Why women live longer than men

Throughout the world, in all countries, we find more widows than widowers. In other words, women live longer than men. Doctors and psychologists have proposed many theories but none of them are really convincing. We now have a probable answer: on the 23rd chromosome, men have a XY pair of chromosomes and women have a XX pair. The X chromosome in women appeared to be "silent" but its role has now been elucidated: as women age, this X chromosome "wakes up" and adds some additional years to the life of women and prevents cognitive decline. You can read the details in the following article.

https://www.science.org/doi/10.1126/sciadv.ads8169

Aging activates escape of the silent X chromosome in the female mouse hippocampus
6 Mar 2025
Margaret Gadek https://orcid.org/0000-0001-5453-6909, Cayce K. Shaw https://orcid.org/0000-0001-6636-9810, Samira Abdulai-Saiku https://orcid.org/0000-0003-3772-4646, Rowan Saloner https://orcid.org/0000-0002-1351-6183, Francesca Marino https://orcid.org/0000-0002-7899-4282, Dan Wang, Luke W. Bonham https://orcid.org/0000-0002-2533-1266, Jennifer S. Yokoyama https://orcid.org/0000-0001-7274-2634, Barbara Panning https://orcid.org/0000-0002-8301-1172, [...] , and Dena B. Dubal https://orcid.org/0000-0001-7504-4372 +3 authors Authors Info & Affiliations
Science Advances

5 Mar 2025

Vol 11, Issue 10

Abstract

Women live longer than men and exhibit less cognitive aging. The X chromosome contributes to sex differences, as females harbor an inactive X (Xi) and active X (Xa), in contrast to males with only an Xa. Thus, reactivation of silent Xi genes may contribute to sex differences. We use allele-specific, single-nucleus RNA sequencing to show that aging remodels transcription of the Xi and Xa across hippocampal cell types. Aging preferentially changed gene expression on the X’s relative to autosomes. Select genes on the Xi underwent activation, with new escape across cells including in the dentate gyrus, critical to learning and memory. Expression of the Xi escapee Plp1, a myelin component, was increased in the aging hippocampus of female mice and parahippocampus of women. AAV-mediated Plp1 elevation in the dentate gyrus of aging male and female mice improved cognition. Understanding how the Xi may confer female advantage could lead to novel targets that counter brain aging and disease in both sexes.

INTRODUCTION

The expansion of aging research to investigate female-specific biology and its mechanistic underpinnings represents a major advance of high importance for human health. True sex differences exist in aging, and understanding what makes women more resilient (or vulnerable) reveals targets for therapeutic paths that may benefit women’s health, men’s health, or both (1).
Women live longer than men, across socioeconomic status, despite famines and epidemics (2), and all around the world (2, 3), suggesting a common biologic contribution to female longevity. Female survival advantage is also variably observed in the animal kingdom (4–7) and extends to many (8–10), but not all (11), strains of mice, including those that are genetically heterogeneous (9). Sex chromosomes affect longevity (10) and may also affect resilience in health span, or the span of time spent healthy for specified outcomes. While female health span differs in its resilience and vulnerabilities across the body (12, 13), the connection between the number of X chromosomes (X’s) and longevity raises the question of whether X’s also affect cognitive resilience.
An increasing body of evidence supports female resilience in brain aging. Women undergo slower molecular brain aging, measured by the epigenetic clock, across brain regions (14), compared to men. Furthermore, women harbor a younger metabolic brain age, measured by positron emission tomography imaging (15). These findings could underlie the striking observation that women show resilience to cognitive decline (16–20) and exhibit higher baseline memory functioning in typical aging of several populations, in the absence of dementia (16–18, 20, 21). Etiologies of female-based cognitive resilience, also observed in aging mice (22), highlight a role for the second X. In normal aging (22), and in models of Alzheimer’s disease (AD) (23), adding a second X improves cognition in male mice and subtracting it worsens cognition in female mice (22, 23), indicating a causal role for the second X in cognitive resilience.
Enriched for neural factors, the X harbors 5% of our genome and has been largely understudied in the aging brain. In female mammals with two X’s, one is silenced through X chromosome inactivation (XCI), resulting in an active (Xa) and inactive X (Xi) (24). While XCI is thought to equalize X-linked gene expression between XX and XY cells, select genes from the Xi escape inactivation, to varying degrees and in a tissue-specific manner (25, 26). Since escape results in expression from both the Xa and Xi (25), a second X selectively increases X-linked gene dose in females and could potentially drive sex bias in cognitive resilience. Whether Xi escape remains unchanged into old age and if aging can affect Xi expression in the female brain are fascinating and unresolved questions. Furthermore, whether Xi expression uniquely and heterogeneously manifests across cell types in a key area of learning and memory targeted by aging, such as the hippocampus, is unknown.
Here, we investigated whether aging modulates expression of genes on the Xi, or the silent X, in the female hippocampus using an approach combining single-nucleus RNA sequencing (snRNA-seq) with allele-specific analysis, thus enabling discrimination of Xi from Xa expression in a cell type–specific manner. We report that aging preferentially modulates differential gene expression on the X compared to autosomes, with a near-global up-regulation of genes from both the Xa and Xi, in a cell type–specific manner. Focused analysis of the Xi revealed many changes including its cell type–specific activation of many genes, with increased baseline escape and new age-induced escape. Further investigation of a robust age-induced escape factor suggested that changes in Xi expression contribute to cognitive resilience in the aging female brain.

RESULTS

Genetic model of nonrandom XCI to detect escapees from Xi

To define and distinguish expression from the Xi and the Xa, we used a genetic model that combines two mouse strains, enabling strain-specific analysis based on differing alleles on the two X’s. Combining mouse strains is commonly used to study allele-specific expression (25, 27–29). In this model, one X is inherited from Mus musculus and the other from Mus castaneus (Fig. 1A), which exhibits single-nucleotide polymorphisms (SNPs) roughly every 264 base pairs (bp). Thus, transcripts can be mapped to either the M. musculus or M. castaneus genome.

Fig. 1. Allele-specific, single-nucleus sequencing of cell types in the young and aging XX hippocampus.
(A) Genetic model for allele-specific investigation of the X. M. musculus XX mice with an Xist deletion were crossed with XY M. castaneus mice. XX progeny without an Xist deletion (left) underwent random XCI. XX progeny with an Xist deletion (right) harbor an Xa chromosome strictly from the M. musculus strain and an Xi chromosome from the M. castaneus. Differences in SNPs between the strains enable a direct measure of expression from each X, the Xa and the Xi. Image credit: D. Velasco. (B) Diagram of the experimental workflow of allele-specific, single-nucleus sequencing. Young (n = 4 mice, age = 3 months) and old (n = 4 mice, age = 22 months) hippocampi from the strain-specific X-detection mice were dissected. Nuclei were dissociated and sequenced with 10x RNA sequencing on the Illumina NovaSeq 6000. Reads were aligned to the combined M. musculus and M. castaneus genome and analyzed by origin from Xa or Xi, cell type, and age. Hippocampi, cells, and NovaSeq were acquired with a ShutterStock Enhanced License. (C) Hippocampal cell types sequenced. Dimension reduction analysis using UMAP revealed 11 distinct cell type populations sequenced plus a mixed category. Clusters are the following: endothelial cells (Endo), astrocytes (Astro), vascular leptomeningial cells (VLMC), microglia (MG), DG neurons, CA1 neurons, glutamatergic undefined neurons (Glut_Undef), CA3 neurons, GABAergic neurons (GABA), oligodendrocyte progenitor cells (OPCs), and oligodendrocytes (Oligo). (D) Matrix plot showing expression relating cell type marker genes (columns), cell clusters (rows), and corresponding genes. (E) Heatmap of cell markers across UMAP clusters. Color intensity reflects extent of gene expression.
Open in viewer
XCI is random (Fig. 1A, bottom left) such that one X is the Xa in approximately half of cells and the other X is the Xa in the remaining half (24), which adds analytical complexity to single-cell studies. To simplify analysis, we implemented a genetic model of nonrandom XCI, with the Xa coming from M. musculus and the Xi coming from M. castaneus (Fig. 1A, bottom right) (25). In this model, the M. musculus X harbors a deletion of Xist, which encodes a long noncoding RNA required in cis for XCI (30). Since only the M. castaneus X can transcribe Xist during development, it is always the Xi in our mouse model. Thus, transcripts with SNPs mapping to the M. castaneus genome arise from the Xi and are designated escapees.
Using snRNA-seq, we profiled over 40,000 nuclei from four young and four old female mouse hippocampi (Fig. 1B). Nuclei were isolated from dissected hippocampi, with similar numbers of nuclei per sample, processed through droplet-based single-cell profiling, and sequenced. The reads were assigned to the M. musculus or M. castaneus genome for further analysis. Similar median unique molecular identifier (UMI) counts and gene counts were assigned to each sample (data S1). Autosomal reads were 56% M. musculus and 44% M. castaneus, with a mild mapping bias toward M. musculus, as observed in other allele-specific sequencing studies (25). In contrast, reads arising from the X were 91.7% M. musculus and 8.3% M. castaneus (including robust Xist expression), consistent with silencing of the M. castaneus X, and with the estimated 3 to 7% of escape in mice (25). These findings validate the genetic model of establishing nonrandom XCI to specifically enable the detection of escapees from the Xi.
Single-nucleus sequencing of the young and old female mouse hippocampus

