Saliva and blood samples in epigenetic age research

Prepared by: Research Desk. Sources are linked for verification.

Data point first: In a within-person study of 284 tissue samples from 83 individuals, common epigenetic clocks returned significantly different estimates from oral samples (cheek swab or saliva) than from blood samples taken from the same people [5]. Sample type is not a neutral detail. It is part of the measurement.

This matters because the same researchers note growing interest in estimating epigenetic age from less-invasive oral tissues, buccal cells or saliva, in both research and commercial settings, while clocks have been trained almost exclusively on blood-based tissues [5]. This explainer covers why tissue type shifts clock output, what the cross-tissue and forensic studies report, and which criteria matter when reading a result by sample type. For the general mechanics of clocks, see our biological age test guide.

Why the sample type changes the answer

An epigenetic clock is a statistical model. It reads DNA methylation at a set of CpG sites and converts the pattern into an age estimate. The model only "knows" the relationships it saw in its training data.

Different cell types have different methylation patterns. Apsley and colleagues note that differentiated cell types across body tissues have unique DNA methylation landscapes and unique age-related changes. Applying a blood-trained clock to an oral sample may therefore introduce bias [5].

Blood samples also vary internally. Changes in the mix of white blood cell types can move clock estimates, a topic covered in epigenetic clocks and blood cell composition. Saliva and buccal samples add another layer, because they contain a different set of cells from blood.

Clocks designed to work across tissues

Some clocks were deliberately built to span tissues.

• The 2013 multi-tissue clock. Horvath's original estimator was developed to estimate DNA methylation age across many human tissues and cell types [1]. Its design and history are covered in The Horvath clock.

• The skin and blood clock (2018). Horvath and colleagues reported that earlier DNA methylation biomarkers had sub-optimal accuracy in fibroblasts and other cell types used in laboratory studies. They built a new estimator based on 391 CpGs for fibroblasts, keratinocytes, buccal cells, endothelial cells, lymphoblastoid cells, skin, blood and saliva samples [6].

• Mixed-tissue deep learning. ResnetAge accepts 22,278 CpG sites as input and supports both 27K and 450K array frameworks. It was trained on 32 public datasets that included whole blood, saliva and mouth tissue. The authors reported a training-set mean absolute error of 1.29 years [7]. A training-set error describes fit to the data the model learned from, not accuracy on new samples. For background on the arrays involved, see EPIC and 450K methylation arrays.

• Buccal-specific models. Researchers have also built aging clocks and classifiers trained directly on buccal methylation data [8]. That approach avoids carrying a blood-trained model over to cheek cells.

The practical point: an epigenetic age from saliva is only as comparable to a blood result as the model's training allows.

What the cross-tissue study found

Among the studies cited here, the Apsley study is the most direct test of this question. The researchers examined common clocks within the same people across five tissue types:

Overall, there were significant within-person differences between oral and blood estimates [5]. The study is a preprint (bioRxiv), so it had not completed journal peer assessment when indexed.

Evidence from forensic age estimation

Forensic science offers a second line of evidence. Labs there estimate the age of the person who left a biological stain, often using small targeted CpG panels rather than whole arrays.

• Accuracy by body fluid. Sutter and colleagues tested published tools based on minisequencing and massively parallel sequencing on blood, saliva, buccal cells and semen. Mean absolute errors ranged from three to five years for blood, saliva and buccal cells; semen showed seven to eight years. Biological replicates from ten individuals (blood, saliva, buccal) or 12 individuals (semen) were used to assess reproducibility [9].

• Tissue-specific CpG combinations. Pfeifer and colleagues tested blood and mouth-swab models built on CpG sites in four genes (ASPA, EDARADD, PDE4-C and ELOVL2), with a different CpG combination for each tissue. Applied in a different laboratory, mean absolute differences were 9.84 years for the blood model and 8.32 years for the buccal model, significantly worse than the published figures. Retraining the models in the new lab improved accuracy [10].

• Instrument effects. So and colleagues reported that different genetic analyzers with different software settings can generate different methylation values, leading to significant errors in age prediction. They tested four correction strategies, including adjustment equations for blood and saliva markers such as ELOVL2, FHL2, KLF14, MIR29B2CHG/C1orf132 and TRIM59 [11].

