key: cord-1032129-zg45xb8x authors: Ji, Lixin; Jazwinski, S. Michal; Kim, Sangkyu title: Frailty and Biological Age date: 2021-08-17 journal: Ann Geriatr Med Res DOI: 10.4235/agmr.21.0080 sha: 2dccb79a21f9ced3135df9872ed21916b5b3acdd doc_id: 1032129 cord_uid: zg45xb8x A reliable model of biological age is instrumental in the field of geriatrics and gerontology. This model should account for the heterogeneity and plasticity of aging and also accurately predict aging-related adverse outcomes. Epigenetic age models are based on DNA methylation levels at selected genomic sites and can be significant predictors of mortality and healthy/unhealthy aging. However, the biological function of DNA methylation at selected sites is yet to be determined. Frailty is a syndrome resulting from decreased physiological reserves and resilience. The frailty index is a probability-based extension of the concept of frailty. Defined as the proportion of health deficits, the frailty index quantifies the progression of unhealthy aging. The frailty index is currently the best predictor of mortality. It is associated with various biological factors and provides insight into the biological processes of aging. Investigation of the multi-omics factors associated with the frailty index will provide further insight. Different types of cells age differently. 11, 12) Heterogeneity in tissue aging contributes to the heterogeneity of aging in different organs and body systems. Thus, no single biological marker of aging can faithfully represent organismal aging. Biomarkers of aging have long been sought from age-related changes; however, the relevance of all age-related changes to the biology of aging remains unknown. As some age-related changes may not be relevant to the biology of aging while others constitute the true mechanisms of aging, multiple biomarkers of aging are used to estimate or models the true biological age. Epigenetic age-based models of biological age were first proposed a decade ago, and all such models are based on DNA methylation data. 13) Another type of biological age models that has been gaining momentum is based on frailty. 14) These frailty-based models use health deficits or health-related changes found in aging individuals. We will first discuss models based on epigenetic age before reviewing frailty-based models of biological age. Genomic DNA methylation sites that we are currently focusing on are conveniently called CpG sites as the methylation of cytosine occurs predominantly in CpG dinucleotides ("p" represents the phosphodiester bond linking the nucleotides). A recent study compiled 11 epigenetic age models, 13) one of which was based on 71 CpG sites in the leukocyte genome. 15) These CpGs were selected using a regularized regression method, which showed that their DNA methylation levels were highly correlated with chronological age. The predicted epigenetic age was obtained using actual methylation data in the regression model. Individuals with epigenetic ages greater and less than their calendar ages are categorized as fast aging and slow aging, respectively. Similarly, DNA methylation age and related measures were calculated based on 353 CpGs whose methylation levels were highly correlated with chronological age in multiple tissues. 16) A change in epigenetic age depends on a change in the methylation levels of selected CpGs. As shown in Fig. 1 , a decrease in the overall DNA methylation level of 353 CpGs resulted in decreased epigenetic age. Since the genomic DNA methylation level tends to decrease with increasing chronological age, [17] [18] [19] the epigenetic age is likely to decrease with increasing chronological age among the oldest of older individuals. 20) Recent epigenetic age models combine health-related data with DNA methylation data using complex multi-step mathematical or statistical approaches. "DNAm PhenoAge" is based on 513 CpGs associated with the Gompertz model-based phenotypic age. 21) This phenotypic age incorporates laboratory blood test findings and chronological age. Other models use CpGs as proxy markers for age-related changes or biological events. For example, age-correlated CpGs have been combined with biological data to build a mitotic-like clock model to predict cancer risk. 22) "GrimAge" uses CpGs that are selected by regressing plasma proteins and smoking on 485,000 CpG sites and chronological age. 23) Epigenetic-based models of age are associated with mortality. [24] [25] [26] Slow epigenetic aging is associated with healthy aging-related factors such as healthy diet, physical activity, lifestyle, and low morbidity. 23, 27, 28) This is potentially interesting in that healthy aging is determined largely by non-genetic factors, including various environmental factors and lifestyle, and DNA methylation is an epigenetic interface between genes and non-genetic determinants. Nevertheless, the biological function of the DNA methylation level in CpGs used in the models remains unknown. The human genome contains an estimated 28 million CpGs. 29) A cross-sectional compilation found that approximately 11% of 485,577 CpGs examined were significantly correlated with chronological age and approximately 30% of them also significantly changed longitudinally. 