Nuclei clustered into different cell types using the Leiden algorithm in Scanpy (Fig. 1C) (31). Using a comparison to the Allen Brain Atlas 10x hippocampus and cortex datasets, differentially expressed genes (DEGs) were assigned to corresponding groups (data S2) (32). Clusters with similar cell marker identities were combined, and cell types were assigned using the most DEGs in the cluster compared to all others. Neuronal cell types included glutamatergic neurons such as CA1, CA3, and dentate gyrus (DG) neurons. Other undefined glutamatergic neurons (Glut_Undef) and GABAergic (GABA) neurons were also present. Glial cell types included astrocytes (Astro), oligodendrocytes (Oligo), oligodendrocyte progenitor cells (OPC), and microglia (MG). Finally, vascular leptomeningeal cells (VLMC) and endothelial cells (Endo) were identified. Transcriptionally indistinct mixed nuclei made up 0.9% of the nuclei population. The distinctly identified clusters were further validated with known marker genes from other mouse brain single-cell datasets (Fig. 1, D and E) (32–35). Expression of marker genes across clusters showed expected patterns. Thus, the cell clustering analysis of our snRNA-seq dataset reflected the spectrum of cell types typically present in the hippocampus, indicating a robust model to probe gene expression across hippocampal cell types in aging.
Enrichment of DEGs on the X in aging hippocampus

We first conducted a differential gene expression analysis of combined cell types and individual cell types. Representation of hippocampal neuronal and glial cell types was similar in young and old cell populations and across samples (fig. S1) (data S3). Using the decoupler and pyDEseq2 programs (36, 37), we interrogated aging-induced differential gene expression via pseudobulking expression by cell type and sample. In combined cell types, 926 genes were significantly differentially expressed in old compared to young nuclei (Fig. 2A) (data S4). Approximately 50% of these DEGs were previously identified to change in similar directions in datasets of the aging male (38) and female (33) mouse brain (Fig. 2A, top hits in red) (data S5). Also in the combined cell types, among the 926 DEGs, 29 were from the X (Fig. 2A, in blue).

Fig. 2. Aging induces differential gene expression in the hippocampus across autosomes and the X’s.
(A) Aging significantly altered 926 DEGs (log2 fold change > 0.1 or < −0.1 and adjusted P < 0.05) in the hippocampus, including 29 expressed from the Xa and Xi in blue, when considering all cell types. Top 10 DEGs matching the described published datasets are in red. (B) DEGs from the autosome and X’s, normalized to total expressed genes (*P = 0.025, χ2 test). (C) Violin plot showing differentially expressed alleles by cell type and log2 fold change. Distributions of significant autosomal DEGs (light gray, left of violin plot) and distributions of significant X DEGs (color, right of violin plot) are shown. Proportion of significant autosomal (A) and X (X) genes over the total detected genes are shown comparing autosome and X significant DEG proportion (χ2 tests: All *P = 0.0253, Astro ns, CA1 ***P = 0.0005, CA3 *P = 0.0225, DG ***P = 0.0004, GABA ****P < 0.0001, Glut ns, MG *P = 0.0375, OPC ns, Oligo *P = 0.0452. (D) Top 10 GO terms (rows) from X genes significantly expressed in each cell type by −log(adjusted P value). (E) Number of DEGs up- and down-regulated in the cell type–specific volcano plots in (F) to (N). Classifications include the following: Neuronal, ≥2/5 neuronal cell types ≤1 glial; Glial, ≥2/4 glial cell types ≤1 neuronal genes; Multiple, ≥2 cell types; Cell type–specific, 1 cell type only. Two genes in the multiple cell types category, Xa Pcdh11x and Xa Sytl5, were excluded because they increased in one cell type but decreased in another. (F to N) Volcano plots of significant DEGs on Xa and Xi in cell types including (F) CA1 neurons, (G) CA3 neurons, (H) DG neurons, (I) undefined glutamatergic neurons, (J) GABAergic neurons, (K) microglia, (L) oligodendrocytes, (M) oligodendrocyte progenitor cells (OPC), and (N) astrocytes. A630012P03Rik, C430049B03Rik, 4930500G05Rik, and 5330434G04Rik are abbreviated as A63Rik, C43Rik, 493Rik, and 533Rik, respectively. Image credit: E. Kakoshina.
Open in viewer
As global changes to X regulation were previously identified as a feature of female hypothalamic aging (33), we wondered whether aging preferentially alters DEGs on the X compared to autosomes in the hippocampus. When normalizing for total genes of each chromosome, more genes changed on the X (Xa and Xi combined) than on autosomes (Fig. 2B). This finding extended to analyses of individual autosomes of similar size compared to the X (fig. S2). DEG overrepresentation on the X similarly extended to most cell types in the aging, compared to young, hippocampus (Fig. 2C). That is, similar to the combined all-cell population, CA1 neurons, CA3 neurons, DG neurons, GABAergic neurons, microglia, and oligodendrocytes individually showed an increased number of significant DEGs from the X, compared to autosomes. Notably, DG neuronal nuclei harbored the highest number of significant DEGs, from both autosomes (14%) and X’s (20%), followed by Oligos (Fig. 2C), suggesting that they may be particularly sensitive to age-induced changes on the X.
Following the finding that aging-induced DEGs were enriched on the X, we next assessed cell type–specific RNA expression of X-linked genes, applying genome-wide corrections. First, using g:Profiler, we conducted a gene ontology (GO) term analysis of the X-linked genes significantly enriched in each cell type compared to other cell types in our dataset (39). The top predicted structures and functions across cell types were the synapse and synaptic organization (Fig. 2D), consistent with the high proportion of genes related to cognition on the X (40). Thus, the enrichment of neural functions on the X highlights the high value for its study in the hippocampus and for cognitive aging.
Aging-induced DEGs from Xa and Xi across hippocampal cell types

We assessed significant X-linked DEGs, for those broadly shared across cell types like neurons and glia, and for cell type–specific populations. DEGs, measured from the Xa or Xi, were classified as neuronal, glial, multiple, or cell type specific according to inclusion criteria (Fig. 2E). Aging nearly globally resulted in up-regulated neuronal and glial DEGs—and largely up-regulated multiple and cell type–specific DEGs—from both the Xa and Xi (Fig. 2E). Notably, most X-linked DEGs changed in a cell type–specific manner, highlighting the heterogeneity of aging effects within the hippocampus. Even so, the convergent aging-induced increase in X-linked expression across cell types suggests common chromatin regulators and transcription factors, and less transcriptional repression, across the X’s.
To investigate how aging alters predicted cellular components and functions referable to the X, we performed individual GO term enrichment analyses of significant X-linked DEGs that change with aging for each cell type (fig. S3). GO terms related to synaptic regulation, activity, and organization were the most robust and remarkably conserved pathways among nearly all cell types (fig. S3). Cell type–specific differences also emerged, such as GTPase (guanosine triphosphatase) binding (CA3, DG, Oligos) or constituents of the myelin sheath (Astro and Oligos) (fig. S3). Rho GTPase signaling integrates synaptic structure and function through AMPA receptor trafficking and represents a treatment avenue for cognitive disorders (41, 42), while myelination decreases with aging and contributes to cognitive decline (43). Since synapses and myelin are fundamental units and substrates of neural function, and targets of aging-induced dysfunction, these enrichments collectively suggest a molecular basis for XX-based resilience against cognitive decline in aging. Details of aging-induced X-linked DEGs that were most common to neuronal and glial cell populations are reported in Supplementary Text.
We identified the aging-induced DEGs derived from Xa, present in the most cell types; all notably harbor mutations in humans that cause intellectual disability: Dmd (44, 45), Cnksr2 (46, 47), and Pak3 (48, 49). These X genes increased in at least seven cell types (Fig. 2, F to N) and have known hippocampal functions. Dmd encodes dystrophin, a scaffold protein at inhibitory synapses (50). Cnksr2 is also a synaptic scaffolding protein for granule cells of the DG (51). Pak3 is a synaptic connectivity protein that forms spines (52) and synapses (53), and contributes to cognition (54). Thus, synaptic functions of aging-induced Xa genes are prominent and show human relevance as demonstrated by intellectual disability resulting from their loss of function or pathogenic variation.
The aging-induced DEGs derived from the Xi represented in most cell types were Ftx, Gpm6b, and Plp1. These genes increased on the Xi in three to four cell types that were either neuronal (Ftx) or glial (Gpm6b, Plp1) biased (Fig. 2, F to N). Ftx, the long noncoding RNA, inhibits hippocampal apoptosis (55) and ferroptosis (56). Also, as a positive regulator of Xist (57), its age-induced increase might influence XCI regulation. Members of the proteolipid protein family, Gpm6b and Plp1 (58), both contribute to myelination, and human mutations in PLP1 cause hypomyelination and intellectual disability (59). Since defects in myelin maintenance contribute to age-related cognitive decline (43), it is interesting to speculate that modest increases in these factors from the Xi could attenuate cognitive deficits.
Since snRNA-seq provides a shallow depth of coverage and allele-specific analysis requires doubling the genes detected (one from each allele), together they necessarily constrain both coverage and statistical power. Thus, cell type–specific DEG analyses yield high specificity but low sensitivity for detecting gene changes, particularly from genes on the Xi, which are expressed at lower levels than on the Xa. Given our focus on detecting X escape from XCI, we complemented our study with additional analyses.

