factors is supported by the observation methylation (allele 1) and opening of the chromatin by that deficiency of Dnmt1 causes loss of imprints post- imprints (for example, Cdkn1c requires a maternal So it is likelyacetylation that andimprinted demethylation (allele DMRs 2). The transcriptional are geneticallyzygotically, andor the imprints cannot be restored by competence of allele 2 is indicated by the binding of a Dnmt1, Dnmt3a or Dnmt3b54. methylation imprint to be expressed;see below).In epigeneticallytranscription modified complex. so that de novo methylationDMRs are generally CpG rich and often fulfil the criteria for CpG islands (see below). However, autoso- addition to methylation imprints, differential replica- can occur. Geneticing new insights modification about the role of has methylation been previouslymal CpG islands do not become methylated de novo. tion of DNA is also apparently erased in both germ postulated toimprints be (for due example, to Cdkn1cstretchesrequires a ofmaternal uniqueSo it isdirect likely that imprinted DMRs are genetically or methylation imprint to be expressed;see below).In epigenetically modified so that de novo methylation 24 Three lines;different in the enzymes female germ carry line this out coincides withThreerepeats different that oftenaddition enzymes to flank methylation DMRs imprints, carry differential.More outreplica- recent can work occur. Genetichas modification has been previously tion of DNA is also apparently erased in both germ postulated to be due to stretches of unique direct CpGdemethylation, methylation but in in mammals the male germline it occurs CpGshown methylation that lines;the inrepeats the female in aregermmammals linenot this necessarily coincides with uniquerepeats that often to flank DMRs24.More recent work has 49 demethylation, but in the male germline it occurs shown that the repeats are not necessarily unique to substantiallyde novo and later, maintenance after birth . DMRs butde novo thatsubstantially and clusters maintenance later, after of birth known49. repeat families,DMRs but such that clusters of known repeat families, such

DNMT1 maintenance methylation, central for maintaining DNMT3A, DNMT3B de novo methylation, particularly cellularBox identity2 | DNA methylation and demethylation important during development De novo methylation DNA methylation in mammals occurs in the Box 2 | DNA methylation and demethylation dinucleotide CpG. Methyl groups can Dnmt3b Dnmt3a be introduced into unmethylated DNA by the De novo methylation de novo methylation enzymes Dnmt3a and Dnmt1 DNA methylation in mammals occurs in theDnmt3b (and perhaps others). When DNA is Active demethylationdinucleotide CpG. MethylMaintenance groups methylation can replicated, the methyl group on the template Dnmt3b Dnmt3a Methylates Cytosine adjacent to strand is recognized and a new one is 5-methylcytosinebe inintroduced newly into unmethylated DNA by theintroduced on the opposite (daughter) strand by the enzyme Dnmt1, which can be associated synthesized DNAde novostrandsmethylation of enzymes Dnmt3a andwith the replication machinery. In the presence Dnmt1dividing cells of Dnmt1, hemi-methylated DNA becomes Dnmt3b (andPassive demethylation perhaps others). When DNAfully is methylated and so DNA methylation Active demethylation Reik & Walter 2001Maintenance Nature Review Genetics methylation Reik & Walter 2001 Nature Review Genetics patterns tend to be maintained (maintenance replicated, the methyl group on the templatemethylation). Demethylation can occur in the strand is recognized and a new one is absence of Dnmt1 with continued rounds of DNA replication (passive demethylation), as introduced on the opposite (daughter) strandwell as actively (without DNA replication). The nature of demethylases is unknown. Epigenetic marks: DNA is not naked by the enzyme Dnmt1, which can be associated Distinct chromatin configurations give rise to diverse phenotypes Histonewith the modifications replication machinery. In the presence 24 | JANUARY 2001 | VO LUME 2 www.nature.com/reviews/genetics of Dnmt1, hemi-methylated© 2001 Macmillan DNA Magazines becomes Ltd Passive demethylation fully methylated and so DNA methylation Histone modifications patterns tend to be maintained (maintenance methylation). Demethylation can occur in the absence of Dnmt1 with continued rounds of DNA methylation DNA replication (passive demethylation), as well as actively (without DNA replication). • Chromatin polymerThe is composednature of of demethylases is unknown. nucleosome units

• Each unit consists of an octamer of core histone proteins wrapped by 147bp of DNA Allis et al 2015 Allis et al 2015 24 | JANUARY 2001 | VO LUME 2 www.nature.com/reviews/genetics © 2001 Macmillan Magazines Ltd Histone Modiﬁcations Histone Acetylation >50 different kinds of post-transcriptional modiﬁcations Usually associated with activation of gene expression

• Majority of know covalent modiﬁcations occur on histone tails (some do occur in globular domains)

• Many occur at speciﬁc sites and residues

RESEARCH ARTICLE

E017 E002 a IMR90 E008 track to learn the model) and repressor NRSF; transcribed states show- E001 E015 E014 E016 ES cell E003 E024 E020 E019 ing H3K79me1 and H3K79me2 and associated with the 59 ends of genes E018 iPSC E021 E022 E007 E009 E010 E013 ES-deriv. E012 and introns; and a large number of putative regulatory and neighbour- E011 E004 E005 E006 E062 E034 E045 E033 ing regions showing diverse acetylation marks even in the absence of E044 E043 E039 Blood & E041 E042 E040 Verdin & Ott (2015)T cell NatureE037 Reviews Molecular Cell Biology E048 the H3K4 methylation signatures characteristic of enhancer and pro- E038 E047 E029 E031 Allis et al 2015 E035 HSC & E051 E050 E036 moter regions. E032 B cell E046 E030 E026 E049 E025 Mesench. E023 E052 We used chromatin states to study the relationship between histone E055 E056 E059 E061 E057 Epithelial E058 E028 E027 modification patterns, RNA expression levels, DNA methylation and E054 E053 E112 E093 E071 19,23,44,45 E074 E068 E069 DNA accessibility. Consistent with previous studies , we found E072 E067 Brain E073 E070 E082 E081 E063 E100 low DNA methylation and high accessibility in promoter states, high E108 E107 Muscle E089 E090 E083 E104 E095 Heart E105 DNA methylation and low accessibility in transcribed states, and inter- E065 E078 E076 TissuesE103 can be described in terms Sm. musc. E111 X-chromosome inactivation E092 E085 E084 mediate DNAmethylation and accessibility inenhancer states (Fig. 4d, e E109 E106 E075 E101 Digestive E102 E110 Example of epigenetic mechanisms E077 E079 and Extended Data Fig. 3a, b). These differences in methylation level E094 E099 E086 E088 E097 of their epigenome E087 E080 Other E091 were stronger for higher-expression genes than for lower-expression E066 E098 E096 E113

Chromatin state annotations in 127 epigenomes E114 E115 E116 E117 genes, leading to a more pronounced DNA methylation profile (Extended E118 E119 E120 Vastly moreENCODE E121complex than description of genomic DNA sequence E122 E123 E124 2012 E125 Data Fig. 3c, Supplementary Fig. 5 and Supplementary Table 4f). Genes E126 E127 E128 E129 FAM205B ATP8B5P SIT1 ~3.5-MbNPR2 RECK regionRNF38 onMELK chromosome PAX5 9 GRHPR FRMPD1 SHB ALDH1B1 proximal to H3K27ac-marked enhancers show significantly higher expres- Chrom. states RefSeq genes FAM205B ATP8B5P SIT1 NPR2 RECK RNF38 MELK PAX5 GRHPR FRMPD1 SHB ALDH1B1 RNA-seq sion levels (Extended Data Fig. 3d), and conversely, higher-expression b H3K36me3 H4K20me1 genes were significantly more likely to be neighbouring H3K27ac- https://www.youtube.com/watch?v=BD6h-wDj7bw H3K79me2 H3K79me1 containing enhancers (Extended Data Fig. 3e). H3K9me1 DNase DGF Chromatin states sometimes captured differences in RNA express- Input H3K4me3 ion that are missed by DNA methylation or accessibility. For example, H3K9ac H3K56ac H2A.Z TxFlnk, Enh, TssBiv and BivFlnk states show similar distributions of H2AK9ac H2BK5ac H3K4me2 DNA accessibility but widely differing enrichments for expressed genes H3K18ac H3K4me1 (Fig. 4c, d). Enh and ReprPC states show intermediate DNA methyla- H3K27ac H4K5ac H4K8ac tion, but very different distributions of DNA accessibility and different H3K4ac H3K14ac H3K23ac enrichments for expressed genes (Fig. 4c–e). Lack of DNA methylation,

IMR90 fetal lung f broblasts H2AK5ac H4K91ac typically associated with de-repression, is associated with both the active H2BK120ac H2BK12ac H2BK15ac TssA promoter state and the bivalent TssBiv and BivFlnk states. Bivalent H2BK20ac H3K27me3 H3K9me3 states TssBiv and BivFlnk also show overall lower DNA methylation WGBS Hi-C and higher DNA accessibility than enhancer states Enh and EnhG, and Roadmap Epigenomics Consortium 2015 Nature cd binding by both activating and repressive regulatory factors (Extended H3K4me1 Data Fig. 2b). These results also held for alternative methylation mea- surement platforms (Extended Data Fig. 4a–c), and for the 18-state chro-

33 data sets in IMR90 lung fbroblasts matin state model (Extended Data Fig. 4d, e). Overall, these results H3K4me3 highlight the complex relationship between DNA methylation, DNA

semonegipe ecnere semonegipe accessibility and RNA transcription and the value of interpreting DNA methylation and DNA accessibility in the context of integrated chro- DNase matin states that better distinguish active and repressed regions. Given the intermediate methylation levels of tissue-specific enhan- fer 721 fer cer regions, we directly annotated intermediate methylation regions, RNA-seq 31,46 WGBS based on 25 complementary DNA methylation assays of MeDIP and MRE-seq22,39 from 9 reference epigenomes47. This resulted in more Genome-wide measurements for all marks H3K4me1 than 18,000 intermediate methylation regions, showing 57% CpG meth- WGBS H3K4me3 ylation on average, that are strongly enriched in genes, enhancer chro- DNase Individual mark data sets across epigenomes RNA-seq matin states (EnhBiv, EnhG, Enh) and evolutionarily conserved regions. Chromatin states Intermediate methylation was associated with intermediate levels of Figure 3 | Epigenomic information across tissues and marks. a, Chromatin active histone modifications and DNase I hypersensitivity. Near TSSs, state annotations across 127 reference epigenomes (rows, Fig. 2) in a ,3.5-Mb intermediate methylation correlated with intermediate gene expres- region on chromosome 9. Promoters are primarily constitutive (red vertical sion, and in exons it was associated with an intermediate level of exon lines), while enhancers are highly dynamic (dispersed yellow regions). inclusion47. Intermediate methylation signatures were equally strong b, Signal tracks for IMR90 showing RNA-seq, a total of 28 histone modification within tissue samples, peripheral blood and purified cell types, suggest- marks, whole-genome bisulfite DNA methylation, DNA accessibility, digital ing that intermediate methylation is not simply reflecting differential genomic footprints (DGF), input DNA and chromatin conformation information72. c, Individual epigenomic marks across all epigenomes in which methylation between cell types, but probably reflects a stable state of they are available. d, Relationship of figure panels highlights data set dimensions. cell-to-cell variability within a population of cells of the same type. Epigenomic differences during lineage specification accessibility (Extended Data Fig. 3a), lower methylation (Extended Data We next studied the relationship between DNA methylation dynam- Fig. 3b) and higher transcription factor binding (Extended Data Fig. 2c) ics and histone modifications across 95 epigenomes with methylation than enhancers lacking H3K27ac. In a subset of 7 epigenomes with an data, extending previous studies that focused on individual lineages19,48–50. average of 24 epigenomic marks, we learned separate 50-state chro- We found that the distribution of methylation levels for CpGs in some matin state models based on all the available histone marks and DNA chromatin states varied significantly across tissue and cell type (Fig. 4g, accessibility in each epigenome (Supplementary Fig. 4), which addi- Extended Data Fig. 4f and Supplementary Table 4a). For example, tionally distinguished: a DNase state with distinct transcription factor TssAFlnk states were largely unmethylated in terminally differentiated binding enrichments (Supplementary Fig. 4f), including for mediator/ cells and tissues, but frequently methylated for several pluripotent and cohesin components43 (even though CTCF was not included as an input embryonic-stem-cell-derived cells (Bonferroni-corrected F-test P , 0.01);

CpG methylation,* % SEE COMMENTARY No. Locus Accession Function Chr. loc. CpGs Twin A Twin B Change

Alu-Sx AC006271 — 19p13.2 30 72.1 55.3 16.8 Alu-Sp U14572 — 19p13ϩ3 17 68.1 29.9 38.2 Alu-E2F3 AF547386 — 6p22 2 81.8 100 18.2 PKD1P2 NGϽ002795 Polycystic kidney disease 1 16p13 7 90.9 83.5 7.4 Alu-Sc U14571 — 8p11ϩ1 4 66.6 63.3 3.3 GUK1 BC006249 Guanylate kinase 1 1q42 18 2.8 9.7 6.9

Chr. loc., chromosome location; —, unknown. *Twelve clones per locus.

