Trajectories of aging: How systems biology in yeast can illuminate mechanisms of personalized aging

Abstract

All organisms age, but the extent to which all organisms age the same way remains a fundamental unanswered question in biology. Across species, it is now clear that at least some aspects of aging are highly conserved and are perhaps universal, but other mechanisms of aging are private to individual species or sets of closely related species. Within the same species, however, it has generally been assumed that the molecular mechanisms of aging are largely invariant from one individual to the next. With the development of new tools for studying aging at the individual cell level in budding yeast, recent data has called this assumption into question. There is emerging evidence that individual yeast mother cells may undergo fundamentally different trajectories of aging. Individual trajectories of aging are difficult to study by traditional population level assays, but through the application of systems biology approaches combined with novel microfluidic technologies, it is now possible to observe and study these phenomena in real time. Understanding the spectrum of mechanisms that determine how different individuals age is a necessary step toward the goal of personalized geroscience, where healthy longevity is optimized for each individual.

1. Introduction

Aging is a biological certainty. Every person who has lived beyond reproductive maturity has experienced the biological aging process. The declines in function and phenotypic changes that go along with aging are easily recognizable: wrinkled skin, graying hair, stooped posture, impaired vision and hearing, loss of muscle mass, and increased risk for a variety of pathological conditions. Many of these phenotypic changes with age are also strikingly conserved across evolutionary distance, easily seen in other mammals such as dogs or mice, and in some cases in invertebrate organisms such as fruit flies or nematode worms [1]. This conservation is particularly obvious when considering the increase in risk of mortality with age, which while occurring on chronologically different scales, displays strikingly similar mathematical properties across species (Figure 1).

Figure 1: Population mortality shows similar patterns across a broad evolutionary distance.

A) Survival curves for four of the most commonly used model organisms for aging research: budding yeast, C. elegans, fruit flies and mice. For most organisms, aging is measured as a function of time, but for budding yeast it is measured as a function of the number of replications a cell has undergone. B) The mortality rate for organisms begins low, but increases in a log-linear fashion that can be modeled by the Gompertz–Makeham law of mortality.

Aging is also an individual phenomenon, with numerous incompletely penetrant phenotypes at the population level. Not every person ages at the same rate or in exactly the same way. Some people’s hair will become gray in their thirties, others in their sixties, and others will lose it altogether. Some people will experience age-associated co-morbidities, such as cognitive decline and heart disease, in their 40s and 50s, while others will live disease-free into their 70s and 80s. Indeed, it could be argued that every person is experiencing their own trajectory of aging, which while having similarity in some respects to the population as a whole, is also unique to them individually.

This concept of “trajectories of aging” raises the interesting question as to whether the biological mechanisms of aging are fundamentally the same or distinct in different individuals (Figure 2). Is it the case that biological aging is invariant at the cellular and molecular levels, but due to genetic, environmental, and stochastic variation, the phenotypic manifestations of aging become personalized? Or could it be that the aging can take on distinctly different features even at the single cell level. Certainly, to some extent, the molecular aging process of a neuron is likely to be different from the aging process of a skin fibroblast or a muscle stem cell. But, more fundamentally, the question is whether – when we look across different individuals of the same species – is the cellular, tissue, and whole organism aging process essentially the same or different. Even within a population of isogenic organisms that experience identical environments, some organisms undergo rapid aging and death, while others grow old slowly and live far longer. What is it that causes some organisms to age faster than others, and do they merely experience the same process on fast-forward, or do they follow a different trajectory completely?

Figure2:

Variation in traits and population aging trajectories. A) Even isogenic cells have variations in phenotypes based on protein fluctuations or random noise. Variation in traits can be assumed to be normally distributed with a mean μ and standard deviation σ. B) Isogenic cells could follow a single, unified trajectory of aging where the variation in lifespan or traits is a function of the normally distributed population. C) Alternatively, cells could follow divergent trajectories as they age where each trajectory could be characterized by a dominant failure mechanism that would result in different phenotypes. Each trajectory would have a unique μ and σ, and thus still have variation within the population following a specific trajectory.

