A cellular identity crisis? Plasticity changes during aging and rejuvenation

In this review, Gorelov and Hochedlinger discuss the cellular differentiation programs that contribute to aging-related cellular plasticity, including dedifferentiation, biased differentiation, lineage infidelty, and malignant transformation. They further explore the mechanistic nuances of cellular reprogramming strategies that could allow for the feasible induction of rejuvenation in aged cells.

Abstract

Cellular plasticity in adult multicellular organisms is a protective mechanism that allows certain tissues to regenerate in response to injury. Considering that aging involves exposure to repeated injuries over a lifetime, it is conceivable that cell identity itself is more malleable—and potentially erroneous—with age. In this review, we summarize and critically discuss the available evidence that cells undergo age-related shifts in identity, with an emphasis on those that contribute to age-associated pathologies, including neurodegeneration and cancer. Specifically, we focus on reported instances of programs associated with dedifferentiation, biased differentiation, acquisition of features from alternative lineages, and entry into a preneoplastic state. As some of the most promising approaches to rejuvenate cells reportedly also elicit transient changes to cell identity, we further discuss whether cell state change and rejuvenation can be uncoupled to yield more tractable therapeutic strategies.

The development of multicellular organisms from a single-celled embryo exemplifies the incredible potential of cellular plasticity (the ability of cells to change their functional identity), given that millions of differentiated cells are gradually produced from a small number of uncommitted pluripotent stem cells. Cellular plasticity is generally thought to decrease with development, as specialized tissues are formed and maintained—in many cases—by a hierarchy of adult stem, progenitor, and differentiated cells.

However, increasing evidence in the context of injury and disease has demonstrated that adult cells retain a degree of plasticity beyond steady-state differentiation. Notably, this plasticity can be beneficial or maladaptive depending on the context. For example, as it relates to adaptive changes, committed tissue progenitors and even terminally differentiated cells can undergo dedifferentiation to a stem cell-like state followed by redifferentiation into mature cells or direct transdifferentiation between mature cell types in order to regenerate damaged or lost tissue (Tata and Rajagopal 2016; Traxler et al. 2023). In contrast, similar changes can be pathological, most notably in the context of cancer when cells often gain features of different cell types or lose features of their own cell type of origin through neoplastic transformation (Roy and Hebrok 2015; Gardner et al. 2024). Changes to cellular plasticity can also be recapitulated in vitro with transcription factor- and small molecule-based approaches, such as pluripotent stem cell differentiation and the reprogramming of adult cells to a pluripotent state (Cohen and Melton 2011; Oh and Jang 2019; Deng et al. 2021; Wang et al. 2023a).

The question of whether cellular plasticity is impacted by aging has only recently garnered attention, with the application of single-cell RNA sequencing technologies across life span in both humans and many model systems, including flies, mice, rats, and primates. These resources have demonstrated profound age-associated changes including altered transcriptional profiles, changes in cell type composition, and the emergence of novel populations of cells (Enge et al. 2017; Angelidis et al. 2019; Dulken et al. 2019; Hernando-Herraez et al. 2019; Kimmel et al. 2019; Ma et al. 2020; Tabula Muris Consortium 2020; Wang et al. 2020; Zhang et al. 2022; Lu et al. 2023b; Nikopoulou et al. 2023). Together, these findings raise fundamental questions concerning how cellular plasticity varies with age, and whether and how this contributes to age-associated pathologies.

Cell identity has previously been defined as a combination of cell fate or cell type (i.e., a macrophage, adipocyte, or hematopoietic stem cell) and cell state (i.e., a myeloid-biased, senescent, or dedifferentiated cell) (Kimmel et al. 2019; Traxler et al. 2023). Some of the cellular shifts observed during aging have been previously characterized and reviewed elsewhere, including the emergence of proinflammatory cells (Ferrucci and Fabbri 2018; Franceschi et al. 2018), fibrotic phenotypes (Selman and Pardo 2021), and senescent cells (Olan and Narita 2022; Ring et al. 2022). However, changes in cell identity as they relate to differentiation potential remain incompletely explored. Hence, our review focuses on four specific shifts in cell identity that can be classified as age-associated programs linked to (1) dedifferentiation, (2) biased differentiation, (3) activation of alternative lineage programs, and (4) entry into a preneoplastic state (Fig. 1). Importantly, this review covers both cell-intrinsic and cell-extrinsic mechanisms that drive these changes. Finally, we conclude by reviewing and critically discussing how cellular plasticity is currently being leveraged to induce rejuvenation using transcription factor- and epigenetic modifier-based approaches in aged cells.

Figure 1.

Schematic describing different shifts in cellular identity that occur with age, including programs linked to dedifferentiation, biased differentiation, acquisition of features of alternative lineages, and entry into a preneoplastic state. Thin black arrows denote normal differentiation, and thick gray arrows indicate age-related changes. Below each shift are representative examples of changes observed in vivo that are described in further detail in the text. (MuSCs) Muscle stem cells, (CHIP) clonal hematopoiesis of indeterminate potential, (ISCs) intestinal stem cells, (LT-HSCs) long-term hematopoietic stem cells, (FAPs) fibroadipogenic progenitors, (BMSCs) bone marrow stromal cells, (NSCs) neural stem cells.

Programs linked to dedifferentiation

Dedifferentiation is defined in the context of this review as molecular, cellular, and morphological changes that together cause a functional shift toward a more immature identity within the same cellular lineage. Functional dedifferentiation has become a well-studied phenomenon both in malignant transformation and in response to tissue injury. In the case of the former, it is appreciated that certain cancers acquire features of stem or progenitor cells to facilitate transformation (discussed in more detail in “Preneoplastic State”) (Ben-Porath et al. 2008). In the context of injury, more committed cells such as progenitor cells or terminally differentiated cells typically reacquire stem cell potential to repair or replace damaged tissue (Tata and Rajagopal 2016). Aging can be conceptualized as the accumulation of cellular damage over time (Gladyshev et al. 2021). As such, a key question in the field is whether dedifferentiation could represent a physiologic response to aging. In support of this notion, transcriptional changes consistent with a less differentiated state have been observed in a variety of aging systems, including pancreatic β cells, enterocytes from flies, and old long-term murine HSCs (LT-HSCs) (Talchai et al. 2012; Kowalczyk et al. 2015; Flint Brodsly et al. 2019). Importantly, in each of these cases, the aged cell types are associated with functional deterioration when compared with their young counterparts, suggesting that these transcriptional programs are maladaptive as opposed to reparative (Ben-Porath et al. 2008).

Using a tractable in vitro system to evaluate dedifferentiation in the context of aging, Mertens et al. (2021) isolated fibroblasts from patients with sporadic late-onset Alzheimer's disease (LOAD), a highly age-associated condition, and transdifferentiated these into induced GABAergic and glutamatergic neurons (iNs) with defined transcription factors (Pang et al. 2011b). Compared with iNs derived from healthy age-matched controls, iNs from patients with LOAD exhibited several molecular and cellular changes consistent with a dedifferentiated state. For example, LOAD iNs downregulated genes involved in mature neuronal differentiation as well as upregulated genes involved in neuronal precursor cell states and stem cell signaling. Moreover, these iNs were less functional, less differentiated morphologically, and showed signs of a nonneuronal metabolic state (Mertens et al. 2021). More recent work has suggested that alternative splicing of the key metabolic enzyme PKM underlies many of these pathological changes. Specifically, PKM is spliced into the PKM2 isoform in LOAD iNs, which not only promotes the nonneuronal metabolic state of glycolysis in the cytoplasm but also enters the nucleus to interact with transcription factors STAT3 and HIF1α to drive the transcription of genes involved in neuronal fate instability. Moreover, the chemical inhibition of PKM2 in iNs derived from LOAD patients shifts their transcriptional profile closer to fully differentiated iN samples derived from healthy age-matched controls (Traxler et al. 2022). Interestingly, many of the transcriptional and metabolic changes observed in the iN in vitro model have been corroborated in postmortem brain tissue from subjects with LOAD (Mertens et al. 2021). However, whether the molecular, cellular, and morphological resemblance to a more immature neuronal state results in functionally dedifferentiated cells that, for example, can redifferentiate into other neuronal subtypes (i.e., motor or dopaminergic neurons) has not been observed in vitro or in vivo. A comprehensive review on the topic of neuronal fate instability within the context of neurodegeneration has been covered elsewhere, and it is important to note that, to our knowledge, this phenomenon has not yet been observed in healthy neuronal aging (Mertens et al. 2015; Huh et al. 2016; Traxler et al. 2023).

