DNA methylation was measured by luminometric methylation assay between the 2 time points for the Iceland cohort (circles). Each circle represents an individual, and they are ranked on the x-axis based on the magnitude and direction of the change. The y-axis shows the absolute change between time points in the global percentage of methylation. The range of results from permutations of the time points within individuals, emulating the null hypothesis of no time effects, are shown as well (tinted region).
Change in HpaII methylation as measured by luminometric methylation assay between the 2 time points (average 16-year interval) for the Utah cohort, sorted by family. Each circle represents an individual.
Bjornsson HT, Sigurdsson MI, Fallin MD, Irizarry RA, Aspelund T, Cui H, Yu W, Rongione MA, Ekström TJ, Harris TB, Launer LJ, Eiriksdottir G, Leppert MF, Sapienza C, Gudnason V, Feinberg AP. Intra-individual Change Over Time in DNA Methylation With Familial Clustering. JAMA. 2008;299(24):2877–2883. doi:10.1001/jama.299.24.2877
Author Affiliations: Departments of Medicine (Drs Bjornsson, Sigurdsson, Cui, Yu, and Feinberg, and Mr Rongione), Epidemiology (Dr Fallin), and Biostatistics (Drs Fallin and Irizarry), and Center for Epigenetics (Drs Bjornsson, Sigurdsson, Fallin, Irizarry, Cui, Yu, and Feinberg and Mr Rongione), Johns Hopkins University, Baltimore, Maryland; Hjartavernd, Reykjavik, Iceland (Drs Aspelund, Eiriksdottir, and Gudnason); Department of Clinical Neuroscience, Karolinska Institute, Karolinska University Hospital-Solna, Stockholm, Sweden (Dr Ekström); Laboratory of Epidemiology, Demography, and Biometry, Intramural Research Program, National Institute on Aging, Bethesda, Maryland (Drs Harris and Launer); Department of Human Genetics and Eccles Institute of Human Genetics, University of Utah, Salt Lake City (Dr Leppert); and Fels Institute for Cancer Research and Department of Pathology, Temple University Medical School, Philadelphia, Pennsylvania (Dr Sapienza).
Context Changes over time in epigenetic marks, which are modifications of DNA such as by DNA methylation, may help explain the late onset of common human diseases. However, changes in methylation or other epigenetic marks over time in a given individual have not yet been investigated.
Objectives To determine whether there are longitudinal changes in global DNA methylation in individuals and to evaluate whether methylation maintenance demonstrates familial clustering.
Design, Setting, and Participants We measured global DNA methylation by luminometric methylation assay, a quantitative measurement of genome-wide DNA methylation, on DNA sampled at 2 visits on average 11 years apart in 111 individuals from an Icelandic cohort (1991 and 2002-2005) and on average 16 years apart in 126 individuals from a Utah sample (1982-1985 and 1997-2005).
Main Outcome Measure Global methylation changes over time.
Results Twenty-nine percent of Icelandic individuals showed greater than 10% methylation change over time (P < .001). The family-based Utah sample also showed intra-individual changes over time, and further demonstrated familial clustering of methylation change (P = .003). The family showing the greatest global methylation loss also demonstrated the greatest loss of gene-specific methylation by a separate methylation assay.
Conclusion These data indicate that methylation changes over time and suggest that methylation maintenance may be under genetic control.
