The work builds on a technique known as direct conversion, in which specialized cells are pushed into a new identity without passing through the pluripotent state that defines induced pluripotent stem cells, or iPSCs. When somatic cells are reprogrammed into iPSCs, their DNA methylation age — a molecular readout of biological aging — is rewound to an embryonic state. In contrast, when cells are converted directly into post-mitotic neurons, they largely retain the age of the cells they came from. The new study sits intriguingly between these two poles: induced neural stem cells, or iNSCs, are self-renewing like stem cells, yet they are generated without a pluripotent intermediate.

The research team, led by investigators at the University of Bonn, used a Sendai virus system to deliver just two transcription factors, SOX2 and cMYC, into erythroid progenitor cells derived from easily accessible blood samples. Within days of infection, the blood cells downregulated the erythroid marker CD71, and by day twelve, the emerging cells expressed the neural stem cell markers DACH1 and PAX6. Morphologically, the cultures already resembled neuroepithelial tissue within the first two weeks. On the surface, the fate switch appeared to happen quickly.

But the molecular data told a different story. Using the epigenetic clock algorithm developed by Steve Horvath and colleagues, the researchers measured DNA methylation age across the conversion time course. While the starting blood cells correlated precisely with the chronological age of their donors, the converted iNSCs at low passage carried a DNA methylation age averaging only about thirteen percent of the donor’s chronological age — and in established lines, less than five percent. Remarkably, this de-aging did not happen in a burst. Instead, it stretched across approximately fifty days, with some individual lines requiring around seven weeks to reach roughly twenty percent of their original epigenetic age.

This slow kinetics stands in sharp contrast to classic iPSC reprogramming, where the epigenetic age reset is largely complete within about twenty days. The protracted timeline suggests that the molecular machinery responsible for erasing age-related methylation signatures operates on a different schedule when pluripotency is not engaged. The researchers also compared their iNSCs with isogenic iPSC-derived neural stem cells generated from the same blood samples, providing a matched reference for cells that have undergone a complete age reset through the pluripotent route.

Perhaps the most unexpected result came from experiments designed to test whether cell division drives the de-aging process. Because both iPSC reprogramming and iNSC conversion involve proliferating cells, while direct conversion into non-dividing neurons preserves epigenetic age, the team hypothesized that mitotic activity might be the engine of rejuvenation. To test this, they slowed proliferation during conversion using either glycerol or thymidine, both of which significantly reduced accumulated population doublings. Yet the epigenetic de-aging continued essentially unabated. Even when the researchers forced the converting cells to differentiate into neurons by overexpressing NGN2 — producing cultures dominated by TUBB3-positive neurons — the DNA methylation age continued to decline at a rate indistinguishable from proliferating controls.

This finding implies that epigenetic de-aging during neural conversion is not simply a byproduct of stem cell self-renewal. It may instead be driven by the sustained presence of the conversion factors themselves, or by molecular processes that operate independently of both cell division and fate consolidation. The researchers note that SOX2 is the one factor common to their system and to a separate study in which human fibroblasts were converted into neural plate border stem cells using BRN2, KLF4, ZIC3, and SOX2 — a conversion that also produced dramatically de-aged cells. That convergence makes SOX2 a compelling candidate for future investigations into the mechanisms of somatic cell rejuvenation.

The consolidation of neural identity itself proved equally drawn out. Genome-wide analysis of DNA methylation and RNA sequencing revealed that while early neural markers appeared within days, the full transcriptional and epigenetic remodeling continued for weeks. Genes associated with neurogenesis, axon development, and neuron projection morphogenesis were still being upregulated at the late, transgene-free stage, long after the cells had adopted a stable neural stem cell morphology. Pathways related to cell migration, adhesion, and the mesenchymal-to-epithelial transition — reflecting the dramatic shift from freely circulating blood progenitors to adherent neuroepithelial cells — were regulated across nearly all stages of conversion.

Intriguingly, the transcriptomic signatures that distinguish young from old donors faded faster than the epigenetic age markers. At the starting blood cell stage, more than 1,800 genes differed in expression between cells from newborns and those from elderly donors. By day fourteen of conversion, only twenty such genes remained differentially expressed, and at low passage, just twelve. This suggests that transcriptional de-aging may be achieved more rapidly than the deeper methylation-level reset, and that the overexpressed transcription factors may directly drive the early transcriptomic changes rather than acting through epigenetic intermediaries.