A closer look at the etiology of cells acquiring a dedifferentiated profile in vivo is similarly based on the upregulation of programs used in development, including embryogenesis. For example, insulin-secreting pancreatic β cells are known to lose nuclear expression of the β-cell identity transcription factor FoxO1 with age (Song et al. 2022). In a mouse model of FoxO1 deficiency in β cells, aged, but not young, animals accumulated pancreatic cells that lacked expression of key differentiated β-cell markers such as insulin but instead upregulated markers associated with endocrine progenitor cells, including Neurogenin-3. Interestingly, using lineage tracing, the investigators were able to confirm that the source of these phenotypically dedifferentiated pancreatic cells in aging FoxO1-deficient mice were originally insulin-producing mature β cells. Importantly, as it relates to function, these cells were then able to redifferentiate into atypical hormone-producing cells, secreting glucagon, which is typically produced from a different mature endocrine cell type termed α cell (Talchai et al. 2012). It is hypothesized that FoxO1 maintains β-cell identity by repressing factors associated with endocrine progenitor cells, including Neurogenin-3 (Kitamura 2013).

To build on these findings in a nontransgenic system, researchers have analyzed unperturbed aging human pancreatic tissue. Enge et al. (2017) evaluated pancreata from humans across life span. Although they did not comment on the appearance of dedifferentiated cells, they did observe increasing levels of mature endocrine cells that expressed both insulin and glucagon, which could potentially be the redifferentiated progeny of dedifferentiated cells (Enge et al. 2017). More recently, Song et al. (2022) evaluated human pancreas samples with immunofluorescence and observed age-related increases in endocrine cells that lost expression of both differentiation markers of hormone secretion (i.e., insulin) and key β-cell identity factors, like FoxO1 and Ucn3, suggestive of a molecularly dedifferentiated state. Interestingly, these findings have also been observed in islets isolated from subjects with type 2 diabetes (T2D) as well as mouse models of the same condition, which is notable given that age is a risk factor for T2D development (Talchai et al. 2012; Cinti et al. 2016). Intriguingly, at least in mouse models of T2D, insulin therapy allowed for redifferentiation of dedifferentiated pancreatic β cells (Wang et al. 2014). However, it will be important to further characterize aberrant human β cells that have lost expression of mature markers to see whether they are analogous to the more plastic dedifferentiated population observed in the FoxO1-deficient mice. Similar to the LOAD iN system, without demonstrating the ability of these aberrant pancreatic cells to redifferentiate, it is difficult to determine whether a functionally dedifferentiated state is attained during aging of the pancreas.

The upregulation of transcriptional programs associated with embryogenesis and development has also been observed in bulk analyses of aging tissues and cells. Data sets examining changes in gene expression across development and aging have noted trends of “reversion,” whereby transcriptional modules normally upregulated during development (i.e., differentiation-associated programs) are downregulated during aging and vice versa. This phenomenon was initially observed in the prefrontal cortex of humans and macaques as well as in mouse livers (Loerch et al. 2008; Somel et al. 2010; Colantuoni et al. 2011; Dönertaş et al. 2017; Anisimova et al. 2020). Building off these findings, Izgi et al. (2022) sought to examine whether this trend occurred globally across a variety of mouse tissues. Specifically, the investigators analyzed the cerebral cortex, lung, liver, and muscle using bulk RNA-seq over the course of development (0–2 months) and aging (3–30 months). A subtle transcriptional reversion trend was observed in the liver and muscle (Izgi et al. 2022). Given this study was only done with a limited number of mice over a wide range of ages using bulk analysis, these provocative findings certainly warrant further examination. It will also be important to assess whether the isolated examples of molecular dedifferentiation described above represent a broader phenomenon across tissues.

In addition to changes to the transcriptome, changes in DNA methylation at key developmental genes have been observed with age (Rakyan et al. 2010; Teschendorff et al. 2010). For instance, a study investigating age-specific changes in DNA methylation across 185 mammalian species found that CpGs near developmental genes changed in an age-dependent manner. Interestingly, many of these CpGs associated with developmental genes are the basis of biological age predictors, termed epigenetic clocks, which use the methylation status of a small subset of CpGs to accurately predict chronological age (Box 1; Moqri et al. 2022; Lu et al. 2023a).

Box 1. DNA methylation clocks.

DNA methylation clocks, or epigenetic clocks, are based on the finding that methylation at certain CpG sites change as a function of age across tissues and species (Horvath and Raj 2018). Building on this observation, Steve Horvath (Horvath 2013) used DNA methylation sequencing data sets across life span to develop penalized regression models that predict “DNA methylation age” or “epigenetic age,” which is highly correlated with a sample's chronological age. Specifically, these models select a subset of age-associated CpGs identified through training data sets and then predict a novel sample's DNA methylation age by assessing those same sites (Horvath 2013). In addition to Horvath’s (2013) original pan-tissue human clock, clocks have been generated for specific organs and model systems (Petkovich et al. 2017; Meer et al. 2018; Coninx et al. 2020). Moreover, so-called universal clocks that are agnostic to tissue or mammalian species have been developed (Lu et al. 2023a). This latter category of clocks suggests that there is a conserved feature of age-related changes in DNA methylation across species. Critically, these clocks can also register interventions that are known to increase longevity (heterochronic parabiosis and calorie restriction) as well as those known to accelerate the aging process (progerias) (Petkovich et al. 2017; Horvath et al. 2018; Zhang et al. 2023).

Of note, a subset of CpG sites consistently enriched in DNA methylation clocks was found to be targets of the Polycomb repressive complex 2 (PRC2), which is essential for development and developmental gene expression across species (Loh and Veenstra 2022). In fact, accurate DNA methylation clocks have been built solely based on methylation changes at PRC2 target sites (Moqri et al. 2022). This finding is notable in light of emerging research suggesting that developmental pathways are aberrantly regulated with aging (Izgi et al. 2022; Liu et al. 2022a). However, whether these changes in DNA methylation at developmental genes have any functional consequences at the level of gene expression and cell identity or are simply stochastic noise remains an open question in the field. We direct interested readers to a recent discussion on this topic (Bell 2024).

Complementing these global analyses, aberrant activation of specific genes involved in embryogenesis and development has been observed in aging model systems. For example, in aged human fibroblasts and blood samples as well as in senescent human mesenchymal progenitor cells (hMPCs), the expression of the pregnancy-specific β-1-glycoprotein (PSG) family is reportedly upregulated due to broad changes in chromatin structure. The PSG family of genes is specifically expressed in the syncytiotrophoblasts of the placenta during pregnancy. Ectopic expression of PSG4 in young hMPCs, where the gene is typically silenced, triggers certain hallmarks of aging such as increased senescence and DNA damage. Detailed chromatin analyses of these cells suggests that the misexpression of genes, including PSG4, is the result of a changing epigenetic landscape, including the derepression of heterochromatin and an elevated degree of repression in typically euchromatic regions (Liu et al. 2022a). Similar observations have been reported for other genes, such as the Hox gene family, using different aging systems (Schwörer et al. 2016; Yang et al. 2023). This topic has been extensively reviewed by Liu et al. (2022b).