we performed ANOVA (see Table 4, which is published as sup- single-copy genes. To confirm that these sequences featured DNA porting information on the PNAS web site). ANOVA was used to methylation differences in the twins, we carried out bisulfite compare two parameters: individual epigenetic values, to compare genomic sequencing of multiple clones. The analysis of four dif- epigenetic variability in the entire population, and ESD between ferent Alu sequences and two single-copy genes demonstrated that twins, as a measure of variability between twin pairs. First, we in those MZ twin pairs from which they were isolated, one sibling analyzed the descriptive epigenetic values for each individual and had dense CpG hypermethylation, whereas the other was predom- compared it with the rest of the population, organized into young inantly hypomethylated at that particular sequence (Fig. 2B). Most (Ͻ28 years old) vs. old (Ͼ28 years old) age-groups. Statistical importantly, for both Alus and single-copy genes, differential Epigeneticanalysis of individual descriptive parameters differences provides information methylation arise was associated over with a different life expression time of that Tissues have unique epigenomes about variability within the whole population and in the two age particular sequence in the MZ twin pair, the presence of DNA groups. Second, we used the ESDInteractions to test whether older vs. youngerwith themethylation environment being associated with silencing or reduced expression. MZ twins are significantly more different within their pairs. We Table 1 summarizes the results. found that the epigenetic variability among individuals is high and

(chromatin states) MEDICAL SCIENCES RESEARCH ARTICLE similar (see Table 5, which is published as supporting information on the PNAS web site), regardless of the age group to which individuals belong (Pearson test, P Ͼ 0.05). In contrast, the variance corresponding to the ESD in the older MZ twin group is signifi- Oesophagus Brain RESEARCH ARTICLE cantly higher than that obtained for the younger group (Pearson Heart Angular gyrus Aorta Anterior caudate test, P Ͻ 0.05). These results suggest that older twins are epige-Chr1 Left ventricle Brain Spinal cord Cingulate gyrus Right ventricle Stomach netically more different between pairs than younger twins, and this Right atrium Hippocampus middle Thymus Adrenal difference is not associated with an increased variance in the Thymus E017Heart Inferior temporal lobe E002 a IMR90 E008 track to learn the model) and repressor NRSF; transcribed states show- E001 E015 Aorta Kidney Lung E014 descriptive epigenetic parameters of the older population. Substantia nigra E016 ES cell E003 Right, left, E024Lung E020 Renal cortex, Adipose E019 ing H3K79me1 and H3K79me2 and associated with the 59 ends of genes Dorsolateral Right,E018 left Finally, we also found that those twin pairs who, according to the iPSC E021 Renal pelvis E022 E007 Prefrontal cortex E009 Breast CordE010 blood E013 Small intestine questionnaire, had spent less of their lifetime together andϽor had Myoepithelial ES-deriv. E012 + and introns; and a large number of putative regulatory and neighbour- Blood E011 B cellsE004 (CD19 ) + E005 + vHMEC Stem cells (CD34 ) T cellsE006 (CD3 ) + E062 Large intestine amoredifferentnaturalhealth–medicalhistorywerethosewho E034 Progenitor enriched B cells (CD19 ) E045 + + + E033 ing regions showing diverse acetylation marks even in the absence of E044Liver Luminal epithelial T cells (CD3 , CD4 , CD8 ) E043 Skeletal muscle + Blood & E039 Granuloytes (CD15 ) E041 also showed the greatest differences in levels of 5mC DNA and E042 Back, trunk, arm, leg Duodenum mucosa E040 PBMCs T cell SpleenE037 E048 the H3K4 methylation signatures characteristic of enhancer and pro- + E038 NK cells (CD56 ) E047 acetylation of histones H3 and H4 levels (Pearson test, P Ͻ 0.05). Liver E029 Gonad PlacentaE031 Chr3 E035 Ovaries, testes Stomach mucosa HSC & E051 E050 Spleen E036 moter regions. E032 B cell E046 Sigmoid colon E030 E026 Duodenum smooth muscle E049 E025 Genomic Screening and Loci Identification of DNA Methylation Dif- Mesench. E023 Ovary E052 We used chromatin states to study the relationship between histone E055 Stomach E056 E059 Smooth muscle E061 ferences in MZ Twin Pairs. We next examined where in the MZ twin Colon E057 Epithelial E058 Smooth muscle E028 E027 modification patterns, RNA expression levels, DNA methylation and Kidney E054 Mucosa E053 genomes these epigenetic differences arose by using a global E112 E093 E071 19,23,44,45 Pancreas E074 Osteoblasts E068 E069 DNA accessibility. Consistent with previous studies , we found E072 methylation DNA fingerprinting technique, AIMS (9). The AIMS E067 Small intestine Brain E073 Rectum E070 E082 Smooth muscle E081 Psoas muscle Germinal matrixE063 approach provides a methylation fingerprint constituted by multiple E100 low DNA methylation and high accessibility in promoter states, high Mucosa E108 E107 Muscle E089 Muscle E090 E083 anonymous bands or tags, representing DNA sequences flanked by E104 E095 Heart E105 DNA methylation and low accessibility in transcribed states, and inter-Chr12 E065 E078 E076 two methylated sites, which can be isolated. An illustrative AIMS iPS cells E103 Sm. musc. E111 E092 6.9, 18c, 19.11, 20b, 15b E085 E084 mediate DNAmethylation and accessibility inenhancer states (Fig. 4d, e E109 A E106 result in a MZ twin pair is shown in Fig. 2 .Approximately600 E075 Trophoblast E101 Digestive E102 E110 E077 E079 and Extended DataAIMS Fig. bands 3a, were b). These resolved differences in the gels, and in wemethylation found that between level ES cells E094 Epigenetic Predictor of Age E099 E086 H1, H9, I3, W A7, HUES6, E088 E097 HUES48, HUES64, 4star E087 0.5% and 35% of the bands were different (on the basis of their E080 Other E091 were stronger for higher-expression genes than for lower-expression E066 E098 Neuronal progenitors E096 E113 presence or absence) within MZ twin pairs. Those MZ twins with

Chromatin state annotations in 127 epigenomes E114 E115 E116 Mesodermal progenitors E117 Skin genes, leading to a more pronounced DNA methylation profile (Extended Ganglion eminence E118 ES cell E119 the most differential AIMS bands were those with the greatestChr17 MassArray (Sequenom) for the Edaradd gene and by pyrosequenc- modelE120 was fit on all but one subject and its prediction was related ENCODE E121 Mesenchymal stem cells E122 Marrow derived Skin keratinocyte derived primary E123 E124 derived 2012 E125 Data Fig. 3c, Supplementary Fig. 5 and Supplementary Table 4f). Genes ing for NPTX2 and Tom1L1. For NPTX2, the pyrosequencing toE126 the truly observed age of the left-outSkin fbroblasts subject. The predicted differences in 5mC DNA levels and acetylation levels of histones H3 Ectoderm cultured neurospheresE127 mesenchymal cells E128 E129 methodEndoderm provided methylation data for five additional CpGCortex sites derived in primaryvalues FAM205B areChondrocytes highly ATP8B5P correlatedSIT1 NPR2 RECK withRNF38Skin theMELK melanocytes observed PAX5 ageGRHPR in malesFRMPD1 SHB ALDH1B1 and H4 (Pearson test, P Ͻ 0.05). Most importantly, the twin pairs 2 proximal to H3K27ac-marked enhancers show significantly higher expres- the promoter.Molecular The results of Cell the validation experiments correlatedculturedChrom. neurospheres states(r = 0.83, p = 3.3610 13, n = 47, Figure S2), females (r = 0.75, with the most differential AIMS bands corresponded to MZ twins RefSeq genes FAM205B24ATP8B5P SIT1 NPR2 RECK RNF38 MELK PAX5 GRHPR FRMPD1 SHB ALDH1B1 sion levels (Extended Data Fig. 3d), and conversely, higher-expression 3-year old twins 50-year old twins Figure 1 | Tissues and cellvery types strongly profiled with in the the arrayRoadmap data Epigenomics for all three genesimmune (Edaradd lineages,RNA-seqp ES= 2.4 cells610 and iPS, n cells,= 19, and Figure differentiated S3), and lineages in the derived combined from sample who were older, had spent less of their lifetimes together, or had b H3K36me3 216 Monozygotic twins Fig. 3. Mapping chromosomal regions with differential DNA methylation in r = 0.96,Genome-wide NPTX2 r = 0.92, Tom1L1 Methylationr = 0.90, n = 23), Profilesproviding a H4K20me1 and(r = 0.83, Humanp = 2.2610 Aging, n = 66, Figure 3). For the male only or genes were significantly more likely to be neighbouring H3K27ac- Consortium.RoadmapPrimary Epigenomics tissues and cellConsortium types representative 2015 Nature of all major lineages ES cells. Box colours match groups shown in Fig. 2b. Epigenome identifiers different natural health–medical histories (Pearson test, P Ͻ 0.05). MZ twins by using comparative genomic hybridization for methylated DNA. technical replication of the array data in the twin sample. The H3K79me2female only models, the average absolute differences between the in the human body were profiled, including multiple brain, heart, muscle, (EIDs, Fig. 2c)H3K79me1 for each sample are shown in Extended Data Fig. 1. containing enhancersWe then (Extended selected Data a subset Fig. of 3e). 53 AIMS bands differentially Competitive hybridization onto normal metaphase chromosomes of the AIMS correlation between the degree of methylation and age of all three H3K9me1predicted and the observed age (the error) are 5.3 years and 6.2 Methylation gastrointestinal tract, adipose, skin and reproductive samples, as well as DNase present in MZ twin pairs for further characterization. These bandsFraga etproducts al 2005 generated PNAS from 3- and 50-year-old twin pairs. Examples of the hybrid- genes was preserved in the subset of twins and was also found in the years,DGF and for the combined sample this is 5.2 years. Even when Chromatin states sometimes captured differences in RNA express- Input were cloned, sequenced, and blasted against multiple sequence ization of chromosomes 1, 3, 12, and 17 are displayed. The 50-year-old twin pair expression. In addition,independent we study the male role sample, of regulatory providing regions a biological in human replication.277 DNA In methylationH3K4me3only the data male sets, and and female 166 RNA-seq replication data samples sets, encom- were used, ion that are missed by DNA methylation or accessibility. For example, shows abundant changes in the pattern of DNA methylation observed by the H3K9ac databases. We found that 43% (23 of 53) of the clones matched Alu disease by relating ourfemales, epigenomic Edaradd annotations and Tom1L1 to genetic were variants significantly asso- correlatedpassing with a totalH3K56ac ofdiscarding 150.21 billion all twin mapped data, the sequencing accuracy readsof the correspond- model remained at ARTICLES presence of green and red signals that indicate hypermethylation and hypom- Molecular Cell H2A.Z TxFlnk, Enh, TssBivsequences and (mainly BivFlnk from states the Alu show 6, Alu similar 7, and distributions Alu 8 subfamilies), of age, but NPTX2 was not. The correlation results are shown in H2AK9ac5.3 years, and the predicted values correlated highly with the ethylation events, whereas the 3-year-old twins have a very similar distribution of ciated with common traits and disorders. These analyses demonstrate ing to 3,174-fold coverage of the human genome.213 Figure 2. A multivariate linear regression model using Edaradd, H2BK5acobserved age (r = 0.85, p = 1.701610 , n = 45, Figure S4). DNA accessibility9% but (5 of widely 53) matched differing other enrichments repetitive sequences for expressed (2 LINEs, genes 2 MER, DNA methylation indicated by the presence of the yellow color obtained by equal the importance and wide applicability of our data resource, and lead to Here, weH3K4me2 focus on a subset of 1,936 data sets (Fig. 2) comprising 111 H3K18acTo test whether additional data pointsexample, on the microarrayFigure could 2B highlights two individualsand 1whose MIR), 34% age (18 isof 53) matched ESTs deposited in databases, amounts of the green and red dyes. Significant DNA methylation changes are Edaradd squared and NPTX2 showed that these two markers H3K4me1 (Fig. 4c, d). Enh and ReprPC states show intermediate DNA methyla- important insights into epigenomics, dGenome-wideifferentiation2 and disease. Specific Methylationreference epigenomes Profiles (Fig. 2a–d), andwhich we Human define as having Aging a core set explain 76% (or R = 0.76) of the variance in age of males and 70% H3K27acimprove the accuracy of the model, we performed lasso penalized and 13% (7 of 53) of the clones corresponded to identified indicated as thick red and green blocks in the ideograms. Figure 1 Maternal care alters cytosine highlights of our findings are given below. of five histoneH4K5ac modification marks (Fig. 2e). The fivevastly marks consist over- of: or underpredicted ontion, the but basis very different of their distributionsmethylationa of DNA accessibility and different in females. When considering males and females together the model H4K8acregression to screen for the top predictors of age [13,14]. The top 1 . Histone mark combinations show distinct levels of DNA methyla- H3K4ac 11 14 15 16 17 18 19 110 11 2 methylation of GR promoter. (a) Sequence map histone H3 lysineH3K14acfive 4 predictors trimethylation were (H3K4me3), tested, and associatedonlydata. three with wereTo pro- examine found to whether these differencesMolecular reflect Celltrue biolog- explained 73% of the variance in age. H3K23ac10,24 enrichments for expressed genes (Fig. 4c–e). Lack of DNA methylation,

