Abstract:Aging has been a fascinating subject since ancient times. It is a process of gradual decline in the general function of an organism, manifested by age-dependent alteration and destabilization at the whole systems level [1]. It is the primary risk factor for many complex human diseases, including cardiovascular diseases, neural degenerative diseases, type 2 diabetes, and various cancers [2]. As aging is a system wide phenomenon, a systems biology perspective is needed to comprehensively and unbiasedly understand the regulation and intervention of aging. Below we summarize our attempts in this direction and then discuss the great potential of quantitative imaging in associating changes in molecular networks with the aging phenome. First, by integrating interactome data with age-dependent gene expression data, we have constructed an active subnetwork for aging using protein-protein interaction networks by selecting interactions linking genes that are transcriptionally correlated or anti-correlated during the aging process [2, 3]. In both the human brain aging network and, the fruitfly aging network we have found transcriptionally anticorrelated gene clusters, corresponding to alternative cellular states, such as proliferation versus differentiation states or oxidative metabolic versus reductive metabolic states [3]. Surprisingly, the genes connecting transcriptionally anti-correlated modules are largely transcriptional and epigenetic regulators, including those regulating histone H3 lysine 27 trimethylation (H3K27me3). We have further observed that the gene encoding the H3K27me3 demethylase UTX displayed a significant aging-dependent expression increase in the human brain, while its methytransferase Ezh2 decreases [4]. This aging-dependent increase also exists for the C. elegans ortholog utx-1 gene, and reducing utx-1 gene expression by RNAi delays the aging process and extends C. elegans lifespan by 30%. This lifespan extension requires a functional daf-16 gene and acts through the IGF-1 pathway [4]. To identify common pathways affected by different dietary and energy interventions that modulate aging and lifespan, we obtained hepatic gene expression profiles of middle age mice on high-fat or low fat diet, with or without caloric restriction (CR), and with or without voluntary exercises. Pathway analysis showed that those pathways whose gene expression levels are associated with the mean lifespan across these different regimens are enriched for aging regulatory functions. This has allowed us to identify a set of known as well as novel pathways that are commonly targeted by different dietary and energy interventions and potentially mediate their effects on lifespan [5]. Most integrative analyses on aging, including our own works described above, have relied solely on molecular changes in the aging process. Moreover, to date, aging research has almost solely relied on mortality and lifespan as a gold standard for pro- or anti-aging determination. However, whether other phenotypes can replace mortality and lifespan as more robust aging and age evaluation benchmarks, in particular at the individual level; whether any individuals age faster or slower than the average; whether any parts of the body age faster or slower; and whether all aging-related molecular changes contribute to all individuals and all aging phenotypes equally or differentially, are questions that can only be effectively answered if the whole spectrum of the aging phenome be quantified and associated with molecular changes. The recent development of automated high throughput image acquisition technology and computational imaging informatics now allow for an unbiased and quantitative aging phenome to be automatically analyzed, and perhaps in the near future for this to take place in a high-throughput manner. C. elegans is the most commonly used model organism for studying the molecular mechanisms of aging. We have, therefore, developed ‘WormFarm’, an integrated microfluidic device for culturing and monitoring C. elegans [6]. WormFarm allows for automated removal of progeny and efficient control of environmental conditions. In addition, we have developed computational image analysis algorithms to automatically analyze the video footage and quantify survival rate (by computationally counting live and dead worms) and other phenotypes, such as worm size, length, width, motility, shape and content density. Using these quantified phenotypes, we have found long-lived age-1(RNAi) and Y82E9BR.3 (RNAi) worms significantly thinner and more mobile than their control groups. In contrast, the short-lived sptf-3(RNAi) worms is exactly the opposite. We have also examined the age dependent changes in worms expressing a mitochondrial-targeted GFP in muscle cells (Pmyo-3::mito::GFP) and have found a continuous decline in fluorescent intensity of GFP with age in worms. These phenotypic data demonstrate yet another significant advantage of the WormFarm culturing and imaging system, that is, automatic, simultaneous and quantitative analysis of multiple aging-related phenotypes [6]. With the knowledge that aging phenomes can be obtained by quantitative imaging analysis, we decided to use a 3D facial image capturing system to collect > 300 3D human facial images, well-distributed across 17–77 years of age. By analyzing these 3D facial images, we generated the first comprehensive atlas of the human facial aging phenome [7]. There are at least two types of phenotypes and/or biomarkers required to measure aging; one to measure the onset of aging and the other to reflect physiological age [2]. After aligning and transforming all vertices' values, we used the information of all 3D geometric vertices to extract aging-related facial changes. In this way, we have quantified facial morphological features and found that eye slopes, mouth width, nose width and the distance between the mouth and nose to be highly associated with age. In addition to predefined shapes and metrics, we have also uncovered other de novo aging related patterns. Most importantly, the 3D facial image data have allowed us to construct a robust age predictor, and based on the deviation between predicted and chronological age, to identify slow and fast agers. The classifications of fast and slow agers have also shown highly significant consistency with physiological indicators for health and age in participants' blood samples, suggesting that age predicted based on 3D facial morphology reliably reflects the biological age of an individual [7]. The aging “phenomics” now enabled by automatic and unbiased detection, quantification and analysis has emerged as a promising new dimension of information for the systems level analysis of aging. Integrating quantitative aging phenomics data with molecular and network changes will not only allow a more comprehensive understanding of aging at a higher level, but will also provide answers to questions that could not be addressed before, and thus generate new insights into the old problem of aging.