Although the work done in accelerated models of aging as well as some of the global analyses implicate the dysregulation of embryonic and developmental pathways in aging, robust signs of dedifferentiation beyond the dysregulation of small groups of marker genes have yet to be observed even with the growing number of single-cell data sets of aging tissues (Enge et al. 2017; Angelidis et al. 2019; Dulken et al. 2019; Hernando-Herraez et al. 2019; Kimmel et al. 2019; Ma et al. 2020; Tabula Muris Consortium 2020; Wang et al. 2020; Zhang et al. 2022; Lu et al. 2023b; Nikopoulou et al. 2023). Therefore, it remains unclear whether the widespread reactivation of these pathways results in a bona fide less differentiated cell state or simply reflects the emergence of dysfunctional cells harboring age-dependent molecular defects; for example, the loss of heterochromatin maintenance or transcriptional fidelity. It is therefore our view that additional work is required to demonstrate that this phenomenon is (1) widespread and (2) leads to functional changes in physiological aging in order for it to be considered a robust shift in cellular identity.

Biased differentiation

Tissue-specific stem and progenitor cell hierarchies established during development are critical for maintaining tissue homeostasis in the adult. The continuous production of highly differentiated cell types that maintain complex regenerative tissues relies on a balanced differentiation from monopotent, bipotent, or multipotent stem and progenitor cells. However, with age, these differentiation hierarchies often break down, resulting in biased differentiation that can have profound consequences for organ function.

Perhaps the best-characterized example of age-dependent biased differentiation is found in the hematopoietic system, where there is a progressive accumulation of mature myeloid cells (monocytes and granulocytes) and a concomitant decrease of mature lymphoid cells (B and T cells) with age (Young et al. 2021). These changes in differentiation can have wide-ranging consequences including the age-associated increase in susceptibility to infections, loss of response to vaccines, and increased risk of myeloid malignancies (Leins et al. 2018; Mejia-Ramirez and Florian 2020). Interestingly, the biased production of myeloid and lymphoid cells has been traced back to the long-term hematopoietic stem cell (LT-HSC) population within the bone marrow, which was long known to become more myeloid-biased with age. This conclusion is based on the observation that LT-HSCs from old mice express elevated levels of myeloid genes and when transplanted into recipient mice preferentially engraft into the myeloid lineage (Rossi et al. 2005; Dykstra et al. 2011).

Importantly, biased differentiation within the hematopoietic system has been observed in both mice and humans, which motivated studies to define the underlying mechanisms with the goal to improve immune function (Pang et al. 2011a). Multiple studies over the past decade indicate that biased differentiation in the blood system can be attributed to both cell-intrinsic factors, including LT-HSC proliferative history (Beerman et al. 2013; Kirschner et al. 2017), replicative stress (Beerman et al. 2013; Flach et al. 2014), mitochondrial metabolic dysfunction (Tothova et al. 2007; Norddahl et al. 2011; Mohrin et al. 2015; Luchsinger et al. 2016; Ho et al. 2017; Luo et al. 2019; Sun et al. 2021; Girotra et al. 2023), mTOR activation (Chen et al. 2009), loss of cellular polarity (Florian et al. 2012), noncanonical Wnt signaling (Florian et al. 2013), and epigenetic drift (Rimmelé et al. 2014; Sun et al. 2014; Adelman et al. 2019; Xu et al. 2020), and cell-extrinsic factors such as the loss of bone marrow innervation (Maryanovich et al. 2018; Ho et al. 2019), age-associated changes in endothelial cells (Poulos et al. 2017), increases in inflammatory cytokines (Ergen et al. 2012; Pietras et al. 2016; Frisch et al. 2019; Pioli et al. 2019; Yamashita and Passegué 2019; Mitchell et al. 2023), and mesenchymal-derived growth factors (Guidi et al. 2017; Young et al. 2021). Most recently, aldehyde-induced DNA damage (Wang et al. 2023b), noncanonical activity of RNA-modifying enzymes (Wang et al. 2023c; He et al. 2024), loss of niche-derived Netrin-1 (Ramalingam et al. 2023), and the microbiome of aged mice (Zeng et al. 2023) have all been suggested to contribute to this bias as well. Whether these diverse factors are related to each other and how they contribute to these shifts in the LT-HSC compartment are only beginning to be understood. For example, the microbiome found in the aged, and therefore more permeable, intestine contributes to elevated levels of inflammatory cytokines such as IL-1β and TNF in the bone marrow niche (Pietras et al. 2016; Yamashita and Passegué 2019; Mitchell et al. 2023; Zeng et al. 2023). Similarly, both aldehyde-induced DNA damage and LT-HSC replicative history are known to converge on elevated levels of p53 signaling, which then drives a myeloid bias (Kirschner et al. 2017; Wang et al. 2023b). Continuing to generate a unified understanding of all these seemingly disparate features will be critical.

It is important to note that the majority of studies investigating the aging LT-HSC compartment relied on bulk transcriptional analysis or the evaluation of peripheral blood to infer myeloid skewing of LT-HSCs. However, these analyses were unable to address some fundamental questions regarding the origin of this population with age: (1) Do myeloid-biased HSCs represent a cell state that is acquired with age? (2) Does aging equally affect all LT-HSCs or only a subpopulation? (3) Is the perceived overabundance of myeloid-biased LT-HSCs simply due to a loss of LT-HSCs with a lymphoid or pan-lineage bias? Recently, single-cell RNA sequencing as well as single-cell transplantation experiments have provided some insights into these questions.

One feature that is becoming increasingly clear from single-cell studies is the large extent of heterogeneity within the aging LT-HSC compartment. Notably, there is a mix of both myeloid-biased LT-HSCs and those with pan-lineage potential (Kirschner et al. 2017; Mann et al. 2018; Yamamoto et al. 2018; Mansell et al. 2021; Kim et al. 2022). Interestingly, myeloid-biased LT-HSCs are also found in the young LT-HSC compartment. What differ between young and old mice are the proportions of these HSC populations, with myeloid-biased LT-HSCs making up the majority of the aged compartment and a minority of the young. These studies therefore imply that the myeloid-biased LT-HSC population expands with age. The selective expansion of these cells is further exacerbated by the age-related loss of pan-lineage LT-HSCs that equally repopulate the myeloid and lymphoid lineages. Recent work by Young et al. (2021) suggested that the age-related loss of stromal IGF1 results in the subsequent depletion of lymphoid-biased HSCs, thus partially explaining the myeloid bias. Together, this body of work suggests that the myeloid shift results from both an expansion of myeloid-biased HSCs and a depletion of HSCs with a lymphoid bias.