The implications extend beyond basic biology. Because iNSCs generated even from centenarian donors lack classic cellular aging hallmarks — including nuclear lamina abnormalities, impaired autophagy, mitochondrial dysfunction, and senescence markers — they closely resemble their iPSC-derived counterparts while offering a faster and more direct route to patient-specific neural cells. The extended fifty-day window of active de-aging also provides researchers with an unprecedented opportunity to dissect the mechanisms of epigenetic rejuvenation in real time, potentially offering an alternative to partial or interrupted reprogramming strategies currently being explored for anti-aging therapies. As the authors conclude, their conversion system may serve as a blueprint for future studies seeking to identify the true drivers of age reversal in somatic cells — a question that has become one of the most urgent in the biology of aging.

Subject of Research: Epigenetic de-aging and fate acquisition during direct conversion of human blood cells into induced neural stem cells

Article Title: Protracted Fate Acquisition and Epigenetic De‐Aging During Induced Neural Stem Cell Conversion of Human Blood Cells

Article References: Berg, L. J., Franzen, J., Buness, A., Konang, R., Sheng, C., Peitz, M., Till, A., Wagner, W., & Brüstle, O. (2026). Protracted Fate Acquisition and Epigenetic De‐Aging During Induced Neural Stem Cell Conversion of Human Blood Cells. Aging Cell, 25(10), Article e70751. https://doi.org/10.1111/acel.70751

Image Credits: AI Generated

DOI: 10.1111/acel.70751

Keywords: induced neural stem cells, epigenetic clock, DNA methylation age, cellular reprogramming, SOX2, cMYC, direct conversion, aging, neural stem cells, iPSC, rejuvenation, transcription factors

The research, published in the open-access journal Epigenetics Communications, represents the first systematic characterization of DNA methylation age in cerebrospinal fluid during the acute recovery period after aneurysmal subarachnoid hemorrhage. DNA methylation age, often called epigenetic age, is estimated from chemical tags called methyl groups that attach to specific positions in the genome, known as CpG sites. Over the past decade, scientists have developed several so-called epigenetic clocks, mathematical models that convert methylation patterns into an age estimate. When a person’s epigenetic age exceeds their chronological age, a phenomenon called age acceleration, it has been linked to cancer, Parkinson’s disease, cardiovascular disease, and all-cause mortality. What remained unknown was whether these clocks, most of which were trained on blood or a mixture of tissues, could produce meaningful readings in cerebrospinal fluid, a tissue that normally contains fewer than five cells per milliliter.

Lead author Lacey W. Heinsberg of the University of Pittsburgh and colleagues tackled this question using an extraordinary dataset: serial cerebrospinal fluid samples from 273 patients recovering from aneurysmal subarachnoid hemorrhage, yielding 850 individual measurements collected over the first 14 days after the hemorrhage. The samples came from external ventricular drains placed as part of standard clinical care to relieve pressure inside the skull, meaning the researchers could track the molecular environment of the injured central nervous system without any additional invasive procedures. For a subset of 72 participants, the team also had blood samples collected during the same early window, allowing a direct comparison between the epigenetic age signatures of the brain’s local environment and the peripheral circulation.

The researchers applied four different epigenetic clocks to their data: the Horvath clock, built from 353 CpG sites and designed as a pan-tissue estimator; the Hannum clock, built from 71 sites and trained on blood; the Levine clock, built from 513 sites and designed to estimate a broader biological metric called phenotypic age; and the Zhang clock, also known as the Improved Precision clock, built from 514 sites and trained on 13,661 blood and saliva samples, by far the largest training dataset of the four. Across all cerebrospinal fluid samples, DNA methylation age correlated moderately to strongly with chronological age, with Pearson correlation coefficients ranging from 0.66 for the Levine clock to a striking 0.97 for the Zhang clock. In blood, the correlations were similarly strong, ranging from 0.83 to 0.97, confirming that even in the chaotic aftermath of a brain hemorrhage, the fundamental relationship between methylation patterns and age remains intact.

