Which methylation test should you buy?
If you want to know your MTHFR genotype, you do not need a methylation test at all: those are two or three single-nucleotide positions that are already present in any whole-genome sequence or consumer genotyping array you own. If you want to know how your DNA is methylated, you need an assay that reads 5-methylcytosine directly, which means a bisulfite or enzymatic conversion followed by an array or sequencing. Our recommendation for most technically fluent readers is an Illumina Infinium MethylationEPIC v2.0 array run on blood, purchased from a vendor that returns the raw .idat files. It costs a few hundred dollars in reagents and gives you roughly 935,000 CpG measurements you can analyze with open-source tools for the rest of your life.
The two products that share one name
The “genetic methylation test” sold by most consumer companies is a genotype panel. It reports variants in genes whose protein products participate in one-carbon metabolism.
- MTHFR
- MTR
- MTRR
- BHMT
- CBS
- COMT
- AHCY
It tells you nothing about methylation state. It tells you which alleles you inherited, which is a fixed property of your genome and never changes.
The other product measures DNA methylation itself, the presence of a methyl group on cytosine in a CpG dinucleotide. This varies by tissue and by cell type. It also varies by age and to some extent by exposure. That is the measurement behind epigenetic clocks and behind essentially all of the published literature on methylation and aging.
If a vendor advertises “methylation” and its deliverable is a list of gene names and risk alleles, you are buying the first product. Price it accordingly.
If you want MTHFR genotype, query your existing data
The two commonly reported MTHFR variants are rs1801133 (677C>T, p.Ala222Val) and rs1801131 (1298A>C). Both are on chromosome 1 in the standard consumer arrays and in any 30x whole-genome VCF. Extract them directly:
bcftools view -r chr1:11790000-11810000 sample.vcf.gz | \
bcftools query -f '%ID\t%POS\t%REF\t%ALT[\t%GT\t%DP]\n' | \
grep -E 'rs1801133|rs1801131'
Check the depth field. Below roughly 10x, a heterozygous call at a single site is unreliable, and this is exactly the situation where a consumer report can hand you a confident-sounding wrong answer. If the position is not in your VCF, confirm it was covered rather than assuming homozygous reference: a missing call and a reference call look identical in a filtered VCF. Interpretation of these variants in the context of folate status or homocysteine is a clinical question and belongs with a physician. A report generator cannot answer it.
If you want measured methylation, insist on IDATs
The EPIC v2.0 array is the practical default. It interrogates about 935,000 CpG sites out of roughly 28 million in the human genome. The coverage is weighted toward promoters, enhancers, and the CpG islands used by published clocks. A single sample of whole blood is extracted to DNA, bisulfite-converted, and hybridized. That produces two .idat files per sample, one for the green channel and one for the red. Those files are the artifact that matters. Beta values, clock ages, and PDF reports are all derived and all reproducible from the IDATs.
Ask the vendor three questions before paying. Which array version, EPIC v1 or v2? EPIC v2 dropped and renamed a meaningful number of v1 probes, so a clock trained on 450k or EPIC v1 probes needs remapping. Will you receive the raw IDATs and the sample sheet? Do they run technical replicates? Without a replicate you have no estimate of measurement noise for your own sample.
Sequencing is the alternative. Enzymatic methyl-seq (EM-seq) avoids the DNA damage caused by sodium bisulfite and gives better library complexity at equal input. Whole-genome EM-seq at 30x covers all 28 million CpGs, but costs several-fold more than an array and has higher per-site variance. Reduced-representation bisulfite sequencing (RRBS) is cheaper and concentrates reads on CpG islands, at the cost of poor overlap with the specific probes that clocks expect. For clock calculation and for comparability with published cohorts, the array wins. For discovery work across non-island regions, take the sequencing.
The pipeline we would run
Start in R with SeSAMe, which handles EPIC v2 manifests and masks problematic probes by default:
library(sesame)
sesameDataCache()
betas <- openSesame("idat_dir", prep = "QCDPB", func = getBetas)
The prep string controls the preprocessing chain. It covers probe quality masking and dye bias correction, then detection p-value filtering and background subtraction with noob. Drop probes with detection p above 0.05, and drop cross-reactive and SNP-overlapping probes. The alternative stack is minfi with preprocessFunnorm(), which is better documented and worse at EPIC v2 manifest handling.
Two things are non-negotiable after preprocessing. First, estimate cell composition. Whole blood is a mixture, and a shift in neutrophil-to-lymphocyte ratio will move hundreds of CpGs and move your clock estimate with them. Use a reference-based deconvolution such as estimateCellCounts2() in FlowSorted.Blood.EPIC and report the fractions alongside any age estimate. Second, use M-values for any statistical testing. The M-value is the logit of beta, and beta values are heteroscedastic at the extremes.
Then compute clocks with the methylclock package or the original coefficient tables. Run several: Horvath multi-tissue, Hannum, PhenoAge, GrimAge where you can get access, and a pace-of-aging estimator. Disagreement between them is informative, because they were trained on different outcomes.
