Microscopy-Based Models Identified Different Signs of Aging Across 40 Tissue Types

Microscopy-Based Models Identified Different Signs of Aging Across 40 Tissue Types

On August 14, Nature Medicine published a study in which the researchers trained a separate model for each tissue type to estimate age from a stained section viewed under a microscope. They compared the estimates with pathology findings and telomere length in 25 713 images representing 40 tissue types from 983 deceased donors. They also used blood gene activity data to train a separate model.

Chronological age measures the time elapsed since birth. A histological section is a thin, stained slice of tissue that reveals cells, fibers, organ linings, and small blood vessels. The material came from the GTEx project, which collected tissues from the same deceased donors, along with gene activity data and pathologists' notes.

For each of the 40 tissue types, the team trained a separate model to predict the donor's chronological age from the structure of the section. The mean absolute error was 4.88 years. The authors then calculated the tissue age gap, defined as the difference between the model estimate and chronological age. If a section from a 55-year-old person resembles sections from older people, the gap increases. In GTEx, larger gaps were associated with shorter telomeres, the protective regions at the ends of chromosomes, and with pathological features visible in the same sections. These features included loss of myelin, the sheath surrounding nerve fibers, in the cerebellum, as well as changes in the aortic wall.

The authors also compared 19 histological clocks with 28 DNA methylation clocks, which measure chemical marks on DNA that change with age. Their age gaps showed almost no agreement in GTEx: the correlation coefficient was 0.09, whereas 1 would indicate perfect agreement. In an independent study of lung tissue, the coefficient was higher, at 0.30–0.47. DNA methylation and visible tissue structure therefore capture partly different aspects of tissue condition. In independent datasets of brain, lung, and skin sections, the correlations between predicted and chronological age were 0.56, 0.76, and 0.46, respectively.

The authors then used the previously calculated tissue age gaps as targets for models based on blood data. In the GTEx dataset, they linked blood gene activity to the age gaps of specific tissues, then applied the models to 1 205 external samples. The model estimated a larger brain age gap in people with stroke. In people with Crohn's disease, a chronic inflammatory bowel disease, the estimated gaps were larger for the esophagus, stomach, small intestine, and large intestine.

This approach gives each tissue a separate age estimate based on its microscopic structure. The blood-based models look for evidence of this tissue signal in a more accessible sample.