Dept of Biology, Leidy Laboratory
University of Pennsylvania
Philadelphia, PA 19104-6018
Laboratory animals maintained on a diet restricted in calories (CR) live longer than ad libitum fed animals, and many of the physiological markers of senescence appear on a delayed schedule. On its face, this effect attests to the plasticity of senescent processes under genetic control, and thus supports the notion of senescence as an adaptation. The difficulty of reconciling CR data with prevailing theories of the evolution of senescence is highlighted, and other failings of these theories are summarized. An adaptive model of senescence solves these problems, but detailed implementation remains a substantial challenge.
Caloric Restriction and the Case for an Adaptive Theory of Senescence
Joshua Mitteldorf, U of Pennsylvania
Experimental links between caloric restriction (CR) and longevity were first reported in the 1930s and have been studied at an accelerating pace ever since (Weindruch & Walford, 1988; Yu, 1993). The effect is robust, and has been observed in many species. Compared to ad libitum fed subjects, animals fed on a nutritionally complete diet with 40 to 80% of the ad libitum caloric content live 20 to 80% longer. Because maximum lifespan and average lifespan both increase, the effect has been interpreted as a retardation in the rate of aging. Many physiological markers of senescence are forestalled, while other aspects of vitality are enhanced: CR animals have strengthened immune function, greater stamina, healthier cardiovascular systems, and better resistance to carcinogenic insults, with no loss in activity level. Reproductive function is suppressed for the duration of the caloric restriction, but fecundity returns if full feeding is resumed.
Several authors (Holliday, 1989; Austad, 1995; Masoro & Austad, 1996) have sought to provide an evolutionary rationale for this effect: Random periods of temporary food scarcity are plausibly assumed to be a common feature of animal niches. Such times are not propitious for reproduction, but survival takes on an enhanced importance, and those individuals who persist to reproduce after the scarcity has ended realize a fitness bonus, as their offspring are delivered into a world where food is plentiful and competition has been culled by starvation. Hence a response that enhances survivability and delays reproduction is likely to have substantial adaptive value.
If this is the correct explanation, then the question arises, why is it that senescence is suspended only during times of caloric stress? If the metabolism can exert such control over the aging process, why has evolution reserved this adaptation for times of famine, and not implemented the program to enhance survival in times of plenty?
There can be only two sorts of answers to this question. The first posits that longevity cannot be increased without a concomitant depression of fertility. Senescence is the result of optimization over the tradeoff between fertility and longevity, and the difference between normal and CR animals results from a shifted balance between the values of fertility and longevity in times of scarcity. This paradigm requires an ironclad link between fertility and senescence, and experimental support for such a link is wanting.
The second is that senescence in itself is an adaptation that has been affirmatively fashioned by selection. This idea has been dismissed on general theoretical grounds: senescence has only negative impact on the fitness of the individual, so that any theory of senescence as an adaptation must invoke selection at the population or species level.
Hence, of three modern theories for the evolution of senescence, two fall firmly in the first class, invoking pleiotropic links, direct and indirect, between fertility and senescence. The third theory derives from mutation/selection balance, and does not support any explanation for the CR effect. Theories of the second class have almost no currency in the literature. In the pages that follow, we catalog some failed predictions of each of the prevailing theories, with special attention to CR. This motivates a preliminary discussion of prospects for a theory of senescence as an adaptation.