IMR90 fetal lung f broblasts Fraga et al. PNAS July 26, 2005 vol. 102 no. 30 10607 tion and accessibility, and predict differences in RNA expression moter regionsH2AK5accontribute; H3 lysine significantly 4 monomethylation to the regression (H3K4me1), model: Edaradd, associ- NPTX2, Ͼ Ͼ Ͼ Ͼ of the exon 17 GR promoter including the 17 H4K91ac 10 typically associated with de-repression, is associated with both the active levels that are not reflected in either accessibility or methylation. ated with enhancerH2BK120ac regions ; H3 lysine 36 trimethylationical (H3K36me3), differences in the state of the individual (i.e., versus measure- CpG dinucleotides (bold) and the NGFI-A binding A leave-one-out analysis forms an accurate epigenetic H2BK12acand ELN. The first two predictors were alreadyGenome-wide part of the model. Methylation Profiles and Human Aging . Megabase-scale regions with distinct epigenomic signatures show associatedH2BK15ac with transcribed regions; H3 lysine 27 trimethylation TssA promoter state and the bivalent TssBiv and BivFlnk states. Bivalent 16 H2BK20acUsing the microarray methylation datament for these error two genes, or the intrinsic variability), we usedMaternal the aging model to behavior influences adult behaviorregion (encircled). (b,c) Methylation analysis of predictor of age 25 strong differencesEpigenetic in activity, gene density and nuclear drift lamina asso- (H3K27me3),is anH3K27me3 associatedaverage indicator error with is Polycomb 4.7 years repression (r = 0.77, p =; and1.029 H36of10 lysine207, 9nage= 34). states TssBiv and BivFlnk also show overall lower DNA methylation the 17 CpG dinucleotides of the exon 1 GR To provide an unbiased estimate of predictive accuracy for age, H3K9me3 quantify each26 individual’s apparent methylomic aging rate 7 ciations, suggesting distinct chromosomal domains. trimethylationWGBSAdding (H3K9me3), the ELN associated methylation with heterochromatindata improved the regions accuracy. of our we used a leave-one-out analysis where the multivariate regression Hi-C and higher DNA accessibility than enhancer states Enh and EnhG, and promoter region from adult high- and low-LG- . Approximately 5% of each reference epigenome shows enhancer and Selected epigenomesmodel, also reducing contain the a subset average oferror additional to 3.5 epigenomic years (r = 0.87, (AMAR), defined as the ratio of the predicted age, basedthrough 1681 on epigenetic programming ccc promoter signatures, which are twofold enriched for evolutionarily p = 2.2610211, n = 34, Figure S5). Results were nearly identical example,bindingFigure by both 2B activating highlights and repressive two individuals regulatory factors whose (Extended age is ABN offspring (6–10 clones sequenced/animal; marks,cd including: acetylation marks H3K27ac and H3K9ac, associated 1741 ctctgctagtgtgacacactt1cg2cgcaactc3cgcagttgg4cggg5cg6cggaccacccctg7c H3K4me1when all twin samples were treated as unrelatedmethylation27–29 individuals, data, and to the chronological age. We then tested for n =4 animals/group; *P < 0.01). (b) Percentage conserved non-exonic elements on average. with increased activation of enhancer and promoter regions (Fig.2f); Data Fig. 2b). These results also held 1801 g forgctctgc alternative8cggctggctgtcaccct methylation9cgggggctctggctgc mea- 10cgaccca11cgggg12cgggct when averaged7,18 values for each pair were used. The distribution of vastly over- or underpredicted on the basis of their methylation . Epigenomic data sets can be imputed at high resolutionEdaradd from exist- DNase hypersensitivity , denoting regions of accessibleassociations chromatin between AMAR and possibly relevant clinical 1861 c13cgag14cggtt ccaagcct15cggagtggg16cggggg17cgggagggagcctgggagaa of cytosine residues that were methylated (mean methylation values for ELN was considered too narrow for further surement platforms (Extended Data Fig. 4a–c), and for the 18-state chro- ing data, completing missing marks in additional cell types, and commonly associated with regulator binding (Fig. 2g); DNA methyla- ± s.e.m.) for the first 15 CpG dinucleotides validation using pyrosequencing or MassArrayfactors, analysis. including genderdata. and Toexamine BMI. Analysis whether of ethnicity these and differences reflect true biolog- Non-attentiveLow LG-ABN mothers providing a more robust signal even for observed data sets. tion,typically associated with repressed regulatory33 data sets regions in IMR90 or lung active fbroblasts gene matin state model (Extended Data Fig. 4d, e). Overall, these results H3K4me3 5« 3« (*P < 0.05). (c) Percentage of methylated . Dynamics of epigenomic marks in their relevant chromatin states transcripts4,30 and profiled using whole-genome bisulfitediabetes sequencing status wasical not possible differenceshighlight due to in thecorrelations the complex state relationship ofwith the batch individual between DNA (i.e., methylation, versus measure- DNA AttentiveHigh LG-ABN mothers 19 20 cytosines (mean ± s.e.m.) for the 5′ (site 16) and allow a data-driven approach to learn biologically meaningful rela- (WGBS) ,reduced-representationbisulfitesequencing(RRBS)Discussion , and 1st week of life ARTICLES semonegipe ecnere semonegipe accessibility and RNA transcription and the value of interpreting DNA tionships between cell types, tissues and lineages. mCRF-combined31 methylation-sensitive restrictionvariables enzyme (MRE) (Figure22 S3). We found that gender, but not BMI hadbc* 3′ (site 17) CpG dinucleotides within the NGFI-A In this high density, genomewide screen of CpG methylation of ment error or intrinsic variability), we100 used the aging model to 100 21 methylation and DNA accessibility6 in the contextLow-LG/ABN of integrated chro- binding sequence (*P < 0.0001). (d) The effect . Enhancers with coordinated activity patterns across tissues are enriched and immunoprecipitationDNasetwins, we identified based 88assays CpG (Fig. sites 2h); near andsignificant 80 RNA genes expres- for which contributions the to aging rate (F test, p = 6 3 10Ͻ , * 8 High-LG-ABN * for common gene functions and human phenotypes, suggesting that sion levels , measuredpercent using methylation RNA-seq in saliva and gene is significantly expression correlated microarrays with age. quantifymatin each states individual’s that better distinguishapparent active methylomic and repressed regions. aging rate of cross-fostering the offspring of high- and low- p > 0.05, Experimental Procedures). The methylome of men 80 80 they represent coordinately regulated modules. (Fig. 2i). Our definitionThese are of highly 111 reference enriched epigenomes for genes known is very to similar influence to age- ingGiven (>3 the fold) intermediate of NGFI-A methylation protein levels to ofthe tissue-specific hippocampal enhan- exon 1 GR * LG-ABN mothers on cytosine methylation of the . Regulatory motifs are enriched in tissue-specific enhancers, enhancer that used by therelated International diseases—mainly Human Epigenome cardiovascular Consortium and neurological (IHEC), disease. (AMAR), defined as the ratio of the predicted age, based on7 * fer 721 fer appeared to age approximately 4% faster than that of women NPTX2 cer regions, we directly annotated intermediate60 methylation regions, 60 5′ and 3′ CpG dinucleotides within the NGFI-A modules and DNA accessibility footprints, providing an important which requiredRNA-seqTen RNA-seq, of these WGBS 88 CpG and sites H3K27ac were shown that earlier are only to be available correlated with promoter in the adult offspring of TSA-treated low-LG-ABN moth- +(Figure+ 2D), even thoughmethylation the overall data, distributions to the chronologicalof age were age.* We then tested31,46 for binding sequence of the exon 1 GR promoter resource for gene-regulatory studies. in a subset of epigenomesage in whole here. blood Lastly, and in an isolated additional CD4 16and histone CD14 modi-cells as well WGBS based on 25 complementary DNA methylation* assays of MeDIP 7 ers compared with the vehicle-treated offspring* of low-LG-ABN gene in adult hippocampi (n =5 animals/group). . Genetic variants associated with diverse traits show epigenomic enrich- fication marks[8]. on Weaverage validated were three profiled genes across in a sample 7 deeplynot of unrelated covered significantly cell males and different betweenand theMRE-seq men22,39 andfrom women 9 reference in the epigenomes40 47. This resulted in more 40 females, which confirmed our findings in these replicate samples. associations between AMAR and possibly relevant clinical L-L: animals born to and reared by low-LG-ABN ments in trait-relevant tissues, providing an important resource for types (Fig. 2j). Genome-wide measurements for all marks H3K4me1 mothers (Fig. 3); there were no significant differences between TSA- C-methylation (%) cohort (p > 0.05, KS test). Likewise,than 18,000 the intermediate validation methylation cohort regions,C-methylation (%) showing 57% CpG meth- understanding the molecular basis of human disease. We jointlyWGBS processedRemarkably, and the analysed methylation our111 values reference for the epigenomesvalidated genes are factors, including gender and BMI. Analysis20 of ethnicity and 20 mothers; H-H: animals born to and reared by linear with age over a span of five9,23 decades and in three separate H3K4me3 treated offspring of low-LG-ABN mothers and either TSA- or vehi- with 16 additional epigenomes from ENCODE .Wegeneratedgenome- DNase ylation on average, that are strongly enriched in genes, enhancer chro- Individual mark data sets across epigenomes confirmed the increased aging rate for men (p < 0.05), but was high-LG-ABN mothers; H-L: animals born to sample sets. Based on this observation, we were able to build a 0 0 Reference epigenome mapping across tissues and cell types RNA-seq wide normalized coverage tracks, peaks and broad enriched domains diabetesmatin statuscle-treated states was (EnhBiv, offspringnot possibleEnhG, Enh)of high-LG-ABN anddue evolutionarily to correlations1234 dams. conserved56789101112131415 As with regions. expected, batch TSA 5« 3« high-LG-ABN mothers and reared by low-LG-ABN model that can7,32 predict the age of ainconclusive subject basedChromatin states on for the BMI (p > 0.05). This complements a previous Glucocorticoid Receptor promoter (GR) in offspring promoter % Methylation n=656 The REMCs generated a total of 2,805 genome-wide data sets, includ- for ChIP-seq andmethylation DNase-seq status, normalizedof just two cytosines gene expression in the genome, valuesexplaining for Intermediate methylation was associated with intermediate levels of mothers; L-H: animals born to low-LG-ABN 33 31,34,35 variables (trFigureeatment S3 did). We not found change that histone gender, H3-K9 but acetylation not BMIRegion or had NGFI-A CpG dinucleotide ing 1,821 histone modification data sets, 360 DNA accessibility data sets, RNA-seq , and73% fractional of the variance methylation in age. levels for eachfinding CpG site of an. epigenetic signal for BMI that does not change Figure 3 | Epigenomic information across tissues and marks. a, Chromatin active histone modifications and DNase I hypersensitivity. Near TSSs, 6 mothers and reared by high-LG-ABN mothers. Of the validated genes, Neuronalwith Pentraxin age II (Feinberg (NPTX2) et al., 2010). binding in the adult offspring of high-LG-ABN mothers, because 318 | NATURE | VOL 518 | 19 FEBRUARY 2015 Tom1L1 state annotations across 127 reference epigenomes (rows, Fig. 2) in asignificant3.5-Mb contributions to aging rate (F test, p = 6 3 10Ͻ , Low-LG/ABN (e) Percentage of cytosine methylation (mean ± methylation has been shown to be upregulated in pancreatic , intermediate methylation correlatedde with intermediate gene expres- ©2015 Macmillan Publishers Limited. All rights reserved As genetic associations have been previously reported with High-LG/ABN s.e.m.) of the 5 and 3 CpG dinucleotides within region oncancer chromosome [15], and its 9. expression Promoters is increased are primarily in Parkinson’s constitutive disease (redp vertical > 0.05,sion,theExperimental andGR in exon exons 1 it7 waspromoter Procedures associatedhttp://www.nature.com/natureneuroscience region with). an Thein intermediatethe methylome offspring level of of high-LG-ABN exonof men ′ ′ 5« CpG dinucleotide 3« CpG dinucleotide lines), while[16]. enhancersIts methylation are status highly was dynamic recentlyhuman shown (dispersed to be longevity correlated yellow regions). and aging phenotypesinclusionmothers47. (Atzmon Intermediateis normally et al., methylationassociated 2006; signatureswith acetylated were equally histones strong and highly 5« CpG dinucleotide 3« CpG dinucleotide the NGFI-A binding region of the exon 17 GR b, Signalwith tracks age for in blood IMR90 aswell showing [8]. Mutations RNA-seq, in the a total Edar of associated 28 histone modificationappeared to age approximately 4% faster than that of women 100 promoter gene in the offspring of high- or low- Suh et al., 2008; Willcox et al., 2008within; tissueWheeler samples, et al., peripheral 2009), weblood100 and purified cell types,100 suggest- 100 death domain (Edaradd) can cause loss of hair, sweat glands, and bound with NGFI-A. 80 marks, whole-genome bisulfite DNA methylation, DNA accessibility, digital 80 80 80 LG-ABN mothers (n =5 animals/group; teeth [17], and it can reduce the speed of wound healing [18]. (Figure 2D),ing that even intermediate though methylation the overall is not distributions simply reflecting of differential age were 60 genomic footprints (DGF), input DNA andexamined chromatin whether conformation the model couldTo distinguish determine aging whether rates TSA for60 treatment reverses60 the maternal effect 60 P < 0.001) as a function of age. There were no Further research should focus on their role in ageing, and age- 40 72 methylation between cell types, but40 probably reflects a stable40 state of 40 informationrelated. diseases.c, Individual epigenomic marksindividuals across all with epigenomes differentnot in which genetic significantly variants. different For this purpose, between we the men and women in the * differences at any postnatal age in level of on methylation within specific20 CpG dinucleotides20 on the exon 1 20 * * 20 The lack of epigenetic drift within each monozygotic pair cell-to-cell variability within a population of cells of the same type. 7 expression in offspring cytosine methylation of the 3 CpG (site 17). GR promoter GR promoter ′ they are available. d, Relationship of figure panels highlights data set dimensions. 0 0 0 0

