Aging disrupts spatiotemporal coordination in the cycling murine ovary Authors: Tammy C. T. Lan, Alison Kochersberger, Ruth Raichur, Sophia Szady, Hien Tran, Radiana Simeonova, Ashley Helmuth, Andrew Minagar, Maria José Orozco Fuentes, Vipin Kumar, Giovanni Marrero, Irving Barrera, Sarah Mangiameli, Alex K. Shalek, Pardis C. Sabeti, Fei Chen, David S. Fischer, Jennifer L. Garrison, Hattie Chung Published: 2026 ABSTRACT Throughout the female reproductive lifespan, the ovary undergoes hundreds of cycles of follicle development, ovulation and tissue regeneration. How aging disrupts the coordination of such precise, multicellular interactions across time and space is not well understood. Using Slide-seq, a near-cellular spatial transcriptomics method, here we profile 22 mouse ovaries across the reproductive cycle and chronological ages, capturing 610,620 spots across 69 spatial profiles. We develop a novel segmentation pipeline to examine the multicellular dynamics of 358 oocytes, 668 follicles and 236 corpora lutea to find that aging impairs the spatial and temporal coordination required for folliculogenesis even before reproductive cycles cease. These disruptions are characterized by altered immune cell dynamics, inflammatory signaling and global tissue disorganization, which impair the cyclic remodeling required for ovarian function. Our findings reveal how multicellular niches orchestrate ovarian function and demonstrate that age-related breakdown in tissue organization precedes the end of fertility. Main The ovary is a highly dynamic organ that undergoes repeated cycles of phenotypic switching to orchestrate oocyte maturation, ovulation and endocrine signaling throughout the reproductive cycle1,2. This ‘homeostatic cycling’ occurs approximately 150–650 times in mice and 300–400 times in humans throughout the female reproductive lifespan3,4. Chronological aging leads to a progressive decline in reproductive function, ultimately culminating in the cessation of these cycles. Here we investigate the molecular interactions underlying this transition, focusing on the depletion of the ovarian reserve, the loss of precision in cellular interactions and the accumulation of tissue-wide inflammation and fibrosis5–10. Central to ovarian function are follicles, which are multicellular structures that develop growing oocytes and transform into corpora lutea (CL) after ovulation. Folliculogenesis begins with primordial follicles, consisting of a dormant oocyte encased by flattened granulosa cells (GCs). Follicle activation triggers GC proliferation, first forming preantral follicles that undergo morphological and functional changes to support oocyte growth and then transitioning to the gonadotropin-dependent and steroidogenic antral stage11,12. Notably, less than 1% of recruited follicles reach ovulation while the rest undergo atresia (that is, controlled follicle death)13. After ovulation, follicular remnants transform into the CL, a transient progesterone-producing endocrine structure essential for pregnancy and cycle regulation12,14. Follicle maturation spans multiple reproductive cycles, approximately 18 days in mice and 375 days in humans, creating an extraordinarily complex tissue environment where dozens (mice) or thousands (humans) of follicles at varying developmental stages share the same physical space with multiple CLs15,16. The molecular mechanisms coordinating such asynchronous spatiotemporal organization remain poorly defined, underscoring the need for spatially resolved profiling to decode the dynamics of these cellular ecosystems. Prior studies of ovarian aging have focused primarily on follicle depletion and oocyte-intrinsic changes, such as aneuploidy17,18. However, mounting evidence suggests that dysfunction across the entire ovarian tissue environment also contributes substantially to reproductive decline7,9,18–20. Known disruptions include GC signaling21–23, extracellular matrix (ECM) remodeling24–27 and aberrant immune infiltration9,28–30, yet how these events collectively drive ovarian aging remains unknown. Although single-cell RNA sequencing (scRNA-seq) is a powerful tool for uncovering cellular heterogeneity and its regulatory mechanisms, it is limited in uncovering insights for ovarian research. Droplet-based scRNA-seq methods often fail to capture mature oocytes due to their large size, which can exceed 70 µm in diameter, leading to a bias toward early-stage oocytes31–34. Furthermore, scRNA-seq dissociates interacting cell types, thereby losing the spatial context of the follicular niche, whereas coarse spatial transcriptomics lacks the resolution needed to characterize follicular architectures20,35. Here we present a spatiotemporal atlas of the naturally cycling C57BL/6 J mouse ovary throughout chronological aging. Using Slide-seq36,37, we generated 69 spatial transcriptomic profiles from 22 biological samples, capturing 610,620 spots at 10 µm resolution during the estrus and metestrus stages. To analyze the cooperative nature of multicellular ovarian structures, we developed a segmentation pipeline that identified 358 primordial and growing oocytes, 668 follicles of which 301 enclosed oocytes and 236 CLs, enabling a high-resolution analysis of their in situ molecular signatures. Our results reveal that aging disrupts the temporal dynamics of both folliculogenesis and CL function. This dysregulation aligns with heightened inflammatory signaling and a global decline in tissue organization, which may compound ovarian dysfunction. Our work demonstrates how the breakdown of coordinated multicellular niches contributes to the age-related decline of reproductive and endocrine function. Results Spatial transcriptomics reveals multicellular organization of the cycling and aging ovary To characterize the multicellular architecture of the ovary, we generated spatial transcriptomics profiles from mice across three reproductive age groups: young (10–12 weeks), middle-aged (36–40 weeks) and old (52–54 weeks) mice38. Mice were harvested during the ovulatory (estrus) and post-ovulatory (metestrus) phases, with active cycling confirmed via longitudinal estrous staging (Fig. 1a, Extended Data Fig. 1a, Supplementary Tables 1 and 2 and Methods). Unsupervised clustering and post hoc annotation resolved all major ovarian cell populations, including oocytes (Gdf9), multiple GC subtypes, theca cells (Col1a2) and two distinct luteal populations (Parm1 or Akr1c18). We also identified epithelial (Epcam), fibroblast (Pdgfra), smooth muscle (Tagln), immune (Lyz2), endothelial (Flt1) and stromal (Aldh1a1) cells (Fig. 1b–d, Extended Data Fig. 1b, Supplementary Tables 3–5 and Supplementary Note). To validate these Slide-seq annotations, we also performed whole-ovary single-nucleus RNA-seq (snRNA-seq). Although snRNA-seq recovered a consistent set of cell types, it showed a notable loss of oocytes and cumulus cells, underscoring the advantage of our spatial profiling for capturing a more comprehensive ovarian cellular atlas (Extended Data Fig. 1c,d).Fig. 1A spatiotemporal map of the cycling and aging ovary identifies spatial niches driving folliculogenesis.a, Experimental overview. Mice were staged by vaginal cytology and collected at estrus and metestrus (n = 22; young 10–12 weeks, middle-aged 36–40 weeks, old 52–54 weeks). Multiple nonadjacent sections per ovary were profiled using Slide-seq, with a minimum gap of 200 µm to avoid resampling the same follicle or CL. Adjacent sections were used for histology. After clustering and cell-type annotation, multicellular segments were defined using an automated segmentation pipeline. b, UMAP of Slide-seq RNA profiles (n = 610,620 dots at 10 µm resolution) across 69 pucks, colored by post hoc annotation (legend); GC, granulosa cell. c, Mean expression (log-normalized counts, dot color) and proportion of expressing cells (dot size) of marker genes (columns) used to annotate cell types (rows). Spot counts in Supplementary Table 4. d, Mean cell-type composition across young (n = 32), middle (n = 16) and old (n = 21) pucks. e, Schematic of folliculogenesis and CL formation. GCs proliferate from a single layer around a dormant oocyte in primordial follicles to form preantral follicles, which transition to gonadotropin-dependent antral follicles with fluid-filled antral cavity formation and theca recruitment. GCs differentiate into multiple subtypes. Atresia can occur at any stage of folliculogenesis. After ovulation, GCs luteinize to form the CL, which is subsequently cleared. f, Mean expression (log-normalized counts and variance-scaled) of key hormone receptors and synthesis genes (columns) across segment types (rows), grouped by gonadotropin-independent, gonadotropin-dependent and luteinization stages. Segment counts: preantral (n = 238), antral mitotic (n = 191), antral luteinizing (n = 43), atretic (n = 181), CL—Parm1 (n = 87), CL—Akr1c18 (n = 147); n = 22 mice across all ages. g, Representative H&E sections with corresponding spatial segment maps showing annotated preantral follicles (purple), antral follicles (green-blue) and CLs (yellow). Asterisks denote structures present in either H&E or Slide-seq due to sectioning differences of sampling, rather than an error in the segmentation. h, Segmentation accuracy assessed across n = 347 segments with an available adjacent H&E section for validation; the majority (n = 311) were correct. Errors included merged segments with other segments, misannotations with incorrect segment types and missed detections by the pipeline. Cohen’s kappa = 0.85. i, Follicle counts from paired H&E sections and Slide-seq segments across pucks (dots) and their means (bars) for young (n = 32), middle (n = 16) and old (n = 21) pucks. Error bars represent standard error of the mean (s.e.m.). Significance was determined by two-tailed t-test; all other bar groups are not significant. A spatiotemporal map of the cycling and aging ovary identifies spatial niches driving folliculogenesis. a, Experimental overview. Mice were staged by vaginal cytology and collected at estrus and metestrus (n = 22; young 10–12 weeks, middle-aged 36–40 weeks, old 52–54 weeks). Multiple nonadjacent sections per ovary were profiled using Slide-seq, with a minimum gap of 200 µm to avoid resampling the same follicle or CL. Adjacent sections were used for histology. After clustering and cell-type annotation, multicellular segments were defined using an automated segmentation pipeline. b, UMAP of Slide-seq RNA profiles (n = 610,620 dots at 10 µm resolution) across 69 pucks, colored by post hoc annotation (legend); GC, granulosa cell. c, Mean expression (log-normalized counts, dot color) and proportion of expressing cells (dot size) of marker genes (columns) used to annotate cell types (rows). Spot counts in Supplementary Table 4. d, Mean cell-type composition across young (n = 32), middle (n = 16) and old (n = 21) pucks. e, Schematic of folliculogenesis and CL formation. GCs proliferate from a single layer around a dormant oocyte in primordial follicles to form preantral follicles, which transition to gonadotropin-dependent antral follicles with fluid-filled antral cavity formation and theca recruitment. GCs differentiate into multiple subtypes. Atresia can occur at any stage of folliculogenesis. After ovulation, GCs luteinize to form the CL, which is subsequently cleared. f, Mean expression (log-normalized counts and variance-scaled) of key hormone receptors and synthesis genes (columns) across segment types (rows), grouped by gonadotropin-independent, gonadotropin-dependent and luteinization stages. Segment counts: preantral (n = 238), antral mitotic (n = 191), antral luteinizing (n = 43), atretic (n = 181), CL—Parm1 (n = 87), CL—Akr1c18 (n = 147); n = 22 mice across all ages. g, Representative H&E sections with corresponding spatial segment maps showing annotated preantral follicles (purple), antral follicles (green-blue) and CLs (yellow). Asterisks denote structures present in either H&E or Slide-seq due to sectioning differences of sampling, rather than an error in the segmentation. h, Segmentation accuracy assessed across n = 347 segments with an available adjacent H&E section for validation; the majority (n = 311) were correct. Errors included merged segments with other segments, misannotations with incorrect segment types and missed detections by the pipeline. Cohen’s kappa = 0.85. i, Follicle counts from paired H&E sections and Slide-seq segments across pucks (dots) and their means (bars) for young (n = 32), middle (n = 16) and old (n = 21) pucks. Error bars represent standard error of the mean (s.e.m.). Significance was determined by two-tailed t-test; all other bar groups are not significant. We next developed a segmentation pipeline to group anatomically and functionally related spatial spots, namely oocytes, follicles and CLs. Our method identified follicular segments across distinct stages of folliculogenesis, including preantral (n = 238), antral mitotic (n = 191), antral luteinizing (n = 43) and atretic (n = 181) follicles, supported by the stage-specific expression of known hormone receptors and synthesis genes (Fig. 1e,f, Extended Data Fig. 1e–g, Supplementary Table 6 and Methods)39,40. We also identified two spatially segregated CL types, classified as Parm1 (n = 87) and Akr1c18 (n = 147) (Fig. 1h). Although current resolution limits precluded the identification of squamous GCs in primordial follicles, we identified primordial oocytes through a separate pipeline that expressed additional marker genes such as Sohlh1 and Nobox41 (Extended Data Fig. 1h and Methods). All segmentation results for follicles and CLs were validated by manual annotation of adjacent H&E-stained sections (Fig. 1g–i, Extended