Xi escape from XCI, with reference to Xa, in the aging hippocampus

We next implemented a well-established and conservative analysis pipeline (25) to define X escape and assess how Xi expression changes across cell types and age, in reference to Xa. This better powered and more sensitive computational approach (Fig. 3A), compared to single-nucleus DEG analysis, aggregates reads broadly across and within cell types (pseudobulking) and enables a quantitative detection of escape from XCI (data S6). X escape status was calculated using an escape proportion and its corresponding 99% confidence intervals (CIs). To meet criteria as an escapee, a lower bound of the CI must be greater than 0, and the median escape value (the escape proportion) must be greater than 0.05 (Fig. 3A).

Fig. 3. Xi escape from XCI, with reference to Xa, in the aging hippocampus across cell types.
(A) Schematic of escape ratio and CI calculation with an example calculation for hypothetical gene “X” showing age-induced escape. (B) Sankey plot showing the classification of X escapee expression of 708 detected X genes in young (middle) and old (right) populations for bulk cell populations. The gene expression is first classified as being baseline escapees (red) or inactive genes (blue). Escapee expression in aging is further divided into increased (purple), maintained (red), and lost escape with age (maroon). The Xi genes are further divided to note if they become new escapees (orange). (C) Bulk cell population, Xi escape 99% CIs of age-activated escapees (orange) and increased escape (purple). A gene is an escapee if the lower bound of its CI is greater than 0 (lower dotted line) and its median is greater than 0.05 (higher dotted line). (D to L) Sankey plots of Xi escape genes with age (left) and their escape CIs (right) in different cell types including (D) CA1 neuron, (E) CA3 neurons, (F) DG neurons, (G) undefined glutamatergic neurons, (H) GABAergic neurons, (I) Microglia, (J) Oligodendrocytes, (K) Oligodendrocyte progenitor cells (OPC), and (L) Astrocytes. 1810030O07Rik and 2010308F09Rik are abbreviated as 181Rik and 201Rik, respectively.
Open in viewer
Baseline escape genes were identified as those that showed escape in the young life stage, and differences between escapee behavior in young and old cohorts were categorized. Genes exhibited increased escape from baseline if they escaped in both young and old samples, but the lower bound of their old CI was greater than the upper bound of their young CI. Baseline escapees increased (Fig. 3B, purple), lost (Fig. 3B, maroon), or maintained (Fig. 3B, red) their level of expression with age. In addition, aging activated expression of previously silent Xi-linked genes to induce their new escape (Fig. 3B, orange).
With this framework for Xi escape across aging in place, we examined escape in a population of combined cell types (Fig. 3B). Of the 73 baseline escapees identified, 4 increased expression with age, 51 maintained expression, and 18 lost expression with aging. Remarkably, aging induced new expression of 19 genes from Xi that were previously inactive in the young life stage. Of particular interest were age-induced (orange) and increased escape with age (purple) genes (Fig. 3C), since their Xi expression was increased in hippocampal cells (Fig. 3, D to L).
We next determined whether there was cell type variation in how aging affected Xi-linked gene expression (Fig. 3, D to L, and fig. S4, A to D). In the DG and glutamatergic neurons, aging both robustly activated and repressed Xi-linked genes, indicating dynamic age-induced remodeling of fundamental neural units underlying learning and memory. In comparison, in CA1 and CA3 (which receive and refine DG input), GABAergic cells, and OPCs, aging predominantly repressed Xi-linked gene expression. In contrast, aging markedly activated Xi-linked gene expression in microglia, astrocytes, and oligodendrocytes, potentially influencing neuroimmune-mediated functions and proper myelination. Thus, the effects of aging on the Xi varied by cell type, highlighting the complexity of gene regulation of the highly and structurally condensed Xi.
To maintain XCI-mediated transcriptional silencing, the Xi is tightly compacted by heterochromatin, is coated by Xist RNA, and exhibits altered DNA and histone methylation patterns compared to the Xa. Since aging altered expression of Xi-linked genes, we wondered whether it influences regulators of XCI. While Xist was abundant in every cell, as anticipated, a further increase in already high expression was only detected on Xi in microglia (Fig. 2K). Because Xist deletion from Xa was used to enforce nonrandom XCI, our method does not capture any Xist expression from the Xa, which may explain why we do not observe the previously reported increase in abundance of this noncoding RNA in several wild-type hippocampal cell types (33). Nonetheless, and interestingly, aging changed expression of other XCI regulators from the Xi, including Jpx and Ftx. Aging increased Jpx expression from the Xi in glutamatergic neurons (Fig. 3G) and decreased it in CA3 neurons (fig. S4) Furthermore, aging consistently increased Xi expression of Ftx in CA1, DG, and glutamatergic neurons (Fig. 3, D, F, and G, and fig. S4), key cell types for learning and memory in the hippocampus. Our results converge with previous findings that aging also increased total Jpx and Ftx expression in the female hypothalamus (33); our data suggest that this increase was from Xi. Whether these XCI regulators affect the XCI status of Xi, or whether they exert XCI-independent functions in the aging hippocampus, remains to be determined.
We examined whether the same genes could be targets of aging-induced Xi modulation across multiple cell types of the hippocampus. Approximately half (44%) of Xi-mediated changes in a gene’s expression extended across two to eight cell types in the given escape pattern. For example, aging increased Plp1 expression on the Xi consistently across seven cell types (CA1 neurons, DG neurons, undefined glutamatergic neurons, GABAergic neurons, oligodendrocytes, microglia, and astrocytes) (Fig. 3, D, F to J, and L, and fig. S4). Relatedly, aging increased expression of Gpm6b in glial cells (oligodendrocytes, OPCs, and astrocytes), induced new escape in MG, and decreased expression in neuronal cells (CA1, CA3, and undefined glutamatergic neurons) (Fig. 3, D, E, G, and J to L, and fig. S4). Thus, as top examples, Plp1 and Gpm6b were targets of Xi modulation in aging, each in seven of nine cell types, suggesting their heightened accessibility to activation or repression on the silent X in aging.
Cell type–specific repression and activation of the Xi

To shore up findings of Xi escape, calculated with reference to Xa, we next further examined gene changes by focusing on Xi-only expression. In addition to understanding aging-induced Xi activation through escapee expression increases and new escape with aging, we extended analyses to aging-induced repression by querying X genes that lost their baseline escape status with aging.

In a combined cell type analysis, escapee genes from the Xi identified to increase expression, lose expression, newly escape in aging, or fit multiple categories (mixed escape)—observed in at least two hippocampal cell types (fig. S4, B to D)—were examined for their fraction of cells and mean Xi expression in young and old mice (Fig. 4A). Broadly, aging modulated escapee expression (Fig. 4A). Notably, the escapee Ftx solely increased expression with aging across multiple cell types—and at very high levels. Mixed escape expression patterns showed opposing Xi changes with age in different cell types that canceled out the signal in this combined cell type analysis.