A context for theoretical consideration of the aging problem is created by the quantification of reproductive value as a function of life history variables (Fisher, 1930, 1958; Charlesworth, 1980; Stearns, 1992). Several book-length reviews are available describing the three prevailing theories, and summarizing pertinent experimental results (Rose, 1991; Finch, 1990; Wachter & Finch, 1997). The first theory, historically is based on antagonistic pleiotropy (AP): genes of mixed effect that promote vitality and fertility early in life, but which happen to have a corrosive effect on fitness at advanced ages. A variant of the AP theory with much current appeal is the disposable soma (DS) theory, based on compromise in the allocation of resources required for the body's maintenance and repair functions. The third theory, invoking mutation/selection balance, is referred to as accumulated mutations (AM). We consider each of these in turn, following a preliminary note of clarification:
The question of senescence is sometimes posed thus: why does the soma decay and mortality increase after reproductive capacity has been lost? This formulation begs the question of reproductive senescence, which is far the more interesting one. An individual who has lost the capacity to reproduce is already an evolutionary nonentity (except in species that nurture their young), and any marginal selective breeze is sufficient to blow him down; however the loss of reproductive capacity over time and the loss of vitality while fertility is still substantial carries a real fitness cost, the persistence of which poses a theoretical challenge. We concern ourselves here exclusively with the latter question, and senescence will be defined as the decline with age of fertility, and survivability insofar as it affects the future capacity to reproduce.
In the DS theory, competing demands for an organism's metabolic resources result in a compromise between repair and maintenance of the soma on the one hand, present survival, a vigorous metabolism and reproductive function on the other. Reproductive value is maximized by enhancing vitality and fertility in early life, diverting some resources away from repair and maintenance in the process. The body is permitted to decay with age because a more effective repair system would entail a metabolic investment that cannot be justified when the same resources could be used for activity and procreation in the present (Kirkwood, 1992).
The DS theory may be extended to provide an account for CR phenomena: when nutrition is scarce, the relative value of present and future reproduction shifts so that the investment of metabolic resources in the future is more worthwhile. The organism is able to forestall senescence by de-investing in present fertility.
Although this idea has been in the literature for over a decade, no detailed model of the process has been proposed. We can easily see why: First, the original tradeoff in Kirkwood's model must be narrowed to balance repair/maintenance against fertility alone. This is because CR animals are able to accomplish their life extension while enhancing all other aspects of present vitality, including endurance, activity level, and disease resistance. Second, the presumed candidate for the "scarce resource" which serves as the basis of the tradeoff is caloric energy; however the ability of CR animals to extend life while challenged by drastically lower caloric energy makes this possibility implausible on its face. Typically, a rodent that begins CR as an adult might weigh 75% of an ad libitum control, while consuming 50% as many calories; thus the total energy per unit of body mass is reduced by a third. Any model that finds more food energy available for somatic repair under these circumstances must specify that more than 1/3 of calories in the ad libitum animals was tied up in reproduction. This figure might be appropriate for lactating females (de Paolo, 1993) but cannot be justified for females that are not breeding, and certainly not for males. Male and female lifespans respond comparably to CR, but fertility in females is curtailed to a much greater extent than in males, and male caloric investment in fertility was presumably smaller to begin with.
Even if the parameters of a DS model were fit to the CR data, without regard to metabolic reality, the theory must predict a curve of lifespan vs caloric intake that falls when feeding levels are sufficiently low: at some point, no more energy is being used for reproduction, and every calorie of which the animal is deprived must reduce the quantity available for somatic repair. But in reality, maximum lifespan continues to increase with decreased caloric consumption right down to the level where other animals are dying of starvation (Kristal and Yu, 1993).
Physical activity is another competing demand for caloric energy. The DS theory would thus predict that exercise should shorten lifespan, especially in CR animals, where all available caloric energy is going into repair and maintenance. The opposite is observed. In experiments with CR rodents, the longest-lived test animals are those which consume less than half the daily calories of the controls and in addition are induced to spend much extra time on the treadwheel (McCarter, 1993).
Some but not all of these inconsistencies could be dispelled if a candidate other than caloric energy could be identified for the conserved resource that is the basis of the DS tradeoff. Imagine that another "scarce resource" could be specified, the supply of which is the same for CR and control animals in the lab. With this modification of the theory, the basic observation of increased longevity with decreased caloric energy would be less stark a contradiction. However, questions would still remain: If it is indeed the diversion of resources into reproductive function in the present that is the root cause of senescence, why is it that breeding is not observed to shorten maximum lifespan? And more fundamentally, why is it that the organism is able to accomplish so much better somatic repair under extreme stress than it can when resources are plentiful and few physical demands are made on the metabolism?
