Evolutionary Biology of Aging: Future Directions

P088387-Ch08.qxd 10/31/05 11:27 AM Page 217

Chapter 8

Daniel E. L. Promislow, Kenneth M. Fedorka, and Joep M. S. Burger

I. Introduction antagonistic pleiotropic effects (Campisi, 2003). Over the past two decades, we have seen Serious effort has been expended in extraordinary progress in evolutionary testing these classic evolutionary theo- studies of senescence. Beginning with ries of senescence (e.g., Hughes & the early quantitative genetic tests of Charlesworth, 1994; Promislow et al., theories of senescence (Edney & Gill, 1996; Rose & Charlesworth, 1980). The 1968; Luckinbill et al., 1984; Rose & field is now moving beyond tests of Charlesworth, 1980), we have progressed existing hypotheses, but what exactly is to the point where evolutionary biolo- the future direction for the evolutionary gists rely on state-of-the-art molecular studies of aging? We do not have a scien- tools, and the ties between evolutionary tific crystal ball, and making definite pre- and molecular approaches in the study of dictions places us somewhere between senescence are often seamless. More hubris and folly. Nevertheless, with this than perhaps any other research area in caveat in mind, we hope that the ideas evolutionary ecology, molecular biolo- presented here may spur the next genera- gists working on senescence appreciate tion of biogerontologists to consider evo- how evolutionary biology contributes to lutionary studies of senescence. our understanding of senescence. Many What differentiates evolutionary studies will tell you about George Williams’ of aging from studies that take a strictly antagonistic pleiotropy theory of senes- molecular, physiological, or demographic cence (Williams, 1957), or perhaps Tom approach? One distinction is that non- Kirkwood’s disposable soma theory evolutionary biologists frequently ask (Kirkwood, 1977), and the especially proximate questions (e.g., how do specific well-rounded researcher might mention biological processes change with age, that molecular biologists have identified and how do genes affect these changes?), individual genes that appear to exhibit whereas evolutionary biologists are

217 P088387-Ch08.qxd 10/31/05 11:27 AM Page 218

218 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

interested in ultimate questions (why has weak selection and could spread through aging evolved, and why have particular random genetic drift. Over evolutionary genes evolved to influence longevity?). time, early-acting deleterious mutations The molecular biologist is trying to iden- will continually be removed by selection, tify the mechanisms that cause aging. whereas late-acting ones will accumulate. The evolutionary biologist is trying to According to Medawar’s mutation accu- place those mechanisms in a broader per- mulation theory, it is inevitable that we spective, asking why those mechanisms carry a relatively high load of late-acting and not others are important to aging, and deleterious mutations inherited from our how those mechanisms are shaped by ancestors. As we age, we experience the other forces—mutation, selection, genetic effects of these mutations, which cause drift—acting over evolutionary time. a decrease in rates of survival and/or In the past few years, the lines between fertility. these two fields have begun to blur. A decade later, George Williams devel- Molecular geneticists working on aging oped his antagonistic pleiotropy theory, in have used evolutionary theory to motivate which he argued that senescence would studies of single genes (Campisi, 2003; arise if late-acting deleterious mutations Walker et al., 2000), and evolutionary biol- were actually favored by selection due to ogists have embraced techniques that are their early-acting beneficial effects firmly rooted in modern molecular biol- (Williams, 1957). In this case, senescence ogy (e.g., Pletcher et al., 2002; Tatar et al., evolves due to tradeoffs between early-age 2001). benefits and late-age costs, an idea that Although evolutionary biologists now was further developed by Tom Kirkwood rely on 21st-century techniques, the in his disposable soma theory (Kirkwood, greatest evolutionary contributions to 1977). A half-century after Medawar aging studies date back half a century. and Williams, evolutionary biologists are Until the middle of the 20th century, the still trying to determine which of these standard argument for the evolution of theories provides the best explana- aging was that it was good for the species tion for senescence (Charlesworth, 2001; (Weismann, 1891). In the 1940s, we began Charlesworth & Hughes, 1996; Hughes to move beyond that argument, which we et al., 2002; Partridge & Gems, 2002a; now recognize as fallacious. Following Snoke & Promislow, 2003). More recently, from the insights of Fisher (1930) and molecular gerontologists have begun to Haldane (1941), in 1946, Medawar argued embrace these ideas, with a particular that because the strength of selection interest in finding genes with antagonistic declines with age, senescence will arise as pleiotropic effects (Campisi, 2003; Walker an inevitable consequence of this decline et al., 2000). (Medawar, 1946). Consider a novel, germ- The current focus in the evolutionary line mutation that reduces survival at just biology of aging encompasses three main one age. If it reduces survival before the areas: age at maturity, then the probability of surviving to produce offspring at any age 1. Ongoing studies are trying to is uniformly reduced. Such a deleterious characterize the genetic architecture mutation would experience strong nega- of aging (see Chapter 7, Mackay et al.), tive selection and eventually be lost from asking not only which model can explain the population. In contrast, a mutation genetic variation for senescence, but that reduces survival at some late age, also whether genes that have been found after most individuals in the population to extend life span in the lab (so called had died, would experience only very “aging genes”) show allelic variation that P088387-Ch08.qxd 10/31/05 11:27 AM Page 219

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 219

correlates with longevity in wild-caught mathematical model for the evolu- isolates, and whether the effect of aging tion of senescence was developed by genes depends on the sex or external W. D. Hamilton (1966). Hamilton started environment in which they are expressed with the well-known Euler-Lotka equa- (Fry et al., 1998; Leips & Mackay, 2000; tion (Euler, 1760; Lotka, 1925): Nuzhdin et al., 1997). 2. A second body of work has focused 1 erxl(x)m(x) (1) specifically on the shape of mortality trajectories (see Chapter 1, Gavrilov and in which r is the intrinsic rate of increase Gavrilova), asking why mortality curves in a population (also called the Malthusian increase exponentially with age (Abrams parameter, and used as a measure of & Ludwig, 1995; Charlesworth, 2001) and Darwinian fitness), l(x) is the probability why mortality rates appear to decelerate of surviving from birth to age x, and m(x) late in life in some cases (Mueller et al., is the number of daughters that a female 2003; Mueller & Rose, 1996; Service, produces at age x. Hamilton’s model used 2000; Vaupel et al., 1998, but see Finch & this equation to provide exact descriptions Pike, 1996; Linnen et al., 2001). of how the strength of selection acting on 3. Finally, many evolutionary biologists rates of mortality or fecundity would have become interested in the central role change with age. One can think of the that the endocrine system may play in strength of selection as a measure of how determining the evolution of the suite of fast a new mutation will be fixed or lost in traits that make up an individual’s “life a population. In particular, the strength of history strategy,” including development selection acting on P(x), the survival rate time, age at maturity, growth rate, body from age x to x 1, is given by size, fecundity, and, of course, life span (Tatar et al., 2003). The evolution of aging eryl(y)m(y) in general, and these three subjects in r y x 1 (2) particular, have all been reviewed recently lnp(x) xerxl(x)m(x) in other sources (e.g., Promislow & Bronikowski, in press; Tatar et al., 2003) The strength of selection acting on and in this book (see Chapter 15, Tu et al; fecundity is given by Chapter 19, Miller and Austad; Chapter 20, Carter and Sonntag). Accordingly, r e rxl(x) rather than going over well-tilled ground, (3) rx in the rest of this chapter, we will look at m(x) xe l(x)m(x) five areas that are less studied at present but which we think may provide fertile The important point that these equa- soil for the growth of future tions illustrate is that the demographic evolutionary studies of aging. These parameters themselves determine the include (1) molecular evolution and gene rate at which selection declines with age. networks; (2) the intersection of We show how the strength of selection physiology and demography; (3) parasites declines for one particular set of values and immunity; (4) sexual selection and for age-specific survival and fecundity sexual conflict; and (5) genetic variation in in Figure 8.1. These equations do not natural populations. include information about such factors as social behavior (e.g., mate choice, par- The rationale for focusing on these parent-offspring conflict), host-parasite ticular areas comes from our thinking interactions, variation in the environ- about early models of aging. The first ment, physiology, or the underlying P088387-Ch08.qxd 10/31/05 11:27 AM Page 220

220 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

1 0.1 predictions about the shape of human A mortality curves (Lee, 2003). And most 0.8 0.08 recently, Jim Vaupel and colleagues have created models to demonstrate that 0.6 0.06 species with indeterminate growth (i.e., no asymptotic size as adults) may evolve 0.4 0.04 “negative senescence,” where mortality Survivorship rates actually decline with age (Vaupel 0.2 0.02 et al., 2004). In the following sections, we

0 0 explore a range of biological phenomena B that may allow us to further extend classic models of aging. Clearly there is much 0.04 Annual Fecundity work to be done, and we are optimistic that as we unite new molecular tools and 0.03 new evolutionary ideas, the coming years will bring a more comprehensive under- 0.02 standing of the evolution of senescence. Selection intensity 0.01 II. Genetics of Senescence 0 0 20 40 60 80 100 Age (years) For almost 20 years, evolutionary biologists and molecular biologists working Figure 8.1 (a) Age-specific fecundity (dashed line) on the biology of aging appeared to be and survivorship (solid line) for a hypothetical human population. (b) Intensity of selection acting working on utterly different problems, on a mutation that decreases age-specific fecundity with little communication between the (dashed line) or survival (solid line) for a single age two groups (perhaps not unlike the gap class. Note the dramatic decline in the intensity of between those working on cellular sen- selection on survival after age at maturity, eventu- escence and those working on aging in ally reaching zero after the last age at reproduction. animal models; Campisi, 2001). This has changed in the past few years, with genetic architecture of these demo- evolutionary biologists now embracing graphic traits. All of these factors could the newest molecular techniques, and alter the shape of the curves described by molecular biologists starting to test evolu- equations (2) and (3). tionary hypotheses. For example, a collab- Researchers interested in the evolution orative effort among evolutionary and of aging have begun to develop ways to molecular biologists led to the first large- extend Hamilton’s equations, enhancing scale microarray analysis of patterns of both biological realism and predictive gene transcription associated with senes- power. For example, Peter Abrams cence in the entire fly genome (Pletcher showed that if one adds density depend- et al., 2002). The sequencing of entire ence to classic models for the evolution genomes has made it extremely easy to of aging, in some cases extrinsic mortal- identify genes associated with aging that ity rates no longer influence the evolu- occur across taxonomically diverse organ- tion of rates of senescence (Abrams, isms (and so, by inference, to identify 1993). Ron Lee has developed models that aging processes that have deep evolution- incorporate interactions between parents ary roots). George Martin wondered and their offspring into models of aging; whether genes associated with aging these models make impressively accurate were likely to be specific to each species P088387-Ch08.qxd 10/31/05 11:27 AM Page 221

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 221

(“private” mechanisms) or constant selectionist debate in molecular evolution across evolutionary time (“public” mech- mirrors that of the mutation accumula- anisms) (Martin, 1997). At least two gene tion–antagonistic pleiotropy debate in pathways associated with aging—sirtuins senescence. After a long argument between and insulin signaling—appear to be public neutralists and selectionists, most evolu- mechanisms (Partridge & Gems, 2002b; tionary biologists now accept that some Tissenbaum & Guarente, 2001). genes have evolved under a neutral sce- Current work on the molecular genetics nario whereas others have evolved due pri- of aging is focused on characterizing the marily to selective forces. Nevertheless, function of the genes and gene pathways the debate started by Kimura fueled that have already been identified, and on decades of exciting and productive finding new genes (see Chapter 7, Mackay research. Similarly, in the world of evolu- et al.). Clearly, this is where the action is. tionary gerontology, it will likely turn out So where does evolutionary biology fit that some genes have evolved due to muta- within this decidedly molecular enter- tion accumulation and others due to antag- prise? In the introduction, we discussed onistic pleiotropy, but the debate over how the strength of selection changes which is the better explanatory hypothesis with age. One challenge is to determine has sparked invaluable research progress. how the strength of selection acts not only However, we see the parallel between on demographic traits, but also on the molecular evolution and aging research individual genes that shape those demo- as having far more than just historical graphic traits. If we can do this, we may be interest. Molecular evolution has led to able to develop an evolutionary genetic fundamental advances in the way that model of aging that allows us to actually we understand the biology of organisms, predict what kinds of genes should be and some of these advances could inform associated with aging. A more refined future studies in the biology of aging. analysis of selection at the level of individ- Work on the genomes of a diversity of ual genes and genomes will come about in organisms has found that substantial por- two ways—first, by combining molecular tions of the genome are often made up of evolutionary studies of gene sequences or transposable elements (TEs). These “jump- genome structure with analyses of life ing genes” can function as parasites in the span, and second, through the study of genomes of their host and can lead to sub- gene- and protein-interaction networks. stantial increases in background mutation rates (McDonald, 1993). To the extent that aging is affected by age-related increases in A. Molecular Evolution and Aging somatic mutation rates, TEs may turn Molecular evolutionists study the process out to play an important role in the of evolution by analyzing variation in aging process. Surprisingly, relatively few DNA and protein sequences within and studies have examined the relationship among species. Many workers in the field between TEs and aging (Nikitin & have focused on the historical and current Shmookler Reis, 1997; Woodruff & patterns of selection that act on genes—an Nikitin, 1995). One study makes the inter- area of inquiry that began in earnest with esting (and cautionary) point that when the debate over whether most genetic vari- genes associated with longevity in ation was due to selection or had occurred Drosophila are identified by knocking in the absence of selection due to the the gene out with a P-element (a type of effects of random drift of neutral or nearly TE found in flies), the P-element inser- neutral mutations (Kimura, 1968). In tion alone may influence life span, some ways, the history of the neutralist– independent of the effects of the target P088387-Ch08.qxd 10/31/05 11:27 AM Page 222

222 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

gene (Kaiser et al., 1997). It would be deep evolutionary roots, it would be of useful to quantify the extent to which age- interest to look for common genetic or related accumulation of somatic muta- epigenetic causes of this pattern. tions accounts for the aging process and to Finally, comparisons of whole genomes examine the role that TEs play in this across species have been used to infer process. physiological process from genetic pat- Molecular studies have found that over tern. For example, Eisen and Hanawalt evolutionary history, some genes have (1999) determined the presence or experienced periods of very strong selec- absence of various DNA repair genes tion, whereas others appear to have been across more than 20 species of microbes. strongly shaped by drift (Li, 1997). Might On the basis of their “phylogenomic” this variation in selection translate to dif- analysis, they were able not only to deter- ferences not just among genes but also mine the degree of ubiquity of different among tissues? Early studies examined repair genes or gene pathways, but also to age-related declines in specific tissues, predict the repair phenotypes of different such as the early 20th-century work of microbes based on their underlying geno- Krumbiegel (1929) on fat body in aging types. Interestingly, they found that Drosophila. More recent studies have although some repair processes, such as found that the age-related rate of decline those associated with the recA gene, are varies among tissue types. For example, in strikingly constant across taxa, others nematodes, muscle cells appear to age at a have evolved relatively recently. The much faster rate than nerve cells (Herndon genetic basis of repair differs among et al., 2002). Furthermore, the effect of species, and in some cases, specific repair gene signaling on longevity is often tissue- mechanisms have evolved convergently specific (Hwangbo et al., 2004; Libina in different taxa. These findings should et al., 2003). One obvious challenge is to serve as a warning to biogerontologists. determine whether tissues differ in their Although some aging genes, such as those rate of aging because of different patterns in the insulin signaling pathway, may of selection acting on different tissues. For have deep evolutionary roots, our current example, the fitness consequences of a cell focus on these ubiquitous pathways may failing to produce hair follicles are likely lead us to overlook many others that to be quite different than the conse- evolve rapidly and are highly variable quences of a pancreatic cell failing to pro- among species. We are confident that duce insulin. If tissue-specific variation in “phylogenomic” approaches will lead to rates of aging (what we might call “senes- profound insights into the evolutionary cent heterochrony”—the opposite of the history of genetic mechanisms of aging in “one hoss shay” effect1) turns out to have the coming years.

1The observed patterns of senescent B. Gene Networks heterochrony contrast with the classic Classic studies of aging relied on forward example of the “One Hoss Shay,” made or reverse genetic techniques to identify famous in Oliver Wendel Holmes’ “The individual genes that extended longevity. Deacon’s Masterpiece”: But a one-gene/one-trait perspective is The poor old chaise in a heap or mound, As if it had been to the mill and ground! clearly an oversimplification (Lewontin & You see, of course, if you’re not a dunce, White, 1960; Wright, 1932). The past How it went to pieces all at once, – few years have given us a new appre- All at once, and nothing first, – ciation for the complex interactions Just as bubbles do when they burst. among genes and proteins that affect the P088387-Ch08.qxd 10/31/05 11:27 AM Page 223

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 223

formation of the final phenotype (Gibson hypothesize that more highly connected & Honeycutt, 2002; Wolf et al., 2000). genes or proteins are under stronger Already, studies have shown that in selection. This idea is supported by find- worms (Shook & Johnson, 1999) and flies ings that more highly connected pro- (Jackson et al., 2002; Leips & Mackay, teins evolve more slowly (Fraser et al., 2000), the way in which a particular 2002), that genes that produce these pro- allele affects longevity can depend on the teins are more likely to have a lethal presence of specific alleles at other loci phenotype when knocked out (Jeong (Spencer et al., 2003). et al., 2001), and that these genes are However, the shift from thinking less likely to be lost over evolutionary about single genes to epistatic interac- time (Krylov et al., 2003). tions between pairs of loci is still an In light of these studies, we might oversimplification. We need to begin expect that the structure of gene- and thinking about age-related changes to protein-interaction networks may influ- whole networks of interacting elements. ence which genes are associated with Recent studies have shown that when senescent decline (Sozou & Kirkwood, genes or proteins interact within a com- 2001). In a study of the yeast protein– plex network, the network structure protein interaction network, Promislow itself can make the network resilient to (2004) found that proteins with relatively damage in a way that would not be pos- high connectivity were more likely to sible if all the elements in the network be associated with replicative aging were operating independently (Albert than proteins with fewer interactions. et al., 2000; Flatt, 2005; Maslov et al., Furthermore, aging genes tended to 2004; Siegal & Bergman, 2002; Wagner, have a higher degree of functional 2000). pleiotropy than expected by chance. Networks can describe a wide array of Although Promislow (2004) argues that interactions, from the regulatory inter- these results are consistent with the actions among genes, to social interac- antagonistic pleiotropy theory of senes- tions among individuals, to transfer of cence, just why we observe these pat- electricity from power stations to users. terns is still an open question. In general, networks consist of “nodes” Molecular studies have identified or “vertices” connected to each other by complex pathways that affect senes- “edges.” The number of edges that a cence, best exemplified by work on particular node has is called its “degree” the insulin-like/insulin growth factor or “connectivity,” and the frequency signaling pathway in C. elegans and distribution of connectivity across all D. melanogaster (see Chapter 13, nodes in a network is the network’s Henderson et al.; Chapter 15, Tu et al.). degree distribution (Albert & Barabási, We now face the exciting challenge of 2002). Biological networks typically using classical genetics approaches (Van have a degree distribution that approxi- Swinderen and Greenspan, 2005) and mates a power law (Albert & Barabási, microarray studies to describe the larger 2002), such that the majority of nodes networks of interacting genes that have just one or two edges, but some might affect aging. At the same time, may have tens or hundreds of edges. we need theoretical models to predict Studies on network robustness illustrate how the network of nodes with which a that the strength of selection acting on single gene interacts can be used to pre- a single gene depends largely on the dict (1) whether this gene will be associ- network context within which that ated with longevity and (2) how likely gene functions. For example, we might the gene is to fail as the organism ages. P088387-Ch08.qxd 10/31/05 11:27 AM Page 224

224 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

III. From Physiology to demographic and physiological parame- Demography ters, the results have sometimes been counterintuitive. Recently, Reznick In his book on the evolution of aging, et al. (2004) found that guppies from Michael Rose defines senescence as “a high-predation environments had higher persistent decline in the age-specific fit- rates of decline in neuromuscular func- ness components of an organism due to tion, but they lived longer and had lower internal physiological deterioration” rates of reproductive aging than guppies (our italics, page 20, Rose, 1991). Most from low-predation environments. Thus, evolutionary biologists studying senes- there may not be a simple one-to- cence have focused on the decline in age- one relationship between physiological specific fitness components (mortality senescence and demographic decline. and fecundity). Whereas physiologists The task of mapping out the relation- have focused on age-related changes in ship between genotype, physiological a wide array of physiological systems senescence and demographic senescence (Masoro, 1995), much less attention has is no small challenge, but drawing these been devoted to testing the hypothesis connections is crucially important. In the that demographic senescence is due to coming decades, the social and medical internal physiological deterioration (see costs associated with physiological Williams, 1999), or to exploring the pos- decline in aging humans will increase rap- sibility that physiological homeostasis idly. We may gain most if we turn at least may even limit the senescent decline in some of our attention from demographic survival or fecundity (Kowald & quantity of life to physiological quality of Kirkwood, 1996). life in evolutionary studies of aging. We In standard evolutionary models, the see three primary areas where a more currency that measures how likely a integrated approach, one that unites gene is to make it into subsequent gener- demography and physiology, can lead to a ations is made up solely of age-specific more comprehensive understanding of the survival and fecundity (Equation 1). biology of aging. In particular, we need to Evolutionary biologists are interested in (1) develop more physiologically based changes in gene frequencies over time, theoretical models of senescence; (2) use so it seems logical to measure senes- classical quantitative genetic approaches cence by observing age-specific declines to determine whether the genes that in survival and fecundity. Until recently, determine rates of physiological decline the physiological factors that are pre- are the same ones that determine rates of sumably the proximate cause of age- demographic decline; and (3) include both related changes in survival or fecundity physiological and demographic measures have been considered of secondary in molecular studies that are searching for importance among evolutionary geron- specific genes and gene pathways that can tologists. In fact, some models suggest slow the aging process. In the following demographic senescence may evolve section, we use the term physiology in its even in the absence of physiological broadest sense—that is, the overall func- decline. Houston and McNamara (1999) tioning of an organism. This includes showed that mortality rates can increase organ performance, system function such with age solely as a result of how indi- as immunity (see Section IV), cell mor- viduals optimize reproductive effort, and phology, organismal behavior, and so without any age-related deterioration in forth. We need a broad definition as these physiological state. Where evolutionary non-demographic traits may overlap and biologists have looked at both interact in complex ways to ultimately P088387-Ch08.qxd 10/31/05 11:27 AM Page 225

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 225

determine the fitness parameters of age- could be both functionally and geneti- specific fertility, fecundity, and survival. cally independent (see Figure 8.2c). In a study on age-related changes in heart failure rate in Drosophila, Wessells and col- A. Physiological Models of Senescence leagues (2004) showed that long-lived Within the theoretical literature on insulin signaling mutants had much aging, not all studies have ignored physi- lower rates of heart failure. Although this ology. For example, Mangel (2001) incor- study demonstrates that a gene that influ- porated an organism’s energy level and ences mortality rates also affects a physi- the accumulated level of metabolic dam- ological parameter, we do not know if age as physiological states into life- heart failure plays any causal role in history models. These models predict aging and/or death in fruit flies (the case how caloric restriction and reproduction illustrated in Figure 8.2b). affect the shape of the mortality curve. A physiologically structured model by Mangel and Bonsall (2004) predicts that the actual shape of the mortality curve can depend on physiological processes, such as growth and the level of repair. As information becomes available about the genetic control of declining organ function (e.g., Wessells et al., 2004), we should be able to construct ever- more realistic models for the way in which physiological senescence relates to demographic senescence.

B. Quantitative Genetic Analyses of Physiology and Demography We discussed earlier the need to develop a more complex and subtle model for the genetic architecture of aging. We would further suggest that a complete understanding of the genetics of demographic aging must include physiology. But the genetic relationship between physiological and demographic senescence may be complicated. Physiological senescence may be the functional intermediary between so-called “longevity genes” and Figure 8.2 The influence of physiology in the genetics of senescence. The figure presents a demographic senescence (see Figure 8.2a). schematic for three possible scenarios: (a) Genes Alternatively, the same genes may regu- influence physiological processes, which then lead late rates of senescence in both demogra- to downstream effects on demographic senescence phy and certain physiological processes in mortality or fecundity. (b) The same genes that independently, such that the demo- influence senescence in physiological processes independently determine rates of senescence in graphic decline may not be specifically demographic traits. (c) Different genes determine caused by those physiological processes rates of senescence in physiological and demo- (see Figure 8.2b). Last, the two processes graphic processes. P088387-Ch08.qxd 10/31/05 11:27 AM Page 226

226 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

Quantitative genetic studies have when physiological processes are incor- demonstrated that there is genetic porated into Hamilton’s (1966) theoreti- variation for the rate of decline cal model? in age-specific survival (Hughes & Charlesworth, 1994; Promislow et al., C. Molecular Genetic Analyses of 1996) and fecundity (Tatar et al., 1996). Physiology and Demography There is ample evidence to suggest that there is genetic variation for physiolog- As with evolutionary geneticists, molec- ical traits in diverse taxa, from dairy ular biologists working on aging have cattle to Drosophila (Kiddy, 1979; also tended to focus on demographic Zera & Harshman, 2001). And we also traits. This has been motivated in large know that genetic variation in longevity part by the search for genes that will is correlated with underlying physio- make organisms live longer, no matter logical differences (e.g., Djawdan etal., what the physiological state of the ani- 1998; Gibbs et al., 1997). But informa- mal. Of course, most biogerontologists tion about genetic variation for the age- hope to find ways to increase not only related rate of decline in physiological the quantity of life, but also the quality function is rare (Wessells et al., 2004), of life at late ages (Arantes-Oliveira and we know little about the genetic et al., 2003). But to do this, we need a correlation with longevity and mortal- balanced research program that includes ity parameters. both demographic and physiological per- To map the relation between genotype spectives. and physiological and demographic Fortunately, we are beginning to see senescence, we need to address a series of just such a shift in focus. For example, specific questions: Do genotypes that Huang and colleagues (2004) found show a fast decline in physiological traits that the age-related decline in pharyn- also have a higher rate at which intrinsic geal pumping and body movement in mortality increases with age? If an inter- C. elegans was positively correlated with vention extends life span through a life span among a series of mutants. This decrease in the rate of increase in age- suggests either that the decline in physio- specific mortality rate, does it also logical processes causes a reduction in decrease the rate of physiological deterio- survival probability (see Figure 8.2a), or ration? Alternatively, if an intervention that the declines in physiological function extends life span through a decrease in and survival are regulated by a shared the initial mortality rate, does the rate of mechanism (see Figure 8.2b). Future work physiological deterioration remain the should focus on attempts to test these same (see Chapter 1, Gavrilov and two hypotheses explicitly and should Gavrilova)? Is there a shared regulatory include demographic parameters other system that mediates the rate of deterio- than longevity. Interestingly, Huang and ration for all physiological traits as pre- colleagues (2004) found that the age- dicted by Williams (1957), or is the related decline in pharyngeal pumping genetic basis for the rate at which and body movement were not correlated one trait deteriorates independent of with self-fertile reproductive span. The the genetic basis for the rate at authors speculate that other measures of which another trait deteriorates? And reproductive aging may correlate with finally, if the link between physiology physiological decline. Alternatively, genes and demography is not as straightforward that affect aging may have pleiotropic as generally assumed, how does the effects on body movement and survival, shape of the selection curve change but not on fecundity. P088387-Ch08.qxd 10/31/05 11:27 AM Page 227

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 227

We are also likely to see substantial lapping effects on survival and immune progress from studies of cellular physiol- function. For instance, the secretion of ogy. Using a histological approach to study juvenile hormone (JH) is crucial for sev- aging in C. elegans, Herndon and col- eral important reproductive pathways in leagues (2002) found that long-lived age-1 insects, including gametogenesis and sper- mutants showed a slower rate of deteriora- matophore production (Wigglesworth, tion in cell ultrastructure. Interestingly, 1965). However, increased JH titers have these lower rates were only seen in certain been shown to decrease immune function cell types. in the mealworm beetle (Rantala et al., Taken together, these two studies in 2003; Rolff & Siva-Jothy, 2002) and to C. elegans show that physiology is corre- decrease longevity in monarch butterflies lated with some but not all demographic (Herman & Tatar, 2001). This apparent traits (Huang et al., 2004), and that physiological antagonism between repro- demography is correlated with some but duction, immune function, and survivor- not all physiological traits (Herndon et al., ship may play an important role in how 2002). The challenge now before us is for insects age. theoretical, quantitative, and molecular Although the proximate effects of para- geneticists to develop an integrated sites on longevity are often clear and research program that incorporates genet- straightforward, there are more subtle ics, physiology and demography to create and interesting relationships between a more integrated research program in hosts and their parasites that may biogerontology. develop over evolutionary time. Parasites may play an important role in the evolution of a striking variety of biological IV. Parasites and Immune traits, including the existence of sexual Function reproduction (Hamilton, 1980); the dramatic, dimorphic coloration in birds In the previous section, we argue that a (Hamilton & Zuk, 1982); and the ability critical challenge for evolutionary geron- of organisms to invade novel habitats tologists is to bridge the gap between (Torchin et al., 2003; Wolfe, 2002). And physiology and demography in studies of recent work on parasites and life-history aging, and we propose some explicit ways strategies (Rolff & Siva-Jothy, 2003; in which this might be accomplished. Williams & Day, 2001) suggests that para- One candidate for a ubiquitous factor that sites may influence the way that natural might tie together physiological senes- selection shapes patterns of senescence. cence and demographic senescence is parasites. A. From Immunosenescence to We all have first-hand experience of the Demographic Senescence deleterious effects of parasites, and not surprisingly, there is abundant evidence We discussed the possibility that physio- for their life-shortening effects in model logical decline may give rise to decline in organisms. Likewise, when individuals fitness traits. One potentially important are deprived of their normal bacterial flora physiological factor is the age-related they often show increased life span (e.g., decline in immunocompetence, or Croll et al., 1977; Garigan et al., 2002; “immunosenescence” (Walford, 1969). We Houthoofd et al., 2002; Larsen & Clarke, have a pretty good idea of the proximate 2002; Min & Benzer, 1997). Recently, causes of immunosenescence in humans. researchers have begun to identify molec- For example, the thymus, where T cells ular pathways that appear to have over- mature, exhibits degenerative changes P088387-Ch08.qxd 10/31/05 11:27 AM Page 228

