MainAging is the biggest risk factor for most chronic diseases1. Current aging research has advanced our molecular understanding of aging, including both cell-intrinsic processes such as telomere shortening, senescence and autophagy, as well as systemic pathways such as the insulin/IGF-1 axis1,2,3,4. Molecular aging clocks have improved prediction of risk of mortality and diseases5,6. Recent studies further show that molecular aging unfolds nonlinearly, pointing to the possibility of time-period-specific targeting7,8.While we have advanced our knowledge of molecular, cellular and systemic changes during aging, a critical missing gap remains: how the physical structure of tissues, the spatial organization of cells, vasculature and extracellular matrix, and the basis of their function, deteriorate with aging (structural aging). Our current understanding of structural aging comes from manual inspection of histology images revealing both global deterioration patterns such as fiber fragmentation9,10, atrophy11 and fibrosis12 as well as tissue-specific patterns such as ovarian follicle loss13, bone porosity14, brain cortical thinning and demyelination, muscle atrophy15, arterial stiffening16, heart fibrosis17,18, thymic involution19 and lens clouding20. These structural changes cause multiple known organ function loss during age, such as ovarian follicle loss reducing fertility and cardiac fibrosis increasing heart failure risk. While scan-based histology clocks are trained on chronological age21,22, there is no systematic study asking how organ structure changes during aging. This gap has persisted due to lack of large-scale high-resolution tissue images across the body and computational tools to quantify these tissue substructures.These limitations have given rise to several fundamental open questions about aging: When and how do different tissue physical structures deteriorate during aging? What are their molecular drivers? Are there tissue-specific periods of accelerated structural aging (ASA), and what molecular dysregulation drives this acceleration? Which organs are early-agers versus late-agers? Do organs age independently or in coordination? How do clinical factors accelerate or protect against structural aging? And, finally, how do germline variants affect organ-specific structural aging?We present PathStAR (Pathology-based Structural Aging Rate), a framework that quantifies tissue structural aging from histopathology images without training on chronological age. We applied it to ~25k post-mortem biopsies across 40 tissues from 970 Genotype-Tissue Expression (GTEx) donors aged 21–70 years. Unlike age-prediction clocks, PathStAR resolves when structure changes: it recovers the ovary’s nonlinear decline, fertility loss in the 30s and menopause in the 50s. It then maps three temporal aging programs, coordinated cross-organ deterioration, molecular drivers of each accelerated period, and germline variants tied to organ-specific trajectories, establishing structural aging as a quantifiable dimension of aging that complements molecular clocks.ResultOverview of the PathStAR pipelineBased on the notion that histology images capture architecture features critical for tissue function, we developed a computational framework called PathStAR (Fig. 1a–c), that quantify how and when this tissue structure changes during aging from high-resolution hematoxylin and eosin (H&E) images of large-scale post-mortem biopsies.Fig. 1: Overview of PathStAR pipeline steps and the GTEx cohort used here.Full size imagea–c, PathStAR computational pipeline: PathStAR starts by extracting morphology feature of 256 × 256 pixels patch level using a pretrained pathology foundation model (UNI), which are aggregated (mean pooling) to generate slide-level representation (a). It then computes structural aging rate by quantifying the rate of morphological change per year using sliding 10-year windows, generating age-specific trajectories that reveal periods of ASA (b). Finally, it computes individual delta-structural aging scores (red) measuring each sample’s deviation from the population aging trajectory (blue) (c). d, Number of samples available for this study per tissue type. e, Uniform Manifold Approximation and Projection (UMAP) based on whole-slide level representation cluster different tissue types, demonstrating its ability to capture tissue type-specific morphological properties. f, Morphological patch level features from UNI clustered using spatial information and K-means, where different colors represent different clusters. Figure created in BioRender; Alvarez, K. https://biorender.com/5o5bsgk (2026).PathStAR was applied on the GTEx cohort23, post-mortem biopsies from 970 nondiseased individuals (age range 21–70 years, 66.4% male donors; Extended Data Fig. 1), spanning 40 tissue types (Fig. 1d). GTEx provided a total of 25,306 organ biopsy whole-slide scans (×20 magnification), from which we extracted 30.3 million patch images (512 × 512 pixels). Pathology annotations from GTEx for these images confirmed canonical aging-related increase in atrophy, fibrosis, cysts and atherosclerosis and decrease in ovum, spermatogenesis and protective layers including mucosa, endometrium and myometrium (Supplementary Figs. 1 and 2), demonstrating quality to develop a systematic method. The tissue-specific quality control analysis is presented in Supplementary Note 1.PathStAR comprises three steps (see details in the Methods):(1)Step 1: feature extraction from whole-slide images. After background removal, each slide is segmented into patches, each encoded into a 1,024-dimensional embedding via UNI24, a vision transformer (ViT) pretrained on more than 100,000 whole-slide images. Mean pooling yields one slide-level representation per sample (Fig. 1a). These capture tissue-specific morphology, separating distinct tissues and substructures in low-dimensional space (Fig. 1e,f). We tested multiple encoders to ensure results are not encoder-specific artifacts (Supplementary Notes 2).