30 This plot was generated using the data from eight autistic subjects available in the MethylationDataExample55 and SampleAnnotatio-nExample55 datasets provided. 16) DNA methylation age was calculated in R using the tutorial provided. The y-axis shows mean DNAm-Age±standard error corresponding to three groups on the x-axis: the autistic subjects ("initial"), after adding 0.1 to the original beta values of 353 CpGs ("+.1_all"), or subtracting 0.1 from the original beta values of 353 CpGs ("-.1_all"). www.e-agmr.org CpGs in the genome is correlated with chronological age, and 0.9 million of them will change longitudinally. With that many age-associated CpGs, one can select any subset of CpGs to predict chronological age or any age-associated changes. One can even select a subset of CpGs in the genome to predict any randomly generated numeric variable ( Table 1 ). Note that in Table 1 , the variable importance scores generated using a random forest method for the randomly generated numeric variable are comparable to those for the frailty index. Similar results were obtained using the regularized regression method. The abundance of CpGs associated with chronological age may also explain why subsets of selected CpGs used in epigenetic age models overlap sparsely (Fig. 2) . Thus, the biological roles of these CpGs, if any, remain unknown. DNA methylation affects important biological phenomena such as gene expression, DNA imprinting, X-chromosome inactivation, and genome stability. 31) Of these, differential gene expression is most relevant to the biology of aging. Changes in the other phenomena tend to have drastic, often pathological, effects, although their contributions to the biology of aging cannot be completely ruled out. The transcriptional regulation of gene expression is primarily mediated by DNA sequence-based genetic factors. DNA methylation also affects gene transcription via epigenetic mechanisms. Following the erasure of DNA methylation patterns of the previous generation, genomic DNA of fertilized eggs is methylated de novo in a tissue-specific manner during implantation and differentia-tion-the erasure seems incomplete, which explains in part why DNA methylation is heritable. DNA methylation may serve as a safety lock to ensure the long-term repression of certain genes. 34) For genes whose 32, 33) expression is responsive to environmental The calendar ages (60-103 years) of 211 participants in the Louisiana Healthy Aging Study 72) were regressed on DNA methylation values (beta) of 10,000 randomly selected CpG sites using a random forest regression method in R. b) The same random forest as in a) , with a frailty index, FI34, instead of calendar age. c) The same random forest as in a) , with a variable of randomly generated numbers instead of calendar age. Scaled permutation-based variable importance scores calculated using the random forest. A Venn diagram of three sets of genes. SH_353 represents genes annotated for 353 CpGs used in the DNA methylation age model. 16) GH_54 represents genes annotated for 71 CpGs used to determine the apparent methylomic aging rate. 15) MS.VG_1000 represents top 1,000 genes whose transcript levels are important in predictive modeling of chronological age. 37 cues, DNA methylation is also a way of non-genetically attuning gene expression to changing milieus. In general, hypermethylation is associated with no or lower gene transcription and hypomethylation with higher transcription, although deviant cases exist. 35, 36) In view of DNA methylation as a modulator of gene transcription, it is interesting to determine whether age-associated DNA methylation of CpGs used in some of the epigenetic age models also show age-associated differential gene transcription. If DNA methylation of CpGs is strongly associated with chronological age, transcription of the genes annotated for CpGs may be strongly associated with chronological age. A recent comprehensive compilation of multi-tissue transcriptomic data 37) excluded genes with low expression to analyze a total of 13,388 genes to determine differentially expressed genes (DEGs) that were important in predictive modeling of chronological age. Among the 54 genes annotated for the 72 CpGs described above, only six ranked in the top 1,000 DEGs associated with chronological age. Similarly, among the 353 genes annotated for the 353 CpGs, only 11 ranked in the top 1,000 DEGs (Fig. 2) . These results indicated that the transcription of most genes annotated in the model CpGs was not highly associated with chronological age or was not important in the predictive modeling of chronological age. In other words, these findings are at odds with the expected biological role of DNA methylation. Thus, no convincing evidence exists to support the mainstream role of DNA methylation in CpG sites used in some of the epigenetic age models. Common among older adults, frailty is often used interchangeably with disability, comorbidity, or even old age. One practicable definition of frailty is age greater than 65 years and dependence on others to perform activities of daily living. 38) Disability is defined as the inability to perform activities of daily living, while frailty along with other morbidities contributes to disability. 