Together, these studies show that tissue, chemistry, instrument and laboratory each contribute to the final number. Repeat-test variability is covered in biological age test accuracy.

Is saliva good enough for methylation work?

Saliva can yield usable DNA. In a small feasibility study of 12 adults with hepatocellular carcinoma and 12 matched controls, saliva samples were analysed for methylation at 1,359 CpG sites. DNA purity was reported as good, with a mean OD260/280 ratio of 1.78 [12]. That study concerned cancer markers, not aging clocks, but it shows saliva can support methylation profiling.

The open question is calibration rather than feasibility. A saliva sample can be measured well and still be scored by a model tuned on a different tissue.

How these tests are regulated

Laboratory-developed tests are a regulatory category that FDA describes on its LDT page [2]. On 19 September 2025, FDA published a Federal Register notice implementing the vacatur of its rule on laboratory-developed tests [3]. Separately, CMS administers CLIA laboratory quality standards [4]. CLIA concerns laboratory operations rather than whether a clock measures aging. Our page on whether biological age tests are FDA approved explains the details.

Questions to ask about any sample-based result

The research above translates into neutral criteria for reading saliva, cheek-swab and blood-spot results:

1. Clock disclosed, with its training tissue. A clock trained on saliva or multiple tissues is a different instrument from a blood-only clock applied to saliva [5][6].

2. Sample tissue stated clearly. Saliva and buccal cells are distinct sample types and were analysed separately in cross-tissue work [5].

3. Like-for-like repeats. Retesting with the same tissue, lab and platform reduces the risk of mixing tissue, cross-lab and instrument shifts with any real change [5][10][11].

4. Laboratory status stated. Look for a named laboratory and its CLIA certification status [4].

5. Price transparency. Kit price, repeat-test pricing and any subscription terms shown before checkout.

6. Data privacy. Clear terms on how long the sample and methylation data are kept, whether they are used in research, and how deletion is requested.

Further reading on ageing biology

Readers who want more depth than a single explainer can turn to books on ageing biology and epigenetics. A few neutral criteria separate a useful book from a promotional one:

• Publication date. Cross-tissue clock work is recent; the studies above date from 2013 to 2025, so a newer edition is more likely to cover saliva and buccal models.

• Primary citations. Look for endnotes that point to named, peer-indexed studies rather than anecdotes.

• Measurement focus. The most useful titles explain how clocks are trained and validated, including tissue and laboratory effects, not just what a number "means".

• Author background. Prefer authors with a track record in genomics, epidemiology or statistics.

• Format and price. Print, e-book and audio editions differ in cost and in how easily tables and figures can be read.

See our about page for how the site works and how it makes money.

Frequently asked questions

Does saliva give the same epigenetic age as blood?

Not necessarily. A within-person study found significant differences between oral-sample and blood-sample estimates from common clocks [5]. Saliva and blood contain different cell types with distinct methylation patterns.

Are cheek swabs and saliva the same thing?

No. Buccal swabs collect cheek epithelial cells, while saliva is a separate sample type. The Apsley study analysed them as distinct tissues [5], and some models are trained specifically on buccal data [8].

Why would the same sample score differently in two labs?

Targeted assays are sensitive to instruments, software settings and lab conditions. A forensic study reported cross-lab errors of 8.32 to 9.84 years before retraining [10], and another reported instrument-specific methylation shifts [11].

Are any clocks designed for saliva?

Yes. The skin and blood clock was built to include buccal cells and saliva [6], and ResnetAge was trained on datasets that included saliva and mouth tissue [7].

Related on this site

• How old are you, biologically?

• Biological age tests: how epigenetic clocks work and what they mean

• Are biological age tests FDA approved? How aging tests are regulated

• Epigenetic clocks and blood cell composition: why cell mix matters

• The Horvath clock: the 2013 multi-tissue epigenetic age estimator

• Biological age test accuracy: reliability, noise and repeat results

Last updated: 26 October 2026.

Not medical advice. This page reports published research and regulatory records for information only.

Research-only notice: Content on this site is provided for research and educational purposes only and is not intended for human use. It is not medical advice.

Translation notice: Some content on this site may have been machine-translated using Google Translate and may contain errors.