The comparison between tissues proved especially revealing. For the 72 participants with both cerebrospinal fluid and blood data, the two tissues produced correlated but clearly distinct age estimates in the Horvath, Hannum, and Levine clocks, with correlations ranging from 0.69 to 0.87. The Zhang clock, however, produced a near-perfect correlation of 0.98 between the two tissues, suggesting that its estimates are essentially interchangeable regardless of whether the DNA comes from the fluid surrounding the brain or from the bloodstream. This robustness extended to the thorniest technical challenge in the study: cell-type heterogeneity. After an aneurysm ruptures, cerebrospinal fluid becomes heavily contaminated with blood cells that gradually clear during recovery, while cells originating in the brain and from the ruptured vessel appear in the mix. Because methylation patterns differ across cell types, these shifting proportions can distort age estimates. No reference-based deconvolution method exists for cerebrospinal fluid, so the team used a reference-free approach to estimate cell-type proportions and repeated their analyses with and without this adjustment.

The results of that sensitivity analysis were decisive. The Zhang clock’s data distributions looked essentially the same whether the estimates came from cerebrospinal fluid or blood, and whether or not cell-type heterogeneity was accounted for. The other clocks showed greater sensitivity to tissue source and cell composition. The authors conclude that, of the four clocks examined, the Zhang clock is the most robust to cell-type heterogeneity and is the recommended choice for complex tissues such as cerebrospinal fluid. This finding is particularly notable because the Zhang clock, like the others, was never trained on cerebrospinal fluid data at all. Its apparent resilience likely stems from the sheer scale of its training dataset, which taught the model to distinguish age-related methylation signals from tissue-specific noise more effectively than its predecessors.

To assess whether epigenetic age shifts during recovery, the team used group-based trajectory analysis, a statistical method that infers distinct subgroups of patients following different temporal patterns. For the Horvath clock, both adjusted and unadjusted for cell-type heterogeneity, the analysis identified four distinct but flat trajectory groups, meaning age acceleration did not change over the 14-day window. The Zhang clock, unadjusted, showed the same pattern of four stable groups. The Hannum clock showed modest temporal variation in its unadjusted form, but once cell-type heterogeneity was controlled, that variation washed out into flat trajectories as well. The Levine clock’s trajectory models failed the study’s posterior quality-control checks due to uncertainty in classifying middle groups, underscoring that this clock is the least well suited to this tissue. Taken together, the trajectories suggest that epigenetic age, as measured by these clocks, is not materially perturbed by the acute pathological environment after aneurysmal subarachnoid hemorrhage.

The study also surfaced intriguing associations between epigenetic aging and patient characteristics. Sex was associated with trajectory group membership in the Hannum clock, with the proportion of women declining as age acceleration rose: the lowest-acceleration group was 93.3 percent female, while the highest-acceleration group was only 16.7 percent female. This aligns with a broader literature showing that men tend to have higher DNA methylation age than women. Independent of trajectory groups, sex was also associated with age acceleration in the Horvath and Hannum clocks, race was associated with acceleration in the Hannum and Levine clocks, and smoking was associated with Levine clock acceleration after cell-type adjustment. Notably, none of the trajectory groups differed meaningfully in injury severity, body mass index, or other clinical characteristics, a somewhat surprising result the authors highlight as deserving further study.

The practical implications cut in two directions. On one hand, the stability of cerebrospinal fluid epigenetic age during recovery suggests that researchers designing future studies could gain statistical power by collecting a single measurement from more participants rather than generating expensive longitudinal methylation data for each patient. On the other hand, the very stability that makes the measurement reliable also suggests that age acceleration from these particular clocks is unlikely to add predictive value for recovery outcomes beyond what chronological age already provides. The authors argue that a clock specifically trained on cerebrospinal fluid methylation data from the acute period after hemorrhage would have the greatest clinical potential, joining a growing movement toward tissue- and disease-specific epigenetic clocks, such as those developed for placental aging and for hippocampal and cortical tissue in Alzheimer’s models.