Epigenetic marks are modifications of DNA or associated proteins, other than the DNA sequence itself, that are heritable through cell division. These include DNA methylation, a covalent modification of cytosine; histone modifications affecting the nucleosomes around which the DNA is coiled; and alterations in nucleosomal packing or higher-order folding of chromatin. We and others have suggested that epigenetics might play a role in the etiology of common human diseases.1- 3 In a recent review in JAMA by 1 of us (A.P.F.), it was noted that epigenetics stands at the epicenter of modern medicine because it unites nuclear reprogramming during development, environmentally induced changes on the body, and the ability of cells to respond appropriately to external stimuli.4 That is because epigenetic changes, unlike the DNA sequence, distinguish one tissue type from another and dietary and other environmental exposures alter the epigenetic program; the ability of genes to alter their expression is controlled by epigenetic factors such as DNA methylation.5,6 Diseases in which epigenetic change has been shown to play a major role include cancer and some disorders of the immune system, and epigenetic defects may also contribute to chronic diseases such as diabetes, bipolar disorder, and autism and loss of normal responsiveness to stress that accompanies aging.7- 9
A critical underpinning of the epigenetic hypothesis of common disease is that epigenetic marks change in the same individual over time. Recently, 14 monozygotic twins of 40 pairs tested were retrospectively found to be discordant in the degree of total DNA methylation and histone acetylation, with a preponderance older than 28 years.10 In contrast, a recent analysis of about 1 Mb of genomic DNA encompassing 40 000 CpG dinucleotides found no difference in DNA methylation related to age, although those data were based on average values rather than paired sampling in the same individuals over time.11 To prove that epigenetic marks change in an individual, a prospective study design is needed, and the only such study to date found no consistent methylation changes at 2 individual loci.12 We performed a direct examination of methylation in the same individuals over time to resolve this important question.
Icelandic samples were from the Age, Gene/Environment Susceptibility (AGES)–Reykjavik Study, which is described in detail elsewhere.13 In brief, the AGES study constitutes visit 7 (in 2002-2005) of the Reykjavik Heart Study, which began with 18 000 residents of Reykjavik recruited in 1967. The AGES study recruited 5758 of the surviving members, who were aged 69 to 96 years in 2002. Of these, 638 provided a DNA sample in 1991 as part of the sixth study visit and, therefore, have DNA from 2 time points available for methylation analysis.
The 111 analyzed herein represent a sample of 61 individuals from the 638 with the largest amount of DNA in the study repository as well as an additional 50 chosen to represent surviving (all-cause) cancer cases within the cohort with the largest amount of DNA in the repository. These samples were 50% male, with an average age at first sampling of 74.6 years (SD, 2.9 years; range, 70-82 years) and an average time between sampling of 11 years.
These 111 were not statistically significantly different from the 638 samples with 2 DNA visits regarding cholesterol, triglycerides, and C-reactive protein levels; blood pressure; smoking; coronary heart disease; diabetes; or stroke prevalence as of the last visit. They were statistically significantly older by 0.6 years; however, this rounds to the same age (84 years in both groups at the second visit). The intra-individual change in methylation over time was not statistically significantly different between samples from (all-cause) cancer and noncancer samples.
Utah samples were from the Salt Lake City CEPH pedigrees collected between 1982 and 1984 (680 individuals from 48 three-generation families) as previously described.12 Because these were family sets, there was a broad range of ages (5-72 years), with the time between sampling an average of 16 years. All families were recontacted and 25 agreed to participate in a second sample collection, of whom 21 had sufficient DNA at both time points for more than 1 family member to be included in our analysis. Disease status was not a consideration in selection of individuals or families for analysis of longitudinal methylation changes. Neither sex ratio nor family size were significantly different (t test = 0.30, P = .77; t test = 0.15, P = .89, respectively) between the collection of families analyzed in this report (mean of 7.1 female and 7.5 male participants per family) and the families from whom collection of second blood DNA samples had not yet been obtained (mean of 7.2 female and 7.5 male participants per family).
All DNA was from unfractionated peripheral blood cells (nontransformed cells). Institutional written informed consent was obtained from all participants. This methylation study was reviewed and approved by the appropriate institutional review boards at the University of Utah, the Icelandic National Bioethics Committee, and the Johns Hopkins Bloomberg School of Public Health. All Utah study participants provided informed consent under University of Utah institutional review board approved protocol number 6090-96. Participant recruitment in the AGES cohort (Iceland) and sample sharing for this project were approved by the Icelandic National Bioethics Committee (FS-04-001). For cell fractionation, Dynabeads (Invitrogen, Carlsbad, California) were used on buffy coats isolated using Ficoll-Paque Plus (GE-Healthcare, Piscataway, New Jersey).
