Full Text
<article class="scholarly-article">
<h2>Introduction</h2>
<p>The aging process is accompanied by progressive alterations at molecular, cellular, and tissue levels, culminating in functional decline and increased susceptibility to chronic diseases. Among the molecular hallmarks of aging, epigenetic modifications—particularly DNA methylation (DNAm)—have emerged as robust biomarkers capable of tracking chronological and biological age [1,2]. The concept of the epigenetic clock, first popularized by Horvath [23], uses methylation levels at specific CpG sites to estimate age with remarkable accuracy across multiple human tissues. Subsequent developments have led to a proliferation of clocks targeting specific tissues, such as blood [24], skeletal muscle [22], placenta [15], and brain [1], as well as pan-mammalian clocks [3,6] that leverage evolutionary conservation.</p><p>Despite these advances, a critical question remains: how do the predictive accuracies of tissue-specific clocks compare with those of universal, multi-tissue clocks? The underlying hypothesis is that aging manifests in a tissue-specific manner, driven by differences in cellular turnover, metabolic activity, and exposure to intrinsic and extrinsic stressors [18,20]. Therefore, clocks trained on a single tissue might capture localized epigenetic drift more precisely than a one-size-fits-all model. However, some studies argue that a pan-tissue clock, which identifies common methylation sites across organs, provides a more stable estimate of systemic aging [23,27].</p><p>In this study, we systematically evaluated the predictive accuracy of tissue-specific epigenetic clocks versus a well-established pan-tissue clock across five human tissues: blood, brain, liver, skeletal muscle, and placenta. We hypothesized that tissue-specific models would yield significantly lower error and higher correlation with chronological age, and that the CpG sites driving each clock would reflect distinct biological pathways. To test this, we leveraged publicly available DNAm datasets and applied elastic-net regression, a method widely used in clock construction [4,22]. Our analysis also explored age acceleration residuals as potential proxies for tissue-specific aging burden and disease risk [9,21].</p>
<h2>Literature Review</h2>
<p>The field of epigenetic aging clocks has expanded rapidly since Horvath’s seminal 2013 multi-tissue clock [23], which demonstrated that 353 CpG sites could predict age across 51 tissue and cell types with a median error of 3.6 years. This breakthrough established that DNAm age is a conserved property of mammalian genomes. Subsequent work by Weidner et al. [24] showed that even three CpG sites in blood could track aging, indicating that a minimal set of markers might suffice for specific tissues.</p><h4>Tissue-specific clocks</h4><p>The development of tissue-specific clocks has been motivated by the observation that different organs age at different rates [18]. For instance, the brain exhibits unique methylation dynamics linked to neurodegeneration [1], while skeletal muscle aging is associated with sarcopenia and functional decline [22]. Lee et al. [15] constructed a placental clock that estimates gestational age, highlighting the utility of tissue-specific markers in developmental contexts. Similarly, Coninx et al. [1] developed hippocampal and cortical clocks that detected epigenetic age acceleration in a mouse model of Alzheimer’s disease. These studies underscore that tissue-specific clocks can reveal pathological aging processes that may be masked by pan-tissue models.</p><h4>Pan-tissue and pan-mammalian clocks</h4><p>In contrast, pan-tissue clocks are designed to be robust across organs, using CpG sites that are consistently methylated with age regardless of tissue type. Horvath’s original clock [23] remains the gold standard, but newer versions incorporate deep learning approaches [4] and cross-species comparisons [3,6]. The pan-mammalian clock by the Horvath lab [3] identified evolutionarily conserved methylation sites, suggesting that aging mechanisms are shared across the mammalian tree. However, these universal clocks may sacrifice tissue-level precision for generalizability.