39) Frailty may develop from diminished physiological reserve before disability, comorbidity, or other adverse outcomes. 40) Frailty is a multidimensional phenotype that manifests through various signs, symptoms, or other health-related events. 41) To operationalize frailty as a clinical syndrome, the Fried frailty uses five phenotypic criteria: (1) weight loss ( > 10 lb in the previous year); (2) self-reported exhaustion; (3) muscle weakness (the lowest quintile of hand grip strength); (4) slow gait speed (the lowest quintile in walking time per 15 feet); and (5) low physical activity (the lowest quintile in kilocalories expended per week). Individuals meeting three or more of these criteria are considered frail. Related to the Fried frailty, Edmonton frailty uses nine criteria: cognition, general health status, functional independence, social support, medication use, nutrition, mood, continence, and functional performance. 42) The maximum score of Edmonton frailty is 17. Geriatric status classification in clinical settings requires a streamlined operationalization of frailty. Based on activities of daily living, a clinical frailty scale has been proposed to classify older people. 43) A recent version of the Clinical Frailty Scale (CFS) has nine categories, from "very fit" to "terminally ill. " 44) It relies on the level of dependence in daily living activities and clinical judgment. The CFS is simple, inexpensive, and easy to perform but has been criticized as potentially subjective. However, a recent study of the CFS reported a general agreement of participants in diverse medical professions and health occupations in the classification of scenario cases. 45) The CFS was a significant predictor of survival of patients with coronavirus disease (COVID-19). 46) and the outcome of coronary artery bypass grafting. 47) The criteria for frailty scales described above yield semiquantitative measures. By including more health items and using a continuous scoring method, one can build a fully quantitative model of frailty. The frailty index is defined as the proportion of health deficits among a set of surveyed health items. 48, 49) The deficit index uses the same calculation method as the frailty index. 50) The health items include various symptoms, signs, disabilities, and diseases. 48) These data are collected from medical history questionnaires, clinical assessments, laboratory measurements, and other data collection instruments. As the proportion of health deficits that an aging individual carries at the time of survey, raw frailty index scores range from 0 (no deficit in all health items surveyed) to 1 (at least one problem in every health item). Unlike frailty scales, calculation of the frailty index does not require a fixed set of specific health items. When the number of health items is sufficiently large (typically ≥ 20), different frailty indices based on different sets of health items yield comparable properties and performances. However, including health items with predictive power can significantly improve the performance of a frailty index. 51) The end of aging is death, and the oldest of older adults (those aged > 90 years) are most prone to this and other adverse outcomes. The frailty index shows an age-dependent exponential increase with mortality (Fig. 3) . Frailty index scores vary among aged peers and may decline individually while their group average increases over time, accounting for the heterogeneity and plasticity of aging. 52) In the general population, chronological age is a strong www.e-agmr.org predictor of mortality and other time-dependent phenomena. The Klemera-Doubal models of biological age, termed therein Equation 25 (BE) and Equation 34 (BEC), outperformed chronological age and biological age measures based on multiple linear regression or principal component analysis in predicting mortality. 53, 54) This finding makes sense in that both health data and chronological age are used in the derivation of BE and BEC. In particular, BEC explicitly incorporates chronological age as an additional biomarker to improve its performance over BE. After adjusting for age and sex, however, the frailty index outperformed not only DNA methylation age but also both BE and BEC in predicting mortality in the oldest of older adults (Fig. 4) . Thus, the frailty index is the best predictor of age-related mortality. The frailty index quantifies the extent of unhealthy aging. In other words, it is an unhealthy aging index (and subtraction of it from 1 gives a healthy aging index). Therefore, examining biological factors associated with the frailty index can provide insights into the biological mechanisms of unhealthy (or healthy) aging. The resting metabolic rate (RMR) estimates the amount of energy used for the maintenance of body systems under resting conditions. RMR, which comprises 60%-70% of the total daily energy expenditure, decreases with increasing age in the general population. 55) In the oldest of older adults, however, RMR increases as the frailty index increases. 