The luminometric methylation assay (LUMA) protocol has been described in detail previously.14 We modified the protocol to minimize effects of degradation on quantification by adding additional measurements for free DNA ends. (Details are available from the authors on request.) We performed a mixing study using predetermined proportions of either fully methylated (SssI methylase) or unmethylated λ phage DNA and measured the proportions by LUMA, using this standard curve to convert HpaII/MspI ratios into HpaII methylation. The assay was linear at the range of 0% to 100% methylation (R2 = 0.984; R = 0.992). Furthermore, the assay was validated by demonstrating the marked hypomethylation found in the previously described DNA methyltransferase I double knockout cell line compared with the parent cell line (HCT116).
Experiments with 3 separate digestions and measurements on 25 samples revealed that the average variance of the assay was 2%. Measurements of HpaII methylation of whole blood from 7 individuals sampled 2 to 4 times over 30 days confirmed the stability of methylation in both total buffy coat white blood cell DNA as well as fractionated T cells, the predominant cell population, showing no significant change in methylation. Furthermore, repeat assays performed 1 year later on the original samples from the 9 individuals from the Icelandic cohort who showed the greatest difference in methylation between time points confirmed the initial measurements.
DNA (0.5 μg) was bisulfite-treated with the EZ DNA methylation kit (Zymo Research, Orange, California). DNA methylation analysis of individual genes was performed using the commercially available Illumina GoldenGate Methylation Solution,15 using the current Cancer Panel I platform (Illumina, San Diego, California), which probes 1505 CpG loci selected from 807 genes. The Illumina assay has been validated by both bisulfite sequencing and methyl-specific polymerase chain reaction15 as well as by quantitative bisulfite pyrosequencing.16
To assess the statistical significance of intra-individual changes in methylation over time in the Iceland sample, we performed permutations of the time labels to generated random draws from the null hypothesis of no time effects. There were 3 measurements at time 1 and 3 measurements at time 2 for each individual. When there is no true change over time, these 6 values should be estimates of the same underlying value and simply reflect random and measurement error. Therefore, at each permutation, we shuffled these 6 measurements within individuals and randomly assigned 3 to time 1 and the other 3 to time 2. We then averaged the 3 values for each permuted time point and then calculated the difference between time 2 and time 1.
To estimate how likely our observation is to be due to chance if there are no true changes in methylation over time, we also calculated the ratio of variance in methylation across all 6 measures over the variance within each time point (R = Varbetween/Varwithin). We compared the ratio in the observed data (R = 11.23) with the distribution of R values from each of 10 000 permutations. None of the 10 000 showed an R as extreme as that observed (P<.001). These analyses were performed using SAS version 9.1 (SAS Institute Inc, Cary, North Carolina).
To estimate heritability of methylation change in the Utah family sample, we first calculated the change between time 2 and time 1, adjusted for time 1 values, to accommodate the influence of time 1 values on the amount of change possible. Residual values for each sample after regressing methylation values at time 2 onto those at time 1 were then used to calculate maximum likelihood estimates of heritability, using variance components models in the ASSOC program of the SAGE package, version 126.96.36.199P <.05 was considered statistically significant.
To ensure that readings from different samples were comparable for the Illumina gene-specific data set, we quantile-normalized the data in the following way: we combined the red/cy5 (methylated) and green/cy3 (unmethylated) intensities into 1 vector that should be proportional to the amount of DNA in the sample. Because these amounts should be the same, these vectors were quantile-normalized18 then separated back into individual intensities. Log ratios were then formed that should be proportional to the log proportion of methylated targets and that made these values symmetric and close to normally distributed across samples for a given probe. To assess the consistency in changes in methylation across time points for family 21, a t test was computed for the difference in log ratios for individuals. To assess the size of the change, the log ratios were back-transformed to percentages [red/(red + green)] as done by Illumina15 and computed the differences in percentages in methylation. To quantify changes, we found that it was important to look at differences in percentages instead of differences in log ratios because a change from, for example, 0.1% methylation to 1% methylation, while 10-fold, is not likely to be biologically meaningful.