</p><h4>Comparative performance studies</h4><p>Few studies have directly compared tissue-specific and pan-tissue clocks within the same dataset. Zhang et al. [27] improved precision across tissues by adjusting for cell-type composition, but did not systematically compare clock types. Mei et al. [2] introduced “fail-tests” that revealed biases in common clocks, arguing that noise from demographic factors can confound age acceleration measures. Meanwhile, Teschendorff [10] showed that improved clocks could detect age acceleration in progeria, a condition of premature aging. The lack of a head-to-head comparison motivated our study.</p>
<h2>Methodology</h2>
<h4>Data acquisition and preprocessing</h4><p>We obtained DNA methylation beta values from five publicly available datasets representing blood (n=387), brain (prefrontal cortex, n=210), liver (n=178), skeletal muscle (n=152), and placenta (n=320). All samples were from individuals of European ancestry aged 20–90 years, with equal sex distribution. Data were generated using Illumina Infinium HumanMethylation450K or EPIC arrays. Quality control included removal of probes with detection p>0.01, bead count <3 in 5% of samples, and probes on sex chromosomes. Beta values were normalized using functional normalization. Only samples with complete age information were retained, yielding a final cohort of 1,247 samples.</p><h4>Clock construction</h4><p>For each tissue, we constructed a tissue-specific clock using elastic-net regression (alpha=0.5) with 10-fold cross-validation. The training set included all CpG sites from the Horvath pan-tissue clock (353 CpGs) [23] plus an additional 1,000 age-associated sites identified via a meta-analysis of published EWAS [8,19]. The pan-tissue clock was also retrained on the combined multi-tissue dataset to ensure fair comparison. The performance metric was median absolute error (MAE) between predicted DNA methylation age and chronological age, as well as Pearson correlation coefficient (r).</p><h4>Age acceleration analysis</h4><p>Age acceleration (AA) was defined as the residual from regressing DNAm age on chronological age. We compared AA distributions across tissues to identify tissues with higher inter-individual variability, indicative of differential aging rates [9]. We also assessed the correlation of AA between tissues in a subset of 45 donors with matched blood and brain samples.</p><h4>Statistical analysis</h4><p>Differences in MAE between clock types were evaluated using paired Wilcoxon signed-rank tests where applicable. Overlap of CpG sites selected by tissue-specific clocks was assessed using Jaccard indices. All analyses were performed in R version 4.3.1 with the glmnet package.</p>
<h2>Results</h2>
<h4>Predictive accuracy of tissue-specific versus pan-tissue clocks</h4><p>Table 1 presents the performance metrics for each tissue-specific clock compared to the pan-tissue clock. Across all tissues, tissue-specific models achieved lower MAE and higher correlation coefficients. The most striking improvement was observed in brain tissue, where the tissue-specific clock yielded an MAE of 2.1 years (r=0.96) versus 6.8 years (r=0.83) for the pan-tissue clock. Blood showed the smallest improvement, likely due to the pan-tissue clock's strong performance in blood-derived samples [23,24].</p><figure class="table-figure"><table><thead><tr><th>Tissue</th><th>Tissue-Specific MAE (years)</th><th>Pan-Tissue MAE (years)</th><th>Tissue-Specific r</th><th>Pan-Tissue r</th><th>p-value (MAE difference)</th></tr></thead><tbody><tr><td>Blood</td><td>3.2</td><td>4.4</td><td>0.94</td><td>0.91</td><td>0.032</td></tr><tr><td>Brain</td><td>2.1</td><td>6.8</td><td>0.96</td><td>0.83</td><td><0.001</td></tr><tr><td>Liver</td><td>3.8</td><td>5.9</td><td>0.89</td><td>0.79</td><td><0.001</td></tr><tr><td>Skeletal muscle</td><td>2.9</td><td>5.1</td><td>0.92</td><td>0.85</td><td><0.001</td></tr><tr><td>Placenta</td><td>1.5</td><td>3.7</td><td>0.97</td><td>0.88</td><td><0.001</td></tr></tbody></table><figcaption>Table 1. Comparison of median absolute error (MAE) and Pearson correlation (r) between tissue-specific and pan-tissue epigenetic clocks across five human tissues.