🧬 Aging Wiki

qi-23k-epidermis-clock-2026

Properties1
aliasesQi 23k epidermis clock, 23k-CpG skin clock, Qi 2026 epidermis clock, 23k methylation clock skin

11 min read

Qi 23k Epidermis Clock (2026)

A first-generation DNA methylation clock trained specifically on human epidermis, extending and cross-validating the earlier bormann-epidermis-clock-2016 training set across multiple Fitzpatrick phototypes (I–VI) and ethnicities without detectable bias. The clock uses 173 CpG features selected by elastic-net from a panel of 23,845 probes shared between the original 27k array (Bormann 2016) and modern Illumina 450k arrays. It is the first published skin-specific methylation clock validated across the full phototype spectrum, and serves as the readout tool in the Qi 2026 topical DHM serum trial 1.

Identity and Origin

  • Primary citation: Qi M et al. (2026) “Epigenetic Skin Aging and Its Reversal to Improve Skin Longevity across Ethnicities and Phototypes Using a Dihydromyricetin-Containing Serum: Results from a Prospective, Single-Cohort Study.” Dermatology and Therapy — doi:10.1007/s13555-026-01764-4 · PMID 42034839
  • Archive status: not in a local paper archive as of 2026-05-19 (paper published 2026-04-25; too recent for OpenAlex ingest). Open-access CC BY-NC 4.0; Springer full text requires authentication even for OA articles (verified 2026-05-19). PDF verification of numerics is queued for R44/R45 when archive ingest completes or direct PDF access is obtained.
  • Clock type: First-generation (chronological-age-trained), tissue-specific
  • Developer group: Beiersdorf AG / DKFZ collaboration (8 of 15 authors are Beiersdorf AG employees; both corresponding authors Falckenhayn C and Grönniger E are Beiersdorf employees)
  • Training basis: Bormann 2016 cohort extended; 23,845-probe panel is the intersection of the Bormann 27k array and modern 450k arrays
  • Feature selection: Elastic-net penalised regression → 173 final CpG features
  • Input modality: Illumina 450k-compatible array (tape-strip epidermis sampling)
  • Output units: Estimated chronological age in years (absolute biological age, not a rate)

Methodology

Probe panel and training

The clock was built from the intersection of probes on the Bormann 2016 27k array and the Illumina 450k array — yielding 23,845 candidate probes. Elastic-net regression on the extended Bormann cohort (108 white females, ages 18–78) selected 173 CpGs. The training target was chronological age — the same frame used by Horvath 2013 and Bormann 2016, but applied to epidermis-specific methylation patterns rather than pan-tissue blood or multi-tissue profiles 2.

The biological basis of the selected CpGs is described by Qi 2026 as concentrated in bivalent chromatin regions — loci marked by both H3K4me3 (active) and H3K27me3 (repressive) histone marks, typically poised developmental enhancers. Age-dependent hypermethylation at these bivalent loci is conserved across Fitzpatrick phototypes I–VI in the pilot cohort (n=17, multi-ethnic) 1. The hypermethylated genes are described as important for keratinocyte vitality; the full list is in Qi 2026 supplementary material and has not been independently verified here. needs-pdf-verification

Sampling approach

The clock is calibrated to adhesive tape-strip epidermis samples — a non-invasive method collecting upper stratum corneum and outer viable epidermis. The tape-strip approach was also used by tapelift-clock-2026 (Rodríguez-Paredes M et al. 2026, NPJ Precision Oncology, doi:10.1038/s41698-026-01302-7) and the independent MitraSolo / MitraCluster clocks (Menendez Vazquez 2025, NPJ Aging, doi:10.1038/s41514-025-00314-0), establishing tape-stripping as an emerging standard for non-invasive epigenetic skin sampling 3.

Performance

Cross-validation (Bormann cohort)

MetricValueNotes
Cross-validation error5.66 yrglmnet cv.glmnet with type.measure="mse" on Bormann 2016 training set; paper labels this as “cross-validation error (type measure=‘mse’) of 5.66 years” — not explicitly labeled MAE; likely RMSE or MSE-based CV loss rather than MAE
Training targetChronological age
Training n108 white females, ages 18–78Bormann 2016 cohort

Multi-ethnic / multi-Fitzpatrick validation (Qi 2026 pilot)

MetricValueNotes
Validation MAE4.88 yrn=17 multi-ethnic pilot (ages 24–69, Fitzpatrick I–VI)
Ethnic bias detectedNoneNo significant MAE difference by phototype or ethnic group
Baseline MAE on product-use cohort6.2 yrn=60 Brazilian volunteers, ages 40–70; higher than pilot MAE, likely reflecting age-range compression

The lower MAE in the cross-ethnic validation cohort vs. the training-cohort cross-validation (4.88 yr vs. 5.66 yr) may reflect cohort composition and age-range differences. This directional result supports generalisability, but the validation n=17 is small and the result should be treated as preliminary. needs-replication