Data Fig. 1i, Supplementary Figs. 1 and 2, Supplementary Table 5 and Methods). Furthermore, to provide orthogonal validation of atretic follicles beyond transcriptional signatures32, we examined the spatial distribution of immune cells within follicles. As expected9, atretic follicles exhibited significant immune infiltration toward the follicle center consistent with clearance, whereas healthy follicles had minimal immune presence (Extended Data Fig. 1j and Supplementary Table 7; P < 0.05 antral versus atretic follicles, Tukey’s test). Altogether, these findings demonstrate that our segmentation pipeline accurately captures the multicellular organization and functional heterogeneity of the ovary from spatial transcriptomics data. Oocytes follow coordinated transcriptional programs during follicular development Although oocytes are traditionally viewed as transcriptionally silent storage units for maternal RNA42, recent evidence reveals active transcriptional regulation throughout their maturation43–45. Indeed, our in situ profiling of oocyte transcriptomes showed that canonical oocyte marker genes were differentially expressed across their follicle stages (Fig. 2a–c and Supplementary Table 8)31,46–48. Further analysis via non-negative matrix factorization (NMF) revealed five distinct oocyte gene programs across their maturation (Fig. 2d, Methods, Extended Data Fig. 2a–c and Supplementary Table 8). Oocyte program 2, characterized by zona pellucida glycoproteins genes Zp2 and Zp3, was upregulated in primordial and preantral oocytes, suggesting the synthesis of components essential for follicular assembly following primordial activation49,50. In contrast, oocyte program 4, enriched in Bmp15 and F-box proteins genes (Fbxw19, Fbxw21, Fbxw24) encoding components of the SCF ubiquitin E3 ligase complex controlling cell cycle progression51,52, emerged in mature antral oocytes, reflecting shifts required for meiotic progression. In mature oocytes, we observed a significant increase in oocyte program 0 (Bcl2l10, Btg4, Tcl1), signaling readiness for growth following the gonadotropin surge (Fig. 2e, Extended Data Fig. 2b and Supplementary Table 8; P = 0.0001 preantral versus antral; two-tailed t-test)53–55. Conversely, atretic oocytes failed to maintain the expression of these programs, showing a loss of Padi6 expression and a premature induction of oocyte program 0 (Fig. 2a,e, Extended Data Fig. 2b and Supplementary Table 8; P = 0.0203 antral versus atretic; two-tailed t-test). Altogether, our findings reveal that oocyte transcription may be dynamically regulated throughout folliculogenesis to govern the switch from oocyte dormancy to maturation.Fig. 2Oocyte transcriptional dynamics across folliculogenesis and age.a, Mean expression (log-normalized counts and variance-scaled) of oocyte marker genes (columns) across follicle stages (rows). n = 358 oocytes across 22 mice, all age groups; oocyte counts per follicle type in Supplementary Table 8. b, Principal component (PC) embedding of n = 358 oocyte RNA profiles (dot) colored by its enclosing follicle stage. c, Representative spatial plots of two pucks showing oocytes (colored dots) within their tissue context, colored by enclosing follicle stage; surrounding GC spots shown in gray. d, Expression (color) of oocyte gene programs. Top contributing genes for each program are indicated; gray text box. e, Violin plot of median expression (color) and distribution (kernel density) of oocyte gene programs (columns) across follicle stages (rows). n = 358 oocytes across 22 mice; oocyte counts per follicle type in Supplementary Table 8. *P = 0.0203, ***P = 0.0001. f, Mean oocyte count per puck (y axis) by enclosing follicle type (color) across age groups (x axis). Oocyte counts by age: young (n = 229), middle-aged (n = 89) and old (n = 40). g, Differentially expressed oocyte genes across age. Select genes (y axis) showing significant monotonic changes in oocytes (dots) across age (colors) are shown; two-tailed t-test. Differentially expressed genes were identified between young and old non-primordial oocytes (two-tailed t-test, Benjamini–Hochberg correction), followed by pairwise testing across young, middle and old to identify monotonic trends. Non-primordial oocyte counts by age are shown in Supplementary Table 9. Oocyte transcriptional dynamics across folliculogenesis and age. a, Mean expression (log-normalized counts and variance-scaled) of oocyte marker genes (columns) across follicle stages (rows). n = 358 oocytes across 22 mice, all age groups; oocyte counts per follicle type in Supplementary Table 8. b, Principal component (PC) embedding of n = 358 oocyte RNA profiles (dot) colored by its enclosing follicle stage. c, Representative spatial plots of two pucks showing oocytes (colored dots) within their tissue context, colored by enclosing follicle stage; surrounding GC spots shown in gray. d, Expression (color) of oocyte gene programs. Top contributing genes for each program are indicated; gray text box. e, Violin plot of median expression (color) and distribution (kernel density) of oocyte gene programs (columns) across follicle stages (rows). n = 358 oocytes across 22 mice; oocyte counts per follicle type in Supplementary Table 8. *P = 0.0203, ***P = 0.0001. f, Mean oocyte count per puck (y axis) by enclosing follicle type (color) across age groups (x axis). Oocyte counts by age: young (n = 229), middle-aged (n = 89) and old (n = 40). g, Differentially expressed oocyte genes across age. Select genes (y axis) showing significant monotonic changes in oocytes (dots) across age (colors) are shown; two-tailed t-test. Differentially expressed genes were identified between young and old non-primordial oocytes (two-tailed t-test, Benjamini–Hochberg correction), followed by pairwise testing across young, middle and old to identify monotonic trends. Non-primordial oocyte counts by age are shown in Supplementary Table 9. Aging alters regulators of oocyte maturation Aging leads to both the depletion and functional decline of maturing oocytes56,57. Consistent with this, we observed a sharp age-associated decrease in oocyte counts across all follicle stages (Fig. 2f and Supplementary Table 5). Differential expression analysis between maturing oocytes of young and old animals revealed transcriptional alterations, notably the upregulation of the genes encoding the nuclear androgen receptor (Ar) and the progesterone-metabolizing enzyme (Akr1c18) (Fig. 2g, Extended Data Fig. 2d, Supplementary Tables 9 and 10 and Methods). Given their roles in oocyte maturation and meiosis58–60, their increase may reflect an adaptive shift in response to the altered endocrine landscape of the aging ovary61,62. Conversely, we observed reduced expression of the insulin-degrading enzyme gene Ide. Because Ide modulates insulin signaling to maintain oocyte quality, attenuating insulin signaling during healthy folliculogenesis63,64, its downregulation suggests a disruption in metabolic homeostasis64,65. We also detected the downregulation of the Nlrp9b family genes (Nlrp9a/b/c), components of the subcortical maternal complex thought to facilitate the oocyte-to-embryo transition66–68 (Extended Data Fig. 2e and Supplementary Table 11). Importantly, these trends were consistent in oocytes across follicle types, suggesting that these changes reflect oocyte-intrinsic changes rather than shifts in follicular composition with age (Extended Data Fig. 2f and Supplementary Table 11). Together, our results suggest that aging impairs the integration of external cues that guide oocyte maturation, providing a potential mechanistic basis for the age-related decline in oocyte quality. Molecular programs define spatially organized follicle lifecycles in young ovaries Next, we investigated how GC states change throughout folliculogenesis. To compare GC organization across hundreds of follicles of varying sizes, we devised a normalized, scaled radial distance from the follicle center. Quantifying GC composition along this radial axis revealed extensive intra-follicle heterogeneity, indicating that transitions between follicle stages are not driven by a synchronized follicle-wide switch, but instead may emerge through the local propagation of changes in cell–cell interactions (Fig. 3a,b, Extended Data Fig. 3a, Supplementary Table 6 and Methods).Fig. 3Spatiotemporal coordination of GCs within the follicular niche.a, Representative spatial plot showing follicular cell types (colors), including preantral, antral early mitotic, antral mitotic, antral luteinizing and atretic GC subtypes, alongside oocyte and theca cell spots. b, Radial distribution of GC subtypes (colored lines, y axis) across follicle types (panels). Lines indicate mean proportion of each subtype across 10 scaled radial bins (x axis; 0, follicle center; 1, edge); shading, s.e.m. across follicles. Significant monotonic radial trends are indicated (preantral follicles: GC—preantral *P = 0.0283, antral mitotic follicles: GC—antral mitotic *P = 0.0479, GC—antral luteinizing *P = 0.0479, antral luteinizing follicles: GC—antral luteinizing *P = 0.0045, atretic follicles: GC—atretic *P < 0.0001, two-sided Spearman correlation). n = 653 follicles from 22 mice, all ages; follicle counts by type in Supplementary Table 6. c, Spatial organization of GC cell cycle states. Left: representative spatial plot with cycling (G2M/S, pink) and non-cycling (G1, gray) GCs. Right: mean radial distribution of cycling vs non-cycling GCs (y axis); shading, s.e.m. across follicles. Significant monotonic trends indicated (cycling: *P < 0.0001, non-cycling: *P < 0.0001, two-sided Spearman correlation; n = 668 follicles). d, Diffusion pseudotime (DPT) across non-atretic and atretic GC trajectories. Left: representative spatial plot with DPT values (colorbar) along non-atretic (purple) and atretic (orange) paths. Right: mean DPT (y axis) across radial bins; shading, s.e.m. across follicles. Significant trends indicated (atretic: *P < 0.0001, non-atretic: *P = 0.0001, two-sided Spearman correlation). n = 653 total follicles from 22 mice, all ages; follicle counts by type in Supplementary Table 7. e, Immunofluorescence of Ki-67 (yellow) and αSMA (red) in a young antral follicle. Scale bar, 100 µm. Representative result across two independent experiments. f, Fluorescence in situ hybridization of Cyp19a1 (yellow) mRNA in a young ovary. Scale bar, 100 µm. Representative result across four independent experiments. g, GC gene programs (GEPs). Left: representative spatial patterns of two distinct GEPs (colorbar), FSH regulation (left) and luteinization (right), with top contributing genes in gray. Right: mean program usage score (y axis) across radial bins (x axis) for preantral (purple), antral mitotic (maroon), antral luteinizing (teal) and atretic (orange) follicles (shading, s.e.m.). n = 653 follicles across 22 mice, all ages; follicle counts by type in Supplementary Table 6. h, Dynamics of GC GEPs across pseudotime (DPT), follicle type and estrous stage in young (n = 9) and old (n = 7) mice. Top: kernel density distribution of follicle types across normalized pseudotime; estrous stage composition below. Middle and bottom: mean normalized GEP usage (color) in non-atretic (middle) and atretic (bottom) follicles across pseudotime (n = 500 bins). Follicle counts in Supplementary Table 12. i, Age-associated changes in GEP usage by follicle type and estrous stage. Mean difference in GEP usage (columns) between old (n = 7) and young (n = 9) mice within follicle types (rows), in estrus (upper half) and metestrus (lower half). Red = higher usage in old; blue = lower. Significance from linear models (GEP approximate age; Methods), Benjamini–Hochberg FDR-adjusted (*q < 0.05, **q < 0.01, ***q < 0.001). Follicle counts in Supplementary Table 12. j, Mean expression (log-normalized counts, dot color) and proportion of expressing cells (dot size) of marker genes (columns) used to annotate atretic GC subtypes (rows). n = 17,175 spots across 22 mice, all ages; spot counts by GC type in Supplementary Table 13. k, Partition-based graph abstraction (PAGA) of GC state transitions. l, Distribution of normalized pseudotime (DPT; x axis) across three atretic GC clusters. m, Mean GC type composition across follicle stages in young (n = 9) and old (n = 7) mice. Follicle counts are shown in Supplementary Table 14. Spatiotemporal coordination of GCs within the follicular niche. a, Representative spatial plot showing follicular cell types (colors), including preantral, antral early mitotic, antral mitotic, antral luteinizing and atretic GC subtypes, alongside oocyte and theca cell spots. b, Radial distribution of GC subtypes (colored lines, y axis) across follicle types (panels). Lines indicate mean proportion of each subtype across 10 scaled radial bins (x axis; 0, follicle center; 1, edge); shading, s.e.m. across follicles. Significant monotonic radial trends are indicated (preantral follicles: GC—preantral *P = 0.0283, antral mitotic follicles: GC—antral mitotic *P = 0.0479, GC—antral luteinizing *P = 0.0479, antral luteinizing follicles: GC—antral luteinizing *P = 0.0045, atretic follicles: GC—atretic *P < 0.0001, two-sided Spearman correlation). n = 653 follicles from 22 mice, all ages; follicle counts by type in Supplementary Table 6. c, Spatial organization of GC cell cycle states. Left: representative spatial plot with cycling (G2M/S, pink) and non-cycling (G1, gray) GCs. Right: mean radial distribution of cycling vs non-cycling GCs (y axis); shading, s.e.m. across follicles. Significant monotonic trends indicated (cycling: *P < 0.0001, non-cycling: *P < 0.0001, two-sided Spearman correlation; n = 668 follicles). d, Diffusion pseudotime (DPT) across