obtained whole-exome sequences for 252 of the individuals in C-methylation (%) Jay Smith/DISCOVERC-methylation (%) Figure 2. Percentage methylation versus age for three markers contrasts with a previous study [3]. The main difference between cohort (pGR > promoter, 0.05, KS we test). mapped Likewise, the differencesL-L H-H the H-L validation L-Hin methylationL-L H-H cohort H-L using L-H the E20 P1 P6 P21 P90 E20 P1 P6 P21 P90 validated in three sampleAge sets. Original twin samples are blue, male the two studies is that we focused on CpGour sites methylome close to functional study at 15 coverage. After sequence process- 3 Epigenomicsodium bisulfite differences technique, during focusing lineage onspecification the NGFI-A consensus Age (d) controlFigure samples are 2. red, Model female control Predictions samples green. and Linear Clinical trendlines Variablesgene transcription start sites whereas Fraga and colleagues confirmed the increased aging rate for men (p < 0.05), but was are shown in the colors of the individual sample sets a) Edaradd investigated random sites, most of whichFigureing were 4. and located Multitissue quality in non- Support control, these sequences yielded 10,694 r = 20.81 (twins), r = 20.73 (male controls), r = 20.75 (female controls)accessibility (Extended Data Fig. 3a), lower methylation (Extended Data Wesequence next studied within the relationshipthe exon 1 betweenregion DNA (Fig. methylation 1a). Statistical dynam- analysis of Bocklandt et(A) al A2011 flow chart PLoS of theONE data (green boxes) and analysesfunctional repeated (red ovals) sequencesHannum used (e.g., to Alu repeats).et al 2013 This suggests Molecular that inconclusive Cell for BMI (p > 0.05). This7 complements a previous Weaver et al 2004 Nature Neuroscience b) NPTX2 Figurer = 0.52 (twins), 3. Geneticr = 0.79 (male Effects controls), on Methylomicr = 0.03 (femaleFig. 3b) Aging and higher transcription factor(A)common Predictions binding (Extended of single-nucleotide age made Data by Fig. the2c) full agingvariants model across on the TCGA the control population generate aging predictions (blue boxes). while drift may occur randomly with age in non-coding, repeat- icsthe and data histone across modifications all 17 sites across revealed 95 epigenomes a significant with methylation effect of Group controls) c)(A) Tom1L1 We surveyedr = 20.70 (twins), genomicr = 2 variants0.49 (male for controls), an association with age-associated samples.(Experimental There is a high Procedures correlationfinding between). As of a chronological negative an epigenetic control, and predicted we signalage, confirmedadulthood. for BMI Gene that expression does is not controlled19,48–50 change by the epigenome, which is sites of the exon 1 GR promoter sequence (Fig. 1b,c). A two-way r = 20.24(B) (female A comparison controls). of predicted and actual agesthan for enhancersrich all individuals DNA regions, lacking based the H3K27ac. critical on regulatory the In a portions subset of of the 7 genome epigenomes with an data, extending previous studies that focused on individual lineages . 7 doi:10.1371/journal.pone.0014821.g002methylation markers. Eight genetic variants, correspondingremain underto strict 14 epigenetic meQTLs, controlbut throughout each tissue life. has a different linear intercept( andF = slope. 93.2, P < 0.0001), Treatmentcomprised ( ofF = chromatin 52.8, P < structure0.0001)17 andand Region DNA methylation18.We ANOVA revealed a highly significant effect of Group (F = 55.9, aging model. average of 24 epigenomic marks, wethat learned none separate of the 50-state geneticwith chro- variants age (WeFeinberg were found significant that et the al., distribution predictors 2010). of methylationof levels for CpGs in some were chosen for validation. Of these, seven were significant in the validation (B) After adjusting the intercept and slope of each(F tissue,= 30.4, the errorP < of0.0001), the model astested well the as ahypothesis significant that Group maternal × careTreatment alters DNA × methylation of P < 0.0001) and Region (F = 27.7, P < 0.0001), as well as a significant (C) Out-of-samplecohort and two showed predictions an association for individuals withmatin AMAR. in the state validation models cohort. based on all theis available similarage itself, to thathistone which of the marks original is to and be whole-blood DNA expectedchromatin data. since Age predictions the states genome varied made significantly sequence on across tissue and cell type (Fig. 4g, PLoS ONE | www.plosone.org 3 June 2011 | Volume 6 | Issue 6 | e14821 As genetic associations have been previously reported with (B) A plot of the trend between the methylationaccessibility marker cg27367526 in each epigenome (STEAP2) (Supplementarycancer samples are Fig. presented 4), which in Figure addi- S2. ExtendedRegion Datainteraction Fig. 4f and(F = Supplementary the2.1, GRP =0exon.01), 1 Table7 promoter, Group 4a). For× andTr example, eatmentthat these interac- changes are stably main- Group × Region interaction effect (F = 27.7, P <0.0001). (D) Apparent methylomic aging rate (AMAR) for each individual, based on the is considered to be relatively static over Group the coursePublishing Nature of a©2004 lifetime. tionally distinguished: a DNase state with distinct transcription factor tained into adulthood and associated with differences in GR expres- Importantly, the cytosine residue within the 5′ CpG dinucleotide agingand model age. The without state clinical of variant variables. rs42663 The (GTPBP10) distribution causes of aging an offset rates in shows this (C)On Agethe predictions other madehand, on one matchedhuman might normal expect longevityTssAFlnk andtion tumor to ( findF samplesstates= and genetic19.9, were from agingP largely variants TCGA.< 0.0001), phenotypes unmethylated that Group in × (terminallyAtzmonRegion differentiatedinteraction et al., 2006 (F =; 4.1, fasterrelationship. aging for men than women. A tablebinding of the markers enrichments used (Supplementary in the aging Predictions Fig. 4f), are including adjusted for for mediator/ the linear offsetcellsP of and< the0.0001)parent tissues, tissue but and frequently (breast, Treatmentsion methylated and HPA forRegion responses several interaction pluripotentto stress. and (F = 2.8, (site 16) of the NGFI-A consensus sequence (Fig. 1c) is always 43 modulate the methylationSuh of et age-associated al., 2008; Willcox markers, et i.e., al., meth- 2008; ×Wheeler et al., 2009), we model(C) Ais second provided example in Table for S3 cg18404041. andcohesin rs2230534 components (ITIH1, NEK4).(even thoughkidney, CTCF lung, was or not skin). included as an input embryonic-stem-cell-derived cells (Bonferroni-corrected F-test P , 0.01); methylated in the offspring of low-LG-ABN mothers, and rarely ylation quantitative-trait loci or meQTLsP < 0.0001). (Bell et al., The 2011 results). Testing again revealed significant differences in the SeeSee also alsoFiguresTable S2 S4 andfor a S3 tableand ofTable confirmed S3. genetic associations. (D) Tumor samples show a significantexamined increase whether in AMAR. the model couldRESULTS distinguish aging rates for methylated in the offspring of high-LG-ABN dams. In contrast, the 320 | NATURE | VOL 518 | 19See FEBRUARYeach also Figure genetic 2015 S4 and variantTablefor S5. associationmethylation with the top of age-associated a number of regions of the exon 17 GR promoter http://www.nature.com/natureneuroscience Maternal care and methylation of exon 17 promoter 3′ CpG dinucleotide (site 17) remains methylated, regardless of dif- ©2015 Macmillanindividuals Publishers Limited. with All rights different reserved genetic variants. For this purpose, we DNA. This meQTL thus captures a cis relationship in which methylation markers, we identifiedsequence 303 meQTLs (Fig. 4 ()Experimental with significantDNA methylation differences is a stable, within epigenomic the 5′ CpG mark (site at CpG dinucleotides ferences in maternal care. Dissected hippocampi inevitably contain obtained whole-exome sequences for 252 of the individuals in 18–20 Wers140692 validated influences this model the on methylation the secondary state cohort, of MBD4 consisting. That DNAProcedures damage through, FDR < oxidative 0.05, Figure stress 3 in16)A). mammalian Forand validation, 3′ CpG cells (site (weHart- selected 17)often dinucleotides associated with of stable the variationsNGFI-A in consensus gene transcription .Two glial cells as well as neurons. Considering the pronounced effect of ofMBD4 an additionalplays a role 174 in independent human aging samples. is supported These by previous samples wigeight and genetic Schlepegrell, variants 1995our (corresponding; methylomeKarthikeyansequence et to study al., 14 ( 2002 meQTLs)Fig. at). 4b15 These).3 Statisticaltocoverage. testkinds in of analysis changes After of sequence in the DNA data methylation process-from these are knowntwo to affect gene maternal care on the methylation status of the 5′ CpG dinucleotide werework processedFigure linking 2.MBD4 in Model theto same DNA Predictions repair, manner asas well and the as Clinicalprimary work showing cohort Variables that and meQTLsa validation represent cohort genetic ofing 322 variants individuals and that qualitysites appear from revealed our to control, broadly methylation a highlyinflu- these study.ex significantpression: sequences regional, effect non-site yielded of Group specific 10,694 ( F DNA= 43.8, methylation around a of the NGFI-A response element (>90%), the effect of maternal care 19 weremutations(A) then Aused flow andknockdowns chart to predict of the ofage dataMBD4 based (greenlead on to boxes) increased the original and genomic analyses model (redenceThis ovals) the analysis aging used methylome found to that and seven may ofP be eight < good0.0001), genetic candidates variants Treatment for (corre-promoter (F = 65.3,and site-specificP < 0.0001) methylation. and Hy Regionpomethylation of CpG must include neuronal as well as glial cells; both populations express common single-nucleotide variantsdinucleotides across of regulatory the regions population of genes is associated with active GR23,24 and NGFI-A25 genes. (i.e.,instability asgenerate trained (Bellacosa aging on the predictions et original al., 1999; cohort). (blueBertoni boxes). et The al., predictions 2009). were furthersponding age-associated to seven disease meQTLs) and remained longevityhighly research. significant in the Figure 5 TSA eliminates the maternal effect on hippocampal GR (Experimental(F = Procedures113.3, P < 0.0001),). As ach as negativero wellmatin as structurea significant control, and we Group transcriptional confirmed × Treatment activity18,20expression.Thus, the and HPA responses to stress. (a) Top: a representative western highlyOf(B) accurate, the Aseven comparison withvalidated a ofcorrelation meQTLs, predicted three between and were actual ageidentified ages and for predictedthat all had individualsvalidation basedcohort on the (FDR < 0.05, Table S4). While all of these vari- interaction (F = 16.0, Pmethylation < 0.0001), pattern Group is a stable × Region signature interaction of the epigenomic status of a Cross-fostering reveals epigenetic marking by maternal behavior a statistically significant association not only with age but also A Multitissue Diagnosticthat none of the genetic variants were significant predictors of blot showing absolute levels of electrophoresed hippocampal GR age ofaging 91% and model. an error of 4.9 years (Figure 2C). The significance ants acted in cis with their meQTLs (within 150 kbp),regulatory we sequence. We focused on the methylation state of the exon Our findings suggest that specific sites within the exon 1 GR pro- with aging rate (AMAR, FDR < 0.05, Figures 3B and 3C). One is Our aging model was derived from whole(F blood,= 37.8, whichP < is0.0001) advan- and Treatment × Region interaction (F = 4.5, immunoreactivity (IR) from vehicle- and TSA (100 ng/ml)-treated adult7 of the(C) aging Out-of-sample model was also predictions confirmed for by individuals the data set in presented the validationconfirmed cohort. that none directlyage itself, modified which the CpGis to site be or expected associated1 GR sincepromoter, the which genome is activated sequence in the hippocampus in offspring moter are differentially methylated as a function of maternal behav- the genetic marker rs2230534, which is a synonymous mutation tageous in the design of practical diagnosticsP =0.04). andPost-hoc for testing analysis7 revealed that TSA treatment signifi- offspring of high- or low-LG-ABN mothers. Molecular weight markers in Heyn(D) et Apparent al., verifying methylomic the age aging association rate (AMAR) of 70 for of each the individual, 71 probe based sequence on the of the associated methylation marker. of high-LG-ABN mothers. ior, but these findings are merely correlational. To directly examine in the gene NEK4, and has a cis association with the methylation samples collected fromis other considered studies.cantly To to investigate decreased be relatively whether the degree static of over cytosine the course methylation of a lifetime.within the 5′ (SeeBlue, Santa Cruz Biotech) correspond to a single major band at 92 markersaging (Heyn model et al., without 2012). clinical Furthermore, variables. the The model distribution was able of agingThe rates methylation shows marker cg27193080 was one of those foundTo determine whether DNA methylation of specific targetkDa. Thesites middle on the panel relation shows between the membrane maternal reprobed behavior for α and-tubulin DNA IR, methylation marker cg18404041. The NEK family of kinases plays a key role our aging model was representativeOn the other(site of otherhand, 16) tissues, CpG one17 mightdinucleotide we ob- expect of to the find NGFI-A genetic binding variants region that of the to fully separate old and young individuals in the Heyn et al. to be significantly associated with age (p < 10Ͻ ), and its meth-the GR promoter change in response to maternal care,illustrating we mapped absolutewithin levels the exonof electrophoresed 1 promoter, wehippocampal performed proteinan adoption bound study in in cell-cyclefaster aging regulation for men and than cancer women. (Moniz A table et al., of 2011 the). markers The tained used DNA in the methylation aging profiles for 368exon individuals 1 GR in promoter the control in the offspring of low-LG-ABN mothers in 7 modulate the methylation7 of age-associateddifferences in the markers,methylation i.e.,status meth-of individual cytosinesto the transferwithin membrane.which the biologicalMolecular offspringweight markers of high- correspond or low-LG-ABN to a single mothers study,secondmodel even variant for isprovided profiles is rs2818384, obtained in Table which via S3. bisulfate is a synonymous sequencing mutation rather categoryylation fractionof The Cancer was found Genome to be Atlasinfluenced (TCGA) by ( theCollins single-nucleo- and comparison to vehicle-treated21 low-LG-ABN mothers (*P < 0.001). ylation quantitative-trait loci or meQTLsthe CpG dinucleotides (Bell et al., of 2011 the exon). Testing 17 promoter from majorhippocampal band at ∼60were kDa cross-fostered and the intensity to either of high- the signal or low-LG-ABN was similar dams in all within 12 h thanin the gene bead-chipJAKMIP3 technologyand has a usedcis association in this study with (Figure the methyl- S2). Barker,tide polymorphism 2007 ),Group including (SNP)Publishing 83 breast, variantNature ©2004 183 rs140692 kidney, (p 60 < lung, 10Ͻ and)(Figure 42 3B). See also Figures S2 and S3 and Table S3. TSA treatment producedtissue ‘demethylation’ from the adult offspring of the 5of′ high-CpG and(site low-LG-ABN 16) lanes. mothers. The lowerof panel birth shows9.Cross-fostering quantitative produced densitometric a pattern analysis of exon (relative 1 promoter ation marker cg05652533. Copy-number variants in JAKMIP3 skinThis samples. meQTL An was aging particularly modeleach basedgenetic interesting on bothvariant as our both primary for the association SNP and and the with the top age-associated 7 and 3 CpG (site 17) dinucleotidesWe used sodium in the bisulfite offspring mapping of21,22 low-LG-ABN,with a particularoptical interest density, in methylationROD) of GR IRthat levels was associatedfrom samples with ( nthe =5 rearing animals/group; mother (F = 4.8, Methylomehave been Aging previously Rate associatedand Its Associations with glioblastoma (Xiong validationmethylation cohorts marker demonstrated mapped tostrong the predictivegene′ methyl-CpG power for binding methylation markers, we identifiedthe region 303 around meQTLs the NGFI-A (Experimental consensus sequence (*FPig. < 1a0.001).). The (b)P Plasma< 0.05; Fcorticosteroneig. 1d) and thus responses reversed7 the(mean difference ± s.e.m.) in methylation to a at Whileet al.,the aging2010). model The final is able variant to predict found the to influenceage of most AMAR individ- is chronologicaldomain protein age in4 ( theseMBD4 samples, with the (expectedmothers, SNP in an value and intron Rhypomethylation = and 0.72), the meth- of the 3′ CpG (site 17) dinucleotide results showed significant differences in the methylation20-min of specific period ofspecific restraint cytosines, stress (solid notably bar) at thein vehicle- 5′ CpG dinucleotideand TSA (site 16) of the ualsrs42663, withWe high which validated accuracy, is a missense it this is equally model mutation valuable on in the as gene secondarya toolGTPBP10 for identi- cohort,althoughylation consisting eachmarker tissue just upstream hadProcedures a clear of thelinearin coding, the FDR offset offspring region), < (intercept 0.05, oneofFigure andhigh-LG-ABN of the 3fewA). For mothers validation, (Fig. 4bwe). selected These findings (100 ng/ml)-treated adult offspring of high- or low-LG-ABN mothers andof associates an additional with cg27367526 174 inindependent the gene STEAP2. samples. STEAP2 Theseslope) from samples the expectationeight (Figure genetic 4A).suggest This variants offset that wasTSA (corresponding consis- treatment can reverse to 14 the meQTLs) hypermethylated to test status in of fying individual outliers who do not follow the expectation. For known genes encoding a protein that can bind to methylated (n = 10 animals/group; *P < 0.01). is knownwere to processed play a role in in maintaining the same homeostasis manner ofas iron the and primarytent within cohort a tissue, and evena across validation differentthe cohort batchesexon 1 of7 ofGR 322 the promoter TCGA individuals848 in the from offspring our methylation of low-LG-ABN study. mothers. VOLUME 7 | NUMBER 8 | AUGUST 2004 NATURE NEUROSCIENCE copper—metals that serve as essential components of the mito- data. We adjusted for each tissue trend using a simple linear were then used to predict age based on the original model This analysisTSA found treatment that resulted seven in of a more eight extensive genetic change variants in DNA (corre- methyla- chondrial respiratory chain (Ohgami et al., 2006). Studies have Molecularmodel, producing Cell 49, 359–367, age predictions January with 24,tion 2013an error thanª2013 comparable maternal Elsevier toInc.care 361per se, since the 3′ CpG (site 17) dinu- shown(i.e., that as perturbations trained on of the iron originalconcentrations cohort). can induce The predictionsthat found in blood were (Figuresponding 4B). Furthermore, to seven predicted meQTLs) AMARs remained highly significant in the cleotide, which is unaffected by maternal behavior, is partially NGFI-A response element, we found that methylation of the cyto- highly accurate, with a correlation between age and predicted validation cohort‘demethylated’ (FDR < in 0.05, responseTable to S4 TSA). While treatment all of these in both vari- cohorts sine within the 5′ CpG dinucleotide (site 16) completely eliminated 362ageMolecular of 91% Cell and49, 359–367, an error January of 4.9 24, years 2013 ª (2013Figure Elsevier 2C). Inc. The significance ants acted(Fig. in 4bcis). Also,with as in their the original meQTLs study (within (Fig. 1b), 150 maternal kbp), care we altered the binding of NGFI-A, whereas methylation of the cytosine within of the aging model was also confirmed by the data set presented confirmedthe that methylation none directly status modified of other the CpG CpG dinucleotides site or associated in the exon 17 the 3′ CpG dinucleotide (site 17) only slightly reduced NGFI-A pro- in Heyn et al., verifying the age association of 70 of the 71 probe sequencesequence; of in the the associated case of sites 1, methylation 2, 5, 12, 14 and marker. 15, these effects were tein binding (I.C.G.W., M.S. & M.J.M., unpublished data). markers (Heyn et al., 2012). Furthermore, the model was able The methylationsimilarly reversed marker withcg27193080 central TSA was infusion. one of The those significance found of these sites for transcription factor binding is17 currently unknown and Reversal of maternal effect on GR expression to fully separate old and young individuals in the Heyn et al. to be significantly associated with age (p < 10Ͻ ), and its meth- thus a focus of ongoing studies. Thus, stable DNA methylation GR gene expression in the hippocampus is increased in the adult study, even for profiles obtained via bisulfate sequencing rather ylation fractionmarking was by found maternal to be behavior influenced is reversible by the in single-nucleo- the adult offspring offspring of high- compared with low-LG-ABN mothers7,9.We 21 than the bead-chip technology used in this study (Figure S2). tide polymorphismhippocampus (SNP) by variantpharmacological rs140692 modulation (p < 10Ͻ of)(Figure chromatin 3B). struc- suggest that such differences are mediated by the differential This meQTLture. was While particularly TSA altered interesting the methylation as both of the theSNP both andthe 5 the′ and 3′ methylation of the 5′ CpG dinucleotide (site 16) of the NGFI-A Methylome Aging Rate and Its Associations methylationCpG marker sites within mapped the NGFI-A to the response gene methyl-CpG element, the former binding appears consensus sequence in the exon 17 GR promoter and the subse- While the aging model is able to predict the age of most individ- domain proteinto be critical 4 (MBD4 for, the with effect the SNPon NGFI-A in an intronbinding and to the exon meth- 17 pro- quent alteration of histone acetylation and NGFI-A binding to the moter. In a previous in vitro study using electrophilic mobility shift exon 17 sequence. If the differential epigenetic marking regulates uals with high accuracy, it is equally valuable as a tool for identi- ylation marker just upstream of the coding region), one of the few31 assays (EMSA) with purified recombinant NGFI-A protein and the expression of the exon 17 GR promoter in high- versus low-LG fying individual outliers who do not follow the expectation. For known genesdifferentially encoding methylated a protein olig thatonucleotide can bind sequences to methylated containing the offspring, then reversal of the epigenetic marking should be accom-