The question of time-dependent epigenetic changes was examined directly by the luminometric methylation assay (LUMA), a global measure of HpaII/MspI methylation, which cleaves 5-′CCGG-3′ sites that are, respectively, dependent and independent of methylation of the internal cytosine. The higher this ratio, the more methylated is the DNA template. We first performed an analysis of the method, mapping all HpaII sites in the human genome, which showed 15-fold enrichment of CpG islands, potentially important regulatory sequences. We also modified LUMA to minimize the possibility of error due to DNA degradation.
We first analyzed DNA from 111 participants in the AGES Reykjavik Study.13 The sorted change values are shown in red in Figure 1. While the mean intra-individual difference between HpaII methylation over an average of 11 years was 0, a wide range of changes were observed (greatest loss = −0.30; greatest gain = 0.26), with 70 individuals (63%) showing a change of at least 0.05 in either direction, 33 individuals (30%) showing a change of 0.10 or more, and 9 (8.1%) showing a change of at least 0.20. Since roughly the same number of participants showed a decrease in methylation as an increase, this substantial intra-individual change would likely be missed by age-specific cross-sectional analysis.
To gauge whether the differences observed were due to measurement variation or chance, we performed permutations of the data to simulate no change over time (Figure 1). None of the 10 000 permuted data sets showed a distribution of change in methylation over time as extreme as the observed data (P < .001). In addition, the coefficient of variation for within-participant triplicate measures of LUMA was only 2.4% compared with a coefficient of variation across individuals of 10.2%.
Inflammatory markers, such as erythrocyte sedimentation rate, C-reactive protein level, and white blood cell count, were also available for these participants at both visits (measured by the Westergren method, Hitachi 912 [Hoffman-LaRoche, Basel, Switzerland], and Coulter Counter [Beckman Coulter, Fullerton, California], respectively). These measures showed no relationship to DNA methylation levels, defined as the residual after adjustment of the change in methylation by the time 1 value, to accommodate any dependency on time 1 values, indicating that the methylation changes were not due to an inflammatory state or redistribution of white blood cells. Furthermore, age and length of storage were not correlated with change in methylation. The length of time between measures was slightly correlated with change (in either direction, using the absolute value of the residual) (ρ = 0.14), although this was not statistically significant (P = .17). This trend may be expected if methylation is indeed changing over time in individuals.
To confirm these results, we examined DNA from a second cohort of 126 individuals from a collection of Utah pedigrees that had been sampled twice over an average of 16 years. Like the Icelandic population, a wide range of changes were observed in this sample (min = −0.49; max = 0.39), with 50 individuals (40%) showing a change of at least 0.05, 23 (18%) showing a change of at least 0.10, and 13 (10%) showing a change of at least 0.20 between time points.
An additional advantage of the Utah cohort is the inclusion of families, allowing estimation of familial correlations in methylation change over time. Many showed clustering of methylation change in most or all family members (Figure 2). This clustering occurred for both decreased and increased methylation. In general, the familial correlations in methylation were more striking at time 2 compared with time 1, indicating that the differences were not due to acquired instability of the DNA due to longer storage, which was also demonstrated directly. While shared family environment could explain this clustering, most families contained 2 generations of adults (ie, the average ages of offspring sampled at times 1 and 2 were 17 and 32 years, respectively), who likely did not share households during the majority of the time between samplings. This suggests that the stringency of global methylation pattern maintenance is itself a heritable trait. Such familiality was most striking for decreased methylation, since 7 of the 13 most extreme decreases in methylation over time were within 2 families, families 21 and 9 (Figure 2). To assess the significance of this clustering, we calculated the heritability of methylation change based on these family data. To account for any correlation between time 2 and time 1 LUMA values, the change in methylation was adjusted by the time 1 value and each participant's residual value was used as the phenotype for heritability analysis. The heritability estimate was 0.99 (P < .001). This familiality was not limited to a single family and remained statistically significant (h2 = 0.743; P = .003) after removal of family 21. This suggests that although family 21 is clearly an outlier in the amount of change and the tight clustering among family members, the rest of the families show clustering within vs across families, supporting a heritable component to methylation stability.
