While without clear consensus (Cohen et al., 2020), ageing is generally defined as “a decline in function of the organism during adulthood” (López-Otín et al., 2023), “characterised by an [age-dependent] progressive loss of physiological integrity, leading to impaired function and increased vulnerability to death” (Lopez-Otin et al., 2013). This definition largely reflects the way we, as humans, experience ageing. However, there are limitations to its validity, especially as it implies that the process of ageing is only somewhat distinguishable from the mere passing of time (i.e., as time passes, an organism is ageing).

By placing time as central in ageing, the traditional understanding and modelling of this phenomenon does not address how the rate of ageing varies amongst individuals. Indeed, human-like ageing is most apparent demographically (when individuality/individual variance is not exactly captured well) and is seen as a force of mortality that steadily and predictably increases over a given lifespan, progressing exponentially with age (Gompertz, 1825). Consequently, this demographic perception of ageing strongly shapes our mechanistic understanding of the involved processes (i.e., the mechanisms of ageing are a phenomenon of chronology). Chronological age, while considered an “imperfect surrogate measure of the ageing process” (Horvath and Raj, 2018), is nevertheless still used as the main parameter of ageing, largely due to its simplicity. However, the need for non-invasive, conserved, quick-to-measure, and time-independent biomarkers of ageing has been recognized now for more than 30 years (Baker and Sprott, 1988; Warner, 2004). Such biomarkers are defined as “biological parameters of an organism that either alone or in some multivariate composite will, in the absence of disease, better predict functional capability at some late age than will chronological age” (Baker and Sprott, 1988). It is well observed, most notably in isogenic populations and highly controlled environments, that individuals do not age at the same rate (Zhang et al., 2016). This implies a difference between the chronological age of an organism - the time elapsed from birth—and its biological age—the age in terms of its physiology and associated instant risk of death.

Nine hallmarks of ageing—recently expanded to twelve—are broadly used by the ageing research community as molecular markers associated with biological ageing (Lopez-Otin et al., 2013; López-Otín et al., 2023). These twelve hallmarks were further divided into three categories: primary, antagonistic, and integrative. The primary hallmarks are processes affected by the accumulation of damage over time and, “unambiguously contribute to the ageing process”; this includes: genetic instability, telomere attrition, epigenetic alterations, loss of proteostasis, and disabled macroautophagy. Antagonistic hallmarks group damage-response processes that have beneficial effects when active at low levels, but damage the organism if continuously present (i.e., mitochondrial dysfunction, deregulated nutrient sensing, cellular senescence). Finally, the integrative hallmarks (stem cell exhaustion, altered intracellular communication, chronic inflammation, and dysbiosis) arise at a systemic level through the non-reversible combination of the primary and antagonistic markers. The hallmarks are evolutionarily conserved (Lemoine, 2021), alter with increasing age, and when intervened upon, the lifespan of an individual increases (Lopez-Otin et al., 2013). In 2018, a similar review of the literature was used to define six transcriptional hallmarks of ageing (Frenk and Houseley, 2018). While underlining the difficulty in defining a “consensus signature” across species and tissues, the six hallmarks defined here are: downregulation of mitochondrial proteins, downregulation of ribosomes, reduction in growth factor signalling, dysregulation of gene expression, dysregulation of immune genes and stress and DNA damage (for the last, they specify that data does not give way to a clear consensus, as in some cases the expression of DNA damage genes is above average, and in others, it is below). Notably, it has been recently shown that at least some of these six hallmarks display asynchronous and non-linear progress across tissues and organs in mice (Schaum et al., 2020). This observation only further provokes a sense of doubt if we are to continue conceptualising ageing as linear and coupled to the chronological life of the organism. To account for inter-individual variability during ageing, the concept of frailty was introduced in the late 20th century as a previously unobserved individual modulator of the force of mortality (Vaupel et al., 1979). Frailty was first adopted by clinical settings and subsequently, experimental biology (Faya-Robles, 2018). In these contexts, the concept was initially used to identify multiple frailty phenotypes; these phenotypes later developed into frailty indexes. The indexes are based on sets of biological/physiological/behavioural parameters and these parameters allow for the prediction of the mortality risk of an individual independently of their chronological age (Fulop et al., 2010; de Vries et al., 2011; Dent et al., 2016). Research that involves model organisms now make use of these indexes as well (Whitehead et al., 2014; Baumann et al., 2018; Heinze-Milne et al., 2019). The latest technological/conceptual advancement in this field is “ageing clocks”—e.g., the mammal epigenetic clock that identifies 5-cytosine methylation of CpG sites (Bocklandt et al., 2011; Hannum et al., 2013; Horvath, 2013; Horvath and Raj, 2018) and transcriptomic clocks in Caenorhabditis elegans (Tarkhov et al., 2019; Meyer and Schumacher, 2021). These are yet another attempt to discriminate between biological and chronological age.

