Atavistic Genetic Expression Dissociation (AGED) During Aging: Meta-Phylostratigraphic Evidence of Cellular and Tissue-Level Phylogenetic Dissociation

Aging is commonly attributed to accumulated damage, or evolved antagonistic genetic trade‐offs, which lead to an accumulation of damage causing misexpression of genes necessary for longevity. We propose an atavistic dysregulation of gene expression at cellular and tissue levels during aging, framing aging as a gradual regression toward ancestral cellular states. Similarly to the atavistic model of cancer, in which cells revert to unicellular‐like behavior, aging may result from the breakdown of coordinated morphogenetic control, leading organs and tissues toward less integrated, ancient unicellular states. We suggest that aging may involve a progressive reversal of the well‐known ontogenetic tracing of prior phylogenetic embryonic characteristics. Moreover, aging could involve a loss of large‐scale coordination, with tissues reverting to ancient gene expression to different degrees. We tested this hypothesis using a meta‐phylostratigraphic analysis, finding: (1) An atavistic over‐representation of differential expression in the most ancient genes and under‐representation in the evolutionary youngest genes for two multi‐tissue aging databases, and tissues covering skin, ovarian, immune, senescent and mesenchymal‐senescent cells; (2) No significant atavistic over‐representation of the differential gene expression during aging of brain cells and mesenchymal stem cells; (3) overall age‐dependent increase of heterogeneity in the direction of the phylogenetic position of tissues' transcriptional profiles; (4) and an overall negative evolutionary age mean shift toward the most ancient genes. Our analyses suggest that aging involves uncoordinated and tissue‐specific phylogenetic changes in gene expression.

Understanding aging as a structured, heterogeneous atavistic process opens new avenues for rejuvenation, focusing on restoring multicellular coherence in evolutionarily youthful gene expression.

Most complex organisms experience a sequential life course consisting of embryonic development stages, growth to maturity, and then functional decline and ultimately death. Aging leads to progressive deterioration of cellular structures and biological functions in living systems (de Magalhães 2011, 2023; Gladyshev et al. 2021; Austad and Hoffman 2018). The aging process is influenced by diverse elements including genetic inheritance, environmental conditions, and behavioral choices (Partridge et al. 2018). The resistance of biological systems to noise and damage decreases with age (Rubin 2007, 2006; Rubin et al. 1996), leading to health issues including cardiovascular complications, cancer, neurodegenerative disorders, metabolic disruptions such as type II diabetes, and compromised immune responses to pathogens, which strongly impact quality of life (Gladyshev et al. 2021; Partridge et al. 2018; López‐Otín et al. 2023; Hayflick 2007).

The scientific community has been focused on two main theoretical frameworks to understand the aging process: damage‐based (Gladyshev et al. 2021; López‐Otín et al. 2023) and programmatic (de Magalhães 2011, 2023; Gladyshev et al. 2021; Kirkwood and Melov 2011; Lidsky et al. 2023, 2022; Lidsky and Andino 2022, 2020) theories. Damage‐based theories hold that aging results from the inevitable accumulation of molecular damage over time. Essential cellular components including genetic stability, telomere length, mitochondrial activity, and protein homeostasis experience severe degradation. In contrast, programmatic theories propose that aging is the evolutionary result of a genetically encoded biological trajectory. This perspective suggests that the aging process is encoded in our genome, because it provides a reproductive advantage at the lineage level, rather than merely representing stochastic damage accumulation over time (de Magalhães and Church 2005; Gems 2022; Skulachev and Skulachev 2014).

