Alzheimer’s disease through the lens of mesenchymal drift: the collapse of cellular identity

Abstract

Cellular identity is maintained by precise epigenetic regulation, but chronic stress and aging can disrupt this control, leading to mesenchymal drift (MD). Unlike the coordinated process of epithelial-mesenchymal transition (EMT), MD is a stochastic, lineage-independent phenomenon marked by increased epigenetic noise and the widespread activation of profibrotic gene programs. As originally defined, MD requires two components to co-occur: a compromised original cellular identity together with the acquisition or intensification of mesenchymal traits. Applying the MD framework to Alzheimer’s disease (AD) provides a new perspective on its pathogenesis. In AD, brain cells do not merely survive or die due to protein toxicity; instead, they lose their specialized identities in an uncoordinated fashion. This transcriptomic drift has been reported across major cell types in the central nervous system: neural progenitor cells differentiate prematurely, mature neurons undergo transcriptomic dedifferentiation, endothelial cells develop features of vascular fibrosis, and glial cells lose their homeostatic functions. Direct enrichment of the MD signature in the AD brain, however, has so far been demonstrated only in astrocytes, oligodendrocytes, and pericytes, and, in our own reanalysis, in human-induced pluripotent stem cell (hiPSC)-derived neural progenitor cells carrying an APP mutation; we distinguish this evidence from observations documenting identity loss or a reactive state alone. We further caution that the novel cell populations identified in recent single-cell transcriptomic studies should not be assumed to represent genuinely new cell types: such clusters may equally reflect a blurring of existing cell lineages, or technical artifacts that produce the same appearance. A drifting transcriptomic profile also does not always signify a corresponding physiological change. These possibilities warrant careful attention, and we therefore emphasize the critical need to integrate transcriptomic data with in vivo functional and morphological validation. This integration is essential to determine whether these identity shifts contribute to AD progression or merely reflect underlying epigenetic instability.

INTRODUCTION

The central nervous system relies on the precise maintenance of diverse, specialized cell identities. For decades, Alzheimer’s disease (AD) research operated on a binary model of cell survival versus death, primarily attributing neuronal loss to amyloid-β (Aβ) plaques and tau tangles. However, advances in single-cell multi-omics have challenged this view, indicating that, rather than simply dying, brain cells exposed to chronic stress and aging gradually lose their intrinsic identities (1-4).

Preserving cellular identity is essential for the proper functioning of multicellular organisms. Highly differentiated cells, such as epithelial cells, neurons, and endothelial cells, are maintained in their specific roles by strict epigenetic and transcriptional controls. Nevertheless, cellular states maintain a degree of plasticity, enabling tissues to develop, repair, and adapt. Although this plasticity is essential for survival, it can be exploited during aging and disease pathogenesis (5, 6). Two critical concepts that define this phenotypic fluidity are the epithelial-mesenchymal transition (EMT) and the more recently described mesenchymal drift (MD) (7-10). EMT is a highly orchestrated, genetically programmed switch essential for embryonic development and tissue repair, but it is frequently co-opted in organ fibrosis and carcinoma progression (8, 11). In contrast, MD is a stochastic, nonadaptive loss of cellular identity caused by aging-related epigenetic noise and chronic environmental stress (10, 12). EMT is limited to specific lineages and involves distinct morphological changes, whereas MD is a systemic, lineage-independent process that significantly affects non-epithelial brain cell types, such as astrocytes and oligodendrocytes (7, 12).

This review explores MD as an emerging concept to explain the widespread loss of cell-type-specific transcriptional profiles and their replacement by a generic mesenchymal gene program. We first set out the operational definition of MD and differentiate it from the regulated EMT. We then examine the evidence linking MD to an uncoordinated loss of identity across neuronal, endothelial, and glial cells in the AD brain. We conclude that interpreting AD through the lens of MD requires a reassessment of transcriptomic data, highlighting that gradual loss of cellular identity, rather than acute cell death alone, may be central to neurodegenerative pathology.