Fig. 4. Activation and repression of Xi-only expression in aging and across cell types.
(A) Combined, bulk cell analysis of Xi expression showing genes with lost escape (maroon), escapee expression increased (purple), new escape with aging (orange), or mixed escape, observed in at least two hippocampal cell types in young and old populations. Circle color indicates expression. Circle size indicates the proportion of cells in the group expressing the gene. Image credit: D. Velasco. (B) Cell type–specific repression, or lost escape of Xi with age observed in at least two cell types. Young (top) and old (young) Xi expression of the escapee gene. (C) Median escape values of Xi genes that undergo repression, or loss of escape with age, in at least two cell types (n = 100 data points, representing 29 genes, each in the cell types with escapee expression lost, ****P < 0.0001, paired two-tailed t test). (D) GO term analysis of Xi genes that underwent repression, or of loss of escape, in at least one cell type by −log(adjusted P value). (E) Cell type–specific increase in escapee expression from Xi with age in any cell type. (F) Median escape values of Xi genes that undergo increased escape expression with age, in at least two cell types (n = 8 data points, representing three genes, each in the cell types with increased escape expression with age, ***P = 0.0003, paired two-tailed t test). (G) Cell type–specific Xi expression of genes that undergo new escape with aging in at least two cell types. (H) Median escape values of Xi genes that undergo new escape with aging, in at least two cell types (n = 53 data points, representing 18 genes, each in the cell types with new escape with aging, ****P < 0.0001, paired two-tailed t test). (I) GO terms of combined Xi genes that underwent escapee expression increased and new escape with aging, in at least one cell type by −log(adjusted P value).
Open in viewer
To increase resolution of Xi-only expression in aging, we next used a cell type–specific approach and queried loss of escapee expression occurring in at least two cell types (Fig. 4B and fig. S4). Aging repressed Xi through loss of escapee expression, resulting in loss of escape status, depicted by reduced gene expression and fraction of cells of the depicted cell type and life stage (Fig. 4B). Parallel to patterns identified in escape analysis, aging repressed Xi expression and cell fraction of Gpm6b in neurons, and boosted them in glia (Fig. 4, B and E, and fig. S4, B to D). Notably, aging repressed Pak3 across most cell types on the Xi (Fig. 4B and fig. S4C) and activated it on the Xa (Fig. 2). To further confirm the escapee’s loss of expression, we graphed X escape for each aging-mediated Xi gene that lost escape status in a cell type. As expected, in their specific cell type, aging uniformly repressed these escapee genes (Fig. 4C).
We wondered what functions aging-mediated repression of the Xi could contribute to a hippocampal cell. A GO term analysis of these Xi genes in at least one cell type (fig. S4C) highlighted several terms, including those regulating chromosomes (Fig. 4D). A conspicuous contributor to this prediction is Atrx, a central player in genome stability that represses numerous regions of chromosomes and contributes to silencing of Xi in development (60). Thus, aging-mediated repression of the repressor Atrx on Xi could potentially contribute to new escape from the Xi.
Similar to our analysis of Xi repression, we conducted an analysis of Xi-only activation in aging, using the same cell type–specific approach (Fig. 4, E to H, and fig. S4, B and D). Aging activated Xi through increased escapee expression (Fig. 4, E and F) and new escape with aging (Fig. 4, G and H), depicted by augmented expression and fraction of cells in both groups of genes. As highlighted in the DEG and escape analyses, aging increased Ftx escape from the Xi in many cell types (Fig. 4E and fig. S4B). To further confirm this finding, we graphed X escape for each Xi-linked escapee that increased expression in a cell type. As expected, in their specific cell type, aging uniformly increased baseline expression of these escapee genes (Fig. 4F).
Several genes showed new escape with aging—indicating activation of the Xi—across many cell types. Among them, and confirming our previous analysis, Plp1 notably exhibited new escape with aging in six cell types and increased escapee expression in the remaining (Fig. 4G and fig. S4D). To further confirm that genes inactive in the young life stage underwent new escape with aging, we graphed X escape for each Xi gene of this category in a cell type. As expected, in their specific cell type, aging robustly activated these genes—going from silent in young mice to new escapee expression in old mice (Fig. 4H).
To probe putative functions conferred by aging-mediated activation of Xi (increased escapee expression and new escape with aging) in the hippocampus, we performed a GO term analysis of this class with genes changed in at least one or more cell type (fig. S4, B and D). An enrichment of terms related to XCI, such as Ftx, Jpx, and Rlim (57, 61, 62), suggests that aging-induced Xi activation could alter its transcriptional landscape. This probably contributes to other terms enriching for broad neural functions (synaptic and endosomal membranes, neuronal cell body, and protein binding) (Fig. 4I). Atrx was implicated in multiple GO terms relating to chromosomal regulation, indicating that the repressor Atrx is repressed in some cell types and activated in others (60). Other escapees, Plp1 and Gpm6b, are relevant to the cellular component organization term, as components of myelin and the membrane of many central nervous system cells (63, 64). Finally, Atp7a and Gria3 are known to contribute synaptic integrity, signaling, and neuronal activation (65, 66). Thus, aging-mediated activation of the Xi may set off a cascade of transcriptional escape to ultimately alter neuronal and glial functions.
Human relevance of DEG and X escape analysis findings

The X is enriched for genes with neural functions, and the findings of our computational analyses highlighted broad neural functions alongside synaptic signaling and neural structural components. Thus, we wondered whether specific aging-induced changes in Xi-linked gene expression are enriched for genes represented among human, X-linked conditions of intellectual disability. In other words, if the select Xi genes identified here were particularly important to cognitive functions, then their mutations would predict cognitive dysfunction. To that end, we systematically searched the literature for the 100 Xi genes that exhibited age-induced changes in at least one or more cell types. Mutation of nearly half (49%) of these genes caused intellectual disability, typically in males since they lack a compensatory X (table S1) (67–115). This finding highlights the exceptionally important role of the silent X, changed by aging, in potentially influencing underlying substrates of cognitive function, one of our most valued outputs of brain function.
Female sex increases Plp1 expression in the hippocampus of mice and parahippocampus of humans

Aging selectively activated Xi genes in both single-nucleus differential gene expression and X escape analyses, and among them, Plp1 either was highly activated from Xi or showed increased escapee expression across cell types. PLP1 is also implicated in Pelizaeus-Merzbacher disease, a leukodystrophy that causes intellectual disability, among other neurological deficits; thus, it has high relevance for human cognition (81, 115). We therefore focused further studies on Plp1. Aging increased Xi expression of Plp1 in DG neurons, oligodendrocytes, and astrocytes in the DEG analysis (Fig. 2, H, L, and N). The Plp1 escapee ratio was highest (Fig. 5A), with the most robust Xi expression (Fig. 5B) in old oligodendrocytes. As PLP1 is intrinsic to oligodendrocyte structure and function, its strong increase from the Xi with aging suggests that it may enhance aging oligodendrocytes (58).

Fig. 5. Activation of hippocampal Plp1 Xi expression across cell types in XX mice and increased Plp1 in aging female mice and humans.
(A) Plp1 X-escape 99% CIs by cell type and age in XX mouse hippocampi. CIs with lower bounds above 0 (lower dotted line), with medians above 0.05 (upper dotted line), show escape from baseline Xi Plp1 expression. (B) Mean expression of Xi Plp1 in young and old hippocampal cell types. The fraction of cells in the group expressing Plp1 correlates with circle size. (C) Relative Plp1 mRNA expression in the whole hippocampus of young (age 3 months; XY, n = 13; XX, n = 15) and old (age 20 to 35 months; XY, n = 10; XX, n = 15) XY and XX mice. Means are relative to XX young mice, arbitrarily defined as 1. Old XX compared to young XX mice (***P = 0.0002, Bonferroni-corrected unpaired two-tailed t test). Old XX compared to old XY mice (**P = 0.0014, Bonferroni-corrected unpaired two-tailed t test). (D) Relative PLP1 mRNA expression in the parahippocampal gyrus from postmortem tissue of old men and women (XY = 68, XX = 116). Data were normalized, log2-transformed, and adjusted to account for relevant covariates including batch, age at death, race, postmortem interval, RIN, exonic rate, and rRNA rate. Box plots represent the median and 25th and 75th percentiles, and box hinges represent the interquartile range of the two middle quartiles within a group. Min and max data points define the extent of whiskers (error bars) (**P = 0.008).
Open in viewer
We further measured Plp1 in mice and PLP1 in humans. To confirm the aging-mediated increase in Plp1 across cell types in XX mice, we compared its bulk expression in young compared to old mice in the XX and XY hippocampus (Fig. 5C) using a separate cohort. Using quantitative polymerase chain reaction (qPCR), we found that aging increased Plp1 mRNA in XX, but not XY, mice. To assess whether this sex difference in Plp1 extends to humans, we then assessed its expression in the parahippocampus of aging humans using the Mount Sinai Brain Bank (MSBB) cohort. This region surrounds the hippocampus and is involved in spatial memory, information, and context. Sequenced samples were analyzed and normalized for library size, and adjusted for relevant covariates, including batch, age at death, race, postmortem interval, RNA integrity number (RIN), exonic rate, and ribosomal RNA (rRNA) rate. In parallel to mice, older women showed increased PLP1 expression in the parahippocampus, compared to older men (Fig. 5D). Increased PLP1 in women was restricted to the parahippocampal region and not observed in the frontal pole, superior temporal cortex, or inferior frontal gyrus (fig. S5). Together, these findings confirm increased Plp1/PLP1 expression in the hippocampus or its surrounding area of aging females in both mice and humans.
AAV-mediated Plp1 overexpression in oligodendrocytes of the mouse hippocampus