We have recently suggested that the dynamics of single cell aging within an isogenic population could be explained by two competing models: a “Cell Cycle” model where cells move through the same physiological states but at different rates, and a “Waddington” model where cells have the potential to follow different paths or succumb to different physiological failures [2]. Either of these models could explain why different individuals appear to age at different rates, or at least reach senescence at different times; however, they differ substantially in their mechanistic implications for biological aging and the goal of intervening in the aging process to increase lifespan and healthspan. Here we will discuss these implications and how the combination of large-scale, systems biology approaches with new microfluidic technologies are beginning to inform the field about how cellular aging works at both the individual and population levels.

2. Yeast as a model to study aging

The budding yeast Saccharomyces cerevisiae provides a unique opportunity to study individual trajectories of aging at the cellular and organismal level simultaneously [2]. Aging in yeast can be studied by measuring the replicative lifespan (RLS) of mother cells, which is defined as the number of daughters produced by a mother cell prior to irreversible cell cycle arrest [3, 4]. Traditionally, RLS has been assessed by manual micromanipulation of daughter cells away from mother cells on the surface of a nutrient agar substrate while counting the number of daughter cells given off by each mother cell [5, 6]. Because cell division in budding yeast is morphologically asymmetric, it is relatively trivial to discriminate the mother and daughter cells by visual inspection.

The RLS of nearly every viable single gene deletion mutant in yeast has been quantified, yielding numerous insights into the genetic and molecular mechanisms of replicative aging [7, 8], at least some of which are conserved in multicellular eukaryotes [9]. Interestingly, despite the identification of more than 200 genes that regulate lifespan in yeast, a unifying model for yeast aging has yet to emerge. Instead, multiple different causes of yeast aging have been identified [10–13]. These include genomic instability at the ribosomal DNA [14–17], mitochondrial dysfunction [18, 19], accumulation of oxidatively damaged and misfolded proteins within the mother cell [20], decreased histone abundance and loss of chromatin at the telomeres [21–23], and a failure to maintain pH homeostasis in the vacuole [24, 25].

While it has been proposed that at least some of the age-related failure points in yeast are interconnected [26–29], genetic interaction data suggest the existence of multiple distinct longevity pathways determining RLS [10, 11]. Such evidence arises primarily from combining two or more lifespan altering mutations or interventions and observing their interaction in terms of replicative longevity. For example, if two different mutations each extend lifespan and combining them results in an even greater lifespan extension, this can be interpreted as the two corresponding genes acting in different longevity pathways, whereas if the combination results in no further lifespan extension, this could be interpreted as the two genes acting in the same longevity pathway. Of course, such interpretations are far from definitive, but serve as a useful starting point for building genetic models of lifespan determination, and when combined with complementary molecular data can provide a framework for understanding mechanisms of aging.

Although there have been some controversies [30–33], the combined literature generally points to the existence of multiple genetically distinct longevity pathways in yeast [7]. Among these, some of the best studied include the rDNA/Sir2 pathway [34], the calorie restriction/mTOR/mRNA translation pathway [35], the proteasome pathway [36], and the vacuole/mitochondria pathway [25]. While it remains to be determined whether, and to what extent, these pathways converge on similar downstream targets, new findings suggest the possibility that these genetic relationships, which have all been characterized at the population level, may in fact reflect fundamentally distinct aging trajectories at the individual cell level (discussed further below).

3. Population-level systems biology approaches to understanding aging in yeast

Large-scale systems approaches have been particularly fruitful in recent years to provide an understanding of the biology of aging yeast cells, even though they have focused thus far primarily on population measures. This work can largely be separated into genetic perturbations, or population-wide analysis of the transcriptome and proteome.

3.1. Genetic approaches

The development of tools in both yeast and C. elegans for large-scale gene deletion or gene knockdown in the late 1990s provided opportunities for researchers to perform the first large-scale, unbiased genetic screens for longevity [9]. In yeast, this was made possible through the construction of the yeast ORF deletion collections, a set of libraries each containing several thousand single gene deletions in either MATa or MATα haploid mating type, as homozygous diploid deletions, or as heterozygous diploid deletions [37–39].

