Aging is a universal and inexorable biological process characterized by a progressive, time-dependent decline in physiological function, which heightens an organism’s susceptibility to internal and external stressors, thereby increasing the risk of disease and mortality. Aging is a multifactorial phenomenon driven by the accumulation of damage across multiple molecular and cellular levels. These “hallmarks of aging” include genomic instability, telomere attrition, epigenetic alterations, loss of proteostasis, mitochondrial dysfunction, and cellular senescence (1). However, for decades, in both clinical practice and biomedical research, the complexity of this process has been simplified into a single, convenient metric: chronological age (2). Defined simply as the passage of calendar years since birth, age has long served as a primary variable in an array of clinical and population-level analyses, for instance, in demographic studies to forecast societal healthcare needs and in epidemiological surveillance to track the incidence of age-related diseases, such as cancer and dementia (3, 4). The pervasive reliance on chronological age is based on the assumption that the aging process occurs uniformly across all individuals, overlooking significant variations. Nevertheless, its widespread use has historically stemmed from the absence of reliable indicators, measurement methods, and standards to quantify the progression of an individual’s unique aging process. Consequently, chronological age often serves as an imperfect proxy for functional capacity, resilience, and vulnerability to adverse health outcomes, providing an incomplete picture of the health status (4, 5). The heterogeneity observed in older adults highlights the limitations of chronological age. Individuals sharing the same chronological age often exhibit divergent physiological states: one may manifest high physical and cognitive resilience, whereas another suffers from multimorbidity and severe functional decline.

In contrast with its chronological counterpart, biological aging reflects the underlying rate of physiological deterioration and cumulative burden of molecular and cellular damage. It represents an individual’s functional age and provides a more accurate measure of their ability to maintain a stable internal environment and the risk of morbidity and mortality (1, 2). Consequently, gerontology and geriatric medicine need to adopt sophisticated and clinically relevant measures to capture the essence of biological aging. The development and validation of robust biomarkers for aging are paramount for advancing our understanding of this process and developing targeted interventions.

One of the most compelling and extensively validated measures of frailty is the Frailty Index (FI) (6-8). Based on the cumulative deficit model, the FI quantifies an individual’s state of vulnerability by systematically summing a diverse range of health deficits. These include diseases, symptoms, signs, laboratory abnormalities, and disabilities, with a typical index comprising 30∼70 variables. An individual’s score, calculated as the proportion of deficits present out of the total considered, provides a continuous, sensitive, and dynamic measure of biological age. Its strength lies in recognizing that aging is not the failure of a single organ system but a widespread loss of physiological complexity. Although FI is a powerful integrative tool, its application is challenging. Several of its components rely on variables that can be influenced by subjective patient reporting (e.g., feelings of fatigue) or clinical interpretation, rather than purely objective measurements. Furthermore, its accuracy depends on a thorough and precise clinical diagnosis of each deficit. This highlights a critical imperative for the field: whereas the FI provides an excellent high-level view of an individual’s frailty, its goal is to identify and validate precise molecular-level biomarkers. Such objective indicators would not only complement and refine FI but also enable us to probe the fundamental biological mechanisms of aging, transitioning from a description of vulnerability to a mechanistic understanding of its origins.

Ultimately, the transition from a chronological to a biological framework of aging is not merely an academic exercise but a necessity. This is an essential step toward achieving more precise, effective, and equitable healthcare for the rapidly growing global population of older adults, enabling us to move beyond extending lifespan to enhance healthspan and gain deeper insights into mechanisms of human aging.

Numerous studies have utilized markers of cellular senescence to assess aging at both the cellular and organismal levels (9). Recently, the focus has expanded to epigenetic clocks as robust indicators of biological age (10). Furthermore, large-scale plasma proteomics studies have helped identify novel aging biomarkers and develop sophisticated models for disease prediction (11).