Abstract:Aging process potentially drives numerous diseases development1Lopez-Otin C. Blasco M.A. Partridge L. Serrano M. Kroemer G. The hallmarks of aging.Cell. 2013; 153: 1194-1217Summary Full Text Full Text PDF PubMed Scopus (8537) Google Scholar and manifests great heterogeneity among individuals. Thus, monitoring and quantifying aging rate is a fundamental question. Scientists have constructed a variety of aging clocks by machine learning to predict chronological age. The residue between prediction and chronological age followed by regressing out age effects is taken as the biological aging rate. DNA methylation clocks2Horvath S. DNA methylation age of human tissues and cell types.Genome Biol. 2013; 14: R115Crossref PubMed Scopus (3300) Google Scholar are the mostly widely recognized aging clocks. However, bio-sample collection and high-throughput sequencing are often invasive, time-consuming and expensive, which may hinder the translational application of aging clocks. What's more, the relatively small sample size of omics-based aging clocks may affect the model generalization and limits the application of powerful deep learning methods. To tackle the above problems, we have built facial aging clocks by support vector regression (SVR)3Chen W.Y. Qian W. Wu G. et al.Three-dimensional human facial morphologies as robust aging markers.Cell Res. 2015; 25: 574-587Crossref PubMed Scopus (66) Google Scholar and deep convolutional neuron networks (CNN)4Xia X. Chen X.W. Wu G. et al.Three-dimensional facial-image analysis to predict heterogeneity of the human ageing rate and the impact of lifestyle.Nat Metab. 2020; 2: 946-957Crossref PubMed Scopus (23) Google Scholar with MAD as 6.0 and 2.8 years respectively. Aging rate derived from facial aging clock shows significant correlation with transcriptome and identifies molecular mediators of lifestyles.4Xia X. Chen X.W. Wu G. et al.Three-dimensional facial-image analysis to predict heterogeneity of the human ageing rate and the impact of lifestyle.Nat Metab. 2020; 2: 946-957Crossref PubMed Scopus (23) Google Scholar These indicate that facial aging clock can accurately capture the biological age and detect associated molecular patterns. In addition, facial aging clock trained on perceived age creatively avoids the dilemma of taking model error as the true aging rate. Age based on only face appearance estimated by human judges are defined as perceived age and have been shown to be a biological age of the face. As perceived age trained facial aging clocks are actually trained on biological age instead of age, it is not surprising that such a clock is more associated with health indicators than chronological age train models. Such a clock can also reflect how humans judge other people's age. Studies have shown that numerous diseases associate with aging rate. DNA methylation aging clock acceleration is observed in diseases like tumors,5Kling T. Wenger A. Caren H. DNA methylation-based age estimation in pediatric healthy tissues and brain tumors.Aging. 2020; 12: 21037-21056Crossref PubMed Scopus (16) Google Scholar combat PTSD,6Yang R.T. Wu G.W.Y. Verhoeven J.E. et al.A DNA methylation clock associated with age-related illnesses and mortality is accelerated in men with combat PTSD (vol 52, pg 541, 2020).Mol Psychiatry. 2021; 26: 5010Crossref PubMed Scopus (2) Google Scholar and even can reflect mortality risk.7Zheng Y.N. Joyce B.T. Colicino E. et al.Blood epigenetic age may predict cancer incidence and mortality.eBioMedicine. 2016; 5: 68-73Summary Full Text Full Text PDF PubMed Scopus (127) Google Scholar Facial aging clock acceleration is also observed for systemic inflammation, high blood cholesterol and uric acid.3Chen W.Y. Qian W. Wu G. et al.Three-dimensional human facial morphologies as robust aging markers.Cell Res. 2015; 25: 574-587Crossref PubMed Scopus (66) Google Scholar,4Xia X. Chen X.W. Wu G. et al.Three-dimensional facial-image analysis to predict heterogeneity of the human ageing rate and the impact of lifestyle.Nat Metab. 