In addition to these changes in cell type composition with age, work conducted by Yamamoto et al. (2013, 2018) discovered a shift in cellular plasticity within myeloid-biased HSCs that arises with age. Using single-cell serial transplantation from HSCs derived from aged mice, the investigators demonstrated that a subset of HSCs with myeloid restriction in a primary transplant acquired pan-lineage potential in a secondary transplant. Importantly, this population of so-called “latent HSCs” appears to be absent in young mice subjected to the same serial transplantation scheme, even though these mice also had myeloid-restricted HSCs present in primary transplants (Yamamoto et al. 2013, 2018). In support of these observations, an independent study observed that the transplantation of HSCs enriched for myeloid differentiation potential into recipients results in defective lymphoid repopulation in a primary transplant, though this defect was lost upon secondary transplantation (Kim et al. 2022). It will be quite interesting to understand whether the mechanisms that allow a subset of myeloid-biased LT-HSCs to regain multipotent potential with age are due to cell-intrinsic factors, such as age-dependent or transplantation-induced epigenetic changes, or cell-extrinsic factors, such as the prolonged exposure to a youthful microenvironment including stromal IGF1. Moreover, it should be informative to further examine the mechanistic effects of systemic interventions that are known to rebalance the myeloid bias, such as heterochronic parabiosis and treatment with senolytics and rapamycin (Chen et al. 2009; Chang et al. 2016; Ma et al. 2022). Addressing these questions is especially important in light of recent efforts to target myeloid-biased HSCs with cytolytic antibody cocktails in mice, which reportedly led to a partial restoration of lymphopoiesis as well as an improved response to vaccine and viral infection (Ross et al. 2024).

Critically, HSCs are not the only stem cell type that undergoes shifts in differentiation potential with age. Fibroadipogenic progenitors found in muscle favor fibroblast differentiation with age, potentially contributing to the age-associated increase in muscle fibrosis (Lukjanenko et al. 2019). Bone marrow stromal cells, which give rise to osteoblasts and adipocytes, produce more adipocytes with age and thus contribute to the increasingly inflammatory environment in the bone marrow niche (Aaron et al. 2022). Lgr5⁺ intestinal stem cells produce elevated numbers of secretory cells, specifically Paneth and goblet cells, with age. This defect has been linked to impaired Wnt and Notch signaling as well as metabolic reprogramming in old ISCs. Notably, this shift can be partially reversed by treatment with metformin (Nalapareddy et al. 2017; Choi et al. 2023). Last, neural stem cells, which typically give rise to neurons and astrocytes, turn on a transcriptional program related to astrocyte development at the expense of neurogenesis, potentially playing a role in age-associated cognitive decline (White et al. 2020).

To further dissect the mechanisms that drive the observed biases in cell states in solid tissues, it should be informative to apply some of the approaches from the hematopoietic system. For example, it will be important to better define the heterogeneity of diverse solid tissue stem cells and determine whether aging entails a selective expansion of stem cell subpopulations akin to the hematopoietic system, a shift in cell state among all stem cells, or a combination of both. Understanding these nuances in cellular biases and how they relate to cell state will be critical for designing strategies to counter them.

Activation of alternative lineage programs

Age-associated shifts in cell type composition, such as increased immune cell and fibrotic infiltration, contribute to diseases of aging, raising the question of where these shifts arise from (Wansleeben et al. 2014; Angelidis et al. 2019; Dulken et al. 2019; Tabula Muris Consortium 2020; Isola et al. 2024). One possibility is that these cell populations are deposited via cellular migration or the expansion of existing cell populations within tissues. Another possibility is that certain resident cells adopt alternative fates with age, which we explore below.

Seminal work by Brack et al. (2007) ∼15 years ago demonstrated that aged, but not young, mononucleated myogenic cells undergo conversion to a fibrotic cell state when forced to proliferate ex vivo. This shift in cell state was defined by both morphological criteria and the expression of fibroblast-associated markers. The underlying process appears to be cell-extrinsic, as it was attributed to systemic factors driving Wnt–TGFβ2 signaling as well as changes within the aged muscle extracellular matrix (Brack et al. 2007; Biressi et al. 2014; Stearns-Reider et al. 2017). Importantly, these findings are consistent with the age-associated increases in muscle fibrosis and defective regeneration of muscle stem cells in response to injury. Of note, the fibrotic phenotype can be somewhat reversed with systemic interventions including heterochronic parabiosis, suggesting that this aberrant cell fate change may be targetable (Brack et al. 2007).

Other examples of cells assuming molecular features of alternative fates with age include dermal fibroblasts turning on markers of adipogenic cells, cardiac fibroblasts expressing osteogenic markers, tracheal stem cells forming structures resembling submucosal glands, and glomerular parietal epithelial cells of the kidney activating a mesenchymal profile. Critically, each of these shifts has been linked to age-related defects, including reductions in dermal thickness, calcification of the heart, squamous cell carcinomas, and glomerulosclerosis, respectively (Roeder et al. 2015; Salzer et al. 2018; Vidal et al. 2019; Aros et al. 2020). However, as these studies have remained largely descriptive, it will be critical to understand how meaningful these purported shifts in cell fate are by conducting genetic lineage tracing as well as functional experiments to probe the robustness of the observed transcriptional changes as well as their causal role in aging.

Mechanistically, the aforementioned studies point to transcriptional noise or transcriptional heterogeneity as drivers of altered cell fate. Specifically, aged cells expressing alternative lineage genes exhibited increased levels of transcriptional entropy or increased cell-to-cell transcriptional variability (Enge et al. 2017; Salzer et al. 2018; Vidal et al. 2019). For example, in studies evaluating shifts in fate in aged fibroblast populations toward adipogenic or osteogenic profiles, genes involved in fibroblast identity were more variably and lowly expressed on average in old fibroblasts than in young fibroblasts (Salzer et al. 2018; Vidal et al. 2019). Consistent with these observations, transcriptional initiation and elongation have recently been recognized as mechanisms commonly disrupted with age across a variety of species (Gyenis et al. 2023; Papadakis et al. 2023; Sen et al. 2023). Accordingly, disruption of H3K36 methylation, which maintains transcriptional fidelity in cells, has been causally linked to life span extension in yeast and worms (Sen et al. 2016). Of relevance, multiple groups have demonstrated that H3K36 methylation is critical for the maintenance of mammalian cell identity, suggesting that age-dependent changes in H3K36 methylation and transcriptional fidelity may lead to changes in cell state (Hoetker et al. 2023; Kovatcheva et al. 2023). Interestingly, the adipogenic shift observed in aged dermal fibroblasts is alleviated by calorie restriction (Salzer et al. 2018). Given that caloric restriction is known to modulate the epigenome, it will be intriguing to test whether there is a link between H3K36 methylation, calorie restriction, and cell identity maintenance (Zhai et al. 2023).

Although studies focused on the activation of lineage-inappropriate genes have been lacking in physiological models of aging, several recent publications have explored this question using models of accelerated aging. For example, mouse models with shortened telomeres, including telomerase-deficient mice and mice with a defective telomere Shelterin complex, phenocopy the epithelial-to-mesenchymal transition associated with aged kidneys. Glomerular epithelial cells with shortened telomeres upregulated key mesenchymal transcription factors, including Twist, Snail, and Zeb1. Importantly, epithelial cells isolated from the kidneys of telomerase-deficient mice were able to revert back and undergo a mesenchymal-to-epithelial transition when telomerase was overexpressed (Saraswati et al. 2021). However, it remains unknown whether the telomere shortening associated with aging drives this process under homeostatic conditions.