Molecular Cell 49NA,TURE 359–367, NEUROSCIENCE January 24,VOLUME 2013 7ª |2013NUMBER Elsevier 8 | AUGUST Inc. 3612004 851 ARTICLES ARTICLES ARTICLES

Figure 6 Methylation of odorant receptor genes in sperm DNA from a Olfr151 (over all CpG sites) b Olfr151 (individual CpG sites) conditioned F0 and odor naive F1 males. (a) Bisulfite sequencing of F0-Prop-Sperm * F0-Ace-Sperm CpG di-nucleotides in the Olfr151 (M71) gene in F0 sperm revealed that 100 100 Childhood care influencesF0-Ace adult mouse DNA (n =behavior 12) was hypomethylated compared with Environmentala Dorsal *bulb conditionsb Dorsal can bulb chave g h ARTICLES a b c6,000 g h that of F0-Prop mice (n = 10) (t test, P = 0.0323, t16 = 2.344). 6,000 8,000 8,000 ARTICLES ARTICLE(b) A particularS CpG di-nucleotide in the Olfr151 (M71) gene in 80 ARTICLE80 ARTICLESS through epigenetic programmingF0 sperm was hypomethylated in F0-Ace mice (n = 12) compared with transgenerational effects *** * *** * **** **** **** **** F0-Prop mice (n = 10) (P = 0.003, Bonferroni corrected). (c) We found F1 6,000 6,000 no differences in methylation between F0-Ace ( = 12) and F0-Prop 60 60 4,000 n Percentage methylation Percentage methylation 4,000 Figure 6( n Methylation = 10) mice acrossof odorant all of receptorthe CpG di-nucleotidesgenesFigure in sperm 4 queried Behavioral DNA infrom the Olfr6 sensitivity Olfr151and (overneuroanatomical all CpG sites) conditionedchangesOlfr151 (individualare fear inherited response CpG sites) a b CpG1CpG2CpG3CpG4CpG5CpG6CpG7CpG8CpG9 * gene in F0 sperm ( > 0.05). (d) Across *specific CpG di-nucleotidesprimers in used for bisulfiteF0-Ace- mapping were identicalF0-Prop-Sperm to the published a Figure 1 b Behavioral sensitivity to odor is specific to the paternally conditioned F0 and odor naiveP F1 males. (a) Bisulfite sequencing of F0-Prop-Sperm Sperm a 300 b 300 4,000 4,000 abthe Olfr6 gene, we found no differencesin inF2 methylation and IVF-derived between F0-Ace generations.19 (a,b*) Responses of F2-C57Bl/6JF0-Ace-Sperm* 300 conditioned odor.150 (a,b) Responses of individual C57Bl/6J F1 male CpG di-nucleotides in the Olfr151 (M71) gene in F0 sperm revealed that Medial bulb (pixels) 60 n.s. 100 Medial bulb (pixels) * n.s. 100 (pixels) ab sequence , thus eliminating potential primer bias between subjects in (pixels) ( = 12) and F0-Prop ( = 10) mice (Bonferroni corrected). (e) Bisulfite Olfr6 (over all CpG sites) d Olfr6 (individual CpG sites) e d f e* f2,000 * 1.8 F0-Ace mousen DNA 1.2(n = 12) nwas hypomethylatedmales compared revealed with that F2-Ace-C57c mice had an enhancedd 200 sensitivity* to 200 offspring conceived100 after the F0 male was fear conditioned2,000 with sequencing of the Olfr151 (M71) gene in F1 sperm revealed thatsodium F1-Ace bisulfite mapping. F0-Prop-Sperm 200 2,000 50 that of F0-Prop mice (n = 10) (t test, P = 0.0323, t16 = 2.344). F0-Ace-Sperm acetophenone. F1-Ace-C57 mice had an enhanced sensitivity to 2,000 1.6 mouse DNA (n = 4) was hypomethylatedacetophenone compared with that compared of F1-Prop with100 F2-Prop-C57 mice (a).100100 In contrast, 100 (b) A particular CpG di-nucleotide in the Olfr151 (M71) gene in The rat homolog80 of the exon 1 NR3C180 promoter, the exon 1 50

1.0 Dorsal glomerulus area F 7 acetophenone (a), but not to propanol (controlDorsal glomerulus area odor, b) compared with Medial glomerulus area propanol 1.4 mice (n50 = 4) (t test, P = 0.0153, t14 = 2.763). (fControl) Bisulfite sequencingSuicide nonabused Suicide abused 100 Medial glomerulus area 40 n.s. F0 sperm was hypomethylated in F0-Ace miceF2-Prop-C57 (n = 12) compared mice with had an enhanced sensitivity to propanol0 compared 0 of CpG di-nucleotides in the Olfr151 (M71) gene in F1 sperm region,revealed is differentially methylated as**acetophenone a## function of variations in F1-Home-C57 mice0 (F1-Ace-C57,0 n = 16; F1-Home-C57,0 n = 13; t test,0 0