Architecture and molecular program of the epithelial-mesenchymal transition

The conceptual acceptance of EMT was greatly advanced in 1982 by a seminal discovery by Gary Greenburg and Elizabeth D. Hay (13). Previously, cell biology was dominated by the assumption that fully differentiated epithelial cells were fixed in their identity and could not change their phenotype (14). Greenburg and Hay introduced a three-dimensional collagen gel culture system, demonstrating that epithelial cells could break down their intercellular junctions, lose apical-basal polarity, stretch pseudopodia and filopodia, and migrate as individual cells. This finding revealed that the extracellular environment can unlock cellular plasticity, thereby establishing the basic concept of EMT. This concept is crucial for understanding cellular plasticity in metazoan biology. Epithelial cells are characterized by tight intercellular junctions, including adherens junctions, tight junctions, and desmosomes, which together form a continuous cell sheet anchored to a basement membrane (15-17). These cells exhibit distinct apical-basal polarity, characterized by an asymmetric distribution of membrane proteins and cytoskeletal elements. This asymmetry maintains structural integrity and enables directional transport across tissue barriers. Adherens junctions are primarily maintained by E-cadherin, a transmembrane glycoprotein. Its homophilic extracellular interactions bind adjacent epithelial cells, while its intracellular domain connects to the actin cytoskeleton via catenins. This complex functions as both a physical adhesive and a signaling hub, preserving the quiescent, differentiated epithelial state (18, 19). During EMT, this architecture undergoes extensive biochemical and structural reorganization. Cell-to-cell and cell-to-matrix contacts are systematically dismantled, beginning with the functional downregulation of E-cadherin, a process known as the cadherin switch. Simultaneously, mesenchymal cadherins, particularly N-cadherin, are upregulated. N-cadherin mediates weak, more transient homophilic interactions, thereby facilitating dynamic cellular remodeling and increased motility (8, 20). A parallel process, endothelial-to-mesenchymal transition (EndoMT), occurs in the vascular endothelium, leading to the functional loss of vascular endothelial (VE)-cadherin (21, 22). The loss of apical-basal polarity is accompanied by the acquisition of front-rear polarity as the cell elongates and reorganizes its actin cytoskeleton to form specialized protrusive structures, such as lamellipodia and filopodia, at the leading edge (8, 23). Within the cell, epithelial intermediate filaments, composed of cytokeratins, are replaced by vimentin. Vimentin provides the mechanical strength necessary for active migration through dense extracellular matrices. Additionally, cells undergoing EMT secrete matrix metalloproteinases (MMPs). These MMPs degrade the basement membrane, enabling the cells to delaminate from the original tissue layer and invade the surrounding stroma. This phenotypic transformation is regulated by a multi-tiered network operating at transcriptional, post-transcriptional, and post-translational levels (24, 25). Key regulators of this network are EMT-inducing transcription factors, particularly members of the SNAIL, ZEB, and TWIST families (8, 26). For more details on the molecular and epigenetic mechanisms orchestrated by these regulators, see other reviews on this topic (8, 27, 28).

Epithelial-mesenchymal transition in development, tissue repair, and disease

During early embryonic development, stationary epithelial cells break down their rigid intercellular junctions and adopt a highly motile, spindle-shaped mesenchymal phenotype, enabling complex tissue reorganization. This developmental EMT, driven by diverse morphogenetic signaling pathways, permits cells to migrate over long distances and coordinate critical events such as gastrulation and neural crest cell dissemination (8). Importantly, this process is not unidirectional; after migration, cells undergo mesenchymal-epithelial transition (MET) to re-form structured, differentiated epithelial tissues (8, 29). Thus, both EMT and MET are indispensable and mutually complementary processes essential for embryonic development. Beyond embryogenesis, this cellular plasticity persists in adult organisms as a crucial mechanism for maintaining physiological homeostasis and facilitating tissue repair. When an epithelial barrier is compromised, local signals prompt cells at the wound edge to enter a transient, metastable mesenchymal state, enabling collective migration to seal the defect (8, 30). In this physiological context, EMT is a self-limiting response: once inflammatory signals diminish and tissue integrity is restored, cells return to their quiescent epithelial state. However, chronic microenvironmental stress, such as sustained inflammation or a malignant tumor stroma, can disrupt this regulation, leading to severe pathologies. In cancer, tumor cells use EMT not only to acquire the motility required to breach basement membranes and invade adjacent tissues, but also to enhance survival. Pathological EMT enables cancer cells to suppress senescence, resist apoptotic cell death, and develop immunosuppressive properties. These changes allow malignant cells to evade immune surveillance and establish a protective, immunotolerant microenvironment that supports their survival during systemic metastasis (11, 26, 27, 31). The cellular plasticity seen in cancer metastasis and progressive organ fibrosis can be interpreted not as a new disease mechanism, but as a reactivation of embryonic developmental programs in adults (8). This perspective, that pathological states may recapitulate developmental transitions, carries potential clinical implications. Focusing on the broader dynamics of cellular plasticity, rather than on individual signaling molecules, may enable the development of new therapeutic strategies. For example, strategies that block EMT pathways could neutralize the invasive and immunosuppressive properties of cancer cells. Alternatively, inducing MET could revert metastatic cells to a benign, drug-susceptible epithelial state.