228 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

throughout life. Consequently, thymic tis- demands of an aging immune system. sue loses the capacity to influence a vari- Alternatively, the immune changes seen ety of important immune functions, late in life may merely be a superficial including the repopulation of T cells marker of other underlying causes of in the lymph nodes (Hirokawa & aging. If so, then immunosenescence Makinodan, 1975). Similarly, the sources may serve as a useful biomarker of of B cells, which fine-tune an antigen physiological age for future studies. match for invading pathogens, begin to disappear over time (Leslie, 2004). This B. Immunocompetence Tradeoffs decline, coupled with a change in T-cell population with age, may lead to a Hamilton showed that the age-related decreased antibody response to most anti- decline in the strength of selection can be gens (Weksler & Schwab, 1992). These predicted solely on the basis of age-specific age-specific changes in the immune sys- survival and fecundity (see Equations 2 and tem mark a dramatic physiological decline 3). Previous models have argued that these late in life that may greatly contribute to two traits should be negatively correlated patterns of demographic senescence. because the amount of resources that can The cellular details differ, but this gen- be invested in both is finite (de Jong & van eral pattern of immunosenescence is Noordwijk, 1992). This tradeoff underlies seen across an impressively broad array Kirkwood’s disposable soma theory for the of species, including fish, birds, and rep- evolution of senescence (Kirkwood, 1977). tiles (Torroba & Zapata, 2003). Insect However, this framework may be incom- species, including bumble bees, crickets, plete. We suggest here that in addition to and dragonflies, also exhibit a marked investment in survival and reproduction, it age-related change in a variety of may be worth considering other intermedi- immune components, often leading to ate physiological components as distinct increased rates of parasitism and mortal- model elements, such as neuron, circula- ity (Adamo et al., 2001; Doums et al., tory, or immune function. For instance, if 2002; Rolff, 2001). Even in the nematode, investment in immune function is directly C. elegans, a positive association between correlated with both age-specific reproduc- age and susceptibility to bacterial infection and age-specific survival (e.g., if the tion is found (Laws et al., 2004). Whether physiological costs of immunity affect a causal relationship exists between fecundity and survival simultaneously), immunosenescence and the age-specific immune function may account for the cor- decline in fitness, however, is currently relation between fecundity and survival. unknown. This, in turn, will influence the age-related Immune senescence could contribute decline in the strength of selection. to a general physiological decline if, To avoid parasites, potential hosts invest as the immune system ages, it requires in a variety of defenses, some of which pre- increased resources to maintain the sta- vent infection in the first place and others tus quo. Consequently, there may be of which fight off the parasite once fewer resources available for other costly an infection has taken hold. But these physiological systems. For example, in defenses can be costly. For example, the collared flycatcher (Ficedula albicol- increasing investment in immunity is lis), older females tend to suffer from a often paid for with lower fecundity (Zuk & decline both in humoral immune func- Stoehr, 2002). Conversely, among individu- tion and offspring size (Cichon et al., als who increase their investment in repro- 2003). However, more work is needed duction, we see a decrease not only in to determine whether the decline in off- survivorship (Partridge & Harvey, 1985), spring size is due to the increased energy but also in immunocompetence (e.g., P088387-Ch08.qxd 10/31/05 11:27 AM Page 229

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 229

Fedorka et al., 2004). Most examples of ies have shown that many of the genes these immune costs of reproduction come that affect longevity play an integral role from insects. However, a recent study in in immune defense (Caruso et al., 2001; women from 18th- and 19th-century Garsin et al., 2003; Ivanova et al., 1998; Finland (Helle et al., 2004) found Lagaay et al., 1991; Laws et al., 2004). In that women who had born twins at some some cases, immunity may be the proxi- time in their reproductive life span were mate mechanism that gives rise to genes more likely to succumb to an infectious with antagonistic pleiotropic effects. disease after menopause than women who For example, human centenarian stud- had only given birth to singletons. This ies have uncovered a strong associa- result held even after controlling for the tion between Major Histocompatibility total number of offspring born to each Complex (MHC) haplotypes and life span woman. (Caruso et al., 2001; Ivanova et al., 1998; There is still much work to be done as Lagaay et al., 1991). At least one human we try to sort out how costs of parasitism leukocyte antigen haplotype, 8.1 AH, and immunity are translated into the cur- may provide a selective advantage in rency of demographic senescence. One early infancy by providing protection important problem will be to determine from infectious disease (Caruso et al., how the physiological costs of mounting 2000). However, this haplotype is also an immune response and the mortality associated with susceptibility to several costs of being parasitized change with age. autoimmune disorders that occur during Such changes can have complicated and reproductive adulthood, such as sarcoido- nonintuitive consequences on mortality sis and systemic lupus erythematosus rates. In traditional models of either (Lio et al., 1997; Price et al., 1999). host–parasite coevolution or senescence, Interestingly, this haplotype also provides as background mortality rates increase, a late-life, sex-specific advantage for male virulence and rates of aging also increase, carriers, whereas female carriers continue respectively. In two separate theoretical to suffer from early morbidity and mor- studies, Williams and Day (2001; 2003) tality late in life (Caruso et al., 2000). demonstrate that when extrinsic sources Such tradeoffs may turn out to be quite of mortality interact with the host’s phys- common for genes associated with iological state in a nonadditive fashion, immune function. Another example comes virulence and rates of aging may actually from the MHC mutation C28Y. This decrease as background mortality mutation increases the intestinal absorp- rates increase. We now need theory that tion of iron, which is crucial in many combines Williams and Day’s models of immune pathways (Salter-Cid et al., 2000). parasitism (Williams & Day, 2001) and However, the mutation is also associated senescence (Williams & Day, 2003) and with haemochromatosis in homozygotes, a empirical tests of the existing theory. disease characterized by a reduced life Such models should help us determine expectancy due to an excess accumulation how selection should shape patterns of of iron in the organs and concomitant age-specific investment in immunity and reduction in immunity (Waheed et al., age-specific survival, and how the two 1997). Researchers have also suggested may interact. relationships between tuberculosis and Tay-Sachs disease (Spyropoulos, 1988), and cholera and cystic fibrosis (Gabriel C. Parasites and the Genetics of Aging et al., 1994). In the past few years, biolo- Studies of aging and immunity may also gists have begun to focus on the evolution- further our understanding of the genetic ary consequences of immunity-related architecture of senescence. Recent stud- tradeoffs (reviewed in DeVeale et al., 2004), P088387-Ch08.qxd 10/31/05 11:27 AM Page 230

230 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

although a detailed connection to senes- inhibits the pro-inflammatory immune cence has yet to be made. Just how com- response, which is believed to contribute mon these tradeoffs are, and whether some substantially to mortality in late life. tradeoffs show age-specificity consistent These patterns are not surprising consid- with the antagonistic pleiotropy theory of ering that sex differences in immune senescence, remains to be seen. Finding function are widespread. Male mammals such genes will provide an interesting chal- often have a lower level of immunocom- lenge for those interested in developing an petence (e.g., Klein & Nelson, 1997) and a evolutionary perspective on immunosenes- higher rate of parasitic infection when cence. compared to females (Moore & Wilson, Parasites may turn out to influence the 2002). Invertebrate males show a similar genetics of aging in an even more general pattern (Adamo et al., 2001; Kurtz & fashion. A recent study found that the Sauer, 2001; Kurtz et al., 2000; Radhika ability of some genes to increase longevity et al., 1998; Rolff, 2001), although due to depended on the presence of the bacterial the complex nature of life-history trade- flora (Brummel et al., 2004). For example, offs, sex differences in immune defense flies carrying the DJ817 mutant, which are often difficult to predict (Doums are normally long-lived, lost their life et al., 2002; Moret & Schmid-Hempel, span advantage when treated with anti- 2000; Zuk et al., 2004; Zuk & Stoehr, biotics. The same study showed that 2002). These sex-specific patterns suggest D. melanogaster males deprived of their that the selection pressures that influ- normal bacterial complement in early ence immunity and longevity are differ- adulthood suffered a significant reduction ent for males and females. in life expectancy. In light of this interest- We have suggested here that sex- ing set of results, we should now set out to specific differences in immune function determine how many genes associated may account, in part, for sex differences with longevity depend on the presence of in longevity. In the following section, we bacteria for their phenotype, and of the move away from immunity and explore many bacteria found in Drosophila, which causes of sex-related differences in ones are early-age symbionts and why. longevity in greater detail.

D. Parasites and Sex Differences V. Sex, Sexual Selection, in Longevity and Sexual Conflict Finally, studies of parasites, immunity, and senescence may help us to under- Sex differences in longevity are common. stand why males and females often have Female mammals generally live longer quite different patterns of aging. Studies than their male counterparts (Promislow, of immunosenescence have found that 1992), with a few interesting exceptions, many of the MHC haplotypes associated such as among anthropoid primates with with longevity are sex-dependent (Lagaay extended paternal care, where males live et al., 1991; Lio et al., 2002). Ivanova and longer than females (Allman et al., 1998). colleagues (1998) found that of the three Conversely, in birds (Promislow et al., MHC alleles linked to human longevity 1992) and nematodes (McCulloch & in their study, two had a sex-dependent Gems, 2003), males tend to outlive effect. Similarly, Lio and colleagues females. Many specific mechanisms have (2002) found a male-biased pattern among been proposed to account for these pat- centenarians for polymorphisms in the terns, from differences in hormon profiles promoter region of the IL-10 gene. IL-10 (e.g., testosterone levels in males) to P088387-Ch08.qxd 10/31/05 11:27 AM Page 231

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 231

differences in predation risk (e.g., vulnera- able for immune defense or other traits, bility of nesting females). leading to reduced survival (Andersson, It is likely that at least part of these dif- 1994). These changes in survival rates can ferences is due to a basic disparity in then lead to sex-specific differences in the reproductive strategies. A female’s optimal declining force of selection with age. At reproductive strategy may be very differ- this point, we need sex-specific models to ent than that of her male partner (Rice, help us determine how sex differences in 1996). These differences may lead to very the risk of mortality might lead to sex different age-specific selection pressures differences in rates of aging. acting on the two sexes. We are only beginning to understand the full implica- B. Female Mate Choice tions of these differences. In the following section, we describe three specific areas There is extensive evidence from differ- that could further our understanding of ent species suggesting that females will sex-specific patterns of senescence, includ- often choose to mate with males based ing costs of reproduction, the proximate on their genetic quality (Andersson, and evolutionary consequences of female 1994). In so doing, choosy females may choice, and intersexual conflict. affect the evolution of senescence. For instance, according to Hamilton and Zuk (1982), female birds may use a male’s A. Cost of Reproduction bright coloration to assess his underly- Reproduction can increase the mortality ing ability to resist parasitic infection. rate in many organisms (e.g., Fedorka If female choice thereby increases et al., 2004; Sgrò & Partridge, 1999), but immunocompetence in the population, these costs can differ dramatically mean survival rates may increase, and between the sexes in a range of organisms selection on senescence will change (Lyons & Dunne, 2003; Michener & accordingly. Locklear, 1990; Rocheleau & Houle, Female choice may have a more direct 2001). Males and females may differ in effect on senescence in a population if the resources they allocate to gamete pro- females prefer to mate with males of a duction, parental care, and mating effort. particular age. Some theoretical studies Dissimilar reproductive investment can, have suggested that females might choose in turn, lead to differences in the risk of to mate with older males because they predation or sexually transmitted disease, have proven their ability to live long or in the resources available for somatic (Beck & Powell, 2000; but see Hansen & maintenance. The evolution of sexually Price, 1995; Kokko, 1998). In their com- dimorphic traits through female choice or puter simulations, Beck and colleagues male competition exemplifies these sex- (2002) found that when females were specific costs. To attract females or out- allowed to choose males based on their compete other males for access to mates, age, they typically evolved a preference males often develop exaggerated second- for older males. In a genetically hetero- ary sexual characteristics, such as bright geneous population, older males will have plumage or coloration, conspicuous call- a higher proportion of alleles associated ing songs, or large antlers. However, with increased life span than younger these traits may also attract predators or males. Thus, female preference for older encumber the male when trying to escape males leads not only to lower mortality from harm. Furthermore, these traits rates in the choosy female’s offspring, but come at a large physiological cost and will actually increase mean life span in may reduce the resources that are avail- the entire cohort over evolutionary time. P088387-Ch08.qxd 10/31/05 11:27 AM Page 232

232 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

The theoretical findings that choosiness wall with needle-like genitalia, leading to can lead to increased longevity are consis- high rates of female mortality (Stutt & tent with experimental studies in Siva-Jothy, 2001). Conflict can even take Drosophila. Promislow and colleagues place at a molecular level. In Drosophila (1998) compared mortality rates in strains spp., males pass accessory gland proteins that were maintained under enforced (Acps) during copulation that may monogamy for multiple generations decrease a female’s sexual receptivity, with strains where female choice and incapacitate previously donated sperm, male–male competition were allowed. prevent future sperm displacement (see They found that the strains with higher Chapman et al., 1995, and references levels of sexual selection evolved lower therein), and increase the rate of oviposi- mortality rates. Similarly, among birds, tion (Heifetz et al., 2000). Moreover, Acps female survival rates are highest in those tend to decrease a female’s overall life species that appear to have been under expectancy in a dose-dependent manner stronger sexual selection (Promislow (Chapman et al., 1995; Lung et al., 2002). et al., 1992). Further empirical work is And tying Acps more directly to needed to determine whether female longevity, Moshitzky and colleagues preference for older males, in particular, (1996) have shown that one of the sex can alter patterns of senescence in natural peptides, a protein found in the male ejac- populations. ulate, can induce the biosynthesis of juvenile hormone, which is associated with longevity (see Chapter 15, Tu et al.). C. Sexual Conflict Further experiments are needed to Sexual conflict has also been suggested determine whether sexual conflict can as playing a significant role in the evolu- affect rates of senescence and differences tion of senescence (Promislow, 2003; in patterns of senescence between the Svensson & Sheldon, 1998). Sexual con- sexes. Are rates of aging higher in species flict arises when the optimal reproduc- with greater sexual conflict? Do species tive strategy differs between the sexes. with higher levels of conflict show greater For example, a female may enhance her sex-differences in rates of aging than fitness by mating multiple times with species with lower levels of conflict? And different males, but this may reduce the are genes associated with conflict (such as fitness of each of her mates by diluting Acps in Drosophila) also associated with his sperm with those of his rivals. These variation in longevity in either sex? conflicts can give rise to a situation where males, in an effort to maximize their fitness, evolve the capacity to dra- VI. Genetic Variation in Natural matically reduce female fitness. Populations We have long known that male behavior can be detrimental to females: Parker In a separate chapter in this and Thompson (1980) observed that, in volume, Brunet-Rossinni and Austad (see the quest for mating opportunities, com- Chapter 9) examine evidence for senes- peting male dung flies (Scatophaga sterco- cence in natural populations. Both lab raria) would often drown potential mates, and natural populations generally show leading to an obvious conflict of interest similar Gompertz-like increases in mor- between the dying female and the tality. However, several studies have overzealous male. Similarly, male bed shown that when we bring organisms bugs (Cimex lectularius) forcibly insemi- into a lab environment, we often inad- nate females by piercing their abdominal vertently expose them to novel selection P088387-Ch08.qxd 10/31/05 11:27 AM Page 233

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 233

pressures that lead to shorter life span argue that the life-extending effects of [Sgrò & Partridge, 2000; Linnen et al., novel mutants or artificial selection sim- 2001]. Thus, what we learn about the ply restore short-lived lab strains to the genetics of aging in the lab may not longevity of their wild relatives (Linnen always transfer to the wild. et al., 2001). For many years, senescence was Recent studies from worms and flies assumed to be rare in the wild (Comfort, point to two specific factors that we need 1979) because organisms were most likely to consider before assuming that lab to die by accident before they would have results necessarily apply to wild popula- a chance to senesce. However, later stud- tions. First, the effect of a mutation can ies demonstrated that senescence is not depend on the environment in which an an artifact created by ideal laboratory con- organism lives, due to gene by environ- ditions but is actually common in natural ment (G E) interaction. The impor- populations of mammals (Promislow, tance of G E interactions in the genet- 1991) and birds (Ricklefs, 1998; 2000). We ics of longevity has been explored in great even see evidence for senescence in natu- detail by Mackay and her co-workers ral populations of insects. Bonduriansky (Chapter 7, Mackay et al.). Second, some and Brassil (2002) demonstrated aging in studies appear to find life extension at no wild populations of very short-lived male cost. For instance, the daf-2 mutation in antler flies, and Carey (2002) was even C. elegans can greatly extend longevity in able to find evidence of aging in the the laboratory without apparent loss of famously short-lived mayfly. These new fertility or activity (Kenyon et al., 1993). studies on aging in nature in short-lived However, when the long-lived daf-2 animals suggest the possibility for devel- mutant is combined with the wildtype oping free-living model systems to study strain in the same culture, the long-lived the genetics of aging. mutant is always outcompeted (Jenkins A few studies have demonstrated a et al., 2004). In general, it may turn out genetic basis to variation in rates of aging that when placed in a natural environ- in natural populations (e.g., Bronikowski ment, genes that extend life span in the et al., 2002; Reznick et al., 2004; Tatar lab may have unanticipated negative con- et al., 1997). Not surprisingly, most of sequences for fitness. what we know about the genetics of Several studies have now tried to deter- aging comes from lab studies. But we mine whether alleles associated with need to be especially cautious in assum- longevity in the lab have similar func- ing that what we learn in the lab trans- tions in natural populations. Schmidt and lates directly to the field. We usually colleagues (2000) showed that there is assume that an allele that increases geographic variation in the frequency longevity in a lab incubator will of methuselah, a Drosophila gene that also increase longevity in the wild. can substantially increase longevity (Lin Unfortunately, when we introduce natu- et al., 1998). However, clinal variation ral populations into the lab, the change in longevity was not correlated with cli- in environment can select for quite dra- nal variation in single nucleotide poly- matic changes in demographic character- morphisms at the methuselah gene. istics. Lab-adapted organisms typically Other studies have had greater luck in mature earlier and have increased early- applying lab-based results to field popula- age fecundity and greatly shortened life tions. Geiger-Thornsberry and Mackay span (Clark, 1987; Houle & Rowe, 2003; (2004) studied genetic variation for genes Miller et al., 2002; Promislow & Tatar, associated with aging in the lab in wild- 1998; Sgrò & Partridge, 2000). One might derived inbred strains of Drosophila. P088387-Ch08.qxd 10/31/05 11:27 AM Page 234

234 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

Using quantitative complementation studying short-lived species, studying tests (see Chapter 7, Mackay et al.), long-lived organisms in the wild (or the they found that there was standing lab, for that matter) could add a new genetic variation at these loci in natural dimension to aging research (Holmes & populations, and that this genetic varia- Ottinger, 2003). tion was correlated with variation in longevity among strains. We now need to determine whether Geiger-Thornsberry VII. Conclusions and Mackay’s findings with the InR and Adh genes are the exception or the rule, The most important conceptual advances and how selection has shaped these loci in our understanding of the evolution of over evolutionary time. senescence came from the initial models It is clearly important to determine (Medawar, 1946; 1952; Williams 1957). whether genes identified in lab-reared Since the time of Medawar’s and populations of flies are effective at extend- Williams’s pioneering efforts, a half- ing life span in natural populations. But century of evolutionary research has led to an even greater challenge beckons. Can extraordinary advances in our understand- we translate lab-based findings to natural ing of evolutionary patterns and processes, populations of different taxa? Can the from adaptation to altruism, from specia- genetic results for flies or worms in the tion to sex ratios. However, with few lab be applied to natural populations of exceptions, these enormous conceptual vertebrates? Researchers have suggested advances have occurred independently of that various fish species, including killi- evolutionary studies of aging. fish (Herrera & Jagadeeswaran, 2004), In this chapter, we have explored five zebrafish (Gerhard, 2003; Gerhard & areas that have excellent potential to Cheng, 2002), and guppies (Reznick, further our understanding of the evolu- 1997), may be ideal model systems to help tion of senescence. These include the bridge the gap between invertebrate and genetic architecture of senescence, the mammalian taxa in studies of aging. relationship between physiological and Reznick recently used a natural popula- demographic decline, the importance of tion of guppies to test the hypothesis that parasites and the immune system, sex rates of senescence should be highest in differences in behavior and aging, and populations with high extrinsic mortality studies of aging in the wild. Some of (Reznick et al., 2004). Interestingly, the these areas include those in which we data did not support this classic hypothe- have already seen important advances by sis, which reiterates the point we made evolutionary biologists working outside earlier in this chapter that we need more of the field of aging. All are linked by the biologically realistic models for the evolu- general problem of which forces will tion of senescence. shape the age-related decline in the force Researchers have also argued that birds of selection. (Holmes et al., 2001), bats (Wilkinson & We hope that the questions that we South, 2002), and other “slow-aging” have posed here will encourage more evo- organisms (Austad, 2001) may be valuable lutionary biologists to take on the prob- model systems for the study of senes- lem of the evolution of senescence. In the cence. By extending the range of taxa that meantime, most of the action in we use to study aging, we should be able biogerontological research appears to be to determine the extent to which mecha- found in the molecular labs, where new nisms of aging are shared among taxa. genes and gene pathways that affect aging Although there are distinct advantages to are being uncovered at an extraordinary P088387-Ch08.qxd 10/31/05 11:27 AM Page 235

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 235

rate. One might ask, then, whether there survival in anthropoid primates: caretakers is still a need for an evolutionary perspec- live longer. Proceedings of the National tive in this fast-paced world of sequenced Academy of Sciences of the USA, 95, genomes and microarrays. We believe that 6866–6869. the most exciting challenge facing all of us Andersson, M. (1994). Sexual selection: monographs in behavior and ecology. is to develop an integrated research pro- Princeton, NJ: Princeton University Press. gram, where molecular and evolutionary Arantes-Oliveira, N., Berman, J. R., & biologists work hand in hand—molecular Kenyon, C. (2003). Healthy animals with biologists asking evolutionary questions, extreme longevity. Science, 302, 611. evolutionary biologists using state-of-the- Austad, S. N. (2001). An experimental art molecular tools. With such an integra- paradigm for the study of slowly-aging tion, as old problems are solved, new and organisms. Experimental Gerontology, 36, interesting questions will arise at an even 599–605. faster rate, ensuring that the study of aging Beck, C. W., & Powell, L. A. (2000). Evolution continues to be exciting and enlightening. of female mate choice based on male age: are older males better mates? Evolutionary Ecology Research, 2, 107–118. Acknowledgements Beck, C. W., Shapiro, B., Choksi, S., & We thank Thomas Flatt and members of the Promislow, D. E. L. (2002). A genetic Promislow lab for thoughtful discussions. algorithm approach to study the evolution Richard Miller, Steve Austad, Edward of female preference based on male age. Masoro, and one anonymous reviewer pro- Evolutionary Ecology Research, 4, 275–292. vided valuable comments on a previous draft Bonduriansky, R., & Brassil, C. E. (2002). of this manuscript. This work was supported Rapid and costly ageing in wild male flies. by an NSF Minority post-doctoral fellowship Nature, 420, 377–377. from the National Science Foundation (KF) Bronikowski, A. M., Alberts, S. C., Altmann, J., and a grant from the Ellison Medical Packer, C., Carey, K. D., & Tatar, M. (2002). Foundation (DP). The aging baboon: comparative demography in a non-human primate. Proceedings of the References National Academy of Sciences of the USA, 99, 9591–9595. Abrams, P. A. (1993). Does increased Brummel, T., Ching, A., Seroude, L., mortality favor the evolution of more rapid Simon, A. F., & Benzer, S. (2004). Drosophila senescence? Evolution, 47, 877–887. life span enhancement by exogenous bacteria. Abrams, P. A., & Ludwig, D. (1995). Proceedings of the National Academy of Optimality theory, Gompertz’ law, and the Sciences of the USA, 101, 12974–12979. disposable-soma theory of senescence. Campisi, J. (2001). From cells to organisms: can Evolution, 49, 1055–1066. we learn about aging from cells in culture? Adamo, S. A., Jensen, M., & Younger, M. Experimental Gerontology, 36, 607–618. (2001). Changes in lifetime Campisi, J. (2003). Cellular senescence and immunocompetence in male and female apoptosis: how cellular responses might Gryllus texensis (formerly G. integer): trade- influence aging phenotypes. Experimental offs between immunity and reproduction. Gerontology, 38, 5–11. Animal Behaviour, 62, 417–425. Carey, J. R. (2002). Longevity minimalists: life Albert, R., & Barabási, A.-L. (2002). Statistical table studies of two species of northern mechanics of complex networks. Reviews Michigan adult mayflies. Experimental of Modern Physics, 74, 47–97. Gerontology, 37, 567–570. Albert, R., Jeong, H., & Barabasi, A. L. (2000). Caruso, C., Candore, G., Colonna Romano, G., Error and attack tolerance of complex Lio, D., Bonafe, M., Valensin, S., & networks. Nature, 406, 378–382. Franceschi, C. (2000). HLA, aging, and Allman, J., Rosin, A., Kumar, R., & longevity: a critical reappraisal. Human Hasenstaub, A. (1998). Parenting and Immunology, 61, 942–949. P088387-Ch08.qxd 10/31/05 11:27 AM Page 236

236 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

Caruso, C., Candore, G., Romano, G. C., Edney, E. B., & Gill, R. W. (1968). Evolution of Lio, D., Bonafe, M., Valensin, S., & senescence and specific longevity. Nature, Franceschi, C. (2001). Immunogenetics of 220, 281–282. longevity. Is major histocompatibility Eisen, J. A., & Hanawalt, P. C. (1999). A complex polymorphism relevant to the phylogenomic study of DNA repair genes, control of human longevity? A review of proteins, and processes. Mutation Research, literature data. Mechanisms of Ageing and 435, 171–213. Development, 122, 445–462. Euler, L. (1760). Rechèrches générales sur la Chapman, T., Liddle, L. F., Kalb, J. M., mortalité et la multiplication. Mémoires de Wolfner, M. F., & Partridge, L. (1995). Cost l’Académie Royale des Sciences et Belles of mating in Drosophila melanogaster is Lettres, 16, 144–164. mediated by male accessory gland products. Fedorka, K. M., Zuk, M., & Mousseau, T. A. Nature, 373, 241–244. (2004). Immune suppression and the cost of Charlesworth, B. (2001). Patterns of age- reproduction in the ground cricket, specific means and genetic variances of Allonemobius socius. Evolution, 58, mortality rates predicted by the mutation 2478–2485. accumulation theory of aging. Journal of Finch, C. E., & Pike, M. C. (1996). Maximum Theoretical Biology, 210, 47–65. life span predictions from the Gompertz Charlesworth, B., & Hughes, K. A. (1996). mortality model. Journal of Gerontology Age-specific inbreeding depression and Series A Biological Sciences and Medical components of genetic variance in relation Sciences, 51, B183–B194. to the evolution of senescence. Proceedings Fisher, R. A. (1930). The genetical theory of of the National Academy of Sciences of the natural selection. Oxford: Clarendon Press. USA, 93, 6140–6145. Flatt, T. (2005). The evolutionary genetics of Cichon, M., Sendecka, J., & Gustafsson, L. canalization. Quarterly Review of Biology, (2003). Age-related decline in humoral 80, 287–316. immune function in Collared Flycatchers. Fraser, H. B., Hirsh, A. E., Steinmetz, L. M., Journal of Evolutionary Biology, 16, Scharfe, C., & Feldman, M. W. (2002). 1205–1210. Evolutionary rate in the protein interaction Clark, A. G. (1987). Senescence and the network. Science, 296, 750–752. genetic correlation hang-up. American Fry, J. D., Nuzhdin, S. V., Pasyukova, E. G., & Naturalist, 129, 932–940. Mackay, T. F. (1998). QTL mapping of Comfort, A. (1979). The biology of senescence. genotype-environment interaction for Edinburgh: Churchill Livingstone. fitness in Drosophila melanogaster. Croll, N. A., Smith, J. M., & Zuckerman, B. M. Genetics Research, 71, 133–141. (1977). Behavioral parameters in aging of Gabriel, S. E., Brigman, K. N., Koller, B. H., C. elegans. Journal of Nematology, 9, Boucher, R. C., & Stutts, M. J. (1994). 266–267. Cystic fibrosis heterozygote resistance to de Jong, G., & van Noordwijk, A. J. (1992). cholera toxin in the cystic fibrosis mouse Acquisition and allocation of resources— model. Science, 266, 107–109. genetic (co)variances, selection, and life Garigan, D., Hsu, A. L., Fraser, A. G., histories. American Naturalist, 139, 749–770. Kamath, R. S., Ahringer, J., & Kenyon, C. DeVeale, B., Brummel, T., & Seroude, L. (2002). Genetic analysis of tissue aging in (2004). Immunity and aging: the enemy C. elegans: a role for heat-shock factor and within? Aging Cell, 3, 195–208. bacterial proliferation. Genetics, 161, Djawdan, M., Chippindale, A. K., 1101–1112. Rose, M. R., & Bradley, T. J. (1998). Garsin, D. A., Villanueva, J. M., Begun, J., Metabolic reserves and evolved stress Kim, D. H., Sifri, C. D., Calderwood, S. B., resistance in Drosophila melanogaster. Ruvkun, G., & Ausubel, F. M. (2003). Long- Physiological Zoology, 71, 584–594. lived C. elegans daf-2 mutants are resistant Doums, C., Moret, Y., Benelli, E., & Schmid- to bacterial pathogens. Science, 300, 1921. Hempel, P. (2002). Senescence of immune Geiger-Thornsberry, G. L., & Mackay, T. F. C. defense in Bombus workers. Ecological (2004). Quantitative trait loci affecting Entomology, 27, 138–144. natural variation in Drosophila longevity. P088387-Ch08.qxd 10/31/05 11:27 AM Page 237