To investigate the effects of increased Plp1 on cognition in aging, we constructed an adeno-associated virus (AAV) to overexpress Plp1. We focused on oligodendrocytes because the highest overall expression and most robust aging-induced increase in Plp1 was observed in this cell type. We used an AAV2/5 Plp1 mRNA construct driven by Mbp, an oligodendrocyte-specific promoter, followed by a ubiquitous SV40 promoter transcribing green fluorescent protein (GFP) to mark successful transfection (Fig. 6A). The control AAV was identical to the Plp1-OE AAV, except for the deletion of Plp1. In vitro infection of a mixed glial population including oligodendrocytes confirmed that Plp1-OE compared to control AAV increased Plp1 mRNA overexpression in both XY and XX cells (Fig. 6, B and C).

Fig. 6. AAV-mediated Plp1 overexpression in vitro and in vivo.
(A) Schematic of AAV2/5 viruses for Plp1 overexpression (Plp1-OE) (top) and control (bottom). The Plp1-OE virus contains an oligodendrocyte-specific Mbp promoter driving Plp1, followed by an SV40 promoter transcribing GFP. The control virus is the same virus without Plp1. (B) Relative Plp1 mRNA expression in XY and XX mixed glia infected with Control (n = 11 wells) and Plp1-OE (n = 12 wells) viruses (*P = 0.0154, two-tailed t test) representing two independent experiments separated by sex, shown in (C). (C) Relative Plp1 mRNA expression in (left) XY mixed glia infected with Control (n = 5 wells) and Plp1-OE (n = 6 wells) viruses (*P = 0.0409, one-tailed t test) and (right) XX mixed glia mixed glia infected with Control (n = 6 wells) and Plp1-OE (n = 6 wells) viruses (*P = 0.0386 one-tailed t test). (D) Schematic of in vivo and ex vivo AAV hippocampal experiments in old XY and XX mice. Image credit: A. Moreno. (E) Representative 20× stitched image of the hippocampus with GFP infection largely in the DG region from the Plp1-OE virus (green) and DAPI (blue). Scale bar, 100 μm. (F) Representative 40× image of the DG with GFP infection from the Plp1-OE virus (green), OLIG2 oligodendrocyte marker (red), and DAPI (blue), indicating infected oligodendrocytes (merged, yellow). Scale bar, 100 μm. (G) Quantitation of the percentage of total cells counted in the DG expressing OLIG2 (oligodendrocyte marker) and GFP (marking AAV infection) and showing colocalization of both markers (yellow), indicating AAV-infected oligodendrocytes (n = 3 XX mice).
Open in viewer
Following in vitro validation, we performed in vivo AAV studies in aging XY and XX mice. We specifically injected the control and Plp1-OE AAV into the DG of the mouse hippocampus to investigate whether Plp1 mediates behavioral changes in the aging brain (Fig. 6D). We focused on the DG since it is a key region in cognitive functions like spatial memory (116) and also harbored the most DEGs in our pseudobulk analysis (Fig. 2, C and H), suggesting that it is particularly sensitive to aging. Using immunohistochemistry, we examined AAV infection of oligodendrocytes with immunofluorescence for GFP (Fig. 6, E and F) and OLIG2 (Fig. 6F), an oligodendrocyte marker. GFP (green) overlapped with OLIG2(red) in 6.8% of cells in the DG and 24.3% of oligodendrocytes within the region (Fig. 6G), confirming that the AAV infected oligodendrocytes.
Cell type–specific up-regulation of Plp1 improves cognition in aging XY and XX mice

Finally, we tested whether increasing Plp1—through cell type–specific elevation in oligodendrocytes of the hippocampus—improved cognition in the aging XY and XX brain. To this end, we conducted a battery of behavioral tests in old male and female mice transfected with AAV2/5 with an empty vector (control) or Plp1 overexpression (Plp1-OE) (Fig. 7A). Plp1-OE in oligodendrocytes did not alter anxiety-like behavior in the elevated plus maze (EPM) (Fig. 7B) or total activity in the open field (Fig. 7C) in either XY or XX mice. To probe spatial and working memory, domains preferentially targeted by aging (117), we then assessed mice in the two-trial large Y-maze. Plp1-OE in oligodendrocytes increased exploration in the novel compared to the familiar arm in both XY (Fig. 7D) and XX (Fig. 7E) mice, indicating that it improved cognition in aging mice of both sexes. Thus, Plp1 elevation in the oligodendrocytes of the hippocampus—recapitulating a component of aging-induced activation of the silent Xi in females—enhanced learning and memory in aging brains of both XY and XX mice.

Fig. 7. Elevating Plp1 in oligodendrocytes of the DG improves cognition in old XY and XX mice.
(A) Experimental paradigm. Following AAV infection with AAV-control (Control) or AAV-Plp1-OE (Plp1-OE), old XX and XY mice underwent behavioral and cognitive tests including EPM, open field, and the two-trial large Y-maze. Image credit: A. Moreno and D. Velasco. (B) Percent time in the open arm of the EPM did not differ between old Control and Plp1-OE mice that were either XY (left) or XX (right) mice (age = 20 months XY; 18.5 months XX) (n = 18 mice per experimental group) (P = 0.9071 XY; P = 0.6775 XX, two-tailed unpaired t test). (C) Total activity in the open field did not differ between old Control and Plp1-OE mice that were either XY (left) or XX (right) (age = 20 months XY; 18.5 months) (n = 18 mice per experimental group) (P = 0.8559 XY; P = 0.2802 XX, two-tailed unpaired t test). (D) In old XY mice, Plp1-OE increased the duration in the novel compared to the familiar arm versus Controls in the two-trial large Y-maze across 5 min (left) (*P = 0.0193 treatment effect, two-way ANOVA) and averaged over 5 min (right) (***P = 0.0005, two-tailed unpaired t test) (age = 20.5 months) (n = 15 Control mice, n = 17 Plp1-OE mice). (E) In old XX mice, Plp1-OE increased the duration in the novel compared to the familiar arm versus Controls in the two-trial large Y-maze across 5 min (left) (**P = 0.0020 treatment effect, two-way ANOVA) and averaged over 5 min (right) (****P < 0.0001, two-tailed unpaired t test) (age = 19 months) (n = 17 Control mice, n = 17 Plp1-OE mice).
Open in viewer
DISCUSSION

Our data show that aging remodels Xi, the silent X, across cell types of the female mouse hippocampus. Aging preferentially altered DEGs on the X compared to autosomes, and nearly globally up-regulated neuronal and glial DEGs from both Xa and Xi, in a cell type–specific manner. Focused analysis of Xi showed changes in escapee expression including new escape with aging, and collectively predicted transcriptional regulation of XCI and broad neural structures and functions including synaptic integrity, signaling, and constituents of myelin. Plp1, a component of myelin highly expressed in oligodendrocytes, was among the most common and robustly activated Xi genes with age, and also increased in the aging parahippocampus of women. AAV-mediated elevation of Plp1 in oligodendrocytes of the DG, a key memory region targeted by aging, improved cognition in old mice of both sexes. Thus, recapitulation of a component of aging-induced activation of the silent X in females promoted better function of the old brain.

The study of female-specific biology is historically underrepresented in science and medicine but is essential and expanding fervently. This area of investigation is important because it (i) promotes an understanding of how fundamental biology contributes to health and disease in women and (ii) lays the groundwork to dissect sex differences between men and women. Unraveling the sex-based biology of what makes one sex more resilient or vulnerable paves the path to new and personalized treatments for both sexes. Our investigation of how aging affects the Xa and Xi chromosomes is specific to females and therefore forms a basis to understand sex differences, since typical males lack a second X.