What a clock result does and does not tell you
Epigenetic age estimates are real measurements with real limits, and the limits are mostly about tissue. The multi-tissue clocks are not uniformly accurate across tissues. A clock recalibrated on human cortex substantially outperforms the pan-tissue clock in brain and reduces the error that comes from applying blood-trained weights elsewhere 1. The cerebellum is an extreme case: it appears epigenetically younger than other tissues in the same donor by a large margin 2. Healthy breast tissue runs older than blood from the same women 3. A blood measurement is a blood measurement.
Newer models do better on the chronological target. Training on large cohorts with elastic net and improved feature selection pushes prediction of chronological age and of biological-age proxies beyond the first-generation clocks 4. That accuracy is a property of the population average. It does not guarantee anything about your single sample.
Acceleration signals do appear in disease states. Brain tissue in Huntington’s disease shows accelerated epigenetic aging together with broad methylation disruption 5, and a meta-analysis across brain datasets reports accelerated epigenetic aging in neurodegeneration 6. Distinct methylation profiles have been reported in women with diminished ovarian reserve 7. None of these findings make a clock a diagnostic test in an individual. If a result concerns you, that conversation belongs with a clinician who can order validated assays.
Change over time is the interesting part, and it is measurable. A dietary intervention has been shown to shift epigenetic age as assessed by the clock 8. The clock concept also transfers across species, with multi-tissue predictors built in mouse 9, which is why intervention studies can be run at all. The practical implication for you is that one array is a point estimate and two arrays a year apart are a trajectory.
A short comparison checklist
Rank any vendor on five things. Check whether you get raw IDATs, which array version they run, and whether they report cell composition. Then check whether they disclose the exact clock implementations and coefficient sources and whether they run a technical replicate. A vendor that scores on all five is worth several times one that returns a score alone. A genotype panel that reports MTHFR and calls itself a methylation test scores zero, because it never measured methylation.
Questions people also ask
Is an MTHFR test the same as a methylation test?
No. An MTHFR test reads your DNA sequence at one or two positions. A methylation test measures chemical modification of cytosine across hundreds of thousands of sites. They answer unrelated questions.
Which methylation array should I choose, EPIC v1 or EPIC v2?
Choose EPIC v2 for new sampling, since it is the version that will remain available and comparable to future runs. Keep in mind that some v1 probes were removed, so clock implementations need EPIC v2 aware manifests.
Can I get methylation data from my existing 23andMe or whole-genome file?
No. Standard genotyping and standard sequencing both destroy methylation information during library preparation or amplification. Reading methylation requires a dedicated conversion chemistry, or nanopore sequencing that calls base modifications directly.
How reproducible are epigenetic age estimates?
Single-CpG measurements on arrays carry meaningful technical noise, and first-generation clocks inherit it. Running technical replicates and using principal-component-based clock formulations both reduce the variance. Treat a single estimate as having an uncertainty of a few years.
Does methylation age change if I change my behavior?
It can move. A controlled dietary intervention shifted epigenetic age measured by a clock 8. Magnitudes are modest and individual response varies, so treat any personal change as a hypothesis to retest rather than a result.
Footnotes
-
Gemma L Shireby, Jonathan P Davies, Paul T Francis, et al. Recalibrating the epigenetic clock: implications for assessing biological age in the human cortex. Brain, 2020. https://doi.org/10.1093/brain/awaa334 ↩
-
Steve Horvath, Vei Mah, Ake T. Lu, et al. The cerebellum ages slowly according to the epigenetic clock. Aging, 2015. https://doi.org/10.18632/aging.100742 ↩
-
Mary E. Sehl, Jill E. Henry, Anna Maria Storniolo, et al. DNA methylation age is elevated in breast tissue of healthy women. Breast Cancer Research and Treatment, 2017. https://doi.org/10.1007/s10549-017-4218-4 ↩
-
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 ↩
-
Steve Horvath, Peter Langfelder, Seung Kwak, et al. Huntington’s disease accelerates epigenetic aging of human brain and disrupts DNA methylation levels. Aging, 2016. https://doi.org/10.18632/aging.101005 ↩
-
Camilla Pellegrini, Chiara Pirazzini, Claudia Sala, et al. A Meta-Analysis of Brain DNA Methylation Across Sex, Age, and Alzheimer’s Disease Points for Accelerated Epigenetic Aging in Neurodegeneration. Frontiers in Aging Neuroscience, 2021. https://doi.org/10.3389/fnagi.2021.639428 ↩
-
Kristina W. Olsen, Juan Castillo-Fernandez, Andrew Cho Chan, et al. Identification of a unique epigenetic profile in women with diminished ovarian reserve. Fertility and Sterility, 2021. https://doi.org/10.1016/j.fertnstert.2020.09.009 ↩
-
Chanachai Sae‐Lee, Sarah Corsi, Timothy M. Barrow, et al. Dietary Intervention Modifies DNA Methylation Age Assessed by the Epigenetic Clock. Molecular Nutrition & Food Research, 2018. https://doi.org/10.1002/mnfr.201800092 ↩ ↩2
-
BI Ageing Clock Team, Thomas M. Stubbs, Marc Jan Bonder, et al. Multi-tissue DNA methylation age predictor in mouse. Genome Biology, 2017. https://doi.org/10.1186/s13059-017-1203-5 ↩
Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.