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 237

Mechanisms of Ageing and Development, Herrera, M., & Jagadeeswaran, P. (2004). 125, 179–189. Annual fish as a genetic model for aging. Gerhard, G. S. (2003). Comparative aspects of Journals of Gerontology Series A-Biological zebrafish (Danio rerio) as a model for aging Sciences and Medical Sciences, 59, research. Experimental Gerontology, 38, 101–107. 1333–1341. Hirokawa, K., & Makinodan, T. (1975). Gerhard, G. S., & Cheng, K. C. (2002). A call Thymic involution: effect on T cell to fins! Zebrafish as a gerontological model. differentiation. Journal of Immunology, Aging Cell, 1, 104–111. 114, 1659–1664. Gibbs, A. G., Chippindale, A. K., & Rose, M. R. Holmes, D. J., & Ottinger, M. A. (2003). Birds (1997). Physiological mechanisms of evolved as long-lived animal models for the study of desiccation resistance in Drosophila aging. Experimental Gerontology, 38, melanogaster. Journal of Experimental 1365–1375. Biology, 200, 1821–1832. Holmes, D. J., Fluckiger, R., & Austad, S. N. Gibson, G., & Honeycutt, E. (2002). The (2001). Comparative biology of aging in evolution of developmental regulatory birds: an update. Experimental pathways. Current Opinions in Genetics Gerontology, 36, 869–883. and Development, 12, 695–700. Houle, D., & Rowe, L. (2003). Natural selection Haldane, J. B. S. (1941). New paths in in a bottle. American Naturalist, 161, 50–67. genetics. London: Allen and Unwin. Houston, A. I., & McNamara, J. M. (1999). Hamilton, W. D. (1966). The moulding of Models of adaptive behaviour. Cambridge, senescence by natural selection. Journal of UK: Cambridge University Press. Theoretical Biology, 12, 12–45. Houthoofd, K., Braeckman, B. P., Lenaerts, I., Hamilton, W. D. (1980). Sex versus non-sex Brys, K., De Vreese, A., Van Eygen, S., & versus parasite. Oikos, 35, 282–290. Vanfleteren, J. R. (2002). Axenic growth up- Hamilton, W. D., & Zuk, M. (1982). Heritable regulates mass-specific metabolic rate, true fitness and bright birds: a role for stress resistance, and extends life span in parasites? Science, 218, 384–387. C. elegans. Experimental Gerontology, Hansen, T. F., & Price, D. K. (1995). Good 37, 1371–1378. genes and old age: do old mates provide Huang, C., Xiong, C. J., & Kornfeld, K. (2004). superior genes? Journal of Evolutionary Measurements of age-related changes of Biology, 8, 759–778. physiological processes that predict lifespan Heifetz, Y., Lung, O., Frongillo, E. A., Jr., & of C. elegans. Proceedings of the National Wolfner, M. F. (2000). The Drosophila Academy of Sciences of the USA, 101, seminal fluid protein Acp26Aa stimulates 8084–8089. release of oocytes by the ovary. Current Hughes, K. A., & Charlesworth, B. (1994). A Biology, 10, 99–102. genetic analysis of senescence in Helle, S., Lummaa, V., & Jokela, J. (2004). Drosophila. Nature, 367, 64–66. Accelerated immunosenescence in Hughes, K. A., Alipaz, J. A., Drnevich, J. M., & preindustrial twin mothers. Proceedings of Reynolds, R. M. (2002). A test of the National Academy of Sciences of the evolutionary theories of aging. Proceedings USA, 101, 12391–12396. of the National Academy of Sciences of the Herman, W. S., & Tatar, M. (2001). Juvenile USA, 99, 14286–14291. hormone regulation of longevity in Hwangbo, D. S., Gersham, B., Tu, M. P., the migratory monarch butterfly. Palmer, M., & Tatar, M. (2004). Drosophila Proceedings of the Royal Society of dFOXO controls lifespan and regulates London Series B Biological Sciences, insulin signalling in brain and fat body. 268, 2509–2514. Nature, 429, 562–566. Herndon, L. A., Schmeissner, P. J., Ivanova, R., Henon, N., Lepage, V., Charron, D., Dudaronek, J. M., Brown, P. A., Listner, K. M., Vicaut, E., & Schachter, F. (1998). HLA-DR Sakano, Y., Paupard, M. C., Hall, D. H., & alleles display sex-dependent effects on Driscoll, M. (2002). Stochastic and genetic survival and discriminate between individual factors influence tissue-specific decline in and familial longevity. Human Molecular ageing C. elegans. Nature, 419, 808–814. Genetics, 7, 187–194. P088387-Ch08.qxd 10/31/05 11:27 AM Page 238

238 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

Jackson, A. U., Galecki, A. T., Burke, D. T., & sequence divergence, gene dispensability, Miller, R. A. (2002). Mouse loci associated expression level, and interactivity are with life span exhibit sex-specific and correlated in eukaryotic evolution. Genome epistatic effects. Journal of Gerontology Research, 13, 2229–2235. Series A Biological Sciences and Medical Kurtz, J., & Sauer, K. P. (2001). Gender Sciences, 57, B9–B15. differences in phenoloxidase activity of Jenkins, N. L., McColl, G., & Lithgow, G. J. Panorpa vulgaris hemocytes. Journal of (2004). Fitness cost of extended lifespan Invertebrate Pathology, 78, 53–55. in C. elegans. Proceedings of the Royal Kurtz, J., Wiesner, A., Gotz, P., & Sauer, K. P. Society of London Series B Biological (2000). Gender differences and individual Sciences, 271, 2523–2526. variation in the immune system of the Jeong, H., Mason, S. P., Barabasi, A. L., & scorpionfly Panorpa vulgaris (Insecta: Oltvai, Z. N. (2001). Lethality and Mecoptera). Developmental and centrality in protein networks. Nature, Comparative Immunology, 24, 1–12. 411, 41–42. Lagaay, A. M., D’Amaro, J., Ligthart, G. J., Kaiser, M., Gasser, M., Ackermann, R., & Schreuder, G. M., van Rood, J. J., & Stearns, S. C. (1997). P-element inserts in Hijmans, W. (1991). Longevity and heredity transgenic flies—A cautionary tale. in humans. Association with the human Heredity, 78, 1–11. leukocyte antigen phenotype. Annals of Kenyon, C., Chang, J., Gensch, E., Rudner, A., the New York Academy of Sciences, 621, & Tabtiang, R. (1993). A C. elegans mutant 78–89. that lives twice as long as wild type. Larsen, P. L., & Clarke, C. F. (2002). Nature, 366, 461–464. Extension of life-span in C. elegans by a Kiddy, C. A. (1979). Review of research on diet lacking coenzyme Q. Science, 295, genetic variation in physiological 120–123. characteristics related to performance in Laws, T. R., Harding, S. V., Smith, M. P., dairy cattle. Journal of Dairy Science, 62, Atkins, T. P., & Titball, R. W. (2004). Age 818–824. influences resistance of C. elegans to Kimura, M. (1968). Evolutionary rate at the killing by pathogenic bacteria. FEMS molecular level. Nature, 217, 624–626. Microbiology Letters, 234, 281–287. Kirkwood, T. B. L. (1977). Evolution and Lee, R. D. (2003). Rethinking the ageing. Nature, 270, 301–304. evolutionary theory of aging: transfers, Klein, S. L., & Nelson, R. J. (1997). Sex not births, shape senescence in social differences in immunocompetence differ species. Proceedings of the National between two Peromyscus species. Academy of Sciences of the USA, 100, American Journal of Physiology, 273, 9637–9642. R655–R660. Leips, J., & Mackay, T. F. (2000). Quantitative Kokko, H. (1998). Good genes, old age and trait loci for life span in Drosophila life-history trade-offs. Evolutionary melanogaster: interactions with genetic Ecology, 12, 739–750. background and larval density. Genetics, Kowald, A., & Kirkwood, T. B. (1996). A 155, 1773–1788. network theory of ageing: the interactions Leslie, M. (2004). All pain, no gain. Science of of defective mitochondria, aberrant Aging Knowledge Environment, 27, 1–4. proteins, free radicals and scavengers in the Lewontin, R. C., & White, M. J. D. (1960). ageing process. Mutation Research, 316, Interaction between inversion 209–236. polymorphisms and two chromosome pairs Krumbiegel, I. (1929). Untersuchungen über in the grasshopper, Moraba scurra. die Einwirkung der fortpflanzung auf Evolution, 14, 116–129. Altern und Lebensdauer der Insekten: Li, W.-H. (1997). Molecular evolution. Carabas und Drosophila. Zoologische Sunderland, MA: Sinauer Associates. Jahrbuch (Anatomie und Ontogenie der Libina, N., Berman, J. R., & Kenyon, C. Tiere), 51, 111–162. (2003). Tissue-specific activities of Krylov, D. M., Wolf, Y. I., Rogozin, I. B., & C. elegans DAF-16 in the regulation of Koonin, E. V. (2003). Gene loss, protein lifespan. Cell, 115, 489–502. P088387-Ch08.qxd 10/31/05 11:27 AM Page 239

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 239

Lin, Y. J., Seroude, L., & Benzer, S. (1998). Maslov, S., Sneppen, K., Eriksen, K. A., & Extended life-span and stress resistance in Yan, K. K. (2004). Upstream plasticity and the Drosophila mutant methuselah. downstream robustness in evolution of Science, 282, 943–946. molecular networks. BMC Evolutionary Linnen, C., Tatar, M., & Promislow, D. E. L. Biology, 4, 9. (2001). Cultural artifacts: a comparison of Masoro, E. J. (Ed.) (1995). Handbook of senescence in natural, lab-adapted and Physiology (Section 11: Aging). New York: artificially selected lines of Drosophila Oxford University Press. melanogaster. Evolutionary Ecology McCulloch, D., & Gems, D. (2003). Evolution Research, 3, 877–888. of male longevity bias in nematodes. Aging Lio, D., Candore, G., Romano, G. C., Cell, 2, 165–173. D’Anna, C., Gervasi, F., Di Lorenzo, G., McDonald, J. F. (1993). Evolution and Modica, M. A., Potestio, M., & Caruso, C. consequences of transposable elements. (1997). Modification of cytokine patterns Current Opinions in Genetics and in subjects bearing the HLA-B8,DR3 Development, 3, 855–864. phenotype: implications for Medawar, P. B. (1946). Old age and natural autoimmunity. Cytokines Cell Mol. Ther. death. Modern Quarterly, 2, 30–49. 3, 217–224. Medawar, P. B. (1952). An unsolved problem Lio, D., Scola, L., Crivello, A., Colonna- in biology. London: H. K. Lewis. Romano, G., Candore, G., Bonafe, M., Michener, G. R., & Locklear, L. (1990). Cavallone, L., Franceschi, C., & Caruso, C. Differential costs of reproductive effort for (2002). Gender-specific association between male and female Richardsons ground- -1082 IL-10 promoter polymorphism and squirrels. Ecology, 71, 855–868. longevity. Genes Immun, 3, 30–33. Miller, R. A., Harper, J. M., Dysko, R. C., Lotka, A. J. (1925). Elements of physical Durkee, S. J., & Austad, S. N. (2002). biology. Baltimore, MD: Williams & Longer life spans and delayed maturation in Watkins. wild-derived mice. Experimental Biology Luckinbill, L. S., Arking, R., Clare, M. J., and Medicine, 227, 500–508. Cirocco, W. J., & Buck, S. A. (1984). Min, K. T., & Benzer, S. (1997). Wolbachia, Selection for delayed senescence in normally a symbiont of Drosophila, can be Drosophila melanogaster. Evolution, 38, virulent, causing degeneration and early 996–1003. death. Proceedings of the National Academy Lung, O., Tram, U., Finnerty, C. M., Eipper- of Sciences of the USA, 94, 10792–10796. Mains, M. A., Kalb, J. M., & Wolfner, M. F. Moore, S. L., & Wilson, K. (2002). Parasites as (2002). The Drosophila melanogaster a viability cost of sexual selection in seminal fluid protein Acp62F is a protease natural populations of mammals. Science, inhibitor that is toxic upon ectopic 297, 2015–2018. expression. Genetics, 160, 211–224. Moret, Y., & Schmid-Hempel, P. (2000). Lyons, D. O., & Dunne, J. J. (2003). Survival for immunity: the price of Reproductive costs to male and female worm immune system activation for bumblebee pipefish. Journal of Fish Biology, 62, workers. Science, 290, 1166–1168. 767–773. Moshitzky, P., Fleischmann, I., Chaimov, N., Mangel, M. (2001). Complex adaptive Saudan, P., Klauser, S., Kubli, E., & systems, aging and longevity. Journal of Applebaum, S. W. (1996). Sex-peptide Theoretical Biology, 213, 559–571. activates juvenile hormone biosynthesis in Mangel, M., & Bonsall, M. B. (2004). The the Drosophila melanogaster corpus shape of things to come: using models with allatum. Archives of Insect Biochemistry physiological structure to predict mortality and Physiology, 32, 363–374. trajectories. Theoretical Population Mueller, L. D., Drapeau, M. D., Adams, C. S., Biology, 65, 353–359. Hammerle, C. W., Doyal, K. M., Jazayeri, A. J., Martin, G. M. (1997). The Werner mutation: Ly, T., Beguwala, S. A., Mamidi, A. R., & does it lead to a “public” or “private” Rose, M. R. (2003). Statistical tests of mechanism of aging? Molecular Medicine, demographic heterogeneity theories. 3, 356–358. Exp Gerontol, 38, 373–386. P088387-Ch08.qxd 10/31/05 11:27 AM Page 240

240 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

Mueller, L. D., & Rose, M. R. (1996). Proceedings of the Royal Society Series B Evolutionary theory predicts late-life (London), 271, 1225–1234. mortality plateaus. Proceedings of the Promislow, D. E. L., & Bronikowski, A. M. (in National Academy of Sciences of the USA, press). Evolutionary genetics of senescence. 93, 15249–15253. In C. W. Fox and J. B. Wolf (Eds.), Nikitin, A. G., & Shmookler Reis, R. J. (1997). Evolutionary genetics: concepts and case Role of transposable elements in age-related studies. Oxford: Oxford University Press. genomic instability. Genetics Research, 69, Promislow, D. E. L., & Tatar, M. (1998). 183–195. Mutation and senescence: where genetics Nuzhdin, S. V., Pasyukova, E. G., Dilda, C. L., and demography meet. Genetica, 102/103, Zeng, Z.-B., & Mackay, T. F. C. (1997). Sex- 299–314. specific quantitative trait loci affecting Promislow, D. E. L., Montgomerie, R. D., & longevity in Drosophila melanogaster. Martin, T. E. (1992). Mortality costs of Proceedings of the National Academy of sexual dimorphism in birds. Proceedings of Sciences of the USA, 94, 9734–9739. the Royal Society (London): Biological Parker, G. A., & Thompson, E. A. (1980). Sciences, 250, 143–150. Dung fly struggles—a test of the war of Promislow, D. E. L., Smith, E. A., & Pearse, L. attrition. Behavioral Ecology and (1998). Adult fitness consequences of sexual Sociobiology, 7, 37–44. selection in Drosophila melanogaster. Partridge, L., & Gems, D. (2002a). The Proceedings of the National Academy of evolution of longevity. Current Biology, 12, Sciences of the USA, 95, 10687–10692. R544–R546. Promislow, D. E. L., Tatar, M., Khazaeli, A., & Partridge, L., & Gems, D. (2002b). Curtsinger, J. W. (1996). Age-specific Mechanisms of ageing: public or private? patterns of genetic variance in Drosophila Nature Reviews Genetics, 3, 165–175. melanogaster. I. Mortality. Genetics, 143, Partridge, L., & Harvey, P. (1985). Costs of 839–848. reproduction. Nature, 316, 20. Radhika, M., Abdul Nazar, A. K., Pletcher, S. D., Macdonald, S. J., Marguerie, R., Munuswamy, N., & Nellaippan, K. (1998). Certa, U., Stearns, S. C., Goldstein, D. B., & Sex-linked differences in phenoloxidase in the Partridge, L. (2002). Genome-wide transcript fairy shrimp Streptocephalus dichotomous profiles in aging and calorically restricted Baird and their possible role (Crustacea: Drosophila melanogaster. Current Biology, Anostraca). Hydrobiologia, 377, 161–164. 12, 712–723. Rantala, M. J., Vainikka, A., & Kortet, R. Price, P., Witt, C., Allcock, R., Sayer, D., (2003). The role of juvenile hormone in Garlepp, M., Kok, C. C., French, M., immune function and pheromone Mallal, S., & Christiansen, F. (1999). The production trade-offs: a test of the genetic basis for the association of the 8.1 immunocompetence handicap principle. ancestral haplotype (A1, B8, DR3) with Proceedings of the Royal Society of London multiple immunopathological diseases. Series B Biological Sciences, 270, 2257–2261. Immunol Rev, 167, 257–274. Reznick, D. N. (1997). Life history evolution Promislow, D. E. L. (1991). Senescence in in guppies (Poecila reticulata): guppies as a natural populations of mammals: a model for studying the evolutionary comparative study. Evolution, 45, biology of aging. Experimental 1869–1887. Gerontology, 32, 245–258. Promislow, D. E. L. (1992). Costs of sexual Reznick, D. N., Bryant, M. J., Roff, D., selection in natural populations of Ghalambor, C. K., & Ghalambor, D. E. mammals. Proceedings of the Royal Society (2004). Effect of extrinsic mortality on (London): Biological Sciences, 247, 203–210. the evolution of senescence in guppies. Promislow, D. E. L. (2003). Mate choice, Nature, 431, 1095–1099. sexual conflict, and evolution of Rice, W. R. (1996). Sexually antagonistic male senescence. Behavioral Genetics, 33, adaptation triggered by experimental arrest 191–201. of female evolution. Nature, 381, 232–234. Promislow, D. E. L. (2004). Protein networks, Ricklefs, R. E. (1998). Evolutionary theories of pleiotropy and the evolution of senescence. aging: confirmation of a fundamental P088387-Ch08.qxd 10/31/05 11:27 AM Page 241

CHAPTER 8 / Evolutionary Biology of Aging: Future Directions 241

prediction, with implications for the pleiotropy and epistasis. Genetics, 153, genetic basis and evolution of life span. 1233–1243. American Naturalist, 152, 24–44. Siegal, M. L., & Bergman, A. (2002). Ricklefs, R. E. (2000). Intrinsic aging-related Waddington’s canalization revisited: mortality in birds. Journal of Avian developmental stability and evolution. Biology, 31, 103–111. Proceedings of the National Academy of Rocheleau, A. F., & Houle, G. (2001). Sciences of the USA, 99, 10528–10532. Different cost of reproduction for the males Snoke, M. S., & Promislow, D. E. (2003). and females of the rare dioecious shrub Quantitative genetic tests of recent Corema conradii (Empetraceae). American senescence theory: age-specific mortality Journal of Botany, 88, 659–666. and male fertility in Drosophila Rolff, J. (2001). Effects of age and gender on melanogaster. Heredity, 91, 546–556. immune function of dragonflies (Odonata, Sozou, P. D., & Kirkwood, T. B. (2001). Lestidae) from a wild population. Canadian A stochastic model of cell replicative Journal of Zoology, 12, 2176–2180. senescence based on telomere shortening, Rolff, J., & Siva-Jothy, M. T. (2002). oxidative stress, and somatic mutations in Copulation corrupts immunity: a nuclear and mitochondrial DNA. Journal of mechanism for a cost of mating in insects. Theoretical Biology, 213, 573–586. Proceedings of the National Academy of Spencer, C. C., Howell, C. E., Wright, A. R., & Sciences of the USA, 99, 9916–9918. Promislow, D. E. L. (2003). Testing an ‘aging Rolff, J., & Siva-Jothy, M. T. (2003). gene’ in long-lived Drosophila strains: Invertebrate ecological immunology. increased longevity depends on sex and Science, 301, 472–475. genetic background. Aging Cell, 2, 123–130. Rose, M. R. (1991). Evolutionary biology of Spyropoulos, B. (1988). Tay-Sachs carriers and aging. Oxford: Oxford University Press. tuberculosis resistance. Nature, 331, 666. Rose, M. R., & Charlesworth, B. (1980). A test Stutt, A. D., & Siva-Jothy, M. T. (2001). of evolutionary theories of senescence. Traumatic insemination and sexual conflict Nature, 287, 141–142. in the bed bug Cimex lectularius. Salter-Cid, L., Peterson, P. A., & Yang, Y. Proceedings of the National Academy of (2000). The major histocompatibility Sciences of the USA, 98, 5683–5687. complex-encoded HFE in iron homeostasis Svensson, E., & Sheldon, B. C. (1998). The and immune function. Immunology social context of life history evolution. Research, 22, 43–59. Oikos, 83, 466–477. Schmidt, P. S., Duvernell, D. D., & Eanes, W. F. Tatar, M., Bartke, A., & Antebi, A. (2003). The (2000). Adaptive evolution of a candidate endocrine regulation of aging by insulin-like gene for aging in Drosophila. Proceedings of signals. Science, 299, 1346–1351. the National Academy of Sciences of the Tatar, M., Gray, D. W., & Carey, J. R. (1997). USA, 97, 10861–10865. Altitudinal variation for senescence in Service, P. M. (2000). Heterogeneity in Melanoplus grasshoppers. Oecologia, 111, individual mortality risk and its importance 357–364. for evolutionary studies of senescence. Tatar, M., Kopelman, A., Epstein, D., Tu, M. P., American Naturalist, 156, 1–13. Yin, C. M., & Garofalo, R. S. (2001). Sgrò, C. M., & Partridge, L. (1999). A delayed A mutant Drosophila insulin receptor wave of death from reproduction in homolog that extends life-span and impairs Drosophila. Science, 286, 2521–2524. neuroendocrine function. Science, 292, Sgrò, C. M., & Partridge, L. (2000). 107–110. Evolutionary responses of the life history of Tatar, M., Promislow, D. E. L., Khazaeli, A., & wild-caught Drosophila melanogaster to Curtsinger, J. W. (1996). Age-specific two standard methods of laboratory culture. patterns of genetic variance in Drosophila American Naturalist, 156, 341–353. melanogaster. II. Fecundity and its genetic Shook, D. R., & Johnson, T. E. (1999). correlation with age-specific mortality. Quantitative trait loci affecting survival Genetics, 143, 849–858. and fertility-related traits in C. elegans Tissenbaum, H. A., & Guarente, L. (2001). show genotype-environment interactions, Increased dosage of a sir-2 gene extends P088387-Ch08.qxd 10/31/05 11:27 AM Page 242

242 D. E. L. Promislow, K. M. Fedorka, and J. M. S Burger

lifespan in C. elegans. Nature, 410, regulation of heart function in aging fruit 227–230. flies. Nature Genetics, 36, 1275–1281. Torchin, M. E., Lafferty, K. D., Dobson, A. P., Wigglesworth, V. B. (1965). Juvenile hormone. McKenzie, V. J., & Kuris, A. M. (2003). Nature, 208, 522. Introduced species and their missing Wilkinson, G. S., & South, J. M. (2002). Life parasites. Nature, 421, 628–630. history, ecology and longevity in bats. Torroba, M., & Zapata, A. G. (2003). Aging of Aging Cell, 1, 124–131. the vertebrate immune system. Microscopy Williams, G. C. (1957). Pleiotropy, natural Research and Technique, 62, 477–481. selection, and the evolution of senescence. van Swinderen, B., & Greenspan, R. J. (2005). Evolution, 11, 398–411. Flexibility in a gene network affecting a Williams, G. C. (1999). The Tithonus error in simple behavior in Drosophila melanogaster. modern gerontology. Quarterly Review of Genetics, 169, 2151–2163. Biology, 74, 405–415. Vaupel, J. W., Baudisch, A., Dolling, M., Williams, P. D., & Day, T. (2001). Interactions Roach, D. A., & Gampe, J. (2004). The case between sources of mortality and the for negative senescence. Theoretical evolution of parasite virulence. Proceedings Population Biology, 65, 339–351. of the Royal Society London Series B Vaupel, J. W., Carey, J. R., Christensen, K., Biological Sciences, 268, 2331–2337. Johnson, T. E., Yashin, A. I., Holm, N. V., Williams, P. D., & Day, T. (2003). Iachine, I. A., Kannisto, V., Khazaeli, A. A., Antagonistic pleiotropy, mortality Liedo, P., Longo, V. D., Zeng, Y., source interactions, and the evolutionary Manton, K. G., & Curtsinger, J. W. (1998). theory of senescence. Evolution, 57, Biodemographic trajectories of longevity. 1478–1488. Science, 280, 855–860. Wolf, J. B., Brodie, E. D. I., & Wade, M. J. Wagner, A. (2000). Robustness against (2000). Epistasis and the evolutionary mutations in genetic networks of yeast. process. Oxford: Oxford University Press. Nature Genetics, 24, 355–361. Wolfe, L. M. (2002). Why alien invaders Waheed, A., Parkkila, S., Zhou, X. Y., succeed: support for the escape-from-enemy Tomatsu, S., Tsuchihashi, Z., Feder, J. N., hypothesis. American Naturalist, 160, Schatzman, R. C., Britton, R. S., 705–711. Bacon, B. R., & Sly, W. S. (1997). Hereditary Woodruff, R. C., & Nikitin, A. G. (1995). P hemochromatosis: effects of C282Y and DNA element movement in somatic cells H63D mutations on association with reduces lifespan in Drosophila beta2-microglobulin, intracellular melanogaster: evidence in support of the processing, and cell surface expression somatic mutation theory of aging. of the HFE protein in COS-7 cells. Mutation Research, 338, 35–42. Proceedings of the National Academy of Wright, S. (1932). The roles of mutation, Sciences of the USA, 94, 12384–12389. inbreeding, crossbreeding and selection in Walford, R. L. (1969). The immunological evolution. Proceedings of the Sixth theory of aging. Copenhagen: Munksgaard. International Congress of Genetics, 1, Walker, D. W., McColl, G., Jenkins, N. L., 356–366. Harris, J., & Lithgow, G. J. (2000). Evolution Zera, A. J., & Harshman, L. G. (2001). The of lifespan in C. elegans. Nature, 405, physiology of life history trade-offs in 296–297. animals. Annual Review of Ecology and Weismann, A. (1891). Essay on heredity and Systematics, 32, 95–126. kindred biological problems (2nd ed., Vol. 1.). Zuk, M., & Stoehr, A. M. (2002). Immune Oxford: Clarendon Press. defense and host life history. American Weksler, M. E., & Schwab, R. (1992). The Naturalist, 160, S9–S22. immunogenetics of immune senescence. Zuk, M., Simmons, L. W., Rotenberry, J. T., Experimental and Clinical & Stoehr, A. M. (2004). Sex differences Immunogenetics, 9, 182–187. in immunity in two species of field Wessells, R. J., Fitzgerald, E., Cypser, J. R., crickets. Canadian Journal of Zoology, Tatar, M., & Bodmer, R. (2004). Insulin 82, 627–634. P088387-Ch09.qxd 10/31/05 11:29 AM Page 243

Chapter 9

Senescence in Wild Populations of Mammals and Birds

Anja K. Brunet-Rossinni and Steven N. Austad

I. Introduction It has long been conjectured, particularly in the biomedical community, that Senescence or, synonymously in this animals in nature do not experience chapter, aging can be defined as the pro- senescence (Comfort, 1979, Hayflick, gressive deterioration in physiological 2000; Medawar, 1952). Among ecologists, function that accompanies increasing such a general assumption has never been adult age. Like the proverbial elephant advanced; however, some animal groups, described by blind men, it has many particularly birds and fishes, have been visible forms, depending on your identified as likely to die at a constant perspective. For instance, to a geriatri- rate in nature, irrespective of age (Deevey, cian, senescence is evident as an 1947). Even this more restricted state- age-related increase in frailty or vulner- ment has met with subsequent skepti- ability and a mounting incidence and cism from field biologists. Botkin and severity of degenerative diseases. To a Miller (1974) pointed out that if annual demographer, it is most easily measured mortality rate were indeed independent as an age-related increase in the proba- of age, then one or more of the royal alba- bility of death, and to an evolutionary trosses seen by Captain Cook during his biologist, senescence might describe the first visit to New Zealand in 1769 should progressive decline in age-specific still be alive today! As remarkable as Darwinian fitness components. These avian longevity might be, that is a hard are all valid embodiments of senes- tale to swallow. The Guinness Book of cence; however, some will be more eas- Records lists the oldest royal albatross ily observed and measured in natural identified to date as 53 years old. populations and less easily confounded The notion that animals do not senesce by other phenomena than others. Also, in nature arises from an implicit belief various signs of senescence may appear that life in the wild is so nasty and at different ages. brutish, so beset by predation, pestilence,