The study is not without limitations. Blood samples were processed on separate plates from cerebrospinal fluid samples, preventing adjustment for potential plate batch effects, though the authors note that correlation coefficients are invariant to changes in origin and scale, and the near-perfect Zhang clock concordance between tissues supports the validity of their comparisons. The blood subset was small, at 72 participants with a single time point, precluding trajectory comparisons in blood. And because no other cohort with serial cerebrospinal fluid methylation data exists, the findings could not be replicated in an independent sample. Still, as the first window into how the epigenetic clocks behave in the fluid that cushions an injured brain, the work establishes a methodological foundation: the Zhang clock delivers precise, stable, tissue-agnostic age estimates even amid the cellular turbulence of a bleeding stroke, while the field moves toward clocks tailored to the tissues and diseases that matter most.

Subject of Research: DNA methylation age in cerebrospinal fluid during recovery from aneurysmal subarachnoid hemorrhage

Article Title: Characterization of cerebrospinal fluid DNA methylation age during the acute recovery period following aneurysmal subarachnoid hemorrhage

Article References: Characterization of cerebrospinal fluid DNA methylation age during the acute recovery period following aneurysmal subarachnoid hemorrhage. (n.d.). https://doi.org/10.1186/s43682-021-00002-6

Image Credits: AI Generated

DOI: 10.1186/s43682-021-00002-6

Keywords: epigenetic clocks, DNA methylation age, cerebrospinal fluid, aneurysmal subarachnoid hemorrhage, biological aging, age acceleration, cell-type heterogeneity, Zhang clock, Horvath clock, stroke recovery, epigenetics, biomarkers

The study, led by Thomas Jonkman, Erik van Zwet, and Bastiaan Heijmans of Leiden University Medical Center together with Anne Richmond and Riccardo Marioni of the University of Edinburgh, drew on DNA methylation data from 4,058 whole blood samples collected across six Dutch biobanks in the BIOS Consortium. Participants ranged from 18 to 87 years old. Using the EpiDISH deconvolution algorithm, the team estimated the proportions of twelve distinct blood cell types in each sample, including neutrophils, monocytes, natural killer cells, and naive and memory subsets of B cells and T cells. Neutrophils dominated the average blood profile, making up 56.8 percent of cells, while the lymphocyte fractions showed clear age-related shifts: naive CD4 and CD8 T cells declined with age, whereas memory T cell fractions increased.

A central methodological hurdle the researchers had to overcome is collinearity. Blood cell fractions are not independent measurements; they are biologically correlated and mathematically constrained to sum to 100 percent. As the authors demonstrate with a series of example regressions, this makes effect sizes for individual cell types deeply ambiguous. When the neutrophil fraction was modeled alone, its apparent effect on age was nearly zero, but adjusting for memory CD4 T cells flipped the estimate positive, and adjusting for naive CD8 T cells flipped it negative. In multivariable models, whichever cell fraction was omitted to satisfy the mathematical constraint had its effect absorbed into the intercept, systematically skewing the coefficients of the remaining fractions. The total variance explained stayed constant, but the interpretation of individual effects became meaningless.

To escape this trap, the team turned to principal component analysis. Because twelve cell fractions contain only eleven independent pieces of information, the researchers computed eleven principal components, each an uncorrelated linear combination of the original fractions. The leading components proved biologically interpretable: the first was almost perfectly correlated with the neutrophil fraction, while the second captured the balance between naive and memory T cells, one of the most striking signatures of immunosenescence. Crucially, because principal components are orthogonal to one another, the effect of each can be estimated independently and reliably, a property the authors verified by showing that removing a component left the remaining effect sizes unchanged.

Applying this framework, the researchers found that cell composition explains a remarkable share of the variation in DNA methylation age itself. Across the six clocks studied, including the first-generation Hannum, Horvath, and Zhang clocks and the second-generation PhenoAge, GrimAge, and DunedinPACE measures, cell counts accounted for 44 to 53 percent of the variance in predicted age, with the naive-to-memory T cell balance the dominant contributor. Calendar age itself was 43.5 percent explainable by cell composition. Yet when the team examined age acceleration, the residual difference between predicted and chronological age, the picture changed dramatically. Associations weakened sharply, explaining at most 21 percent of variance, and the specific cells involved shifted: for second-generation clocks, neutrophils took center stage while T cell contributions largely faded.