If ageing as we have presented it thus far is complex and difficult to define, understanding its evolutionary origins has revealed extremely complex problems for the past 150 years. Although it was first seen as an adaptive force of evolution immediately following the publication of Darwin’s work (Weismann, 1882), most of the 20th century has seen the development and adoption of theories that propose that ageing is a mere by-product of evolution (i.e., not adaptive). Ageing, as it has been theorised, exists through evolutionary time not because it is advantageous nor is it a process in itself but because the age-dependent selective pressure creates a “selection shadow” (Haldane, 1941). The “selection shadow,” in theory, allows for the accumulation of 1) deleterious mutations acting only late in life (Medawar, 1952) or 2) genes with early life benefits showing negative effects on fitness, again, only later (Williams, 1957).

In this Hypothesis and Theory article, we present and discuss a decade of work on the Smurf phenotype, initially described in Drosophila melanogaster as an age-related increase of intestinal permeability. This increase in intestinal permeability is made observable by feeding the flies a blue food dye that leaks only in individuals about to die of natural causes, hence the name “Smurf” (Rera et al., 2011). As our work progressed, we ascertained that the use of such a phenotype allows for the identification of individuals, within a population, who will soon die from natural causes. These same individuals show the physiological hallmarks traditionally associated with ageing (Rera et al., 2012). This prompted us to develop a theoretical framework whereby ageing is made of two consecutive and necessary phases. The transition between these two phases can be detected experimentally by the Smurf phenotype. This phenotype distinguishes a first phase (in which the risk of reaching the point of transition increases with time) from a second phase, where individuals show the long-described properties of ageing. Notably, differentiating between Smurf and non-Smurf individuals within an ageing population allows for the deconvolution of the “ageing transcriptome” into its chronological and biological components (Zane et al., 2023). Thus, our conception of ageing better accounts for biological disposition and heterogeneity of a population. Using this framework, we have developed and published approaches that seem to reconcile some of the empirical vs. theoretical discrepancies mentioned in the aforementioned paragraphs. Precisely, we have modelled longevity curves of these two consecutive phases (that, as we have observed, constitute a lifespan). An abrupt transition is distinguishable and the parameters of this model are experimentally quantifiable (Tricoire and Rera, 2015). Additionally, this theoretical framework allows for the conceptualisation of ageing within evolutionary theory as something that has been and is directly selected, rather than a mere by-product of other processes under selection (Roget, 2018; Méléard et al., 2019; Roget et al., 2024).

Conceiving ageing as a continuous, uninterrupted process that unfolds through time and, at large, weakens the physiology of the organism, aligns with the very personal and intimately known experience of human ageing. Despite our experiential perception of ageing, the mechanisms of such a phenomenon are enigmatic; ageing, through an evolutionary perspective, is rather counterintuitive: why and how might a process detrimental to the individual persist through evolutionary time and pervade a wide array of species? Prevailing theories see ageing as somewhat accidental, or a corollary to other things that directly benefit fitness. At the individual level, ageing is largely indistinguishable from the simple passing of time—the organism progressively, and irreversibly declines. Our work suggests otherwise and we aimed to present data that outlines the foundation of a “new ageing framework.”