Test-retest reliability

Test-retest ICC has not been reported for the 23k clock. This is a notable gap. The Higgins-Chen 2022 analysis of blood-based methylation clocks demonstrated that conventional (non-PC) clock implementations can have substantial technical noise, substantially reducing interpretable ICC 4. Whether the 23k epidermis clock suffers from similar noise is unknown. In contrast, the independent MitraCluster clocks report intra-individual prediction variation <2 years, suggesting tape-strip-based clocks can achieve usable reproducibility 3. The 23k clock’s ICC should be established before it is used as a primary endpoint in powered trials. needs-replication

Comparison to Other Skin-Specific Clocks

ClockYearFeaturesTraining approachMAEEthnic validationTest-retest
bormann-epidermis-clock-20162016Subset of 27k probesAging Cell; 108 white femalesNot stated separatelyWhite-only cohortNot reported
tapelift-clock-20262026TapeLift panelTape-strip trainedNot extractedNo bias in Qi 2026 cohortNot reported
Qi 23k clock (this page)2026173 / 23,845Elastic-net on 450k-compatible probes5.66 yr (CV); 4.88 yr (multi-ethnic)Fitzpatrick I–VI, no bias detectedNot reported
[[menendez-vazquez-2025-mitrasolo-mitraclusterMitraSolo / MitraCluster]]2025Enzymatic methyl-seq462 epidermal samples~4 yrNot specified
horvath-clock-20132013353 CpGsPan-tissue / blood-trained~3.6 yr (pan-tissue)Validated across multiple tissuesHigh ICC in blood

The Qi 23k clock and the Bormann 2016 clock share a training lineage and probe panel — they are not independent. The TapeLift clock and MitraSolo/MitraCluster clocks are conceptually similar but derive from independent training pipelines. A systematic comparison of all skin-specific clocks on a common held-out multi-ethnic cohort has not been published as of 2026-05-19. needs-replication

Intervention Responsiveness

Qi 2026 DHM topical serum — primary signal

In the Qi 2026 product-use phase (n=60 Brazilian volunteers, ages 40–70, Fitzpatrick I–VI; 8-week twice-daily application of a topical serum containing DHM + hyaluronic acid + glycine saponin + enoxolone, with SPF 50+ sunscreen co-applied daily), the 23k clock showed 1:

EndpointResultp-value
Mean DNAm-age change (paired Wilcoxon, n=60)−2.08 yr (paper text rounds to “around 2.1 years”; Fig 2a label: “Δ pred. age = −2.08 y”)0.029
Responders (≥5 yr reduction)40% of participants—
CI for mean changeNot reported—

Critical design caveat: This was an open-label single-arm study with no vehicle control and no randomisation. The 2.1 yr reduction cannot be attributed to DHM alone — it could reflect contributions from HA, sunscreen-mediated UV reduction, regression to the mean, or tape-strip site drift between visits. The authors explicitly acknowledge the absence of a vehicle arm. This should be interpreted as a positive cosmetic-grade signal rather than aging-RCT-grade evidence that DHM reverses epigenetic aging. See qi-2026-dhm-epigenetic-skin-aging for full design details and limitations. needs-replication

The proposed mechanism is dihydromyricetin (DHM / ampelopsin) acting as a DNMT1 inhibitor, partially reversing the age-associated bivalent-region hypermethylation documented in the pilot cohort. The DNMT-inhibitor characterisation derives from Falckenhayn 2023/2024 (in vitro) 5. Topical penetration to viable basal keratinocytes at concentrations sufficient for DNMT inhibition in vivo has not been directly demonstrated. no-mechanism

Bienkowska 2026 population-cohort support

The Bienkowska 2026 observational study (n=851, Clinical Epigenetics, doi:10.1186/s13148-026-02101-4) examined 326 lifestyle, physiological, and pharmacological variables against skin DNAm age. DHM was identified among the factors associated with methylation patterns consistent with decelerated epigenetic aging 6. This is from a Beiersdorf-funded observational analysis (the same group as Qi 2026) and does not constitute independent replication, but the population-scale n=851 is substantially larger than Qi 2026 and uses a correlational rather than interventional design. The specific clock used (23k clock vs Bormann) and the effect magnitude are not fully extracted here. needs-pdf-verification

Conflict-of-Interest Profile

This clock was developed at Beiersdorf AG (Hamburg). The critical considerations for wiki users:

  1. 8 of 15 authors are Beiersdorf AG employees (confirmed via PubMed PMID 42034839 COI disclosure); both co-corresponding authors (Falckenhayn C, Grönniger E) are Beiersdorf employees. Guadanhim, Frey, and Lyko received consultation fees from Beiersdorf; Rodríguez-Paredes received speaker fees from Beiersdorf SA. Only 3 authors (Raddatz, Feng, Wanitphakdeedecha) declared no commercial relationship.
  2. The active ingredient (DHM), the multi-ingredient serum formulation, and the 23k clock used to measure efficacy were all developed in-house at Beiersdorf / DKFZ. A company measuring its own product’s effect with its own tool is a maximal conflict structure — not misconduct per se, but a strong reason for independent replication.
  3. The Bienkowska 2026 supporting study (n=851) is from the same group and should not be treated as fully independent external validation.
  4. The appropriate use of this clock in the wiki is as a signal of skin-specific epigenetic aging biology (bivalent-region CpG drift is cross-ethnic and conserved) and as a candidate readout tool whose intervention-responsiveness is plausible but requires vehicle-controlled replication.