The X, a major source of sex difference due to a second X in female mammals, represents 5% of the genome in men and women, and its study is growing rapidly, particularly in brain aging and age-related neurodegenerative conditions such as AD. Historically, analytic challenges posed by X hemizygosity in males, random XCI and baseline X escape in females, shared sequences between X and Y, and limited representation of the X in genome-wide association studies often resulted in X exclusion in studies (118, 119). However, expanding tool kits and varied sequencing approaches offer innovative and complimentary opportunities to study the X with high accuracy in the brain, as evidenced by several recent studies (23, 120–122).
Our findings remarkably converge upon several targets identified in recent, unbiased, and large-scale human studies of the X in aging and AD. In these studies, either X expression (23) or its genetic variation (120–122) associated with genes coding for synaptic (GRIA3, DMD, TENM1, FRMPD4, IL1RAPL1, GRIPAP1), kinase or phosphatase (PNCK, IRAK1, WNK3, MTM1), ubiquitin ligase or deubiquitinase (RLIM, OTUD5), DNA polymerase or transcription (AFF2, POLA1), and sodium and potassium transporter (SLC9A7) factors—all of which were modulated in our study of aging-induced remodeling of Xa and Xi. Notably, genetic variation in SLC9A7, which underwent new escape with aging in oligodendrocytes (fig. S4D), associated with AD risk in a meta-analysis of over 1 million individuals (122). It will be particularly interesting to know how genetic variation alters cell type–specific SLC9A7 levels and function, and how that links to AD risk. This, along with the finding that around half of the aging-induced targets we identified on the Xi cause human intellectual disability if mutated (table S1), highlights the exceptional role of the X factors identified in contributing to cognition.
XCI was proposed in 1961 by Mary Lyon (123), followed by several decades of mechanistic advances (24, 124–130). Newer tools such as single cell and snRNA-seq now enable more nuanced study of XCI escape across cell types (131–134), and allele-specific approaches enable direct measure of Xi expression (25, 27–29, 135–139). Our study uniquely combines snRNA-seq with an allele-specific approach to understand how aging alters Xa and Xi expression across cell types of the hippocampus, a region critical to cognition and targeted by aging, AD, and other neurodegenerative diseases.
Aging-induced activation of Xi, either through increased escapee expression or new escape altogether, increases the dose of activated X genes in cell types of the XX hippocampus. In other words, the aging female brain carries higher X expression due to the second “inactive” X. It is interesting to speculate that the X increase may benefit the aging brain since it is enriched for cognition-related genes. This may, in part, underlie why adding a second X to XY male mice improves, while subtracting one from XX female mice worsens, hippocampal-dependent cognitive functions in aging and an AD model (22, 23). It may also link with female-based resilience to cognitive aging (16–20) and higher baseline memory functions in typical aging and early AD (16, 20, 21, 140) observed in women, across many populations. However, X genes also harbor immune-related functions that exacerbate phenotypes, as noted with Tlr7 in the brains of aging male mice (141) and Kdm6a in lymphocytes of female mice in an experimental model of autoimmune encephalitis (142). In addition, an X factor, USP11, promotes tau modification and pathology in female mice (143). Whether these factors escaped from the Xi is unclear in these studies, but it is important to note that while an increase in Xi gene expression could confer cognitive resilience at large, not all X factors will uniformly benefit the brain at all life stages and conditions. Each will require causal testing for their cell type–specific role(s) in brain aging and neurodegenerative diseases. What Xi activation broadly means for women’s brain health—or for other systems of the body—is now a critical area of investigation.
Here, increasing Plp1 expression in oligodendrocytes of the DG, which recapitulates a component of age-induced Xi activation, improved cognition in old XX and XY mice. First, Plp1 is robustly expressed in oligodendrocytes, where it promotes myelination, a principal regulator of axonal conductance, and target of age-induced degeneration (144–147) that links to cognition (148). Oligodendrocyte vulnerability in aging impairs myelin plasticity through decreased remyelination and regenerative capacities alongside attenuated metabolic shuttling of lactate and pyruvate to axons (43). Therefore, increasing Plp1 in this cell type may boost the functions of oligodendrocytes and counter age-induced vulnerabilities.
Increasing Plp1 in the small hippocampal subregion of the DG improved cognitive functions, consistent with findings that interventions in cells forming a functional hub can modulate larger networks (149). Furthermore, Plp1-mediated cognitive improvement was acute and occurred when the intervention was introduced during old age, not requiring lifetime exposure to a transgenic alteration. This is important because treatments for cognitive aging, and for neurodegenerative diseases like AD, will require treatment late in life, and perhaps after symptoms emerge.
Beyond Plp1, aging activated the escape of several genes from the “inactive” Xi, suggesting its heightened accessibility to transcription. Epigenetic alterations are a hallmark of aging (150), and XCI is maintained by epigenetic programs of methylation, heterochromatin, and noncoding RNAs. It follows that the process of aging itself alters these programs across the genome including on the X (151), causing an overall loss of repression and increased transcription (150, 152, 153). More specifically, aging increases variability of the methylation (151) that widely silences genes on Xi (154), as observed in demethylation of regions escaping XCI in human leukocytes (153). Furthermore, aging increases chromatin accessibility (152), promoting increased gene expression (155), across the genome and particularly on topological-associated domain-like structures with high chromatin accessibility where escapees tend to cluster (156), as observed in hematopoietic stem cells (152). Collectively, these examples provide mechanistic insight into how aging may activate Xi in the brain, possibilities that require further study in neural cell types.
Given the density and high volume of epigenetic repression involved in maintaining XCI, the silent X may unintentionally but preferentially be affected by the global alterations of aging. Alternatively, or additionally, the aging-driven activation of Xi may offer an evolutionary advantage to females by increasing X dose, and thus potentially conferring resilience to cognitive decline. In the context of the debated “grandmother hypothesis” (157–161), a sharp mind and knowledge of resources could enhance the survival, reproductive success, and genetic fitness of a female’s children, grandchildren, and local community. Whether or not activation of the Xi is an intended or unintended consequence of aging, this remarkable process of partial “awakening” with age, first noted in the in 1980s in the liver (162), requires further investigation. The timing of Xi activation, its mechanistic and epigenetic orchestration, its conservation and overlap with humans (who show more X escape at baseline than mice) (163–165), and its functional significance for the aging female brain are all high-value areas of future research.
Emulating the effects of escape from XCI or even directly activating Xi genes could be a potential therapeutic strategy in brain aging and neurodegenerative diseases. Treatments simulating escape from XCI have been explored to treat Rett syndrome and other X-linked conditions. Since Rett syndrome is caused by a mutation in MEPC2 on one X (166, 167), therapies to reactivate the silenced wild-type MEPC2 or replace the mutated gene have been proposed (166, 168, 169). Our studies suggest that targeted elevation of Plp1, which is down-regulated in some studies of AD (170)—or perhaps targeting a number of other activated Xi genes—might improve cognitive deficits in aging and neurodegenerative disease.
In our analysis, most DEGs (Fig. 2E) and escapees (fig. S4) were cell type specific (60% and 56%, respectively), yielding high resolution of the Xi with aging. Notably, cell type classification was recapitulated by manual and automated (171, 172) annotation with 96% cell class overlap (data S7). While snRNA-seq datasets are inherently confounded by doublets and ambient RNA to some degree (173), we largely accounted for these confounders using Scrublet (174) and Cell Ranger, balancing resolution of cell type specificity while maintaining sufficient power to detect allele-specific reads.
Our study has several caveats and limitations. First, our experiments did not include males, who harbor a single Xa. Whether aging similarly increases X expression in males, as it does in females, remains an important question. Second, we did not study how parent-of-X origin, known to influence X gene expression in neurons (175), influences the age-induced modulation of the Xa or Xi in females. Third, in our allele-specific analysis, which uses two different reference genomes, mapping bias toward M. musculus may have occurred since it is a more refined and established genome than that of M. castaneus, which is more newly generated (176, 177). This could result in capturing less of the Xi and lead to a more conservative analysis. Fourth, while a great many SNPs differentiate the two mouse genomes, increasing discrimination of the Xa from Xi, some genomic areas show scant SNPs, which could lead to decreased discrimination in some genes. However, this genetic structure, consistent between young and old mice, would not influence age-induced escape findings. Fifth, genetic deletion of Xist on the Xa chromosome, used to enforce nonrandom XCI, may have precluded our detection of age-induced Xist changes. Finally, like for any snRNA-seq library, selection bias of RNAs captured in our study was unavoidable, as was the twofold decrease in power resulting from measuring each gene twice to assess allele-specific expression.
In summary, our study reveals aging-induced activation, repression, and remodeling of the “silent” Xi and its cell type specificity, with high resolution. Since Xi activation increases X dose, and X genes disproportionately influence neural and cognitive functions, our findings may imply a mechanism for female-based resilience to cognitive aging and increased baseline functions in early AD (16–21). Understanding how Xi escapees, such as Plp1, can contribute to counteracting vulnerabilities of brain aging may pave the way to therapeutic targets that benefit one or both sexes.
MATERIALS AND METHODS