56) One interpretation of this is that a higher RMR is required for this population to maintain homeodynamics as their health deteriorates. Some factors also show sex specifici-ties; for instance, the association of frailty index with body composition (fat and fat-free mass) is specific to female nonagenarians, whereas circulating creatine kinase (CK) is specific to male nonagenarians. Detailed analyses of these sex specificities led to the identification of the associations of UCP2 and UCP3 with frailty index in women 57) and XRCC6 and LASS1 in men. 58) UCP2 and UCP3, which encode uncoupling proteins in the mitochondria, function as metabolite transporters important in energy metabolism. In contrast, XRRC6 and LASS1, which encode the protein Ku70 and ceramide synthase, respectively, are involved in programmed cell death. Muscle damage caused by strenuous physical activity or exercise is a major factor leading to elevated blood CK levels. 59) Thus, one interpretation of the male-specific findings is that an elevated CK level reflects an increased number of damaged muscle cells undergoing programmed death. 58) The genetic basis of longevity is not as substantial as previously estimated, indicating that non-genetic and environmental factors are much more influential in aging. 60) Frailty is associated with various transcriptomic, proteomic, and metabolomic factors. [61] [62] [63] [64] [65] [66] [67] [68] Overall, the number and species of identified omics factors vary across studies and no omics factors have been convincingly replicated in multiple independent studies. Nevertheless, several studies merit further investigation. CK is a major protein associated with the Fried frailty scale. 63) Although it is unknown whether the association is specific to males, the proteomics finding is consistent with the previous association using the frailty index, 58) as described above. In their proteomics study, Sathyan et al. 62, 69) analyzed plasma proteomic profiles using the same technology but different de- pendent variables. Based on the frailty index as a dependent variable, the fatty acid-binding proteins showed the most significant association, while proteins that showed enrichment in the bioinformatic analysis were involved in lipid metabolism, cell-to-cell signaling, and interactions. 62) For chronological age, the most significantly associated proteins were pleiotrophin, WNT1-inducible-signaling pathway protein 2, chordin-like protein 1, transgelin, and R-spondin-1, while the bioinformatically enriched proteins were related to inflammatory response, organismal injury and abnormalities, and cell and organismal survival. 69) Thus, these two studies further demonstrated that the biological factors and pathways associated with the frailty index differed from those associated with chronological age. In other words, the biological interpretations or hypotheses to test depend on which dependent variable is used in the association analysis. Therefore, it is critical to use a reliable and accurate metric of biological age. While single omics analysis is useful, aging is a complex phenomenon that occurs across interconnected, yet heterogeneous, biological systems. The simultaneous analysis of multiple omics data obtained from the same specimens can greatly enhance the accuracy of data analysis and interpretation. 70, 71) Therefore, integrative multi-omics using a reliable measure of biological age may be a more useful approach for elucidating the biology of aging. One fruitful approach to understanding the biology of aging is to establish a reliable measure of biological age and study its associated biological factors. Epigenetic age models, which are based on DNA methylation data, are promising in that lower epigenetic ages are associated with healthy aging and DNA methylation is considered an interface between the genome and environment. However, the biological significance and function of epigenetic model ages and model CpGs are yet to be elucidated. Frailty reflects the deteriorating physiological processes of aging. The extent of frailty is fully quantifiable by calculating the proportion of various health deficits. The frailty index is the best predictor of mortality, especially among older patients. Thus, the frailty index is a simple . BEC28 is the Klemera-Doubal biological age estimate using 28 health items. 51, 53) FI28 is the frailty index based on the same 28 health items. 51,53) *p≤0.05, **p≤0.01, ***p≤0.001. mathematical and easily rationalized method of quantifying biological age. Various biological factors are associated with frailty index, providing valuable insights into various aspects of aging. Simultaneous analysis of multi-omics datasets using the frailty index may be a fruitful approach to understanding the biology of aging. Longevity, senescence, and the genome Biologic versus chronologic age Interventions to slow aging in humans: are we ready? Caloric restriction delays disease onset and mortality in rhesus monkeys Animal models of aging research: implications for human aging and age-related diseases Extrachromosomal rDNA circles: a cause of aging in yeast daf-2, an insulin receptor-like gene that regulates longevity and diapause in Caenorhabditis elegans Extended life-span and stress resistance in the Drosophila mutant methuselah A single combination gene therapy treats multiple