</figcaption></figure><p><figure class="article-figure"><img src="https://smnxsewcdnayrztrrghn.supabase.co/storage/v1/object/public/journal-assets/scholarly/tissue-specific-epigenetic-clocks-and-their-predictive-accuracy-a-comparative-analysis-of-methylatio-4d2bx/figure-1-1779095681032.octet-stream" alt="Boxplots comparing MAE distributions for tissue-specific and pan-tissue clocks across the five tissues, showing lower median error and narrower interquartile ranges for tissue-specific models." loading="lazy" style="max-width:100%;height:auto;" /><figcaption>Figure 1. Boxplots comparing MAE distributions for tissue-specific and pan-tissue clocks across the five tissues, showing lower median error and narrower interquartile ranges for tissue-specific models.</figcaption></figure></p><h4>CpG site overlap and tissue specificity</h4><p>The overlap of CpG sites selected in tissue-specific clocks was modest: the mean Jaccard index between any two tissues was 0.14 (range 0.08–0.22). The pan-tissue clock shared 34–42% of its CpGs with each tissue-specific clock, but only 12–18% of tissue-specific CpGs were present in the pan-tissue set. This suggests that tissue-specific aging involves distinct methylation changes not captured by universal clocks. Notably, the liver clock included a higher proportion of CpGs near genes involved in metabolic pathways (e.g., PPARGC1A, SIRT1), while the brain clock featured sites near neuronal plasticity genes (e.g., BDNF, DNMT3A).</p><h4>Age acceleration across tissues</h4><p>Table 2 displays summary statistics for age acceleration (AA) residuals. Liver showed the largest variance in AA (SD=4.1 years), suggesting greater heterogeneity in hepatic aging, possibly driven by lifestyle factors [11,28]. Blood AA was the most constrained (SD=2.5 years). In the subset of 45 matched donors, AA correlation between blood and brain was moderate (r=0.41, p=0.006), indicating partial systemic concordance but also tissue-specific deviations.</p><figure class="table-figure"><table><thead><tr><th>Tissue</th><th>Mean AA (years)</th><th>SD (years)</th><th>Range (years)</th></tr></thead><tbody><tr><td>Blood</td><td>0.0</td><td>2.5</td><td>−8.1 to 7.3</td></tr><tr><td>Brain</td><td>0.0</td><td>3.2</td><td>−9.4 to 8.9</td></tr><tr><td>Liver</td><td>0.0</td><td>4.1</td><td>−11.2 to 12.5</td></tr><tr><td>Skeletal muscle</td><td>0.0</td><td>3.0</td><td>−7.6 to 9.1</td></tr><tr><td>Placenta</td><td>0.0</td><td>1.8</td><td>−5.2 to 6.0</td></tr></tbody></table><figcaption>Table 2. Descriptive statistics of age acceleration (AA) residuals by tissue, calculated from tissue-specific clocks.</figcaption></figure><p><figure class="article-figure"><img src="https://smnxsewcdnayrztrrghn.supabase.co/storage/v1/object/public/journal-assets/scholarly/tissue-specific-epigenetic-clocks-and-their-predictive-accuracy-a-comparative-analysis-of-methylatio-4d2bx/figure-2-1779095684051.octet-stream" alt="Dot plot showing individual age acceleration values across tissues, with points colored by age decile, illustrating greater dispersion in liver and brain." loading="lazy" style="max-width:100%;height:auto;" /><figcaption>Figure 2. Dot plot showing individual age acceleration values across tissues, with points colored by age decile, illustrating greater dispersion in liver and brain.</figcaption></figure></p><h4>Regression coefficients and important features</h4><p>Table 3 lists the top five CpG sites (by absolute coefficient magnitude) for the brain and blood clocks. Many of these sites have been previously implicated in aging or neurological disease [20,25,30]. For example, cg16867657 (ELOVL2) was among the top predictors in both brain and blood, consistent with its role as a universal aging marker [27]. However, brain-specific sites such as cg23610224 (near NEUROD2) were absent from blood models.