To gain insight into gene-specific methylation changes, we examined a panel of 1505 CpG dinucleotides in 807 genes (approximately 3% of known human genes) using the recently released Golden Gate methylation assay15 (cancer panel 1). We analyzed a subset of 41 individuals at each of 2 time points, representing 17, 5, and 19 individuals showing the greatest loss, least change, or greatest gain, respectively, in global DNA methylation as measured by LUMA. When individuals were analyzed based on the difference in methylation over time (distance computed based on all genes tested), there was tight clustering of family 21 (P < .001), which also showed the greatest change by LUMA. Furthermore, of the 50 CpG probes that showed the greatest change over time in the 5 members of family 21, 49 showed methylation loss with age (P < .001; 25 expected by chance) (Table 1). There was a small but statistically significant enrichment for CpG probes within imprinted genes in this subset of genes (5/50 compared with 28/807 on the array; P < .047). Furthermore, comparing the 50 CpGs with the greatest difference across all individuals (not in family 21), 13 were shared with family 21, which was statistically significant (P < .001) (Table 2). There were a number of immunological mediators among these genes (Table 2), which is intriguing given the suggested role of altered DNA methylation in immunological disease.19,20 Furthermore, among the genes that changed the most, family 21 had more genes in common with the most extreme outliers from families 3 and 9 than would be expected by chance alone.
In this study, we observed time-dependent changes in global DNA methylation within the same individual in 2 separate populations in widely separated geographic locations, with 8% to 10% of individuals in both populations showing changes greater than 20% over an 11- to 16-year span. These changes showed familial clustering of both increased and decreased methylation and were most marked (>30%) in a family with 5 individuals showing loss of methylation over time, in whom methylation alterations were confirmed by examination of approximately 1500 CpG dinucleotides in 807 arbitrarily selected genes. The enrichment for imprinted genes was intriguing given the sensitivity of imprinted genes to both in vitro fertilization in humans21 and dietary modification in mice.22
These data support the idea of age-related loss of normal epigenetic patterns as a mechanism for late onset of common human diseases (common disease genetic and epigenetic model),1 which could arise through the loss of functionally important epigenetic modifications as well as through the release of epigenetic buffering of intrinsic genetic variation.23,24 In that regard, it is particularly interesting that many of the genes showing common variation are involved in immune system modulation and, thus, might reflect temporally acquired changes in the cell type that was studied (lymphocytes). However, lymphoid tissues might also act as a good surrogate tissue for changes in other target tissues, as for loss of imprinting of IGF2, one of the best-studied epimutations, the defect is found in both lymphocytes as well as colon and changes of either are associated with increased colorectal cancer risk.25
The familial clustering of methylation changes also raises the possibility that methylation stability might be directly related to genetic variation, such as in genes controlling 1-carbon metabolism or DNA methyltransferase activity. Consistent with this idea, gene-environment interactions affecting folate biosynthesis are linked to risk of colorectal neoplasia.26 The mechanism could involve altered methylation of specific genes, such as that leading to loss of imprinting of IGF2 associated with colorectal cancer risk.27
Both losses and gains of DNA methylation were observed over time in different individuals, and both could contribute to disease, which subsequent studies will need to determine. For example, cancer is associated with both hypomethylation and hypermethylation through activation of oncogenes and silencing of tumor suppressor genes, respectively.28 Similarly, animal studies have shown that a loss of DNA methylation increases intestinal adenoma initiation and a gain of DNA methylation increases adenoma progression.29 Similarly, both hypomethylation and hypermethylation could lead to autoimmune disease by activating autoreactivity genes or silencing histocompatibility genes.19,30 Our data stand in contrast to the observation of Eckhardt et al11 that there are no changes in DNA methylation over time. In that study, values were averaged across individuals for a given age group, while our data suggest considerable interindividual age variation, with differences accruing over time within individuals that would be missed by group averaging.