Through our characterization of the Smurf phenotype (Figure 1) in D. melanogaster (Rera et al., 2012), we have demonstrated that the physiological Smurf state encompasses a biological state in which the necessary and sufficient parameters that lead to death are met. Our experiments show that every individual turns Smurf before death, but not all flies do so at the same age, even in isogenic populations. Drosophila females turn Smurf on average 3 days before dying (approximately 1 day in males), and physiological reversion to a non-Smurf state has not been observed yet. The risk of becoming a Smurf increases with age, leading to an increase in the proportion (not the absolute number) of Smurf individuals in a population with time (Figure 4). Interestingly, those properties are conserved by Smurfs in C. elegans and D. rerio (Dambroise et al., 2016) and a short-lived strain of M. musculus (Cansell et al., 2023).

All data collected in the past 10 years and presented here support a biphasic model of ageing from flies to mice. According to this framework, an organism ages by following two distinct phases in its life: a first, non-Smurf phase, where the organism is healthy and the risk of mortality is null, yet faces an increasing probability of becoming Smurf with time; and a second, shorter Smurf phase, in which the organism displays the markers of ageing and is committed to death in a specific amount of time (this time depends on the species and is subject to sexual dimorphism).

This framework reinterprets traditional assumptions of ageing (i.e., that the risk of mortality increases exponentially at the population level and that hallmarks of ageing increase progressively through time) and conceptualises the process of ageing as rather something that is constitutive of two phases: a linear one in which the risk of becoming Smurf increases with time and a following exponential phase in which there is an abrupt expression of ageing hallmarks. Our mathematical enunciation of this model removes the assumption of population homogeneity at a given age using parameters that can be quantified experimentally.

Our framework does not invalidate the importance of time in the ageing process. However, it strongly distinguishes between two different “times”: the chronological, calendar one, and the biological one, which is attuned to the individual. Both affect ageing and must be taken into account. Chronological time acts as a risk factor for the entrance into this second phase, while biological time (measured by whether an individual is Smurf or not) more accurately predicts the timing of appearance for the hallmarks of ageing and death. The Smurf model therefore places biological time at the centre of the ageing process, and views chronological time as a covariate.

In our recent study (Zane et al., 2023), we observed that transcriptional noise progressively increases with chronological time in non-Smurfs. This transcriptional noise is not further affected by the biological time of the organism, as Smurfs do not show an increase in transcriptional noise compared to their age-matched counterparts. We therefore hypothesise that an increase in transcriptional noise is one of the mechanisms through which chronological time acts as a risk factor for entrance into the Smurf phase. The non-Smurf phase can thus be conceptualised (within the framework of our model) as a resilience phase, by which the organism can cope with dysfunctions and genetic network alterations up to the moment when perturbations reach a non-tolerance level (Cohen et al., 2022)—i.e., “Smurf Transition Point,” which is yet to be identified—leading to a general stereotyped collapse (i.e., the Smurf phase). A similar view of ageing is supported by the inflammageing theory (Kirkwood and Franceschi, 1992; Franceschi et al., 2000). This theory proposes that ageing is due to “negative hits” affecting the organism’s ageing network, with chronic inflammation being the primary and common hit across individuals. In this view, the causes of collapse can be different in different individuals, but lead to the same outcome (i.e., the onset of the hallmarks of ageing). Recently, it has also been suggested that bioelectric networks act as body coordinators, modulating communications between cells and tissue (Levin, 2023). Ageing would then begin following the miscoordination of such signals, leading to divergence and aberrancies in tissue and organ behaviours (Pio-Lopez and Levin, 2024). Could this be linked to the transcriptional noise increase observed with time? This is, in any case, reminiscent of complex systems theory of ageing and the network theory of ageing proposed 30 years ago (Kowald and Kirkwood, 2016).

The examination of what happens at the moment of the transition, or just before (i.e., the pre-Smurf phase) would help reveal the causes of the organismal collapse. Practically, this is a challenging task, as it requires in vivo non-destructive reporters of the individual biological time of pre-Smurfs (i.e., biologically old flies just before they turn Smurf). We hope to identify putative reporter candidates with already available data by further assessing the expression of genes in very old Smurfs (which are statistically enriched in pre-Smurfs) and comparing them to the ones in Smurfs.