Finch (1990, pp 547-550) reviews evidence from lab animals concerning the cost of reproduction, and concludes that reproductive stress entails an increase in concurrent mortality rates, but is not associated with future mortality, as would be expected if the rate of aging were affected. In humans, there is contradictory evidence, associating reproduction with both longer and shorter lifespans. In some studies, sexual activity extends the lives of males, while in others the siring of offspring shortens lifespans (Reznick, 1985). There is evidence that eunuchs live longer than other human males (Hamilton & Mestler, 1969), but that frequency of male sexual activity in humans is negatively correlated with mortality (Davey Smith, 1997). Bearing children and nursing may be associated with stress factors that accelerate senescence, but they may also be linked to hormones that preserve the cardiovascular system from aging (Finch, 1990 pp 338-345). There is epidemiological evidence that childbearing may forestall aging in women (Perls et al, 1997; Mitteldorf, 2000), and historical evidence in which childbearing is associated with shorter lifespan (Westendorp & Kirkwood, 1997). The elementary observation that women have longer natural lifespans than men (Finch 1990, p 352), despite their greater investment in reproductive metabolism has never been explained in the context of the DS theory. The fact that this may be part of a general evolutionary stratagem extending the lifespan of the caregiving gender (Waldron, 1987) strengthens the impression that rates of senescence are elastic, and that lifespans are easily extended under genetic control. Although there may be considerable ambiguity surrounding detailed interpretation of these data, the DS theory would predict a strong and manifest link between fertility and senescence, and the observations do not support this inference.
The AP theory, in its original formulation (Medawar, 1952; Williams, 1957) assumed the existence of genes that enhance the organism's early-life fitness but detract from late-life fitness. The same logic that casts doubt on the DS theory above carries over to AP, but with one difference: The DS theory, in combination with CR data, implies that reproduction should significantly shorten lifespan; the corresponding prediction for AP is that genes affecting lifespan should have a reciprocal effect on fertility.
Evidence has been eagerly sought for this connection and some results that confirm a genetic link between fertility and senescence are uncritically reported as vindication for the AP theory. In the oldest and most extensive experiments involving fruitflies bred for longevity, Rose (1984) reported early results in which fertility decreased in parallel with mortality. However, as the experiment continued, fertility was seen to turn around and increase again, even as mortality continued to decline (Leroi et al, 1994). Rose's experiments have now produced a decade of incidentally advancing fertility, as he selects his flies for longevity (Rose, Nussbaum & Chippindale, 1996). Rose himself remains a knowledgeable and committed advocate for pleiotropy, and he continues to describe the data as complex and nuanced; but in the hands of a skeptic, his data, in combination with the CR results, might be promoted as a mortal blow to the AP theory.
In other animals, the evidence is no more supportive. Nematodes bred for longevity show a tendency to suppressed fertility; however single genes have been found that extend life in C. elegans for which no adverse pleiotropic effect has been identified (Lithgow, 1996). Recently, a single-site mutation was discovered that extends the life of laboratory mice (Migliaccio et al, 1999). The mechanism of action of this gene is already well-understood: it controls apoptosis in cells that have been affected by free radical damage. Apparently, cells in an early stage of senescence are programmed for apoptosis long before their usefulness to the body has ended. Two implications suggest themselves: First, it is unlikely that such a mechanism carries any benefit to early fertility; hence we have an example of an important mammalian senescence gene which is not pleiotropic. Second, apoptosis is unquestionably an adaptation, a purposeful piece of life's developmental program. The association of apoptosis with accelerated mortality suggests that senescence may be part of life's program.