Finally, the implications of these results are potentially profound for population-based studies of human disease. The epigenome changes in individuals over time, which might directly influence disease phenotype. Epigenetic changes also might reflect age-related or environmental exposures. Thus, including epigenetic measurements in epidemiological studies could open a molecular window into potential genome exposures as well as mechanisms.
Corresponding Authors: Vilmundur Gudnason, MD, PhD, Hjartavernd, Holtasmari 1, 201 Kopavogur, Iceland (email@example.com); Andrew P. Feinberg, MD, MPH, 1064 Ross Research Bldg, 720 Rutland Ave, Baltimore, MD 21209 (firstname.lastname@example.org).
Author Contributions: Drs Bjornsson, Sigurdsson, and Fallin contributed equally to this work. Drs Bjornsson, Feinberg, Fallin, Irizarry, Aspelund, Gudnason, and Feinberg had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Bjornsson, Fallin, Leppert, Sapienza, Gudnason, Feinberg.
Acquisition of data: Bjornsson, Sigurdsson, Cui, Yu, Rongione, Ekström, Harris, Launer, Eiriksdottir, Leppert, Sapienza, Gudnason.
Analysis and interpretation of data: Bjornsson, Sigurdsson, Fallin, Irizarry, Aspelund, Rongione, Sapienza, Gudnason, Feinberg.
Drafting of the manuscript: Bjornsson, Fallin, Aspelund, Feinberg.
Critical revision of the manuscript for important intellectual content: Bjornsson, Sigurdsson, Fallin, Irizarry, Aspelund, Cui, Yu, Rongione, Ekström, Harris, Launer, Eiriksdottir, Leppert, Sapienza, Gudnason, Feinberg.
Statistical analysis: Bjornsson, Fallin, Irizarry, Aspelund, Sapienza.
Obtained funding: Fallin, Launer, Sapienza, Gudnason, Feinberg.
Administrative, technical, or material support: Sigurdsson, Fallin, Cui, Yu, Rongione, Ekström, Harris, Eiriksdottir, Leppert, Sapienza, Gudnason, Feinberg.
Study supervision: Bjornsson, Fallin, Feinberg.
Financial Disclosures: None reported.
Funding/Support: This work was supported by National Institutes of Health (NIH) grants P50-HG003233 (Dr Feinberg) and R01-ES015211 (Dr Fallin), the Swedish Cancer Foundation (Dr Ekström), NIH contract N01-AG-12100 (Drs Harris, Launer, and Gudnason), the Icelandic Parliament (Dr Gudnason), Huntsman General Clinical Research Center grant M01-RR00064 (Dr Leppert), and the W. M. Keck Foundation and George S. and Delores Doré Eccles Foundation (Dr Leppert). Dr Bjornsson was supported by the Fulbright Foundation and Dr Sigurdsson was supported by a 2006 grant from the Icelandic Student Innovation Fund.
Role of the Sponsor: Drs Harris and Launer from the NIH helped design the AGES cohort, but otherwise, the study's funders had no role in the design and conduct of the study, in the collection, analysis, and interpretation of the data, or in the preparation, review, or approval of the manuscript.
Additional Contributions: We thank Andreas Peiffer, MD, and Melissa M. Dixon, MS, University of Utah, for technical assistance. No compensation was received.