🧬 Aging Wiki

skin-methylation-clocks

Properties1
aliasesskin clocks, skin epigenetic age, skin DNAmAge predictors, skin methylation clock MOC

8 min read

Skin methylation clocks

Class MOC consolidating the four DNA methylation aging predictors trained on epidermal tissue that are currently seeded as atomic pages in this wiki. All quantitative claims, primary-source footnotes, and methodology details live on those four child pages — follow the wikilinks. This page provides cross-clock comparison, lineage structure, methodological dimensions, and open-questions navigation.

Quick-reference comparison table

ClockPaper / yearArchitectureTraining nError metricTissue / samplingLineageAtomic page
Bormann 2016Bormann F et al., Aging Cell 2016SVM on full 450k probe space (no compact CpG list)n=108 white femalesLOOCV MAE <5.25 yr; Vandiver indep MAE 6.72 yrEpidermis — punch biopsy + suction blister (invasive)Beiersdorf-DKFZ (founding)bormann-epidermis-clock-2016
Qi 23k 2026Qi M et al., Dermatology and Therapy 2026Elastic-net; 173 CpGs from 23,845-probe panel (Infinium EPICv2)Extends Bormann n=108; multi-ethnic pilot n=17CV error 5.66 yr (glmnet MSE); MAE 4.88 yr (multi-ethnic validation)Epidermis — tape-strip (non-invasive)Beiersdorf-DKFZ (successor)qi-23k-epidermis-clock-2026
TapeLift 2026Rodríguez-Paredes M et al., npj Precision Oncology 2026Dual: elastic-net 157 CpGs + PC clock 5,021 CpGsn=71 train / 18 val / 19 indep test (Caucasian-only)Elastic-net RMSE 6.2 yr; PC clock RMSE 5.1 yr (indep test)Epidermis — adhesive tape-strip forehead (non-invasive)DKFZ / Rodríguez-Paredes (distinct subgroup)tapelift-clock-2026
MitraSolo / MitraCluster 2025Menendez Vazquez A et al., npj Aging 2025Two-clock framework: MitraSolo (CpG-based) + MitraCluster (region-based); architecture unconfirmed no-fulltext-accessn=462 enzymatic methyl-seq samplesMAE ~4 yr (abstract-reported); intra-individual variation <2 yrEpidermis — adhesive tape-strip (non-invasive); generalizes across anatomical sitesMitra Bio (independent)menendez-vazquez-2025-mitrasolo-mitracluster

Three independent research lineages

Understanding which groups produced which clocks is critical for interpreting “independent validation” claims.

1. Beiersdorf-DKFZ lineage

The dominant lineage, comprising four publications from the same industrial-academic collaboration:

  • Bormann 2016 — founded the tissue-specific epidermis-clock field; SVM on 450k arrays; training cohort n=108 white females; established that pan-tissue clocks (Horvath 2013) underperform ~1.8-fold on epidermis.
  • Falckenhayn 2024 (Frontiers in Aging) — identified DHM as a natural DNMT1 inhibitor and showed Bormann-clock age reduction in keratinocyte cultures; provided mechanistic rationale for DHM-as-epigenetic-rejuvenator. Cross-linked via dihydromyricetin.
  • Qi 23k 2026 — elastic-net successor to Bormann; 173-CpG portable model; first cross-Fitzpatrick-validated skin clock; contains a single-arm open-label DHM serum trial (n=60) as the intervention arm.
  • Bienkowska 2026 (Clinical Epigenetics; n=851) — large cross-sectional landscape study using “a published skin-specific epigenetic clock” to characterize 37 accelerating/decelerating lifestyle and physiological factors.

Critically: Bienkowska 2026 is NOT independent validation of Qi 2026 — it is a same-group study from the same research program. The same applies to the Falckenhayn 2024 → Qi 2026 mechanistic chain. The only fully independent validation of any Bormann-lineage clock is the Vandiver 2015 dataset (n=18) applied in the original Bormann 2016 paper.