Mesenchymal drift represents an entropic decay of cellular identity distinct from the orchestrated epithelial-mesenchymal transition

MD is a recently defined paradigm in cellular aging and degenerative pathology (7, 12). While classical developmental biology emphasized coordinated, deterministic lineage switches, recent research has focused on the systemic, progressive erosion of cell-type-specific identity throughout the lifespan. Large-scale, unbiased meta-analyses of single-cell RNA sequencing (scRNA-seq) datasets from various tissues and age-related disease models have revealed a conserved transcriptomic trajectory: as organisms age or experience chronic degenerative conditions, specialized non-mesenchymal cell types systematically lose their unique gene expression profiles and concurrently upregulate a generic mesenchymal and profibrotic gene program (7). Lu et al. define MD as a state in which two components are documented: a compromised original cellular identity and the acquisition or intensification of mesenchymal traits (7). Under this criterion, loss of cell identity and epigenetic erosion correspond to the first component alone and do not, by themselves, establish MD. Cellular dedifferentiation, by contrast, describes movement toward a less mature state along a cell’s own lineage and does not require the acquisition of mesenchymal traits.

Unlike EMT, an adaptive evolutionary program that achieves specific morphological outcomes, MD is a nonadaptive, entropic feature of biological aging. MD is driven by the progressive accumulation of epigenetic noise, including loss of heterochromatin stability, abnormal DNA methylation, and altered histone modifications, such as the loss of repressive marks like H3K9me3 or H3K27me3, resulting from lifelong exposure to systemic stressors (10, 12, 32-34). These stressors include chronic low-grade inflammation (“inflammaging”), the ongoing production of reactive oxygen species (ROS), and proteotoxic stress (34-39). Under persistent stress, the epigenetic boundaries that maintain cell-type-specific gene repression become compromised. This leads to the widespread derepression of genes typically silenced in specialized parenchymal cells, including those encoding fibrillar collagens, extracellular matrix remodeling proteins, and inflammatory secretomes (12).

This transcriptomic drift is closely associated with clinical severity. Individuals with elevated MD signatures in diseases such as idiopathic pulmonary fibrosis, liver cirrhosis, and advanced renal disease exhibit accelerated disease progression, impaired organ function, and elevated mortality rates (7). A key feature of MD is its notable susceptibility to therapeutic intervention through partial epigenetic reprogramming. Recent studies indicate that such reprogramming can resolve MD’s transcriptomic signatures by eliminating accumulated epigenetic noise and suppressing the hyper-activated mesenchymal gene program (7). Therefore, reversing this transcriptomic drift through epigenetic reprogramming offers a promising strategy to mitigate age-associated pathologies and restore tissue homeostasis (40-42).