AD is the most common cause of dementia, pathologically characterized by the extracellular accumulation of amyloid-beta (Aβ) plaques and the intracellular formation of neurofibrillary tangles (NFTs), which are composed of hyperphosphorylated tau protein. Clinically, this neurodegenerative disorder is characterized by a progressive decline in memory and other cognitive functions, ultimately leading to a complete loss of independence (36). The prevalence of AD rises dramatically with increasing age. Furthermore, the observation of limited Aβ deposition and NFTs in the brains of cognitively normal older adults suggests a potential continuum between normative brain aging and AD pathology (37). Consequently, AD has historically been conceptualized as a form of “accelerated aging.” However, a substantial body of subsequent research has demonstrated that AD is not a mere extension of normal aging but a distinct and altered pathological process. Recent advances in single-cell analysis of human postmortem brain tissue have provided strong evidence supporting this modern perspective. In one landmark study, the reconstruction of a cellular manifold from 1.65 million single-nucleus transcriptomes, obtained from the postmortem brain tissue of 437 older adults, revealed that aging and the accumulation of AD pathology proceed along two independent trajectories (38). Notably, the trajectory representing progression to AD was characterized by distinct clusters of microglia and astrocytes exhibiting transcriptional changes in inflammatory and metabolic pathways that were not observed in normal aging. These findings suggest that AD is a pathologically transformed process at the cellular level. This divergence is also evident in epigenomic studies. Using multiregional single-cell ATAC-seq analysis, Liu et al. (39) demonstrated that AD progression was associated with extensive changes in chromatin accessibility and widespread epigenomic erosion. Conversely, some cognitively resilient older individuals exhibit suppression or delay of these epigenetic alterations, suggesting that AD is driven by the selective vulnerability of specific cell populations and dysregulation of key regulatory mechanisms (40).

More sophisticated aging models are required to delineate the differential pathways between aging and AD. To date, established aging clocks rely on DNA methylation, bulk transcriptomic profiles from RNA-seq, and proteomic and metabolomic markers in the blood. Although these approaches are valuable for estimating biological age at the organismal level, they inherently lack the resolution to capture cell-type-specific changes and heterogeneity at the tissue level. To overcome this limitation, Muralidharan et al. (41) recently constructed cell-type-specific aging clocks by analyzing single-nucleus transcriptome data from the postmortem prefrontal cortex of individuals aged 18∼94 years. This analysis revealed that major brain cell types, including neurons, astrocytes, microglia, and oligodendrocytes, exhibit distinct aging rates and divergent transcriptomic patterns. Specifically, the clocks of oligodendrocyte progenitor cells, microglia, and excitatory neurons demonstrated the highest concordance with the chronological age. These clocks demonstrated consistent predictive power when validated against independent, single-nucleus transcriptome datasets. This study confirmed the existence of cell-type-specific aging dynamics and showed that single-cell analysis can capture cellular-level differences that are obscured by bulk data. For example, the microglial clock identified age-related upregulation of genes involved in inflammatory responses and cytokine production, whereas neuronal and glial clocks captured changes in synaptic transmission pathways. When applied to a cohort of patients with AD, clocks predicted that microglia and oligodendrocytes would be significantly “older” than their chronological ages, providing direct evidence that specific glial cell populations are disproportionately vulnerable to pathology-associated aging.

These findings indicate that AD is not a result of accelerated normal aging. Instead, it is a transformed pathological process that occurs when specific cell types become disproportionately vulnerable to age-related changes (Fig.1). Therefore, future research on AD must expand beyond models that merely recapitulate the accumulation of Aβ and tau. It is imperative to develop next-generation models that can precisely simulate the aging process and incorporate cell-type-specific vulnerabilities that underlie this devastating disease.

