From microscopic tissue changes to a simple blood sample, researchers show how artificial intelligence can map organ-specific aging and uncover biological signatures linked to disease.
Study: Histological aging signatures for monitoring tissue-specific aging and disease. Image Credit: Digital Photo / Shutterstock
A recent study published in the journal Nature Medicine highlighted age-related molecular changes across different human tissues based on histological examination and deep learning (DL) algorithms.
The analysis of histological images, ribonucleic acid sequencing (RNA-seq) data, and comparisons with deoxyribonucleic acid (DNA) methylation clocks revealed tissue-specific alterations associated with biological aging. Histologically derived biological age was associated with greater changes in transcriptional pathways and signatures than chronological age.
They also found blood-derived tissue age-gap predictions associated with chronic diseases in organs beyond the primary disease site, suggesting that the comprehensive approach could help identify disease-related aging patterns across organs from minimally invasive blood samples.
Age-related physiological declines are associated with molecular alterations in several biological pathways. Such alterations influence biological age, which may vary even among two individuals born in the same year. Age-related alterations could increase an individual’s risk of chronic diseases. An improved understanding of these changes could help develop more targeted strategies based on personalized risk assessments.
About the study
In the present study, researchers examined age-related molecular alterations in human tissues. To do so, they analyzed 25,712 whole-slide histopathological images (WSIs) of 40 different tissues across 29 organs of 983 deceased individuals (mean age, 53 years). The postmortem samples were obtained from the Genotype-Tissue Expression (GTEx) project.
Using DL algorithms, the team quantified morphological changes and developed ‘tissue clocks’ that predicted biological age based on tissue structure. Regression models were trained on morphological features from WSIs to predict age, with age gaps calculated as the difference between predicted and chronological age.
The team also analyzed previously available telomere-length measurements from 6,197 corresponding tissue samples. They additionally examined associations between gene expression and age gaps.
The researchers also investigated whether demographic, medical, and lifestyle factors could influence the biological age of different tissues. Integrating histological and transcriptomic data helped them predict age gaps from blood samples. To externally validate the blood-based predictors, the team analyzed gene-expression data from blood samples from 577 healthy individuals and 628 people with seven chronic diseases or stroke from the ARCHS4 database.
Results
The tissue clocks correlated with established biomarkers of aging, such as telomere attrition, subclinical pathologies, and comorbid conditions. Higher histological age gaps were associated with shorter telomeres and greater comorbidity burden.
The findings were pronounced in the esophagus, stomach, pancreas, prostate, and kidney. While DNA methylation clocks performed comparably to or better than telomere length, histological age gaps were more consistently associated with comorbidity burden.
Biological age gaps captured features less evident with chronological age. For instance, the team observed that age gaps captured muscle atrophy more clearly than chronological age alone. Samples with wider age gaps also showed more pronounced tissue-specific pathological changes.
The cerebellar samples of such individuals showed discoloration associated with myelin loss and ischemic changes. Likewise, aorta samples from people with wider gaps exhibited thickened walls with loss of integrity, changes associated with atherosclerosis and other vascular disorders.
Several organs showed specific morphological alterations. While fat infiltration was observed in skeletal muscles, uterine samples showed microvascular rarefaction. The DL model achieved a mean absolute error (MAE) of 4.88 years. Combined with a coefficient of determination of 0.69, these findings suggest that the model outperformed classical models, while its performance was comparable to that of foundation models.
In external validation, GTEx clocks showed tissue-specific correlations between predicted and chronological age (Pearson r = 0.76, 0.56, and 0.46 for lung, brain, and skin, respectively) and, in the lung cohort, moderate agreement with DNA methylation clocks (r = 0.30- 0.47). Separately, within GTEx, blood-based predictions performed particularly well for the systemic age gap, gastrointestinal tract, and spleen. Predicted age gaps were negatively associated with tissue telomere length and showed enrichment for tissue pathologies.
ARCHS4 analyses showed significant differences in blood-derived predicted tissue age gaps between disease and healthy groups. Blood samples from people with stroke showed elevated predicted brain age gaps, while those from people with cystic fibrosis showed elevated predicted kidney age gaps. Positive predictive values (PPVs) for disease classification using thresholded blood-derived tissue age gaps exceeded 0.3 in several cases, suggesting potential applicability to population-level screening.
Age-related alterations were also associated with changes in the expression of several genes. In fact, even genes typically not expressed in the affected tissue were altered. For example, the EYA transcriptional coactivator and phosphatase 4 (EYA4) was upregulated in adipose tissue from people with higher biological ages, although it is typically expressed in the tongue, muscle, and brain.
In the exploratory GTEx clinical-factor analysis, higher cerebellar age gaps were associated with unexplained seizures, while accelerated prostate aging was associated with hypertension in men; no formal statistical testing was performed for these associations. In addition, samples with wider age gaps showed upregulation of pathways related to inflammation and apoptosis. In contrast, metabolic processes such as adipogenesis and oxidative phosphorylation were downregulated in these samples.
Conclusion
The findings highlight organ-level molecular and structural changes associated with biological aging and existing disease. If confirmed in larger prospective cohort studies integrating genetic and longitudinal data, clinicians could potentially infer tissue-specific biological age from minimally invasive blood tests. However, the cross-sectional postmortem design prevents causal inference, and external blood-based age gaps could not be directly calibrated against paired tissue histology.
Prospective longitudinal studies are needed to establish whether these blood-derived signatures precede disease onset and can support early detection or future risk prediction.
Scientists map blood chemistry links to hundreds of diseases