Fertility and longevity are exactly the components of individual fitness, on which the pressure of natural selection has been strong and direct. How can it be that wild populations quite typically encompass significant diversity in these traits, which have been subject to such direct and sustained selection pressure? Data from breeding experiments constitute powerful prima facie evidence that both senescence and limits on fertility are adaptations, which can only have been selected for their positive fitness value at the population level. Kenyon has stated quite forthrightly, "It's inescapable that aging is regulated deliberately by genes." (Pennisi, 1998).
The loose association between reduced fertility and enhanced longevity found in some artificial selection experiments does not compel the conclusion that antagonistic pleiotropy is a root cause of senescence, but is open to other obvious interpretations. For example, in a population that sustains a steady-state diversity, it must be expected that any trait that decreases fitness cannot long persist unless it is associated with a trait that increases fitness. Thus, it may be that pleiotropy is an explanation for the intraspecies variability in the genetic component of senescence, though not the fundamental explanation for its existence. A more speculative possibility is that senescence, once having been imposed on the genome by group selection effects, is maintained and protected from individual selection pressure by pleiotropic links. Pleiotropy would then be a mechanism for the enforcement of the communal interest in preventing drift toward ever longer lifespans via individual selection. In this view, antagonistic pleiotropy is a result, not a cause of the evolution of senescence.
According to the AM theory (Edney & Gill, 1968), the metabolic properties that lead to senescence are flaws in the genome, mutations that accumulate faster than selection can eliminate them. In every area of the genome, selection is constantly weeding out deleterious mutations. At any given time, there is a steady-state load of unfavorable mutations, recently acquired and not yet eliminated. The AM theory says that because the force of selection on traits that only appear at advanced age is always weaker than selection for traits that manifest early, the steady-state load of such traits is bound to be heavier at advanced ages. This is adduced as an evolutionary explanation for senescence.
The CR effect seems incongruous in the context of an AM theory of senescence. That a CR effect could have evolved separately to combat each constellation of accumulated mutations, since the time that those mutations appeared but before selection could have eliminated them - this basic result is unexpected at best. Is it plausible that across so many species natural selection has found the complex response for alleviating senescent damage in times of caloric stress, but has been unable in the same timespan to eliminate from the genome the mutations that caused the senescent damage in the first place?
No theory of the CR effect premised upon AM as a context has ever been proposed, and it is difficult to imagine what such a theory might look like. To explain the CR phenomena, AM has no recourse but to fall back on pleiotropy. The difficulties then are identical to those described above for the AP theory.
In other respects as well, AM has failed to find experimental support. Promislow et al (1996) perform a direct test of this theory when they measure additive genetic variance of traits affecting mortality as a function of age in fruitflies. They report that, contrary to prediction, the variance actually declines with age.
The AM theory suggests, for each species, a recent, stochastic origin for the genes accounting for senescence, since these mutations are in steady state with (weak) selection. But Clark (2000, and references therein) finds evidence that prevalent senescent mechanisms are highly conserved over a broad range of taxa, and that some of these mechanisms can be traced to the dawn of eukaryotic life. "The high degree of conservation of the genes underlying senescence, across species so widely separated in evolutionary time, is one of the major arguments in favor of the early emergence of genes controlling senescence and compulsory death, as opposed to their gradual accumulation." (Clark, 1999 p 190)
The mechanisms of senescence, common to all higher life forms, are oxidative damage, telomeric limits on reproduction, and apoptosis. (Leist & Nicotera, 1997) The latter two are clearly purposeful metabolic systems, and their cellular machinery and genetic basis provides the vision of commonality that is the basis for Clark's argument. But even the former - oxidative damage - is not easily reconciled with the AM picture. All cells have elaborate machinery for molecular repair and maintenance. In protozoa, the machinery operates with adequate efficiency to insure a robust metabolism after millions of cell divisions. In higher life forms, a very similar machinery operates with great efficiency in early life, but slackens over time, and allows oxidative damage to accumulate. It is the burden of the AM theory to explain why, of all mutations that affect late-life fitness, the one that turns down the control knob for molecular maintenance functions has been so ubiquitous.