A novel direction with respect to aging is suggested by a focus on morphogenesis and morphostasis (cellular self‐assembly into complex anatomical forms, and continued maintenance of the correct structure) as active navigation of anatomical morphospace (Fields and Levin 2022; Harris 2018; Pezzulo and Levin 2016, 2015; Rasskin‐Gutman and Izpisua‐Belmonte 2004; Stone 1997). This framework sees embryogenesis, regeneration, and cancer suppression as a continuous dynamic set of decisions made by the cellular collective to reach and maintain organ‐level target morphologies (Fields and Levin 2022; Pezzulo and Levin 2015; Levin 2023a). Numerous tools from behavioral neuroscience and cybernetics have been deployed in morphogenetic systems to probe the mechanisms that coordinate cells toward common anatomical endpoints (Pezzulo and Levin 2016, 2015; Levin 2023b; Friston et al. 2015). The ability to reach and maintain species‐specific large‐scale anatomical states reliably, despite noise and perturbations, is a kind of homeostatic dynamic that has been modeled as a collective intelligence (McMillen and Levin 2024), with many applications in biomedicine across birth defects and regeneration (Lagasse and Levin 2023; Levin 2024). Moreover, this approach has provided novel ways to address cancer as a loss of coordination of cells toward tissue‐level homeostatic goals (Levin 2021; Moore et al. 2017; Chernet and Levin 2013). From this perspective, one can ask what kind of information‐processing disorders could contribute to aging as a long‐term loss of the ability to organize cells and subcellular materials toward maintenance of a healthy complex form.

We recently proposed that aging is the result of the loss of morphostatic information (Pio‐Lopez and Levin 2024; Pio‐Lopez et al. 2024). In this framework, aging is an emergent failure mode of information‐processing collectives that lose morphogenetic guidance after the construction of the adult body structure has been completed (Pio‐Lopez and Levin 2024; Pio‐Lopez et al. 2024). What kind of directional changes might be observed in cells and tissues after the primary anatomical goals have been met, in addition to loss of precision and random alterations? Moreover, when not bound to a single (body‐wide) anatomical setpoint as occurs during embryogenesis, would different cells in the body eventually adopt distinct setpoints, thus resulting in the loss of coherent function as observed in aging?

The setpoints for anatomical homeostasis in evolved creatures are shaped by evolution (Newman and Bhat 2009; McGhee 2007; Ollé‐Vila et al. 2016; Duran‐Nebreda et al. 2016; Raff 1996; Gilbert et al. 1996; Jeffery and Raff 1983); specifically, they are determined by phylogenetic position and the body plan it specifies. Could the kind of dissociation of self‐model from reality that is observed in certain psychological disease states (Nijenhuis et al. 1998; Waller et al. 2001) have a somatic counterpart, in which the actual phylogenetic age of cells ( Homo sapiens ) becomes dissociated from the effective physiological phylogenetic age due to the expression of more ancient genes? We sought to ask: what is the relationship between an organism's age and the phylogenetic information its tissues attempt to implement? Using transcriptional profiling datasets, and phylostratigraphy (Domazet‐Lošo et al. 2007; Domazet‐Lošo and Tautz 2010; Lineweaver and Davies 2021) to analyze the evolutionary age of genes being expressed in specific tissues, we tested two hypotheses. First, that aging involves a roll‐back of transcriptional profiles to include more ancient genes (in effect, a reversal of the well‐known “ontogeny recapitulates phylogeny” dynamic with respect to embryonic patterns). Second, that this would include a significant loss of coordination across the body, in which different tissues ended up phenocopying the transcriptional states of different positions across phylogenetic history, similarly to what occurs in cancer (Levin 2021; Bussey et al. 2017; Davies and Lineweaver 2011; Rubin 1985).