Distinguishing mesenchymal drift from the epithelial-mesenchymal transition

EMT and MD are distinct biological processes, though both involve acquiring mesenchymal transcriptomic traits. A primary distinction lies in their target cell populations: classical EMT is restricted to organized epithelial and endothelial layers undergoing physical transformation, whereas MD is lineage-agnostic and systemic, occurring in diverse cell types beyond epithelial origins (8, 12). In advanced neurodegenerative conditions such as AD, MD is particularly evident in specialized brain parenchymal cells, including astrocytes and oligodendrocytes, shifting them from homeostatic, neuroprotective states to dysfunctional, fibrotic-like transcriptomic profiles (7, 12). Additionally, the nature of these processes and their primary triggers are fundamentally opposed. EMT functions as a highly programmed, biologically orchestrated switch, activated by specific developmental or pathological pathways. It relies on defined signaling cascades, such as acute transforming growth factor β (TGF-β), Wnt, and Notch, as well as master transcription factors, including SNAIL and TWIST, to achieve anatomical restructuring (8). In contrast, MD is a stochastic, entropic, and maladaptive process marked by the gradual accumulation of epigenetic noise. It is triggered by chronic environmental stressors, including aging, ROS, endoplasmic reticulum (ER) stress, senescence-associated secretory phenotype (SASP), and chronic inflammation (7, 12). Although molecular factors such as TGF-β are involved in both processes, their timing differs. In EMT, TGF-β acts as a transient, programmed cue, whereas in MD, it functions as a persistent, low-grade stressor within the SASP, gradually degrading epigenetic features (8, 12). In summary, EMT represents a coordinated cellular transformation, while MD reflects an uncoordinated decay of cellular identity. A critical distinction lies in the resulting phenotypes, their reversibility, and the extent of supporting morphological evidence. EMT generates highly mobile, resilient cells capable of systemic spread and colonization. This process is naturally reversible via MET when inductive signals, such as those at metastatic sites, are withdrawn (8). In contrast, MD produces dysfunctional, fibrotic-like cells that degrade local tissue architecture and impair organ function. Reversal appears more challenging; targeted epigenetic interventions, such as partial reprogramming with Oct4, Sox2, Klf4, and c-Myc (OSKM), have been shown to reduce an MD signature (7, 12) (Fig. 1). While EMT is a well-characterized structural transformation supported by extensive morphological validation, MD is currently defined primarily by transcriptomic profiles. Although single-cell transcriptomics revealed the upregulation of motility and extracellular matrix-associated genes in drifting cells, it remains unclear whether these non-epithelial cells acquire true mesenchymal motility. Therefore, further morphological and functional validation is needed to bridge the genotype-phenotype gap and understand the physical consequences of MD.

Fig. 1.

Conceptual comparison of cellular identity transitions: epithelial-mesenchymal transition vs. mesenchymal drift. Epithelial-mesenchymal transition (EMT) is an orchestrated, programmed, and naturally reversible process restricted to specific epithelial or endothelial lineages and triggered by defined factors such as TGF-β, Wnt, and SNAIL. In contrast, mesenchymal drift (MD) represents a stochastic, entropic decay of cellular identity, driven by chronic stressors like aging, ROS, and epigenetic noise. Unlike EMT, MD is a maladaptive and uncoordinated process toward a de-identified, fibrotic-like state. Partial reprogramming with OSKM is shown here as one demonstrated approach that reduces an MD signature.

Evidence for mesenchymal drift and loss of cellular identity across neuronal, endothelial, and glial cells in Alzheimer’s disease

Historically, AD progression has been characterized by the linear accumulation of Aβ plaques and neurofibrillary tangles, ultimately leading to neuronal death (43-45). However, recent advancements in scRNA-seq, epigenomics (e.g., single-cell assays for transposase-accessible chromatin with sequencing [scATAC-seq]), and multi-omics have reshaped this perspective. Current research suggests that brain cells do not simply survive or die; rather, they experience a profound loss of cellular identity under disease conditions (2, 46). This phenomenon can be interpreted within the MD framework, which describes an entropic decay driven by chronic environmental stress, metabolic alterations, and epigenetic erosion, rather than a coordinated biological process. In AD, changes consistent with this framework have been reported across multiple brain cell types, including premature neuronal differentiation, pathological glial reactivity, and vascular fibrosis. The strength of this evidence differs by cell type. In a reanalysis of published single-nucleus RNA-sequencing data from AD and control brain, Lu et al. found significant upregulation of MD genes in astrocytes, oligodendrocytes, and pericytes, but not in neurons, microglia, or oligodendrocyte progenitor cells (OPCs); this pattern persisted after restricting the comparison to age-matched donors (7). Direct, signature-level evidence for MD in the AD brain is therefore established for these three cell types alone. This does not exclude MD in the others, given the limited power of snRNA-seq for sparsely represented cell types and the dependence of any enrichment score on the reference gene set. We therefore distinguish this direct evidence from other observations of neuronal identity loss or microglial reactivity, which are consistent with MD but have not been tested against the signature itself.