F0-Prop mice (n = 10) (P = 0.003, Bonferroni corrected). (c) We found 80 80 propanol Percentage OPS to 1.2 that two40 particular0.8 CpG di-nucleotideswith in the F2-Ace-C57 Olfr151 (M71) gene mice were (b; F2-Prop-C57,4,17,22 n = 8; F2-Ace-C57, n = 12; OPS Percentage OPS to 0 = 0.043, t = 2.123). (c,d) Responses of M71-LacZ F1 male offspring 30 no differences in methylation between F0-Ace ( = 12) and F0-Prop 60 F1-Home–10060 F1-PropF1-Home–100 F1-AceF1-Prop P 27 –50F1-Ace n Percentage methylation maternal care . Cytosine methylationPercentage methylation is a highly stable epigenetic 1.0 hypomethylated in F1-Ace mice (n = 4) compared with F1-Prop mice acetophenone (n = 10) mice across30 all of the CpG di-nucleotidesto acetophenone: queried in the Olfr6 t test, P = 0.0158, t18 = 2.664; OPS toF1-Home-C57 propanol: F1-Ace-C57 t test, F1-Home-C57 F1-Ace-C57 conceived after the F0 male was fear conditioned with acetophenone or Percentage OPS to (n = 4) (P = 0.002,0.6 Bonferroni corrected). Data are presented as Percentage OPS to && ## mark that&& regulates ##60 gene expression& ## via60 itsCpG1 effectsCpG2CpG3CpG4CpG5 onCpG6 transcriptionCpG7CpG8CpG9 –100 –100 Percentage methylation 20 0.8 gene in F0 sperm (P > 0.05). (d) Across specific CpG di-nucleotides in F0-Prop- F0-Ace- Percentage methylation propanol. F1-Ace-M71 mice had an enhanced sensitivity to acetophenone (c), mean o20 s.e.m. *P ARTICLE< 0.05 after correction. =S 0.0343, = 2.302). (c23,24–f). SpermF2-Ace-M71Sperm micec whose600 F0 generation25 d 600 P t17 ## F2-Prop-C57 F2-Ace-C57 F2-Prop-C57F1-Ace-M71F2-Ace-C57 promoter (GR) promoter factor& binding . We used sodium** bisulﬁte mapping to examine the Olfactory system glomeruli (M71) F1-Ace-M71 F1-Ace-M71 F1-Ace-M71 the Olfr6 gene, we found no differences in methylation between F0-Ace CpG1CpG2CpG3CpG4*CpG5CpG6CpG7CpG8**CpG9 * but not to propanol (d), compared with F1-Home-M71,F1-Prop-M71 and F1-Prop-M71 F1-Prop-M71 F1-Prop-M71 F1-Prop-M71 /GAPDH (log(conc)) 0.6 0.4 /GAPDH (log(conc)) F0-Ace- F1-Home-M71 F1-Home-M71 F1-Home-M71 F1-Home-M71 Methylation Figure 3 Neuroanatomical characteristics of the olfactory system in Methylated clones (%) Methylated 10 F0-Prop-Sperm Sperm Figure 3 Neuroanatomical characteristics of the olfactory system in 10 (n = 12) and F0-PropF (n = 10) mice (Bonferronimale corrected). had been (e) Bisulfite conditioned to acetophenoneOlfr6 (over all CpG sites) had larger dorsalOlfr6 (individual and CpGmedial sites) mice. In contrast, F1-Prop-M71 mice had an enhanced sensitivity to

promoter (GR) promoter methylationc status of individual CpGd dinucleotides400 in the NR3C1 400 total 0.4 sequencing of the Olfr151 (M71) gene in F1 sperm revealed that F1-Ace F2 F1 males afterF0-Prop-Sperm paternal F0 olfactoryF1 males fear after conditioning. paternal F0 (olfactorya–f) -galactosidase fear conditioning. (a–f) -galactosidase 0 Gene expression B B Glucocorticoid Receptor propanol (d), but not acetophenone (c) (F1-Home-M71, = 11; quantitativeGR1 0.2 PCR was performedM71 for the glomeruli Olfr151 gene. in We the did notolfactory bulb than F2-Prop-M71 mice whose n

GR F0-Ace-Sperm 0 (frequency) Methylation promoter sequence100 Olfr151 extracted (over all CpG from sites) the hippocampal100 Olfr151 (individual tissue CpG sites) of the same g 0.2 mouse DNA (n = 4) 1was 2 hypomethylated 3 4 5 6 7 compared 8 9 10 11with 12 that 1314 of 15F1-Prop 16 17 18 19 20e 21 22 23 24 25 26 27 28 29 30 31f 32200 33 34 35 36 37 38 200 i i observe any differencesFigure 4 in Behavioral histone-mediated sensitivity epigenetic and neuroanatomical signatures changes are inheritedstaining revealedconditionedF1-Prop-Sperm that fear offspring responsestaining of F0 revealedmales trained that offspring to acetophenone of F0F1-Ace-M71, males (F1-Ace-M71) trained n4,000 = 13;to acetophenoneF1-Prop-M71, had larger**** n dorsal =(F1-Ace-M71) 9; OPS to acetophenone: had larger dorsal mice (n = Glucocorticoid Receptor 4) (t test, P = 0.0153, t14 = 2.763).F0 generation (f) Bisulfite sequencing had been conditioned to propanol. Scalea bar represents b 350 350 0 around the M710 locus when chromatin was immunoprecipitatedsubjectsCpG used site for glucocorticoid* receptor100 expressionF1-Ace-Sperm analysis. Sodium propanol Suicide Suicide in F2 and IVF-derived generations. (a,b) Responses 100of F2-C57Bl/6J 300 150 Medial ANOVA, P = 0.003, F2,30 = 6.874; F1-Home-M71 versus F1-Ace-M71, *** ** *** ** ControlControl Suicideof CpGSuicide di-nucleotides in theControl Olfr151 (M71)Suicide gene inSuicide F1 sperm revealed and medialacetophenone 0 acetophenone-responding* and medialc 0 glomeruli acetophenone-responding (M71d glomeruli) in glomeruli the olfactory (M71 bulbglomeruli) compared in the with olfactory bulb compared with abused with antibodies that recognize histone200 modifications Mm. (g) Dorsal thatbisulfite either M71 per- mapping glomerular80 revealed area ainsignificant F2 generation effect80 *(M71-LacZ: on the percentage* of * nonabusedthat twoabused particular CpG di-nucleotidesmales revealednonabused in thethat Olfr151 F2-Ace-C57abused (M71) gene mice were had an enhanced sensitivity to Percentage OPS to Percentage OPS to 100 P < 0.05; F1-Ace-M713,000 versus F1-Prop-M71, P < 0.01; OPS to propanol: nonabused mit (acetylated H3) or repress (H3trimethyl K27) to transcription F1-Prop-M71200 and F1-Home-M71F1-Prop-M71 mice. Scale and barF1-Home-M71 represents 1mice. mm. Scale (g) Dorsal bar represents M71 glomerular 1 mm. (areag) Dorsal in F1 M71 glomerular300 area in F1 300 hypomethylated in F1-Aceacetophenone mice (n = 4) compared comparedF2-Prop, with with F1-Propn =F2-Prop-C57 7; mice methylatedF2-Ace, mice n (= clonesa ).8; In tcontrast, test, (that P is, < the0.0001, number t–20013 of = clones5.926). with at least one –200 ANOVA, P = 0.020, F2,26 = 4.541; F1-Ace-M71 versus F1-Prop-M71, 80 50 (n = 4) ((PSupplementary = 0.002, Bonferroni Fig. 7 corrected).). The fact thatData the are M71 presented locus wasas not epige- 80 generation100 (M71-LacZ: F1-Home,generation n = 38; (M71-LacZ: F1-Ace, n F1-Home, = 38; F1-Prop, n = 38;< 0.05). n F1-Ace,= 18; Dataarea ANOVA, are n =presented 38; P F1-Prop, < as 0.0001, mean n =s.e.m. F18;2,91 *ANOVA, = < 0.05, P ** < 0.0001,< 0.01. F2,91 = Figure 1 Hippocampal glucocorticoid receptorF2-Prop-C57 expression. ( amice,b(h)Mean±) hadMedial an enhanced M71 methylatedglomerular sensitivity to CpG propanolarea60 site in comparedF2 divided generation by the total(M71-LacZ: number60 ofF2-Prop, clones) between 0 P 2,000 o P P Percentage methylation netically marked via histones in the F0 sperm could indicate that we Percentage methylation Figure 2 Methylation of the NR3C1 promotermean o s.e.m. in the*P < hippocampus. 0.05with after F2-Ace-C57 correction. Twenty mice clones (b; F2-Prop-C57, were sequenced = 8; for F2-Ace-C57, each subject =for 12; methylation OPS15.53; mapping.F1-Home-M71 (a) Mean versus ± s.e.m.15.53; F1-Ace-M71, propanol F1-Home-M71 < 0.0001; versus F1-Ace-M71 F1-Ace-M71, versus < 0.0001; F1-Prop-M71, F1-Ace-M71 < 0.05).versus F1-Prop-M71, 250 < 0.05). 250 s.e.m. expression levels of total glucocorticoid receptor (GR) mRNAn = 6; (a F2-Ace,)and n = 10;n t test, P = 0.0006,n t = 4.44). 0(i) Dorsal M71 –50 P P P P did not immunoprecipitate with the relevant histone-modificationgroups ( F 3.47, P 0.05).14Post hocacetophenone testsCpG1CpG2 revealedCpG3CpG4CpG5CpG6 aCpG7 significantCpG8CpG9 o F1-Ace-M71 F1-Prop-M71 F1-Ace-M71 F1-Prop-M71

percentage of methylated clones for suicide victims with a history of childhood abuse (n 12), suicide victims withoutF0-Ace- a history ofPercentage methylation childhood abuse (n 12) to acetophenone: test, = 0.0158, = 2.664; OPS to propanol: test, F1-Home-M71 PropanolF1-Home-M71 Dorsal glomerulus t P t18 Percentage methylation F0-Prop-Sperm Spermt 60 60 Percentage OPS to Acetophenone ¼ Percentage OPS to glucocorticoid receptor 1F (GR1F)antibody in 12 suicide or that the victims epigenetic with basis aglomerular history of this ofinheritance area might¼ differencein IVF not beoffspring between (F1-Prop-IVF, suicide victims(h n) =Medial 23; with–100 F1-Ace-IVF, M71 a history glomerular of n =¼ childhood 16; area( h ) inMedial F1 –100generation M71 glomerular (M71-LacZ: area in F1-Home, F1suggest generation a doublen = 1,00031;(M71-LacZ: dissociation F1-Ace, F1-Home,n and = 40;specificity F1-Prop, n = 31;of the F1-Ace, odor association, n = 40; F1-Prop, and controls (n 12). The methylation percentage was calculatedP = 0.0343, as t17 the = number2.302). ( ofc–f clones). F2-Ace-M71 with at mice least whose one methylated F0 generation CpG site divided byF2-Prop-C57 the total F2-Ace-C57 trained F0F2-Prop-C57 trainedF2-Ace-C57 F0 F2-Prop-M71 F2-Ace-M71 200 childhood abuse, 12 nonabused suicidehistone based, victims instead and 12relying control on other subjects mechanisms, (b). such as DNA = 16; ANOVA,CpG1CpG2CpG3CpG4 CpG5 < CpG60.0001,CpG7CpG8 n = 16; = ANOVA,31.68; PF1-Home-M71 < 0.0001, F versus = 31.68; F1-Ace-M71, F1-Home-M71 < 0.0001; versus F1-Ace-M71, F1-Ace-M71 P < 0.0001;200 F1-Ace-M71 ¼ quantitative PCR was performed for the Olfr151 gene. We did not F1-Ace- n P F2,84 2,84 along with the inheritanceP of a behavioral sensitivity that is specific McGowan et almale 2009 had Nature been Neuroscience tconditioned test, P < to 0.001, acetophenoneabuse t37 compared= had4.083). larger withdorsal F1-Prop-Olfr151(jSperm) Medial nonabused (overand allmedial CpG M71Sperm sites) suicide glomerular victimsOlfr151 (areaP (individual 0.05)in CpG or sites) controls 100 100 Number of M71 OSNs number of clonesOutliers (* indicates excluded fromP r analysis0.05; n.s. includedmethylation indicatesn (as2 not reported control statistically subjects,above) or significant). non-codingn 1 RNA, (b) Methylation as has been ofe the NR3C1 promoter region,f showing the frequency Number of M71 OSNs observe any differences in histone-mediated epigenetic signatures F1-Prop-Sperm¼ Dorsal ¼ 33 ¼ versusall of F1-Prop-M71, them were mated P with< 0.0001). naiveversus females. (F1-Prop-M71,i) F1-Ace-M71 Thus, any Pfindings mice< 0.0001). had that a (largerito) F1-Ace-M71the F0-conditionednumberh of mice M71 odor.had OSNs a larger in the number MOE ofthan M71 OSNs in0 the MOE than 0 of methylation observed at each CpG site fordemonstrated suicide victims forM71 the withKitglomeruli locus a history. IVFin the ofoffspring olfactory childhood bulb (F1-Prop-IVF, abuse, (thanP o F2-Prop-M71 suicide0.05). Theren victims = mice16; was with F1-Ace-IVF,whose no no difference history ofn childhoodin= 19; the percentaget test,F1-Ace-Sperm abuse P and< of 0.001, methylated gef 5,000 suicide victims with a historyaround of childhood the M71 abuselocus when for glucocorticoid chromatin was receptor immunoprecipitated Olfr151: odorant* receptor100 F0 generation had been& conditioned to&& propanol.acetophenone Scale100 bar represents at various stages F1-Prop-M71 ofwe gestation obtained increases wereand M71# F1-Home-M71not glomerular the results *F1-Prop-M71 of miceextremeDias4,000 &(M71-LacZ: Ressler phenotype and F1-Home-M71 2013**** biasingF1-Home,Nature Neuroscienceor a mice n = To 6;(M71-LacZ: furtherF1-Ace, corroborate n F1-Home, = 6; F1-Prop, the n enhanced = ***6; n F1-Ace,= 4; behavioral ANOVA, n = 6; sensitivity F1-Prop, to n the = 4; ANOVA, control subjects1F and (*P ano additional0.05, **Pno 10.001,with suicide antibodies abused victim that withsuicides recognize a history versus histone of childhood controls;t modifications33 = 5.880). abuse,P o that0.05, Data eitherclones Pper-areobetween presented0.001, non-abused suicide as mean victims suicides o withouts.e.m.* versus childhoodP controls;< 0.05,Medial* abuse P o 0.05, and controls ¼ DISCUSSION 200 Mm. (g) Dorsal M71 glomerular area in F2 generationarea and preference (M71-LacZ: for acetophenone previouslyc in adolescent existing offspringd genetic37. This sensitivity. Both C57Bl/6J and M71-LacZ F0-conditioned4,000 odor, we conducted an independent behavioral assay ## and n 3 nonabused suicide victims for overall levels of glucocorticoid P = 0.0001, F2,13 = 18.80; F1-Home-M71P = 0.0001,3,000 F2,13 versus = 18.80; F1-Ace-M71, F1-Home-M71 P < 0.001; versus F1-Ace-M71 F1-Ace-M71, versus P < 0.001; F1-Prop-M71, F1-Ace-M71 versus F1-Prop-M71, P o 0.001, abused suicides versusmit non-abused (acetylated suicides;H3) or repress Bonferroni (H3trimethyl**postP < hoc K27)0.01,comparisons). to transcription***(PP <4 0.001,0.05; Fig. **** 2aP). Methylation< 0.0001. was limited to specific sites, with no 21 ¼ Focusing on classicalF2-Prop, conditioning n = 7; inF2-Ace, an F0 generation n = 8; t test,before P concep- < 0.0001,last t study13 = 5.926).exemplifies how the olfactorymice possess sensitivity the and M71 neuroanatomy odorant receptor in their olfactory epithelium that directly probes behavioral sensitivity using an odor concentration F1-Ace-M71 80 < 0.01).80 Data are presentedP as< 0.01).mean Data s.e.m. are *presented < 0.05, as ** mean < 0.01, o s.e.m. *** *3,000P < < 0.001,0.05, ** ****P < 0.01, < 0.0001. ***P < 0.001, ****P < 0.0001.F1-Ace-M71 F1-Prop-M71 F1-Prop-M71 receptor. * indicates P o 0.05;(Supplementary n.s. indicates Fig. not7). The statistically fact that the significant. M71 locus was notclone epige- showing global methylationP (Fig. 2b). There were no significant area o P P P P F1-Home-M71 F1-Home-M71 tion and using specific(h) Medial odors M71 as the glomerular conditioned areastimuli in allowedF2 generation us to of (M71-LacZ: offspring can F2-Prop, bear imprints ofand parental both experience. can consequently detect acetophenone.2,000 The main difference curve and thearea association time of the mice with these concentrations. neticallytag marked a specific via olfactory histones experience in the F0 and sperm follow could the salience indicate of that expewe - However, it is imperative to realize that all of the aforementioned n = 6; F2-Ace, n = 10; t test, P = 0.0006,correlations t14 = 4.44). between (i) Dorsal levels M71 of exon 1F methylation and age at death 2,000 did notrience immunoprecipitate at the level of behavior with the and relevantneuroanatomy histone-modification through subsequent manipulations of the parental conditionbetweenPropanol have the occurred strains whenAcetophenone is thatthe pups the OSNs ofDorsal glomerulus the1,000 M71-LacZ mice produce We found that F1-Ace males were able to detect acetophenone at lower glomerular area in IVF offspring (F1-Prop-IVF, n = 23; F1-Ace-IVF, n = 16; Percentage methylation (r 0.15, PPercentage methylation 460 0.05), brain pH (r 0.08,trained60P F04 0.05) or postmortem expression. The use of HEK293antibody cells or allowed that the epigenetic us to concurrently basis ofThe this inheritance dorsaleffect and might of medial not group be (M71-specificF 17.12, P o glomeruli0.0001)B-galactosidase and in a the significant olfactory in trainedM71-expressing interaction F0bulbs neurons F2-Prop-M71and can thereforeF2-Ace-M71 be concentrationsMedial glomerulus than F1-Prop males, whereas F1-Prop males detected generations. We tfound test, thatP < the 0.001, F1 and t F2 =generations 4.083). (werej) Medial extremely¼ M71 glomerularor embryos arearea in uteroin , thereby ¼assaying behavior and neuroanatomy 1,000 expression and increasedhistone HPAbased, instead activity, relying we on hypothesized other mechanisms,37 that such intervalas DNA (PMI,¼ r 0.24, P 4 0.05; Table 1). 22 transfect a number of expression vectorssensitive with to the high specific efficiency. odors used to condition The F0between mice. Using CpG a trans- sitein andanimals group that are ( Fextremely13.44, differentvisualizedefP ofromCpG10.0001). CpG2those. DorsalThisCpG3 conceivedCpG4 procedureCpG5 InCpG6 NGFI-A CpG7afterCpG8 allowed h us to examine a seldom studied propanol at lower concentrationsF2-Prop-M71 thanF2-Ace-M71 F1-Ace males (Fig. 2a,b), fur- IVF offspring (F1-Prop-IVF,of F2-Ace-M71 n = 16; F1-Ace-IVF, mice were n = 19; significantly¼F1-Prop- t test, P < 0.001,F1-Ace- increased in size comparedboth F0-Ace5,000 and F0-Prop mice received shocks during conditioning, These data suggest that the effect of paternal olfactory fear con- suicide victims wouldmethylation showgenic decreased mouse(as reported in which expression above) OSNs expressing or bothnon-coding a ofspecific gluco- RNA, odorant as receptorhas been can perturbationSperm to the parent.¼Sperm bothIn other F0-Ace words, the and fetuses F0-Prop in the cited mice received shocks*** during conditioning, These data suggest that the effect of paternal olfactory fear con- Environmentalabsence of plasmid conditions replication during at the transientcriticalt transfection = 5.880). periods Data assay are presentedrecognition as mean elements,o s.e.m.*EpigeneticP < methylation 0.05, was observedfactor abnormalities that at CpGmight sites markedly 12, 13, influence 30, associated the nervous systems of adults; with ther suggesting an enhanced detection sensitivity that is specific to demonstratedbe visualized, for the we Kit 33found locus that33. the behavioralwith those sensitivity of was F2-Prop-M71 accompa- studies were mice directly (Fig. exposed 4c to– theh ).environmental Similar perturbation. results wereThis these data4,000 suggest that these training-specific effects do not occur ditioning on neuroanatomy is associated with increased numbers corticoid receptor and glucocorticoid receptor** < 0.01, 127F ***compared < 0.001, with ****