non-atretic and atretic GC trajectories. Left: representative spatial plot with DPT values (colorbar) along non-atretic (purple) and atretic (orange) paths. Right: mean DPT (y axis) across radial bins; shading, s.e.m. across follicles. Significant trends indicated (atretic: *P < 0.0001, non-atretic: *P = 0.0001, two-sided Spearman correlation). n = 653 total follicles from 22 mice, all ages; follicle counts by type in Supplementary Table 7. e, Immunofluorescence of Ki-67 (yellow) and αSMA (red) in a young antral follicle. Scale bar, 100 µm. Representative result across two independent experiments. f, Fluorescence in situ hybridization of Cyp19a1 (yellow) mRNA in a young ovary. Scale bar, 100 µm. Representative result across four independent experiments. g, GC gene programs (GEPs). Left: representative spatial patterns of two distinct GEPs (colorbar), FSH regulation (left) and luteinization (right), with top contributing genes in gray. Right: mean program usage score (y axis) across radial bins (x axis) for preantral (purple), antral mitotic (maroon), antral luteinizing (teal) and atretic (orange) follicles (shading, s.e.m.). n = 653 follicles across 22 mice, all ages; follicle counts by type in Supplementary Table 6. h, Dynamics of GC GEPs across pseudotime (DPT), follicle type and estrous stage in young (n = 9) and old (n = 7) mice. Top: kernel density distribution of follicle types across normalized pseudotime; estrous stage composition below. Middle and bottom: mean normalized GEP usage (color) in non-atretic (middle) and atretic (bottom) follicles across pseudotime (n = 500 bins). Follicle counts in Supplementary Table 12. i, Age-associated changes in GEP usage by follicle type and estrous stage. Mean difference in GEP usage (columns) between old (n = 7) and young (n = 9) mice within follicle types (rows), in estrus (upper half) and metestrus (lower half). Red = higher usage in old; blue = lower. Significance from linear models (GEP approximate age; Methods), Benjamini–Hochberg FDR-adjusted (*q < 0.05, **q < 0.01, ***q < 0.001). Follicle counts in Supplementary Table 12. j, Mean expression (log-normalized counts, dot color) and proportion of expressing cells (dot size) of marker genes (columns) used to annotate atretic GC subtypes (rows). n = 17,175 spots across 22 mice, all ages; spot counts by GC type in Supplementary Table 13. k, Partition-based graph abstraction (PAGA) of GC state transitions. l, Distribution of normalized pseudotime (DPT; x axis) across three atretic GC clusters. m, Mean GC type composition across follicle stages in young (n = 9) and old (n = 7) mice. Follicle counts are shown in Supplementary Table 14. Notably, our analyses suggested a spatially maintained division of labor of GCs within follicles that balances growth and differentiation. Proliferating, immature GCs were enriched near the follicular center (Fig. 3c), whereas spatial projection of differentiation trajectories showed that mature Cyp19a1+ GCs with higher diffusion pseudotime (DPT) localized to the periphery (Fig. 3d and Methods). These spatial patterns were validated by immunostaining and in situ hybridization (Fig. 3e,f). Analysis of GC gene expression programs (GEPs) revealed functional compartmentalization between inner and outer locations: FSH-responsive programs (GEP2; Inhbb, Fst) were enriched centrally, whereas steroidogenic programs (GEP1; Hsd3b1, Lhcgr, Mro) were localized to the periphery (Fig. 3c–g, Extended Data Fig. 3b–f, Supplementary Tables 4 and 7 and Methods), consistent with prior observations that proximity to the theca-granulosa boundary influences GC specialization69. This radial organization was lost in atretic follicles, which exhibited a global reduction in both programs (Fig. 3g and Supplementary Table 6). Altogether, our findings indicate that GCs maintain a spatially organized division of labor that balances proliferation, differentiation and functionalization during follicle development, a pattern that collapses during atresia. Aging disrupts the temporal trajectory of GC differentiation and alters atresia progression Analysis of transcriptional programs across the follicular lifecycle revealed that aging disrupts the coordinated molecular transitions observed across granulosa cell differentiation. In young ovaries, GEP usage across GC trajectories defined by DPT and follicle type follow an ordered progression: GCs in preantral follicles initiate an early development program (GEP9; Wt1, Wnt4), progress through proliferative expansion (GEP3; Top2a, Mki67) in antral, followed by the induction of follicular dominance (GEP2 and GEP4; Inhbb, Fst; Esr2, Greb1l) and luteinization (GEP1; Lhcgr, Hsd3b1) programs heightened at metestrus (Fig. 3h, Extended Data Fig. 3g,h and Supplementary Table 12). Atretic follicles are characterized by apoptosis-associated programs (GEP10 and GEP11; Thbs1, Pik3ip1; Ghr, Ptprd) (Fig. 3h, Extended Data Fig. 3f,g and Supplementary Table 12). In contrast, aged ovaries displayed altered timing and intensity of GEPs, indicating a breakdown in the normal sequence of GC state transitions across follicle development (Fig. 3i and Extended Data Fig. 3h). For example, aged antral follicles downregulate the luteinization program (GEP1) (Fig. 3h,i, Extended Data Fig. 3g,h and Supplementary Table 12). Notably, atretic follicles were the most affected by aging, with decreased canonical apoptosis-associated programs (GEP10 and GEP11) and aberrant upregulation of the ECM remodeling program (GEP6; Col1a2, Lama2), which is typically specific to preantral follicles in young ovaries. These patterns suggest an age-associated shift from regulated to dysregulated remodeling of the follicular niche, resulting in a ‘molecular blurring’ between the boundaries of developmental and degenerative processes. To further investigate age-associated changes in follicular atresia, we next characterized the heterogeneity of atretic GCs in greater detail. First, we identified three distinct atretic GC states, which reflected metabolic reprogramming (state 1; Cdk8, Hexb, Hsd3b1, Tpt1)70–74, oxidative stress response and matrix remodeling (state 2; Clu, Gpx3, Sparc, Col1a2)75–77, and apoptotic and immune-modulatory signatures (state 3; Foxo1, Ghr, Pik3ip1, Cfh, Cd47)78–82 (Fig. 3j, Supplementary Table 13 and Supplementary Note). Trajectory inference using partition-based graph abstraction (PAGA) and pseudotime analysis supported a progressive transition from state 1 to state 3, with the terminal state 3 enriched during metestrus (Fig. 3k–m and Supplementary Tables 13 and 14). Atresia appeared to originate from multiple non-atretic GC populations, particularly from antral early follicles, consistent with the high rates of follicular loss occurring prior to dominant follicle selection83,84. Aging was associated with a reduction in the frequency of terminal atretic state 3 across follicle stages (Fig. 3m and Supplementary Table 14), suggesting impaired progression to terminal atresia. This shift coincided with the upregulation of complement genes and reduced expression of the phagocytosis inhibitory ‘don’t-eat-me’ signal Cd47 (Extended Data Fig. 3i and Supplementary Table 13). Together, these findings suggest that successful atresia requires coordinated transition into an apoptotic, immune-active state for follicular clearance, and that aging impairs this process, potentially promoting unresolved atresia and accumulation of follicular debris. Multicellular dynamics and interactions shaping folliculogenesis progression Folliculogenesis is driven by coordinated interactions across multiple cell types, extending beyond cell-intrinsic granulosa programs. To resolve this higher-order organization (Fig. 4a and Supplementary Table 6), we ordered follicle segments from reproductively young mice along a computationally inferred trajectory of folliculogenesis (Fig. 4b,c and Supplementary Table 6), visualized in a UMAP embedding (Fig. 4d, Extended Data Fig. 4a, Supplementary Tables 6 and 15 and Methods). Diffusion pseudotime recapitulated canonical transitions from preantral to antral or atretic follicles (Fig. 4e), accompanied by progressive shifts in cell type composition (Extended Data Fig. 4b). To link these compositional changes to intercellular signaling, we applied MultiNicheNet85 to infer receptor-ligand interactions among oocytes, granulosa, theca and immune cells (Methods and Extended Data Fig. 4c). This analysis revealed follicle stage-specific communication patterns, including Gdf9-Fxyd6 between oocytes and GCs in preantral follicles, Fst-Bmpr1b between granulosa and theca cells in antral follicles, and Apoa4-Lrp1 within GCs in atretic follicles (Fig. 4f, Supplementary Table 16 and Methods). The atretic-enriched Apoa4-Lrp1 interaction implicated metabolic reprogramming during follicle regression: Lrp1 regulates lipid transport and glucose metabolism86,87, whereas Apoa4 promotes lipid handling and facilitates glucose uptake88. Notably, LRP1 has been genetically associated with polycystic ovarian syndrome, a disorder characterized by metabolic and endocrine dysfunction89. Together, these results indicate that dynamic, stage-specific intercellular signaling coordinates follicle progression.Fig. 4Aging disrupts temporal coupling in folliculogenesis.a, Mean cell-type composition across preantral (n = 238), antral mitotic (n = 191), atretic (n = 181) and antral luteinizing (n = 43) follicles, and CL (n = 234). b, Schematic of optimal transport inference of multicellular transitions, in which follicle pairs are aligned by minimizing their distributional distance. Transcriptionally similar pairings (for example, GC to GC) incur low cost, whereas dissimilar pairings (for example, theca to GC) incur high cost. c, Pairwise transition matrix of segments from young mice (rows, columns) showing transport costs (blue) organized by unsupervised hierarchical clustering (dendrogram, axes) and annotated by segment type (color). d,e, Interpolated folliculogenesis trajectory in young mice (n = 382 follicles from 8 mice), shown as a UMAP embedding of follicles (dots) colored by segment type (d) and diffusion pseudotime (DPT, colorscale; e). f, UMAPs of follicles (dots) showing mean ligand–receptor expression scores (color) for significant interactions identified by MultiNicheNet. Panel titles show ligand (left) and receptor (right) components colored by the expressing cell type. Follicle counts: preantral n = 144, antral n = 120, atretic n = 160. g–l, Follicle characteristics in young (n = 9) (g–i) and old (n = 7) mice (j–l). g,j, Fraction of follicle types per puck (y axis) across estrus and metestrus (x axis) for individual pucks (dots) and group means (bars) (NS, not significant; two-tailed t-test). Puck counts: young estrus n = 23 in, young metestrus n = 6, old estrus n = 12, old metestrus n = 6. h,k, Boxplots of follicle DPT (y axis) by estrous stage (x axis, bar shading) (**P = 0.002, NS, not significant; two-tailed t-test). Center lines, median; box bounds, first and third quartiles; whiskers, 1.5× interquartile range. Follicle counts: young estrus n = 281, young metestrus n = 66, old estrus n = 58, old metestrus n = 23. i,l, Boxplot of expression of hormone receptor genes (x axis) across follicles types (color) and stage (bar shading); boxplots as in h,k. Follicle—antral mitotic: Amhr2 P = 0.0126, Fshr P = 0.0003, Lhcgr P = 0.0002; Follicle—antral luteinizing: Fshr P = 0.0062, Lhcgr P = 0.0002; Follicle—atretic: Amrh2 P = 0.0348, Fshr P = 0.0003, Lhcgr P < 0.0001, two-tailed t-test; all others were not significant. Follicle counts are shown in Supplementary Table 12. Aging disrupts temporal coupling in folliculogenesis. a, Mean cell-type composition across preantral (n = 238), antral mitotic (n = 191), atretic (n = 181) and antral luteinizing (n = 43) follicles, and CL (n = 234). b, Schematic of optimal transport inference of multicellular transitions, in which follicle pairs are aligned by minimizing their distributional distance. Transcriptionally similar pairings (for example, GC to GC) incur low cost, whereas dissimilar pairings (for example, theca to GC) incur high cost. c, Pairwise transition matrix of segments from young mice (rows, columns) showing transport costs (blue) organized by unsupervised hierarchical clustering (dendrogram, axes) and annotated by segment type (color). d,e, Interpolated folliculogenesis trajectory in young mice (n = 382 follicles from 8 mice), shown as a UMAP embedding of follicles (dots) colored by segment type (d) and diffusion pseudotime (DPT, colorscale; e). f, UMAPs of follicles (dots) showing mean ligand–receptor expression scores (color) for significant interactions identified by MultiNicheNet. Panel titles show ligand (left) and receptor (right) components colored by the expressing cell type. Follicle counts: preantral n = 144, antral n = 120, atretic n = 160. g–l, Follicle characteristics in young (n = 9) (g–i) and old (n = 7) mice (j–l). g,j, Fraction of follicle types per puck (y axis) across estrus and metestrus (x axis) for individual pucks (dots) and group means (bars) (NS, not significant; two-tailed t-test). Puck counts: young estrus n = 23 in, young metestrus n = 6, old estrus n = 12, old metestrus n = 6. h,k, Boxplots of follicle DPT (y axis) by estrous stage (x axis, bar shading) (**P = 0.002, NS, not significant; two-tailed t-test). Center lines, median; box bounds, first and third quartiles; whiskers, 1.5× interquartile range. Follicle counts: young estrus n = 281, young metestrus n = 66, old estrus n = 58, old metestrus n = 23. i,l, Boxplot of expression of hormone receptor genes (x axis) across follicles types (color) and stage (bar shading); boxplots as in h,k. Follicle—antral mitotic: Amhr2 