To date, comprehensive analysis of RLS has been reported only for the haploid deletion collections. This large project took more than a decade to complete, as it involved manual microdissection of daughter cells away from mother cells for nearly 5,000 unique single gene deletion collections. The obvious power of such an approach is that it is both comprehensive and unbiased, in that it does not rely on pre-existing knowledge or hypotheses to predict which genes modify longevity. Several important discoveries resulted from this project [6, 8, 22, 23, 35, 36, 40–48], and nearly 200 lifespan extending mutations were identified [7]. Of particular note, the combined genome-wide replicative longevity screen in yeast and those being carried out in parallel in C. elegans [49–52], allowed for the first quantitative demonstration of conservation of genetic control of longevity between these two evolutionarily divergent organisms [53]. Network analysis of the comprehensive replicative longevity map from analysis of single gene deletions clearly indicates functional clustering of longevity genes [7]. Overrepresented functional categories associated with longevity include mitochondrial translation, TCA cycle, the SAGA transcriptional complex, the proteasome, the cytoplasmic ribosome, and mannosyltransferase enzymes.

While this approach has led to many discoveries, there are some important limitations, and it has been difficult to pinpoint specific mechanisms of aging solely from identifying long-lived gene deletions. By their nature, long-lived deletion mutants represent only a small fraction of potential longevity genes. For example, there are likely to be genes for which reducing expression by 50% increases lifespan but complete deletion is detrimental, and these would not be identified from the deletion set screen. Other genes are known to extend lifespan when overexpressed but not when deleted. Also, while it is relatively easy to determine how a gene deletion is modifying cellular function in young cells (e.g. log phase culture), it is often difficult to know how these gene deletions modify the physiology of aged cells without performing laborious purification procedures (discussed further below). Finally, these studies have all focused on population metrics, so it remains unclear whether lifespan extension at the population level equates to delayed aging in every cell or in only a subset of cells.

3.2. The Omics of Aged Cells

The development of methods to quantify the transcriptome, the proteome and metabolome of cells has been a significant boon to many fields, aging included. As mentioned above, it is relatively trivial to perform -omics type analyses on young cells, since any log phase population will be dominated by young cells and will contain exceedingly few old mother cells. However, obtaining pure populations of sufficiently aged mother cells in sufficient quantity to perform similar experiments has proven quite challenging. In order to obtain sufficient numbers of cells (frequently on the order of 10⁶-10⁸ cells depending on method) these approaches have relied on the purification of old mother cells from within a mixed population of young and old cells. Because yeast cells are typically grown in batch culture, new progeny rapidly outnumber the previously existing cells, and the fraction mother cells of any given replicative age, n, is roughly 1/2ⁿ. To tackle this problem, several methods have been developed that allow enrichment of aged mothers on the scale that is necessary to perform large-scale omics measurements.

The isolation of aged mother cells has primarily relied on two approaches, enrichment by magnetic beads [54] and the Mother Enrichment Program [55]. With the formation of a new bud, the cell wall of the daughter cell is generated de novo, but the cell wall of the mother remains the same. The first approach to mother cell purification took advantage of this fact by doing a single timepoint labelling of batch cultured cells with biotin. When these cells were then placed into normal media, the new younger cells which quickly formed the bulk of the population lacked the surface biotin labelling. Using streptavidin coated magnetic beads, the older mother cells which still retained the biotin-labeled wall could then be selectively removed from the population. The Mother Enrichment Program uses a specific synthetic circuit that is exclusively activated in virgin daughter cells that can be used to kill newborn daughters when they are exposed to estradiol. By preventing new daughters from dividing, the growth rate becomes linear instead of exponential which allows cells to grow to significantly older ages without consuming all media or nutrients. This overcomes one significant hurdle, and as a result many recent papers rely on the Mother Enrichment Program.

Early efforts to quantify alterations to the transcriptome during aging used the biotin enrichment approach and relied primarily on microarrays [56–60]. These approaches identified an age-associated shift in gene transcription resembling the starvation response elicited by cells in batch culture when glucose becomes limited (termed the diauxic shift), and reductions in ribosome biogenesis that were later confirmed by both proteomic and metabolomic methods. These papers relied on repeated purification of aged cells using the biotin approach, and one concern is that the separation of cells could bias the results, as the purification process itself is stressful. Furthermore, as the cells are allowed to grow in flasks for anywhere between 7–10 doublings, the concentration of nutrients will change during this time and might explain some of the observed transcriptional changes.