243 P088387-Ch09.qxd 10/31/05 11:29 AM Page 244

244 A. K. Brunet-Rossinni and S. N. Austad

foul weather, famine, and drought that may say more about the length, design, animals must inevitably die before signs of and intensity of the study than the proba- senescence appear. According to this intu- bility that the population under study did, ition, only under the protected conditions or did not, exhibit signs of senescence. of captivity (or civilization) does aging become manifest. No doubt such a scenario describes life for some species. For II. Evidence of Senescence in Wild instance, multiple field studies of house Populations mice (Mus musculus) indicate that median A. Demographic Senescence longevity in nature is only about 3 to 4 months, with 90 percent of deaths occur- The majority of articles assessing senes- ring by about 6 months of age (Phelan & cence in natural populations have focused Austad, 1989)—ages at which senescence on detecting an age-related increase in is indeed not readily observable even in mortality rate (actuarial or survival senes- the laboratory. This is in contrast to the 2- cence) and/or decrease in reproductive rate to 3-year mean longevity under protected (reproductive senescence). Together we conditions (Turturro et al. 1999). call these measures of demographic senes- On the other hand, the general claim cence. Increasing age-specific mortality in of negligible senescence in the wild, adulthood is widely used as the gold stan- despite its persistence in the biogeron- dard of senescence, both in natural and in tological literature, is quite clearly laboratory populations. This choice stems incorrect, as we will show. We are not from the assumption that age-specific suggesting that animals in nature reach death rate is a good indicator of intrinsic the advanced state of decrepitude found physiological hardiness. As hardiness in the city or the laboratory, but only declines with age, the probability of death that an observable age-related decline in should rise. Such a rise has often been function is not rare. found in natural populations. For instance, In this article, we review evidence for Promislow’s (1991) analysis of 56 pub- senescence in birds and mammals in the lished life tables from mammal field stud- wild. We have limited our survey to birds ies found statistically significant evidence and mammals not because they are for survival senescence in 44 percent (26) uniquely senescence-prone, but to keep of the studies and a nonsignificant trend in the size of our review manageable and the same direction in a further 34 percent also because of the availability of numer- (20) of the studies. Thus, there was at least ous long-term field studies in which some reason to suspect the existence of individuals have been monitored for a senescence in nearly four of five published number of years, sometimes throughout mammal field studies. Later, Gaillard and life, for these two animal groups. Such colleagues (1994) published a slightly more studies are not required to detect senes- conservative reanalysis of the same data cence, but they are the most likely to do but still found statistically compelling evi- so. Such studies also avoid many of the dence for senescence in 42 percent (25 of pitfalls of less intensive studies. We dis- 59) of data sets. cuss the nature of these pitfalls later on. Sometimes, as in the previously It is important to note that detecting mentioned Royal Albatross, survival senescence is the primary goal of few if senescence can be inferred by the lack any field studies. Data relevant to senes- of survivors above a certain age. For cence are usually adventitiously acquired instance, Ian Nisbet has been studying while investigating other issues. Therefore, the common tern, a small seabird, at an the failure to detect senescence in a study island off the Massachusetts coast for P088387-Ch09.qxd 10/31/05 11:29 AM Page 245

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 245

more than 30 years and has individually interpreting the failure to find increas- marked thousands of birds. The annual ing mortality as indicating an absence observed adult survival rate is 0.9. If sur- of senescence. First, for any given cohort, vival rate were independent of age, then nonseasonal, stochastic environmental about 5 percent of adults (assuming adult- factors may override or mask even the hood begins at age 2) should survive to 30 very real effects of age on physical state. years of age and 1 percent should survive As one example, a classic study of house 45 years. However, direct observation sparrows found that a New England win- finds that the 5 percent survival age is 18 ter storm preferentially killed birds that years and the longest-lived bird seen so were particularly large or small relative to far is only 26 years old (Nisbet, 2001). the mean size in the population (Bumpus, Measurement of survival senescence 1898). Body size is not related to age in has some practical difficulties when birds, therefore the storm likely killed applied to wild populations, however. birds with little if any relation to their age. One such complication is seasonality. However the mortality from that single For instance, a number of small mam- event might have overwhelmed any mals develop in the spring, mature in the underlying age-related pattern. Stochastic summer and autumn, and tend to die in climatic events can also depress food the winter. Among species such as these, availability, which is likely to lead to more increasing age-related mortality rate risk-taking by foragers and thus a gener- could indicate senescence but could just ally higher mortality rate. Environmental as easily indicate increasing environmen- events of this type can confound detection tal harshness from summer to autumn to of survival senescence or its absence. winter. That is, even if no physiological Significantly, dead animals are seldom deterioration has occurred in the study recovered in field studies. Death is typi- populations, morality rate could still cally assumed when marked animals dis- increase with age. Consequently, increas- appear from a study area, but they could ing mortality rate with age does not by also have emigrated. As a result, mortality itself unequivocally demonstrate senes- and emigration can easily be conflated. In cence in the wild. addition, if individual identification is by Another problem is environmental tags or bands, these markers can fall off. If change with unknown effects on demog- so, a death will be mistakenly recorded. raphy. For instance, the common tern Third, even if death can be unambiguously colony studied by Nisbet and colleagues detected, mortality rates can easily be grew from about 500 breeders in 1970 to affected by behavior as well as by intrinsic about 4,500 breeders in 1992. An early deterioration. For instance, animals engag- report from this colony suggested that ing in high-risk behaviors will obviously several markers of reproductive function die at higher rates than their more cau- (clutch size, egg volume) decreased with tious peers. If animals in certain age age (Nisbet et al., 1984). However a much classes are particularly prone to risky later report found no evidence of repro- behavior, underlying physical senes- ductive senescence (Nisbet et al., 2002). cence patterns again may be masked Whether this difference was due to the (see Figure 9.1). Are there particular ages increased population density, some other at which animals in nature are prone to environmental factor, or simply more and engage in high-risk behaviors? Indeed, as better information is not clear. in humans, adolescent and newly matur- In addition to complications involving animals frequently take exceptional ing seasonality or environmental change, risks. Typically, birds and mammals dis- several other factors warrant caution in perse from their birth site around the time P088387-Ch09.qxd 10/31/05 11:29 AM Page 246

246 A. K. Brunet-Rossinni and S. N. Austad

of physical senescence. For instance, A among human females, fertility declines are detectable by about age 30, an age of minimal senescence by other measures (Dunson et al., 2004; Shock, 1983). One main difficulty with using reproductive function as a measure of senescence is that, like age-specific mortality, environmental events can mask underlying patterns. Hard times—food shortages, temperature extremes, exceptional preda- tor abundance—rather than aging may B Mortality Rate depress reproduction in a study population. Reproductive performance also has the disadvantage that it is generally much easier to monitor in one sex (females) relative to the other. So, too, there are sub- tleties of reproductive senescence that will be difficult to detect in either field or laboratory. For instance, reproductive senescence may manifest itself as a Figure 9.1 Misleading inference of no senescence decline in the phenotypic quality of off- caused by elevated risk-taking during dispersal or spring from older females rather than a territory acquisition. A. Regression from points straightforward reduction in number of sampled. B. Underlying mortality pattern. •Ages sampled. Dashed line = actual adult mortality tra- eggs or newborns (Saino et al., 2002). jectory, solid line = inferred mortality trajectory As another example, slow growth of from 5 sampled ages. unweaned pups rather than reduced litter size marked reproductive senescence in Virginia opossums (Austad, 1993). of sexual maturity to seek and compete for Both reproductive and survival senes- new territories and/or mates. Venturing cence can also be difficult to detect in through unknown areas and competing for the presence of infectious disease. The territories or mates can be very dangerous. relationship between senescence and dis- In one of the few studies in which this ease is complex (see Masoro, Chapter 2). “dispersal risk” could be quantified, Decreased reproductive performance near female water-voles (Arvicola terrestris) the very end of life might as easily reflect died at at least an 86-fold higher rate dur- infectious status as senescence per se. ing dispersal than when remaining in their For instance, in a long-term study of an natal home range (Leuze, 1980). Any such oceanic bird species, the black-legged kitti- behavior-mediated elevation of mortality wake, Coulson and Fairweather (2001) rate in early life will require that statisti- observed depressed reproductive perform- cal analyses be sensitive to the possibility ance in the final breeding period prior to that temporary, early life mortality spikes death in birds of all ages. Because this was might make a simple Gompertzian analy- observed in young as well as old birds, the sis of the entire adult life somewhat mis- authors interpreted this pattern as a sign of leading (see Figure 9.1). terminal illness, not senescence, although Survival is only part of demography. An these are clearly not mutually exclusive. age-related decline in reproductive per- Such difficulties aside, an extensive but formance can also be a sensitive indicator not exhaustive survey of the literature P088387-Ch09.qxd 10/31/05 11:29 AM Page 247

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 247

turned up evidence for demographic sources. However, we included any senescence (either survival or reproduc- article in which the authors claimed evi- tive) in natural populations of 42 species dence of senescence in natural popula- of mammals and 35 species of birds (sum- tions or in which we could easily detect marized in Tables 9.1 and 9.2). Included such evidence from the published data in this list are the 26 species for which even if the author(s) failed to note it. Promislow (1991) found statistically sig- In addition to the studies we cite, of nificant (P 0.1) evidence of senescence. course, a number of studies have failed to We excluded from this list data on cap- find demographic senescence. However, tive or semi-wild populations and claims because these studies varied in their of senescence based only on secondary intensity and evidentiary methodology, it

Table 9.1 Literature presenting evidence of senescence in wild populations of mammals. Survival senescence refers to an increasing age-specific mortality rate. Measures used as evidence of reproductive senescence are listed for each species and symbol in parenthesis indicates which sex was studied. Arrows indicate direction of change in variable with increasing age. Column labeled as “other” contains information regarding changes in behavior or morphology associated with increasing age.

Species Survival Reproductive Other senescence senescence

Didelphis virginiana Austad, 1993 qinfertility,ppouch young growth rate () Virginia opossum Austad, 1993 Pipistrellus pipistrellus Promislow 1991 Pipistrelle bat Macaca mulatta Promislow 1991 Rhesus macaque Macaca fuscata pbirth rate () Japanese macaque Wolfe & Noyes 1981 Papio hamadryas Bronikowski pmaternity rate, et al. 2002 pfertility () Baboon Packer et al. 1998 Packer et al. 1998 Lutra canadensis Promislow 1991 River otter Martes zibellina Promislow 1991 Sable marten Panthera leo Packer et al. 1998 pmaternity rate () African Lion Promislow 1991 Packer et al. 1998 psurviving cubs/male () Packer et al. 1988 Ursus arctos Promislow 1991 Brown bear Ursus maritimus plitter size and mass () pbody mass () Polar bear Derocher & Stirling 1994 Derocher & Stirling 1994 Alces alces Ericsson plitter size,qoffspring et al. 2001 mortality () Moose Ericsson et al. 2001 pfertility Heard et al. 1997

(continues) P088387-Ch09.qxd 10/31/05 11:29 AM Page 248

248 A. K. Brunet-Rossinni and S. N. Austad

Table 9.1 (Cont’d)

Species Survival Reproductive Other senescence senescence

Dama dama McElligott preproduction et al. 2002 probability () Fallow deer McElligott et al. 2002 Aepyceros melampus Promislow 1991 Impala Hemitragus jemlahicus Promislow 1991 Himalayan tahr Ovis dalli Promislow 1991 Dall’s sheep Kobus kob Promislow 1991 Kob Rupicapra rupicapra Promislow 1991 Chamois Cervus elaphus Clutton-Brock pfecundity, pbody mass et al. 1988 qcalf mortality () Mysterud Red deer Clutton-Brock 1984 et al. 2001 preproduction pcontrol of probability () harems () Clutton-Brock Clutton-Brock et al. 1979 et al. 1979 Tragelaphus Owen-Smith 1993 strepsicerus Greater kudu Syncerus caffer Sinclair, 1977 ppregnancy rate () African buffalo Promislow 1991 Sinclair, 1977, Grimsdell, 1969 Ovis musimon pparental care () Mouflon Réale & Boussès 1995 Hippopotamus Promislow 1991 amphibious Hippopotamus Phacochoerus aethiopicus Promislow 1991 Desert warthog Lutra canadensis Promislow 1991 River otter Martes zibellina Promislow 1991 Sable marten Panthera leo Packer et al. 1998 pmaternity rate () African Lion Promislow 1991 Packer et al. 1998 psurviving cubs/male () Packer et al. 1988 Ursus arctos Promislow 1991 Brown bear Ursus maritimus plitter size and mass () pbody mass () Polar bear Derocher & Stirling 1994 Derocher & Stirling 1994 Lepus europaeus Promislow 1991 European hare

(continues) P088387-Ch09.qxd 10/31/05 11:29 AM Page 249

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 249

Table 9.1 (Cont’d)

Species Survival Reproductive Other senescence senescence

Oryctolagus cuniculus Promislow 1991 European rabbit Sylvilagus floridanus Promislow 1991 Eastern cottontail Equus burchelli Promislow 1991 Burchell’s zebra Callorhinus ursinus Promislow 1991 Northern fur seal Arctocephalus gazelle preproductive rates () Antarctic fur seal Lunn et al. 1994 Phoca hispida Promislow 1991 Ringed seal Pan troglodytes Promislow 1991 pbirth rate () Chimpanzee Sugiyama 1994 Loxodonta Africana Promislow 1991 African elephant Microtus agrestis Promislow 1991 Field vole Apodemus flavicollis Promislow 1991 Yellow-necked mouse Peromyscus maniculatus Millar 1994 Deer mouse Promislow 1991 Peromyscus leucopus plitter size of old, large females White-footed mouse Morris 1996 Tamiasciurus hudsonicus Promislow 1991 Red squirrel Spermophilus colombianus qunsuccessful litters () Colombian ground squirrel Broussard et al. 2003

Table 9.2 Literature presenting evidence of senescence in wild populations of birds. Survival senescence refers to an increasing age-specific mortality rate. Measures used as evidence of reproductive senescence are listed for each species and arrows indicate direction of change in variable with increasing age. Column labeled as “other” contains information regarding changes in behavior or morphology associated with increasing age.

Species Survival Reproductive Other senescence senescence

Aphelocoma coerulescens McDonald et al. 1996 pnumber of fledglings, poffspring

Florida scrub jay survival Fitzpatrick & Woolfenden 1988 Pica pica pclutch size Black-billed Magpie Birkhead & Goodburn 1989

(continues) P088387-Ch09.qxd 10/31/05 11:29 AM Page 250

250 A. K. Brunet-Rossinni and S. N. Austad

Table 9.2 (Cont’d)

Species Survival Reproductive Other senescence senescence

Bucephala clangula pbrood survival Common goldeneye Milonoff et al. 2002 pclutch and brood size Dow & Fredga 1984 Larus glaucescens pegg volume,phatching success Glaucous-winged gull Reid 1988 Sula granti pfledging success Nazca booby Anderson & Apanius, 2003 Larus delawarensis phatching success Ring-billed gull Haymes & Blokpoel 1980 Larus californicus Pugesek 1987 California gull Pugesek et al. 1995 Larus canus Rattiste & Lilleleht 1987 Common gull Fulmarus glacials Dunnet & pbreeding success Ollason 1978 Northern Fulmar Ollason & Dunnet 1978 pfecundity,pbreeding success Ollason & Dunnet 1988 Rissa tridactyla Aebischer & pfledgling production Coulson 1990 Black-legged Coulson & Thomas 1983 Kittiwake Wooller 1976 Somateria Coulson 1984 preproductive output mollissima Common Eider Bailey & Milne 1982 Puffinus tenuirostris Bradley et al. 1989 preproductive performance Short-tailed Wooller et al. 1990 shearwater Phalacrocorax Aebischer 1986 aristotelis Shag Harris et al. 1994 Catharacta skua pclutch volume Great skua Hamer & Furness 1991 pclutch size Ratcliffe et al. 1998 Diomedea exulans Weimerskirch 1992 pegg size,pbreeding success and frequency Wandering albatross Weimerskirch 1992 Sterna hirundo pclutch size,pegg volume Common tern Nisbet et al. 1984 Sterna paradisaea pclutch size and volume Arctic tern Coulson & Horobin 1976 Chen caerulescens phatchability, qbrood loss Snow goose Rockwell et al. 1993

(continues) P088387-Ch09.qxd 10/31/05 11:29 AM Page 251

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 251

Table 9.2 (Cont’d)

Species Survival Reproductive Other senescence senescence

Hirundo rustica Møller & poffspring quality qectoparasite de Lope 1999 load, Barn swallow Saino et al. 2002 psecondary sexual characters, pnumber of fledglings, qfluctuating preproductive value asymmetry, Møller & de Lope 1999 pbody mass, delayed arrival from migration Møller and de Lope 1999; phumoral immunity Saino et al. 2003 Tachycineta bicolor pbreeding performance index Tree swallow Robertson & Rendell 2001 Calidris temminckii Hilden 1978 Temminck’s stint Parus major Dhondt 1989 pnesting success, pbrood size, pjuvenile survival Great tit Dhondt 1989 phatching rate,pfledgling survival Perrins & Moss 1974 Parus caeruleus pnesting success, pclutch and brood size, pjuvenile survival Blue tit Dhondt 1989 Parus montanus Orell & Belda 2002 Willow tit Pyrrhocorax pclutch size,pfledging pyrrhocorax success Red-billed chough Reid et al. 2003b Ficedula albicollis psuccessful broods, phumoral immune pnumber of fledglings, response pclutch size Collared flycatcher Gustafsson & Pärt 1990 Cichon et al. 2003 Ficedula hypoleuca Sternberg 1989 Pied flycatcher Parus atricapillus Loery et al. 1987 Black-capped chickadee Acrocephalus phatching success,pnumber sechellensis of fledglings Seycelles warbler Komdeur 1996 Melospiza melodia pnumber of fledglings pCell-mediated immune response Song sparrow Nol & Smith 1987 Reid et al. 2003a

(continues) P088387-Ch09.qxd 10/31/05 11:29 AM Page 252

252 A. K. Brunet-Rossinni and S. N. Austad

Table 9.2 (Cont’d)

Species Survival Reproductive Other senescence senescence

Anthus spinoletta psecond clutch size Rock pipit Askenmo & Unger 1986 Geospiza conirostris Grant & Grant 1989 Large cactus finch Tetrao tetrix pcopulation success () Black grouse Kruijt & de Vos 1988 Accipiter nisus Newton & preproductive output Rothery 1997 European Sparrowhawk Newton et al. 1981 pclutch size,pnumber of young/nest,pegg size Newton 1988 pannual production of young, preproductive value Newton and Rothery 1997 Aegolius funereus pclutch size () Tengmalm’s owl Laaksonen et al. 2002

is not possible to identify general pat- with age, and there is a tradeoff between terns concerning species that exhibit reproduction and survival, as predicted by senescence in nature relative to those Williams (1957) and empirically verified by that do not. For one thing, failure to many others (e.g., Koivula et al, 2003; observe a phenomenon of interest likely Orell & Belda, 2002), then declining sur- leads to reporting bias. That is, negative vival with age might reveal nothing more results often go unpublished. than increasing reproduction rather than This is not to say that some very thor- an independently deteriorating internal ough studies have failed to find demo- state. Although evolutionary theories of graphic senescence. For instance, a senescence, such as antagonistic pleiotropy 17-year study of Southern elephant seals and mutation accumulation (Medawar, tracked 1,650 individually marked pups 1952; Williams, 1957), do not specifically throughout their lives. Although only predict a decline in fertility or increase in 5 percent of the marked animals survived death risk alone with age, they do predict even to age 10, no statistically significant declining reproductive expectations with increase in age-specific mortality rate age. Reproductive expectations incorporate could be detected even as late as between both reproduction and survival. This is ages 10 and 17 (Pistorius & Bester, 2002). why Partridge and Barton (1996) suggest A 30-year study of the common tern Fisher’s “reproductive value,” which uti- (Sterna hirundo) in which about 60,000 lizes both reproduction and survival to chicks have been banded has not found quantify expected current and future repro- evidence of reproductive senescence even duction as the best metric for assessing among the oldest 5 percent of birds in senescence. An interesting case illustrating the population (Nisbet et al., 2002). the problem and how such an analytical Besides practical difficulties in measur- approach may be useful is Pugesek’s long- ing demographic senescence in nature, term study of California gulls (Pugesek, there are also theoretical difficulties. For 1987; Pugesek & Diem, 1990; Pugesek instance, if animals increase reproduction et al., 1995). Older birds in this study P088387-Ch09.qxd 10/31/05 11:29 AM Page 253

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 253

died at higher rates than younger birds, behavioral factors may overwhelm subtle hence by the standard criterion of survival demographic indicators; therefore, physio- senescence, they aged. However, older logical markers of functional decline birds also raised more chicks to fledging might serve as alternative or complemen- than did younger birds. Feeding and pro- tary sources of information. Note that the tecting young in the nest is energetically indicators used here only need document taxing and physically risky. That is, that later life decline in function has there is a cost to reproduction. Controll- occurred. They do not need to be useful as ing for reproduction by comparing young markers of the rate of functional loss. For (3- to 10-year-old) birds with older (11- to instance, studies of three bird species 17-year-old) ones that fledged the same have presented evidence of a decline in number of young, no difference in yearly immune function with advancing age survival could be detected (Pugesek, 1987). (Cichón et al., 2003; Reid et al., 2003a; A potentially useful way of envisioning Saino et al., 2003). Although the long-held demographic senescence is to assess the view that aging is accompanied by a maintenance of adaptive tradeoffs between monolithic decline in immune function is reproduction and somatic survival as they gradually being supplanted by a more occur in young adults. For instance, nuanced view that aging alters immune Broussard and colleagues (2003) found that response in complex, potentially adaptive the oldest Colombian ground squirrels ways (Effros, 2001), it is probably still a (Spermophilus colombianus) in their safe generalization that deteriorating study population were more likely than resistance to infectious agents is a hall- younger age classes to experience repro- mark of physiological senescence. Indeed, ductive failure. Moreover, young females autopsy data on humans implicated infec- failing to reproduce regained healthy tions in the deaths of a majority of people body mass lost during their reproductive older than 80 years in a Japanese popula- attempt, whereas old females did not. tion (Horiuchi & Wilmoth, 1997). Thus, the adaptive tradeoff between repro- Declining immunity with age seems ductive and somatic investment appears widespread among species. It is well docu- to deteriorate in old age in the Colombian mented in humans and laboratory rodents ground squirrel. It is not unexpected that of course, but it has also been reported tradeoffs like other adaptive traits deterio- in song sparrows (Reid et al., 2003a) rate with age and decaying physiology. and rhesus macaques (Coe & Ershler, This type of deterioration may provide 2001). Particularly interesting field stud- a better indication of the effects of phy- ies combine assessment of immune com- siological degeneration than purely petence with demographic parameters. demographic parameters. Moreover, such Intriguingly, in barn swallows, humoral deteriorating tradeoffs could potentially immunity as measured by antibody produce physiological/demographic pat- response to injection of Newcastle terns that we do not normally recognize as Disease Virus declined in females 3 years senescence (Blarer et al., 1995). old or older relative to 1-and 2-year olds (Saino et al, 2003), but no such decline was observed in males. Survival senes- B. Nondemographic Measures cence could only be detected in birds of Senescence age 5 and older, whereas reproductive The failure to find demographic evidence senescence as measured by number of of senescence does not necessarily mean fledglings produced began to decline by that animals are not experiencing it. As age 4. Thus, the decline in female previously mentioned, environmental or immune function preceded indicators of P088387-Ch09.qxd 10/31/05 11:29 AM Page 254

254 A. K. Brunet-Rossinni and S. N. Austad

demographic senescence. It might there- ter size and body mass increase for fore be assumed that immune function female polar bears (Ursus maritimus) decline is a more sensitive indicator of until ages 14 to 16, and thereafter both senescence than demographic parameters. decline (Derocher & Stirling, 1994). However, the situation is not so sim- However, this is not always the case, as ple. In the collared flycatchers, a much shown by bighorn ewe body mass, which different pattern is seen. Humoral immu- began declining at about age 11 in one nity (response to injected sheep red blood study, about three years before a reduc- cells) did not decline significantly until 5 tion in fertility was observed but four to 6 years of age, whereas reproductive years after survival senescence became and survival performance declined after evident (Jorgenson et al., 1997). In sum, age 3 (Cichón et al., 2003; Møller & the meaning and reliability of age-related De Lope, 1999). In this case, demographic body mass decline as a marker of physi- senescence preceded immunosenescence. cal senescence is far from clear. Thus, one can’t generalize about the sen- Senescence could potentially be seen as sitivity of immune markers of physiolog- well in a reduced ability to provide nour- ical senescence in wild animals. It is ishment to offspring. Mammary glands most useful to have both physiological themselves may senesce due to a decreas- and demographic measures of senescence ing replicative capacity of mammary whenever possible. epithelial cells with age (Daniel, 1977). In A few studies have adduced age-related support of this phenomenon, the lambs of declines in body mass as evidence of old mouflon (Ovis musimon) ewes suckle senescence (Berubé et al., 1999; Derocher less frequently, decrease total suckling & Stirling, 1994; Mysterud et al., 2001). time, and spend more time grazing rela- The general sensitivity of body mass tive to lambs of younger ewes (Réale & decline as an indicator of aging is not Boussès, 1995). clear. There is substantial variability Other nondemographic indicators of among laboratory mouse sexes and senescence exist in specific instances. genotypes in the timing of body mass For instance, the ability to defend a terri- decrease relative to patterns of demo- tory or other more mobile resource such graphic senescence (Turturro et al., as a harem may decline with age. On the 1999). However, in the laboratory, with Scottish island of Rhum, the reproduc- ad lib feeding and minimal energetic tive success of red deer stags declines demands, body mass may decline at a after age 11 due to a decreased ability to more advanced age compared to the wild. fight and control harems (Clutton-Brock Senescence in nature can affect body et al., 1979). However, this is not always mass by decreasing the ability to obtain, true even in harem-defending species, process, and store food (Ericsson et al., as shown by the absence of a similar 2001). For instance, tooth wear may be a phenomenon in the American prong- significant contributor to age-related horm (Byers, 1998). From the female mortality in some mammals (Ericsson side, calf mortality increases for red deer et al. 2001; Skogland, 1988). Tooth wear hinds over 12 years of age. Survival of increases markedly in roe deer after the calves depends partially on its mother’s age of 7 and coincides with a decrease dominance rank and access to a good- in survival rate (Gaillard et al., 1993). quality home range, both of which Indeed, the strongest evidence for changes decline as females age (Clutton-Brock, in body mass relating to senescence are 1984). Therefore, change in dominance when simultaneous changes occur in rank may serve as a marker of senes- demographic parameters. Thus, both lit- cence in some species. P088387-Ch09.qxd 10/31/05 11:29 AM Page 255

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 255

The ability to defend a territory decrease with age. Indeed, Møller and declines with age in a number of species. De Lope (1999) found that both tail and In the small passerine bird, the great wing length decreased in barn swallows tit (Parus major), male territory size 5 years old or older. In addition, these increases until 4 years of age and then aged birds had less perfect symmetry decreases from 5 years throughout the between left and right wing and tail rest of life. Some old great tits forego feathers compared with younger birds. defending a territory altogether, although Finally, as biochemical markers of aging they still manage some breeding by mat- become available, studies that involve ing with females late in the year (Dhondt, recapturing individuals can focus on 1971). Male black grouse (Tetrao tetrix) directly measuring deteriorating physio- defend territories that females attend logy from blood or urine samples, though only to breed. Those 6 years and older assays will need to be validated for the par- seldom occupy the highest-quality terri- ticular study organism. Regardless of the tories on display grounds, likely because measures selected to test for senescence in they cannot defend them (Kruijt & natural populations, it is critical to have de Vos, 1988). By contrast, young male detailed understanding of the dynamics of greater white-lined bats (Saccopteryx the study population, the physiology of the bilineata) lurk on the periphery of older study organism, and the life history trade- males’ breeding territories, copulat- offs individuals face over their life span. ing with females opportunistically. No reports exist of males so old that they can no longer defend their territory. This may III. Patterns of Senescence be a case where significant senescence is not seen in the wild (Heckel & Von Can we glean anything about general pat- Helversen, 2002). Hormones may serve as terns of senescence in wild populations markers of senescence as well. Plasma from the information currently avail- testosterone level in Misaki feral horses able? The first and most important point correlates with age and harem size (Khalil is that one should exercise caution when et al., 1998). If senescent males no longer accepting and comparing published data can mount and maintain appropriate on senescence in the wild (Nisbet, 2001). testosterone levels during the breeding Apparent senescence can arise from season, their ability to recruit a harem short-term climatic events such as may become compromised. Thus, hor- El Niño or increasing population density monal and reproductive aging decline over the course of a study. The literature synchronously. must be interpreted with sensitivity to Sometimes demographic senescence variation among studies in the methods can be observed even when other well- used to collect and analyze demographic characterized indicators of senescence data. Second, senescence, measured as cannot be detected. For instance, com- changes in life-history traits and physical mon terns exhibit survival senescence, function associated with old age, is a yet no decline can be detected over their complex phenomenon. It can defy our lifetime in immunological, endocrinolog- expectations about which species should ical, or reproductive aging (Apanius & and should not show signs of senescence Nisbet, 2003). and at what point during the life course In birds, where renewal of feathers indicators of senescence might become must occur annually after molting, apparent. It is also critical to remember one might expect that the ability to pro- that patterns of senescence are not nec- duce long, healthy feathers might essarily species-specific but may vary P088387-Ch09.qxd 10/31/05 11:29 AM Page 256