P = 0.0126, Fshr P = 0.0003, Lhcgr P = 0.0002; Follicle—antral luteinizing: Fshr P = 0.0062, Lhcgr P = 0.0002; Follicle—atretic: Amrh2 P = 0.0348, Fshr P = 0.0003, Lhcgr P < 0.0001, two-tailed t-test; all others were not significant. Follicle counts are shown in Supplementary Table 12. Aging disrupts temporal coupling in follicle activation, maturation and hormone sensing A hallmark of ovarian aging is the onset of irregular cycling, indicating impairments in the temporal regulation of endocrine signals and ovarian function90,91, as observed during perimenopause in humans. Although the aged mice in our study were confirmed to be cycling, their molecular profiles indicated a loss of temporal coordination between the estrous cycle and ovarian cellular states observed in young ovaries. In young mice, follicle states were tightly coupled to their cycle stage, with the following characteristics at estrus compared to metestrus: preantral follicles were more abundant (Fig. 4g; P < 0.0001, two-tailed t-test; Supplementary Table 17), antral mitotic follicles were smaller (Extended Data Fig. 4d; P = 0.0011, two-tailed t-test; Supplementary Table 12), and follicles exhibited earlier pseudotime (DPT), indicating a less mature distribution of follicle states (Fig. 4h, Extended Data Fig. 4e and Supplementary Table 18; P = 0.002, two-tailed t-test). Correspondingly, antral follicles at estrus showed reduced expression of gonadotropin receptors Lhcgr and Fshr (Fig. 4i; *P < 0.05, **P < 0.01, ***P < 0.001, two-tailed t-test; Supplementary Table 12), suggesting an immature and limited hormone responsiveness, likely reflecting their recent transition from the gonadotropin-independent preantral state11. In contrast, aged mice lacked these stage-dependent patterns across estrus and metestrus: preantral follicle frequencies and antral follicle size did not vary (Fig. 4j; P = 0.942, two-tailed t-test; Extended Data Fig. 4f and Supplementary Tables 12 and 17), and projecting aged follicles onto the young reference trajectory revealed no separation in DPT (Fig. 4k; P = 0.677, two-tailed t-test; Extended Data Fig. 4e,g; Supplementary Table 18; and Methods). Consistently, Fshr and Lhcgr expression in antral follicles did not differ between estrus and metestrus (Fig. 4l and Supplementary Table 12). In summary, we conclude that in young ovaries, gonadotropins synchronize follicle cohorts to coordinate progression towards ovulation or atresia, yielding immature antral follicles at estrus. By contrast, in aged ovaries, follicles at estrus are larger and more mature, consistent with disrupted developmental timing. Although reduced follicle numbers in old ovaries may limit statistical power, the consistent loss of stage-dependent differences in follicle composition, pseudotime and hormone receptor expression indicates that temporal regulation of folliculogenesis becomes uncoupled from the estrous cycle with age, preceding acyclicity and reproductive senescence. Dysregulation of CL clearance with aging may initiate inflammatory remodeling We next examined whether the dynamics of the post-ovulatory luteal cells are also altered with age. Identification of two spatially segregated luteal cell populations with opposing progesterone regulatory programs suggested a temporal phenotypic switch during the luteal lifecycle. Parm1-enriched CLs (‘CL – Parm1’) express progesterone synthesis genes, contain vascular endothelial cells, are enriched at estrus and consist of histologically smaller cells92,93, features consistent with newly formed, hormonally active CLs14,32,94–96. In contrast, Akr1c18-enriched CLs (‘CL – Akr1c18’) express the luteal regression marker Ptgfr, are infiltrated by immune cells, are more frequent at metestrus and contain histologically larger cells92,93, features consistent with luteolysis and regression (Fig. 5a–c, Extended Data Fig. 5a–f and Supplementary Tables 4, 6 and 17)96. In aged ovaries, although these transcriptional distinctions between luteal subtypes were preserved (Extended Data Fig. 5g), Akr1c18 CLs were significantly accumulated at estrus compared to young ovaries (Fig. 5b, P = 0.0055, two-tailed t-test; Supplementary Table 17), suggesting failed timely clearance. Because timely CL regression and the accompanying drop in progesterone are required to initiate the next cycle14,95,97, persistence of these degenerating CLs likely disrupts progesterone homeostasis, potentially impairing cycle progression1,11.Fig. 5Aging disrupts CL clearance.a, Fluorescence in situ hybridization of Parm1 (yellow) and Akr1c18 (teal) in a young ovary. Scale bar, 100 µm. Representative result across four replicates. b, Fraction of each CL type per puck (y axis) during estrus across age groups (x axis) for individual pucks (dots) and mean across pucks (bars). Error bars represent standard deviation across pucks (young n = 23, old n = 12), and significance was determined by two-tailed t-test. c, Cell type composition of CL subtypes. Boxplots show spot counts (y axis, log10 scale) for luteal (all types), immune and vascular cells (color) across CL—Parm1 and CL—Akr1c18 segments (two-tailed t-test). Center lines, median; box bounds, first and third quartiles; whiskers, 1.5× interquartile range. n = 234 CLs from 22 mice, all ages; CL counts in Supplementary Table 6. d, Representative spatial plots across estrous stages and age showing luteal (blues) and immune (yellow) cells; all other cell types, gray. e, Expression of luteolysis-associated receptor genes (x axis) in Akr1c18 CLs across young (n = 9), middle-aged (n = 6) and old (n = 7) mice (color), stratified by estrous stage (shading). Dots represent individual corpora lutea; bars show means; error bars show s.e.m. (young: Prlr ***P < 0.0001, Ptgfr ***P = 0.0007, middle: Prlr ***P = 0.0003, two-tailed t-test; all others are nonsignificant). Akr1c18 CL counts in Supplementary Table 12. f, Mean expression and fraction expressing of CL—Akr1c18 segments (dot size) for select genes (x axis) involved in CL clearance and immune cell recruitment, across young (n = 9), middle-aged (n = 6) and old (n = 7) mice (rows). Counts in Supplementary Table 15. g, Age-associated immune remodeling. Outer ring: fraction of immune (yellow) versus non-immune (gray) spots in estrus (white) and metestrus (black) stages. Inner ring: distribution of immune spots across spatial compartments in young (n = 31), middle (n = 16) and old (n = 21) pucks. Aging disrupts CL clearance. a, Fluorescence in situ hybridization of Parm1 (yellow) and Akr1c18 (teal) in a young ovary. Scale bar, 100 µm. Representative result across four replicates. b, Fraction of each CL type per puck (y axis) during estrus across age groups (x axis) for individual pucks (dots) and mean across pucks (bars). Error bars represent standard deviation across pucks (young n = 23, old n = 12), and significance was determined by two-tailed t-test. c, Cell type composition of CL subtypes. Boxplots show spot counts (y axis, log10 scale) for luteal (all types), immune and vascular cells (color) across CL—Parm1 and CL—Akr1c18 segments (two-tailed t-test). Center lines, median; box bounds, first and third quartiles; whiskers, 1.5× interquartile range. n = 234 CLs from 22 mice, all ages; CL counts in Supplementary Table 6. d, Representative spatial plots across estrous stages and age showing luteal (blues) and immune (yellow) cells; all other cell types, gray. e, Expression of luteolysis-associated receptor genes (x axis) in Akr1c18 CLs across young (n = 9), middle-aged (n = 6) and old (n = 7) mice (color), stratified by estrous stage (shading). Dots represent individual corpora lutea; bars show means; error bars show s.e.m. (young: Prlr ***P < 0.0001, Ptgfr ***P = 0.0007, middle: Prlr ***P = 0.0003, two-tailed t-test; all others are nonsignificant). Akr1c18 CL counts in Supplementary Table 12. f, Mean expression and fraction expressing of CL—Akr1c18 segments (dot size) for select genes (x axis) involved in CL clearance and immune cell recruitment, across young (n = 9), middle-aged (n = 6) and old (n = 7) mice (rows). Counts in Supplementary Table 15. g, Age-associated immune remodeling. Outer ring: fraction of immune (yellow) versus non-immune (gray) spots in estrus (white) and metestrus (black) stages. Inner ring: distribution of immune spots across spatial compartments in young (n = 31), middle (n = 16) and old (n = 21) pucks. Persistent CLs in aged ovaries exhibited dysregulation of luteolytic and inflammatory signaling, underscoring their pathologic role in ovarian remodeling. Under homeostatic conditions of young ovaries, immune cells accumulate within regressing Akr1c18 CLs at estrus to facilitate tissue clearance9 (Fig. 5c, Extended Data Fig. 5h and Supplementary Tables 6 and 12). However, in aged ovaries, immune infiltration within this compartment instead shifted to metestrus (Fig. 5d and Extended Data Fig. 5h; P < 0.005, two-tailed t-test; Supplementary Table 12), suggesting mistimed or sustained inflammatory recruitment. Aged Akr1c18 CLs also failed to induce key prolactin- and prostaglandin-responsive luteolytic genes at metestrus (Prlr, Ptgfr; Fig. 5e and Supplementary Table 12) while instead upregulating inflammatory chemokines Ccl2, Il1b and Ccl328,33,34 (Fig. 5f, Extended Data Fig. 5i and Supplementary Tables 12 and 15). These observations suggest that defective regression of Akr1c18 CLs may be a focal contributor to ovarian inflammaging. Globally, across all ovarian compartments, age-associated immune expansion was most pronounced within CLs (Fig. 5g; Supplementary Table 5). Altogether, our findings suggest that aging disrupts timely CL clearance, generating persistent inflammatory niches that contribute to a malfunctioning endocrine clock and may initiate broader pathologic immune remodeling in the aging ovary. Spatial immune remodeling reveals multiple drivers of ovarian inflammaging To define how aging reshapes global immune organization7,28, we characterized the composition and spatial architecture of immune populations across the ovary. First, we identified key immune subpopulations, including four macrophage subsets (phagocytic, antigen-presenting, efferocytic and metabolically active), dendritic cells, mast cells, plasma cells and T cells, each occupying distinct spatial niches (Fig. 6a,b, Supplementary Tables 6, 17 and 19 and Supplementary Note). Phagocytic and antigen-presenting macrophages (macrophages 1 and 2) were the dominant populations and expanded preferentially in preantral and atretic follicles, and in CLs, with age (Fig. 6b,c and Supplementary Tables 15 and 17). Across all ages, phagocytic macrophages (macrophage 1) localized to the periphery of follicles and CLs, whereas antigen-presenting macrophages (macrophage 2) concentrated toward follicular and CL centers, a compartmentalization accentuated in aged ovaries (Fig. 6d,e and Supplementary Table 15). Immunofluorescence of regressing CLs confirmed CD11b+ macrophage and CD3+ T cell accumulation in these central niches, suggesting their cooperative role in antigen presentation during luteal clearance (Fig. 6f). Although low in abundance, the expansion of T cells in aged CLs (Fig. 6c; P = 0.0079, two-tailed t-test) may represent the anatomical location underlying prior reports of cytotoxic and double negative T cells accumulation in the aging mouse ovary34. Altogether, these data reveal that ovarian aging involves coordinated remodeling of multiple immune niches, pointing to spatially coexisting immune programs that may jointly drive ovarian inflammaging.Fig. 6Age-associated changes in immune composition.a, Mean expression (log-normalized counts, color) and proportion of expressing spots (size) for marker genes (columns) defining immune cell types (rows); spot counts in Supplementary Table 19. b, Spatial and temporal distribution of immune cell types. Left: percentage of immune cell types across spatial structures. Right: percentage across stage and age. Significant age-associated compositional changes identified by scCODA (*FDR < 0.05). Follicle-type and puck counts in Supplementary Tables 6 and 17. c, Immune populations significantly altered with age across spatial structures, shown for individual segments (dots) and group means (bars). Error bars, s.e.m. across young (n = 9), middle-aged (n = 6) and old (n = 7) mice (macrophage 1: follicle—preantral *P = 0.0291, follicle—atretic *P = 0.0291; mast cell: follicle—tretic *P = 0.0375, other **P = 0.0079; plasma cell: follicle—antral mitotic *P = 0.0409, follicle—atretic **P = 0.0040, other ***P = 0.0003; T cell: CL—late **P = 0.0079, two-tailed t-test; all others, nonsignificant). Follicle type by age in Supplementary Table 15. d, Radial distribution of macrophage subtypes (panels) across segment types (colored lines, y axis) in young (n = 9) and old (n = 7) mice across normalized radial bins (x axis; 0, center; 1, edge); shading, s.e.m. Segment frequencies in Supplementary Table 15. e, Representative spatial plots of immune cell distributions in young and old animals. f, Immunofluorescence of immune cells in a regressing CL (dashed line) from a young ovary. Scale bar, 100 µm. Representative result across two independent experiments. Age-associated changes in immune composition. a, Mean expression (log-normalized counts, color) and proportion of expressing spots (size) for marker genes (columns) defining immune cell types (rows); spot counts in Supplementary Table 19. b, Spatial and temporal distribution of immune cell types. Left: percentage of immune cell types across spatial structures. Right: percentage across stage and age. Significant age-associated compositional changes identified by scCODA (*FDR < 0.05). Follicle-type and puck counts in Supplementary