(2)Step 2: construction of structural aging trajectories. We next asked: when does tissue structure change most rapidly during aging? For each tissue, we quantify the per-year rate of structural change, the structural aging rate, by comparing morphological representations between consecutive 10-year age windows using an effect-size measure. Sliding the window in 1-year increments yields a continuous, population-level trajectory (ages ~30–60) that distinguishes rapid from stable periods; intervals of pronounced acceleration define ASA periods.(3)Step 3: individual-level deviation from population trajectories. With population-level trajectories established, we next ask: which individuals are aging faster or slower than expected for their tissue and age? For each sample, we compute an individual structural aging score and measure how much a given tissue in a given individual deviates from its expected aging path (Methods and Fig. 1c, delta-structural aging score). These scores enable identification of individuals with accelerated or protected structural aging, and downstream analysis of the clinical, lifestyle and genetic factors that drive these deviations.PathStAR captures ovary’s functional decline and its molecular correlates without chronological age trainingMorphology captures ovary’s nonlinear functional decline without supervisionThe ovary exhibits a unique functional decline: fertility decreases gradually from the early 30s, marked by rapid follicular depletion accelerating through the late 30s, culminating in menopause at ~50–55 years25,26. Molecular clocks trained on chronological age assume linear aging and thus cannot capture this nonlinear remodeling (Extended Data Fig. 2). We therefore tested whether PathStAR can capture this established nonlinear functional decline.Patch-level UNI features spatially segregated distinct ovary substructures (Fig. 2a–c and Supplementary Fig. 3) and readily distinguished young, middle-aged and old samples, with the first two principal components (PCs) correlating with age (r2 = 0.43), indicating substantial age-related structural change (Fig. 2c). Computing the structural aging rate trajectory across 250 ovary biopsies (ages 21–70; Fig. 2f) yielded a striking bimodal pattern, with two accelerated-structural-aging peaks aligning with reproductive milestones (Fig. 2f and Methods): the first ASA period peaks at ages 35–40, coinciding with fertility decline, and the second at ages 55–60, corresponding to menopause. Ischemic time and other post-mortem artifacts showed minimal correlation with this trajectory (Supplementary Note 3 and Supplementary Fig. 4).Fig. 2: Morphology features capture nonlinear ovary structural aging that bulk-molecular profiles cannot.Full size imagea,b, The morphological patch level features from two ovary samples from both older (a, left, age 60–69) and younger (b, right, age 20–29). The cluster shows that UNI can very well capture the structural property of the tissue. c,e, PC analysis of all morphology features extracted by UNI in 250 ovary samples from whole-slide images separates young versus old samples, without prior training on chronological age: expression (c) and methylation (e) profiles from the matched samples. f, PathStAR applied to UNI features from 250 ovarian samples, age 21–70 years, captures nonlinear structural aging. Structural aging rate was estimated using sliding 10-year age windows (for example, 21–30 versus 31–40), advanced in 1-year increments across the age span. Each data point represents the structural difference between two adjacent windows and is positioned at the lower bound of the older window. Windows containing insufficient donor representation were excluded, yielding 28 retained trajectory points derived from the full cohort. Trajectory points were retained for smoothing when more than 5% of UNI features were significant. Trajectory points were considered significant when ≥5% of UNI features had nominal P 0.15) as well as at least 200 total samples with samples distributed across the whole age span, for calculation of the structural aging trajectory.Fig. 3: PC analysis of UNI-extracted slide-level morphological features for all 40 tissue types (n = 970 individuals, 25,306 biopsies).Full size imageEach point represents one individual’s tissue sample; color indicates chronological age, with darker shades representing younger donors (ages 21–30) and lighter shades representing older donors (ages 60–70). Tissues are arranged by organ system, with each system outlined in a distinct color. Tissue names marked with an asterisk indicate tissues meeting both criteria for age-driven morphological separation: PC1 and PC2 jointly explain more than 15% of variance in chronological age by ordinary least-squares regression (r2 > 0.15, F-test; Supplementary Table 1). Significance of the F-test is indicated by asterisks adjacent to tissue names (*P