256 A. K. Brunet-Rossinni and S. N. Austad

among populations within a species opportunities, often somewhat later than (Austad, 1993). the time of sexual maturation. McElligott Generally, senescence is expected to be and colleagues (2002) evaluated such a more apparent among species of low mor- situation in male fallow deer (Dama tality rates and long life span because dama). They found that social matura- senescence reduces average life span tion began at 4 to 5 years, yet annual more in such species (Ricklefs, 1998) and survival rate did not begin to decrease because high mortality rates significantly until age 9. An even more complex reduce the chances of any individual situation is found in bighorn sheep reaching a senescent age or remaining (Ovis canadensis), in which the onset of alive in a senescent state for long enough survival senescence did not coincide to be detected. However, some short- with age of first reproduction in females, lived animals with high mortality rates but did in males (i.e., rutting decreased have been demonstrated to experience male survival rate) (Jorgenson et al., senescence. For example, small rodents 1997). Thus, no clear pattern about the typically experience high mortality rates timing of senescence’s onset emerges and small mice rarely live more than one from field data. year in the wild. Despite this, Promislow One point that does emerge though is (1991) and Millar (1994) both found that patterns of mortality dynamics in evidence of actuarial senescence in popu- nature can be dramatically different lations of deer mice (Peromyscus manicu- than in captive populations. For instance, latus), and Morris (1996) found evidence animals in captivity often exhibit of reproductive senescence in white- Gompertzian mortality dynamics (Finch, footed mice (Peromyscus leucopus; see 1990)—that is, a monotonically log-linear Table 9.1). increase age-specific mortality rate—but Is there a general age or developmental many long-lived mammals and birds in stage that is a threshold for the onset of nature have a more phase-specific rela- senescence generally? Evolutionary tionship between survival, reproduction, senescence theory (Williams, 1957) sug- and age. That is, survival and reproduc- gests that aging should begin at about tion are relatively low and variable in the time of sexual maturity. Although young adults, both increase and remain observations consistent with the theory steady in prime-aged animals, and then have been reported for a German popula- both decrease again in oldest individuals tion of pied flycatchers (Sternberg, 1989) (Berubé et al., 1999; Caughley, 1966; as well as Florida scrub jays (McDonald Festa-Bianchet et al., 2003). Some of the et al., 1996), it was not found in any difference between captive and wild pop- of five populations of three species of ulations may be explained by the greater ungulates (roe deer, bighorn sheep, phenotypic variability within populations Pyrenean chamois) (Loison et al., 1999). in nature due to some combination of dif- In a particularly thorough long-term (12 ferential genetic endowment, consequent to 22 years) study, age at first reproduc- differential access to important resources, tion in all five populations was 2 years, and discrete social roles. A case in point yet annual female survival rate high is the previously mentioned study of remained high (0.9) with no statistical fallow bucks (McElligott et al., 2002). decline at least until age 7. Another Overall, mortality dynamics fit well with potential stage at which senescence a Gompertz model. However, when the could conceivably begin is “social matu- population was subdivided into nonrepro- rity,” the age at which animals are capable ducers (animals that had ceased to breed of actively competing for reproductive later in life) and producers (animals still P088387-Ch09.qxd 10/31/05 11:29 AM Page 257

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 257

actively reproducing) irrespective of age, Although the previous discussion and nonreproducers’ survival declined after our summary tables have primarily age 9 but reproducers showed no change focused on species patterns, it is worth in survival even to much later ages. recalling that not all populations of a A related question concerns whether species will necessarily exhibit the same there are general patterns in the timing of pattern. For example, Sanz and Moreno the beginning of reproductive senescence (2000) found no evidence of declining age- vis à vis survival senescence? To take specific survival rate of female pied fly- one well-known example, in humans, catchers (Ficedula hypoleuca) breeding in age-specific mortality increases from the central Spain, the southern part of this age of 10 to 11 years in societies with species’ breeding range, whereas Sternberg access to modern medicine (Finch, 1990), (1989), studying the same species in whereas subtle evidence of reproductive Germany, did observe a steady decline in senescence only becomes apparent by annual survival after age 1. The former about age 30 (vom Saal et al., 1994). authors suggest that the lack of observed Is this appearance of survival senescence senescence in their population relative to prior to reproductive senescence a general Sternberg’s may be due to its shorter phenomenon? Apparently not. Although migration route to wintering grounds in such a pattern is seen in female bighorn West Africa. Loison and colleagues (1999) sheep (Berubé et al., 1999, Jorgenson also found marked differences in senes- et al., 1997), European sparrowhawks cence patterns among different popula- (Newton & Rothery, 1997), European red tions of roe deer and bighorn sheep. deer (Clutton-Brock et al., 1988), baboons Gaps in our knowledge of senescence and African lions (Packer et al., 1998), the patterns in wild populations stem reverse—reproductive senescence preced- from the limited number of field studies ing survival senescence—is seen in barn systematically addressing the issue. swallows (Møller & De Lope, 1999; Saino However, there are also methodological et al., 2003) and wandering albatrosses difficulties associated with assessing (Weimerskirch, 1992). The difference senescence in natural populations that between onset of reproductive and sur- may continue to make the accumulation vival senescence can be substantial. In of new knowledge slow. Reliable studies bighorn ewes, survival senescence appears need to be based on long-term observa- by age 7 to 8 but reproductive senescence tions and/or experiments, enhanced by only after 13 years (Berubé et al., 1999; detailed knowledge of the mechanisms Jorgenson et al., 1997). Similarly, both affecting age-specific mortality and male and female red deer (Cervus fecundity, of the life-history tradeoffs at elaphus) experience survival senescence play in the population, and of the causes after 8 years of age, but reproductive and dynamics of these tradeoffs (Blarer senescence does not appear until age 12, et al., 1995). as evidenced by declining female fecundity and increasing calf mortality (Clutton- Brock, 1984) and the loss of males’ ability IV. Methodological Difficulties in to control harems (Clutton-Brock et al., Evaluating Senescence in Wild 1979). In contrast, wandering albatrosses Populations (Diomedea exulans) exhibit decreased egg size, breeding success, and breeding The scarcity of studies on senescence in frequency after age 20 but survival natural populations testifies to the logis- decreases significantly only after age 27 tical difficulty of observing and investi- (Weimerskirch, 1992). gating it. Some authors urge readers to be P088387-Ch09.qxd 10/31/05 11:29 AM Page 258

258 A. K. Brunet-Rossinni and S. N. Austad

careful when accepting published data on rather than absolute age estimation is senescence in wild populations (Gaillard sufficient. Even when absolute age esti- et al., 1994; Nisbet, 2001) and cite a vari- mation is required, rough precision may ety of methodological and interpretation suffice. For instance, if bowhead whales problems that must be overcome before indeed frequently live 150 to 200 years, we can develop a true understanding of as one report suggests, then an age esti- how, when, and why senescence occurs mator such as amino acid racemization in the wild. Potential methodological of the eye lens may be sufficiently pre- problems include the lack of visible, cise for most demographic analyses, even anatomical markers of aging, the use of if its margin of error is as much as a cross-sectional rather than longitudinal decade (George et al., 1999). life tables, small sample sizes of individu- Even well-established age estimators als at older ages, and the assumption that can be flawed. For example, the correla- an undetected individual is a dead indi- tion between growth layers in dentine and vidual. Interpretation problems include age used to establish calendar age of demographic heterogeneity, assumptions southern elephant seals was constructed of temporal constancy of environmental with animals only up to 8 years of age. and demographic conditions, and the It is known from individually marked choice of variables to measure senes- individuals that this species can live in cence. We will discuss some of these excess of 23 years (Hindell & Little, 1988). problems briefly. Whether extrapolation from known to The first question to ask before evalu- older unknown ages is warranted will ating any type of senescence in wild pop- depend on the biology of the species and ulations is whether you can be confident marker in question. Some traditional in assigning calendar age to study ani- markers, such as scale rings of fishes, mals. The gold standard should continue have failed validation tests in at least to be unique marking of individuals as some species (Nedreaas, 1990), whereas close to birth as possible. Even gold stan- newer techniques, such as growth rings of dards are not perfect, however, as indi- fish otoliths, have been extensively vali- vidual marks such as bands or tags can dated (Cailliet et al., 2001). be lost over time. Frequently, surrogate Cross-sectional or “snapshot” studies markers of calendar age, such as growth are the easiest and quickest way to inves- rings or plumage changes, are used as age tigate senescence in the wild. That is, estimators, sometimes without extensive demographic parameters are estimated validation. In fact, there are very few reli- from the age distribution of a current able markers of age in the wild, and even population. Indeed, life tables used in the ones that exist are often species-, human demography are virtually always population-, and even sex-specific. Thus, cross-sectional. Cohort tables, in which a their uncritical use can be deceptive. population of individuals born at the Tooth wear categories, for example, are same time is tracked throughout life, often used to estimate age in populations will generally be preferred for studies of of ungulates. However, tooth wear wild populations. Problems with cross- changes have been shown to decelerate sectional life tables have to do with the with age and differ between male and assumptions required for using such life female Norwegian red deer (Loe et al., tables as surrogates for cohort tables. 2003). On the other hand, the precision These include assumptions such as a required of an age estimator depends on stationary population age distribution the precision necessary to answer the and consistency of mortality patterns research question. Sometimes relative over time that are unlikely to be even P088387-Ch09.qxd 10/31/05 11:29 AM Page 259

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 259

approximately met in wild populations ual females, a researcher would have an (Gaillard et al., 1994). inaccurate picture concerning reproduc- If individuals in a population are tive senescence. not identified, even serial snapshots of a Even longitudinal studies are not fool- single cohort can mislead as a conse- proof, of course. For instance, the length quence of, say, population heterogeneity. of sampling intervals must be meaning- Consider, for instance, a hypothetical ful within the context of the life of cohort of five females that were born the the study organism. Sampling wild mice same year (see Figure 9.2). A researcher annually would clearly be useless given averages the number of offspring of the that their average longevity is only 3 to females at 1 year of age and then again at 4 months (Phelan & Austad, 1989). In 6 years in an attempt to detect reproduc- addition, there is the question of how tive senescence. Only three of the origi- long a longitudinal study should last in nal five females make it to age 6. The order to obtain accurate measures of results from these two snapshots appear demographic parameters and actually to indicate a decline in reproduction detect individuals that reach a senescent with age—that is, reproductive senes- stage. Climatic variability and changes cence. However, the reproduction of in population density are only two possi- individual females never changed over ble confounders of field demography. time. A more accurate assessment of Consider a population of fulmars in the population would be that for some England that has been tagged and reason—a tradeoff between reproduction tracked since 1950 (Dunnet & Ollason, and survival or genetic heterogeneity or 1978). Based on data from 1950 to 1962, chance—females with low reproductive researchers calculated the average adult rates live longer. Without data on sur- longevity to be 15.6 1.9 years (Dunnet vival and number of offspring for individ- et al., 1963). Based on data from 1950 to 1970, adult longevity increased to 25.0 4.3 years for males and 22.3 4.2 years for females (Cormack, 1973). 10 Finally, based on even later data (1958 to 1974), which now included only birds wearing a style of leg bands that were less likely to fall off, adult longevity was estimated at a substantially shorter 19.9 1.8 years (Dunnet & Ollason, 1978). Although the analytical methods differed somewhat in each of these studies, Fecundity (Number of offspring) 12345 6 the increasing study length combined Year with improved survival data from better Mean fecundity at age 1: Mean fecundity at age 6: (9 + 7 + 6 + 5 + 2)/5 = 5.8 (6 + 5 + 2)/3 = 4.3 leg bands account for at least part of the progressive changes in longevity Figure 9.2 Misleading inference from two cross- sectional samples (year 1 and 6) of cohort from one estimates. hypothetical population. Each line represents one Although the length of a longitudinal individual. Height of line on the y-axis represents study is important for evaluating fecundity of that individual which remains con- trends in senescence, so is the number of stant with age, length of the line along the x-axis individuals marked in the population. represents individual longevity. An apparent decline in reproduction is caused not by individual Needless to say, this is not a problem senescence, but rather by eariler death of more specific to field studies. Sample sizes fecund individuals. limit demographic analyses in both field P088387-Ch09.qxd 10/31/05 11:29 AM Page 260

260 A. K. Brunet-Rossinni and S. N. Austad

and lab. For instance, if only 50 individu- been seen in snowshoe hares (Lepus amer- als are included in a study, a real 1 per- icanus), which experience dramatic and cent difference in mortality rate is regular population cycles. That is, periods impossible to detect (Promislow et al., of very high hare density are followed by 1999). Also, because fewer and fewer population crashes to low densities, individuals are alive at later ages, the which in turn are followed by rebounds to statistical power to determine late-life high density again. During a 16-year demographic trends inevitably decreases. breeding program, Sinclair and colleagues For instance, in a typical study, only (2003) found that genetic lineages estab- 4 percent of the willow tits (Parus mon- lished from females of the high-density tanus) reached an age at which demo- phase had significantly lower reproductive graphic senescence could be detected output and longevity than lineages estab- (Orell & Belda, 2002). lished from females of the low-density Population heterogeneity can also phase. Furthermore, high-phase females obscure underlying demographic patterns showed declining reproductive output (Carnes & Olshansky, 2001; Festa- with age, whereas reproductive output Bianchet et al., 2003; McDonald et al., remained constant throughout life in low- 1996; Service, 2000; Vaupel & Yashin, phase females. 1985), as the previous example with fal- Despite the manifold complications low deer indicated. This interpretation associated with the study of senescence problem is not unique to field studies but in nature, an increasing number of stud- may be exaggerated in nature relative to ies are addressing the issue, with progres- the laboratory. Such heterogeneity can sively more sophisticated techniques. arise from genotypic or phenotypic differences among individuals. Consider that it is likely that only the strongest, healthi- V. Conclusions est individuals in a population will make it to old age. As individuals of low quality Studies of senescence in natural popula- are eliminated from the population, we tions of mammals and birds have see a resulting increase in survival rate. improved significantly in recent years This increase could be interpreted as neg- with the introduction of new method- ative senescence. Yet it does not represent ological and analytical techniques and individuals becoming more robust as they the continuation of long-term research get older. It is the result of a selection projects. Clearly, senescence occurs in process that has changed the nature of the the wild, although just as clearly it is sample being measured and it thus may not ubiquitous. Only by increasing the obscure a different underlying pattern if breadth of study species and focusing the same individuals were followed longi- specifically on the question of senes- tudinally. For instance, McDonald and cence will we eventually be able to com- colleagues (1996) found evidence of sur- prehensively assess interspecies patterns vival senescence in Florida scrub jays. The of senescence. Further research is also evidence was most clear when social role necessary to enhance our understanding was controlled for and a homogenous sub- of variation in senescence patterns population of high-quality individuals among populations of the same species. with the highest annual fledging rates The future of this line of research will be and greatest longevity were identified. based on longitudinal data from large Another example in which phenotypic samples of marked individuals in popu- heterogeneity may be misleading with lations that are well characterized and respect to demographic senescence may organisms for which we have a sound P088387-Ch09.qxd 10/31/05 11:29 AM Page 261

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 261

understanding of physiology, behavior, evolutionary causes from phenotypic and life-history options, constraints, and patterns can be misleading. Proceedings of tradeoffs. the Royal Society of London B, 262, 305–312. Acknowledgments Botkin, D. B., & Miller, R. S. (1974). Mortality rates and survival of birds. American We thank Ed Masoro, Daniel Promislow, Naturalist, 108, 181–192. Victor Apanius, and an anonymous reviewer Bradley, J. S., Wooller, R. D., Skira, I. J., & for helpful comments on an earlier version of Serventy, D. L. (1989). Age-dependent this manuscript. Preparation of the manuscript survival of breeding short-tailed was supported in part by an Ellison Senior shearwaters Puffinus tenuirostris. Journal Scholar Award and NIH grant AG022873. of Animal Ecology, 58, 175–188. Bronikowski, A. M., Alberts, S. C., Altmann, References J., Packer, C., Carey, K. D., & Tatar, M. (2002). The aging baboon: comparative Aebischer, N. J. (1986). Retrospective demography in a non-human primate. investigation of an ecological disaster in Proceedings of the National Academy of the shag, Phalacrocorax aristotelis: a general Sciences of the USA, 99, 9591–9595. method based on long-term marking. Broussard, D. R., Risch, T. S., Dobson, F. S., & Journal of Animal Ecology, 55, 613–629. Murie, J. O. (2003). Senescence and age- Aebischer, N. J., & Coulson, J. C. (1990). related reproduction of female Colombian Survival of Kittiwakes in relation to sex ground squirrels. Journal of Animal and position in colony. Journal of Animal Ecology, 72, 212–219. Ecology, 59, 1063–1071. Bumpus, H. C. (1898). The elimination of the Anderson, D. J., & Apanius, V. (2003). unfit as illustrated by the introduced Actuarial and reproductive senescence in sparrow, Passer domesticus. Biology a long-lived seabird: preliminary evidence. lectures, Marine Biological Laboratory, Experimental Gerontology, 38, 757–760. Woods Hole, MA. Apanius, V., & Nisbet, I. C. T. (2003). Serum Byers, J. (1998). The American Pronghorm. immunoglobulin G levels in very old Chicago: University of Chicago Press. common terns Sterna hirundo. Cailliet, G. M., Andrews, A. H., Burton, E. J., Experimental Gerontology, 38, 761–764. Watters, D. L., Kline, D. E., & Ferry- Askenmo, C., & Unger, U. (1986). How to be Graham, L. A. (2001). Age determination double-brooded: trends and timing of and validation studies of marine fishes: do breeding performance in the rock pipit. deep-dwellers live longer? Experimental Ornis Scandinavica, 17, 237–244. Gerontology, 36, 739–764. Austad, S. N. (1993). Retarded aging rate in an Carnes, B. A., & Olshansky, S. J. (2001). insular population of opossums. Journal of Heterogeneity and its biodemographic Zoology, 229, 695–708. implications for longevity and mortality. Bailey, S. R., & Milne, H. (1982). The influence Experimental Gerontology, 36, 419–430. of female age on breeding in the Eider Caughley, G. (1966). Mortality patterns in Somateria mollissima. Bird Study, 29, mammals. Ecology, 47, 906–918. 55–66. Cichón, M., Sendecka, J., & Gustafsson, L. Berubé, C. H., Festa-Bianchet, M., & (2003). Age-related decline in humoral Jorgenson, J. T. (1999). Individual differences, immune function in collared flycatchers. longevity and reproductive senescence in Journal of Evolutionary Biology, 16, bighorn ewes. Ecology, 80, 2555–2565. 1205–1210. Birkhead, T. R., & Goodburn, S. F. (1989). Clutton-Brock, T. H. (1984). Reproductive effort Magpie. In I. Newton (Ed.), Lifetime and terminal investment in iteroparous reproduction in birds (pp. 173–182). animals. American Naturalist, 123, 212–229. London: Academic Press. Clutton-Brock, T. H., Albon, S. D., Blarer, A., Doebeli, M., & Stearns, S. C. Gibson, R. M., & Guinness, F. E. (1979). (1995). Diagnosing senescence: inferring The logical stag: adaptive aspects of P088387-Ch09.qxd 10/31/05 11:29 AM Page 262

262 A. K. Brunet-Rossinni and S. N. Austad

fighting in red deer (Cervus elaphus L.). Dhondt, A. A. (1971). Some factors Animal Behavior, 27, 211–225. influencing territory in the great tit, Parus Clutton-Brock, T. H., Albon, S. D., & major. Gerfaut, 61, 125–135. Guiness, F. E. (1988). Reproductive success Dhondt, A. A. (1989). The effect of old age on in male and female red deer. In T. H. the reproduction of great and blue tit. Ibis, Clutton-Brock (Ed.), Reproductive success: 131, 268–280. studies of individual variation in Dow, H., & Fredga, S. (1984). Factors affecting contrasting breeding systems (pp. 325–343). reproductive output of the Goldeneye duck Chicago: University of Chicago Press. Bucephala clangula. Journal of Animal Coe, C. L., & Ershler, W. D. (2001). Intrinsic Ecology, 53, 679–692. and environmental influences on immune Dunnet, G. M., & Ollason, J. C. (1978). The senescence in the aged monkey. Physiology estimation of survival rate in the Fulmar, and Behavior, 73, 379–384. Fulmarus glacialis. Journal of Animal Comfort, A. 1979. The Biology of Ecology, 47, 507–520. Senescence (3rd ed.). Edinburgh: Churchill Dunnet, G. M., Anderson, A., & Livingston. Cormack, R. M. (1963). A study of survival Cormack, R. M. (1973). Commonsense of adult fulmars with observations on the estimates from capture-recapture studies. prelaying exodus. British Birds, 56, 2–18. In M. S. Bartlett & R. W. Hiorns (Eds.), Dunson, D. B., Baird, D. D., & Colombo, B. The mathematical theory of the dynamics (2004). Increased infertility with age in men of biological populations (pp. 225–234). and women. Obstetrics and Gynecology, London and New York: Academic Press. 103, 51–56. Coulson, J. C. (1984). The population dynamics Effros, R. (2001). Immune system activity. In of the Eider duck Somateria mollissima and E. J. Masoro & S. N. Austad (Eds.), evidence of extensive non-breeding by adult Handbook of the biology of aging (5th ed., ducks. Ibis, 126, 525–543. pp. 324–350). San Diego: Academic Press. Coulson, J. C., & Fairweather, J. A. (2001). Ericsson, G., Wallin, K., Ball, J., & Broberg, M. Reduced reproductive performance prior to (2001). Age-related reproductive effort and death in the black-legged kittiwake: senescence in free-ranging moose, Alces senescence or terminal illness? Journal of alces. Ecology, 82, 1613–1620. Avian Biology, 32, 146–152. Festa-Bianchet, M., Gaillard, J.-M., & Côté, S. Coulson, J. C., & Horobin, J. (1976). The D. (2003). Variable age structure and influence of age on the breeding biology and apparent density dependence in survival of survival of the Arctic tern (Sterna adult ungulates. Journal of Animal Ecology, paradisaea). Journal of Zoology, 178, 72, 640–649. 247–260. Finch, C. E. (1990). Longevity, senescence, Coulson, J. C., & Wooller, R. D. (1976). and the genome. Chicago: University of Differential survival rates among breeding Chicago Press. kittiwake gulls Rissa tridactyla. Journal of Fitzpatrick, J. W., & Woolfenden, G. E. (1988). Animal Ecology, 45, 205–213. Components of lifetime reproductive Daniel, C. W. (1977). Cell longevity in vivo. success in the Florida scrub jay. In In C. E. Finch & L. Hayflick (Eds.), T. H. Clutton-Brock (Ed.), Reproductive Handbook of the biology of aging (1st ed., success: studies of individual variation in pp. 122–158). New York: Van Nostrand contrasting breeding systems (pp. 305–320). Reinhold. Chicago: University of Chicago Press. Deevey, E. S. Jr. (1947). Life tables for natural Gaillard, J.-M., Allainé, D., Pontier, D., populations of animals. Quarterly Review Yoccoz, N. G., & Promislow, D. E. L. of Biology, 22, 238–314. (1994). Senescence in natural populations of Derocher, A. E., & Stirling, I. (1994). Age- mammals: a reanalysis. Evolution, 48, specific reproductive performance of 509–516. female polar bears (Ursus maritimus). Gaillard, J.-M., Delorme, D., Boutin, J.-M., Van Journal of Zoology (London), 234, Laere, G., Boisaubert, B., & Pradel, R. (1993). 527–536. Roe deer survival patterns: a comparative P088387-Ch09.qxd 10/31/05 11:29 AM Page 263

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 263

analysis of contrasting populations. Journal Jorgenson, J. T., Festa-Bianchet, M. of Animal Ecology, 62, 778–791. Gaillard, J.-M., & Wishart, W. D. (1997). George, J. C., Bada, J., Zeh, J., Scott, L., Effects of age, sex, disease and density on Brown, S. E., O’Hara, T., & Suydam, R. survival of bighorn sheep. Ecology, 78, (1999). Age and growth estimates of 1019–1032. bowhead whales (Balaena mysticetus) via Khalil, A. M., Murakami, N., & Kaseda, Y. aspartic acid racemization. Canadian (1998). Relationship between plasma Journal of Zoology, 77, 571–580. testosterone concentration and age, Grant, B. R., & Grant, P. R. (1989). Evolutionary breeding season and harem size in Misaki dynamics of a natural population: the large feral horses. Journal of Veterinary Medical cactus finch of the Galapagos Islands. Science, 60, 643–645. Chicago: University of Chicago Press. Koivula, M., Koskela, E., Mappes, T., & Gustafsson, L., & Pärt, T. (1990). Acceleration Oksanen, T. A. (2003). Cost of reproduction of senescence in the collared flycatcher in the wild: manipulation of reproductive Ficedula albicollis by reproductive costs. effort in the bank vole. Ecology, 84, 398–405. Nature, 347, 279–281. Komdeur, J. (1996). Influence of age on Hamer, K. C., & Furness, R. W. (1991). reproductive performance in the Seychelles Age-specific breeding performance and warbler. Behavioral Ecology, 7, 417–425. reproductive effort in great skuas, Kruijt, J. P., & de Vos, G. J. (1988). Individual Catharacta skua. Journal of Animal variation in reproductive success in Ecology, 60, 693–704. male black grouse, Tetrao tetrix.In Harris, M. P., Buckland, S. T., Russell, S. M., & T. H. Clutton-Brock (Ed.), Reproductive Wanless, S. (1994). Year- and age-related success: studies of individual variation variation in the survival of adult European in contrasting breeding systems shags over a 24-year period. Condor, 96, (pp. 279–290). Chicago: University of 600–605. Chicago Press. Hayflick, L. (2000). The future of ageing. Laaksonen, T., Korpimäki, E., & Nature, 408, 267–269. Hakkarainen, H. (2002). Interactive effects Haymes, G. T., & Blokpoel, H. (1980). The of parental age and environmental variation influence of age on the breeding biology of on the breeding performance of ring-billed gulls. Wilson Bulletin, 92, Tengmalm’s owls. Journal of Animal 221–228. Ecology, 71, 23–31. Heard, D., Barry, S., Watts, G., & Child, K. Leuze, C. C. C. K. (1980). The application of (1997). Fertility of female moose (Alces radio tracking and its effect on the alces) in relation to age and body behavioral ecology of the water vole, composition. Alces, 33, 165–176. Arvicola terrestris. In C. J. Amlamer & Heckel, G., & Von Helversen, O. (2002). D. W. MacDonald (Eds.), A handbook on Male tactics and reproductive success in biotelemetry and radio tracking the harem polygynous bat Saccopteryx (pp. 361–366). Oxford: Pergamon Press. bilineata. Behavioral Ecology, 13, 750–756. Loe, L. E., Mysterud, A., Langvatn, R., & Hilden, O. (1978). Population dynamics in Stenseth, N. C. (2003). Decelerating and sex Temminck’s stint Calidris temminckii. dependent toothware in Norwegian red Oikos, 30, 17–28. deer. Oecologia, 135, 346–353. Hindell, M. A., & Little, G. J. (1988). Loery, G., Pollock, K. H., Nichols, J. D., & Longevity, fertility and philopatry of two Hines, J. E.(1987). Age-specificity of black- female southern elephant seals (Mirounga capped chickadee survival rates: analysis of leonina) at Macquarie Island. Marine capture-recapture data. Ecology, 68, Mammal Science, 4, 168–171. 1038–1044. Horiuchi, S, & Wilmoth, J. R. (1997). Age Loison, A., Festa-Bianchet, M., Gaillard, J.-M., patterns of the life table aging rate for major Jorgenson, J. T., & Jullien, J.-M. (1999). causes of death in Japan, 1951–1990. Age-specific survival in five populations of Journals of Gerontology. Series A, Biological ungulates: evidence of senescence. Ecology, Sciences and Medical Sciences, 51, B67–B77. 80, 2539–2554. P088387-Ch09.qxd 10/31/05 11:29 AM Page 264

264 A. K. Brunet-Rossinni and S. N. Austad

Lunn, N. J., Boyd, I. L., & Croxall, J. P. (1994). Newton, I., Marquiss, M., & Moss, D. (1981). Reproductive performance of male Age and breeding in sparrowhawks Antarctic fur seals: the influence of age, Accipiter nisus. Journal of Animal Ecology, breeding experience, environmental 51, 327–343. variation and individual quality. Journal of Nisbet, I. C. T. (2001). Detecting and Animal Ecology, 63, 827–840. measuring senescence in wild birds: McDonald, D. B., Fitzpatrick, J. W., & experience with long-lived seabirds. Woolfenden, G. E. (1996). Actuarial Experimental Gerontology, 36, 833–843. senescence and demographic heterogeneity Nisbet, I. C. T., Apanius, V., & Friar, M. S. in the Florida scrub jay. Ecology, 77, (2002). Breeding performance of very old 2373–2381. common terns. Journal of Field McElligott, A. G., Altwegg, R., & Hayden, T. J. Ornithology, 73, 117–124. (2002). Age-specific survival and Nisbet, I. C. T., Winchell, J. M., & Heise, A. E. reproductive probabilities: evidence for (1984). Influence of age on the breeding senescence in male fallow deer (Dama biology of common terns. Colonial dama). Proceedings of the Royal Society of Waterbirds, 7, 117–126. London B, 269, 1129–1137. Nol, E., & Smith, J. N. M. (1987). Effects of Medawar, P. B. (1952). An unsolved problem age and breeding experience on seasonal of biology. London: Lewis. reproductive success in the song sparrow. Milonoff, M., Pöysä, H., & Runko, P. (2002). Journal of Animal Ecology, 56, 301–313. Reproductive performance of common Ollason, J. C., & Dunnet, G. M. (1978). Age, goldeneye Bucephala clangula females in experience and other factors affecting the relation to age and lifespan. Ibis, 144, breeding success of the fulmar, Fulmarus 585–592. glacialis, in Orkney. Journal of Animal Millar, J. S. (1994). Senescence in a Ecology, 47, 961–976. population of small mammals? Ecoscience, Ollason, J. C., & Dunnett, G. M. (1988). 1, 317–321. Variation in breeding success in Fulmars. Morris, D. W. (1996). State-dependent life In T. H. Clutton-Brock (Ed.), Reproductive history and senescence of white-footed success: studies of individual variation mice. Ecoscience, 3, 1–6. in contrasting breeding systems Møller, A. P., & De Lope, F. (1999). (pp. 263–278). Chicago: University of Senescence in a short-lived migratory bird: Chicago Press. age-dependent morphology, migration, Orell, M., & Belda, E. J. (2002). Delayed cost reproduction and parasitism. Journal of of reproduction and senescence in the Animal Ecology, 68, 163–171. willow tit Parus montanus. Journal of Mysterud, A., Yoccoz, N. G., Stenseth, N. C., Animal Ecology, 71, 55–64. & Langvatn, R. (2001). Effects of age, sex Owen-Smith, N. (1993). Comparative and density on body weight of Norwegian mortality rates of male and female kudus: red deer: evidence of density-dependent the costs of sexual size dimorphism. senescence. Proceedings of the Royal Journal of Animal Ecology, 62, 428–440. Society of London B, 268, 911–919. Packer, C., Herbst, L., Pusey, A. E., Bygott, J. D., Nedreaas, K. (1990). Age determination of Hanby, J. P., Cairns, S. J., & Mulder, M. B. Northeast Atlantic Sebastes species. (1988). Reproductive success of lions. In Journal du Conseil CIEM, 47, 208–230. T. H. Clutton-Brock (Ed.), Reproductive Newton, I. (1988). Age and reproduction in success: studies of individual variation the sparrowhawk. In T. H. Clutton-Brock in contrasting breeding systems (Ed.), Reproductive success: studies of (pp. 363–383). Chicago: University of individual variation in contrasting Chicago Press. breeding systems (pp. 201–219). Chicago: Packer, C., Tatar, M., & Collins, A. (1998). University of Chicago Press. Reproductive cessation in female Newton, I., & Rothery, P. (1997). Senescence mammals. Nature, 392, 807–810. and reproductive value in sparrowhawks. Partridge, L., & Barton, N. H. (1996). On Ecology, 78, 1000–1008. measuring the rate of ageing. Proceedings of P088387-Ch09.qxd 10/31/05 11:29 AM Page 265

