What a Methylation Report Tells You

Oak
methylation epigenetics epigenetic-clocks mthfr bioinformatics
A glowing living tree trunk covered in shifting points of light stands beside a dark unlit stone pillar in a night forest.

If you search for a methylation report you will find two unrelated products under one name. The first reads a handful of SNPs (MTHFR C677T, MTRR A66G, COMT V158M) out of a 23andMe or AncestryDNA file and colors them red, yellow, green. That is a genotype lookup dressed up as a physiology report, and the colors are editorial. The second measures actual 5-methylcytosine at hundreds of thousands of CpG sites in your DNA and reports epigenetic clocks and cell composition. That one is a measurement. If you want to know something about your current state rather than your invariant germline, you want the second, and you want the raw IDAT files that come with it.

The SNP panel version, and why we don’t run it

The genetic “methylation pathway” report takes variants in folate and one-carbon metabolism genes and assigns each a status. The underlying genotype call is usually fine. The interpretation layered on top is where it falls apart.

Three problems. First, coverage: consumer arrays genotype maybe 600k–900k positions, and the specific SNP a report wants is often absent, so the vendor imputes or reports “not detected” as if it were informative. Second, effect size: MTHFR C677T homozygosity reduces enzyme activity in vitro, but in populations with adequate folate the downstream effect on plasma homocysteine is modest and highly dependent on folate status. Third, and this is the one that matters, the report is static. Your C677T genotype at 45 is the same as it was at 5. A report that cannot change cannot tell you whether anything you did worked.

If you care about one-carbon metabolism, the direct measurements are serum homocysteine, serum folate, vitamin B12 or holotranscobalamin, and methylmalonic acid. Those are blood tests that move. Elevated homocysteine with a known C677T genotype is a conversation with a physician, not a supplement search. We will not tell you what to take, and neither should a web report.

What the epigenetic version measures

DNA methylation at CpG dinucleotides changes with age, cell type, and exposure, and a large fraction of the age signal is reproducible enough to build predictors on 1. Horvath’s 2013 multi-tissue clock used 353 CpGs selected by elastic net and predicted chronological age with a median absolute error near 3.6 years across many tissues 2. That launched the field. Since then the useful models have moved away from predicting chronological age (which your birth certificate already does for free) toward predicting mortality, morbidity, and rate of change.

The three families worth knowing:

  • Chronological-age clocks (Horvath, Hannum, and successors trained on much larger cohorts). Zhang and colleagues showed that training on ~13,000 samples produces markedly tighter chronological predictions, which also means less residual left over to interpret as “biological age” 3. Bernabeu and colleagues took this further, separating chronological and biological age prediction and improving both 4.
  • Phenotype/mortality clocks (PhenoAge, GrimAge). Trained against clinical biomarkers or time-to-death rather than birthdate. These carry most of the association with age-related disease outcomes 5.
  • Pace-of-aging measures. DunedinPACE was trained on the rate of change in 19 organ-system biomarkers across two decades in a birth cohort, and reports a rate rather than an age. Its test-retest reliability is high (ICC around 0.96), which is what you want for a longitudinal metric 6.

We would rank DunedinPACE and the mortality-trained clocks over Horvath age for anyone tracking themselves over time. A single Horvath number is close to uninterpretable at the individual level. A rate, measured three times over three years on the same platform with the same pipeline, is at least a trend you can reason about.

The files to demand

Whatever vendor runs the array, ask for the raw .idat files, both _Grn.idat and _Red.idat, plus the sample sheet with sentrix ID and position. A PDF with a number on it is not a result you can reprocess when the field’s models improve. If you are getting whole-genome bisulfite or EM-seq instead, ask for the FASTQs or at minimum the Bismark coverage files (.cov.gz, chromosome/start/end/percent-methylated/count-methylated/count-unmethylated).

Note that ordinary short-read WGS gives you no methylation information at all. Bisulfite conversion destroys it unless you do it deliberately. Nanopore sequencing does carry 5mC natively, and modkit pileup on a dorado-basecalled BAM gives you per-CpG methylation genome-wide, which is the most information-dense option if you have it.

Processing IDATs yourself

We use sesame in R over minfi, mainly for its masking of problematic probes and its handling of EPIC v2.

library(sesame)
sesameDataCache()

# EPICv2: preprocessing code includes the T (probe type) step
betas <- openSesame("idats/", prep = "TQCDPB", func = getBetas)
qcs   <- openSesame("idats/", prep = "", func = sesameQC_calcStats)

Check before you interpret anything: fraction of probes with detection p > 0.05 (we drop samples above ~5% failure), mean intensity, predicted sex against known sex, and the control probe metrics. Then estimate cell composition, because whole blood methylation is dominated by the proportion of neutrophils, monocytes, and lymphocytes in the tube. EpiDISH with the centDHSbloodDMC reference gives you fractions in a line, and if you skip this step you will interpret an infection or a hard training week as epigenetic aging.