Neural progenitors and mature neurons: identity loss with limited evidence of mesenchymal gain

Epigenetic mechanisms, particularly the repression of inappropriate genes by DNA methyltransferases like DNMT3A, are crucial for maintaining highly specialized neuronal states. However, in the AD brain, these regulatory controls often fail. This failure may involve abnormal DNA demethylation and derepression by ten-eleven translocation (TET) enzymes, which leads to transcriptomic instability and the loss of cellular boundaries (47, 48). Loss of REST, a key transcriptional repressor, further permits inappropriate gene expression (49, 50). Meyer et al. showed that this epigenetic breakdown drives human-induced pluripotent stem cell (hiPSC)-derived neural progenitor cells (NPCs) into premature differentiation, potentially depleting the progenitor pool in both sporadic and APOE4-associated AD models (51). Expanding on progenitor exhaustion, Cho et al. found that the APPswe mutation increases mitochondrial ROS in the same model system, which is necessary but not sufficient on its own to prematurely drive NPCs out of their proliferative state, leaving them vulnerable and immature (52). These mutant NPCs display disrupted transcriptomic profiles, sharing expression signatures even with non-neuronal cells such as astrocytes and oligodendrocyte progenitor cells (OPCs). In our own reanalysis, gene set enrichment analysis (GSEA) reveals notable enrichment for the MD signature in APPswe NPCs, although acquisition of phenotypic mesenchymal properties remains untested (Fig. 2). Furthermore, control NPCs sorted for high mitochondrial ROS also exhibit a significant, though less pronounced, enrichment for this MD signature. These results suggest that elevated ROS may be a major contributor to transcriptomic drift, which the APPswe mutation further exacerbates. Notably, because NPCs are rarely represented in adult post-mortem brain tissue, this in vitro result addresses a population that a previous in vivo analysis could not test (7).

Fig. 2.

Transcriptomic evidence of mesenchymal drift in Alzheimer’s disease neural progenitor cells. (A) The schematic outlines the experimental design comparing control and APPswe mutant hiPSC-derived neural progenitor cells (NPCs). (B) Gene set enrichment analysis (GSEA) shows significant upregulation of the MD signature in APPswe NPCs relative to controls. (C) The experimental workflow details NPC isolation based on mitochondrial ROS levels. Control NPCs were sorted into Low-ROS and High-ROS groups using fluorescence-activated cell sorting (FACS) with MitoSOX staining. (D) GSEA demonstrates that High-ROS NPCs are highly enriched for the MD signature compared to Low-ROS NPCs. The GSEA results presented in this figure were generated through our independent reanalysis of previously published datasets (52).

This lineage smearing, in which cells lose distinct boundaries and adopt mixed transcriptomic programs, is consistent with the uncoordinated identity decay observed in MD. Mature neurons are also affected: Lucciola et al. reported that AD neurons upregulate early developmental and cancer-associated transcription factors, especially RUNX1 and YY1 (53). These changes illustrate how chronic stress may drive mature neurons to lose their specialized functions and drift toward a dedifferentiated state. Although these findings establish a strong transcriptomic basis for neuronal identity loss, significant gaps remain in directly linking functional neuronal abnormalities to transcriptomic drift. For example, Ghatak et al. found that hiPSC-derived AD neurons exhibit marked functional deficits, including hyperexcitability, increased sodium current density, and reduced neurite length (54). Similarly, Cho et al. observed enhanced neuronal activity and firing rates in AD cerebral organoids (52). However, hyperexcitability might reflect an acute response to toxicity or localized ion channel disruption, rather than a sustained loss of transcriptional identity. Therefore, further research is needed to determine whether these functional changes correlate with transcriptional drift at the single-cell level, as this alignment is crucial for distinguishing MD from acute physiological state changes.