There is additional evidence from evolutionary history that senescence is an adaptation regulated by genes, rather than a failing or limitation of evolution. Lower organisms that reproduce asexually but conjugate occasionally evince an analog of senescence, mediated through telomeres. Telomeres are molecular tails of the chromosome that shorten with each cell division; in both protists and multi-celled organisms, cells can no longer fission after the telomeres are depleted. (Cells of higher plants and animals have telomeres of sufficient length that replication over a lifespan does not usually approach the telomeric limit on cell divisions; hence telomeres are thought not to be a generally significant mechanism of senescence in higher organisms.) All eukaryotic cells have genes coding for telomerase, an enzyme which serves as an effective antidote to this form of senescence. However, in protists the enzyme is made available during conjugation only, and not during mitosis. Hence telomeres shorten with each cell division, but are replenished and the clock is reset when conjugation occurs.
This mechanism gives every appearance of being a purposeful metabolic function. The AM and AP theories are not applicable to telomeric senescence, and, in fact, no evolutionary theory of protozoan senescence based in individual selection has ever been proposed. Like senescence in higher organisms, telomeric senescence contributes only negatively to individual fitness. The only conceivable benefit of this well-developed adaptation is the enhancement of genetic diversity in the population. If group selection is accepted as the explanation for protozoa, it becomes the more plausible that the model may be extended to higher organisms.
Particular disparities between the AP and AM theories and experiment have been raised in the past, and some have not been resolved over time. Pletcher & Curtsinger (1998) have found the two existing theories incapable of explaining the shape of mortality curves at advanced age. And though Mueller and Rose (1996) claim to have a model which resolves this disparity, their model takes a fixed reproductive senescence timescale as a starting point, begging the question, as we have noted above ("preliminary note").
Aging in most higher organisms is not a gradual decline, but a rapid deterioration that appears suddenly in midlife. Mortality is usually modeled as a Gompertz (1825) function, increasing exponentially with age. Abrams and Ludwig (1995) find that the steep increase in mortality with age typical of many species in nature does not derive readily from a tradeoff model based on reproductive value of investments in fertility vs repair.
In semelparous organisms, we are offered a spectacle of senescent mechanisms that are common to iteroparous relatives, but which operate on a greatly accelerated schedule. The usual explanation for this effect invokes metabolic costs of all-out reproductive effort (Stearns, 1992, pp 186-193). However, symptoms of deterioration first appear only after reproduction is complete (Finch, 1990, pp 83-93). Decline proceeds rapidly at a time when reproduction has ended. Annual plants can be induced to survive during an extended period of intense reproductive effort by the removal of flowers and fruits before the seeds mature (Leopold, 1961). Semelparous female octopi can be reprogrammed for an extended life by removal of their optic glands after spawning is complete (Wodinsky, 1977). These observations make clear that accelerated senescence in semelparous species is an example of programmed death, and its interpretation as an incidental cost of intense reproductive effort derives from a theoretical bias.
Promislow et. al. (1996) report experiments measuring additive genetic variance in fertility and mortality as a function of age, and interpret them to the detriment of the AP and AM theories. In introducing these results, they opine that "...the main evolutionary models of senescence are antagonistic pleiotropy and mutation accumulation, neither of which has substantial experimental support."
Kenyon (1996a, 1996b) has studied the genetics and biochemistry of aging in nematodes, and finds evidence for senescence as a metabolic function under genetic control. She and Clark (1996, 1999, 2000) have been bold advocates for the proposition that aging is an adaptation.
Bowles (1998) attempts a coherent theory of evolution of aging as an adaptation, derived from metabolic evidence, evolutionary history, and comparative zoology. This essay by a self-taught bioligist is of uneven quality, but includes gems of evolutionary insight.