Recent work by Bansal et al. (2015) and Zhang et al. (2016) also describes two phases in the context of ageing. However, their conceptualization partially differs from the Smurf model proposed here. Bansal et al. (2015) illustrate two phases based on the organism’s ability to cope with environmental stresses, where interventions increasing lifespan increase the period of sensitivity to those (“gerospan”). Our model, instead, focuses on ageing in physiological conditions; while we can suppose that the “gerospan” phase described is enriched in Smurf individuals, not enough data are available at the moment to affirm that the two are biologically overlapping. In this sense, our work may compare more directly to that of Zhang et al. (2016), as they study physiological ageing in C. elegans. It is true that Zhang et al. (2016) describe a rather constant duration of lifespan, while we describe a rather constant duration of time individuals spend in poor health (Smurf phase); however, the scaling they describe seems to be similar to the changes in non-Smurf to Smurf transition rates that we describe between strains with significantly different lifespans (Dambroise et al., 2016). Interestingly, part of their five measured hallmarks might segregate with non-Smurf individuals—i.e., as a function of chronological time only- which would explain why they are able to identify two phases but the last phase is of variable duration.

The Smurf phenotype may not be perfect. As all biological processes, it is a continuous phenotype (Martins et al., 2018), i.e., the release of the blue dye from the intestine is a gradient. Even though there is individual variability in the intensity of the intestinal leakage, we define a Smurf as soon as there is extended coloration visible throughout the body. This binary classification has shown to be relevant both mathematically (Tricoire and Rera, 2015) and molecularly (Rera et al., 2012; Zane et al., 2023). Longevity curves are successfully modelled when it is assumed that the Smurf transition is binary and precedes death within a constant expected remaining amount of time. Finally, individuals who we have been labelled as Smurf show the stereotypical hallmarks of ageing when characterised molecularly. All of this causes reason to believe in the validity of a binary classification.

The assay is simple yet nevertheless presents experimental caveats that may contribute to reproducibility issues. Two teams have reported that only a fraction of individuals turn Smurf prior to death; we feel that addressing this incongruity is best achieved through discussion with these teams, which is an avenue that lies in our future and will undeniably further reveal the unaddressed complexities of ageing (Bitner et al., 2020; Landis et al., 2021; Tower, 2023). It should be recognized, however, that the Smurf assay allows the community of researchers studying ageing to use a physiological, evolutionarily conserved marker, which predicts a high risk of impending death while bringing together other so-called physiological or transcriptional hallmarks of ageing (Rera et al., 2012; Clark et al., 2015; Dambroise et al., 2016; Cansell et al., 2023) even when death is caused by acute stresses (Rera et al., 2012; Katzenberger et al., 2015).

The identification of the two-phase ageing process, something that appears to be highly stereotyped and conserved in Drosophila, nematodes, zebrafish and now mice, (reminiscent of what is generally defined as “age-related changes”) prompted us to examine the evolution of ageing. Our resulting evolutionary model shows formally and numerically that an organism capable of reproducing and maintaining itself will necessarily evolve towards a life-history trait demarcated by two consecutive phases: a first healthy phase, followed by a second one of decreased fertility and life expectancy. We interpret the results from this model as reminiscent of the Smurf phase.

In summary, our framework challenges the commonly held belief that ageing markers progressively increase with time in an individual, while reinterpreting this phenomenon as the progressive increase of Smurf individuals within a given population (and as something that has been positively selected for through evolutionary time). Whether the Smurf phase is programmed (and to what extent) remains to be determined. It is nevertheless a highly stereotyped phase both molecularly and physiologically. The evolutionary conservation of ageing, when viewed from the bi-phasal perspective, notably questions the anti-ageing approaches that have driven most of ageing research in the past 20 years. Indeed, if ageing is mostly considered an accident of evolution, why not attempt to “cure” it? If, instead, ageing follows a stereotypical, evolutionarily conserved path, selected for through evolution, can we justify our efforts to seek something that might not, at certain points in time within a lifespan, exist (i.e., a cure)? The framework we propose allows for the identification of factors involved in the risk of entering the second phase of life while providing clues that the latter is likely a phase of chronic disease and higher risk of impending death (and also a time in which interventions might prove ineffective). This framework for studying ageing thus favours the future outlining of a unified model of multiple, age-related chronic conditions that might require distinct interventions depending on whether an organism is in the first or the second phase of its life.