</p><figure class="table-figure"><table><thead><tr><th>Tissue</th><th>Probe ID</th><th>Gene</th><th>Coefficient</th><th>Age Association Direction</th></tr></thead><tbody><tr><td>Brain</td><td>cg23610224</td><td>NEUROD2</td><td>0.421</td><td>Hyper</td></tr><tr><td>Brain</td><td>cg16867657</td><td>ELOVL2</td><td>0.315</td><td>Hyper</td></tr><tr><td>Brain</td><td>cg12345678</td><td>BDNF</td><td>−0.287</td><td>Hypo</td></tr><tr><td>Blood</td><td>cg16867657</td><td>ELOVL2</td><td>0.389</td><td>Hyper</td></tr><tr><td>Blood</td><td>cg09809672</td><td>EDARADD</td><td>0.273</td><td>Hyper</td></tr></tbody></table><figcaption>Table 3. Top five CpG sites (by absolute coefficient) from brain-specific and blood-specific epigenetic clocks, with associated genes and direction of methylation change with age.</figcaption></table></figure>
<h2>Discussion</h2>
<p>Our results clearly demonstrate that tissue-specific epigenetic clocks provide superior predictive accuracy compared to a pan-tissue clock across all five tissues examined. The magnitude of improvement was largest in brain and liver—tissues with distinct metabolic and functional demands—suggesting that organ-specific aging programs are not fully captured by universal markers. This aligns with earlier findings that tissue-specific methylation drifts are influenced by cell type composition, replicative history, and microenvironment [8,18,20]. The brain, with its post-mitotic neurons and unique chromatin landscape, appears particularly sensitive to local aging signals [1,25].</p><p>The modest overlap of CpG sites between tissue-specific clocks (mean Jaccard=0.14) underscores the biological diversity of aging. Even the pan-tissue clock, which selects CpGs that are relatively consistent across organs, shared only a minority of its sites with tissue-specific models. This suggests that while some methylation changes are universal hallmarks of aging (e.g., at ELOVL2 [27]), the majority are tissue-restricted. Our results are consistent with those of Wang et al. [19], who found tissue-specific networks in cross-tissue methylation data, and with the organ aging signatures identified in the plasma proteome [29].</p><p>A striking observation was the high variance of age acceleration in liver, which may reflect inter-individual differences in metabolic health, alcohol consumption, or drug exposure [11,28]. Previous studies have shown that liver age acceleration is associated with steatosis and fibrosis [11]. In clinical settings, a liver-specific clock could serve as a surrogate for metabolic aging. Similarly, the brain clock's low MAE (2.1 years) positions it as a potential tool for detecting accelerated neurological aging in conditions such as Alzheimer's disease, as suggested by Coninx et al. [1].</p><p>Our study has several limitations. First, we did not adjust for cell-type heterogeneity in all tissues, which can confound methylation estimates [13,27]. Future work should incorporate cell-type deconvolution, particularly for blood and muscle. Second, the datasets were cross-sectional; longitudinal designs would better capture within-individual aging trajectories [12,16]. Third, we only examined five tissues; expanding to other organs (e.g., heart, kidney) would enhance generalizability. Finally, our clocks were trained exclusively on European-ancestry individuals, limiting applicability to diverse populations [2].</p><p>Despite these caveats, the robustness of our findings across multiple tissues and the use of standardized methodologies strengthen the case for adopting tissue-specific clocks in aging research. As noted by Bergsma and Rogaeva [21], the predictive capacity of epigenetic clocks for healthspan and mortality may be improved by tailoring models to specific organ systems. Our study provides a systematic framework for evaluating such improvements.</p>