In a separate, provocative model of accelerated aging, Yang et al. (2023) recently developed the “inducible changes to the epigenome” (ICE) system, where double-stranded breaks are transiently introduced in largely noncoding regions of the genome. This approach is based on the hypothesis that the relocalization of chromatin factors to the sites of DNA breaks away from regions that are critical to cell identity and fitness-phenocopies certain aspects of aging (Lu et al. 2023c). Fittingly, these mice exhibit molecular, cellular, and physiologic hallmarks of aging, and of relevance to this section, ICE mice exhibit the epithelial-to-mesenchymal transition of glomerular epithelial cells seen in aged mice (Roeder et al. 2015). Interestingly, upon detailed mechanistic examination of this model, fibroblasts derived from ICE mice upregulated genes involved in neurogenesis and lost repressive histone marks at associated promoters. When ICE fibroblasts were reprogramed into iNs using defined neurogenic transcription factors, the resultant cultures derepressed neuronal genes and gave rise to neurons more efficiently compared with controls. When Yang et al. (2023) explored the conservation of these observations in vivo, they found that the muscle of ICE mice appeared to have lost activating marks at genes associated with myocyte identity, whereas they seemingly have gained activating marks at genes critical for immune cells. Although this finding points to a possible change in cell fate, it will be important to rule out that age-related changes in cell type composition or an infiltration of immune cells accounted for these molecular changes rather than muscle cells adopting features of immune cells (Tobin et al. 2021). This is particularly important in light of recent commentary about the ICE model, which calls into question the extent of genotoxic stress induced in these mice as well as subsequent cellular death and inflammation, which could drive some of the observed phenotypes (Timmons and Brenner 2024; Yang et al. 2024). Thus, although these experiments suggest that age-related changes to the epigenome can drive the expression of alternative lineage programs, moving forward, it will be critical to confirm the epigenetic and cellular changes described in this study using physiologically aged mice.

Epigenetic changes leading to the misexpression of cell fate regulators have also been implicated in other models of aging, including senescence and laminopathies. For instance, in senescent human fibroblasts, genes involved in stratum corneum differentiation (LCE2 skin genes) and innate immune signaling (the NLRP3 gene, encoding a subunit of the inflammasome), among others, become aberrantly expressed. The mechanisms of derepression vary but include chromatin decompaction, p53 and C/EBPβ signaling, and disruption of topologically associated domains. Notably, in the case of the LCE2 skin genes, the epigenetic mechanisms responsible for re-expression were distinct from how LCE2 is typically regulated during homeostatic differentiation of the stratum corneum, highlighting the abnormal nature of this senescence-associated program (Tomimatsu et al. 2022).

Last, laminopathies caused by mutations in nuclear lamins can phenocopy certain aspects of aging on a molecular and functional level. Molecular changes include the disruption of lamin-associated domains, leading to the loss of peripheral heterochromatin; physiological changes include the development of osteoporosis, atherosclerosis, and myotonia. Shah et al. (2021) investigated why a particular subset of laminopathies appeared to selectively affect certain organs, specifically the heart. To do so, they generated human induced pluripotent stem cell (hiPSC)-derived cardiomyocytes, hepatocytes, and adipocytes carrying a specific mutation known to primarily affect the heart despite its ubiquitous expression across the body. They observed that genes associated with alternative, nonmyocyte lineages lost contact with lamins and were subsequently re-expressed in mutant hiPSC-derived cardiomyocytes. Accordingly, LMNA mutant cardiomyocytes were less functional than controls. This was not the case for the hiPSC-derived mutant hepatocytes and adipocytes, which appeared to maintain their cellular identity, and fittingly, these organs exhibit minimal pathology in patients. Importantly, on a transcriptional level, nonmyocyte genes were also found to be upregulated in myocardium biopsied from patients with LMNA mutations, suggesting conservation of this phenotype in humans (Shah et al. 2021).

Interestingly, some of this work suggests that aged cells become more plastic with age, which stands in contrast to a growing body of work demonstrating that aged cells are less efficient at undergoing iPSC reprogramming (Box 2; Li et al. 2009; Lapasset et al. 2011; Ravaioli et al. 2018; Grigorash et al. 2023; Yang et al. 2023). It will therefore be critical to functionally characterize these “confused” cells with robust assays to understand whether they simply express markers of different lineages due to transcriptional or epigenetic noise but without functional consequences or in fact undergo a real change in cell identity and contribute to age-associated pathologies. Although caution should be exercised when extending the findings from these accelerated models of aging to physiological aging, these systems do provide helpful information on relevant mechanisms worth further exploration. Specifically, it will be important to continue to dissect the mechanisms underlying the increased levels of transcriptional and epigenetic noise that likely drive the aberrant expression of lineage-inappropriate markers.

Box 2. Are aged cells more plastic? Lessons from iPSC reprogramming.

The finding that fibroblasts from prematurely aged ICE mice are more plastic as determined by their more efficient reprogramming into induced neurons stands in contrast to a body of work that suggests that aged cells are less efficient at reprogramming into iPSCs (Li et al. 2009; Lapasset et al. 2011; Ravaioli et al. 2018; Grigorash et al. 2023; Yang et al. 2023). However, Mahmoudi et al. (2019a) recently examined the impact of aging on iPSC reprogramming more systematically by evaluating the reprogramming potential of 108 mouse fibroblast samples derived from young, middle-aged, and old animals. Of note, although the mean reprogramming efficiency did not change across age groups, the amount of variability between samples increased dramatically, with some aged fibroblasts undergoing more efficient reprogramming than their young counterparts, and others being substantially worse. This finding suggests that stochastic aging trajectories of fibroblasts may dictate reprogramming efficiency (Mahmoudi et al. 2019a). Indeed, single-cell transcriptional and secretomics analyses demonstrated that the proportion of activated fibroblasts as well as the inflammatory cytokines they produced (e.g., IL-6) correlated well with reprogramming efficiency (Brady et al. 2013; Mosteiro et al. 2016; Mahmoudi et al. 2019a). These results are in line with a growing body of evidence showing that aging is accompanied by an increasing degree of cellular and molecular heterogeneity (Enge et al. 2017; Salzer et al. 2018; Angelidis et al. 2019; Hernando-Herraez et al. 2019; Kimmel et al. 2019). Therefore, although bulk cultures overall may lose plasticity, rare cells within these cultures may become more plastic, and this in turn may facilitate pathologic shifts in cellular identity. However, additional work is certainly needed to understand at a single-cell level how increasing levels of heterogeneity are linked to cellular plasticity.

Preneoplastic state

Aging is considered one of the most important risk factors for cancer, with individuals >70 years old accounting for approximately half its mortality burden (Editorial 2022). This age-dependent component of cancer suggests that cells undergo changes with age that make them more amenable to malignant transformation. Consistently, many of the hallmarks of cancer overlap with the hallmarks of aging, including increased levels of DNA damage, changes to the epigenome, dysregulated interactions with the microenvironment, and most recently the acquisition of cellular plasticity (Dos Santos et al. 2023; López-Otín et al. 2023). It is therefore of interest to better understand how these cell-intrinsic and cell-extrinsic forces experienced with age predispose cells to malignant transformation.

It is well established that, with age, cells acquire genetic changes including somatic mutations, copy number variation, and chromosomal abnormalities (Schumacher et al. 2021). Population-wide sequencing studies have determined that a subset of these mutations, oftentimes in cancer-related genes, leads to clonal outgrowths among stem and progenitor cells. Although in some cases these clones can be beneficial by conferring resistance to disease or injury, in many cases these clonally expanded populations of cells are at a higher risk of converting into frank malignancies (Zhu et al. 2019a; Wang et al. 2023d). This phenomenon has been observed in multiple tissues but has been mostly studied in the hematopoietic system in the form of clonal hematopoiesis of indeterminate potential (CHIP) (Genovese et al. 2014; Jaiswal et al. 2014; Xie et al. 2014; Marongiu et al. 2023). In CHIP, the age-associated accumulation of mutations in epigenetic modifiers, proteins involved in growth signaling, and the DNA damage response pathway in HSCs leads to clonal outgrowth. The prevalence of CHIP increases with age and is thought to be 10%–20% in subjects >70 years old but can be much higher depending on the sequencing method used (Jaiswal et al. 2014). Although subjects with CHIP are not considered to actively carry a blood malignancy, some studies estimate that CHIP is associated with an ∼10-fold increased relative risk of developing cancer (Genovese et al. 2014; Jaiswal et al. 2014). However, it is important to note that the rate of clonal outgrowths as well as the risk of transformation are variable, dependent on both the particular genetic mutation and cell-extrinsic factors including the surrounding microenvironment (Jaiswal and Ebert 2019; Zhang et al. 2019). Although this topic has been extensively reviewed elsewhere, it remains a striking example of how an age-associated process—in this case, the accrual of mutations—can lead cells to enter a preneoplastic state (Jaiswal and Ebert 2019; Kapadia and Goodell 2024).