Animals

All mouse studies were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of California, San Francisco (IACUC protocol number: AN204838). The studies complied with the National Institutes of Health guidelines. Mice were on a 12-hour light/dark cycle, in standard housing groups of five mice per cage with ad libitum access to food and water.

snRNA-seq mice

Female M. musculus XistloxP+/−,Zp3-cre mice (30, 178) with a congenic C57BL/6J background were crossed with male M. castaneus mice obtained from The Jackson Laboratory (strain 000928). XX progeny carrying the XistloxP+/−,Zp3-cre allele were used for snRNA-seq. Each experimental group consisted of age-matched, littermate controls. Four young (3 months) and four old (22 months) mice were euthanized. Their fresh hippocampi were dissected and frozen for downstream studies.
Plp1 overexpression in mice

Aged male (19 months) and female (17.5 months) mice from a congenic C57BL/6J background were used for Plp1 overexpression studies. Plp1 was up-regulated using an AAV2/5 vector that encodes Plp1 [National Center for Biotechnology Information (NCBI) Reference Sequence: NM_011123; 894 bp] from Applied Biological Materials (catalog number: #370271040110). The cytomegalovirus promoter was changed to an Mbp promoter. Downstream of Plp1, Gfp with an SV40 promoter is present. An additional control virus was used with the same sequence as the overexpression virus but no Plp1.

Mice were anesthetized (2 to 3% isoflurane) and placed in a stereotaxic frame using ear bars and tooth bar. AAV (5 μl per hemisphere) was stereotaxically injected into each DG of the hippocampus using the coordinates anterior-posterior (AP) = −2.1, medial-lateral (ML) = ±1.7, and dorsal-ventral (DV) = 1.9 at 3 μl/min, allowing for 10 min of diffusion. Experimenters were blinded to the identity of the viral injection.

Sequencing and bioinformatics

snRNA preparation

Nuclei were extracted using the Nuclei EZ Prep kit (Millipore Sigma), following the manufacturer’s protocol. Frozen hippocampi were homogenized in lysis buffer and centrifuged twice to isolate the nuclei from the cell debris. The nuclei were resuspended in glycerol-containing buffer and stored at −80°C. The libraries were prepped with the goal of 5000 captured nuclei per sample using the Chromium Single Cell 3′ Library and Gel Bead Kit v3.1, Chromium Next GEM Chip G, and Dual Index Kit TT Set A (10x Genomics). The samples were sequenced by Novogene USA on an Illumina NovaSeq 6000 with paired-end 150-bp reads PE150 (Novogene) with the goal of 100,000 reads per target captured cell.

snRNA-seq data analysis

A combined M. musculus (GRCm39) and M. castaneus (CAST_EiJ_v1.111) genome was created using the mkref function of Cell Ranger (Cell Ranger/8.0 10x Genomics). FASTQ files were aligned to the combined reference genome using Cell Ranger Count v8.0.0 with the -include introns flag. Combined feature-barcode matrixes were formed using the aggregate function of Cell Ranger (Aggr v8.0.0). The resultant feature-barcode matrix was analyzed using Scanpy (31). The matrix was filtered by removing barcodes with less than 100 corresponding genes and removing genes associated with fewer than three barcodes.
Scrublet was used to remove predicted doublets from the dataset (174). Data were normalized to total counts and logarithmized. Dimension reduction analysis was conducted using uniform manifold approximation and projection (UMAP). The Leiden algorithm was used to cluster the cells at a 0.5 resolution. The Wilcoxian method was used to determine the top 100 DEGs in each group. The genes were compared to clusters in the Allen Brain Atlas 10x hippocampus and cortex dataset, and clusters were assigned (32). Clusters with similar markers were combined for downstream analysis. Automated annotation was conducted with CellTypist (171, 172).
For DEG analysis, barcodes were pseudobulked by sample and cell cluster for downstream analysis using the decoupler and pyDEseq2 programs (36, 37) after filtering out genes with a minimum of 10 reads across samples and 10 counts in an individual sample. Genes were considered to be differentially expressed if the adjusted P value was <0.05, and the log2 fold change was >0.1 or <−0.1. Typical to DEseq2 pipelines, the P values are calculated with the Wald test and are adjusted using the Benjamini-Hochberg procedure.
GO term analysis was conducted for each cell type. Genes were ranked using the Scanpy function “rank_genes_groups” and grouped by cell type. The top 350 X chromosome genes (ranked by P value and log2 fold change) were evaluated with the g:Profiler if their adjusted P value was <0.05, and the log2 fold change was >0.1 or <−0.1 (39). P values are adjusted with the g:SCS method (179). GO term analysis was also conducted on significant (adjusted P value was <0.05, and the log2 fold change was >0.1 or <−0.1) X DEGs using g:Profiler (39).
Computational escape

Escape was assessed using a method adapted from Berletch et al. (25). The M. musculus and M. castaneus normalized reads were summed for each gene, and the escape ratio was evaluated by taking the proportion of M. castaneus reads from the total reads for the gene (Eq. 1).
Xa is the total normalized reads assigned to the M. musculus genome for the given cell type and age context.

Xi is the total normalized reads assigned to the M. castaneus genome for the given cell type and age context.

Pesc is the ratio of Xi reads
The escape ratio was corrected for mapping bias using the methods explained by Berletch et al. (25) using the ratio of the total autosomal reads to the M. musculus and M. castaneus genomes (Eq. 2).
ATcast is the total normalized autosomal reads from the M. castaneus genome across all genes.

ATmus is the total normalized autosomal reads from the M. musculus genome across all genes.

Pescadj is the adjusted escape ratio
A 99% CI was conducted for each escape ratio (Eq. 3)
If the lower bound of the CI was greater than 0 and the corrected escape value was greater than 0.05, the gene was considered to escape. If Xa or Xi was lower than the 5th percentile for X allelic reads in all cell types, the gene was not considered an escapee. If Xa was lower than the 5th percentile for X allelic reads in all cell types, age-related escape trends were not calculated. Escape was assessed by cell type and age group. GO term analysis was conducted on escapees following certain trends with aging using the program g:Profiler (39).
Human PLP1 analysis

MSBB cohort data were accessed from the Accelerating Medicines Partnership Alzheimer’s Disease (AMP-AD) knowledge portal at Synapse (180). Four different cortical regions—Broadmann area 10 (frontal pole), Broadmann area 22 (superior temporal gyrus), Broadmann area 36 (parahippocampal gyrus), and Broadmann area 44 (inferior temporal gyrus)—were analyzed (181). Using the DESeq2 package in R, raw PLP1 count data for each region were normalized for library size and modeled as a function of sex while adjusting for standard covariates used in the MSBB processing pipeline: batch, age at death, race, postmortem interval, RIN, exonic rate, and rRNA rate (180, 182). Sex differences are reported as log2 fold change of normalized PLP1 expression. Cases with available uncensored age-at-death data were included in the final analysis (frontal pole: n = 161, superior temporal gyrus: n = 159, parahippocampal gyrus: n = 184, inferior temporal gyrus: n = 150).
In vitro and in vivo experiments

Immunohistochemistry

AAV-infected mice (2.5 months postinfection) were perfused with a peristaltic pump for 4 min using cold phosphate-buffered saline (PBS) (10 ml/min). Whole brains were collected and postfixed for 48 hours in 4% (w/v) paraformaldehyde. The brains were preserved in 30% (w/v) sucrose in PBS. A freezing sliding microtome (Leica) was used to section whole brains into 40-μm-thick coronal slices and stored in cryoprotective medium at −20°C.