Endothelial cells: transcriptomic drift and barrier failure

Cellular identity drift also significantly affects brain endothelial cells, driving blood-brain barrier (BBB) breakdown and neuroinflammation. Brain endothelial cells normally maintain the BBB through tight junctions and junctional proteins like VE-cadherin. However, with aging and chronic inflammation, these cells show reduced epithelial and endothelial characteristics. They begin to express mesenchymal markers such as fibronectin and α-SMA, accompanied by BBB disruption and vascular fragility. These changes reflect transcriptomic and metabolic reprogramming (21, 22, 55, 56). hiPSC-derived BBB models that reconstitute endothelial, mural, and astrocytic compartments have been shown to recapitulate barrier properties and AD-associated vascular pathology (57). Studies using such models indicate that chronic inflammatory conditions, including exposure to NF-κB-activating cytokines, cause morphological remodeling, junctional disruption, and significant lipid accumulation as well as transcriptomic shifts in these cells. These transcriptomic changes closely resemble the ‘endoM2’ signature, a pathological transcriptomic state found in AD brain vasculature characterized by inflammatory and EndoMT-associated gene expression and an inverse correlation with cognition (58). These findings suggest that inflammation could promote a dysfunctional, mesenchymal-like state in endothelial cells. Transcriptomic evidence of this EndoMT-like state observed in the AD brain is associated with organelle-level stress and alterations in energy metabolism, not merely structural changes. Analyses of AD mouse models and human AD brain single-cell datasets indicate that mitochondrial and endoplasmic reticulum (ER) stress are associated with the expression of the transcription factors FOSB and MEOX1, potentially leading to upregulation of genes involved in lipid oxidation and oxidative phosphorylation (59). This metabolic shift has been proposed to contribute to the mesenchymal component of endothelial drift and to a profibrotic transcriptional program, although fibrosis itself was not assessed histologically (59). This is consistent with the concept that MD reflects the accumulation of epigenetic and metabolic noise under chronic stress.

Glia: reactive states and the boundary with mesenchymal drift

Glial reactivity has traditionally been viewed as an adaptive polarization in response to neuroinflammation, aimed at clearing debris and isolating plaques (60, 61). This view describes a self-limiting response to acute insult. Under chronic stimulation, however, the reaction does not resolve, and the MD framework interprets this sustained state as pathological transcriptomic drift, where glial cells progressively lose their homeostatic support functions. In chronic neurodegenerative conditions such as brain injury or AD, reactive glia alter their neurotrophic and synaptic regulatory roles and instead engage in TGF-β-associated programs that promote extracellular matrix (ECM) production and proliferation (62). Sustained activation of these programs may represent a glial counterpart to mesenchymal transition and can contribute to fibrotic scar formation that persists within cores and impedes neural regeneration (62). High-resolution transcriptomic profiling supports this entropic decay. Mathys et al. and Soreq et al. found that, as AD advances, specific astrocyte and microglial subpopulations undergo extensive transcriptomic reprogramming. Rather than mounting a coordinated defense, these cells downregulate neuro-supportive genes and adopt a generalized stress response (1, 4). Some disease-associated astrocytes upregulate genes for extracellular matrix (ECM) remodeling, transitioning from a supportive to a profibrotic, scar-forming state. Similarly, certain microglial subpopulations markedly increase osteopontin (SPP1) expression (2), a factor shown to drive a mesenchymal, profibrotic phenotype in other cell types (5). Murdock and Tsai further highlighted this lineage-agnostic collapse, noting that even specialized cells, such as cortical interneurons, lose their canonical markers, reflecting a widespread loss of transcriptional identity in the brain parenchyma (2).

De-identified cells and therapeutic implications

The most severe consequence of losing glial and neuronal identity is the emergence of de-identified cells. Xiong et al. mapped late-stage AD brains and observed a significant increase in cell populations lacking clear lineage markers, which were linked to severe epigenetic erosion (46). This includes the random opening of repressed heterochromatin and the closure of active promoters. Such epigenomic smearing across astrocytes, microglia, and oligodendrocytes challenges the traditional neuroinflammation model. Rather than an organized immune response, glial changes in late AD may instead reflect a loss of epigenetic control that permits differentiated cells to deteriorate toward a generic, dysfunctional, mesenchymal-like state (Fig. 3).

Fig. 3.