In addition to genetic changes, the epigenome is known to undergo similar rewiring in aging and cancer progression (López-Otín et al. 2023). Most notably, similarities in global methylation patterns have been observed between naturally aged, senescent, and cancerous cells. Specifically, in all of these cell types, hypomethylation is observed in regions that are gene-poor, are late-replicating, and have low CpG density, whereas hypermethylation is noted at CpG islands in promoter regions for genes involved in development and tumor suppression (Rakyan et al. 2010; Teschendorff et al. 2010; Cruickshanks et al. 2013; Zabransky et al. 2022). In fact, noncancerous gastric epithelia from aged subjects were noted to have hypermethylation at various tumor suppressor genes (So et al. 2006). However, this work is largely descriptive and prompts the question of whether the age-associated changes in the epigenome precede and therefore promote malignant transformation.

Recent work conducted by Tao et al. (2019) investigated this question in the context of BRAF^V600E-induced tumorigenesis in colon adenocarcinomas (COADs). Using organoids derived from young mice, they observed that after introducing a BRAF^V600E mutation, 5 months of additional culturing was required before cells became tumorigenic. During this time, the organoids underwent aberrant hypermethylation at the promoters of various cancer-related genes including negative regulators of Wnt signaling (Tao et al. 2019). Importantly, this methylation pattern is a prominent feature of COADs (Issa 2004). Given that this type of cancer is known to occur in older individuals, the investigators examined whether the long-term culturing of organoids would phenocopy natural aging and cause changes in DNA methylation that facilitate malignant transformation. Indeed, the introduction of BRAF^V600E into organoids that had been passaged for >1 year led to their rapid transformation within 2 weeks (Tao et al. 2019). Although independent work suggested that the same methylation changes are observed in aged, but otherwise healthy, intestinal tissue, it will be important to demonstrate this finding in the context of aged organoids as opposed to those cultured long term (Belshaw et al. 2010). Moreover, the long-term culturing of organoids may select for growth-promoting mutations, and it will therefore be critical to understand the role that genetic mutations play during tumorigenesis in this context, either indirectly by modulating the epigenome or directly by activating oncogenes or disrupting tumor suppressor genes (Zhao et al. 2022). Importantly, in this case, whole-exome sequencing did not identify any known cancer mutations in organoids cultured for >1 year (Tao et al. 2019). As such, this result nicely exemplifies how age-related changes rewire cells into a state that is more susceptible to malignant transformation. It is important to note here that other groups have hypothesized the opposite—that age-associated epigenetic changes may restrain rather than promote cancer—but this work largely remains hypothetical and untested (Johnstone et al. 2022).

In addition to genetic and epigenetic forces, the aging microenvironment can also allow for the outgrowth of malignant cells. Recent work has examined this idea in the context of basal cell carcinoma (BCC), a type of cancer where age is a primary risk factor. Using a transgenic model expressing a mutant form of Smoothened (SmoM2) in epithelial cells, the investigators demonstrated that BCC growth is enabled in ear epidermis but inhibited in back skin epidermis of young mice (Bansaccal et al. 2023). However, when this mutation was induced in aged mice, the mutant back skin cells behaved like ear epidermis cells and underwent reprogramming to an embryonic hair follicle progenitor (EHFP) state that led to tumor initiation. Mechanistic analyses revealed that reprogramming to EHFPs is facilitated in the aged context because the aged dermis is less dense, as the collagen I matrix breaks down with age. Fittingly, EHFP and tumor initiation could be recapitulated in young mice treated with collagenase or UVA exposure, which is known to induce cellular damage in the dermis via DNA breaks, thereby decreasing collagen production.

Recently, “phenotypic plasticity” has been added as a new hallmark of cancer (Hanahan 2022). This is based on the observation that cancer cells acquire the ability to downregulate markers of their tissue of origin and upregulate markers associated with stemness or other tissues to promote tumorigenesis (Köhler et al. 2017; Pastushenko et al. 2021). Interestingly, a recent meta-analysis of aging human healthy tissues and cancers found that 40% of aging tissues downregulated tissue-specific markers with age, a feature that was shared across the majority of examined tumors (Dos Santos et al. 2023). This observation raises the relevant question of whether these “confused” aged cells are the cells of origin of certain age-related cancers and thus represent potential targets for treatment.

Some of the initial work to address this question was conducted in Drosophila. Single-nucleus sequencing of the Drosophila midgut highlighted an age-associated increase in intestinal stem cells (ISCs) and their direct progenitors, enteroblasts (EBs), coupled with a decrease in differentiated enterocytes, suggestive of a dysplastic phenotype (Liu et al. 2022a). This aberrant accumulation of ISCs and EBs is thought to result from the accumulation of oxidative stress with age, which then activates Jun N-terminal kinase (JNK) and Notch/Delta signaling. Increased levels of these signaling pathways cause ISC proliferation and, ultimately, EB misdifferentiation. Over time, the accumulation of these aberrantly stalled stem and progenitor cells has been shown to give rise to lethal tumors within the midgut (Biteau et al. 2008).

More recent work has looked at cancers that arise from the mammary epithelium in mice and humans (Garbe et al. 2012; Pelissier Vatter et al. 2018; Li et al. 2020; Shalabi et al. 2021). The mammary epithelium is composed of two different types of differentiated cells—myoepithelial cells and luminal cells—as well as dedicated progenitors for each lineage, and among these there has been an enormous interest in defining the cells of origin for breast cancers. Interestingly, and somewhat counterintuitively, the putative cell type of origin for breast cancers that express myoepithelial markers is thought to be either luminally biased progenitor cells or mature luminal cells (Molyneux et al. 2010). Supporting this finding, a recent study characterized both luminal and myoepithelial marker expression in malignant breast lesions (i.e., invasive ductal carcinomas). The investigators observed that of all the luminal cells in the tumors, approximately half expressed markers of the myoepithelium, suggestive of a phenotypically plastic cell type giving rise to some of these cancers (Shalabi et al. 2021).

Given that aging is a primary risk factor for sporadic breast cancer, a key question is whether mixed-lineage cells arise with age. Histological analyses as well as single-cell mass cytometry suggested that novel populations of luminal cells arise in the mammary epithelium with age, and these indeed express myoepithelial markers (Pelissier Vatter et al. 2018; Shalabi et al. 2021). In addition to inappropriate marker expression, these aged luminal cells also share certain functional features with myoepithelial cells, including cellular adhesion, signaling, and colony formation potential. Moreover, transcriptional analyses of these cells highlighted similarities with immortalized mammary epithelial cells from aged donors, suggesting that these cells may be transitioning toward malignant transformation (Pelissier Vatter et al. 2018). However, the underlying mechanisms of this fate shift and its causal role in tumorigenesis require further investigation. Interestingly, Shalabi et al. (2021) showed that mammary epithelium from young donors carrying germline mutations in critical DNA damage repair machinery, such as BRCA1 and BRCA2, phenocopy many of these age-related changes in terms of mixed cellular identity, and these cells also have a much higher chance of developing into a cancer.