We tested this hypothesis using a meta‐phylostratigraphic analysis. We applied a phylostratigraphic analysis using RNA‐seq and scRNA‐seq data from 8 different studies and two meta‐analyses including different tissues RNA‐seq during aging (Palmer et al. 2021; Tikhonov et al. 2024), scRNA‐seq data from skin cells (Zou et al. 2021), an aging meta‐analysis of the scRNA‐seq of senescent cells (Avelar et al. 2020), and brain (cortex, hippocampus and cerebellum cells) (González‐Velasco et al. 2020), immune (Lu et al. 2022), ovarian (Jin et al. 2024) and stem cells RNA‐seq (Wagner et al. 2009) (see Table 1). We found: (1) An atavistic over‐representation of differential expression in the most ancient genes and under‐representation in the evolutionary youngest genes for two multi‐tissue aging databases, and tissues covering skin, ovarian, immune, senescent and mesenchymal‐senescent cells; (2) No significant atavistic over‐representation of the differential gene expression during aging of brain cells and mesenchymal stem cells; (3) overall age‐dependent increase of heterogeneity in the direction of the phylogenetic position of tissues' transcriptional profiles; and (4) an overall negative evolutionary age mean shift toward the most ancient genes. Thus, we propose that while cancer is a loss of cellular organization in space, aging involves a loss of cellular organization in (evolutionary) time.

We applied a meta‐phylostratigraphic analysis using RNA‐seq and scRNA‐seq data from 8 different studies two meta‐analyses including different tissues: RNA‐seq during aging (Palmer et al. 2021; Tikhonov et al. 2024), scRNA‐seq data from skin cells (Zou et al. 2021), an aging meta‐analysis of the scRNA‐seq of senescent cells (Avelar et al. 2020), and brain (cortex, hippocampus and cerebellum cells) (González‐Velasco et al. 2020), immune (Lu et al. 2022), ovarian (Jin et al. 2024) and stem cells RNA‐seq (Wagner et al. 2009) (see Table 2).

To analyze the differentially expressed genes (DEGs) in the aging cells under various conditions through phylostratigraphic analysis, we used the evolutionary ages of 19,660 human protein‐coding genes as determined by Litman and Stein (2019). These genes were categorized into 19 major phylostrata, as outlined by Domazet‐Lošo and Tautz (2010). The phylostrata are a hierarchical range of evolutionary origins for: all living organisms (including Eubacteria, Bacteria and their descendants, e.g., unicellulars), Eukaryota, Opisthokonta, Holozoa, Metazoa, Eumetazoa, Bilateria, Deuterostomia, Chordata, Olfactores, Craniata, Euteleostomi, Tetrapoda, Amniota, Mammalia, Boreoeutheria, Eutheria, Euarchontoglires, and Primates.

We quantified the number of genes expressed under the various experimental conditions to assess the atavistic patterns we may find during aging in terms of genetic expression. We then applied an overrepresentation test for each group of genes in each phylostrata compared to the ensemble of ages of all human protein‐coding genes. The overrepresentation test (also known as enrichment analysis) is a widely used statistical method to determine whether a specific set of elements (e.g., genes, proteins, or other features) is significantly over‐represented in a given category compared to a background distribution. Over‐representation tests are extensively used in various scientific fields, particularly in Gene Ontology (GO) Enrichment Analysis where we identify functional categories over‐represented in a set of differentially expressed genes (Ashburner et al. 2000), Pathway Enrichment Analysis where we assess whether biological pathways, such as KEGG pathways, are significantly enriched in a dataset (Kanehisa et al. 2017) and Disease Association Studies where we link genetic variants to diseases by testing for enrichment in disease‐associated categories (Subramanian et al. 2005). This approach is crucial for understanding biological processes, molecular pathways, and functional annotations associated with high‐throughput data.

More specifically, in our case, given a “gene universe” (here the list of all human protein‐coding genes) containing N total genes, this test evaluates whether a specific phylostratigraphic category is statistically overrepresented within a smaller subset of n genes selected from this gene universe in a specific condition. With M genes belonging to a specific phylostratigraphic category in the gene universe, the test calculates the probability of observing k genes from this functional category within the selected subset of n genes randomly, using the hypergeometric distribution: PX=K=MkN−Mn−kNn

A p‐value is then calculated to determine the significance of the over‐representation (X ≥ k) and under‐representation (X ≤ k) of the observed enrichment. Here we use a p‐value inferior to 0.05 to determine the significance, with Benjamini–Hochberg False Discovery Rate (FDR) correction applied to control for multiple testing across all 19 evolutionary age strata within each dataset. We also performed a Mann–Whitney test to assess mean evolutionary age shifts in the overall distribution of evolutionary ages compared to the baseline distribution.