*** 10 30 50 70

27 7 30 50 70 90 40 60 80 *** rodents . Furthermore, fetal origins** of diseases have been1-D proposed 1-E after the 1-B father’s 1-F diet had 1-Gbeen manipulated10 30 50 1-C3,1-C2,1-C3 70 . Second, exposure 1-H to 2 the male offspring: inheritance via the gametes or transmission via a Reduced longevitydecreased associated glucocorticoid** with receptor* cardiovascular transcription disease observedCross-fosteringPoor and in suicideOur diabetes supports observations inheritance of ofthe information behavioral and structuralvalue of the changes odor would specific necessitate female subjecting germ the F0 line. generation Second, to given the0.01% possibilitymale0.03% offspring: 0.06%that mating inheritance with the via the gametes or transmission via a was no difference* between nonabused suicide victims and controls We used a transientGA transfection GG assayspecificAA in humanto AG paternal HEK293 conditioningAA cells AG to or couldAA also AG GGbe inheritedAA via GA the GG 0.001% 0.003% 0.006% % Metylation cg26668828 % Metylation cg13012494 % Metylation cg08942192 for a multitude of disorders as having their roots in the experience anti-androgenic endocrine disruptor% Metylation cg12016809 vinclozolin during embryonic Genotype rs367881881 F0an olfactoryGenotype appetitive rs75666188 odor-conditioning experienceGenotype rs62214051 affectsF0 task. olfactory GenotypeF1 neuroanatomy rs12118280 experience % Metylation cg11763394 Genotype affects rs4849179 F1 neuroanatomyAcetophenone concentration Propanolsocial concentration route that is reminiscent of the transmission of maternal care in victims with(P a4 history0.05; Fig. of childhood 1b). of the abuse fetus wasto the associated parentalOur observations environment withto differ- the whileof the F0-conditioned in behavioral utero35.examine From and a structural transcriptionalgonadalodor beingsex(Minor changes determination Allele: retained A activity; MAF=0.096)specific affects of in(Minor afemale fertilitythe NR3C1Allele: GF2 ;andgerm MAF= generation, promoterbehavior 0.057) line. (MinorSecond, in at Allele: ligated least Gand ;given fourMAF= to 0.11) the aF0 possibility(Minor conditioned Allele: G ; MAF= that 0.17) mating male(Minor withAllele: in A ;thesome MAF= 0.32) way alteredsocial maternal route thatbehavior is reminiscent toward of the transmission of maternal care in to the F0-conditioned odor being retained in the F2 generation, and F0In conditionedthe F019 generation, male in wesome previouslyPreviously way altered reported maternal19, we that reported behavior olfactory toward thatfear the behavioral response27 (increased rodents27. To begin to dissociate these two possibilities, we conducted 1.0 ences in methylationWe examined levelsPembrey occurring the relationshipchemosensory et al only 2006 at European specific betweenperspective, sitesJournal glucocorticoid anti-predatory in of the Human exonthe persistenceGenetics 1behaviorF receptor is transmittedpromoter-less of the structuralsubsequent firefly generations, effects luciferase after andPreviously it expression isIVF, associated argue with vector, againstwe epigenetic reported (pGEM-LUC, social changes that subsequently the behavioral born response offspring, (increasedFigure we wanted 2 Sensitivity torodents accountof F1 males. forTo toward beginany F0-conditioned differences to dissociate odor. Associationthese two possibilities, we conducted conditioning2,9,39 adult males to acetophenone increases FPSHerrera when et al.the in review from gravid female cricketsthe persistence to their offspring of the whenstructural the females effects are after in IVF, the spermargueChr ofagainst 4 descendant social male subsequently offspringChr 6 . bornA recent offspring,Chr study 7 used we wanted Chrto 6account for any differences time with either the concentration of odor on the x axis or an empty NR3C1 promoterexpression (Fig. and 2b). psychiatric An ANOVA diagnoses examining (Table the 1 methylation). Mood disorders of and Promega; Fig. 3a) in the presence or absence of ectopic NGFI-AFPS to acetophenone)19 of F0-Ace conditioned males is comple- experiments with the F2 generation and with IVF-derived mice. Naive Risk Ratio transmission36 and make a strong case forFPS transgenerational to acetophenone) inherit of- F0-Acein maternal conditioned investment males or isinformation comple- transferexperiments about withthe conditioned the F2 generation and with IVF-derived mice. Naive exposed to a high densitytransmission of a predator and. Finally, make indirectly a strong related case forto transgenerationala social69.4345 defeat mb procedure69.4355 inherit mb in mice- startle inand maternal 32.6325found stimuli mb paternal investment are pairedtransmission4.244 ormb with information4.245 ofacetophenone mb 32.7285 transfer mb presentation32.7295 about mb the conditioned. In the chamber was recorded. An aversion index was computed by subtracting CpG dinucleotidessubstance across abuse the disorders exon are 1F NR3C1 risk factorspromoter for suicide revealed and have a been mented by an increase in the number of acetophenone–responsive F1 males (F1-Ace, F1-Prop) were mated with naive females to gener- * our study is a report that supplementing**ance. the mouse Notably, maternal dietour with results depressive-like are highly69.435 mbbehavior 69.436specific mbinmented subsequentlyF1 32.632in generation, themb by 32.633 conceivedolfactory an mb increasewe adultfound sensory offspring. 4.2445 thatin mb theC57Bl/6J numberodor F1-Ace that32.729 of mb might miceacetophenone–responsive (F1-Ace-C57) result from theour amount conditioning of timeF1 spent males protocol.in the (F1-Ace, open chamber F1-Prop) from the timewere spent mated in the with naive females to gener- linked to alterations of glucocorticoid receptorance. Notably,** expression** our results12.There are highly specific in the olfactory sensory odor that might result from our conditioning protocol. significant effect of CpG site (F 13.86, P 0.0001), a significant Gene Granddaughters' Mortality 0.5 o Good Table 1 DemographicUGT2B17 characteristicsshowed andHLA-DQB1 psychiatricenhanced OPS diagnoses (unconditioned)SDK1 M71-expressing toHLA-DQB2 acetophenone OSNs compared in the MOEodor chamber.and M71 (a) When glomerular tested with acetophenone,area in ate F1-Ace F2 adultsmice detected (F2-Ace, F2-Prop) whose F0 ancestors had been condi- ** *** modality toward the F0-conditioned odor, and both F0-Ace and F0-M71-expressingWe found that, similarOSNs to in the the situation MOE in whichand theM71 F0 male glomerular (father) area in ate F2 adults (F2-Ace, F2-Prop) whose F0 ancestors had been condi- were no significant effects¼ of psychopathology, even aftermodality controlling toward the F0-conditioned odor, and both F0-Ace and F0- We found that, similar to theacetophenone situation at ina lower which concentration the F0 male(0.03%) (father) than F1-Prop mice, with ** Prop males were subjected to the same shock trainingCpG conditions that withwas C57Bl/6J conditioned F1-Home to acetophenone, mice the(F1-Home-C57) olfactory F1-Ace mice bulbs.(Fig. in 1 thisa To). No maternally examine differ- whether alterations in the neuro- tioned with either acetophenone or propanol. For the IVF experiment, for childhood abuse status,94 on overall glucocorticoidProp receptor males or were subjectedVOLUME to the 17 | sameNUMBER shock 1 | theJANUARY trainingolfactory 2014 NATURE conditions bulbs. NEUROSCIEN To thatexamineCE was whether conditioned alterations to acetophenone, in theboth neuro groups -F1-Ace eventuallytioned showingmice with inequal either this aversion maternallyacetophenone at the 0.06% concentration or propanol. For the IVF experiment, †0 42 6 8 10 12 14 16 18 20 DMR Abused suicideences Nonabusedbetween groups suicide were Control found when propanol was paired with might be deemed stressful and potentially conveyed to the mother. trained experiment had an enhancedanatomical OPS to acetophenone representation compared of the(P = 0.005 conditioned with Bonferroni odor correction accompa for multiple- spermcomparisons). from F0 males was collected 10 d after the last conditioning day, glucocorticoidAge of Paternal receptor Grandmother 1F expression (years) (P 4 0.05). might be deemed stressful and potentiallyanatomical conveyedthe startle, indicating to representation the mother.that the response oftrained the was specific conditioned experiment to acetophenone had odor an enhanced accompa- OPSsperm to acetophenone from F0 males compared was collected 10 d after the last conditioning day, Figure 3 In vitro analysis of NR3C1 promoter methylation.This (a )argues The relative against the idea that our results might merely be the trans- with F1-Home controls (Fig. 5a). If our behavioral findings were a (b) When tested with propanol, F1-Prop mice detected propanol at a lower Male/female 80 12/0 (Fig. 1b). Similarly,12/0 F1-Ace-M7112/0nied showed the behavioral enhanced OPS sensitivity to aceto- reported above, we used standard and IVF was performed by the Transgenic Mouse Facility at Emory b position of the NR3C1Slowvariant Growth and Period the promoter sequence,mission showing of a the stressfulThis argues paternal against experience the toidea the thatmother our during results thenied mightresult the merely ofbehavioral a ‘social be thetransmission’ sensitivity trans- modewith reported of F1-Home information above, controls transfer, we used we(Fig. concentrationstandard 5a). If our (0.003%)and behavioral IVF than wasF1-Ace findingsperformed mice, with were both by groups athe Transgeniceventually Mouse Facility at Emory 2.0 bcAge (years) 60 34.2 ± 10 33.8 ± 11 35.8 ± 12 location of theGenotyping CpG dinucleotides. and methylation The 255-bp analysis (., solidtime underline) of co-habitation. and To ensure14,000 that our experimental groups4,000 wereB -galactosidasephenone,would have but predicted not staining to propanol, a reversal inB compared-galactosidasenaive of the aboveM71-LacZ with result. F1-Home-M71 staining Instead, F1 males we in found andnaive that showing M71-LacZhad nei equal- aversion F1University males at the that 0.006% in hada location concentration nei- independentUniversity (P = 0.0005 in with ofa location our laboratory independent at Yerkes of our where laboratory at Yerkes where mission of a stressfulPMI (h) paternal 40 experience24.63,500 ± 5.8 to the mother 39.0 ± 25.7 during 23.5 the ± 6.0 result of a ‘social transmission’ mode of information transfer, we 125-bp (,, brokenBecause underline) alterations deletion in glucocorticoid constructs are shown,receptorbalanced alongGood 1F foractivity with both odor could and beshock12,000 exposure, many of our experiments F1-Prop-M71that the F1-Ace-C57(fostered) (Fig. 1c,d). In contrast,ther male been F1-Prop-M71 offspring behaviorally still showed had a higher enhanced tested OPS with, Bonferroni nor exposed correction to,for multipleany of comparisons) the con- (F1-Ace-C57,we conducted n = 16; all of the other studies reported. Subsequently con- 20 3,000ther been behaviorally tested with, nor exposed to, any of the con- we conducted all of the other studies reported. Subsequently con- time of co-habitation.pH ToMean % methylation ensure that 6.3our ± 0.24experimental 6.5 ± 0.29groups were 6.5 ± 0.22 would have predicted a reversalF1-Prop-C57, of the aboven = 16). result.Data are presentedInstead, as we mean found ± s.e.m. (**P < 0.01). specific CpGderived dinucleotides from that nucleotide were methylated sequence in variation eachutilized deletion and/or F0-Prop construct, epigenetic as the mod-control10,000 group rather than F0-Home. OPSto acetophenone to propanol, butthan not F1-Home-C57(fostered) to acetophenone (Fig. offspring1c,d). These (Fig. data 5b ), Childhood abuse/neglect 12/02,500 (100%) 0/12 (0%) 0/12ditioned (0%) odors. We found that the dorsal and medial M71-specific ceived IVF offspring (F1-Ace-IVF and F1-Prop-IVF) were raised to 1.0 as indicated byifications, circles. Boxes we sequenced represent known the NR3C1 or putativepromoter canonical region (solid-balanced from each for 8,000both odor and shockp−value(FDR)=2.4e−12; exposure,ditioned manyp−value(FDR)=2.3e−29; of odors. our experiments Wep− foundvalue(FDR)=1e−39; that that thep− vthedorsalalue(FDR)=8.4e−38; F1-Ace-C57(fostered) and medial M71-specific male offspring ceived stillIVF had offspring a higher (F1-Ace-IVF OPS and F1-Prop-IVF) were raised to To further address this issue, and to address potentialR2=0.56 maternal2,000 suggestingR2=0.85 a biological, R2=0.93rather than social, R2=0.91mode of inheritance. lined box) and noncanonical (broken-lined box) NGFI-A–bindingPoor sites, with 6,000Mood disorder 8/12 (67%) 8/12 (67%) 0/12glomeruli (0%) in the olfactory bulb of F1 offspring of acetophenone- adulthood and tissue was collected in this facility. Risk Ratio subject. No sequence variation was seentransmission, among subjects utilizedwe conducted and allF0-Prop of a cross-fostering as the control study. Sexuallygroup 1,500rathernaiveglomeruli thanFor F0-Home.the in equivalent the olfactory experiment bulb toto visualize ofacetophenone F1 neuroanatomy,offspring than of we acetophenone- F1-Home-C57(fostered)per- adulthood offspring and tissue (Fig. was 5b collected), in this facility. 19 Alcohol/drug abuse disorder 9/12 (75%)90 6/12 (50%) 5/12 (42%) VOLUME 17 | NUMBER 1 | JANUARY 2014 NATURE NEUROSCIENCE the shaded areathe indicating sequences the were beginning identical of to the those exon. published (b,femalec) Mean previously mice ± s.e.m. were conditioned. Moreover, with4,000 acetophenone or left in their1,000 hometrained formed F0 a malessimilar cross-fostering (F1-Ace-M71)trained experiment were F0 males usingsignificantly M71-LacZ (F1-Ace-M71) females,increased were in significantly size When increased tested in in size our behavioralWhen tested assay, in F2-Ace-C57 our behavioral mice assay, exposed F2-Ace-C57 to mice exposed to Grandsons' Mortality To further address this issue, and to address potential maternal suggesting a biological, rather than social, mode of inheritance.