2. DKFZ / RodrĂ­guez-Paredes lineage (TapeLift)

The tapelift-clock-2026 paper (Rodríguez-Paredes et al. 2026, npj Precision Oncology) originates from DKFZ but the Rodríguez-Paredes subgroup operates independently of the Beiersdorf commercial arm. The paper’s primary framing is skin cancer epigenomics; the aging clock is a component within that broader program. The TapeLift clock’s two-variant architecture (compact elastic-net + PC clock) and its tumor-suppressor-gene mitotic-age angle distinguish it methodologically from the pure chronological-age-trained Bormann/Qi clocks.

Cross-ethnic status: The training cohort is Caucasian-only (Fitzpatrick I–IV). The “no non-white bias” attribution frequently applied to TapeLift in narrative summaries originates from Qi 2026’s independent application of the TapeLift clock to an n=17 multi-ethnic pilot — NOT from the Rodríguez-Paredes 2026 paper itself. See tapelift-clock-2026 § Limitations for the full correction.

3. Mitra Bio lineage (MitraSolo / MitraCluster)

menendez-vazquez-2025-mitrasolo-mitracluster is fully independent of both lineages above. Mitra Bio is a commercial entity; all authors except Mahdi Moqri (Stanford) are company employees. The distinguishing technical feature is enzymatic methyl-sequencing rather than Illumina microarray — a different sample-prep and bioinformatics pipeline that may tolerate lower DNA input. The n=462 training set is the largest reported for an epidermal methylation clock as of 2026-05-19. The <2 yr intra-individual variation is the best-characterized noise floor among published skin clocks; however, the full paper is not yet locally available and CpG counts, architecture, and cohort demographics remain unverified (#gap/no-fulltext-access — see menendez-vazquez-2025-mitrasolo-mitracluster).

Methodological dimensions

Tissue sampling invasiveness

The field has moved from invasive to non-invasive:

  • Invasive — Bormann 2016 used punch biopsy and suction blister roof (both require clinical procedure, local anaesthesia, or specialist equipment). Limits scalability to large cohorts and intervention trials.
  • Non-invasive — Qi 2026 (tape-strip), TapeLift 2026 (adhesive gel strip), and MitraSolo/MitraCluster 2025 (tape-strip) all demonstrate that routine adhesive tape collection provides sufficient DNA for clock profiling. This enables at-home or outpatient-scale collection.

Quantification platform

PlatformClocks using itDistinguishing feature
Illumina Infinium 450k arrayBormann 2016, Qi 23k 2026 (via 23,845-probe intersection)Comprehensive; standardized globally; used in Bormann training cohort
Illumina MethylationEPIC v2TapeLift 2026 (535,384 probes queried)Expanded probe space; newer array generation
Enzymatic methyl-sequencingMitraSolo / MitraCluster 2025Sequencing-based; lower DNA input tolerance; different bioinformatics

Interoperability note: The 23,845-probe panel used by Qi 2026 is the intersection of the Bormann 27k array and modern 450k/EPIC chips, making Qi 2026 partially backward-compatible with the Bormann probe space. TapeLift and MitraSolo use non-overlapping platforms; no head-to-head comparison on the same sample set has been published.

Predictor architecture

ArchitectureClock(s)Key property
SVM (full probe space)Bormann 2016No compact CpG list; requires full model parameters to apply; not portable as a coefficient table
Elastic-net (compact CpG list)Qi 23k 173-CpG; TapeLift 157-CpGPortable; publishable as coefficient table; widely used in clock development since Horvath 2013
PC clockTapeLift 5,021-CpG (Higgins-Chen 2022 PC architecture)Better test-retest reliability by regressing out technical noise; trades portability for robustness
Two-clock frameworkMitraSolo (CpG-level) + MitraCluster (region-level)Dual design addresses both single-sample precision (Solo) and bulk/region robustness (Cluster); architecture details pending full-text verification

Intervention responsiveness

Only one of the four skin clocks has a published study reporting a significant clock-age reduction in an interventional context with human participants:

  • Qi 23k 2026 — DHM topical serum, single-arm open-label, n=60, 8 weeks: epidermal DNAm age −2.08 yr (paired Wilcoxon p=0.029). No vehicle control; not a randomized trial. Partial-correlation analysis for sunscreen use is present in the paper. See qi-2026-dhm-epigenetic-skin-aging and dihydromyricetin for mechanistic detail.
  • MitraSolo / MitraCluster — claims a Yamanaka factor rejuvenation signal in vitro; abstract-only as of 2026-05-19; design details unverified. no-fulltext-access
  • Bormann 2016 and TapeLift 2026 — no published intervention-responsiveness studies as of 2026-05-19. needs-replication

Comparison with pan-tissue and blood clocks

The skin-specific clocks sit within the broader epigenetic clock landscape covered in biological-age-measurement. Key contextual points:

  • Pan-tissue clocks (Horvath 2013) underperform ~1.8-fold on epidermis relative to the tissue-specific Bormann clock. This established that tissue-specific training is necessary for accurate skin-clock performance — a principle now generalized to other tissues.
  • None of the skin methylation clocks has been tested in a pre-registered RCT with a hard endpoint; DunedinPACE (blood) remains the only clock that responded to a well-powered RCT (CALERIE-2). No skin-clock equivalent of the CALERIE-2 design has been run.
  • The Higgins-Chen 2022 PC-clock reliability critique (test-retest ICC degradation from technical noise) applies to all array-based clocks. Of the four skin clocks, only TapeLift has a PC-clock variant; the others have no published test-retest ICC data.

Open questions and gaps

  1. No head-to-head benchmark study. All four clocks have been run on distinct training and validation cohorts; no study has run all four on the same sample set. Without a common benchmark, cross-clock MAE comparisons in the table above are not directly comparable (different cohorts, different age distributions, different sample-collection methods).

  2. Intervention-responsiveness evidence is thin and lineage-concentrated. Only the Beiersdorf-DKFZ group has published an intervention study (DHM serum; Qi 2026). An independent group has not replicated the ~2 yr DHM signal. No study has tested topical retinoids, SPF, or lifestyle interventions (exercise, diet) against any skin methylation clock. needs-human-replication

  3. UV / sun-exposure confounding. Photoaging is the dominant external driver of skin epigenetic change, yet most skin clocks were trained on forearm (mixed UV exposure) or forehead (high UV) without systematic UV-history covariate control. Qi 2026 included a sunscreen partial-correlation analysis; other clocks have not addressed this confound. needs-replication

  4. Tissue-region specificity. Face, forearm, scalp, and volar arm differ in UV exposure history and intrinsic aging rates. Whether a clock trained on one region transfers to another has not been systematically characterized. The Menendez Vazquez 2025 abstract claims generalization across anatomical sites — details pending full-text access.

  5. Sex and ethnicity validation. All Bormann-lineage training data originates in white (Caucasian) female cohorts; cross-ethnic validation exists only for Qi 2026 (n=17 pilot) and the TapeLift application by Qi 2026 (same n=17). MitraSolo/MitraCluster cohort demographics are unverified from abstract alone.

  6. Lack of mortality or morbidity outcome validation. None of the four clocks has been validated against photoaging severity scores, keratinocyte cancer risk, wound healing outcomes, or all-cause mortality in a prospective cohort. All four are chronological-age-trained (first generation); no skin-clock mortality-trained equivalent of PhenoAge or GrimAge exists. needs-replication

Cross-references

  • biological-age-measurement — parent clock-landscape MOC; skin clocks are catalogued in the “Tissue-specific clocks” section
  • epigenetic-alterations stub — epigenetic aging hallmark anchor; skin clocks are direct biomarkers of this hallmark
  • epidermis — tissue context for all four clocks
  • skin — broader tissue page
  • dihydromyricetin — the only intervention with a published positive skin-clock signal (Falckenhayn 2024 in vitro + Qi 2026 topical RCT-context)
  • skin-autofluorescence-age-reader — structural skin aging biomarker (collagen-AGE fluorescence); distinct from methylation clocks but covers overlapping skin-aging biology

Last updated 2026-05-19 — R45 post-batch atomic-page integration (Menendez Vazquez 2025 + Bienkowska 2026 propagation). Four atomic skin-clock pages now seeded. No 2024–2026 skin-clock papers identified beyond the four covered here plus Bienkowska 2026 (same-group population study).

Graph View