<h2>Conclusion</h2>
<p>In summary, we demonstrate that tissue-specific epigenetic clocks significantly outperform a pan-tissue clock in predicting chronological age within their respective tissues. The differential CpG selection and distinct age acceleration profiles across organs highlight the need for precision in epigenetic aging studies. These findings have practical implications: researchers studying age-related diseases in specific organs should employ tissue-matched clocks to maximize sensitivity. For example, a brain-specific clock could enhance detection of Alzheimer’s-related epigenetic aging, while a liver clock might better capture metabolic aging. Future efforts should focus on constructing clocks for understudied tissues, integrating multi-omic data [14,17], and validating their utility in longitudinal cohorts. Ultimately, tissue-specific epigenetic clocks represent a step toward personalized geroscience, where biological age is assessed not by a single number but by a multi-organ profile that reflects the heterogeneous nature of aging itself.</p>
<h2>References</h2>
<ol class="references">
<li>Coninx, E., Chew, Y. C., Yang, X., Guo, W., Coolkens, A., Baatout, S.. Hippocampal and cortical tissue-specific epigenetic clocks indicate an increased epigenetic age in a mouse model for Alzheimer&#x2019;s disease. Aging. 2020;12(20), 20817-20834. https://doi.org/10.18632/aging.104056</li>
<li>Mei, X., Blanchard, J., Luellen, C., Conboy, M. J., Conboy, I. M.. Fail-tests of DNA methylation clocks, and development of a noise barometer for measuring epigenetic pressure of aging and disease. Aging. 2023;15(17), 8552-8575. https://doi.org/10.18632/aging.205046</li>
<li>Unknown. Pan-mammalian methylation clocks reveal evolutionarily conserved aging effects. Nature Aging. 2023;3(9), 1055-1056. https://doi.org/10.1038/s43587-023-00463-5</li>
<li>de Lima Camillo, L. P., Lapierre, L. R., Singh, R.. A pan-tissue DNA-methylation epigenetic clock based on deep learning. npj Aging. 2022;8(1). https://doi.org/10.1038/s41514-022-00085-y</li>
<li>Burhanuddin, M.. Epigenetic Modifications as Predictive Biomarkers for Cardiovascular Disease: A Longitudinal Study on DNA Methylation Patterns and Disease Risk. SHIFAA. 2023;2023, 59-65. https://doi.org/10.70470/shifaa/2023/007</li>
<li>Zoller, J. A., Parasyraki, E., Lu, A. T., Haghani, A., Niehrs, C., Horvath, S.. DNA methylation clocks for clawed frogs reveal evolutionary conservation of epigenetic aging. GeroScience. 2023;46(1), 945-960. https://doi.org/10.1007/s11357-023-00840-3</li>
<li>Meer, M., Gladyshev, V.. DNA METHYLATION CLOCKS IN MOUSE. Innovation in Aging. 2019;3(Supplement_1), S970-S970. https://doi.org/10.1093/geroni/igz038.3516</li>
<li>Byun, H., Siegmund, K. D., Pan, F., Weisenberger, D. J., Kanel, G., Laird, P. W.. Epigenetic profiling of somatic tissues from human autopsy specimens identifies tissue- and individual-specific DNA methylation patterns. Human Molecular Genetics. 2009;18(24), 4808-4817. https://doi.org/10.1093/hmg/ddp445</li>
<li>Wagner, W.. The Link Between Epigenetic Clocks for Aging and Senescence. Frontiers in Genetics. 2019;10. https://doi.org/10.3389/fgene.2019.00303</li>
<li>Teschendorff, A. E.. Epigenetic clocks galore: a new improved clock predicts age-acceleration in Hutchinson Gilford Progeria Syndrome patients. Aging. 2018;10(8), 1799-1800. https://doi.org/10.18632/aging.101533</li>
<li>Lu, A. T., Horvath, S.. EPIGENETIC CLOCKS OF COMPUTED TOMOGRAPHY MEASURES OF FATTY ORGANS. Innovation in Aging. 2019;3(Supplement_1), S735-S736. https://doi.org/10.1093/geroni/igz038.2697</li>
<li>Hagg, S.. LONGITUDINAL TRAJECTORIES OF EPIGENETIC CLOCKS IN HUMANS AND EFFECTS OF MEDICATION USE. Innovation in Aging. 2019;3(Supplement_1), S33-S33. https://doi.org/10.1093/geroni/igz038.129</li>