age-related diseases The hallmarks of aging Age mosaicism across multiple scales in adult tissues A single-cell transcriptomic atlas characterizes ageing tissues in the mouse Underlying features of epigenetic aging clocks in vivo and in vitro A comparison of two approaches to measuring frailty in elderly people Genome-wide methylation profiles reveal quantitative views of human aging rates DNA methylation age of human tissues and cell types Genomic 5-methyldeoxycytidine decreases with age Decline in genomic DNA methylation through aging in a cohort of elderly subjects Distinct DNA methylomes of newborns and centenarians Decreased epigenetic age of PBMCs from Italian semi-supercentenarians and their offspring An epigenetic biomarker of aging for lifespan and healthspan Correlation of an epigenetic mitotic clock with cancer risk DNA methylation GrimAge strongly predicts lifespan and healthspan DNA methylation age of blood predicts all-cause mortality in later life DNA methylation levels at individual age-associated CpG sites can be indicative for life expectancy DNA methylation signatures in peripheral blood strongly predict all-cause mortality Epigenetic age acceleration predicts cancer, cardiovascular, and all-cause mortality in a German case cohort Epigenetic clock analysis of diet, exercise, education, and lifestyle factors DNA methylation in human epigenomes depends on local topology of CpG sites Change in epigenome-wide DNA methylation over 9 years and subsequent mortality: results from the In-CHIANTI study The diverse roles of DNA methylation in mammalian development and disease Heritable DNA methylation in CD4+ cells among complex families displays genetic and non-genetic effects DNA methylation: an epigenetic mark of cellular memory G9a-mediated irreversible epigenetic inactivation of Oct-3/4 during early embryogenesis Paradoxical role of DNA methylation in activation of FoxA2 gene expression during endoderm development Principles of DNA methylation and their implications for biology and medicine Modeling the human aging transcriptome across tissues, health status, and sex Who are the frail elderly? Trajectories of disability in the last year of life Preventing frail health Frailty in older adults: evidence for a phenotype Validity and reliability of the Edmonton Frail Scale A brief clinical instrument to classify frailty in elderly people Using the clinical frailty scale in allocating scarce health care resources The clinical frailty scale: do staff agree? The effect of frailty on survival in patients with COVID-19 (COPE): a multicentre, European, observational cohort study Clinical frailty scale and outcome after coronary artery bypass grafting A standard procedure for creating a frailty index Accumulation of deficits as a proxy measure of aging Cumulative deficits and physiological indices as predictors of mortality and long life Feature selection algorithms enhance the accuracy of frailty indexes as measures of biological age Quantitative measures of healthy aging and biological age A new approach to the concept and computation of biological age Modeling the rate of senescence: can estimated biological age predict mortality more accurately than chronological www.e-agmr.org age? Invited review: aging and energy balance An elevation of resting metabolic rate with declining health in nonagenarians may be associated with decreased muscle mass and function in women and men, respectively Single nucleotide polymorphisms linked to mitochondrial uncoupling protein genes UCP2 and UCP3 affect mitochondrial metabolism and healthy aging in female nonagenarians Programmed cell death genes are linked to elevated creatine kinase levels in unhealthy male nonagenarians Creatine-kinase-and exercise-related muscle damage implications for muscle performance and recovery Estimates of the heritability of human longevity are substantially inflated due to assortative mating Proteomic screening of glycoproteins in human plasma for frailty biomarkers Plasma proteomic profile of frailty Characterization of the plasma proteomic profile of frailty phenotype A robust machine learning framework to identify signatures for frailty: a nested case-control study in four aging European cohorts Plasma protein biomarkers of the geriatric syndrome of frailty Identification of frailty-associated genes by coordination analysis of gene expression Frailty in middle age is associated with frailty status and race-specific changes to the transcriptome Frailty markers comprise blood metabolites involved in antioxidation, cognition, and mobility Plasma proteomic profile of age, health span, and all-cause mortality in older adults Transcriptomic and metabolomics data integration Machine learning and integrative analysis of biomedical big data The frailty index outperforms DNA methylation age and its derivatives as an indicator of biological age Association of healthy aging with parental longevity We thank the people of Louisiana for their participation in our study. Lixin Ji participated in the research and manuscript preparation as a recipient of the Acquiring Skills and Practice in Research Excellence (ASPIRE) fellowship at Tulane School of Medicine. The researchers claim no conflicts of interest.