Tables 6 and 17. c, Immune populations significantly altered with age across spatial structures, shown for individual segments (dots) and group means (bars). Error bars, s.e.m. across young (n = 9), middle-aged (n = 6) and old (n = 7) mice (macrophage 1: follicle—preantral *P = 0.0291, follicle—atretic *P = 0.0291; mast cell: follicle—tretic *P = 0.0375, other **P = 0.0079; plasma cell: follicle—antral mitotic *P = 0.0409, follicle—atretic **P = 0.0040, other ***P = 0.0003; T cell: CL—late **P = 0.0079, two-tailed t-test; all others, nonsignificant). Follicle type by age in Supplementary Table 15. d, Radial distribution of macrophage subtypes (panels) across segment types (colored lines, y axis) in young (n = 9) and old (n = 7) mice across normalized radial bins (x axis; 0, center; 1, edge); shading, s.e.m. Segment frequencies in Supplementary Table 15. e, Representative spatial plots of immune cell distributions in young and old animals. f, Immunofluorescence of immune cells in a regressing CL (dashed line) from a young ovary. Scale bar, 100 µm. Representative result across two independent experiments. MNGCs define a specialized macrophage fusion pathology in aging ovaries Among age-expanded immune niches, multinucleated giant cells (MNGCs) emerged as a distinct macrophage-associated pathology. Accumulation of MNGCs is a hallmark of ovarian aging associated with chronic inflammation and fibrosis8,28,98,99, visible as fused cellular clusters on H&E sections (Fig. 7a,b and Extended Data Fig. 6a). Registering adjacent H&E sections to Slide-seq pucks using anatomical landmarks enabled comparing immune spots located within versus outside MNGCs (Fig. 7c, Extended Data Fig. 6b and Methods), which revealed an enrichment of phagocytic macrophages within MNGCs (Fig. 7d). Across macrophage subpopulations, those within MNGCs adopted specialized transcriptional programs distinct from the surrounding stroma, including complement- and fusion-associated genes (Tagln, Acta2, Itgae) in antigen-presenting macrophages, clearance programs in efferocytic macrophages (Gpnmb, Mfge8, Ctsd) and inflammatory-lipid remodeling signatures in metabolically active macrophages (Pla2g7, Lyz2, Lgals3) (Fig. 7e, Extended Data Fig. 6c and Supplementary Table 20). Together, these findings suggest that MNGCs represent specialized inflammatory fusion niches underlying defective tissue clearance, potentially linking immune dysfunction to fibrotic pathology.Fig. 7Global changes in tissue-wide remodeling with aging.a, Representative H&E section showing multinucleated giant cells (MNGCs) in aged ovaries, with manually annotated MNGC masks. b, Distribution of histology sections (y axis) classified as MNGC-high or MNGC-low across young (n = 31), middle (n = 16) and old (n = 21) pucks (x axis). c, Representative registration of MNGC masks from adjacent H&E to Slide-seq using Napari; spots are colored by annotated cell type (colors). d, Mean proportions of immune cell subtypes within MNGC regions in old (n = 21) pucks. e, log2 fold changes of representative genes across macrophage subsets (rows) comparing stromal macrophages inside versus outside MNGCs in aged ovaries. Spot counts in Supplementary Table 20. f, Age-associated compositional changes in stromal populations. Left: proportions (y axis) of select cell types (x axis) across pucks (dots) and group means (bars); significant changes identified by scCODA (Extended Data Fig. 6d; FDR = 0.05). Right: mean proportions of remaining cell types (y axis) across ages (x axis); puck counts: young n = 32, middle n = 16, old n = 21. g, Mean expression (log-normalized, variance-scaled) of ECM- and inflammation-associated genes (columns) across stromal subtypes in young (n = 9), middle-aged (n = 6) and old (n = 7) mice (rows). Genes identified by differential expression analysis between young and old stromal spots; spot counts in Supplementary Table 21. h, Spatial neighborhood analysis of homotypic follicle co-localization. z-scores of co-localization (y axis) shown as a function of spatial radius (x axis; 200 µm steps) relative to a null model (z-score 0, black dashed line) for preantral (top), antral (middle), atretic (bottom) follicles in young (n = 9) and old (n = 7) mice (colors); follicle counts by age in Supplementary Table 16. i, Model summarizing hallmarks of ovarian aging. Aging impacts key processes underlying tissue-wide homeostasis, impacting both the reproductive and endocrine functions of the ovary. Young ovaries maintain temporal regulation of folliculogenesis, spatial domain integrity, immune homeostasis and coordinated remodeling, whereas aged ovaries exhibit disrupted timing, impaired atresia and CL clearance, tissue disorganization, follicle loss, MNGC accumulation, inflammation, fibrosis and epithelial thickening. Global changes in tissue-wide remodeling with aging. a, Representative H&E section showing multinucleated giant cells (MNGCs) in aged ovaries, with manually annotated MNGC masks. b, Distribution of histology sections (y axis) classified as MNGC-high or MNGC-low across young (n = 31), middle (n = 16) and old (n = 21) pucks (x axis). c, Representative registration of MNGC masks from adjacent H&E to Slide-seq using Napari; spots are colored by annotated cell type (colors). d, Mean proportions of immune cell subtypes within MNGC regions in old (n = 21) pucks. e, log2 fold changes of representative genes across macrophage subsets (rows) comparing stromal macrophages inside versus outside MNGCs in aged ovaries. Spot counts in Supplementary Table 20. f, Age-associated compositional changes in stromal populations. Left: proportions (y axis) of select cell types (x axis) across pucks (dots) and group means (bars); significant changes identified by scCODA (Extended Data Fig. 6d; FDR = 0.05). Right: mean proportions of remaining cell types (y axis) across ages (x axis); puck counts: young n = 32, middle n = 16, old n = 21. g, Mean expression (log-normalized, variance-scaled) of ECM- and inflammation-associated genes (columns) across stromal subtypes in young (n = 9), middle-aged (n = 6) and old (n = 7) mice (rows). Genes identified by differential expression analysis between young and old stromal spots; spot counts in Supplementary Table 21. h, Spatial neighborhood analysis of homotypic follicle co-localization. z-scores of co-localization (y axis) shown as a function of spatial radius (x axis; 200 µm steps) relative to a null model (z-score 0, black dashed line) for preantral (top), antral (middle), atretic (bottom) follicles in young (n = 9) and old (n = 7) mice (colors); follicle counts by age in Supplementary Table 16. i, Model summarizing hallmarks of ovarian aging. Aging impacts key processes underlying tissue-wide homeostasis, impacting both the reproductive and endocrine functions of the ovary. Young ovaries maintain temporal regulation of folliculogenesis, spatial domain integrity, immune homeostasis and coordinated remodeling, whereas aged ovaries exhibit disrupted timing, impaired atresia and CL clearance, tissue disorganization, follicle loss, MNGC accumulation, inflammation, fibrosis and epithelial thickening. Aging disrupts interfollicular spatial coupling Age-associated immune remodeling converged with broader stromal disorganization, suggesting organ-level consequences of inflammaging. Across stromal populations residing outside of follicles and CL, we observed a progressive expansion of surface epithelial and immune populations, accompanied by depletion of steroidogenic stromal cells7,100–103 (Fig. 7f, compositional test with scCODA104; Extended Data Fig. 6d; Supplementary Table 5; and Methods). Thickening of the surface epithelium was independently confirmed by morphometric measurements on adjacent H&E sections (Extended Data Fig. 6e and Methods). Cell type compositional shifts coincided with transcriptional changes marked by reduced expression of ECM genes and increased metalloproteases that may indicate aberrant ECM remodeling, as well as elevated fibrotic and inflammatory signatures100 (Fig. 7g, Extended Data Fig. 6f and Supplementary Tables 21 and 22). We next asked whether stromal remodeling alters the spatial coordination of follicle development with aging. To test this, we quantified the spatial co-localization among homotypic follicles, that is follicles in the same developmental stage. At short spatial scales (<500 µm), both preantral and atretic follicles were highly co-localized in young but not aged ovaries (Fig. 7h, Methods and Supplementary Table 16). This is consistent with prior evidence that spatial coupling among preantral follicles promotes their activation and early development105,106, and that loss of preantral clustering may, in part, reflect increased ECM rigidity6,8,100. Co-localization of atretic follicles in young ovaries further suggests spatially coordinated paracrine signaling during follicle selection, potentially shaped by signals emanating from dominant follicles. In contrast, antral follicles showed little short-range in either young or old mice, consistent with their relative spatial isolation as isolated ‘islands’ during dominant follicle selection. Weak co-localization at larger scales (>2,000 µm) likely reflects global ovarian size in young mice (~3,000 µm in diameter) rather than local organization. Together, these results suggest that aging disrupts follicular crowding and spatial coordination, potentially impairing interfollicular communication. Collectively, these findings support a model in which immune remodeling, ECM stiffening and loss of follicular spatial organization represent interconnected layers of ovarian aging, spanning cell-level dysfunction to niche disruption and tissue-level disorganization (Fig. 7i). Discussion Here we present a near-cellular-resolution spatial transcriptomics atlas of the naturally cycling murine ovary across reproductive ages and show that ovarian aging is characterized by the progressive erosion of multiscale spatiotemporal tissue organization. Across cellular, niche and tissue scales, aging disrupted spatial patterning within follicles, temporal coordination of folliculogenesis and CL turnover, and immune–stromal homeostasis, with changes emerging before cessation of cycling. These findings indicate that reproductive aging can be viewed as a progressive breakdown of tissue-level coordination, rather than solely the depletion of the follicle reserve. Our findings suggest that the spatial organization of follicles is an elegant mechanism for survival and resource allocation. By compartmentalizing proliferative activity to inner GCs and hormone-sensing, steroidogenic functions to outer GCs positioned near the vasculature and stroma, follicles optimize resource allocation under the energetic constraints of this competitive environment. We hypothesize that successful spatial compartmentalization of GC function during folliculogenesis could contribute to follicular fate, with dominant follicles effectively maintaining functional polarity, whereas less efficient competitors undergo atresia107,108. More broadly, our findings refine a view of follicular atresia as a heterogeneous process linked to metabolic, immune and tissue remodeling programs, rather than a uniform degenerative endpoint. Although our results suggest a plausible sequence of atresia commitment arising from multiple GC lineages, the specific molecular drivers remain to be experimentally validated109–112. In aged ovaries, the blurring of developmental and degenerative boundaries points to compromised follicle viability. Whether atretic fate is caused primarily by oocyte-intrinsic cues or by a dysregulated follicular environment remains unclear. Successful reproductive cycling depends on precise temporal coordination between local tissue dynamics and systemic endocrine signaling. In young mice, we find that tissue-wide remodeling processes are tightly coupled to the estrous cycle, ensuring synchronized progression of follicular dynamics, atresia and luteal regression97,113–117. With age, however, this synchronization deteriorates: follicles lose stage- and follicle-type-restricted regulation of developmental and atretic programs, and CLs persist across stages. These features parallel luteal-phase alterations observed during perimenopause and support a model in which reproductive aging manifests as a progressive uncoupling of local tissue dynamics from systemic hormonal rhythms. The rapid dynamics of the murine estrous cycle, spanning only 4 to 5 days, further highlight this phenomenon, allowing spatial transcriptomics to capture transient yet meaningful transitions that span heterogeneity across follicle populations. Together, these observations underscore temporal coordination as a central organizing principle of ovarian function that becomes progressively disrupted with age. These findings have implications for understanding and mitigating human reproductive aging, which remains difficult to study at scale with high temporal resolution. The observation that spatial and temporal coordination breakdown prior to cycle cessation suggests potential opportunities for earlier detection and intervention in age-related fertility decline. Additionally, the importance of tissue-wide organization in maintaining ovarian function implies that effective therapeutic strategies may need to target the entire ovarian environment, rather than focusing exclusively on oocytes or individual follicles. Our results also highlight several open questions. How spatial compartmentalization of GC function is established and maintained in a highly proliferative context, whether improved division of labor in the follicular environment would alter atresia, how neighboring follicles coordinate