The wiki should NOT cite Qi 2026 as evidence that DHM is a senotherapeutic compound in the geroprotector sense; see dihydromyricetin for the compound-level evidence assessment.

Limitations and Gaps

  1. Single-paper origin. The 23k clock has not been validated by an independent group using the same probe set. MitraSolo/MitraCluster (Menendez Vazquez 2025) and TapeLift (Rodríguez-Paredes 2026) are conceptually analogous but independently trained — they are not replications of the 23k clock per se. needs-replication

  2. No test-retest ICC. Essential for an intervention biomarker; must be established before this clock is used as a primary endpoint in powered trials. needs-replication

  3. No vehicle-controlled intervention study. The 2.1 yr DNAm-age reduction in Qi 2026 is uncontrolled; the contribution of DHM vs. other formulation components vs. sunscreen vs. measurement noise cannot be disaggregated with the current data. needs-replication

  4. Validation n is small. The multi-ethnic validation cohort is n=17 — too small to detect subtle phototype-specific biases with adequate power. A larger independent validation cohort is required. needs-replication

  5. No longitudinal calibration. The training target is chronological age (cross-sectional); the clock is not known to track biological age acceleration from smoking, UV damage, or disease. Whether it behaves as a biological-age estimator (and not just a chronological-age proxy) beyond the Qi 2026 intervention context is unknown. no-mechanism

  6. Beiersdorf funding throughout. All published supporting evidence to date comes from Beiersdorf-affiliated groups. Independent validation is the single highest-priority next step.

  7. Asian and male volunteer under-representation in the product-use cohort limits external validity.

Cross-references

Footnotes

Footnotes

  1. qi-2026-dhm-epigenetic-skin-aging · doi:10.1007/s13555-026-01764-4 · PMID 42034839 · n=17 (pilot/clock-development, multi-ethnic, Fitzpatrick I–VI, ages 24–69) + n=60 (product-use, Brazilian, ages 40–70, Fitzpatrick I–VI) · prospective single-cohort · in-vivo · model: Homo sapiens (epidermis, tape-strip) · Dermatology and Therapy 2026 · open-access CC BY-NC 4.0 · not in archive (paper too recent); 8 of 15 authors Beiersdorf employees (both corresponding authors); no vehicle control in product-use phase · PDF verified 2026-05-19 against user-provided local copy ↩ ↩2 ↩3

  2. doi:10.1111/acel.12470 · Bormann F et al. · Aging Cell 2016 · n=108 white females, ages 18–78 · observational (cross-sectional, 27k methylation array) · model: Homo sapiens (epidermis, biopsy) · archive: local PDF available (gold OA; citation percentile 100th, 94 citations) · training cohort for the 23k clock ↩

  3. doi:10.1038/s41514-025-00314-0 · Menendez Vazquez A et al. · NPJ Aging 2025 · n=462 epidermal samples · in-vivo (tape-strip, enzymatic methyl-sequencing) · MitraSolo and MitraCluster clocks; MAE ~4 yr; intra-individual variation <2 yr; independent group from Beiersdorf · archive: in archive (pending download) ↩ ↩2

  4. doi:10.1038/s43587-022-00248-2 · Higgins-Chen AT et al. · Nat Aging 2022 · principal-component clock versions substantially improve test-retest reliability; conventional clocks have substantial technical noise · review / methods · the 23k epidermis clock has not been evaluated in a similar reliability framework ↩

  5. doi:10.3389/fragi.2023.1258184 · Falckenhayn C et al. · Frontiers in Aging · received 13 July 2023; accepted 12 December 2023; published online 04 March 2024 (DOI year-in-path “2023” reflects submission year, per Frontiers journal convention; publication year is 2024) · in-vitro characterisation of DHM as a DNMT1 inhibitor in human skin keratinocytes (13 keratinocyte lines, EPIC 850k array, paired with in vivo tape-strip and 3D skin model data) · archive: local PDF available (gold OA; FWCI 29.2; citation percentile 100th) ↩

  6. doi:10.1186/s13148-026-02101-4 · Bienkowska A, Qi M et al. · Clinical Epigenetics 2026 · n=851 participants; 326 lifestyle/physiological/pharmacological variables examined against skin DNAm age · observational · model: Homo sapiens (skin) · DHM identified as associated with decelerated epigenetic aging · not in archive (too recent); Beiersdorf-affiliated group · needs-pdf-verification for clock used and effect magnitude ↩

Graph View

Backlinks