We applied the phylostratigraphic analysis on two multi‐tissue aging datasets. The first signature was extracted from GenAge (Palmer et al. 2021) and has been extracted from 127 publicly available microarray and RNA‐Seq datasets from mice, rats and humans, identifying a transcriptomic signature of aging across species and tissues (brain, heart and muscle). The second one, AgeMeta, is likewise a meta‐analysis of 51 humans scRNA‐seq datasets of different tissues (vastus lateralis, cerebellum, frontal cortex, muscle, brain, adipose tissue, adrenal gland, blood, blood vessel, brain (without cerebellum), esophagus, heart, lung, nerve, pituitary, salivary gland, prostate, testis, and thyroid). We observed in both aging signatures an over‐representation of the most ancient genes in the “All living organisms” strata (see Figure 1). The GenAge aging signature shows more over‐representations, notably in Ospithokonta, Holozoa, Eumetazoa, Bilateria and Euteleostomi strata while the human AgeMeta signature shows only one over‐representation in the unicellular genes. The number of upregulated and downregulated genes in the most ancient genes in the GenAge aging signature is almost equal, suggesting more a reshuffling in the genetic expression during aging in the most ancient genes. The AgeMeta aging signature is dominated by downregulated genes in the “All living organisms” strata.

Interestingly, we also found an under‐representation in the evolutionary youngest genes for the GenAge signature for Mammalia, Boroeutheria, Euarchontoglires, and Primates. Similarly, we also observed an important under‐representation for the AgeMeta signature in Boroeutheria and Primates strata. Overall, the most important changes were noted in the most ancient genes, while we observed an under‐representation of the changes in the youngest ones compared to the expected distribution.

Both meta‐signatures showed significant shifts toward evolutionarily older genes, with mean age shifts of −2.41 and −2.32 categories respectively (see Table 1).

We conclude that there is an atavistic over‐representation of differentially expressed genes in both aging signatures with a heterogeneity in the direction of the genetic changes.

We applied the phylostratographic analysis to skin, ovarian and immune scRNA‐seq. For skin cells, in both age groups, gene expression changes are most significant in the most ancient evolutionary category, for example, “All living organisms”. A shift is observed in gene percentages across aging, with a decrease in downregulated genes in the oldest group. In the 40–70 years age group, genes belonging to “All living organisms” show the highest number of expression changes, with a stronger contribution from downregulated genes. In individuals older than 70, the “All living organisms” and “Euteleostomi” categories showed a significant increase in gene expression changes, particularly upregulation for the latter. A notable decrease in gene expression changes is seen in “All living organisms” in the > 70 group compared to the 40–70 group of fibroblasts with < 35% of the changes in this stratum (see Figure 2). The under‐representation is more widespread for the skin cells. In the 40–70 group of fibroblasts, we have a significant under‐representation for “Eukaryota”, “Bilateria”, “Euteleostomi”, “Boroeutheria” and “Primates”. We have an under‐representation for the > 70 group only for “Eukaryota” and “Primates”.

As in skin cells, the highest number of over‐represented differentially expressed genes in immune cells was found in “All living organisms”. Compared to skin cells 40–70 group, immune cells exhibited fewer gene expression changes in the most ancient genes. Upregulated genes were more important in the most ancient genes (see Figure 2). We found an under‐representation for “Eumetazoa”, “Chordata”, “Mammalia”, “Boroeutheria”, “Euarchontoglires”, and “Primates”.

Ovarian cells showed the highest total number of differentially expressed genes among all cell types. The most significant gene expression changes occurred in ancient evolutionary categories. From “All living organisms” to “Bilateria” strata, we observed an over‐representation of the genetic changes (see Figure 2). Ovarian cells also showed a higher level of gene expression shifts (−1.96, see Table 1), possibly reflecting the pronounced effects of aging in reproductive tissues. The under‐representation is clearly located in the youngest genes, from “Tetrapoda” to “Primates” strata.