The examples we have discussed in this section suggest that DNA damage can directly lead to malignant transformation by giving cells a competitive advantage (i.e., CHIP) or induce shifts in cellular identity (basal-like luminal cells in the mammary epithelium). Additionally, DNA damage may facilitate transformation via indirect mechanisms that rewire the epigenome (DNA methylation changes in COADs) or reprogram the ECM so it loses tumor-suppressive function (dermal damage leading to less collagen production). Moving forward, it will be interesting to understand whether age-related DNA damage plays a central role in promoting entry into a preneoplastic state.

Factor-induced rejuvenation and cell identity changes

One of the best-studied paradigms for cell identity change is the reprogramming of somatic cells into induced pluripotent stem cells (iPSCs). Almost 20 years ago Takahashi and Yamanaka (2006) demonstrated that the expression of the four transcription factors—Oct4, Klf4, Sox2, and c-Myc (OKSM)—can stably reprogram adult fibroblasts into iPSCs, which molecularly and functionally resemble embryo-derived pluripotent stem cells. Intriguingly, several recent studies have suggested that the expression of the same four factors in vivo or in vitro can modulate several hallmarks of aging, including but not limited to the restoration of heterochromatin, decreases in DNA damage levels, and the reduction of the epigenetic clock toward a younger state (Horvath 2013; Ocampo et al. 2016; Mertens et al. 2021). Moreover, iPSCs derived from aged hematopoietic progenitor cells were able to give rise to a youthful hematopoietic system when redifferentiated into HSCs, suggesting that at least some of the mechanisms driving aging are epigenetic and thus are potentially reversible (Wahlestedt et al. 2013, 2017). However, prolonged expression of these reprogramming factors in vivo typically leads to teratoma formation as well as intestinal and hepatic dysfunction, ultimately causing premature death (Abad et al. 2013; Ohnishi et al. 2014; Parras et al. 2023). These findings have prompted researchers to test whether the transient induction of OKSM or subsets thereof are sufficient to elicit a more youthful state without the loss of cellular identity, a phenomenon termed “partial reprogramming.” As reprogramming factor-dependent reversion of aging hallmarks has been extensively reviewed elsewhere, here we focus exclusively on the crucial question of whether these paradigms actually alter cell identity (Cipriano et al. 2024).

Most studies examining cell identity changes during partial reprogramming used tissue-specific expression of OKSM. For example, the transient expression of OKSM in adult cardiomyocytes elicits gene expression changes that are reminiscent of embryonic or neonatal cardiomyocytes, and this has been linked to a re-entry into the cell cycle. However, these transcriptional changes promptly reverted after factor withdrawal (Chen et al. 2021). Brief expression of low levels of OKSM in the liver led to a decreased expression of mature hepatocyte markers and transcription factors as well as the upregulation of genes involved in liver development, leading the investigators to conclude that hepatocytes had been reprogrammed into an “undetermined plastic state.” Similar to the cardiomyocyte example, cells regained their fully differentiated hepatic state once OKSM expression was discontinued (Hishida et al. 2022).

To dissect the kinetics of OKSM-induced rejuvenation and cell identity, researchers have turned to more tractable in vitro models of partial iPSC reprogramming using adult or aged human fibroblasts. Using epigenetic age as a correlate of rejuvenation, multiple groups have demonstrated that partially reprogrammed fibroblasts exhibit an ∼30 year reduction in epigenetic age while still maintaining certain aspects of their fibroblast identity (Olova et al. 2019; Simpson et al. 2021; Gill et al. 2022). However, fibroblast cultures treated in this way also upregulated transcriptional programs and markers associated with iPSCs and concomitantly acquired an epithelial morphology, suggesting a shift in cell identity. To test the fate of cells treated in this manner, Gill et al. (2022) discontinued OKSM expression in partially reprogrammed fibroblasts, which resulted in a reversion to a fibroblast-like state based on transcriptional and morphological criteria, mirroring in vivo studies in cardiomyocytes and hepatocytes (Chen et al. 2021; Hishida et al. 2022). Mechanistically, fibroblast enhancers remained unmethylated during partial reprogramming but gained DNA methylation during the final stages of complete reprogramming toward iPSCs. This suggests that enhancer memory may play a role in identity maintenance during partial reprogramming (Gill et al. 2022). This observation calls to mind the previously reported requirement for Tet-dependent demethylation during OKS-induced axon regeneration in mice (Lu et al. 2020). Although the latter study used a different type of DNA methylation clock, based on ribosomal DNA, for their analysis, it will be interesting to determine whether retinal ganglion-specific enhancers are relevant targets of the Tet enzymes in this model. In addition to the observed reductions in DNA methylation age in these two models, the investigators reported molecular and functional improvements following OKS(M) expression, consistent with a rejuvenating effect.

A critical question emerging from these and other studies is whether the transient loss of cell identity is required for eliciting rejuvenation. Interestingly, certain genetic approaches that alter cell identity in vitro have been shown to modulate various aging hallmarks including DNA methylation clocks, aging transcriptional signatures, and cellular biomarkers. For example, although reprogramming toward iPSCs and dedifferentiation toward adult stem-like cells reset or reduce epigenetic age, respectively, transdifferentiation did not lead to any changes (Fig. 2; Horvath 2013; Huh et al. 2016; Sheng et al. 2018). However, these studies are limited in scope and do not unequivocally prove that cell fate change—namely, dedifferentiation—is required for the rejuvenation process.

Figure 2.

Examples of how forced changes in cell identity in vitro affect cellular DNA methylation age, a key aging biomarker. (Left) Reprogramming differentiated cells into induced pluripotent stem cells (iPSCs) resets the DNA methylation clock back to 0. (Middle) Dedifferentiating peripheral blood cells into induced neural stem cells, without going through a pluripotent state, reduces but does not reset the DNA methylation clock. (Right) Transdifferentiation of fibroblasts into induced neurons does not alter epigenetic age. Importantly, in each of these transitions, other hallmarks of aging, including transcriptional and cellular changes, were shown to track with DNA methylation age. Given the differences in epigenetic age between these dedifferentiation and transdifferentiation approaches, it will be important to understand whether dedifferentiation is an essential part of the rejuvenation process or whether cells can be rejuvenated without shifts in cellular identity.

In an orthogonal effort to address this question, Roux et al. (2022) assessed how different combinations of the canonical reprogramming factors affected rejuvenation and the loss of somatic cell identity using a mesenchymal stem cell (MSC) system. Although each combination of factors reduced MSC transcriptional identity to varying extents, the investigators concluded that identity suppression was not correlated with a reduction in transcriptional aging scores. For example, OS had an effect on the transcriptional aging score similar to that of OKSM, even though it did not suppress mesenchymal identity to the same extent as OKSM (Roux et al. 2022). Of relevance, the expression of OS or any other dual combinations of the OKS factors with AAV vectors was insufficient to facilitate axon regeneration using an optic nerve injury model, suggesting that the effects of reprogramming factors may vary depending on cellular context and delivery method.

Considering the complex effects of OKSM on reprogramming, rejuvenation, and tumorigenesis, it will be important to understand whether other transcription factors not associated with pluripotency and cancer may be equally effective at reversing hallmarks of aging. Toward this goal, a recent study by Jing et al. (2023) searched for genes involved in human mesenchymal stem cell (hMSC) aging. Using a CRISPR activation screen, the investigators identified the transcriptional regulator SOX5 and its target, HMGB2, as two factors that can alleviate multiple features of MSC aging and senescence. SOX5 and HMGB2 are typically expressed in MSCs, but their expression decreases with aging. Importantly, the re-expression of SOX5 in aged, senescent MSCs did not cause dedifferentiation, and SOX5-expressing MSCs were reportedly more effective at chondrogenesis, adipogenesis, and osteogenesis. Furthermore, SOX5-expressing MSCs were not tumorigenic upon transplantation in vivo. To compare this lineage-specific approach with OKSM reprogramming, the investigators used a mouse model of osteoarthritis, an age-related joint disease involving MSCs. Intra-articular injection of SOX5-, HMGB2-, or OKSM-expressing lentivirus into aged, arthritic mice ameliorated arthritic lesions to a similar extent, suggesting that a lineage-specific approach can be as effective as OKSM reprogramming (Jing et al. 2023). More broadly, this study highlights the possibility of identifying other lineage-specific factors that can ameliorate certain effects of aging without altering cell identity and without the risk for tumors (Table 1; Zhu et al. 2019b; Roichman et al. 2021; Nefzger et al. 2022; Ribeiro et al. 2022; Roux et al. 2022; Wendorff et al. 2022).