CHAPTER 9 / Senescence in Wild Populations of Mammals and Birds 265

the Royal Society of London B, 263, Reid, J. M., Bignal, E. M., Bignal, S., 1365–1371. McCracken, D. I., & Monaghan, P. Perrins, C. M., & Moss, D. (1974). Survival of (2003b). Age-specific reproductive young great tits in relation to age of female performance in red-billed choughs parent. Ibis, 116, 220–224. Pyrrhocorax pyrrhocorax: patterns Phelan, J. P., & Austad, S. N. (1989). Natural and processes in a natural population. selection, dietary restriction and extended Journal of Animal Ecology, 72, 765–776. longevity. Growth, Development, & Aging, Reid, W. V. (1988). Age-specific patterns of 53, 4–6. reproduction in the glaucous-winged gull: Pistorius, P. A., & Bester, M. N. (2002). A increased effort with age? Ecology, 69, longitudinal study of senescence in a 1454–1465. pinniped. Canadian Journal of Zoology, 80, Ricklefs, R. E. (1998). Evolutionary theories of 395–401. aging: confirmation of a fundamental Promislow, D. E. L. (1991). Senescence in prediction, with implications for the natural populations of mammals: a genetic basis and evolution of life span. comparative study. Evolution, 45, 1869–1887. American Naturalist, 152, 24–44. Promislow, D. E. L., Tatar, M., Pletcher, S., & Robertson, R. J., & Rendell, W. B. (2001). A Carey, J. (1999). Below-threshold mortality: long-term study of reproductive implications for studies in evolution, performance in tree swallows: the influence ecology and demography. Journal of of age and senescence on output. Journal of Evolutionary Biology, 12, 314–328. Animal Ecology, 70, 1014–1031. Pugesek, B. H. (1987). Age-specific Rockwell, R. F., Cooch, E. G., Thompson, C. B., survivorship in relation to clutch size and & Cooke, F. (1993). Age and reproductive fledging success in California gulls. success in female lesser snow geese: Behavioral Ecology and Sociobiology, 21, experience senescence and the cost of 217–221. philopatry. Journal of Animal Ecology, Pugesek, B. H., & Diem, K. L. (1990). The 62, 323–333. relationship between reproduction and Saino, N., Ambrosini, R., Martinelli, R., & survival in known-aged California gulls. Møller, A. P. (2002). Mate fidelity, Ecology, 71, 811–817. senescence in breeding performance and Pugesek, B. H., Nations, C., Diem, K. L., & reproductive trade-offs in the barn swallow. Pradel, R. (1995). Mark-resighting analysis Journal of Animal Ecology, 71, 309–319. of a California gull population. Journal of Saino, N., Ferrari, R. P., Romano, M., Applied Statistics, 22, 625–639. Rubolini, D., & Møller, A. P. (2003). Ratcliffe, N., Furness, R. W., & Hamer, K. C. Humoral immune response in relation to (1998). The interactive effects of age and senescence, sex and sexual ornamentation food supply on the breeding ecology of great in the barn swallow (Hirundo rustica). skuas. Journal of Animal Ecology, 67, Journal of Evolutionary Biology, 16, 853–862. 1127–1134. Rattiste, K., & Lilleleht, V. (1987). Population Sanz, J. J., & Moreno, J. (2000). Delayed ecology of the common gull Larus canus in senescence in a southern population of the Estonia. Ornis Fennica, 64, 25–26. pied flycatcher (Ficedula hypoleuca). Réale, D., & Boussès, P. (1995). Effect of ewe Ecoscience, 7, 25–31. age and high population density on the Service, P. M. (2000). Heterogeneity in early nursing behavior of mouflon. individual mortality risk and its importance Ethnology, Ecology and Evolution, 7, for evolutionary studies of senescence. 323–334. American Naturalist, 156, 1–13. Reid, J. M., Arcese, P., & Keller, L. F. (2003a). Shock, N (1983). Aging of physiological Inbreeding depresses immune response in systems. Journal of Chronic Diseases, 36, song sparrows (Melospiza melodia): direct 137–142. and inter-generational effects. Proceedings Sinclair, A. R. E. (1977). The Africa buffalo: a of the Royal Society of London B, 270, study of resource limitation of populations. 2151–2157. Chicago: University of Chicago Press. P088387-Ch09.qxd 10/31/05 11:29 AM Page 266

266 A. K. Brunet-Rossinni and S. N. Austad

Sinclair, A. R. E., Chitty, D., Stefan, C. I., & reproductive aging in humans, laboratory Krebs, C. J. (2003). Mammal population rodents, and other selected vertebrates. cycles: evidence for intrinsic differences In E. Knobil & J. D. Neill (Eds.), The during snowshoe hare cycles. Canadian physiology of reproduction (2nd ed., Journal of Zoology, 81, 216–220. pp. 1213–1314). New York: Raven Press. Skogland, T. (1988). Tooth wearing by food Weimerskirch, H. (1992). Reproductive effort limitation and its life history consequences in long-lived birds: age-specific patterns of in wild reindeer. Oikos, 51, 238–242. condition, reproduction, and survival in the Sternberg, H. (1989). Pied flycatcher. In wandering albatross. Oikos, 64, 464–473. L. Newton (Ed.), Lifetime reproduction in Weladji, R. B., Mysterud, A., Holand, Ø., & birds (pp. 55–74). London: Academic Press. Lenvik, D. (2002). Age-related reproductive Sugiyama, Y. (1994). Age-specific birth rate effort in reindeer (Rangifer tarandus): and lifetime reproductive success of evidence of senescence. Oecologia, 131, chimpanzees at Bossou, Guinea. American 79–82. Journal of Primatology, 32, 311–318. Williams, G. C. (1957). Pleiotropy, natural Thomas, C. S. (1983). The relationship selection and the evolution of senescence. between breeding experience, egg volume Evolution, 11, 398–411. and reproductive success of the kittiwake Wolfe, L. D., & Noyes, M. J. S. (1981). Rissa tridactyla. Ibis, 125, 567–574. Reproductive senescence among female Turturro, A., Witt, W. W., Lewis, S., Hass, B. S., Japanese macaques (Macaca fuscata Lipman, R. D., & Hart, R. W. (1999). Growth fuscata). Journal of Mammalogy, 62, curves and survival characteristics of 698–705. the animals used in the biomarkers of aging Wooller, R. D., Bradley, J. S., & Croxall, J. P. program. Journals of Gerontology Biological (1992). Long-term population studies of Sciences, 54A, B492–B501. seabirds. Trends in Ecology and Evolution, Vaupel, J. W., & Yashin, A. I. (1985). 7, 111–114. Heterogeneity’s ruses: some surprising Wooller, R. D., Bradley, J. S., Skira, I. K., & effects of selection on population Serventy, D. L. (1990). Reproductive dynamics. American Statistician, 39, success of short-tailed shearwaters Puffinus 176–185. tenuirostris: relation to their age and Vom Saal, F. S., Finch, C. E., & Nelson, J. F. breeding experience. Journal of Animal (1994). Natural history and mechanisms of Ecology, 59, 161–170. P088387-Ch10.qxd 10/31/05 11:31 AM Page 267

Chapter 10

Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster

James W. Curtsinger, Natalia S. Gavrilova, and Leonid A. Gavrilov

I. Introduction monplace to demographers but foreign to most biologists. Here we review some For the last 15 years there has been a high basic analytic methods from demography level of interest in combining the meth- and lay out essential biological methods ods of biology and demography to investi- and questions, hoping to introduce both gate aging in experimental populations. biologists and demographers to the hybrid The hybrid field of biodemography field. addresses a wide range of questions about The integration of genetic and demo- aging organisms and aging populations, graphic methods requires an experimen- and also attempts to provide insights into tal system that is genetically defined human aging (Wachter & Finch, 1997). and amenable to large-scale popula- A handful of issues have preoccupied tion studies. The fruit fly Drosophila the nascent field: To what extent are melanogaster is an obvious candidate, the genetic phenomena that influence life being one of the premiere experimental histories age-specific in their effects? systems for basic research in genetics. How malleable are the patterns of sur- The genome is completely sequenced, vival and death among the oldest organ- and the flies can be reared in large isms? Why do populations often exhibit numbers (tens of thousands of organ- mortality plateaus? How have observed isms). The nematode Caenorhabditis survival patterns evolved under the influ- elegans and some yeast species also ence of mutation and natural selection? have those desirable characteristics, but To what extent do survival patterns in other standard experimental systems do populations reflect underlying changes in not. The genetics of house mice (Mus individual organisms? All of these ques- musculus) is an important and growing tions are challenging, and none fully area of research, but large-scale answered yet. Addressing them requires a population studies with rodents are set of analytical techniques that are com- impractical. Demography of medflies

267 P088387-Ch10.qxd 10/31/05 11:31 AM Page 268

268 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

(Cerititus capitata) and several para- after emergence, depending on genotype sitic wasp species has been investi- and environmental conditions. gated in large experimental populations (Carey, 2003), but those systems are A. Collection of Survival Data genetically undefined. An interesting feature of Drosophila as an experimen- Survival experiments with laboratory pop- tal model is the similarity of its mortal- ulations of Drosophila are typically longi- ity kinetics to that of humans, first tudinal, large scale, and complete. That is, noted by Raymond Pearl (1922). Both age-synchronized cohorts consisting of species have a relatively short period thousands or tens of thousands of experi- of high initial mortality, followed by a mental animals are established with relatively long period of mortality newly emerged adults and are observed increase, and then deceleration at over time. As the cohorts age, dead ani- advanced ages (although the period of mals are removed, counted, and recorded mortality deceleration and mortality on a daily basis. Observations continue plateau in Drosophila is longer than in until the last fly dies, typically around humans). 100 days after emergence (depending on Here we concentrate on genotype and sample size; see below). D. melanogaster, a holometabolous Experimental populations are maintained insect. Larvae hatch from eggs about under controlled environmental condi- 24 hours after laying, feed voraciously tions, including temperature, light cycle, for a week, and then pupate. Adults and humidity. Initial population density is emerge from the pupal case after a few also controlled, at least approximately; in days of metamorphosis and are sexually smaller experiments, exact numbers of mature within 24 hours. In the wild, flies are counted, whereas in larger experi- D. melanogaster adults probably live ments, density is approximated by one to two weeks. In laboratory culture, volume or weight of anesthetized flies flies are normally maintained on a two- (one large female weighs 1mg., whereas week generation schedule but can live males are typically 30 percent smaller). much longer as adults. In a typical Experimental populations are often outbred population, adults survive 30 housed in cages of one to several liters in to 50 days on average, depending on volume, each holding up to a thousand temperature and other environmental individuals, but half-pint milk bottles and conditions. Inbreeding and increased finger-sized glass vials are sometimes temperature reduce mean adult life used. There is always fresh fly food in the spans, while artificial selection for containers, which serves as both an ovipo- increased life span is capable of dou- sition medium and a source of nutrition bling it. Maximum adult life spans for adults and larvae. Frequent replace- observed in large experiments typically ment of the medium and changing cages exceed 100 days. There is no precise prevents unwanted recruitment of new definition of young, middle-aged, or adults into experimental populations. old adult flies. At two weeks after Populations used for survival studies emergence, metabolic rate and gene typically consist of males and females in expression reach low levels characteris- approximately equal proportions when tic of remaining adult life (Tahoe et al., experiments are initially set up, but 2004; Van Voorhies et al., 2003, 2004). because of differential survival, the sex For females, old age in flies is probably ratio changes over time. In mixed-sex pop- best understood as the age after egg lay- ulation cages, females actively reproduce ing has ceased, usually 40 to 60 days and generally exhibit shorter average life P088387-Ch10.qxd 10/31/05 11:31 AM Page 269

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 269

spans than males (Curtsinger & Khazaeli, arithmetic average survival time, has 2002; Curtsinger et al., 1998; Fukui et al., intuitive appeal as a descriptor of survival 1993, 1995, 1996; Khazaeli & Curtsinger, ability, but the information contained in 2000; Khazaeli et al., 1997; Pletcher, 1996; that summary statistic is limited. The Resler et al., 1998;). Because flies reach most critical limitation in the present sexual maturity soon after emergence, context is that the mean gives little infor- mating behavior begins almost immedi- mation about the age-specificity of sur- ately in mixed-sex populations. It is possi- vival patterns. Two cohorts could have ble to study the survival characteristics of very similar means but experience vastly unmated flies in single-sex populations by different life histories. For instance, if one anesthetizing newly emerged adults and population suffers mortality only at mid- then sorting the sexes under a dissecting dle age, whereas a second experiences microscope when cohorts are initially mortality equally and exclusively at early established (Miyo & Charlesworth, 2004; and late ages, mean life spans in the two Semenchenko et al., 2004). populations will be similar. Maximum There is significant uncontrolled envi- observed life span is also frequently ronmental variation that affects death reported but is similarly uninformative rates in experimental populations of about age-specific events. Drosophila. The magnitude of the variation The central conceptual tool for organiz- is perhaps underappreciated. For instance, ing and analyzing age-specific aspects of it is not unusual to see four- or five-fold survival data in experimental populations variation in individual life spans among of Drosophila and other species (indeed, flies of the same genotype sharing the other objects) is the cohort life table. It is same food and population cage. This is not interesting that Drosophila was the sec- a peculiarity of fly life spans; biologists ond species, after humans, for which such have long recognized that quantitative demographic life tables were constructed traits vary between organisms, even if (Pearl & Parker, 1921). The essential fea- they are genetically identical and reared tures of the life table are that age classes under carefully controlled conditions (for are defined by sampling intervals, and for a review, see Finch & Kirkwood, 2000; each age class (life table row) specific vari- Gavrilov & Gavrilova, 1991). Because of ables (life table columns) are estimated. this irreducible variation, which is not The first variable is the fraction of the well understood, survival experiments total population dying while in age class should be highly replicated, in some cases x, denoted dx. The distribution of dx, a involving hundreds of populations. Ideally, typical example of which is shown in data from genotypes or treatments that are Figure 10.1a, is approximately bell-shaped to be contrasted are collected simul- but not symmetrical, in contrast to the taneously in order to avoid confounding normal curve. The long right-skewed tail uncontrolled environmental variations represents the oldest survivors of the with treatment or genotype effects. cohort and is observed even in genetically homogeneous populations. The second variable, survivorship, is represented as l B. Data Analysis: Mean Life Span and x and is defined as the probability of sur- Survivorship vival from the beginning of the experi- The central problem in survival analysis ment until the beginning age interval x. is to summarize and interpret large That probability is estimated by the amounts of information hidden in the proportion of the initial cohort that survival data. Raw data consist of esti- remains alive at age x. Survival curves, mated ages at death. Mean life span, the which show plots of lx versus x, start at P088387-Ch10.qxd 10/31/05 11:31 AM Page 270

270 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

100 percent and decline to zero at the age depends on the length of the age interval when the last animal in the cohort dies x for which it is calculated, which ham- (see Figure 10.1b). Survival curves have pers both analyses and interpretation. For built-in smoothing because they are non- example, one-day probabilities of death increasing (the proportion of the initial may follow the Gompertz law of mortal- cohort remaining alive at age x can ity, but probabilities of death calculated only be the same or lower at age x x). for other age intervals with the same data For this reason, even relatively small may not (Gavrilov & Gavrilova, 1991; le cohorts produce smoothly declining sur- Bras, 1976). A meaningful descriptor of vivorship curves. Life-table values of lx the dynamics of survival should not xx and dx are related as follows: l depend on the arbitrary choice of age lx dx, where x is the length of the sam- intervals. Another problem is that, by pling (age) interval, typically equal to one definition, qx is bounded by unity, which day for fly experiments. It is important to makes it difficult to scale the variable for emphasize that both lx and dx are cumula- studies of mortality at advanced ages. tive indicators that depend on preceding A more useful indicator of mortality is death rates. Events early in the life his- the instantaneous mortality rate, or haz- tory, such as a temporary epizootic, can ard rate, x, which is defined as follows: affect survivorship and the fraction dying in later age classes, even in old age. In dNx this sense, lx and dx reflect the survival x Nxdx history of the cohort up to and including age x. where Nx is the number alive at age x. The hazard rate does not depend on the C. Data Analysis: Probability of Death length of the age interval; it reflects and Mortality Rate instantaneous risk of death. It has no upper bound and has the dimension of a Unlike survivorship and fraction dying, rate (time 1). One of the first empirical which have “memory,” some other life- estimates of hazard rate was proposed table variables are noncumulative and x by Sacher (1956): better suited to detecting age-specific effects. Age-specific probability of death 1 x x (q ) is defined as the conditional proba- (ln l ln l ) x x x x x bility of dying in the interval x for 2 2 individuals that survive to the beginning of interval x. It is estimated as the 1 lxx ln x number of deaths that occur in age 2 lxx class x, divided by the number of individuals entering class x. An example of This estimate is unbiased for slow age-specific probability of death is changes in hazard rate (Sacher, 1966). A shown in Figure 10.1c. Note that in this simplified version of the Sacher estimate particular example, the age-specific (for small age intervals equal to unity) is probability of death grows monotoni- often used in biological studies of mortal- cally with age up to an advanced age ity: x ln(1qx) (see Carey, 2003) and and then levels off, a phenomenon dis- assumes constant hazard rate in the age cussed in detail later. interval. Although probability of death is useful The Cutler-Ederer (1958) estimate (also and intuitive, it has limitations. The called the actuarial hazard rate) is based main problem is that the value of qx on the assumption that deaths are P088387-Ch10.qxd 10/31/05 11:31 AM Page 271

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 271

B A 100 400

80 300

200

40 Number dying, d(x) Percent survivors, l(x) 100 20

0 0 020406080 020406080 Age, days Age, days

C D 1 1

0.1 0.1

0.01 0.01 Hazard rate in a log scale Sacher estimate Cutler-Ederer estimate One-day probability of death in a log scale 0.001 0.001 020406080 0204060 80 Age, days Age, days Figure 10.1 Life-table variables as a function of adult age, estimated for experimental population of 8,926 D. melanogaster males. (a) Number dying dx; (b) survivorship lx; (c) age-specific probability of death qx; (d) age-specific mortality (hazard) rate x. The subscript “x” indicates adult age in days since eclosion. (Unpublished data of Khazaeli, Gavrilova, Gavrilov, & Curtsinger)

uniformly distributed in the age interval Here, cx is number of censored individu- and that all cases of withdrawal (censor- als during the age interval (for example, ing) occur in the middle of the age number of flies accidentally escaping the interval: cage during food replacement). The hazard rate is measured at the midpoint of d x x the age interval. Gehan and Siddiqui x 2 c d xl x x (1973) used Monte Carlo simulation to x 2 2 show that for samples less than 1,000, P088387-Ch10.qxd 10/31/05 11:31 AM Page 272

272 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

the Sacher method may produce biased Perhaps the most common misunder- results compared to the Cutler-Ederer standing among biologists about survivor- method, whereas for larger samples, the ship and mortality is the widespread Sacher estimate is more accurate. The assumption that rates of senescence can advantage of the Cutler-Ederer estimate be easily seen in the slopes of survivor- is its availability in standard statistical ship curves. The apparent or actuarial packages (such as SAS and Stata), which rate of senescence, defined as the rate at compute actuarial life tables. Despite the which risk of death increases with age, is apparent differences between Cutler- precisely reflected in the slope of the Ederer and Sacher estimates, the methods mortality curve: a steep slope indicates produce very similar results for real data rapid actuarial senescence, a shallow (see Figure 10.1d). Note that the mortal- slope indicates negligible senescence, and ity curve, depicting x as a function of x, a zero slope indicates no senescence. Of describes survival events in true age-spe- course, the slope of the survivorship cific fashion. It clearly illustrates the rate curve bears a mathematical relationship of actuarial senescence, usually defined to the slope of the corresponding mortal- as the slope of the mortality curve, and is ity curve, but not one that is easily particularly useful for examining details grasped by visual inspection. The prob- of death rates among the oldest survivors lem is that even populations that experi- of a cohort. In contrast, the details of ence no apparent senescence (constant shape in a survivorship curve as it probability of death at all ages) will approaches the x-axis are generally indis- exhibit exponentially declining survivor- tinct (but see Pearl & Parker’s method, ship with increasing age. Thus, informa- described below). tion about the rate of senescence is The differences between survivorship present in a survivorship curve only as a and mortality are fundamental. The for- deviation from the exponential, a quanti- mer depends on all previous cohort his- tative measure that is not well suited to tory, whereas the latter reflects risk casual inspection. Pearl and Parker (1924) specific solely to the age group under addressed this problem by examining sur- study. This distinction has often been vivorship in semi-logarithmic plots. This misunderstood or overlooked by biolo- approach may be useful in defining peri- gists. Rose’s (1991) influential text on ods of mortality leveling-off (mortality evolutionary biology of aging contains plateaus): survivorship curves in semi- dozens of figures, extensive discussion of logarithmic scale should be linear if age-specific life-history phenomena, and mortality is constant. Economos (1979, not a single depiction of a mortality 1980) used this method for demonstrating curve, either experimental or theoretical. non-Gompertzian mortality kinetics at Similarly, Kirkwood’s (1999) general text advanced ages, but the technique has not on causes of aging gives considerable been widely used in recent years. notice to age-specific phenomena but There are probably several reasons that employs survivorship rather than mortal- biologists in some fields have not, until ity throughout. In an otherwise excellent recently, adequately appreciated the paper on chromosomal mapping of information that can be gained by esti- genes that influence mean life spans in mating mortality rates. Survivorship Drosophila, Nuzhdin and colleagues curves have intrinsic smoothing, as men- (1997) test an evolutionary model of tioned above, whereas mortality curves senescence by examining age-specific tend to be jumpy. For a single data set variance in lx, when the issue is clearly plotted both ways, the mortality estima- variance in x. tion makes the data look noisy, whereas P088387-Ch10.qxd 10/31/05 11:31 AM Page 273

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 273

the survivorship curve gives an appear- intervals, in which case intervals will ance of orderly behavior. Accurate esti- have to be combined and the accuracy of mation of age-specific mortality rates hazard rate estimation will decline. requires larger sample sizes than those Thus, the minimum sample size of needed for estimating means or survivor- experimental populations for hazard rate ship but provides extra sensitivity in studies may be estimated on the basis of studies of short-term response to phe- expected risk of death during younger nomena such as heat shock (Khazaeli ages, when mortality is low. et al., 1997) and dietary restriction (Mair For example, if the expected risk of et al., 2003; Pletcher, 2002). The sample- death is 1 per 1,000 during a one-day size requirement is especially critical for period, then the sample size should be at the oldest ages; large initial cohort sizes least 1,000. If mortality at younger ages are required in order to have adequate is higher, then smaller sample sizes will numbers of animals alive for estimation suffice. This rule of thumb does not of death rates at the older ages. apply to studies of mortality deceleration In the 1920s, Raymond Pearl, an early and leveling-off. This phenomenon hap- advocate of biostatistics and experimen- pens later in life, after a significant part tal investigation of populations, pub- of population has died and the remaining lished a series of papers on Drosophila number of animals is a small fraction of life spans that employed relatively large the initial cohort. The empirical rule sample sizes. For instance, Pearl and here may be to have at least 50 animals Parker (1924) collected survival data on alive at the age when mortality decelera- about 4,000 flies from two strains. Since tion starts so that hazard rate estima- the 1950s, radiobiologists have routinely tions would not be distorted by small employed large sample sizes to estimate numbers of deaths. If one is interested in mortality rates in survival studies with short-term effects of caloric restriction or experimental organisms. However, in other interventions on mortality kinetics spite of those pioneering efforts, up until at middle ages close to the modal life around 1990, it was standard practice span, then much smaller sample sizes among experimental gerontologists, evo- may be sufficient because numbers of lutionary biologists, and geneticists to organisms at risk and numbers of deaths employ small populations in studies of will be substantial. Drosophila survival, typically on the order of 50 to 100 animals per experimen- D. Smoothing and Model Fitting tal treatment or genotype. Such sample sizes sufficed to give reasonably accurate Two approaches are commonly used to estimates of mean life spans and aestheti- describe trends in the (often noisy) data cally pleasing survivorship curves but on age-specific mortality. One approach provided virtually no information about is to apply a non-parametric smoothing death rates in old age. procedure. For data organized in the form Sample size requirements will depend of a life table, smoothing can be accom- on the specific question being asked. For plished by widening the age intervals. If accurate estimates of hazard rates, it is times to death for each individual in the necessary to have some events (deaths) in sample are known with reasonable accu- each age interval. At younger ages, when racy, and/or small sample size does not mortality rates are low, it would be desir- allow construction of a conventional life able to have at least one death in each table, then the method of hazard rate observation interval. In small samples smoothing using kernel functions may there might be no deaths during some be more appropriate (Ramalu-Hansen, P088387-Ch10.qxd 10/31/05 11:31 AM Page 274

274 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

1983). The latter method is more compu- Differences in frailty might be innate and tationally complex, although special rou- fixed throughout life, or modified over tines are available now in SAS and Stata. the life history. Strehler and Mildvan Applying methods of non-parametric (1960) showed that when there is such smoothing decreases statistical noise and heterogeneity, the observed population facilitates visual inspection of mortality mortality pattern deviates from the plots but does not allow quantitative underlying mortality for individuals. analysis of life-span data. Following Beard (1963), the observed The second major approach for summa- mortality in the population is rizing and simplifying mortality estimates Bx 2 is parametric model fitting, which allows µx Ae /[1 (x)], researchers to describe the observed mortality kinetics using a small number of where A and B are as defined in the parameters of a specified mortality model. Gompertz model, 2 is the variance for Although there are many possible models frailty in the population, and (x) in the literature, three are widely used (A/B)(eBx 1). Note that when 2 0, by biologists. The venerable model of there is no heterogeneity in the popula- Gompertz (1825) specifies exponentially tion, and the logistic reduces to the increasing hazard rate with increasing age: Gompertz model. However, if 2 0, then the logistic curve increases expo- Bx µx Ae nentially at early ages and plateaus at more advanced ages (as x becomes large, 2 where A is initial mortality rate, e is the ux approaches B/ ). Yashin and col- base of the natural logarithms, and B, the leagues (1994) showed that this model slope parameter, controls the rate at applies under two biologically different which mortality increases with age. circumstances: when individuals possess Estimates of A in laboratory populations a fixed frailty from birth that differs from of D. melanogaster are typically in the that of other individuals, and when all range 0.005 to 0.010 per day, whereas individuals start life with identical frail- B often lies in the range 0.04 to 0.10 ties but then randomly acquire differ- per day (Fukui et al, 1993). The Gompertz ences in frailty during adulthood. model produces a straight line in semi-log A third model used by biologists is also plots of hazard rate versus age, with the motivated by the observation that mor- y-intercept estimating the initial mortal- tality data often exhibit plateaus at older ity rate and the slope estimating the rate ages. This approach involves fitting two of senescence. The aging rate is some- curves to the mortality data. Curtsinger times summarized by the mortality rate and colleagues (1992) proposed a two- doubling time (MRDT), defined as stage Gompertz model, in which a ln(2)/B. However, this measure has lim- Gompertz curve is fit to the data at ited applicability to Drosophila because young ages up to some breakpoint age, of non-Gompertzian mortality dynamics and then a second Gompertz curve with at advanced ages; in particular, as B shallower slope is fit to the older ages. approaches zero in old age, the MRDT This model includes five parameters: two approaches infinity. intercept and two slope parameters for A second widely used model is the two Gompertz curves, and a fifth param- logistic, which is motivated by the eter for the breakpoint. Zelterman and possibility that individuals in the same Curtsinger (1994, 1995) applied the population can have different frailties method to fly data, and Vaupel and col- (age-dependent chances of death). leagues (1994) used it for nematodes. P088387-Ch10.qxd 10/31/05 11:31 AM Page 275