developmental states and how degenerating follicular and luteal cells are efficiently cleared from the tissue remain to be determined. Addressing these questions will require perturbation-based approaches coupled with methods to track follicular fate. Most of all, whether interventions targeting cellular turnover or inflammatory pathways preserve ovarian function with age remains to be seen. More broadly, defining how cellular turnover, immune activity and tissue remodeling are coordinated may reveal general principles of tissue maintenance that extend beyond the ovary to aging and regenerative systems. We acknowledge several limitations in our study. Although spatial transcriptomics provides high-resolution insight into tissue organization, it does not capture lineage relationships or fate decisions, limiting our ability to define the precise point of atresia commitment, identify follicles that successfully ovulate or determine the age of a CL. Future studies integrating lineage tracing, functional perturbations and hormone measurements across the full estrous cycle will be critical to resolve these dynamics and their changes with age. Additionally, validation in human tissue will be necessary for translating these findings into clinical applications. Together, our findings suggest that ovarian aging reflects progressive failure of a coordinated spatiotemporal architecture across follicle development, endocrine signaling and tissue remodeling. In this model, local defects in cellular organization and tissue clearance propagate across niches, producing organ-level disorganization. More broadly, these results frame aging as a breakdown of coordinated tissue dynamics, providing a conceptual basis for understanding reproductive decline and opening new avenues for preserving fertility and ovarian endocrine function. Methods Mice All procedures involving animals were approved by and conducted in accordance with Institutional Animal Care and Use Committee guidelines at UCSF (protocol AN194707-01N under Zachary Knight), the Broad Institute (protocol 0296-02-21 under H.C. and F.C.) and Yale University (protocol 2024-20567 under H.C.). Female C57BL/6 J (Jax 000664) mice were bred (UCSF) or purchased (Broad, Yale) and maintained under specific-pathogen-free conditions on a 12-hour light-dark cycle at ambient temperature 21.5 + 1°C and relative humidity between 30% and 70%, with standard chow (20% protein diet and 4.5% fat) and water provided ad libitum. Broad mice were purchased at ages 10, 36 and 50 weeks and acclimated to the vivarium for 2 days (‘Broad cohort 1’) or 2 weeks (‘Broad cohort 2’) upon arrival before staging. Yale mice were purchased at ages 8, 36 and 42 weeks and acclimated to the vivarium for 14 days upon arrival before staging. The numbers of mice per age and stage used to generate data for all experiments are shown in Fig. 1a. Estrous cycle staging mice by vaginal cytology Buck mice Mice collected at UCSF were staged near-daily for 8 consecutive days using vaginal cytology, as previously described118,119. A total volume of 100 µl sterile saline (in 20–50 µl increments) was expelled five to ten times with a P200 pipette into the vaginal canal, taking care to avoid touching the vaginal opening with the pipette tip then expelled onto a glass slide and imaged immediately using Nikon Eclipse Ts2 (10× objective) and a Nikon DS Qi2 camera. Broad and Yale mice Female mice were staged near-daily for 8 consecutive days using vaginal cytology, as previously described118,119. Briefly, mini-tip flocked nasopharyngeal swabs (VWR SLIN501CS01) were moistened with sterile PBS, and approximately one-fourth of the flocked tip was inserted into the vaginal canal of manually restrained mice. Swabs were rotated five times clockwise and five times counterclockwise and then smeared onto a non-treated glass slide (VWR 16004-422). Slides were air-dried and stained with a 0.1% crystal violet solution (Jorgenson J0322) for 1–2 min. Excess stain was rinsed with PBS and sterile water, and the slides were counterstained with 1% eosin (Jorgenson #J0322A) for 10–20 s. Following the counterstaining step, slides were rinsed again, air-dried completely, examined under a light microscope and imaged with a Leica Aperio VERSA Brightfield, Fluorescence & FISH Digital Pathology Scanner, using the 20× objective. Cytological evaluation identified the predominant cell types (epithelial cells, cornified cells and leukocytes), thereby enabling classification of the estrous cycle stages (proestrus, estrus, metestrus and diestrus) as previously described118,119. In brief, proestrus, estrus and diestrus are characterized by a predominance of nucleated epithelial cells, cornified epithelial cells or leukocytes, respectively, whereas metestrus is identified by a mixture of cornified epithelial cells and leukocytes120. The time course of the estrous cycle is in Supplementary Table 1. Tissue collection Mice identified at the appropriate estrous stage were anesthetized with 3% isoflurane for 1 min and maintained at 1% isoflurane via nose cone. Adequate anesthesia was confirmed by absence of pedal reflex. Mice were transcardially perfused with PBS through the left ventricle to remove blood. Whole ovaries were rapidly dissected, embedded in Optimal Cutting Temperature (OCT) compound in cryomolds and flash-frozen on dry ice. Tissues were either processed immediately or stored at −80 °C. Hematoxylin and eosin stain of adjacent histology sections Fresh-frozen tissues were sectioned at 10 µm and stained using a Leica ST5010 Autostainer XL (Leica Biosystems). Sections were immersed in xylene, rehydrated through graded ethanol (100% and 95%) and stained with hematoxylin (Leica Biosystems, 3801571). Nuclei were differentiated with mild acid (Leica Biosystems, 3803598), “blued” in a weakly alkaline solution (Leica Biosystems, 3802918), washed in water and counterstained with eosin (Leica Biosystems, 3801616). Sections were then dehydrated through graded ethanol (100% and 95%), cleared in xylene, dehydrated and coverslipped using a Leica CV5030 automated glass coverslipper. Brightfield images were acquired on a Leica Aperio VERSA Brightfield, Fluorescence & FISH Digital Pathology Scanner, using a 20x objective. RNA-FISH via HCR RNA in situ hybridization was performed on 10 µm fresh-frozen cryosections using the hybridization chain reaction (HCR) v3.0 HCR-Gold RNA-FISH kit (Molecular Instruments) according to the manufacturer’s instructions. Probes targeting mouse transcripts were obtained from Molecular Instruments. Sections were fixed with ice-cold 4% paraformaldehyde (PFA) in PBS for 15 min at 4 °C, dehydrated through graded EtOH (50%, 70%, 100%) and washed in PBS. Samples were pre-hybridized in 50 µl pre-warmed HCR hybridization buffer for 10 min at 37 °C in a humidified chamber, followed by overnight hybridization at 37 °C in a humidified chamber with probe solution (2 µl of each probe in 100 µl HCR Hybridization Buffer). Sections were washed four times with pre-warmed HCR probe wash buffer (15 min each, 37 °C). For signal amplification, sections were incubated in 50 µl HCR Gold Amplifier Buffer for 30 min at room temperature (RT) and then incubated overnight at RT in dark with snap-cooled HCR hairpins h1 and h2 (2 µl each in 100 µl HCR Gold Amplifier Buffer). Sections were washed four times with HCR Gold Amplifier Wash Buffer (15 min each), counterstained with DAPI (1 µg ml−1 in PBS; Miltenyi, 130-111-570), washed with PBS and treated with TrueBlack (1× in 70% ethanol; Biotium, 23007) for 1 min at RT to quench autofluorescence, followed by three PBS washes. Samples were mounted in antifade (Thermo Fisher, P10144) and imaged on an Andor Dragonfly 505 spinning disk confocal. Images were analyzed using ImageJ. Multiplexed immunofluorescence of immune markers Fresh-frozen tissues were embedded in OCT compound, cryosectioned at 10 μm, mounted on SuperFrost microscope slides (VWR, 48311-703) and stored at −80 °C until staining. Slides were fixed in 4% paraformaldehyde (PFA) for 10 min and then transferred to the Molecular Cytology Core Facility at Memorial Sloan Kettering Cancer Center. Slides were blocked for 20 min in PBS containing 0.2% Triton X-100, 5% normal goat serum (NGS) and 5% bovine serum albumin (BSA). Iterative immunofluorescence was performed on the Lunaphore COMET platform over 11 cycles using Alexa Fluor 555- and 647-conjugated secondary antibodies. Primary antibodies: CD11b (Abcam ab182981, 0.125 µg ml−1, cycle 3), Ki-67 (Invitrogen 14-5698-82, 1:2,000 dilution, cycle 5) and CD3 (Roche N03789, 1:2 dilution, cycle 9). Following COMET imaging, slides underwent antigen retrieval (pH 9 buffer, 15 min) on a Leica automated stainer and were subsequently stained with CD45 (Proteintech 60287-1-Ig) using an Alexa Fluor 594-conjugated secondary antibody. Images acquired across platforms were aligned for analysis. Single-nucleus RNA sequencing Nuclei were isolated from three frozen ovarian tissues, one from each age group collected at estrus, using EZ Lysis buffer supplemented with recombinant RNase inhibitor (1 μl ml−1), following previous protocols121. Tissues were transferred to a pre-chilled and cleaned dounce homogenizers containing 2 ml EZ Lysis buffer, homogenized with 20 strokes each of pestle A and pestle B, then brought to 5 ml total volume, incubated on ice for 5 min and centrifuged at 500 g for 5 min at 4 °C. Pellets were washed once in 5 ml EZ Lysis buffer, then resuspended in 200 μl PBS containing RNase inhibitor and filtered through a 40-μm strainer. Single-nucleus libraries were generated using the 10X Genomics Chromium Single Cell 3’ v4 (GEM-X) platform, loading one sample per channel with 20,000 nuclei each. Reverse transcription, cDNA amplification and library construction were performed according to the manufacturer’s protocol. In situ transcriptome processing via Slide-seq Tissue sections were processed and sequenced using Slide-seq v2 as previously described36. Slide-seq arrays were purchased from Curio Biosciences (Curio Seeker) or prepared in-house using custom-synthesized barcoded beads functionalized with the sequence25 (5′TTT_PC_GCCGGTAATACGACTCACTATAGGGCTACACGACGCTCTTCCGATCTJJJJJJJJTCTTCAGCGTTCCCGAGAJJJJJJNNNNNNNVVT30-3′) with a photocleavable linker (PC), a bead barcode sequence (J, 14 bp), a UMI sequence (NNNNNNNVV, 9 bp) and a poly dT tail. The sample table of all Slide-seq data, denoting the puck origin and other metadata, is in Supplementary Table 2. OCT-embedded frozen tissue samples were warmed to −20 °C, sectioned to 10 μm on a cryostat (Leica CM3050S), placed on a pre-cooled Slide-seq array and melted onto the array by warming it briefly. For each sample, 2-3 sections for Slide-seq were collected >200 μm to maximize sampling diversity of tissue sub-structures and consecutive sections were collected for hematoxylin and eosin staining. The affixed tissue section and array was moved into a 1.5 ml Eppendorf tube for tissue digestion and transcriptomic library preparation as previously described36,37. The last library amplification was performed in a final volume of 200 μL of PCR mix, divided into four 0.2 ml tubes and cycled with the following program: one cycle of 98 °C for 2 min, four cycles of 98 °C for 20 s, 65 °C for 45 s, 72 °C for 3 min, nine cycles of 98 °C for 20 s, 67 °C for 20 s, 72 °C for 3 min and one cycle of 72 °C for 5 min. Libraries were sequenced using the following read structure on a NextSeq 1000/2000 (P2; Illumina) or NovaSeq (S4 or X; Illumina): Read1: 42 bp; Read2: 50 bp; Index1: 8 bp, and sequences were processed as previously described121 using the pipeline available at https://github.com/MacoskoLab/slideseq-tools. Pre-processing of single-nucleus and spatial transcriptomics data For snRNA-seq data: reads were demultiplexed with Cellranger (v.6.1.2) mkfastq (10x Genomics)122 and aligned to the mouse mm10 reference genome using STAR v2.7.8a with default parameters123. For spatial data: spatial barcode assignment and bead-based feature extraction were performed using the Slide-seq pipeline (v2022-01-01), resulting in a gene-by-spot expression matrix for downstream analyses123. Initial data processing Gene expression data of individual samples or pucks were concatenated using an outer join. For spatial data, spatial coordinates were assigned by matching barcodes to puck-specific barcode mapping files. The combined dataset was filtered to retain spots with at least 50 detected genes and 150 UMI counts. For spot clustering and embeddings, all gene expression values were scaled and centered, and the top 5,000 highly variable genes were identified from log-normalized counts across batches using Scanpy’s highly_variable_genes function (batch Key=puck)124. Dimensionality reduction was performed by regressing out total UMI counts, followed by principal component analysis (PCA) on highly variable genes. Puck-specific batch effects were corrected using a PyTorch implementation of Harmony125, after which a k-nearest-neighbor graph with 10 neighbors was constructed using the top 40 principal components. Unsupervised spot clusters were identified using the Leiden algorithm126 (resolution = 1.5) and visualized using UMAP127, implemented in Scanpy124 (v1.10.1). Smear removal Smearing artifacts were manually identified and removed from each puck using an interactive Python-based tool implemented with LassoSelector, which was used to exclude spot regions containing technical artifacts. After artifact removal, the dataset was reprocessed for dimensionality reduction and clustering as described above. Unsupervised clustering and cell-type annotation Cell types were annotated using top differentially expressed genes (DEGs) identified for each Leiden