Mesenchymal drift and loss of cellular identity across brain cell types in Alzheimer’s disease. Neuronal cells: NPCs undergo premature differentiation, leading to immature neurons and transcriptomic lineage smearing. Mature neurons exhibit epigenetic breakdown and the upregulation of mesenchymal-associated transcription factors, such as RUNX1 and YY1, as well as phenotypic and functional deficits. Endothelial cells: Aging and chronic inflammation drive brain endothelial cells toward a mesenchymal-like state, in which they lose endothelial identity markers such as VE-Cadherin and acquire mesenchymal markers (fibronectin, α-SMA). This shift is accompanied by the “endoM2” signature, an NF-κB-associated transcriptomic module enriched for EndoMT regulators, and can drive BBB disruption and vascular fibrosis. Glial cells: Homeostatic astrocytes and microglia shift to reactive states, accompanied by extensive transcriptomic reprogramming. Astrocytes downregulate neuro-supportive genes and mount a stress response, with TGF-β signaling activation and ECM remodeling. Concurrently, microglia upregulate tissue fibrosis genes such as SPP1. In late AD, this widespread epigenetic erosion may lead to the emergence of de-identified, dysfunctional, and fibrotic-like cells that lack clear lineage markers.

As mesenchymal drift in vascular, neuronal, and glial cells has been proposed as a contributing factor in AD pathology, strategies to block this process and preserve cellular identity are being explored as novel therapies. For example, repressing ARF6, which regulates actin-dependent protrusive activity and vascular stability (63), inhibits EndoMT in endothelial cells, stabilizes the BBB, and reduces neuroinflammation in a mouse model of neuroinflammatory demyelination (64, 65). This suggests that restoring transcriptomic stability could be a promising therapeutic approach for AD, though this remains to be tested directly in AD models. Further research beyond computational and transcriptomic analyses is also needed to confirm glial reactivity as a feature of MD. Upregulated mesenchymal markers must be validated through in vivo physical, structural, and morphological studies. It remains unclear whether transcriptomically drifting astrocytes and microglia undergo true cytoskeletal changes, dismantle their networks, or deposit fibrotic matrices in the brain, as mesenchymal cells do in other tissues. Bridging the gap between transcriptomic changes and physical morphology is essential to validate MD as a key driver of neurodegenerative pathology.

DISCUSSION

Distinguishing adaptive reactive states from entropic decay

The MD framework introduces a new approach to understanding AD pathogenesis. Recent high-resolution genomic studies reveal significant parallels between AD and MD at the transcriptomic and epigenomic levels, including stochastic loss of epigenetic boundaries, uncoordinated lineage transitions, and cellular dedifferentiation (7, 46). Important distinctions, however, remain. In peripheral organs, MD manifests as visible fibrosis and extensive extracellular matrix accumulation, whereas the structure of the brain may constrain such changes. How MD would manifest in brain tissue is therefore unresolved; vascular instability and glial scar formation are plausible correlates, since both involve ECM remodeling, whereas any relationship to synaptic loss is unclear. AD-related MD may also be shaped by brain-specific stressors, such as Aβ and tau proteotoxicity. Taken together, these considerations argue for treating AD as a progressive, staged loss of cellular identity rather than as a binary outcome of altered function versus cell death.

Over the past decade, single-cell transcriptome studies have consistently revealed an increase in disease-associated cell populations and a decrease in healthy ones (1-4). From an MD perspective, it is critical to acknowledge the biphasic nature of these cellular states. In the early stages of AD, transcriptomic signatures likely indicate a highly programmed, adaptive defense mechanism, mediated by specific pathways like TREM2 signaling. Such processes generate reactive populations, such as disease-associated microglia (DAM) and astrocytes (DAA) (66-68). As the disease progresses to late stages under prolonged chronic stress, these initially protective responses turn into chronic, irreversible processes (69), and cellular identity itself begins to collapse (46). Consequently, the emergence of seemingly novel transcriptomic clusters may not indicate new cell types or functional adaptation. Instead, it could reflect resident brain cells losing their canonical markers and adopting ambiguous identities. Nonetheless, computational clustering algorithms may assign these cells to apparently novel reactive populations. Distinguishing genuine adaptive states from entropic decay is therefore essential to interpreting such data.

Technical explanations for apparently de-identified clusters

Some of the apparently de-identified or intermediate clusters described above may also have technical origins that are worth considering alongside a genuine loss of identity. Low-quality nuclei, marked by elevated mitochondrial or ambient RNA content and low library complexity, tend to lose cell-type-specific signal and aggregate into intermediate-looking clusters. Ambient RNA released during tissue dissociation imposes a shared, lineage-blurring background across droplets, and doublets generate clusters that carry markers of more than one lineage, a pattern that can resemble the lineage smearing described here. Dissociation itself can also induce ex vivo activation programs in glia that are difficult to distinguish from genuine disease-associated states. On the quantification side, reads from highly similar duplicated genes and paralogues cannot always be uniquely assigned, and donor-specific genomic variation can reduce mapping efficiency against a single linear reference, so an apparent loss of a marker gene may in part reflect reference bias rather than transcriptional silencing. Distinguishing genuine identity loss from these artifacts will require orthogonal evidence, such as matched chromatin accessibility profiling, in situ validation of marker loss, or replication across independent cohorts and dissociation protocols.