2020; 2: 946-957Crossref PubMed Scopus (23) Google Scholar Convenient data acquisition and processing, and high accuracy confer promising translational potential of facial aging clocks. Single cell RNA-seq clock can be accelerated by 10–20 years in CoVID and lupus, and decelerated by decades in super centenarians.8Zhu H. Chen J. Liu K. et al.Human PBMC scRNA-seq–based aging clocks reveal ribosome to inflammation balance as a single-cell aging hallmark and super longevity.Sci Adv. 2023; 9eabq7599Crossref Scopus (0) Google Scholar Unlike other omics aging clocks, facial aging clock is suitable to clinical scenarios. Facial images taking and biological age prediction takes only several minutes, thus well fit to daily monitoring of aging rate. Except common clinical evaluation markers of diseases, aging rate dynamics of patients as a novel biomarker can provide doctors evaluation of quantitative patients aging status and objective evaluation of therapies. As aging acceleration often reflects disease risk and progression, facial aging clock application may help doctors to optimize drugs and therapies, thus a better prognosis. Accelerated aging may indicate future risk of some diseases, can thus guide people to alter their lifestyles. Another application scenario may be for clinicians to monitor patients health management and relieving clinicians from overloaded routine monitoring. Overall, we envision the non-Invasive and low-cost facial aging clock and even disease risk assessment to be affordable and practical applications under many clinical scenarios, as the workflow shown in Fig. 1. Furthermore, application of facial aging clocks is not restricted to only hospitals. Facial aging clock can be utilized for evaluating daily anti-aging interventions like calorie restriction, metformin and personalized therapies. If this clock is integrated to smart phones in the soon future, daily monitoring of aging is available to all people under any situations. We believe that facial aging clock will prompt people to get accustomed to monitoring aging for optimum health. From the public health perspective, monitoring aging by facial aging clocks can prevent or suspend diseases progression, and ease the economical and clinical pressure. Additionally, some diseases seem to be linked with face morphology. Various diseases may be manifested through alterations in facial features, such as in the case of Down syndrome.9Kruszka P. Porras A.R. Sobering A.K. et al.Down syndrome in diverse populations.Am J Med Genet. 2017; 173: 42-53Crossref PubMed Scopus (59) Google Scholar Facial appearance has been used as a diagnostic tool to facilitate early detection. Deep learning-based analysis of facial images has shown high accuracy in genetic disorders.10Gurovich Y. Hanani Y. Bar O. et al.Identifying facial phenotypes of genetic disorders using deep learning.Nat Med. 2019; 25: 60-64Crossref PubMed Scopus (333) Google Scholar This implicates that pre-diagnosis and risk assessment of certain diseases are clinically practical based on facial images. With the advance in artificial intelligence and cohort data accumulation, we expect that a plethora of facial models will emerge and assist doctors to prevent, diagnose and manage aging and diseases in the near future. However, there exist some limitations on clinical application of facial aging clocks. The effect of geographical differences on faces may introduce confounding factors. Facial appearance exhibits notable variations across different regions worldwide. Recent study of 3D facial images has further revealed significant disparities in facial morphology between East Asian and European populations.11Zhang M. Wu S. Du S. et al.Genetic variants underlying differences in facial morphology in East Asian and European populations.Nat Genet. 2022; 54: 403-411Crossref PubMed Scopus (8) Google Scholar Environmental and lifestyle factors also influence facial development.12Richmond S. Howe L.J. Lewis S. Stergiakouli E. Zhurov A. Facial genetics: a brief overview.Front Genet. 2018; 9: 462Crossref PubMed Scopus (65) Google Scholar Moreover, while deep learning-based facial image analysis holds promise for future clinical applications, it is important to acknowledge that facial changes associated with certain diseases can be subtle and may overlap with normal variations or non-disease-related factors. Finally, many diseases exhibit diverse manifestations. To enhance the accuracy and reliability of disease identification, it is crucial to adopt a comprehensive approach that integrates multiple diagnostic factors, including medical history, physical examinations and medical tests. J.D.J.H conceived and designed the comment. Y.W., K.M., H.Z. and J.D.J.H. wrote the manuscript. The authors declare no competing interests. This work was supported by grants from National Natural Science Foundation of China (NSFC) (Nos. 92049302 and 32088101), China Ministry of Science and Technology (No. 2020YFA0804000), to J.D.J.H.