None of the three prevailing theories for the evolution of senescence can be reconciled with a broad range of observations about the genetics and developmental biology of aging and death. Moreover, the shape of the observations points in the direction of senescence as an adaptation. The reason for the persistence of the current theories is not their explanatory power, but rather the problematic nature of the theoretical alternative.
Senescence by definition offers no advantage for the individual; it is difficult to imagine any benefit to offspring or other relatives that is not mediated through alleviation of crowding (Weismann, 1889), so that the benefit accrues equally to all conspecifics in a given geographic region. Hence the hypothesis of senescence as an adaptation leads inevitably to the necessity for group selection.
It is interesting to note that the logic that leads from the CR data to group selection does not depend on a theory of senescence. CR animals are more robust in every way than ad libitum controls. Why is the enhanced stamina and disease resistance typical of CR animals not observed in times of abundant nutrition? Either the metabolic cost of reproduction is fundamentally and unavoidably high, so high as to dwarf the cost of a 50% reduction in fuel intake, and equally high in males and females; or else evolution has tempered survivability of individuals in times of plenty, reserving maximal robustness for periods of nutritional stress. This may be a population-level adaptation, distinct from senescence, and akin to the exercise of natural population control, for which claim Wynne-Edwards (1962) was once subject to scorn and derision (Williams, 1966).
The basis for the group benefit is likely to be enhancement of population diversity and a shorter effective generation cycle. Fundamental theoretical considerations suggest that collapse of genetic diversity is a looming danger to ongoing evolution, and senescence contributes to a solution to this problem by limiting the extent to which the progeny of the most successful individuals may dominate a population. It has been proposed (Bull, 1987; Charlesworth, 1993) that evolution may operate directly to maintain diversity, for the long-term health of the population. This is, in fact, a commonly-cited rationale for the ubiquity of sex (Bell, 1982).
The shorter effective generation cycle is not due to enhanced early fertility, but only to the fact that more offspring can survive to maturity when competition is softened by deaths of aging conspecifics. Both enhanced population diversity and reduced generation cycle contribute to an increase in the rate of evolutionary change. A senescing population adapts in response to a changing environment more nimbly than a population that does not age. No detailed model for the evolution of senescence via group selection has yet been put forward, perhaps because of the extraordinary computational requirements for the demonstration of these effects: a large number of groups needs to be followed through many cycles of environmental change, and each group must contain a diverse population of individuals, whose genomes must all be tracked by the model.
The challenges of constructing a plausible evolutionary model that selects for senescence as an adaptation are considerable. But the motivation is correspondingly strong, and there are two hints that the endeavor could meet with success.
First is the new field of evolution of evolvability. Wagner and others have cited evidence from the genotype-phenotype map that the process of evolution is itself highly optimized (Wagner & Altenberg, 1996; Wagner, Chiu & Hansen, 1999; Kirschner & Gerhart, 1998). It is a point not emphasized in these articles, but implicit in the concepts, that evolvability is a trait characterizing a population and not an individual. The concept of individual selection for evolvability has no meaning. Thus the new attention to evolvability creates a context in which senescence may appear as one of many adaptations whose signal effect is to enhance the long-term effectiveness of evolution.
Second is the ever-growing body of literature that make group selection far more plausible and more viable a prospect than it was thirty years ago, when many of our prevailing attitudes were frozen in place. Sober and Wilson (1998) review some of this history. Multi-Level Selection theory (Wilson, 1980; 1997) has been developed expressly to analyze the balance between group and individual selection. In a recent MLS model (Mitteldorf & Wilson, 2000), group selection is an emergent phenomenon in populations that self-organize as they diffuse slowly through a viscous lattice grid. Blind, local altruism is shown to be a viable strategy in this environment. This kind of model seems both sufficiently general to account for the ubiquity of senescence; and the presumed group benefit of senescence, mediated by the alleviation of crowding and consequent enhancement of genetic diversity, is an instance of blind, local altruism. Perhaps the model can be extended to embrace senescence.