levels of luciferase expression in HEK293 cells. Results are shown after the 2,000 50 70 90 for each subject, the genomic sequencescage. targeted They for were binding then bymated the withData odor-naive are presented males as mean20 40for 60 ± s.d.10 d, after500 and used female mice conditionedcompared to propanol with as those our control of the group offspring of home cage or propanol- odors for the first time showed increased OPS to acetophenone com- 0.5 20 40 60 transmission, we conducted a cross-fosteringcompared40 60study. 80 with Sexually those naive of the offspringFor the of equivalent home cage experiment or propanol- to visualizeodors neuroanatomy, for the first time we showed per- increased OPS to acetophenone com- subtraction of expression of the promoter in the antisense orientation. 0 AA AGLuminance (arbitrary units) GG 0 AA GA GG AA GA GG AA GA GG †0 42 6 8 10 12 14 16 18 20 Luminance (arbitrary units)

which the male was removed. Subsequent offspring % Metylation cg10632656 were then divided % Metylation cg12192813 (offspring labeled% Metylation cg24441899 as F1-Prop).trained We % Metylation cg07180897 found F0 malesthat the (F1-Home-M71 increased dorsal and F1-Prop-M71, respectively) pared with F2-Prop-C57 mice, whereas F2-Prop-C57 mice showed (b) The full NR3C1Agepromoter of Paternal was Grandmother either unmethylated (years) (hGR) or completely Genotype rs35307342 trainedGenotype F0kgp6614124 males (F1-Home-M71Genotype rs1105697 Genotypeand rs9274514F1-Prop-M71, respectively) pared with F2-Prop-C57 mice, whereas F2-Prop-C57 mice showed into the followingfemale groups: mice offspring were of hGRconditioned home cage mothers(Minor with Allele: (F1-Home), acetophenoneG ; MAF= 0.33) (Minorand Allele: medial or A ; MAF= left 0.25)glomerular in their(Minor Allele: homearea A ; MAF= persisted 0.24) formed (Minorin F1-Ace Allele: a A ; similarMAF= mice 0.22) even cross-fostering after they experiment using M71-LacZ females, * p < 0.05 ** p < 0.02 *** p < 0.01 255 bp 125 bp (ANOVA, P < 0.0001 for dorsal and medial glomeruli; Fig. 3a–h). an increased OPS to propanol (Fig. 4a,b). This persistent behavioral patch methylated (hGR PatchM) and transfected in the presence or absence (ANOVA,255 bp M P < 0.0001125 bp Mfor dorsal and medial glomeruli; Fig. 3a–h). an increased OPS to propanol (Fig. 4a,b). This persistent behavioral NATURE NEUROSCIENCESlow Growth PeriodVOLUME 12 [ NUMBERoffspring3 of[ acetophenone-conditionedMARCHcage. They 2009 were then mothers mated (F1-Ace), with offspringodor-naive were males cross-fostered for 10 byd, mothersafter 343 conditionedand used tofemale propanol mice (F1-Ace- conditioned to propanol as our control group of NGFI-A. (c) The 255-bp and 125-bp NR3C1 deletion construct were either hGR PatchM This increase in M71 glomerular area was accompanied by a signifi- sensitivity to the F0-conditioned odor was accompanied by corres- 2.0 of home cage mothers cross-fosteredhGR + startingNGFI-A at postnatal day 1 byThis M71(fostered)). increase in M71In contrast, glomerular F1-Prop micearea cross-fostered was accompanied by mothers by a signifi- sensitivity to the F0-conditioned odor was accompanied by corres- unmethylated (255 bp or 125 bp) or methylated (255 bp M or 125 bpwhich M), as the male was removed. Subsequent255 offspringbp + NGFI-A were125 bp then+ NGFI-A divided (offspring labeled as F1-Prop). We found that the increased dorsal mothersPoor conditioned to acetophenone (F1-Home(fostered)), and off- conditioned255 bp M + NGFI-A to acetophenone125 bp M + NGFI-A (F1-Prop-M71(fostered))cant increase in the did numbers not show any of M71 OSNs in the MOE (ANOVA, ponding increases in glomerular size in an independent set of F2 into the following groups:hGR PatchM offspring+ NGFI-A of homecant cage increase mothers in (F1-Home), the numbers ofand M71 medial OSNs glomerular in the MOE area (ANOVA, persisted in pondingF1-Ace mice increases even afterin glomerular they size in an independent set of F2 shown in a, and transfected either in the presence or absencespring of of NGFI-A. acetophenone conditioned mothers cross-fostered by home increases in M71 glomerular areaP < ( Fig.0.0001; 5c–h). Fig. In summary, 3i). these cross- M71-LacZ mice that had no previous exposure to the odors used. offspring of acetophenone-conditioned mothersP < 0.0001; (F1-Ace), Fig. 3 ioffspring). were cross-fostered by mothers conditionedM71-LacZ to propanol mice that (F1-Ace- had no previous exposure to the odors used. 1.0 cage mothers (F1-Ace(fostered)) (Supplementary Fig. 4). Notably, fostering results, taken together with our IVF and F2 studies, strongly of home cage mothers cross-fostered starting at postnatal day 1 by M71(fostered)). In contrast, F1-Prop mice cross-fostered by mothers

Risk Ratio Good 344 VOLUME 12 [ NUMBER 3 [ MARCH 2009 NATURE NEUROSCIENCE mothers conditioned to acetophenone (F1-Home(fostered)), and off-NATUREconditioned NEUROSCIEN to acetophenoneCE VOLUME (F1-Prop-M71(fostered)) 17 | NUMBER 1 | JANUARY did 2014 not show any 91 92 NATUREVOLUME NEUROSCIEN 17 | NUMBERCE 1 | JANUARYVOLUME 2014 17 NATURE| NUMBER NEUROSCIEN 1 | JANUARYCE 2014 91

Granddaughters' Mortality 0.5 spring of acetophenone conditioned mothers cross-fostered by home increases in M71 glomerular area (Fig. 5c–h). In summary, these cross- cage mothers (F1-Ace(fostered)) (Supplementary Fig. 4). Notably, fostering results, taken together with our IVF and F2 studies, strongly †0 42 6 8 10 12 14 16 18 20 Age of Paternal Grandfather (years) Figure 1 The effect of paternal grandparental food supply (good red filled squares, poor green open squares) at different times in their early life on the mortality rate of their grandchildren. Both (a and b) show¼ on the Y-axis the mortality¼ RR of the grandchildren separated92 by sex, first the VOLUME 17 | NUMBER 1 | JANUARY 2014 NATURE NEUROSCIENCE grandsons’ mortality RR and below this, the granddaughters’ mortality RR. The age at which the paternal grandparent was exposed to good or poor food supply is given along the X-axis. The grandchildren’s mean mortality RR results are plotted for both those paternal grandparents who had a good food supply (red) and those who had a poor supply (green) at the specified age on the X-axis. (a) First relates paternal grandfathers’ exposure to his grandsons’ mortality RR and then below the paternal grandmothers’ exposure to her granddaughters’ mortality RR. (b) First relates paternal grandmothers’ exposure to his grandsons’ mortality RR and then below the paternal grandfathers’ exposure to her granddaughters’ mortality RR. The data points were obtained using a 3-year frame, advanced one year at a time, for grandparental age at exposure, to produce rolling means for the grandchild’s (proband’s) mortality RR for both ‘good’ and ‘poor’ ancestral food supply. Exposure to at least 1 year of surfeit of food or to at least 1 year of poor availability during a 3-year period denotes it as exposure to a ‘good’ period or a ‘poor’ period respectively. wFood supply at age 0 years is the mean for the 33-month period from 267 days until the day before the second birthday and therefore includes fetal life. À

European Journal of Human Genetics Summary Open Questions in Epigenetic • Epigenetic marks are involved in development Research

• DNA methylation at CpG sites is associated with down-regulation of gene expression

• Histone acetylation is associated with up-regulation of gene expression

• Tissue-speciﬁc cell-line memory is established and maintained through epigenetic modiﬁcations.

• Epigenetic changes can happen during the lifetime of individuals due to interactions with environment

• Environment can have effects on epigenome/

phenotype that can be inherited transgenerationally Allis et al 2015

Mutations of Epigenetic Machinery Many genes involved in epigenetic machinery have been identiﬁed to be mutated in cancer

Allis et al 2015