Aging Phenomics Enabled by Quantitative Imaging Analysis

Phenotypic Age: a novel signature of mortality and morbidity risk

Three-dimensional Human Facial Morphologies As Robust Aging Markers.

Clinical Application of Facial Aging Clocks.

Models for Evaluation of Biological Age Based on Multi-omics

Molecular and phenotypic biomarkers of aging

Three-dimensional facial-image analysis to predict heterogeneity of the human ageing rate and the impact of lifestyle

Molecular and phenotypic biomarkers of aging [version 1; referees: 3 approved]

Integrated analysis of facial microbiome and skin physio-optical properties unveils cutotype-dependent aging effects

High-dimensional Ageome Representations of Biological Aging across Functional Modules

qMAP Enabled Microanatomical Mapping of Human Skin Aging

A new aging measure captures morbidity and mortality risk across diverse subpopulations from NHANES IV: A cohort study

Proteome-Wide Analysis Reveals an Age-Associated Cellular Phenotype of in Situ Aged Human Fibroblasts

Biologically informed machine learning modeling of immune cells to reveal physiological and pathological aging process

Phenotype correlations reveal the relationships of physiological systems underlying human ageing

Thermal facial image analyses reveal quantitative hallmarks of aging and metabolic diseases

705 Skin Signs, Genes and Environmental Factors Associated with Perceived Facial Age - A Phenotype Effectively Inferred by Machine Learning Based Methods

Integration of skin phenome and microbiome reveals the key role of bacteria in human skin aging

647 Genome-wide Scans Identified Genetic Variants Associated with Facial Aging Traits Quantified by Deep Learning Methods

Epigenomic, Transcriptome And Image-Based Biomarkers Of Aging