Individual sections were blocked with 10% normal donkey serum. The blocked sections were incubated with primary antibodies [rabbit anti-GFP (1:1000, Sigma-Aldrich, G1544) and mouse anti-OLIG2 (1:200, Proteintech, 66513-1-IG)] at 4°C overnight. The sections were washed and incubated with secondary antibodies [donkey anti-rabbit 488, donkey anti-mouse 594 Alexa Fluor–conjugated secondary antibodies (1:1000, Invitrogen A-21206, Invitrogen R37115)] at room temperature for 2 hours, with a final addition of 300 nM 4′,6-diamidino-2-phenylindole (DAPI). Sections were washed and mounted on a microscope (Nikon CSU-W1) using Micro Manger 2.0 Gamma (183). The microscope has a Zyla 4.2 CMOS camera (Andor), piezo XYZ stage (ASI), CSU-W1 Spinning Disk with Borealis upgrade (Yokogowa/Andor), Spectra-X (Lumencor), CSU-W1 Penta Dichroic 405/488/561/640/755, ILE 4 line solid-state Laser Launch a slide with Vectashield.
Sections were imaged on the fluorescence microscope with a spinning disk confocal (405/488/561/640 nm; Andor). Images were taken using the Plan Apo λ 20×/0.75 objective using solid-state lasers 405 and 488 nm and emission filters 447/60 and 525/50 (Semrock) (for DAPI and GFP, respectively). Additional images were taken with the Plan Apo λ 40×/1.3 oil objective using solid-state lasers 405, 488, and 561 nm and emission filters 447/60, 525/50, and 607/36 (Semrock) [for DAPI, GFP, and red fluorescent protein (RFP), respectively]. Images were stitched and processed using Fiji (184). Images (40×) were used for cell counting, described below.
Cell counting

Microscopy images were analyzed using the segment artificial intelligence (AI) analysis from the Nikon NIS Elements AR software version 5.41.00 64bit. The images were preprocessed using the Mexican Hat, Fast Smoothing, and Subtracting backgrounds functions to make the nuclei clearer and increase the ability of the software to identify the nuclei. The segment AI was trained on 512 × 512 segments of the DG until the AI was reliably able to identify about 95% of DG nuclei. The trained segment AI was then applied to the DG images, where it counted the nuclei and measured the intensity of GFP and RFP in the cytoplasm. A threshold of 350 intensity units and 200 intensity units was set for RFP and GFP, respectively. All nuclei that met that threshold were recorded as positive for that signal. Cells that were positive for GFP and RFP were counted as cells with colocalization.

Mixed glial culture

Using methods previously described (185), primary mixed glia were cultured from postnatal 0 to 2 congenic C57BL/6J male XY and female XX mouse pups. Cells were plated at a density of 106 cells/ml in 24-well plates. The cells matured in Dulbecco’s modified Eagle’s medium. At day 5 (division 2) in vitro, they were treated with the Plp1 AAV (described in the Plp1-OE mouse section) at multiplicity of infection (MOI) = 2. At division 4, cells were treated with cytosine arabinoside to reduce glial proliferation (186). At division 10, cells were harvested, and RNA was isolated.
Quantitative PCR

qPCR was conducted on whole hippocampi from young (3 months) (n = 13; XX, n = 15) and old (20 to 35 months) (XY, n = 10; XX, n = 15) XX and XY mice. Mouse Plp1 transcript variant 2 F: 5′CTTCCTTTATGGGGCCCTCC and R: 5′GGTGGTCTTGTAGTCGCCAA primers were used.

Behavior

Behavioral studies were conducted beginning 4 weeks after AAV injection, and experimenters were blinded to the treatments. Studies were conducted on age-matched controls and during the light cycle. All chambers, objects, and areas were cleaned with 70% alcohol between testing sessions.

Elevated plus maze

The EPM was carried out as described (187, 188). For 1 hour before testing, mice were habituated in the testing room in dim light. During testing, mice were placed in the center of the EPM facing the open arm and could explore for 10 min. Their time spent in each region and distance traveled in the open and closed arms were recorded using the Kinder Scientific Elevated Plus-Maze (Chula Vista, CA) and MotorMonitor system.
Open field

The open field was carried out as described (23, 187–189). Mice were acclimated for 30 min and explored the open field for 10 min. The total activity in the open field (clear chamber 41 × 30 cm) was detected by beam breaks and measured with an automated Flex-Field/Open Field Photobeam Activity System (San Diego Instruments).
Two-trial large Y-maze

The mice were evaluated in the two-trial large Y-maze following the described protocols (188, 190, 191). Mice were trained by allowing exploration of one arm of the Y with a visual cue. Following 16 hours after completion of training, the novel arm with a different visual cue was opened and mice were allowed access. Mice then explored all arms of the two-trial large Y-maze. The duration in the novel compared to the familiar arm, a measure of spatial learning and memory, was assessed.
Statistical analysis

Statistical analyses for in vitro and in vivo behavioral experiments were carried out using GraphPad Prism (version 7.0) for t tests and two-way analyses of variance (ANOVAs). All tests were two-tailed unless indicated otherwise. Differences between two means were assessed using unpaired t tests and a two-way ANOVA to assess differences among multiple means for all experiments unless otherwise stated. Error bars represent SEM, and null hypotheses were rejected at or below a P value of 0.05 when rounded to two decimal points.

Acknowledgments

We acknowledge D. Velasco for illustrations in Figs. 1A, 4A, and 7A; E. Kakoshina for formatting Fig. 2; A. Moreno for illustrations in Figs. 6D and 7A; C. Chen and E. Davis for mouse colony management; S. Nanda for helping with the initial round of genomic library processing; and the Center for Advanced Light Microscopy for assistance with imaging.
Funding: This work was supported by NIH grants RF1AG079176-01 and RF1AG068325-01 (to D.B.D.), Simons Foundation grant 811225SPI (to D.B.D.), Simons Foundation Transition to Independence Award (to S.A.-S.), Philanthropy (to D.B.D.), Bakar Aging Research Institute (to F.M. and D.B.D.), Simons Foundation grant SF811217 (to B.A.B.), NIH-NIBIB T32 EB001631 (to L.W.B.), RSNA R&E Foundation RR24-217 (to L.W.B.), NIH-NIA R01AG062588 (to J.S.Y.), Alzheimer’s Association AARF-23-1145318 (to R.S.), and New Vision Research CCAD 2024-001-1 (to R.S.).

Author contributions: Conceptualization: M.G., C.K.S., V.R., and D.B.D. Methodology: M.G., C.K.S., S.A.-S., R.S., L.W.B., B.A.B., K.B.C., V.R., and D.B.D. Investigation: M.G., C.K.S., S.A.-S., R.S., F.M., D.W., and L.W.B. Visualization: M.G., C.K.S., S.A.-S., R.S., F.M., and D.B.D. Supervision: J.S.Y., B.P., B.A.B., K.B.C., V.R., and D.B.D. Writing—original draft: M.G. and D.B.D. Writing—review and editing: M.G., C.K.S., S.A.-S., R.S., F.M., D.W., L.W.B., J.S.Y., B.P., B.A.B., K.B.C., V.R., and D.B.D.

Competing interests: D.B.D. serves on the Board of the Glenn Medical Foundation and consulted for Unity Biotechnology (unrelated to the content of the manuscript) and SV Health Investors (unrelated to the content of the manuscript). J.S.Y. serves on the scientific advisory board for the Epstein Family Alzheimer’s Research Collaboration. All other authors declare that they have no competing interests.

Data and materials availability: All data are available in the main text or Supplementary Materials or as raw and processed files available at GEO series accession number GSE276652. Codes for the sequencing filtering normalization, Xa/Xi expression, and pseudobulking are available at https://github.com/margaretgadek/XCIEscapeAging-/releases/tag/v1.0.0 and https://zenodo.org/records/13750335.
Supplementary Materials

The PDF file includes:

Supplementary Text

Figs. S1 to S5

Table S1

Legends for data S1 to S8

References

Other Supplementary Material for this manuscript includes the following:

REFERENCES AND NOTES

1

D. B. Dubal, C. Murphy, Y. Suh, B. A. Benayoun, Biological sex matters in brain aging. Neuron 113, 2–6 (2025).

2

V. Zarulli, J. A. Barthold Jones, A. Oksuzyan, R. Lindahl-Jacobsen, K. Christensen, J. W. Vaupel, Women live longer than men even during severe famines and epidemics. Proc. Natl. Acad. Sci. U.S.A. 115, E832–E840 (2018).

3

United Nations Department of Economic and Social Affairs, Population Division, World Population Ageing (United Nations Department of Economic and Social Affairs, 2020).

4

A. M. Bronikowski, R. P. Meisel, P. R. Biga, J. R. Walters, J. E. Mank, E. Larschan, G. S. Wilkinson, N. Valenzuela, A. M. Conard, J. P. de Magalhaes, J. E. Duan, A. E. Elias, T. Gamble, R. M. Graze, K. E. Gribble, J. A. Kreiling, N. C. Riddle, Sex-specific aging in animals: Perspective and future directions. Aging Cell 21, e13542 (2022).

(0)eLetters

eLetters is a forum for ongoing peer review. eLetters are not edited, proofread, or indexed, but they are screened. eLetters should provide substantive and scholarly commentary on the article. Neither embedded figures nor equations with special characters can be submitted, and we discourage the use of figures and equations within eLetters in general. If a figure or equation is essential, please include within the text of the eLetter a link to the figure, equation, or full text with special characters at a public repository with versioning, such as Zenodo. Please read our Terms of Service <https://www.science.org/content/page/terms-service> before submitting an eLetter.

Log In to Submit a Response <https://www.science.org/action/ssostart?redirectUri=/doi/full/10.1126/sciadv.ads8169>
No eLetters have been published for this article yet.