Animal models have been foundational to aging studies. Invertebrate models, such as Caenorhabditis elegans and Drosophila melanogaster, offer the advantage of short lifespans, allowing for the rapid observation of aging phenotypes. They are also highly amenable to large-scale screening studies involving genetic manipulation or environmental changes. For instance, landmark studies in C. elegans have demonstrated that deletion of genes such as daf-2 or daf-16 can significantly extend lifespan (42). Similarly, experiments in C. elegans have provided early evidence that caloric restriction can prolong life, revealing the crucial roles of metabolic regulation and autophagy in longevity (43, 44). In D. melanogaster, the inhibition of transposable element activation has been shown to extend lifespan, suggesting that chromatin remodeling and epigenetic regulation are central mechanisms of aging and longevity (45). Despite their utility, invertebrate models have significant limitations in studying brain aging. Their nervous systems are less complex and lack the intricate cortical architecture of the human brain. This evolutionary divergence extends far beyond neuroanatomy. Fundamental differences in metabolism, distinct organization of organ systems, and the absence of an adaptive immune system, which is increasingly recognized as a critical modulator of neuroinflammation and cognitive function during aging, make it difficult to conclude that aging processes are directly analogous to those in humans. Consequently, mouse models have been widely used to study age-related diseases, including neurodegeneration. The advantages of the mouse model include the ability to easily observe age-related declines in performance in behavioral paradigms designed to assess cognitive domains (e.g., memory impairment), physical capacity (e.g., reduced muscle strength), and external appearance (e.g., changes in fur color and texture) (46). However, several significant challenges remain. Acquiring wild-type aged mice without genetic modifications requires at least 24 months, which is both time-consuming and resource-intensive. To bypass these logistical hurdles, some researchers have used strains developed for accelerated aging, such as senescence-accelerated mouse-prone 8 or progeroid models expressing mutated lamin A (progerin) (2, 47). However, a significant caveat is that these models are often closer to disease models than to true aging models. The cellular and tissue microenvironments in these animals can be dominated by aggressive pathological cascades that may differ substantially from the gradual multifactorial accumulation of deficits observed in natural physiological aging processes. Furthermore, fundamental interspecific differences cast doubt on whether these models accurately recapitulate the natural human aging process. Wild-type mice rarely develop AD and lack the genetic complexity of key human risk factors (48, 49). For example, humans carry three common alleles for APOE (APOE2, APOE3, and APOE4), with the APOE4 variant dramatically increasing the risk of disease. In contrast, mice possess only a single functionally distinct Apoe allele. This disparity necessitates the introduction of humanized genes, such as the various APOE variants, through genetic modifications to model this critical aspect of the disease (50). Models such as the 5xFAD mouse exhibit AD pathology at a very early age, Aβ accumulation by 2 months, and memory deficits by 4 months (51). Consequently, although AD is an age-related disease, these models often fail to reflect the underlying aging context.

As recapitulating physiological aging in experimental models remains challenging, a more tractable strategy is to model one or more key factors or hallmarks of the aging process to investigate their pathological roles in AD (Table 1; 52-80). The development of such reductionist models often begins in vitro, where cellular and molecular variables can be precisely controlled and manipulated. This approach can start with establishing techniques to simulate age-associated phenotypes in isolated single-cell types and subsequently progress to more complex co-culture systems to study the impact of aging on cell-cell interactions. By creating robust in vitro platforms that mimic the defined aspects of aging, researchers can begin to elucidate the causal roles of these changes before undertaking more complex in vivo manipulations. Among the various techniques developed for this purpose, direct conversion has emerged as a powerful tool. Recent methodologies have enabled the direct generation of neurons from the skin cells of aged or diseased individuals, bypassing the induced pluripotent stem cell (iPSCs) stage. This process, known as direct conversion, involves the forced expression of key neuronal transcription factors, such as Ngn2 and Ascl1, or the use of specific miRNAs to drive neuronal identity (81-83). It is critical to distinguish between the concepts of ‘reprogramming’—the technical process of altering cell identity—and ‘rejuvenation’—the biological resetting of cellular age. iPSCs generation involves both cellular reprogramming (from the somatic to pluripotent state) and inherent rejuvenation, which erases age-associated signatures. In contrast, direct conversion achieves reprogramming of cell identity (e.g., from fibroblasts to neurons) without inducing cellular rejuvenation. Therefore, directly converted induced neurons (iNs) are uniquely advantageous for aging research because they decouple cell fate changes from age resetting, retaining the donor’s aging signatures (Table 2; 52, 53, 83-88). For example, iNs derived from the fibroblasts of older donors exhibit an increased epigenetic age compared to iPSC-derived neurons (iNEUs), which possess youthful characteristics owing to the rejuvenation that occurs during the iPSC reprogramming (84). Furthermore, iNs from patients with AD exhibit increased senescence markers (including SA-β-gal) and elevated levels of pathological markers, such as Aβ, phosphorylated tau, and mislocalized TAR DNA-binding protein 43, compared to age-matched controls (52, 53). Crucially, they also overcome a major hurdle in disease modeling related to the tau protein. Alternative splicing of the MAPT gene in the adult human brain produces a balanced 1：1 ratio of two major tau isoforms: those with three microtubule-binding repeats (3R) and those with four (4R). This balanced expression is considered a hallmark of mature adult neurons, and its dysregulation is believed to be central to a class of neurodegenerative disorders known as tauopathies (89). However, iNEUs, owing to their rejuvenated embryonic-like state, typically express only the fetal 3R isoform and fail to undergo a developmental switch to produce the 4R tau isoform. By retaining the epigenetic and maturational states of the donor cells, iNs express these adult 4R tau isoforms, making them an invaluable and physiologically relevant model for studying the mechanisms of adult-onset tauopathies (85).