Table 1.

Comparison of OKS(M)-based rejuvenation approaches with other transcription factor or epigenetic modifier-based approaches

Intervention	Rejuvenating effects	Effect on cell identity
OKSM/OKS-based approaches (for review, see Cipriano et al. 2024)	Increased life span in progeroid murine models and geriatric mice Amelioration of aging pathologies in tissues (e.g., pancreas, brain, skin, liver, heart, muscle, and retinal ganglion cells) More youthful molecular state based on transcriptional and epigenetic analyses	Dedifferentiation consistently observed during factor administration, whereas factor withdrawal leads to redifferentiation Teratoma formation upon prolonged OKS(M) expression Hepatic and intestinal toxicity observed
Slug overexpression in aged muscle stem cells (Zhu et al. 2019b)	Repression of senescence markers Increased self renewal in long-term in vitro culture model Improved stem cell engraftment in transplant models	Not assessed
Global overexpression of Sirt6 in mice (Roichman et al. 2021)	Increased life span in naturally aged mice Improved physiological hallmarks of aging, including increased physical activity Amelioration of age-related defects in metabolism primarily through Sirt6's function in the liver and adipose tissue	No difference in neoplasm presence between Sirt6-overexpressing versus control mice at time of death (yet cancer onset appears delayed in Sirt6-overexpressing mice)
Inactivation of Phf6 in murine HSPCs (Wendorff et al. 2022)	Balanced differentiation into myeloid and lymphoid lineages Increased HSC engraftment Youthful transcriptional, immunophenotypic, and functional HSC profile	Phf6 knockout rebalances the HSPC compartment toward a young HSPC cellular distribution No evidence of dedifferentiation or transformation despite Phf6's role in malignancy
Cyclic overexpression of an N-terminal truncated Foxm1 in mice (Ribeiro et al. 2022)	Life span extension in progeroid murine models and naturally aged mice Amelioration of age-related pathologies in the skin, heart, muscle, and adipose tissue	Long-term cyclic induction of Foxm1 did not result in increased rates of transformation compared with control aged mice
Msx1 overexpression in aged murine myogenic cells in vitro (Roux et al. 2022)	Ameliorated age-related defects in myogenic differentiation Reduction in aging myogenic transcriptional score	Msx1-overexpressing cells were able to differentiate more efficiently Msx1 was previously suggested to facilitate myocyte dedifferentiation into a multipotent state (but not a pluripotent state)
Egr1, Irf1, and FosB overexpression in aged murine intestinal stem cell organoids (Nefzger et al. 2022)	Enhanced ISC regenerative potential measured by organoid-forming potential and organoid counts derived from aged ISCs	Not assessed
Sox5 or Hmgb2 overexpression in senescent mesoderm (Jing et al. 2023)	Counteracted senescence in human mesenchymal progenitor cells (hMPCs) Improved osteoarthritis symptoms in aged mice As effective as OKSM overexpression in murine models of osteoarthritis	Sox5 overexpression in hMPCs leads to more efficient differentiation in mesoderm lineage Sox5-overexpressing hMPCs are nontumorigenic when transplanted into immunocompromised mice

It is worth noting that cell-extrinsic systemic interventions (i.e., calorie restriction) have also yielded promising results as they relate to rejuvenation, and more extensive reviews on these topics have been covered elsewhere (Mahmoudi et al. 2019b; Green et al. 2022; Liu et al. 2024). However, how these interventions impact age-associated changes in cell identity remains limited and, where applicable, has been referenced throughout the text.

Conclusions and future outlook

Accumulating evidence suggests that aging affects cell identity maintenance across tissues and organisms (Lukjanenko et al. 2019; Shalabi et al. 2021; Lu et al. 2023a). Although these changes may be rare and subtle, their detection warrants further investigation, as altered cell fate has been implicated in age-related organ dysfunction and diseases including cancer. A major limitation of studies that previously examined changes to cell identity with age, including programs linked to dedifferentiation, biased differentiation, acquisition of alternative lineages, and preneoplastic changes, is that they were largely descriptive in nature. It is therefore imperative to introduce robust genetic approaches such as lineage tracing and lineage ablation as well as cell type-specific functional assays into relevant studies in order to better define the extent of cell fate changes and their implications for aging. Similarly, shifts in cell state that were previously observed in accelerated models of aging including laminopathies, long-term in vitro culture systems, or the recently published ICE model (Tao et al. 2019; Shah et al. 2021; Yang et al. 2023) need to be extended to physiological aging models to appreciate the generality of the observations in an unperturbed context. From a mechanistic point of view, it will be important to understand the driving forces behind the observed shifts in cellular identity. For example, the work discussed in this review has implicated each hallmark of aging in aspects of age-related changes to cell identity, including but not limited to genomic instability, deregulated nutrient sensing, epigenetic alterations, and altered intercellular communication (Brack et al. 2007; Traxler et al. 2022; Liu et al. 2022a; Yang et al. 2023). Although this review highlighted distinct mechanisms underlying particular shifts in cell identity, it is likely that there is also considerable mechanistic overlap among aging hallmarks, as exemplified by increased levels of DNA damage as well as epigenetic and transcriptional noise underlying multiple identity shifts. Nevertheless, our observations provide a useful framework for future studies that explore the causal roles of these processes in age-dependent cell identity control.

Our review also highlights how changes in cell identity appear to be a shared feature of using OKSM-induced partial reprogramming (Roux et al. 2022). Across most examined systems, induction of OKSM leads to transient dedifferentiation or the loss of somatic cell identity followed by redifferentiation after factor removal (Gill et al. 2022). However, the identity of intermediate cells generated during partial reprogramming is only beginning to be understood (Chondronasiou et al. 2022). A more rigorous characterization of these cells is thus necessary to understand how OKSM rewire aging-associated transcriptional and epigenetic patterns as well as to what extent this process actually requires a transient loss of cell identity, which remains controversial (Roux et al. 2022). These unknowns, coupled with the risk of teratoma formation and tissue toxicity, will remain formidable challenges for the clinical translation of these approaches.

To overcome some of these challenges, it should therefore be fruitful to also explore downstream effectors of OKSM-based approaches with the goal of disentangling pathways critical for rejuvenation and those involved in the loss of cell identity. This in turn should allow for more tailored approaches to target particular genes or epigenetic marks aimed at maximizing the effects on rejuvenation while minimizing the effects on cell identity loss. Previously identified epigenetic roadblocks to iPSC reprogramming are obvious candidates to pursue (Brumbaugh et al. 2019). An equally promising and complementary approach will be the use of lineage-specific reprogramming factors, some of which were shown to be as efficacious as OKSM-based approaches to modulate or reverse hallmarks of aging without the unwanted effects on cell identity and tumorigenesis (Table 1). Regardless of the approach, a more detailed understanding of the relationship between any factor-based rejuvenation intervention and cellular identity will be imperative as it relates to their ultimate translatability into the clinic.

Competing interest statement

The authors declare no competing interests.