Human progenitor cells showed a significant atavistic over‐representation in the “All living organisms” evolutionary age (see Figure 2). We observed mostly down‐regulation in the genetic expression in this over‐representation. As in ovarian cells, the under‐representation is located in the evolutionary youngest genes, from “Mammalia” and “Primates”.

In the CellAge senescent signature, gene expression changes are most pronounced in ancient evolutionary categories, with “All living organisms” showing the highest level of differentially expressed genes, dominated by downregulation. The stratum “Eukaryota” also exhibits significant over‐representation (see Figure 3). The under‐representation is also largely located in the youngest genes, from “Tetrapoda” and “Primates” strata.

Mesenchymal senescent cells followed a similar trend, with “All living organisms” stratum showing significant over‐representation, though the magnitude is slightly higher than in the general senescent signature (see Figure 3). We found an under‐representation mainly located in the youngest genes, from “Amniota” to “Primates”.

Tissue‐specific aging signatures exhibited significant negative evolutionary age shifts: skin cells 40–70 (−2.23), ovary (−1.96), and progenitor cells (−1.84) showed strong shifts toward older genes, while CD8+ T‐cells (−1.47) and skin > 70 (−0.77) displayed a more moderate atavistic pattern, indicating tissue‐specific manifestations of evolutionary regression during aging (see Table 2). Senescence models (Mesenchymal senescent, CellAge senescence) gene sets demonstrated consistent shifts toward evolutionarily older genes (−2.41 and −2.32 categories).

Overall, we conclude that there is an atavistic overrepresentation in the differential genetic expression during aging in skin, immune, ovarian and human progenitor cells associated with a heterogeneity in the direction of the changes; sometimes the atavistic change is more about downregulation (skin cells), sometimes the opposite with more upregulation in the most ancient genes (immune and ovarian cells).

Cerebellum, hippocampus and cortex cells exhibited similar patterns, with an over‐representation of differentially expressed genes occurring in the “Euteleostomi” strata. This category showed mainly upregulation for the hippocampus, and more downregulation for the two other types of brain cells, suggesting increased reliance on vertebrate‐specific pathways during aging for the brain. There is no genetic atavistic over‐representation; the DEGs are indeed not significantly over‐represented in the “All living organisms” stratum (Figure 4). The three types of cells showed under‐representation mostly located in the evolutionary youngest genes, from “Amniota” to “Primates” for Cerebellum cells, from “Euarchontoglires” to “Primates” strata for hippocampus cells and from “Mammalia” to “Primates” for cortex cells.

The mesenchymal cells, in contrast, showed no over‐representation nor under‐representation of DEGs in “Euteleostomi” (Figure 4). However, we note that the number of genes analyzed in mesenchymal cells is almost an order of magnitude smaller than that of the other cells. This could have reduced the statistical power to detect potential differences according to evolutionary age.

Brain aging signatures showed moderate evolutionary age shifts. It is significant for cortex and cerebellum cells, with a mean age shift of −0.74 and −0.90 categories respectively and not significant for the hippocampus with a mean age shift of −0.31 (see Table 1), while mesenchymal stem cells showed a non‐significant and minimal shift (−0.27 categories), revealing tissue‐specific manifestations of the atavism theory across different cell types.

We can conclude there is no atavistic over‐representation of the genetic changes in the most ancient genes for brain and mesenchymal stem cells suggesting a differential aging or resistance to aging for the brain tissue and stem cells, as suggested by (Nie et al. 2022; Schaum et al. 2020).

Overall, we found an age‐dependent increase in heterogeneity in the direction of the phylogenetic position of tissues' transcriptional profiles. The number of upregulated and downregulated genes in the most ancient genes in the GenAge aging signature is almost equal, suggesting more of a reshuffling in genetic expression during aging in the most ancient genes.