Two known traps. Preprocessing choice changes clock output: McEwen and colleagues showed systematic differences in estimated methylation age across common normalization methods on EPIC data, which means a between-visit comparison is only valid if you reprocess both visits identically 7. And EPIC v2 dropped and renamed probes relative to EPIC v1, so clocks built on 450k or EPIC v1 CpG sets now have missing inputs. Garma and Quintela-Fandino quantified the overlap loss and the accuracy cost of the imputation strategies people use to fill it 8. Freeze your pipeline, record the array version, and never compare a v1 result to a v2 result without re-running both.

How to read the number you get back

Treat the output as a noisy estimate of a slow-moving quantity. Individual clock estimates carry measurement error of a year or more, and the biological signal you are looking for over 12 months is smaller than that. This is why single-timepoint “your biological age is 39” reports are close to entertainment, and why principal-component and large-cohort reformulations of these clocks exist at all: to cut technical noise so real change is visible 43.

The state of evidence on intervention: methylation age has been used as an endpoint in small self-experiment-style trials, and the results there are hypothesis-generating at best, with tiny n and no control arm 9. Methylation changes with age across many loci and tissues, and the direction and magnitude differ by region, so a single summary number compresses away most of the biology 10. If a report claims a protocol lowered your epigenetic age by five years, ask how many probes were imputed and what the test-retest interval was.

Questions people also ask

Is it worth getting a methylation test? The SNP panel, no. The array, yes, if you will run it more than once on the same platform and keep the IDATs. One measurement gives you a number with wide error bars. Three measurements over several years give you a slope, and slope is where the measures with good reliability like DunedinPACE earn their keep 6.

How do you tell if you have methylation issues? Not from a genotype report. The functional readouts are blood: homocysteine, folate, B12, methylmalonic acid. Abnormal values there need a clinician to work up, because the causes range from dietary intake to absorption problems to kidney function.

Does insurance pay for a methylation test? Generally no for epigenetic clocks, which are not clinical tests. Homocysteine and B12 are ordinary clinical labs and are often covered when there is a documented reason to order them.

Why don’t doctors test for MTHFR? Because for most people the genotype does not change management. Several professional bodies have recommended against routine MTHFR testing in workups where it was once common. Measuring homocysteine directly answers the question the genotype only gestures at.

What is the best vitamin for methylation? We don’t make supplement recommendations. What we will say is that if you take anything that affects folate or B12 status, measure homocysteine and B12 before and after, because otherwise you have no readout at all. Discuss the results with a physician, particularly if B12 is low, since supplementing folate can mask a B12 deficiency.

How much does a methylation test cost? Consumer EPIC array panels run a few hundred dollars. Clinical homocysteine and B12 are tens of dollars each. The SNP-based reports are often $20–50, which is roughly what the information is worth.

Oak builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

  1. Meaghan J. Jones, Sarah J. Goodman, Michael S. Kobor. DNA methylation and healthy human aging. Aging Cell, 2015. https://doi.org/10.1111/acel.12349 ↩

  2. Steve Horvath. DNA methylation age of human tissues and cell types. Genome Biology, 2013. https://doi.org/10.1186/gb-2013-14-10-r115 ↩

  3. Qian Zhang, Costanza L. Vallerga, Rosie M. Walker, et al. Improved precision of epigenetic clock estimates across tissues and its implication for biological ageing. Genome Medicine, 2019. https://doi.org/10.1186/s13073-019-0667-1 ↩ ↩2

  4. Elena Bernabeu, Daniel L. McCartney, Danni A. Gadd, et al. Refining epigenetic prediction of chronological and biological age. Genome Medicine, 2023. https://doi.org/10.1186/s13073-023-01161-y ↩ ↩2

  5. Yasmeen Salameh, Yosra Bejaoui, Nady El Hajj. DNA Methylation Biomarkers in Aging and Age-Related Diseases. Frontiers in Genetics, 2020. https://doi.org/10.3389/fgene.2020.00171 ↩

  6. Daniel W Belsky, Avshalom Caspi, David L Corcoran, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife, 2022. https://doi.org/10.7554/elife.73420 ↩ ↩2

  7. Lisa M McEwen, Meaghan J Jones, David Tse Shen Lin, et al. Systematic evaluation of DNA methylation age estimation with common preprocessing methods and the Infinium MethylationEPIC BeadChip array. Clinical Epigenetics, 2018. https://doi.org/10.1186/s13148-018-0556-2 ↩

  8. Leonardo D. Garma, Miguel Quintela-Fandino. Applicability of epigenetic age models to next-generation methylation arrays. Genome Medicine, 2024. https://doi.org/10.1186/s13073-024-01387-4 ↩

  9. Josh Mitteldorf. A Clinical Trial Using Methylation Age to Evaluate Current Antiaging Practices. Rejuvenation Research, 2019. https://doi.org/10.1089/rej.2018.2083 ↩

  10. Adiv A. Johnson, Kemal Akman, Stuart R.G. Calimport, et al. The Role of DNA Methylation in Aging, Rejuvenation, and Age-Related Disease. Rejuvenation Research, 2012. https://doi.org/10.1089/rej.2012.1324 ↩