Despite these advantages, the utility of iNs remains a matter of debate. Because they originate from fibroblasts, a cell type that may undergo a different aging process than neurons, it remains unclear whether the aged phenotypes observed in iNs truly reflect neuronal aging (90). Besides these conceptual challenges, significant practical hurdles have limited their widespread use. Primary fibroblasts enter replicative senescence after a limited number of divisions, severely restricting the total number of iNs generated from a single donor biopsy. This was compounded by issues related to quality control and reproducibility (90). For example, patient-derived fibroblast lines exhibit high variability in growth rate and neuronal conversion efficiency. This inherent inconsistency between donors makes it difficult to obtain standardized results and to compare findings across studies. Furthermore, iNs, as post-mitotic and highly specialized cells, are fragile and difficult to maintain in a healthy state for extended periods, which may be required to model the slow progression of age-related cellular decline. These limitations directly conflict with the increasing need for large-scale generation of neurons for modern biomedical research. A substantial and consistent supply of cells is essential for robust multiple experiments that are fundamental to rigorous scientific inquiry. More specifically, large cell numbers are a prerequisite for powerful research applications, such as high-throughput drug screening, which involves testing thousands of potential therapeutic compounds, and for comprehensive multiomics analyses (e.g., transcriptomics, proteomics, and metabolomics), which require significant starting material to generate reliable, high-quality data (91, 92). These large-scale approaches are critical for elucidating complex disease mechanisms and identifying novel therapeutic targets. Therefore, whereas iNs offer a unique window into aged cellular states, these limitations constrain their application in the studies necessary to advance the field.

iPSCs are generated from somatic cells, such as fibroblasts or blood cells, by expressing Yamanaka factors (93, 94). Given their theoretically unlimited proliferative potential, iPSCs are an excellent source for the large-scale production of neurons and iNEUs, which are also relatively stable in long-term cultures. However, as previously mentioned, the fundamental limitation of this model is the inherent rejuvenation that occurs during the reprogramming process. This effectively erases the aging signatures of the donor, including epigenetic alterations, and reverts the cells to an embryonic-like state (95). Consequently, whereas iNEUs derived from patients can recapitulate certain genetic aspects of a disease (96-98), the absence of a crucial aging context means that these phenotypes are often restricted to the earliest manifestations of the pathology, reflecting an asymptomatic or prodromal stage rather than a full-blown age-dependent clinical state. They fail to capture the complex interplay between genetic predisposition and decades of cellular stress that drive late-onset neurodegeneration, making their direct application in modeling age-related diseases problematic.

Various attempts have been made to intentionally induce an aging state in rejuvenated iNEUs. However, these strategies often introduce the very same conceptual limitations as those previously discussed for in vivo accelerated aging models. Rather than recapitulating physiological aging, these models tend to impose an aggressive pathology. For instance, introducing a progeria mutation into iPSC-derived dopaminergic neurons induces certain age-associated phenotypes, such as increased mitochondrial reactive oxygen species, but simultaneously triggers early features of PD, thereby confounding physiological aging with overt disease pathology (99). Similar outcomes were observed when telomere shortening was used to mimic premature aging in the same cell type (100). Although these models display specific phenotypes, they suffer from a fundamental confounding factor: mirroring the caveats of their in vivo counterparts, they represent a distinct pathological state rather than the gradual, multifactorial process of physiological aging. This makes it difficult to distinguish phenotypes driven by aging insults from those representing independent pathologies. Consequently, these models have limited utility in clarifying the causal relationship between physiological aging and the development of age-related diseases (Fig. 2).