To overcome limitations with standard old cell purification approaches, bulk culture methods were developed where biotinylated cells are held in place using a magnet, and fresh media is constantly flowed over them washing away newborn daughters [61]. Using this approach mother cells are continuously exposed to fresh media, and concerns about environmental changes or stresses from purification are reduced. This method was used to perform an age-course analysis where samples were obtained every twelve hours for three days, and subjected to both next-gen RNA sequencing and shotgun proteomics [61]. Major findings included a general increase in proteins related to stress response and glycolysis and gluconeogenesis. Interestingly, they also identified an increasing decoupling of the proteome from the transcriptome likely related to changes in ribosome biogenesis. This decoupling of mRNA and protein also resulted in reduced stoichiometry in multi-subunit complexes which may explain why the vacuolar ATPase was previously identified as a failure point during replicative aging [25], and suggests that other complexes could share a similar fate. Leveraging this methodology and experimental setup, a study investigating the metabolome was just published, finding a similar shift in metabolic flux away from fermentation to respiration, along with the associated metabolic intermediates and redox cofactors [62].

Recently, a miniature-chemostat method has been described that relied on similar principles to enable larger, parallelized experiments [63]. To accomplish this, they modified miniature-chemostat to incorporate magnets to hold biotinylated mother cells in place while flowing over fresh media and removing daughter cells. The authors focused on chromatin and transcriptome changes (ATAC-seq and RNA-seq) and examined age-related changes in a short-lived-mutant (sir2∆) and two long-lived mutants (fob1∆, and ubr2∆) in addition to wild-type cells. Critically, when using populations of old cells, some old mothers will be dead, while others remain alive. In order to correct for this problem, the authors added propidium monoazide which can only enter dead cells and binds to DNA preventing inclusion of dead cells in the ATAC-seq data. Using this approach, shifts in nucleosome occupancy with age were detected [63], but not the dramatic reduction in total occupancy as has been previously reported [64].

4. Application of single-cell technologies to aging yeast populations

Although most studies to date have focused on population level changes in gene, protein, or metabolite levels with age, newer single-cell technologies have the potential to provide information on trajectories of aging among individual cells. For example, single-cell or single-nuclei RNA-sequencing could be combined with the cell-sorting methods above to understand how variance in gene expression between cells changes with age. While such approaches are beginning to be used to study changes in cell composition of tissues during aging in multicellular eukaryotes [65–68], to date they have not been applied in yeast to understand individual variation during aging.

If cellular aging in yeast were a mechanistically unitary process, we would predict that the age-associated changes in gene-expression (or protein or metabolite) seen at the population level should be largely recapitulated at the individual-cell level. While some level of stochastic variation would be expected across individuals, the core set of changes caused by the biological aging process should be shared among all individuals and, importantly, should be reproducible across experiments performed under identical conditions (e.g. temperature, media composition, etc.). In contrast, if different sub-populations of cells experience distinct mechanisms of aging then we would predict that distinct age-associated changes would be seen in sub-populations of cells. Specific gene expression (or protein or metabolite) signatures should be apparent for each distinct trajectory of aging, which would always cluster together within the all individuals experiencing that trajectory.

As with the lifespan experiments performed using microdissection (described above), all of the omics studies described here on purified aged mother cells have relied on population-based bulk measurements. While these data can provide informative discoveries about changes in the average abundance of a gene, protein, epigenetic mark, or metabolite with age, they do not inform us about the penetrance of those changes at the individual level.

4.1. Microfluidics allow individual cells to be monitored throughout life

Recently, the adaptation of microfluidic technologies to the study of aging in yeast has allowed for analyses of phenotypes associated with aging at both the population and single-cell levels simultaneously [68]. Several groups have developed microfluidic devices for studying aging in yeast, most of which involve trapping of mother cells within features that leverage the asymmetry between mother and daughter cells [69–76]. As mother cells are larger than daughters, traps have been designed that retain mothers but allow daughters to be removed by flow. Initial designs primarily trapped mothers using a vertical restriction and pinning mothers beneath large columns. The gap between the columns and the floor was sufficient to trap mother cells, but the daughters were too small and were thus removed. More recently, however, designs have largely tended to follow an approach where mothers are pushed by fluid flow up against pillars with small openings that allow buds to be pushed through. Because these small individual traps offer both higher retention as well as increased density of cells within a single field-of-view [77, 78], they have become increasingly popular. One concern with all of these devices, however, is that retention is never perfect, and if cells are lost from traps in a non-random fashion (for example smaller cells may be preferentially lost or retained during aging), losses have the potential to bias experimental results. As a result of this concern, we continue to believe strongly that results from microfluidic systems showing lifespan extension or shortening should be confirmed using traditional microdissection approaches. However, by providing a method to observe multiple fluorescent proteins and physiological parameters in individual cells, microfluidic devices have significant potential to inform our understanding of lifespan heterogeneity and the aging trajectories that cells follow within a population.