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 275

Drapeau and colleagues (2000) employed The nonlinear least squares method a similar method, except older ages were provides an alternative to maximum fit to a linear rather than exponential likelihood. This method is implemented curve. It should be noted that mortality in most statistical software packages and trajectories following the Weibull (power) allows researchers to fit a large variety of law of mortality may resemble a two- nonlinear models. The limitation of this stage Gompertz model in semi-log coor- method is its theoretically less desirable dinates (see Chapter 1 in this volume). optimality properties compared to the The two major methods of parameter maximum likelihood, and less applicabil- estimation for nonlinear models are max- ity to censored data. Both methods are imum likelihood and nonlinear least sensitive to the choice of initial parame- squares. The maximum likelihood ter estimates and outliers. approach is based on maximizing the There is a tradeoff between flexibility likelihood function, or the probability of and convenience of the nonlinear least obtaining a particular set of data given squares method and the accuracy of the the chosen probability model. Maximum maximum likelihood approach. In prac- likelihood methods provide unbiased and tice, the theoretical considerations men- efficient parameter estimates for large tioned above are apparently not crucial, data sets (though the estimates may be and the two approaches generate similar heavily biased for small samples). results. For example, Gehan and Siddiqui Another advantage is that maximum (1973) conducted a simulation study of likelihood generates theoretically more fitting Gompertz and some other hazard accurate confidence bounds for parame- models to survival data. The authors ter estimates. An important property of concluded that the least squares esti- maximum likelihood for survival data is mates are nearly as efficient as maxi- that censored observations can be readily mum likelihood when sample size is 50 introduced (see Filliben, 2004). The limi- or more. They also found that the tation of this method is the need for weighted least squares approach, which specifying the maximum likelihood accounts for systematic decrease of the equations for each particular function sample size with age, generated more not implemented in the standard statisti- efficient but less accurate parameter esti- cal software packages, which often is not mates compared to the nonweighted trivial. Standard statistical packages pro- method. Thus, maximum likelihood is a vide maximum likelihood estimates preferred method in those cases where for a limited number of models. For the statistical software is readily avail- example, the Stata package has a proce- able or the optimization procedure can dure for maximum likelihood estimation be easily implemented. Otherwise, the of Gompertz and logistic models. nonlinear least squares may be a reason- Maximum likelihood estimation of able choice. Gompertz, Gompertz-Makeham, logistic, It is important to recognize the limita- and logistic-Makeham models is imple- tions and pitfalls of model fitting. The mented in WinModest, a program writ- main problem is uneven statistical ten and distributed by S. Pletcher (Baylor power. At young ages, there are relatively College of Medicine, Houston) specifi- few deaths; at the oldest ages, death rates cally for calculating basic statistics, fit- are high, but there are relatively few ting mortality models to survival data, organisms. At middle ages, there are and partitioning mean longevity differ- large numbers of both organisms at risk ences between populations (Pletcher & and deaths, and so statistical power Curtsinger, 2000a). for estimation of mortality rates is P088387-Ch10.qxd 10/31/05 11:31 AM Page 276

276 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

concentrated in those middle age classes. because they potentially disrupt gene Consequently, model fitting to the entire expression or function at the insertion life history can give very accurate site. Screening of P-element inserts led descriptions of the dynamics of middle to the discovery of life-extending age and can be systematically biased at “methuselah” (mth) and “I’m not dead early and late ages. yet” (Indy) single-gene mutations (Lin et al., 1998; Rogina et al., 2000). Clark & Guadalupe (1995) used P-element inser- II. Experimental Evidence for tion lines to investigate the genetic basis Age-Specific Effects of senescence and found that otherwise genetically identical lines differed in sur- If new mutations and genetic variants vivorship and mean life span under the segregating in populations modify influence of P-induced insertions. The chances of survival by a constant factor at authors claimed that some of the P-ele- all ages (a situation known among ment insertions led to reduced post- demographers as “proportional hazards”), reproductive survival without affecting then there is no true age specificity; all is early life history, and that P-element known from observations at a single age. inserts altered the ages at which mortal- However, if genes alter survival charac- ity curves leveled off, though few demo- teristics specifically at certain prescribed graphic details were given. ages or stages of the life cycle, with no effect or very different effect at other ages, then the situation is more complex, B. Mutation Accumulation Experiments and much more interesting. The evolu- The term mutation accumulation refers tionary theory for the evolution of senes- to both a theory of the evolution of cence requires age-specificity of genetic senescence (Medawar, 1952) and an effects (Charlesworth, 1980; Curtsinger, experimental design pioneered in 2001; Hamilton, 1966; Medawar, 1952; Drosophila (Mukai, 1964). It is the latter Williams, 1957). As we discuss below, sense of the term that concerns us for evolutionary models currently under the moment, although the former will be investigation are sensitive to the precise relevant later. The goal of a mutation degree of age specificity. Proving the exis- accumulation experiment is to measure tence of such age-specific genetic varia- the rate at which new genetic variation tion is difficult, especially at the older spontaneously arises in a population, and ages, but mounting evidence suggests to measure the phenotypic effects of that there may be a substantial degree those new mutations. General features of of age specificity of genetic effects in mutation accumulation experiments Drosophila. In the following sections, we using Drosophila are as follows: starting describe several different types of experi- with a single highly inbred line of flies, mental evidence that address that issue. multiple sub-lines are established and maintained separately in small populations for dozens or even hundreds of A. P-Element Tagging generations. Spontaneous germline muta- P-elements are naturally occurring trans- tions occur independently in the various posable genetic elements (transposons) sub-lines, causing them to diverge both specific to Drosophila. Their ability to genetically and phenotypically. The sub- insert into random chromosomal loca- lines are kept at small census numbers tions throughout the genome makes so that new mutations have a reasonable them useful tools for genetic research, chance to increase to fixation within P088387-Ch10.qxd 10/31/05 11:31 AM Page 277

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 277

each particular line by random genetic of mutation affecting early survival than drift. The rate at which sub-lines diverge late survival. Surprisingly, there appeared phenotypically provides an estimate of to be no upward or downward bias of the rate of input of new genetic variation mutational effects on mortality rates affecting the particular trait assayed. (mutations increasing mortality are as fre- The first mutation accumulation quent as mutations decreasing mortality), study of age-specific mortality was exe- contradicting the usual assumption that cuted by Pletcher and colleagues (1998), almost all mutations are deleterious to who established 29 sub-lines of carriers. One possible explanation of this D. melanogaster from a single highly paradox may be related to elimination of inbred progenitor pair. Sub-lines were many deleterious mutations through maintained for 19 generations, and then selective deaths at early larval stages of survival data were collected on approxi- Drosophila development. mately 100,000 flies. Mutational effects Mack and colleagues (2000) and were detected by comparing age-specific Yampolsky and colleagues (2001) used a mortality rates in each sub-line with different experimental design, the “middle that of the progenitor stock, which was class neighborhood” method, to accu- maintained in nonmutating condition by mulate mutations affecting mortality, cryopreservation. Significant mutational fecundity, and male mating ability on a variance for age-specific mortality was genetically heterogeneous background of detected, but only for flies aged less than recently collected flies. They found clear 30 days post-emergence. Most of the new evidence of age-specific effects of new mutations were highly age-specific, each mutations after 20 generations of muta- affecting survival rates over a well- tion accumulation, including many effects defined age window of one or two weeks. limited to middle and advanced ages. This Mutations that affected mortality at all result contrasts with that of Pletcher and ages were also detected, but their contri- colleagues (1998, 1999), who found mostly bution to overall mutational variance early age effects. In both studies, the was small. The conclusion from this degree of age specificity declined in later study is that most new mutations have generations of the experiment. age-specific effects, but the failure to Martorell and colleagues (1998) exe- detect mutational variance at very old cuted a large mutation accumulation ages is difficult to interpret. It is unclear experiment to study life history in whether the failure to detect late-acting D. melanogaster, maintaining 94 sub- mutations is due to smaller sample sizes lines for 80 generations. They found and loss of statistical power, to inher- evidence for small mutational effects on ently lower mutation rates for alleles mean life span, but because mortality that specifically affect old age survival, rates were not assayed, the experiment or a combination of those factors. provides no information about age speci- Pletcher and colleagues (1999) con- ficity of genetic effects. If Pletcher and tinued the mutation accumulation colleagues (1999) are correct about muta- experiment, assaying mortality rates at tions decreasing mortality as often as 47 generations of divergence, and also they increase it, then Martorell and col- jointly analyzing data at three time points leagues (1998) might have underesti- (10, 19, and 47 generations). These assays mated the rate of mutations that modify involved approximately a quarter of a mil- mean life spans. Similar remarks apply to lion flies. Further evidence for highly age- studies of life span and related characters specific mutation was found, and once in flies exposed to mutagenic chemicals again there was evidence for higher levels (Keightley & Ohnishi, 1998). Mutation P088387-Ch10.qxd 10/31/05 11:31 AM Page 278

278 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

accumulation experiments on life- and others like it also demonstrate that history traits have also been executed genetic variation can produce specific using the nematode C. elegans (Keightley late-onset phenotypes in adult et al., 2000). Drosophila. Evidently, lack of cell division in adults does not preclude age- specific effects in older flies. C. Neurogenetics and Gene Expression There is also evidence for age-specific Adult Drosophila are entirely post- genetic effects in modern studies of gene mitotic organisms; that is, all cell divi- expression. It used to be widely assumed sion is completed when the animal meta- that the regulation of gene expression, morphoses from larval to adult stage. which is capable of transforming single This contrasts sharply with other organ- cells into highly differentiated and isms, such as humans, in which cell divi- spatially structured mature organisms, sion continues throughout the adult life becomes chaotic in old age. This view is span. It has been suggested that the lack now rejected, in part because of evidence of cell division in adult flies precludes from Drosophila (Helfand & Rogina, 2000, late-onset genetic effects in Drosophila. 2003; Rogina & Helfand, 1995; Rogina However, recent evidence from several et al., 1998). Regulation of gene expression areas of biology that are not normally throughout the adult life span, including part of the discourse of demography sug- old age, sets the stage for age-specific gests otherwise. genetic effects. DNA microarrays are pow- Neurodegenerative diseases in human, erful tools for the study of genome-wide including Alzheimer’s, Huntington, and patterns of gene expression in Drosophila Parkinson’s disease, are characterized and other organisms. Microarrays have by late onset of pathology. Because been used to detect genes that vary in Drosophila and humans share many expression levels over the lifetimes of functionally and structurally related flies, and to detect genome-wide transcrip- genes, it is possible to model some of the tional responses to experimental treat- human neurodegenerative pathologies by ments that modify life spans (McCarroll creating lines of flies that carry foreign or et al., 2004; Pletcher et al., 2002). Results artificially modified genes (Driscoll & from microarray studies bolster the view Gerstbrein, 2003; Fortini & Bonini, 2000; that gene expression is regulated through- Mutsuddi & Nambu, 1998). Feany and out the adult life span and is therefore Bender (2000) constructed transgenic likely to be subject to genetic modi- flies carrying normal or mutant forms of fication. Tahoe and colleagues (2005) the human gene for -synuclein, a candi- demonstrated that age-specific patterns of date cause of Parkinson’s disease. All gene expression differ between lines of transgenics exhibited normal neural mor- Drosophila with very different mean life phology and geotactic behavior as young spans, and in some cases, including the adult flies, but beginning at 25 days genes encoding anti-microbial peptides, after eclosion, mutant transgenics devel- the line differences are manifest only in oped Parkinson-like neural morphology old age. Such observations do not prove and a dramatic loss of locomotor ability, that there are genetic differences between whereas nonmutant transgenics escaped lines that alter survival specifically at the morphological and behavioral mani- advanced ages, but the observation of late- festations of disease. Of course, the onset transcriptional differences does ren- primary importance of such research der the existence of such effects more is its potential application to treating likely. As more longitudinal studies of human disease, but the -synuclein case genome-wide transcription levels are P088387-Ch10.qxd 10/31/05 11:31 AM Page 279

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 279

published in the next few years, we can likelihood or LOD score indicating the expect a more complete picture of genome probability that a QTL is present at a function and its variability throughout the particular chromosomal position. A typi- adult life span. cal QTL map has peaks and valleys; genes affecting the quantitative trait are most likely to be located in chromo- D. Mortality QTLs somal regions under the peaks, provided Quantitative trait locus (QTL) mapping is that the peaks exceed some likelihood a set of procedures for identifying approxi- threshold. Curtsinger and Khazaeli (2002) mate chromosomal locations of segregat- extended the usual analysis by mapping ing genes that influence polygenic traits QTLs that affect mortality in each week (Mackay, 2001, 2002; see Chapter 8, this of adult life and then adding a third volume). QTLs affecting mean life span in dimension to the QTL map, indicating Drosophila have been identified in a num- age. An example of a three-dimensional ber of studies (Curtsinger et al., 1998; QTL map is shown in Figure 10.2. There De Luca et al., 2003; Forbes et al., 2004; is a QTL that affects age-specific mortal- Leips & Mackay, 2000, 2002; Luckinbill & ity near the left end of chromosome III; Golenberg, 2002; Khazaeli et al., 2005; the QTL has significant effects on Nuzhdin et al., 2005; Nuzhdin et al., 1997; Pasyukova et al., 2000; Resler et al., 1998; Valenzuela et al., 2004; Vieira et al., 2000). In principle, it is possible to apply the 15

methods of QTL mapping to localize 10 genes that affect age-specific mortality rates rather than just mean life spans. 5 LIKELIHOOD RATIO

However, the requirements are strin- 300 9 gent: not only is there the prerequisite 320 7 for large sample sizes, as in any esti- 340 WEEK mation of mortality rates, but it is also 5 360

necessary that the populations be 3 380

genetically highly defined and contain 400 1 a high density of genetic markers for 300 9 QTL localization. To date this has 320 been accomplished in only two cases. 7 CHROM Curtsinger and Khazaeli (2002) identified 340 5 O QTLs that affect age-specific mortality SO 360 MAL P rates in recombinant inbred populations 3 OSIT 380 of D. melanogaster, finding evidence for ION (cM) several genetically variable chromoso- 400 1 mal regions that influence survival in age-specific fashion. The authors also Figure 10.2 Three-dimensional QTL map of age- specific mortality rates for experimental popula- developed a graphical method for pre- tions of male D. melanogaster (Curtsinger & senting age-specific QTL results, as fol- Khazaeli, 2002). The figure shows the chromosomal lows. QTL mapping results are typically location and ontogenic timing of effects of quanti- presented in two-dimensional graphs: tative trait loci that influence weekly mortality the abscissa represents chromosomal rates throughout the adult life span. The peak near the left telomere of chromosome III indicates position, measured in units of recombi- genetic effects on mortality rates primarily in early nation from the left telomere, while the adult life, with no evidence for significant effects ordinate represents a statistical measure, late in adult life. P088387-Ch10.qxd 10/31/05 11:31 AM Page 280

280 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

mortality in the first few weeks of person who noticed that the Gompertz adult life but has no effect on survival at curve is not applicable to extreme old later ages. ages was Benjamin Gompertz himself One other study of age-specific mortality (Gompertz, 1825, 1872; see review by rates using QTL mapping methods is that Olshansky, 1998). In 1867, William of Nuzhdin and colleagues (2005). QTLs Makeham noted that for humans “the affecting weekly mortality rates in both rapidity of the increase in the death rate sexes were mapped in 144 recombinant decelerated beyond age 75” (p. 346). In inbred lines. Twenty-five statistically sig- 1919, Brownlee wondered whether it is nificant QTLs were found; most had posi- “possible that a kind of Indian summer tively correlated effects on mortality at occurs after the age of 85 years is passed, several different ages, but in two cases the and that conditions improve as regards correlations were negative. Overall, the length of life” (p. 385). Perks (1932) results suggest that the standing genetic observed that “the graduated curve [of variation in survival consists of a mixture mortality] starts to decline in the neigh- of transient deleterious mutations that borhood of age 84” (p. 15). Greenwood tend to increase mortality at younger ages, and Irwin (1939) confirmed that “the and a few mutations with opposing age- increase of mortality rate with age specific effects that are maintained by advances at a slackening rate, that nearly balancing selection. The latter are poten- all, perhaps all, methods of graduation of tially examples of antagonistic pleiotropy, the type of Gompertz’s formula overstate although finer genetic resolution will be senile mortality” (p. 14). They also sug- required to rule out the competing link- gested “the possibility that with advanc- age hypothesis. ing age the rate of mortality asymptotes to a finite value” (p. 14), and made the first estimates for the asymptotic value III. Leveling-Off of Mortality of human mortality plateau (expressed Rates in one-year probability of death, qx). According to their estimates of human In many biological species, including mortality plateaus, “the limiting values Drosophila and humans, death rates of qx are 0.439 for women and 0.544 for increase exponentially with age for much men” (Greenwood & Irwin, 1939, p. 21). of the life span. However, at extreme old In 1960, Science published an article on a ages, a “mortality deceleration” occurs— “General theory of mortality and aging” the pace of mortality growth decelerates that listed some “essential observations from an expected exponential curve. which must be taken into account in any Sometimes this mortality deceleration general theory of mortality.” (Strehler & progresses to the extent that mortality Mildvan, 1960, p. 14). The first of “leveling-off” is observed, leading to a these essential observations was the “mortality plateau.” Thus, at extreme old Gompertz law of mortality, while the ages, a paradoxical situation is observed second essential observation stated that when one of the major manifestations of “the Gomperzian period is followed by a aging—increasing death rate—apparently gradual reduction in their rate of increase fades away or even disappears. of the mortality” (Strehler & Mildvan, The phenomenon of mortality deceler- 1960, p.14). This observation of mortality ation has been known for a long time, deceleration was confirmed for several although its mechanisms were not inten- species, including Drosophila and C. ele- sively studied prior to the 1990s. The first gans (Economos, 1979). The author P088387-Ch10.qxd 10/31/05 11:31 AM Page 281

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 281

concluded “that after a certain species- even for initially identical individuals characteristic age, force of mortality and (Gavrilov & Gavrilova, 1991). Thus, a probability of death cease to increase testable prediction from this theory exponentially with age . . . and remain was that mortality deceleration should constant at a high level on the average for be observed even for genetically identical the remainder of the life span.” (p. 74). individuals kept in strictly controlled The author called these findings “a non- laboratory conditions. Shortly there- Gompertzian paradigm for mortality after, Carey and colleagues (1992) and kinetics” (Economos, 1979, p. 74). A year Curtsinger and colleagues (1992) pub- later, the same author analyzed data for lished back-to-back papers in Science thoroughbred horses (mares), Dall moun- demonstrating mortality plateaus in lab- tain sheep, houseflies, and some other oratory populations of medflies and species and came to a conclusion that Drosophila, respectively. The medfly “Gompertz’s law is only an approxima- study employed genetically heteroge- tion, not valid over a certain terminal neous populations, whereas the compan- part of the lifespan, during which force of ion study in Drosophila used highly mortality levels off.” (Economos, 1980, inbred lines that were essentially devoid p. 317). These findings failed, however, to of within-line genetic heterogeneity. receive attention, and the topic stagnated. The medfly and Drosophila experimental papers generated a flurry of criticisms and responses (Carey et al., 1993; A. Recent Studies of Mortality Plateaus Curtsinger et al., 1994; Gavrilov & Prior to 1990, the most popular explana- Gavrilova, 1993; Graves & Mueller, tion of mortality plateaus was based on 1993, 1994; Kowald & Kirkwood, 1993; the idea of initial population hetero- Nusbaum et al., 1993; Robine & Ritchie, geneity, suggested by British actuary 1993; Vaupel & Carey, 1993). Within a Robert Eric Beard (1911–1983). Beard few years, even the most ardent critics developed a mathematical model in were convinced that mortality plateaus which individuals were assumed to have were real phenomena and not merely exponential increase in their risk of artifacts of contamination or declining death as they age, but their initial risks density in population cages (Khazaeli et differed from individual to individual al., 1995a, 1996). Mortality plateaus were and followed a gamma distribution subsequently documented on very large (Beard, 1959, 1963, 1971). This model scales in a variety of experimental produces a logistic function for mortal- species, including yeast, nematodes, ity kinetics that is very close to the Drosophila, medflies, parasitic wasps, exponential function at younger ages, and humans (see Vaupel et al., 1998, for a but then mortality rates decelerate and review). reach a plateau in old age. This compo- Typical characteristics of a mortal- sitional interpretation of mortality ity plateau in Drosophila are shown in plateaus explained them as an artifact of Figure 10.3 (from Pletcher & Curtsinger, mixture, perhaps reducing their intrin- 1998). In this sample of 122,000 males, sic interest to biologists. age-specific mortality increases in approx- The situation changed in 1991, when it imately exponential fashion from emer- was found that the general theory of sys- gence until 60 days. After 60 days, when tems failure (known as reliability theory) 5 percent of the original cohort remains predicts an inevitable mortality leveling- alive, mortality decelerates and remains off as a result of redundancy exhaustion, fairly constant until 80 days of age. Thus, P088387-Ch10.qxd 10/31/05 11:31 AM Page 282

282 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

0.25 10,000 MALES plateaus; it is difficult to see a plateau in the tail of a survivorship curve, even if N = 121,894 0.20 8000 sample sizes are relatively large. NUMBER ALIVE

0.15 6000 B. Explaining Mortality Plateaus 0.10 4000 Although the existence of mortality 0.05 2000 plateaus is now universally accepted,

DAILY MORTALITY RATE explaining why plateaus exist is contro-

0.00 0 versial. It is convenient to define two 0 20406080100 general, non-exclusive classes of expla- AGE (days) nations: population heterogeneity and Figure 10.3 Age-specific mortality rate and sur- individual aging. Heterogeneity refers to vivorship in a large cohort of male D. melanogaster the idea that individuals in a cohort differ (Pletcher & Curtsinger, 1998). Age-specific mortality increases approximately exponentially until in frailty, which is most conveniently 60 days post-eclosion, and then reaches a plateau parameterized as a multiplicative factor on days 60 to 80. of the Gompertz hazard model. The hazard rate of an individual of age x and frailty Z is for a period of 20 days, or about 20 percent Bx of the maximum life span in this particu- x,z ZAe , lar experiment, there is no trend toward increasing mortality with increasing age. where Z is a gamma-distributed random After 80 days the mortality curve shoots variable with mean 1 and variance 2. up, as the last few survivors die. The Under those circumstances, the mean latter behavior is of no particular signifi- age-specific mortality in the population is cance, and is best understood as an arti- given by the logistic equation. Individual fact of finite sample size, occurring when differences in frailty can be genetic or fewer than 10 flies remain alive. environmental in origin and tend to pro- The turnaround in views about applica- duce mortality deceleration. This occurs bility of the Gompertz model, which had because weaker organisms die first, leav- been revered for well over a century, raises ing preferentially more robust members an obvious question: Why was Gompertz of the population alive for later survival widely accepted until recently, and even measurements. The process of sorting raised to the stature of “Gompertz’ law” weaker and stronger individuals by death despite various exceptions being pointed within a generation is often referred to as out? In addition to science’s predilection “demographic selection,” the first part of for simple laws of nature, the likely expla- the term being necessary to distinguish it nation is that most survival experiments from selection of the Darwinian sort. prior to the 1990s had been too small to Frailty may be fixed at birth, or acquired detect plateaus. Mortality plateaus are and modified through life experience, as late-life phenomena. Small experiments mentioned above. For instance, for the fail to detect them because there are few fixed frailty situation, we might imagine survivors to the age at which mortality that a population of flies contains different rates begin to level off. It is also possible genotypes, each with its characteristic haz- that biologists’ habit of examining sur- ard rate. Or, in a genetically homogeneous vivorship curves rather than mortality population such as an inbred line or F1 rates contributed to ignorance about cross between two inbred lines, differences P088387-Ch10.qxd 10/31/05 11:31 AM Page 283

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 283

in frailty between organisms could arise less exposure to infectious agents and less from micro-environmental effects, such as generation of harmful oxygen radicals. slight uncontrolled spatial variation in For humans, a similar hypothesis was temperature or quality of food experienced proposed by Greenwood and Irwin (1939), at pupation sites. In either case, the essen- who suggested that lower-than-expected tial feature of the fixed frailty models is mortality of centenarians could be that the organisms carry a certain frailty explained by their less risky behavior. factor Z with them throughout their lives. There is a growing body of evolutionary In contrast, flies could acquire different theory that addresses ultimate causes frailty factors during their adult lifetimes of mortality plateaus. The basic prob- as a result of exposure to infectious organ- lem to be solved by theoreticians is isms, or differential rates of reproduction. that evolutionary models of age-specific In either case, the logistic model predicts mortality tend to generate very high mor- the expected population mortality dynam- tality rates (“walls of death”) at post- ics (Yashin et al., 1994), and the magnitude reproductive ages (Charlesworth, 1980; of population variance for frailty has a Curtsinger, 1995a,b; Partridge & Barton, strong influence on mortality dynamics. 1993; Pletcher & Curtsinger, 1998). Gavrilov and Gavrilova (1991, 2001; Imagine a population in which there is see Chapter 1, this volume) developed initially no senescence—that is, the haz- several classes of aging models based ard rate is the same for all age classes. on reliability theory. Interestingly, all Over time, new mutations occur, some of these models predict a mortality deceler- which have age-specific effects on sur- ation, no matter what assumptions are vival. Many of the new mutations are made regarding initial population hetero- deleterious at all ages and are quickly geneity or its complete initial homogene- eliminated from the population by natural ity. Moreover, these reliability models of selection. Some mutations, presumably aging produce mortality plateaus as very few, improve survival of carriers at inevitable outcome for any values of con- early ages, are positively selected, and sidered parameters. The only constraint increase in frequency in the population; is that the elementary steps of the multi- this causes an evolutionary lowering of stage destruction process of a system the population mortality curve at juvenile should occur by chance only, independ- and reproductive ages. Some mutations ent of age. The models also predict that increase or decrease mortality specifically an initially homogeneous population will at post-reproductive ages, but because become highly heterogeneous for risk of post-reproductive survival is irrelevant to death over time (acquired heterogeneity). Darwinian fitness, natural selection does Another class of explanations for mor- not discriminate. The net result is that tality plateaus depends not on differences there is no evolutionary force “pushing between individuals, but on changes down” on the late-life part of the mortal- within individuals as they age. If the ity curve. If the majority of mutations that hazard rates for individual organisms affect old-age survival cause a deteriora- decelerate at older ages, then so, too, will tion of vitality, then post-reproductive sur- the observed population mortality. One vival will erode under mutation pressure, can imagine various biological reasons with nothing to stop it from eventually that individual hazard rates might decel- producing a wall of death. This scenario erate. Older flies might incur less physio- presumes the existence of exclusively late- logical and metabolic cost from mating acting mutations, as originally postulated behavior and reproduction, or lower by Medawar (1952), and is known as activity levels in old age might entail the mutation accumulation model of the P088387-Ch10.qxd 10/31/05 11:31 AM Page 284

284 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

evolution of senescence. The central prob- is, the assumption that mutations have lem for evolutionists trying to understand negatively correlated effects on survival at mortality trajectories is to discover some young and old ages. They argued that such means of counteracting the tendency of models easily explain mortality plateaus, recurrent mutation to drive post-reproduc- but their results have been widely criti- tive hazard rate to infinity. cized. Mueller and Rose (1996) assumed One possibility, not widely considered, that every mutation increases survival is that mutations that affect only the in one randomly chosen age class, and old might improve survival as often as reduces it in another; there are no uncondi- they erode it. This might seem at first tionally deleterious mutations in the glance to be nonbiological, violating the model. Charlesworth and Partridge (1997) widely held view that the vast majority noted that the Mueller-Rose model was not of mutations are deleterious to their iterated to equilibrium, and suggested that carriers. However, reasonable scenarios late-life survival rates would approach zero can be imagined; for instance, a mutation in this model as more evolutionary time that reduces mobility in old age might elapsed. In general, the evolutionary equi- increase survival by causing carriers to librium state is difficult to define in numer- generate fewer damaging oxygen radicals. ical simulations of finite populations. There is some suggestion in the results Pletcher and Curtsinger (1998) argued that of mutation accumulation experiments the Mueller-Rose model includes a strange described above that mutations increase feature that biases the results: there is an survival as often as they decrease it, but assumption that when the population mor- it must be admitted that the distribution tality rate is low, new mutations tend to of mutational effects for old-age-specific increase mortality, but when the mortality mutations is not known in detail. rate is high, new mutations tend to make it Abrams and Ludwig (1995) addressed decrease. The net effect is that mortality the mortality plateau problem in an rates are forced toward an intermediate evolutionary context by analyzing an value. Pletcher and Curtsinger (1998) optimality model in which organisms showed that removing that assumption are presumed to allocate resources to leads to a late-life wall of mortality. The either somatic repair or reproduction. most telling critique is by Wachter (1999), The optimal allocation was presumed to who obtained analytical results for a gener- be that which maximizes lifetime repro- alized class of Mueller-Rose–type models ductive output. Abrams and Ludwig and concluded that mortality plateaus can- (1995) found that an optimal allocation not be accounted for by their equilibrium involves declining investment in repair behavior. Wachter (1999) states unequivo- with increasing age, which, the authors cally that the Mueller-Rose model fails in suggest, could lead to late-life mortality this respect. Thus, it seems likely that the plateaus. However, Charlesworth and simulation of Mueller and Rose (1996) pro- Partridge (1997) re-examined the opti- duced transient mortality plateaus that mality model and found that the death were erroneously interpreted as equilib- rate tends to infinity with increasing rium evolutionary states. age. We also note that the optimality Given strong criticisms of the Mueller- approach does not specifically incorpo- Rose simulation model and analytical rate deleterious mutations with age- invalidation of its results, it is surprising specific effects, an important omission. that Drapeau and colleagues (2000), Rose Mueller and Rose (1996) used numerical and Mueller (2000), and Rose and simulations to study the evolution of mor- colleagues (2002) have continued to pro- tality under antagonistic pleiotropy—that mote it. All three of those papers failed to P088387-Ch10.qxd 10/31/05 11:31 AM Page 285