The preceding sections underscore the profound limitations inherent in the current models of physiological aging, from the temporal and financial costs of aging animal cohorts to the correlational nature of human studies. These challenges have created a critical bottleneck, hindering our ability to move beyond identifying associations to establishing the causal mechanisms that drive age-related neurological decline. To overcome this impasse, the future of aging research may not lie in perfecting any single model but in developing an integrative framework that synergistically combines advances in bioinformatics, sophisticated in vitro systems, and large-scale human data to create and validate in vivo models with unprecedented physiological relevance. A foundational pillar of this framework is the continued advancement of high-resolution bioinformatics. The proliferation of single-cell and spatial multiomics (e.g., single-cell RNA sequencing, single-nucleus ATAC sequencing, spatial transcriptomics, and proteomics) applied to postmortem human brain tissue has generated an exquisitely detailed atlas of how aging manifests at the cellular level (32, 35, 38, 39, 41, 101-112) (Table 3). By analyzing these datasets, we can move beyond generalized aging markers to identify robust, cell-type-specific molecular signatures for aged neurons, astrocytes, microglia, and oligodendrocytes. This allowed us to define the appearance of a “physiologically aged” human brain cell from a systems biology perspective, thus providing a critical reference blueprint against which any experimental model must be validated. The next step involves engineering in vitro models that can accurately recapitulate these complex molecular signatures. iPSCs are an ideal starting point for this purpose. Using iPSCs from a cohort of healthy individuals, we can control the vast genetic heterogeneity that plagues human studies and create a standardized, scalable platform. The central challenge is to intentionally induce a state of “physiological aging” in rejuvenated cells. Rather than relying on potent non-physiological insults, this can be achieved through subtle and chronic stressors that mimic the accumulated challenges of life. This could involve long-term culture under conditions of mild metabolic stress, DNA breaks, controlled induction of low-level mitochondrial dysfunction, or the use of advanced clustered regularly interspaced short palindromic repeats (CRISPR)-based epigenetic editors to systematically rewrite methylation patterns to match the “epigenetic age” observed in human brain tissue (113-116). The goal is not to create a caricature of aging but to guide young cells toward the specific molecular state defined by our bioinformatic blueprint.

This framework requires a rigorous validation process. The molecular state of these in vitro aged cells, including their transcriptomes, proteomes, and epigenomes, must be systematically compared with the reference data from the human brain. A model was considered validated only when its multiomic signature demonstrated substantial concordance with that of aged human cells. This data-driven validation step ensures that we are not studying an artifact of the culture dish but a system that genuinely mirrors the key aspects of the human aging process. Human informatics databases, such as the UK Biobank or the Religious Orders Study and Rush Memory and Aging Project, would serve as the ultimate arbiters of a model’s relevance (117, 118). Ultimately, any causal factors identified in vitro must be validated in vivo to confirm their relevance to complex physiological systems. Once a model passes this stringent validation, it transforms from a descriptive tool into a powerful platform for investigating causality. Within this controlled human cell-based system, mechanistic experiments that are impossible to conduct in humans can be performed. Using CRISPR-mediated gene editing, aging markers identified in an initial bioinformatic screening can be systematically perturbed in a cell-type-specific manner. These experiments can directly reveal whether an age-associated change is a driver of functional decline, a compensatory protective response, or a neutral bystander, thus untangling the complex web of causality. Ultimately, this validated physiological aging platform provides an ideal background for investigating the influence of genetic risk factors on neurodegenerative disease. Not all elderly individuals develop age-associated neurodegeneration, such as AD. By introducing genetic risk factors, such as APOE4, into our iPSC lines using isogenic controls, we can investigate how these factors interact with and modify the physiological aging process. This approach enables the systematic dissection of the mechanisms by which genetic predisposition converts the trajectory of normal aging into pathological neurodegeneration. By integrating these disparate fields, this forward-looking framework promises to bridge the causal gap in aging research and pave the way for the development of targeted therapeutics that promote neurological health.