An accepted manuscript in eLife Sciences has described eyeAge, a new clock that uses deep learning to analyze the eye in detail in order to predict chronological age and age acceleration.
Retinal signs of aging
This project was developed with the knowledge that fundus imagery, which shows the blood vessels at the back of the eye, can be a useful diagnostic tool for eye diseases like age-related macular degeneration (AMD) [1] and diabetic retinopathy [2] along with seemingly unrelated diseases, such as hypertension [3] and cancerous tumors in entirely different body parts [4]. The researchers hold that this diagnostic power is because such diseases cause tiny changes in the vascular system that show up in the smallest capillaries first, and retinal capillaries are particularly small.
However, detecting these tiny changes is difficult for human beings even with modern instruments. Therefore, the researchers have employed a deep learning system to build their aging clock, as such systems have been used successfully in diabetic retinopathy [5] and other diseases.
A broad study using two datasets
This study drew its data from the EyePACS database, using images from over 100,000 people for development and about a quarter that many for tuning. To test the robustness of their model, the researchers turned to the UK Biobank, which contained nearly 120,000 relevant images and different demographics.
The UK Biobank cohort also contained extensive genomic data that was useful in comparing eyeAge to PhenoAge, a well-known epigenetic aging clock. Interestingly, while eyeAge was better correlated with chronological aging (at 0.87) than PhenoAge (at 0.82), the two clocks were not as well correlated with each other, at 0.72. Except for their correlations with chronological aging, these two clocks were almost entirely independent of one another.
Connecting visual results to genomics
The researchers used eyeAge data to create an age acceleration measurement, eyeAgeAccel, similar to acceleration measurements in other clocks, and compared this measurement to UK Biobank genomics through a genome-wide association study (GWAS), looking for the genes that were strongly associated with rapid eye aging.
As expected, many of the associated genes were directly linked to the eyes and associated diseases, including AMD, cataract development, and pigmentation. Some of the other genes were linked to cancer development and hearing loss, and another single gene had been implicated in both muscle wasting [6] and Alzheimer’s disease [7].
One particular gene, ALKAL2, was of special concern. The Drosophila version of this gene has been shown to be associated with shorter lifespans, and inhibiting it in these flies was found to increase their lifespans. The researchers also noted that their research confirms previous research showing a human allele that decreases the expression of ALKAL2 and other negatively associated genes in this study [8], marking a prospective target for potential future therapies.
Conclusion
As a visual biomarker that only needs imagery as an input, eyeAge may be a simple tool for researchers and clinicians to measure accelerated aging and gain valuable information about the prognosis of eye diseases. Future research using this clock alongside other clocks will determine its value in measuring the effectiveness of interventions against age-related diseases.
In conclusion, predicted age from retinal images can be used as a biomarker of biological aging that is independent from assessment based on blood markers. This study demonstrates the potential utility of a retinal aging clock for studying aging and age-related diseases and quantitatively measuring aging on very short time-scales, opening avenues for quick and actionable evaluation of gero-protective therapeutics.
Literature
[1] Luu, J., & Palczewski, K. (2018). Human aging and disease: lessons from age-related macular degeneration. Proceedings of the National Academy of Sciences, 115(12), 2866-2872.
[2] Namperumalsamy, P., Kim, R., Vignesh, T. P., Nithya, N., Royes, J., Gijo, T., … & Vijayakumar, V. (2009). Prevalence and risk factors for diabetic retinopathy: a population-based assessment from Theni District, south India. Postgraduate medical journal, 85(1010), 643-648.
[3] Wong, T. Y., & McIntosh, R. (2005). Systemic associations of retinal microvascular signs: a review of recent population‐based studies. Ophthalmic and Physiological Optics, 25(3), 195-204.
[4] Kreusel, K. M., Wiegel, T., Stange, M., Bornfeld, N., Hinkelbein, W., & Foerster, M. H. (2002). Choroidal metastasis in disseminated lung cancer: frequency and risk factors. American journal of ophthalmology, 134(3), 445-447.
[5] Gulshan, V., Peng, L., Coram, M., Stumpe, M. C., Wu, D., Narayanaswamy, A., … & Webster, D. R. (2016). Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. Jama, 316(22), 2402-2410.
[6] Judge, S. M., Deyhle, M. R., Neyroud, D., Nosacka, R. L., D’Lugos, A. C., Cameron, M. E., … & Judge, A. R. (2020). MEF2c-Dependent Downregulation of Myocilin Mediates Cancer-Induced Muscle Wasting and Associates with Cachexia in Patients with CancerLoss of Myoc via MEF2c Mediates Cancer Cachexia. Cancer research, 80(9), 1861-1874.
[7] Xue, F., Tian, J., Yu, C., Du, H., & Guo, L. (2021). Type I interferon response-related microglial Mef2c deregulation at the onset of Alzheimer’s pathology in 5× FAD mice. Neurobiology of disease, 152, 105272.
[8] Woodling, N. S., Aleyakpo, B., Dyson, M. C., Minkley, L. J., Rajasingam, A., Dobson, A. J., … & Partridge, L. (2020). The neuronal receptor tyrosine kinase Alk is a target for longevity. Aging Cell, 19(5), e13137.