Importantly, microfluidic devices also provide the ability to observe physiological changes during replicative aging with much greater precision than has been previously possible, due to the fact that it is only feasible to collect bulk population-level data at intervals of multiple hours. Such precision is important to better understand the kinetics of age-associated changes. For example, the irreversible loss of mitochondrial membrane potential is thought of as a “hallmark of aging” in yeast due to the high levels of petite daughter cells from aged mothers [79–81]. However, observation of cells aged in a microfluidic device showed that the risk of losing mitochondrial membrane potential may be independent of age, with the irreversibility of this change driving the original association with aging [82]. Although the majority of microfluidic devices have focused on replicative as opposed to chronological aging[11], it would be straightforward and useful to couple these studies with chronological aging studies as well.

Through the use of these microfluidic devices, we and others have identified distinct subpopulations of aging cells that show differential or incomplete penetrance of several different physiological traits[83]. Because these changes only affect a subpopulation of cells, measurements which act on the population such as measurements of protein levels for example, will obscure these trends. For example, if some cells activate a specific stress response during aging, while others don’t, population measurements will show a consistent trend towards increasing stress response (Figure 3A). At the single cell level, however, some cells may fail to ever activate this stress response while others will activate it strongly and consistently (Figure 3B). When individual many cells are viewed together, cells could even be segregated by the time of activation (Figure 3C). Interventions, either genetic or environmental, which appear to shift the mean response, could be reducing the fraction of cells which activate the stress response at all (Figure 4A,B) or modulating the extent to which the stress response is activated in that subpopulation (Figure 4B,C). Understanding how genes and environment influence penetrance of specific trajectories is crucial to our understanding of aging and the ability to design personalized strategies to delay age-related declines.

Figure 3: Population level data can hide divergent paths at the single cell level.

A) Data for a hypothetical stress response phenotype that increases with age on the population level. B) Although the stress response increases uniformly at the population level, it is possible that individual cells follow different paths where some cells activate the stress response with age and others don’t. C, D) Kymographs showing individual cells, and the activation of the stress response.

C) The population data from A could be explained by complete penetrance where all cells only moderately activate the stress response. D) The population data from A could also be explained by divergent trajectories where only a sub-population of cells activates the stress response, but they activate it very strongly.

Figure 4: Interventions (environmental or genetic) that affect population aging can act similarly on all cells or only affect the age-related phenotype in a subset of cells.

A) Schematic showing an increasing stress response with age that only affects a sub-population of wild-type cells. B) An intervention that lowers the population mean response (top) by a reduction in the penetrance of the specific trajectory (middle). Cells that proceed down the trajectory activate the stress response just as strongly as wild-type cells C) An intervention that lowers the population mean response (top) by a reduction in the activation within a single trajectory (middle). The fraction of cells that proceed down a specific trajectory is unchanged, but cells that proceed down the trajectory benefit by a reduced stress response.

4.2. Single cell analysis reveals evidence for early bifurcation of aging trajectories

Recent work has provided compelling evidence for alternative trajectories of single cell aging in a microfluidic device [83]. Using a microfluidic system and brightfield imaging of cells during their entire RLS, Jin and colleagues identified states that cells can move through as they age [83]. These states are based on combinations of the cell cycle duration and the daughter cell morphology. The authors report that, in their device, cells start with normal budding durations with daughter cells nearly as large as the mother cell, and as they age mother cells either start to produce daughters with an extended, elliptical morphology (State 1), or daughters that are significantly smaller than the mother cells but that remain circular (State 2). Using these morphological and cell cycle times to classify mother cells as they age, the authors were able to identify trajectories unique to each individual cell and predict the transitions through states and to death.

The trajectories identified by Jin et al. are largely mutually exclusive, and once cells “differentiate” during aging, they are unlikely to transition from elliptical to circular daughters, or vice versa. In this sense, it appears that aging, at least under the conditions employed here, may be similar to a Waddington landscape, where cells moving further down the hill become trapped in deeper and deeper valleys. Once cells begin progressing down one trajectory it is possible from them to jump from one trajectory of death to another, but the odds of such a transition are extremely low. Although the morphological changes reported by Jin et al. have been observed in other microfluidic systems[70, 71, 84], it is still important to determine whether these states are also seen in standard RLS assays, such as microdissection, and to determine whether specific microfluidic designs affect the penetrance of specific states.