cluster using Scanpy’s rank_genes_groups (Wilcoxon rank-sum) and cross-referenced with established marker genes from literature32,35,109,128–134. Additional visualization of marker gene expression in space was used to confirm cell-type-specific identities based on known morphology. GCs were further subclustered by re-running Leiden clustering, followed by DEG analysis and marker-based assignment. The top 50 marker genes identified for each cluster are in Supplementary Table 3. Projection of additional pucks onto the annotated reference atlas To incorporate newly generated Slide-seq pucks (young, metestrus) into the reference atlas, raw gene-by-spot matrices were preprocessed as described above, including spatial coordinate assignment and smear removal. To ensure compatibility with the reference atlas, gene sets were harmonized prior to integration by restricting the query AnnData object to the intersection of genes shared between query and reference. The cleaned query data were projected onto the existing atlas using scvi-tools (v1.1.6), leveraging a pretrained SCVI model to infer a shared representation. Latent embeddings for the query were obtained and concatenated with reference embeddings for downstream analysis. For label transfer, scArches135 (v0.6.1) was used to adapt the pretrained SCVI model using the unsupervised surgery workflow, enabling alignment of query data to the harmonized latent space and probabilistic annotation transfer. Spot-level labels were assigned at both fine-grained and major cell-type levels, and low-confidence predictions were masked using an uncertainty threshold of 0.2. Spatial segmentation was performed as described below. Subclustering of immune cells and GCs To refine cell state annotation, we performed iterative subclustering of immune cells and GCs. AnnData objects were subset to populations of interest, normalized to 10,000 transcripts per cell with highly expressed genes down-weighted (max_fraction = 0.05), log-transformed and scaled to unit variance. Highly variable genes were identified using Scanpy’s highly_variable_genes function (flavor = ‘seurat’, batch_key = ‘segment_label’), followed by PCA and neighborhood graph construction. UMAP was used for visualization and Leiden clustering for subcluster identification. DEGs were identified for each subcluster using the Wilcoxon rank-sum test (min log fold-change ≥ 1). Marker genes used for annotation were required to be (i) specific to the cluster, (ii) highly expressed within that cluster and (iii) supported by prior literature as having cell-type-defining functional roles. Initial immune subclustering resolved macrophages, T cells, plasma cells, dendritic cells and mast cells; macrophages were then reclustered to identify distinct macrophage subtypes. A similar iterative approach was applied to GCs (Extended Data Fig. 1b), with a second round of subclustering focused on atretic GCs to resolve transcriptionally distinct subpopulations. A detailed description of the iterative subclustering strategy, DEG-based marker selection, and literature-based justification for GC and immune subtype annotation are provided in Supplementary Note; top-ranked DEGs for each GC subcluster are in Supplementary Table 3. Comparison of luteal populations with published datasets To assess the concordance of our luteal populations with prior CL classifications, we compared our data with the single-cell RNA-seq dataset from Morris et al.32. For clarity, we excluded the ‘Regressing CL’ cluster reported in Morris et al., as this population derives from post-partum ovaries and was outside our study design. Cell type composition was quantified using contingency tables of cluster identity against estrous stage. To compare gene expression profiles, we visualized DEGs defining luteal subtypes in both our dataset and Morris et al.32, and concordance was assessed by overlap of marker genes and expression patterns across annotated subtypes (Extended Data Fig. 5b–e). Spatial segmentation of oocytes, follicles and CL Code underlying these analyses can be found in II.2_segmentation.ipynb and II.3_segment_extension.ipynb. Oocyte segmentation ‘Oocyte’ and ‘Oocyte - cumulus’ clusters were initially used as putative oocyte spots to account for uncertainty at cluster boundaries in space. For each puck, a spatial neighborhood graph (radius = 50) was constructed and clustered using Leiden126 (resolution = 0.1) to define preliminary oocyte segments. ‘Oocyte - cumulus’ spots were subsequently removed, and segments containing fewer than three spots were excluded. Segments were further refined by subclustering, merging and manual curation based on tissue context. Decontamination of cumulus signals from oocytes To remove residual cumulus contamination from oocytes, NMF136 was applied iteratively to oocyte spots using scikit-learn137 (k = 8 programs). Programs associated with contaminating signals were identified by marker genes, and spots were annotated as contaminants if they either had these programs as their top utilized program or if the program scores reached the following thresholds: Program 1 (Kctd14, Slc18a2, Wt1): > 0.4, Program 2 (Lars2, Hexb, Cdk8): > 0.3, Program 4 (Inhba, Inhbb, Nap1l5): > 0.3, Program 6 (Col1a2, Col3a1, Cfh): > 0.4 (Supplementary Fig. 3a,b). Contaminating spots were removed across two rounds of filtering; in the second round, spots with program 1 (Serpine2, Ccnd2, Kctd14) as the top program or with a score >0.3 were annotated as contaminants (Supplementary Fig. 3c,d) and removed from the oocyte segments. Identifying primordial oocytes To capture small, putatively primordial, oocytes, unassigned ‘Oocyte’ spots or spots expressing Sohlh131,138 located at least 200 from existing oocyte segment centers were selected. For pucks containing at least two such spots, a neighborhood graph (radius 20 µm) was clustered using Leiden124 clustering (resolution = 0.1). Segments with less than 2 spots or lacking Sohlh1 expression were excluded. Segmentation of follicles and CL All GC clusters alongside ‘Cumulus – oocyte’ were used to define putative follicle spots, and luteal clusters to define putative CL spots. Spatial neighborhood graphs (radius 50 µm) were clustered with Leiden126 (resolution = 0.1 for follicles; 0.001 for CLs as they tended to be larger) for each puck, yielding segments. Segments with less than 20 spots were removed, and the remaining segments were refined by subclustering, merging, manual inspection, and extended to include nearby granulosa or luteal spots within 50 µm. These curation choices are documented in the supplementary code. In a few cases, oocyte segments were identified by the oocyte segmentation pipeline, but the corresponding follicle was partially or completely missed because it was too small or irregularly shaped. To recover these follicles, unassigned spots from follicle cell types within 50 µm of an oocyte segment were assigned as new follicle segments and grouped to the nearest oocyte segment. Newly identified follicle segments within 100 µm (minimal spot-to-spot distance between segment pairs) of an existing follicle segment were merged. To recover remaining small follicles, unassigned spots from the ‘Cumulus – oocyte’, ‘GC – preantral’, ‘GC – antral early mitotic’ and ‘GC – atretic’ clusters were treated as putative small follicles. For each puck, a spatial neighborhood graph (radius 20 µm) was constructed for these putative follicle and CL spots and clustered (Leiden resolution = 0.1) to identify additional segments. Segments with less than 5 spots were excluded. Remaining segments were merged with existing follicle segments if within 50 µm of existing segments (minimal spot-to-spot distance across segment pair), and were subsequently refined through subclustering, merging and reassignment to proximal oocytes as appropriate. These curation choices are documented in the supplementary code. Assigning proximal spots to segments Immune, endothelial (all ‘Endothelial’ types), theca (‘Theca’ and ‘Stromal - steroidogenic’) and mesenchymal (‘Adipocyte’, ‘Stromal - nonsteroidogenic’) spots located within 100 µm of a follicle or CL segment were assigned to that segment. Assigning oocytes to follicles Oocyte segments were assigned to the closest follicle segment if it was less than 100 µm away. These assignments were manually curated where appropriate based on manual inspection of the segmentation results and the surrounding tissue; these curation choices are documented in the supplementary code. Assigning segments to follicle and CL types Segments were first assigned as either follicle or CL. A segment was assigned as a CL if its most abundant type was a luteal cell type, or luteal spots comprised >50% of all GCs and luteal spots within that segment (Extended Data Fig. 1f,g). All remaining segments were classified as follicles. This classification was applied to all segments generated from both follicle and CL segmentations to identify cases missed during the initial segmentation. Segments were then assigned more specific state labels. Follicle segments were classified as ‘Follicle – atretic’ if >20% of constituent spots were annotated as ‘GC – atretic’. All other follicle segments were labeled according to their predominant GC subtype (‘GC – preantral’ as ‘Follicle – preantral’, ‘GC – antral luteinizing’ as ‘Follicle – antral luteinizing’, ‘GC – antral early mitotic’ and ‘GC – antral mitotic’ as ‘Follicle – antral mitotic’). CL segments were classified as ‘CL - early’ and ‘CL – late’ according to their predominant luteal cell type. Validation of the segmentation pipeline with manually segmented adjacent H&Es To validate the segmentation pipeline, adjacent H&E-stained sections were manually annotated for preantral follicles, antral follicles and CLs, and compared with segmented spatial plots. For this comparison, ‘follicle – antral luteinizing’, ‘follicle – antral mitotic’ and ‘follicle – atretic’ were grouped as a single follicle – antral’ category, as these states cannot be reliably distinguished by histology. Across young sections, 89.6% of annotated structures were concordant between H&E and spatial segmentation (Fig. 1g,h and Supplementary Figs. 1 and 2). Most discrepancies arose from structures captured in only one of the two adjacent sections because of sectioning offsets; for example, the edge of a follicle was captured in the Slide-seq section, whereas the adjacent H&E section does not contain any cells from that follicle (Supplementary Figs. 1 and 2). Remaining discrepancies reflected merging of neighboring follicles into a single segment due to close spatial proximity (7.9%), missed detection of low spot count segments (1.9%) or incorrect segment identity assignment caused by incorporation of granulosa spots from adjacent structures (0.6%) (Fig. 1h). Cohen’s kappa was calculated with scikit-learn137. Follicle counts and error types were visualized in GraphPad Prism version 10.0.0. Calculating mean expression profiles of oocyte, follicle and CL segments Raw counts of spots assigned to a segment (oocyte, follicle or CL) were normalized per spot to 10,000 and then averaged within each segment. Where appropriate, the resulting segment expression profiles were log1p transformed for downstream analyses. Trajectory inference of GCs GC trajectories were inferred using DPT139. DPT was performed on annotated GCs following PCA and neighborhood graph construction with k = 10 neighbors and 40 PCs. Diffusion components were computed from the neighborhood graph, and the trajectory root was manually selected after inspecting the diffusion component embedding. DPT was calculated in two complementary ways, depending on the analysis. For Fig. 3d, pseudotime was computed separately for GC spots in non-atretic and atretic follicles to assess spatial patterns of trajectories in follicles under developmental versus degenerative conditions. For Fig. 3h, DPT was computed across all GC spots to generate a unified ordering of GC states across the follicular lifecycle and enable comparisons across follicle types, estrous cycle stage and age (Extended Data Fig. 3g). Calculation of cell cycle scores Canonical marker genes of S-phase and G2/M-phase, as previously described140, were used to calculate GC cell cycle scores and classify them into G1, S or G2/M phases. G1 was determined as ‘non-cycling’, whereas S or G2/M phases were determined as ‘cycling’. Identifying gene programs with non-negative factorization Oocytes Gene programs in composite oocytes were identified using NMF (k = 5) in scikit-learn136,137 on log-normalized counts. Gene rank per program was determined by the gene weight matrix. GCs GC GEPs were identified using consensus NMF (cNMF) on raw counts of 5,000 highly variable genes using a Harmony-corrected expression matrix generated with an adapted cNMF-integrated implementation125,141 (negative values set to zero before factorization). cNMF factorized the data into a gene weight matrix, capturing gene contributions to programs, and a usage matrix, capturing program activity per spot. The number of programs was set to k = 16 to balance program stability and reconstruction error based on Pearson correlation across 20 factorization replicates per k. cNMF was then re-run with 200 replicates to generate final consensus programs. Genes were ranked within each program by contribution to the gene weight matrix (Extended Data Fig. 3f), and program activity scores were normalized per spot by total program usage per spot. Spots were assigned a dominant program based on highest normalized usage (Extended Data Fig. 3d). Programs were annotated using top contributing genes and known GC markers and pathways11,20,112,142,143. Radial distance analysis within segments