The divergence between DNA methylation age and age acceleration carries a profound implication. If age acceleration truly represented “biological age,” then a one-year advance in predicted age should reflect the same underlying biology as a one-year excess over chronological age. The data show this is not the case, at least with respect to blood cell composition. The CpG sites making up first-generation clocks responded almost identically to calendar age and to age acceleration, with correlations between the two effect estimates reaching 0.80 to 0.91. For second-generation clocks, those correlations collapsed to between 0.13 and 0.50, and the age-acceleration estimates carried standard errors nearly three times larger, indicating substantial noise layered on top of genuinely distinct biology.

To establish that cell composition actually drives clock readings rather than merely correlating with them, the researchers performed an elegant validation experiment. They downloaded DNA methylation profiles of 56 purified blood cell samples from a public repository and constructed artificial cell mixtures in silico, blending the average methylation profile of each cell type in proportions matching the population mean. They then generated twelve additional mixtures, each time raising one cell fraction by one standard deviation while proportionally shrinking the others. Running the clocks on these synthetic samples reproduced the observational findings: boosting naive CD4 or CD8 T cells lowered predicted age acceleration for the Hannum, PhenoAge, GrimAge, and DunedinPACE clocks, while increasing neutrophils raised it. The Horvath and Zhang clocks remained largely indifferent to these manipulations.

The final question was whether this cell-composition signal contaminates the clinically important associations between epigenetic age acceleration and disease. Using data from 18,859 participants in the Generation Scotland cohort, the team tested the clocks against 176 incident health outcomes, including all-cause mortality. Cell composition itself was significantly associated with twelve outcomes after Bonferroni correction, among them mortality, asthma, and non-Hodgkin lymphoma, with mortality linked to neutrophil-related and regulatory T cell components but notably not to the naive-to-memory T cell ratio. When the researchers adjusted the clock-outcome associations for cell composition, the hazard ratios were attenuated for four of the six clocks, but only modestly, by roughly 2 to 7 percent overall and between 1 and 6 percent for mortality.

That modest attenuation is perhaps the study’s most reassuring finding for the field. Even though blood cell composition drives a substantial fraction of clock variation, it explains less than 10 percent of the association between age acceleration and incident disease. The authors suggest that neutrophil abundance, which does not change with chronological age yet tracks mortality, frailty, and inflammation, may act as a health-relevant biomarker that clocks capture independently of the aging process itself. This raises a deeper conceptual question: if changes in cell composition are part of how aging manifests in blood, for example through chronic inflammation, then adjusting them away in epigenetic association studies may discard biologically meaningful signal rather than confounding.

For the growing consumer market in epigenetic age testing, the message is one of caution and nuance. A high epigenetic age acceleration reading on a second-generation clock may partly reflect an elevated neutrophil fraction rather than accelerated cellular aging, while a Horvath or Zhang reading may be largely insensitive to immune composition but noisier as a health predictor. The researchers provide their principal component loadings and analysis code publicly, allowing other teams to project the same decomposition onto new datasets. As epigenetic clocks continue to migrate from research laboratories into clinics and commercial services, this work makes clear that interpreting what a clock is actually measuring requires looking carefully at the blood it was drawn from, and that the biology of predicted age and the biology of age acceleration are, in important respects, two different stories.

Subject of Research: The contribution of blood cell composition to DNA methylation age and epigenetic age acceleration

Article Title: Blood cell composition reveals distinct biological interpretation of DNA methylation age and age acceleration

Article References: Jonkman, T. H., Richmond, A., BIOS Consortium, Marioni, R. E., van Zwet, E. W., & Heijmans, B. T. (2026). Blood cell composition reveals distinct biological interpretation of DNA methylation age and age acceleration. Genome Medicine, 18(1), Article 142. https://doi.org/10.1186/s13073-026-01751-6

Image Credits: AI Generated

DOI: 10.1186/s13073-026-01751-6

Keywords: epigenetic clocks, DNA methylation age, age acceleration, biological age, blood cell composition, T cells, neutrophils, immunosenescence, principal component analysis, collinearity, Genome Medicine, aging biomarkers