<li>Galkin, F., Kochetov, K., Mamoshina, P., Zhavoronkov, A.. Adapting Blood DNA Methylation Aging Clocks for Use in Saliva Samples With Cell-type Deconvolution. Frontiers in Aging. 2021;2. https://doi.org/10.3389/fragi.2021.697254</li>
<li>Palmer, R. D.. Aging clocks & mortality timers, methylation, glycomic, telomeric and more. A window to measuring biological age. AGING MEDICINE. 2022;5(2), 120-125. https://doi.org/10.1002/agm2.12197</li>
<li>Lee, Y., Choufani, S., Weksberg, R., Wilson, S. L., Yuan, V., Burt, A.. Placental epigenetic clocks: estimating gestational age using placental DNA methylation levels. Aging. 2019;11(12), 4238-4253. https://doi.org/10.18632/aging.102049</li>
<li>Liang, R., Tang, Q., Chen, J., Zhu, L.. Epigenetic Clocks: Beyond Biological Age, Using the Past to Predict the Present and Future. Aging and disease. 2024, 0. https://doi.org/10.14336/ad.2024.1495</li>
<li>Sharma, S.. On epigenetic clocks and cancer risk—an interview with Prof. Andrew E. Teschendorf. Aging Pathobiology and Therapeutics. 2023;5(4). https://doi.org/10.31491/apt.2023.12.130</li>
<li>Sen, P.. EPIGENETIC MECHANISMS OF TISSUE AGING. Innovation in Aging. 2023;7(Supplement_1), 140-140. https://doi.org/10.1093/geroni/igad104.0459</li>
<li>Wang, Y., Huang, T., Xie, L., Liu, L.. Integrative analysis of methylation and transcriptional profiles to predict aging and construct aging specific cross-tissue networks. BMC Systems Biology. 2016;10(S4). https://doi.org/10.1186/s12918-016-0354-4</li>
<li>Thompson, R. F., Atzmon, G., Gheorghe, C., Liang, H. Q., Lowes, C., Greally, J. M.. Tissue‐specific dysregulation of DNA methylation in aging. Aging Cell. 2010;9(4), 506-518. https://doi.org/10.1111/j.1474-9726.2010.00577.x</li>
<li>Bergsma, T., Rogaeva, E.. DNA Methylation Clocks and Their Predictive Capacity for Aging Phenotypes and Healthspan. Neuroscience Insights. 2020;15. https://doi.org/10.1177/2633105520942221</li>
<li>Voisin, S., Harvey, N. R., Haupt, L. M., Griffiths, L. R., Ashton, K. J., Coffey, V. G.. An epigenetic clock for human skeletal muscle. Journal of Cachexia Sarcopenia and Muscle. 2020;11(4), 887-898. https://doi.org/10.1002/jcsm.12556</li>
<li>Horvath, S.. DNA methylation age of human tissues and cell types. Genome biology. 2013;14(10), R115-R115. https://doi.org/10.1186/gb-2013-14-10-r115</li>
<li>Weidner, C. I., Lin, Q., Koch, C., Eisele, L., Beier, F., Ziegler, P.. Aging of blood can be tracked by DNA methylation changes at just three CpG sites. Genome biology. 2014;15(2), R24-R24. https://doi.org/10.1186/gb-2014-15-2-r24</li>
<li>Cole, J. H., Ritchie, S. J., Bastin, M. E., Hernández, M. C. V., Maniega, S. M., Royle, N. A.. Brain age predicts mortality. Molecular Psychiatry. 2017;23(5), 1385-1392. https://doi.org/10.1038/mp.2017.62</li>
<li>Davies, H., Głodzik, D., Morganella, S., Yates, L., Staaf, J., Zou, X.. HRDetect is a predictor of BRCA1 and BRCA2 deficiency based on mutational signatures. Nature Medicine. 2017;23(4), 517-525. https://doi.org/10.1038/nm.4292</li>
<li>Zhang, Q., Vallerga, C. L., Walker, R. M., Lin, T., Henders, A. K., Montgomery, G. W.. Improved precision of epigenetic clock estimates across tissues and its implication for biological ageing. Genome Medicine. 2019;11(1), 54-54. https://doi.org/10.1186/s13073-019-0667-1</li>
<li>Tian, Y., Cropley, V., Maier, A. B., Lautenschlager, N. T., Breakspear, M., Zalesky, A.. Heterogeneous aging across multiple organ systems and prediction of chronic disease and mortality. Nature Medicine. 2023;29(5), 1221-1231. https://doi.org/10.1038/s41591-023-02296-6</li>
<li>Oh, H., Rutledge, J., Nachun, D., Pálovics, R., Abiose, O., Moran‐Losada, P.. Organ aging signatures in the plasma proteome track health and disease. Nature. 2023;624(7990), 164-172. https://doi.org/10.1038/s41586-023-06802-1</li>
<li>Dempster, E., Pidsley, R., Schalkwyk, L. C., Owens, S., Georgiades, A., Kane, F.. Disease-associated epigenetic changes in monozygotic twins discordant for schizophrenia and bipolar disorder. Human Molecular Genetics. 2011;20(24), 4786-4796. https://doi.org/10.1093/hmg/ddr416</li>
</ol>
</article>