Limitations of stem cell-based models for a proposed aging mechanism

A further caveat concerns the stem cell-based models that underlie part of the evidence reviewed above. Reprogramming to pluripotency resets donor age-associated epigenetic and transcriptional features and returns cells to an embryonic-like state. hiPSC-derived neural progenitors and organoids therefore largely lack the aging context in which MD is proposed to arise, and findings from such models cannot on their own be read as demonstrations of age-driven drift (70, 71). This caveat applies unevenly across the studies cited here. The RUNX1 and YY1 evidence for neuronal drift was obtained from induced neurons generated by direct lineage conversion of adult donor fibroblasts, a method that bypasses the pluripotent state and, unlike hiPSC-derived neurons, retains the donor’s transcriptomic and epigenetic aging signature (53). This evidence is therefore less confounded by rejuvenation. By contrast, the neural progenitor data showing APPswe-driven premature differentiation and MD signature enrichment were obtained in hiPSC-derived NPCs (52). We accordingly describe this MD-like signature as a phenocopy driven by the APPswe genotype and elevated mitochondrial ROS rather than as a demonstrated consequence of chronological aging. Two partial mitigations deserve mention, though neither is sufficient on its own. Prolonged in vitro culture is itself a stressful environment, and sustained oxidative, inflammatory, and proteotoxic load may recapitulate some of the drivers that produce drift in vivo (72). In addition, because the mechanistic claim underlying MD concerns chronic stress and epigenetic noise rather than chronological age as such, stress-induced drift in a rejuvenated background can still be informative about mechanism. Closing this gap will require defined long-term culture or chronic-stimulation paradigms benchmarked directly against aged human tissue, using epigenetic clocks, chromatin accessibility profiling, and biochemical readouts of senescence and proteostatic decline, rather than assumed to recapitulate aging (72).

From transcriptomic drift to functional validation

Even where a transcriptomic shift is technically sound and biologically genuine, adopting a new transcriptomic profile does not necessarily result in a corresponding functional change. A cell may display the RNA signature of a motile or fibrotic cell without exhibiting the phenotype that such a signature would predict in vivo: actual migration, matrix deposition, or, in the case of neurons, a loss of specialized function. AD neurons illustrate this dissociation; rather than becoming functionally quiescent, they exhibit aberrant and often excessive activity in vivo (73, 74). Researchers should interpret single-cell transcriptomic data with caution in this regard, as equating transcriptomic shifts directly with physiological changes may misrepresent the underlying disease mechanisms. Establishing a direct causal link between transcriptomic drift and disease pathology requires rigorous experimental validation to determine whether these shifting transcriptomic identities lead to morphological and physiological changes in the brain. For instance, combining spatial transcriptomics (such as Multiplexed Error-Robust FISH [MERFISH]) with high-resolution imaging can verify whether MD signatures correspond to actual structural remodeling within pathological niches such as Aβ plaques (75). Advanced techniques such as Patch-seq can also directly correlate single-cell transcriptomic drift with electrophysiological abnormalities, including hyperexcitability (76). Furthermore, lineage tracing, utilizing in vitro 3D brain organoid models or in vivo models, can determine whether drifting cells genuinely acquire physical motility (77). By integrating genomic profiling with such multimodal functional validations, we can clarify whether these drifting identities drive neurodegeneration or merely reflect background epigenetic instability.

Concluding remarks

Taken together, the MD framework at this stage provides a way of classifying evidence rather than an explanation of AD pathogenesis. It supplies a defined, two-component criterion against which observations of glial reactivity, neuronal dedifferentiation, and endothelial transition can be tested, and it makes explicit which of these currently meet that criterion and which do not. Framing the question in these terms makes clear what evidence would be required to determine whether MD contributes to AD progression or merely reflects background epigenetic instability.