Using this conceptual framework and the states they identified, the authors investigated how several well studied genetic and environmental perturbations affected the single cell progression through aging. Intriguingly they were able to identify a mutation (sir2∆) that affected one aging trajectory more than the other as well as a mutation (sgf73∆) which lowered the hills of the Waddington landscape, and made transitions from one trajectory to another more likely. They similarly used two different environmental perturbations known to affect aging, calorie restriction [85] and supplementation with nicotinamide [86]. Calorie restriction, accomplished by reducing the glucose present in the culture media, not only reduced the fraction of cells that followed the circular daughter trajectory, but also only increased the lifespan of cells with elliptical daughters (State 1). The authors suggest that the elliptical daughter state is likely to be related to increased rDNA circles or reduced rDNA silencing during aging [83, 84], both of which have been previously implicated in yeast replicative aging through the Sir2-dependent longevity pathway [14–16].

These exciting data suggesting the existence of at least two distinct trajectories of aging in yeast may provide an explanation for a previously controversial and surprising interaction between Sir2 and calorie restriction. It was initially proposed that Sir2 and calorie restriction act in the same genetic pathway [85], but these early studies were performed in two different strain backgrounds, so a formal genetic interaction could not be discerned [87]. When both Sir2 overexpression and calorie restriction were performed in a third strain background together, the data were more consistent with the model that Sir2 and calorie restriction acted in different genetic pathways [88]. The work of Jin et al. [83] suggest the possibility that each of these “pathways” really represents distinct trajectories of aging and that manipulations that increase lifespan through these pathways could act by changing either the probability that cells will favor one trajectory over the others or by altering their likelihood to transition from one state to another. Although speculative, this idea is amenable to experimental testing, and it will be of interest to determine how these and other known longevity interventions influence the passage of cells through different states as they age.

5. Conclusion

Advances in the technologies used to study aging are allowing investigators to probe the mechanisms of aging at both the population and single-cell levels simultaneously. These tools, for the first time, provide an opportunity to test the hypothesis that individuals experience different trajectories of aging. If correct, this has important implications both for the understanding the basic mechanisms of aging and for the development of interventions to target the molecular mechanisms of aging. For example, an intervention that delays or prevents mechanisms associated with one trajectory of aging might only increase health and lifespan in those individuals who experience that trajectory. Understanding these phenomena, and how genetic and environmental factors influence which individuals experience different trajectories of aging, will be critical for the future of aging research.

The biology of aging has been described in terms of nine “Hallmarks of Aging” [89]. These hallmarks are genomic instability, telomere attrition, epigenetic alterations, loss of proteostasis, deregulated nutrient sensing, mitochondrial dysfunction, cellular senescence, stem cell exhaustion, and altered intercellular communication. These hallmarks of aging represent much of what we currently understand about evolutionarily conserved aging mechanisms and provide an important framework for conceptualizing aging biology. Importantly, the interaction between distinct hallmarks of aging can also fit within the framework of distinct trajectories of aging. In a particularly simplistic model, each hallmark could underlie a distinct trajectory of aging; however, abundant experimental evidence indicates that these hallmarks interact with each in complicated ways to reinforce age-related declines in function. For example, loss of proteostasis may limit functional lifespan in some cells, but the transition into this state could be made more likely through increased mitochondrial dysfunction, or vice versa.

Budding yeast provide a particularly powerful model system in which to probe these questions. While single-cell methodologies are beginning to be applied to the study of aging in invertebrate and mammalian systems, it remains difficult to parse out the effects of tissue heterogeneity and cell non-autonomous factors. Yeast are unique in that aging is identical at the cellular and organismal levels. Therefore, the individual cells that undergo different trajectories of aging are also the individuals within the population. Thus, yeast have the potential to provide a powerful framework for defining the importance of individual trajectories of aging, their underlying molecular mechanisms, and also for modeling how these mechanisms interact to determine overall function and fitness throughout life. Such an understanding of aging trajectories is a necessary first step toward personalized geroscience, where we begin to classify each individual’s biological age based off of well-defined biomarkers. Only then will it possible to optimize interventions such that each individual can achieve their maximum healthspan.