To quantify spatial organization within follicular segments, radial distances for each spot was calculated as its Euclidean distance from the segment centroid, normalized by the maximum segment radius to yield a ‘scaled radial distance’ ranging from 0 to 1 (Extended Data Fig. 3a). Spots were binned into equal intervals along this scaled radial axis to enable a standardized comparison across segments of varying sizes. For each radial bin, the mean and s.e.m. were calculated for features such as cell-type composition, DPT, cell cycle scores and cNMF-derived GEP, and then aggregated by segment type. Monotonic spatial trends were assessed using the Spearman rank correlation between radial bin position and mean feature values (for example, cell-type composition, cell cycle state, DPT or GEP usage) for each segment type. Correlation coefficients (ρ) and associated P values were calculated using SciPy; bins with fewer than two data points were excluded from the analysis. Analysis of GEPs with pseudotime (DPT) To investigate transcriptional dynamics during follicle development and atresia, GC GEP usage scores were analyzed along DPT, computed as a unified trajectory across all GC spots. Stage stream plots depict follicle type occupancy along pseudotime, estimated using Gaussian kernel density estimates over 400 grid points and displayed as stacked area plots. Estrous stage occupancy (second row, where available) was computed similarly. GEP heatmaps were generated separately for non-atretic and atretic by binning pseudotime into 500 equal-width intervals and calculating mean GEP usage per bin. Age-associated differences in GEP usage were tested within each GEP, follicle type and estrous stage using ordinary least squares (OLS) regression with cluster-robust standard errors (Usage_k ~ age; young as reference), clustering variance by puck to account for within-puck dependence of spots. Strata lacking both age groups were not estimable. Across all tested GEP × follicle type × stage combinations, P values were adjusted for multiple testing (Benjamini–Hochberg; significance set at FDR < 0.05). Analyses were performed using statsmodels (statsmodels.formula.api.ols, statsmodels.stats.multitest.multipletests). For an alternative visualization (Extended Data Fig. 3g), GEP usage (Usage_1 - Usage_16) was summarized using a three-level averaging scheme to respect biological replication. GEP usage scores were first averaged within individual follicle segments to generate follicle-level profiles, then averaged across segments within each puck for each follicle type, estrous stage and age, and finally averaged across pucks to obtain condition-level means for age × stage × follicle type combination. For visualizations, categories were arranged along an interleaved axis: age → follicle type → estrous stage to enable side-by-side comparisons across conditions. Compositional analysis of cell and segment types Global cell-type compositional analysis Cell-type composition was calculated as the fraction of spots assigned to each cell type within each puck, and then averaged across young, middle and old groups. Age-associated compositional changes were tested using scCODA143, with young as the reference condition, and ‘Epithelial – oviduct’ as the reference cell type because of its low variability across conditions. Immune subsets were additionally stratified by segment assignments (follicle, CL or stroma if unassigned). CL-associated endothelial cells were analyzed separately from other endothelial cells. Cell-type composition within segments Within each segment, cell-type composition was calculated as the fraction of spots assigned to each cell type, and then averaged across segment types. To test age-associated differences within follicle types, proportions were transformed using a centered log-ratio and modeled using OLS regression (centered log-ratio ~ age) with cluster-robust standard errors by puck to account for within-sample correlation. As before, P values were adjusted for multiple hypothesis testing (Benjamini–Hochberg; significance, FDR < 0.10). Segment-type composition analysis Segment-type composition was calculated as the fraction of segments of each type within each puck and summarized by estrous stage and age. Mean fractions and s.e.m. were computed across pucks for each condition, and differences between groups were assessed using two-tailed t-tests. Follicle embedding and trajectory inference using optimal transport Code for these analyses is in III_segment_OT_1_pca_coretypes.ipynb and III_segment_OT_2_plot_results.ipynb. Segment-to-segment distance matrix computation Pairwise distances between all follicle and CL segments in young mice were computed using optimal transport, with each segment represented by its oocyte, granulosa or luteal, and theca spots. For each segment pair, a transport matrix was estimated using the Sinkhorn algorithm implemented in moscot144 (epsilon = 0, tau_a = 1, tau_b = 1, rank = 20, initializer = k-means, min_iterations = 5, max_iterations = 200). Segment distance was defined as the sum of the element-wise product of the transport matrix and the corresponding cost matrix of the optimal transport problem. Segment embedding and follicle pseudotime The resulting segment-to-segment distance matrix used as a feature space for embedding. A k-nearest neighbor graph (k = 15) was constructed from the top 50 PCs and visualized using UMAP. Embeddings were generated for all follicle and CL segments in young mice, as well as follicles alone in young mice as the reference trajectory. DPT139 was then computed for young follicles using a manually selected root segment. Trajectory projection of follicles from middle-aged and old mice Distances between segments from young mice and those from middle-aged and old mice were computed using the same optimal transport approach as above. Each segment from middle-aged or old mice was assigned to its closest young segment based on this distance and that young segment’s pseudotime coordinate. Cell–cell communication analysis Code for these analyses is in III.5_nichenet_ageYoung_stageAll_segmentFollicle_withOocyte_byCluster.Rmd and III_segment_OT_2_plot_results_v0.3.ipynb. Cell–cell communication analysis was performed using MultiNicheNet85 on young mice, treating follicle segments as samples. Comparisons were made across coarse follicle stages (preantral, antral, atretic) using oocyte, granulosa, theca and immune spots. Ligand–receptor interactions and sender-receiver cell-type pairs enriched in each follicle stage were prioritized. Follicles that did not meet default cell-type compositional thresholds were discarded (at least 5% of spots per segment mapping to each cell type). To summarize communication states across all follicles, a segment-level ligand–receptor score was calculated by averaging normalized expression of ligand–receptor pairs across relevant cell types within each segment, applying a log1p transformation and summing the resulting ligand–receptor scores to generate the final segment-specific score. Immune infiltration analysis To compare immune cell infiltration across segment types, immune spot composition was analyzed as a function of distance from the segment center. For each scaled radial distance bin (defined above) within a segment, immune infiltration was calculated as the fraction of immune spots among all spots in that bin. Mean immune composition and s.e.m. were then computed across segments within each radial bin. Differences across radial bins were evaluated using Tukey’s honest significant difference test. To control for potential bias from tissue sectioning, additional analyses were repeated on only oocyte-containing follicles to confirm that slicing artifacts did not affect the evaluation of radial infiltration (Extended Data Fig. 1j). Statistical analyses were performed using SciPy145. Radial bins with fewer than two data points were excluded from both statistical and visualization analyses. Age-associated differential gene expression analysis Follicles For follicular hormone-response genes, mean expression and s.e.m. were calculated within each follicle type (preantral, antral mitotic, antral luteinizing and atretic), estrous stage (estrus, metestrus) and age group for Amhr2, Fshr and Lhcgr8,28,98,99,145. Differential expression between estrous stages was performed using two-sided t-tests. Oocytes and stromal cells Differential expression between young and old oocytes or spots was performed in Scanpy using a two-sided t-test with overestimated variance and Benjamini–Hochberg correction. Genes expressed in fewer than 10% of oocytes or spots were excluded. Genes with adjusted P < 0.05 and log2 fold-change >1 were used for further analysis. Oocytes: primordial oocytes were removed due to their low representation in old animals. Stromal cells: ‘Fibroblasts’, ‘Stromal – nonsteroidogenic’ and ‘Stromal – steroidogenic’ were used for differential expression analysis. Identifying MNGCs Code for these analyses is in II.5_registration_v0.8_all_pucks.ipynb and III_MNGC_dotplot_v0.1.ipynb. Annotation of MNGCs in histology sections Adjacent H&E-stained sections were manually annotated for MNGCs based on pigmentation, size and proximity to growing follicles, as previously described8,28,98,99,146. Sections were classified as ‘visible MNGCs’ (high density of light brown clusters; n = 19), ‘possible MNGCs’ (lower density or weaker pigmentation; n = 13) or ‘no MNGCs.’ MNGC segmentation in registered spatial transcriptomics data Samples containing MNGCs were selected for registration to Slide-seq and segmentation using a napari plugin in squidpy147, with anchor points manually defined using clearly visible tissue features in both modalities (such as large oocytes or points on the edges of the tissue section). In the registered space, MNGC regions were segmented based on histology and mapped to spot transcriptomics spots. For immune analysis, immune spots outside follicles and CLs were classified as ‘stromal’ (outside of MNGC segments) and ‘MNGC’ (inside of MNGC segments). Co-localization analysis of segmented tissue structures Code for these analyses is in III_segment_colocalization.ipynb. Within each puck, a spatial neighborhood graph of follicles was constructed using follicle centroids (spatial center point of its constituent spots). Homotypic neighborhood enrichment for preantral, antral and atretic follicles was then computed on this graph using Squidpy147 across spatial radii from 200 µm to 2,000 µm, in 200 µm increments. Measurement of surface epithelium thickness To validate the increased abundance of surface epithelial cells in aged ovaries, epithelial thickness was measured in ImageJ148. For each section, three thickness measurements were taken and averaged; one section per ovary was analyzed and plotted using GraphPad Prism (v10.0.0). Statistics and reproducibility All experiments used C57BL/6 J female mice maintained under standardized husbandry conditions. Sample sizes (n = 22 ovaries across the reproductive lifespan and estrous cycle; 69 spatial transcriptomic profiles) were selected to capture multicellular dynamics across chronological age and hormonal states. No statistical method was used to predetermine sample size; sizes were based on established practice for large-scale spatial transcriptomics and the feasibility of obtaining high-quality in situ profiles across diverse follicular stages. Biological replicates (individual mice and ovaries) were used for all major findings, as indicated in figure legends. Spatial organization, temporal coordination and immune-associated phenotypes were reproducibly observed across independent ovaries within the same age and estrous cycle stage. Data exclusion For Slide-seq analysis, spatial transcriptomic profiles were subjected to predefined quality-control filtering, excluding spots with low transcript counts or poor mapping to tissue architecture. For segmentation analyses, only intact niches containing oocytes (n = 358), follicles (n = 668) and CL (n = 236) were included in multicellular dynamics modeling. Apart from these standard quality-control and biological-integrity filters, no data were excluded. Randomization and blinding Animals were assigned to experimental groups by chronological age and estrous cycle stage and were therefore not randomized. The investigators were not blinded during experiments or outcome assessment, as knowledge of the reproductive stage was essential for tissue collection and interpretation of the spatial niches. Statistical analyses Statistical analyses were performed in Python (v3.10) and Scanpy (v1.10.1) unless otherwise stated. Data distribution was assumed to be normal, but not formally tested. Owing to the overdispersed and non-normal nature of spatial transcriptomic data, non-parametric Wilcoxon rank-sum tests were used for two-group comparisons unless otherwise specified. Multiple testing correction was performed using the Benjamini–Hochberg procedure, with adjusted P < 0.05 considered significant. Specific statistical tests, significance thresholds and definitions of plot elements (for example, box plot whiskers) are provided in the corresponding figure legends. All code for spatial analysis, segmentation and niche identification is available in the GitHub repository (https://github.com/HattieChungLab/ovarian_aging), and primary datasets are deposited in Gene Expression Omnibus (GEO) and cellxgene as stated in the ‘Data availability’ section. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary InformationSupplementary Figs. 1–3 and Note. Reporting Summary Supplementary TablesSupplementary Tables 1–22. Supplementary Figs. 1–3 and Note. Reporting Summary Supplementary Tables 1–22.