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 285

cite the analytical results of Wachter Service (2000a) simulated mortality (1999). Mueller and colleagues (2003) dynamics under the assumption of popu- address the various criticisms, including lation heterogeneity in individual age- Wachter’s (1999) analytical results, but specific risk of death. Heterogeneity was the responses are unconvincing (de Grey, modeled by assigning each individual a 2003a, 2004; Service, 2004). Technical unique Gompertz mortality function, details aside, the broader point is that with means and variances of Gompertz Rose, Mueller, and their associates parameters based on the published litera- endorse individual aging over population ture for Drosophila. He found that the heterogeneity as a general explanation for heterogeneity generated by variation in mortality plateaus, a position that could Gompertz parameters was sufficient to ultimately prove to be correct. They refer explain late-life mortality plateaus and to their argument as “the evolutionary could also account for late-life declines theory” (Rose & Mueller, 2000, p. 1,660), in genetic variance of mortality rates. implying that heterogeneity explanations Similar conclusions were reported by are “un-evolutionary” or “anti-evolution- Pletcher and Curtsinger (2000b). ary.” The nomenclature is unfortunate. The reliability models of multistage Phenotypic variability between organ- destruction (Gavrilov & Gavrilova, 1991, isms, including genetically identical ones, 2001) were recently reformulated in is an essential feature of quantitative mathematical terms of a stochastic genetic variability and micro-evolutionary Markov process (Steinsaltz & Evans, change (Falconer & Mackay, 1996). 2004). The authors define a Markov Labeling the argument “evolutionary” is mortality model as a stochastic process, just a rhetorical device, with few con- which is “killed” at random stopping straints on its use: Graves and Mueller times according to the behavior of a (1993, 1994; see also Curtsinger, 1995a,b) Markov process. A general feature of raised the “evolutionary” flag when they such multistage models is that they usu- argued against the existence of mortality ally produce mortality plateaus, as it was plateaus in Drosophila, a stance that was demonstrated earlier with a more simple eventually abandoned. approach (Gavrilov & Gavrilova, 1991, Pletcher and Curtsinger (1998) pre- 2001). As Steinsaltz and Evans (2004) put sented simulation results for the evolu- it, “the mortality rate stops increasing tion of mortality plateaus, focusing on [with increasing age], not because we positive pleiotropy, in which mutations have selected out an exceptional subset exert positively correlated effects on mor- of the population, but because the condi- tality rates at different ages. In these sim- tion of the survivors is reflective of their ulations, positive pleiotropy seemed to being survivors, even though they started produce mortality plateaus, but, as in any out the same as everyone else.” Thus, simulation of finite populations, the defi- the Markov mortality models explain nition of stable evolutionary state is diffi- mortality plateaus by a type of hetero- cult, and the outcomes were probably geneity in acquired frailty because the transient. Charlesworth (2001) used ana- underlying assumptions are similar to lytical techniques to study a similar situ- the earlier reliability models. ation by assuming that all deleterious In evaluating the various theories, it mutations have deleterious effects at is important to remember that the fact reproductive ages. This assumption pre- that a particular mathematical model or vents mutation frequencies from explod- simulation can fit or “predict” an experi- ing at older ages and, thus, preserves mental outcome is not proof that the mortality plateaus. assumptions of the model are correct. For P088387-Ch10.qxd 10/31/05 11:31 AM Page 286

286 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

example, the venerable Hardy-Weinberg at least temporarily, because the more frail model of population genetics predicts individuals will have been eliminated. certain genotypic frequencies, but obser- The latter pattern was observed, and vation of those frequencies in real popu- was interpreted by Khazaeli and col- lations does not validate the underlying leagues (1995b) as evidence for significant assumptions of the model (random levels of heterogeneity. However, the mating, absence of natural selection, authors retracted that result when it was etc.). Theory guides our thinking, but realized that there was a flaw in the inter- critical tests must come from well- pretation (Curtsinger & Khazaeli, 1997). designed experiments, efforts at which The problem is that exposure to an exter- are described in the next section. nal stress does more than kill the more frail flies; it also induces a stress response in the survivors. This phenomenon, C. Testing the Theories known as hormesis, is well documented Designing critical experiments to in a variety of species and involves a rapid address the causes of mortality plateaus genomic response to severe stress. The has proven to be exceptionally difficult; stress response is an interesting phenome- in fact, all experimental tests in this non, but it creates difficult problems in area are flawed in one way or another. the interpretation of the stress experi- Thus, no final answers can be given at ment. In particular, the post-stress decline present, but it is instructive to review in mortality among experimentals com- the relevant experiments and consider pared to controls could be due to reduced the pitfalls. heterogeneity through elimination of The first experiment specifically weaker flies, hormesis induced among sur- designed to test heterogeneity theory used vivors, or both factors. The experimental lethal stress to manipulate the magnitude design of Khazaeli and colleagues (1995b) of population heterogeneity (Khazaeli does not permit separation of the hetero- et al., 1995b) and was inspired by demo- geneity and hormetic effects, and so the graphic studies of human populations after result is inconclusive regarding hetero- a catastrophe (Vaupel et al., 1987). Using a geneity. Recently, the stress experiment single highly inbred line of flies, multiple was redesigned to correct the confounding age-synchronized cohorts were estab- flaw, and data have been collected in the lished. In control populations, flies were Curtsinger lab on 100,000 male flies of maintained under the usual conditions, one inbred genotype. Five intensities of whereas in experimental populations, flies stress were applied, including one suffi- were subjected to 24 hours of desiccation cient to induce a stress response but not at a young age. About 20 percent of the severe enough to cause immediate deaths. flies died during and immediately after the The mild stress will allow estimation of desiccation stress. Post-stress mortality the hormesis effect independent of the rates are informative about population heterogeneity effect, unconfounding the heterogeneity; in particular, in the absence variables. Data analysis by the authors of population heterogeneity, post-stress of this chapter and Dr. A. Khazaeli is mortality in experimental and control underway. populations is expected to be identical. A different and more benign experi- However, if there is significant population mental design was used by Khazaeli and heterogeneity at the time of the stress, colleagues (1998), who attempted to then post-stress mortality in the experi- manipulate population heterogeneity mental populations is expected to drop by fractionating genetically homoge- below that of the control populations, neous populations. Working with two P088387-Ch10.qxd 10/31/05 11:31 AM Page 287

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 287

highly inbred lines, experimental popula- of Drapeau and colleagues (2000), who tions were subjected to the most strin- argued that there is a close connection gent environmental controls possible, far between frailty and sensitivity to environ- beyond what is normally employed in mental stresses in experimental popula- fly husbandry. Eggs were collected over tions of Drosophila. They further a seven-hour period, instead of the usual suggested that, according to heterogeneity 24 hours. First instar larvae were col- theory, populations differing in tolerance lected from that sample for only three to stress should have different late-life hours, and emerging adults were mortality characteristics, though the collected in three-hour windows. The nature of the expected differences was not result of all this careful timing of devel- spelled out. They compared mortality tra- opment is that within a cohort, adult jectories in fly populations that had been flies experienced larval and pupal envi- selected for resistance to starvation with ronmental conditions that are as similar those of unselected controls. No statisti- as possible. The question then is whether cally significant differences were found, the environmentally “homogeneous” which the authors interpreted as evidence populations exhibit mortality plateaus to against the heterogeneity theory. Service a lesser extent than normal environmen- (2000b) questioned the assertion that the tally “heterogeneous” control popula- populations are expected to differ in late- tions. Khazaeli and colleagues (1998) life mortality, noting that for the logistic found that 93 percent of experimental model the plateau occurs at B/2. populations and 100 percent of control Consequently, populations could differ in populations exhibited statistically signif- the intercept parameter A and have the icant mortality deceleration late in life. same levels of late-life mortality. As noted The authors concluded that reducing by Mueller and colleagues (2000) in their environmental heterogeneity during lar- response to Service (2000b), the force of val and pupal stages has negligible effect this criticism is blunted by the generally on adult mortality trajectories. Drapeau accepted theoretical observation that large and colleagues (2000, p. 72) overstated and biologically unrealistic amounts of this experimental result when they variation in the intercept parameter would wrote that “Khazaeli et al. (1998) found be required to produce mortality plateaus, no evidence to support the hypothesis if that were all that varied between indi- that environmental heterogeneity among viduals. Service (2000b) also noted that if individual flies is a primary factor in 2 is lower in the selected population, determining late-life mortality rates.” then it is expected to have higher mortal- The experiment actually gives informa- ity rate than controls (when all other tion only about larval and pupal stages, parameters are fixed), especially at early and is in the strictest sense relevant only ages, as observed. Service concludes that to the “fixed-heterogeneity” model. The the results of Drapeau and colleagues results are not informative about hetero- (2000) are entirely consistent with the pre- geneity acquired in adulthood, which dictions of the heterogeneity model. de may be substantial. Perhaps a broader Grey (2003b) criticized the use of maxi- lesson from this study is that there is a mum likelihood methods by Drapeau substantial and intrinsic environmental and colleagues (2000) and argued that heterogeneity in experimental popula- heterogeneous Gompertz parameters tions that cannot be removed experimen- could explain the experimental results. tally, even by Herculean efforts. Steinsaltz (2005) reanalyzed the experi- The most widely discussed experimen- mental results of Drapeau and colleagues tal test of heterogeneity theory is that (2000) and questioned the claim that P088387-Ch10.qxd 10/31/05 11:31 AM Page 288

288 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

there is no difference in late-life mortality some genotypes have been eliminated by schedules between populations. The origi- selection. However, several factors could nal claim was based on comparisons cause selected populations to be more of means averaged over populations. heterogeneous, both in genetic and envi- Steinsaltz (2005) noted that the data are ronmental variance. If the selection bimodal, and means are therefore mislead- response entails an increase in frequen- ing. He reanalyzed the data and found that cies of initially rare alleles, genetic vari- populations were actually quite different, ance is expected to increase under selec- the mortality plateau being lower in the tion, a prediction that has been verified selected populations. He concluded that experimentally (Curtsinger & Ming, the experimental results lend mild support 1997). This counterintuitive result occurs to the heterogeneity theory, although the because the contribution to total genetic expected differences in timing of the variance by any particular locus depends plateau were not observed. In sum, the cri- on 2pq, where p and q are allelic frequen- tiques of Drapeau and colleagues (2000) cies (Falconer & Mackay, 1996); rare and are varied and instructive, and illustrate common alleles contribute little to popu- some of the difficulties of the experimen- lation genetic variance, but alleles tal task and complexities of the analysis. at intermediate frequencies potentially Rose and colleagues (2002) studied contribute much. The same effect occurs mortality trajectories in populations of if new mutations increase to appreciable Drosophila that had been artificially frequencies during the selection process. selected for long life and compared them Another factor that complicates matters to unselected control populations. is genetic homeostasis. It is well known Mortality trajectories had previously been that homozygous genotypes generally studied in the same populations by exhibit greater environmental variance Service and colleagues (1998), who than heterozygotes (see review by Phelan invoked a heterogeneity explanation. & Austad, 1994). If selection and/or Rose and colleagues (2002) showed that inbreeding increase homozygosity in control populations consistently exhib- selected populations, then the envi- ited earlier onset of mortality plateaus ronmental component of variance is than selected populations. This result expected to increase. In short, there are was interpreted as being consistent with too many unknown variables in geneti- an “evolutionary” (i.e., individual aging) cally uncharacterized outbred popula- model. The result is suggestive, but not tions to allow critical tests of predictions critical; it is not clear that the observa- of heterogeneity models. A better experi- tions are inconsistent with predictions of mental design is that of Miyo and any particular heterogeneity model. In Charlesworth (2004), who studied mortal- general, we consider it very unlikely that ity rates in hybrid progeny of crosses critical tests of heterogeneity and individ- between inbred lines of Drosophila. ual aging models can be executed with In such populations, all individuals are outbred experimental populations. The genetically alike, except for recent problem is that the variance parameter mutations, and heterozygous at loci that plays a central role in the predictions of differ between parental lines. Miyo and heterogeneity models but is generally Charlesworth (2004) found that popula- unknown in either relative or absolute tions of both mated and unmated hybrid terms for outbred, genetically uncharac- males exhibited mortality plateaus, and terized populations. It is widely assumed suggested that their results were consis- that selected populations are less hetero- tent with underlying heterogeneity of geneous than unselected controls because mortality rates. P088387-Ch10.qxd 10/31/05 11:31 AM Page 289

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 289

In the final analysis, evaluating the var- Institutes of Health. We thank Dr. A. Khazaeli ious heterogeneity models is a purely for comments. quantitative question. No reasonable person would deny that there is some het- References erogeneity for frailty within populations, even genetically homogeneous ones; the Abrams, P. A., & Ludwig, D. (1995). question is whether there is sufficient Optimality theory, Gompertz’ law, and the heterogeneity to produce late-life mortal- disposable soma theory of senescence. ity plateaus. We are optimistic that large- Evolution, 49, 1056–1066. scale, multilevel stress experiments and Beard, R. E. (1959). Note on some mathematical mortality models. In G. E. W. other designs using genetically defined Wolstenholme & M. O’Connor (Eds.), The populations will provide the relevant esti- lifespan of animals. Boston: Little, Brown. mates. On the other hand, if the individ- Beard, R. E. (1963). A theory of mortality ual aging theory is correct, then there based on actuarial, biological, and medical must be some important biological considerations. Proceedings of the processes that differ between organisms International Population Conference, at pre- and post-plateau ages and account New York, 1, 611–625. for the change in mortality trajectory. Beard, R. E. (1971). Some aspects of theories of mortality, cause of death analysis, forecasting and stochastic processes. In W. Brass (Ed.), IV. Conclusions Biological aspects of demography (pp. 57–68). London: Taylor & Francis. The integration of biology and demogra- Brownlee, J. (1919). Notes on the biology of a phy proceeded sporadically for most of the life-table. Journal of the Royal Statistical 20th century. Pearl, Sacher, Strehler, and Society, 82, 34–77. Carey, J. R. (2003). Longevity: the biology and others showed the way toward integration demography of life span. Princeton, NJ: of the fields, but their efforts were not Princeton University Press. always widely appreciated. Now we are in Carey, J. R., Curtsinger, J. W., & Vaupel, J. W. a period of widespread dissemination of (1993). Response to letters. Science, 260, demographic techniques among experi- 1567–1569. mental biologists. The new field of biode- Carey, J. R., Liedo, P., Orozco, D., & mography is flourishing and has rich con- Vaupel, J. W. (1992). Slowing of mortality ceptual bases to draw on in demography, rates at older ages in large medfly cohorts. evolutionary biology, reliability theory, Science, 258, 457–461. and even theoretical physics (Pletcher & Charlesworth, B. (1980). Evolution in age- Neuhauser, 2000). Its first major concep- structured populations. New York: Cambridge University Press. tual challenge is to explain mortality Charlesworth, B. (2001). Patterns of age-specific plateaus. We are optimistic that consensus means and genetic variances of mortality will emerge in this area as experimental rates predicted by mutation-accumulation designs and methods of data analysis theory of ageing. Journal of Theoretical become more sophisticated. Other impor- Biology, 210, 47–65. tant challenges include defining the nature Charlesworth, B., & Partridge, L. (1997). of age-specific genetic variation and Ageing: leveling of the grim reaper. Current explaining the high degree of environmen- Biology, 7, R440–R442. tal variation in demographic parameters. Clark, A. G., & Guadalupe, R. N. (1995). Probing the evolution of senescence in Acknowledgements Drosophila melanogaster with P-element tagging. Genetica, 96, 225–234. Research is supported by grants from the Curtsinger, J. W. (1995a). Density and age- National Institute of Aging at the National specific mortality. Genetica, 96, 179–82. P088387-Ch10.qxd 10/31/05 11:31 AM Page 290

290 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

Curtsinger, J. W. (1995b). Density, mortality, E. G., & Mackay, T. F. C. (2003). Dopa- and the narrow view. Genetica, 96, 187–89. decarboxylase affects variation in Drosophila Curtsinger, J. W. (2001). Genetic theories of longevity. Nature Genetics, 34, 429–433. senescence. International encyclopedia of Drapeau, M. D., Gass, E. K., Simison, M. D., social and behavioral sciences. Oxford: Mueller, L. D., & Rose, M. R. (2000). Pergamon Press. Testing the heterogeneity theory of late-life Curtsinger, J. W., & Khazaeli, A. (1997). A mortality plateaus by using cohorts of reconsideration of stress experiments and Drosophila melanogaster. Experimental population heterogeneity. Experimental Gerontology, 35, 71–84. Gerontology, 32, 727–729. Driscoll, M., & Gerstbrein, B. (2003). Dying Curtsinger, J. W., & Khazaeli, A. A. (2002). for a cause: invertebrate genetics takes on Lifespan, QTLs, age-specificity, and human neurodegeneration. Nature Reviews pleiotropy in Drosophila. Mechanisms of Genetics, 4, 181–194. Ageing and Development, 123, 81–93. Economos, A. C. (1979). A non-Gompertzian Curtsinger, J. W., Fukui, H. H., Resler, A. S., paradigm for mortality kinetics of Kelly, K., & Khazaeli, A. A. (1998). Genetic metazoan animals and failure kinetics of analysis of extended lifespan in Drosophila manufactured products. AGE, 2, 74–76. melanogaster. I. RAPD screen for genetic Economos, A. C. (1980). Kinetics of metazoan divergence between selected and control mortality. Journal of Social and Biological lines. Genetica, 104, 21–32. Structures, 3, 317–329. Curtsinger, J. W., Fukui, H., Townsend, D., & Falconer, D. S., & Mackay, T. F. C. (1996). Vaupel, J. W. (1992). Demography of Introduction to quantitative genetics genotypes: failure of the limited life-span (4th ed.). Harlow, UK: Longman Science paradigm in Drosophila melanogaster. and Tech. Feany, M. B., & Bender, W. W. Science, 258, 461–463. (2000). A Drosophila model of Parkinson’s Curtsinger, J. W., & Ming, R. (1997). disease. Nature, 404, 394–398. Non-linear selection response in Filliben, J. J. (2004). Exploratory data analysis. Drosophila: A strategy for testing the In NIST/SEMATECH e-Handbook of rare-alleles model of quantitative genetic Statistical Methods. Available at variability. Genetica, 99, 59–56. http://www.itl.nist.gov/div898/handbook. Curtsinger, J. W., Service, P., & Prout, T. Finch, C. E., & Kirkwood, T. B. L. (2000). (1994). Antagonistic pleiotropy, reversal of Chance, development, and aging. New dominance, and genetic polymorphism. York: Oxford University Press. American Naturalist, 144, 210–228. Forbes, S. N., Valenzuela, R. K., Keim, P., & Cutler, S. J. , & Ederer, F. (1958). Maximum Service, P. M. (2004). Quantitative trait utilization of the life table methodology in loci affecting life span in replicated analyzing survival. Journal of Chronic populations of Drosophila melanogaster. I. Disease, 8, 699–712. Composite interval mapping. Genetics, de Grey, A. D. (2003a). Overzealous 168, 301–311. maximum-likelihood fitting falsely Fortini, M. E., & Bonini, N. M. (2000). Modeling convicts the slope heterogeneity human neurodegenerative diseases in hypothesis. Experimental Gerontology, Drosophila. Trends in Genetics, 16, 161–167. 38, 921–923. Fukui, H. H., Ackert, L., & Curtsinger, J. W. de Grey, A. D. (2003b). Critique of the (1996). Deceleration of age-specific demographic evidence for “late-life non- mortality rates in chromosomal senescence.” Mathematical Society homozygotes and heterozygotes of Transactions, 31, 452–454. Drosophila melanogaster. Experimental de Grey, A. D. (2004). Reply to Mueller’s and Gerontology, 31, 517–531. Rose’s letter to the editor: models on trial: Fukui, H., Pletcher, S., & Curtsinger, J. W. falsifying overstated claims of generality (1995). Selection for increased longevity in does not falsify correctly-stated ones. Drosophila melanogaster: A response to Experimental Gerontology, 39, 453. Baret and Lints. Gerontology, 41, 65–68. De Luca, M., Roshina, N. V., Geiger- Fukui, H, Xiu, L., & Curtsinger, J. W. (1993). Thornsberry, G. L., Lyman, R. F., Pasyukova, Slowing of age-specific mortality rates in P088387-Ch10.qxd 10/31/05 11:31 AM Page 291

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 291

Drosophila melanogaster. Experimental Caenorhabditis elegans and inferences Gerontology, 28, 585–599. about bivariate distributions of mutation Gavrilov, L. A., & Gavrilova, N. S. (1991). effects. Genetics, 156, 143–154. The biology of lifespan: a quantitative Khazaeli, A. A., & Curtsinger, J. W. (2000). approach. New York: Harwood Academic Genetic analysis of extended lifespan in Publishers. Drosophila melanogaster. III. On the Gavrilov, L. A., & Gavrilova, N. S. (1993). Fruit relationship between artificially selected fly aging and mortality. Science, 260, 1565. lines and wild stocks. Genetica, 109, Gavrilov, L. A., & Gavrilova, N. S. (2001). 245–253. The reliability theory of aging and Khazaeli, A. A., Pletcher, S. D., & Curtsinger, longevity. Journal of Theoretical Biology, J. W. (1998). The fractionation experiment: 213, 527–545. reducing heterogeneity to investigate age- Gehan, E. A., & Siddiqui, M. M. (1973). specific mortality in Drosophila. Simple regression methods for survival Mechanisms of Ageing and Development, time studies. Journal of the American 105, 301–317. Statistical Association, 68, 848–856. Khazaeli, A. A., Tatar, M., Pletcher, S., & Gompertz, B. (1825). On the nature of the Curtsinger, J. W. (1997). Heat-induced function expressive of the law of human longevity extension in Drosophila mortality, and on a new mode of melanogaster I. Longevity, mortality, determining the value of life contingencies. and thermotolerance. Journal of Philosophical Transactions of the Royal Gerontology, Biological Sciences, 52A, Society (London), 115, 513–585. B48–B52. Gompertz, B. (1872). On one uniform law of Khazaeli, A. A., Van Voorhies, W., & mortality from birth to extreme old age, Curtsinger, J. W. (2005). Longevity and and on the law of sickness. Journal of the metabolism in Drosophila melanogaster: Institute of Actuaries, 16, 329–344. positive pleiotropy between life span and Graves, J. L., & Mueller, L. D. (1993). age-specific metabolic rate in populations Population density effects on longevity. artificially selected for long life. Genetics, Genetica, 91, 99–109. 69, 231–242. Graves, J. L., & Mueller, L. D. (1994). Khazaeli, A. A., Xiu, L., & Curtsinger, J. W. Population density effects on longevity (1995a). Effect of adult cohort density on revisited: a note in response to ‘Density and age-specific mortality in Drosophila age-specific mortality’ by J. W. Curtsinger. melanogaster. Journal of Gerontology, Genetica, 96, 183–186. Biological Sciences, 50A, 262–269. Greenwood, M., & Irwin, J. O. (1939). The Khazaeli, A. A., Xiu, L. & Curtsinger, J. W. biostatistics of senility. Human Biology, 11, (1995b). Stress experiments as a means of 1–23. investigating age-specific mortality in Hamilton, W. D. (1966). The molding of Drosophila melanogaster. Experimental senescence by natural selection. Journal of Gerontology, 30, 177–184. Theoretical Biology, 12, 12–45. Khazaeli, A. A., Xiu, L., & Curtsinger, J. W. Helfand, S. L., & Rogina, B. (2000). Regulation (1996). Effect of density on age-specific of gene expression during aging. Results and mortality in Drosophila melanogaster: A Problems in Cell Differentiation, 29, 67–80. density supplementation experiment. Helfand, S. L., & Rogina, B. (2003). Molecular Genetica, 98, 21–31. genetics of aging in the fly: is this the Kirkwood, T. B. L. (1999). Time of our lives: beginning of the end? Bioessays, 25, 134–141. the science of human ageing. New York: Keightley, P. D., & Ohnishi, O. (1998). Oxford University Press. EMS-induced polygenic mutation rates for Kowald, A., & Kirkwood, T. B. L. (1993). nine quantitative characters in Drosophila Explaining fruit fly longevity. Science, 260, melanogaster. Genetics, 148, 753–766. 1664–1665. Keightley, P. D., Davies, E. K., Peters, A. D., le Bras, H. (1976). Lois de mortalité et age & Shaw, R. G. (2000). Properties of limité. Population, 31, 655–692. ethylmethane sulfonate-induced mutations Leips, J., & Mackay, T. F. C. (2000). affecting life-history traits in Quantitative trait loci for lifespan in P088387-Ch10.qxd 10/31/05 11:31 AM Page 292

292 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

Drosophila melanogaster: interactions Mueller, L. D., & Rose, M. R. (1996). with genetic background and larval Evolutionary theory predicts late-life density. Genetics, 155, 1773–1788. mortality plateaus. Proceedings of the Leips, J., & Mackay, T. F. C. (2002). The National Academy of Sciences of the USA, complex genetic architecture of Drosophila 93, 15249–15253. life span. Experimental Aging Research, 28, Mueller, L. D., Drapeau, M. D., Adams, C. S., 361–390. Hammerle, C. W., Doyal, K. M., Jazayeri, A. Lin, Y. J, Seroude, L., & Benzer, S. (1998). J., Ly, T., Beguwala, S. A., Mamidi, A. R., & Extended life-span and stress resistance in Rose, M. R. (2003). Statistical tests of the Drosophila mutant methuselah. demographic heterogeneity theories. Science, 282, 943–946. Experimental Gerontology, 38, 373–386. Luckinbill, L. S., & Golenberg, E. M. (2002). Mueller, L. D., Drapeau, M. D., & Rose, M. R. Genes affecting aging: mapping quantitative (2000). Stress resistance, heterogeneity, trait loci in Drosophila melanogaster using and mortality plateaus: response by the amplified fragment length polymorphisms authors. Experimental Gerontology, 35, (AFLPs). Genetica, 114, 147–156. 1089–1091. Mack, P. D., Lester, V. K., & Promislow, D. E. L. Mukai, T. (1964). The genetic structure of (2000). Age-specific effects of novel natural populations of Drosophila mutations in Drosophila melanogaster. II. melanogaster. I. Spontaneous mutation rate Fecundity and male mating ability. Genetica, of polygenes controlling viability. Genetics, 110, 31–41. 50, 1–19. Mackay, T. F. C. (2001). Quantitative trait loci Mutsuddi, M., & Nambu, J. R. (1998). in Drosophila. Nature Reviews, Genetics, Drosophila degenerates for a good cause. 2, 11–20. Current Biology, 8, R809–811. Mackay, T. F. C. (2002). The nature of Nusbaum, T. J., Graves, J. L., Mueller, L. D., quantitative genetic variation for & Rose, M. R. (1993). Fruit fly aging and Drosophila longevity. Mechanisms of mortality. Science, 260, 1567. Ageing and Development, 123, 95–104. Nuzhdin, S. V., Khazaeli, A. A., & Curtsinger, Mair, W., Goymer, P., Pletcher, S. D., & J. W. (2005). Survival analysis of life span Partridge, L. (2003). Demography of dietary QTLs in Drosophila melanogaster. restriction and death in Drosophila. Genetics, 170, 719–731. Science, 301, 1731–1733. Nuzhdin, S. V., Pasyukova, E. G., Dilda, C., & Makeham, W. M. (1867). On the law of Mackay, T. F. C. (1997). Sex-specific mortality. Journal of the Institute of quantitative trait loci affecting longevity in Actuaries, 13, 325–358. Drosophila melanogaster. Proceedings of Martorell, C., Toro, M. A., & Gallego, C. the National Academy of Sciences of the (1998). Spontaneous mutation for life- USA, 94, 9734–9739. history traits in Drosophila melanogaster. Olshansky, S. J. (1998). On the biodemography Genetica, 102/103, 315–324. of aging: a review essay. Population and McCarrol, S. A., Murphy, C. T., Zou, S., Development Review, 24, 381–393. Pletcher, S. D., Chin, C. S., Jan, Y. N., Partridge, L., & Barton, N. H. (1993). Kenyon, C., Bargmann, C. I., & Li, H. Optimality, mutation, and the evolution of (2004). Comparing genomic expression aging. Nature, 362, 305–311. patterns across species identifies shared Pasyukova, E. G., Vieira, C., & Mackay, T. F. C. transcriptional profiles in aging. Nature (2000). Deficiency mapping of quantitative Genetics, 36, 197–204. trait loci affecting longevity in Drosophila Medawar, P. B. (1952). An unsolved problem melanogaster. Genetics, 156, 1129–1146. of biology. London: H. K. Lewis. Pearl, R. (1922). Experimental studies on the Miyo, T., & Charlesworth, B. (2004). Age- duration of life. VI. A comparison of the specific mortality rates of reproducing ad laws of mortality in Drosophila and in non-reproducing males of Drosophila man. American Naturalist, 56, 398–405. melanogaster. Proceedings of the Royal Pearl, R., & Parker, S. (1921). Experimental Society of London, Series B, 271, 2517–2522. studies on the duration of life. I. P088387-Ch10.qxd 10/31/05 11:31 AM Page 293

CHAPTER 10 / Biodemography of Aging and Age-Specific Mortality in Drosophila melanogaster 293

Introductory discussion of the duration of Pletcher, S. D., & Neuhauser, C. (2000). life in Drosophila. American Naturalist, Biological aging: criteria for modeling and a 55, 481–500. new mechanistic model. International Pearl, R., & Parker, S. (1924). Experimental Journal of Modern Physics C, 11, 525–546. studies on the duration of life. IX. New life Ramalu-Hansen, H. (1983). Smoothing counting tables for Drosophila. American Naturalist, process intensities by means of kernel 58, 71–82. functions. Annals of Statistics, 11, 453–466. Perks, W. (1932). On some experiments in the Resler, A. S., Kelly, K., Cantor, G., graduation of mortality statistics. Journal Khazaeli, A. A., Tatar, M., & Curtsinger, J. W. of the Institute of Actuaries, 63, 12–57. (1998). Genetic analysis of extended Phelan, J. P., & Austad, S. N. (1994). Selecting lifespan in Drosophila melanogaster. II. animal models of human aging: inbred Replication of the backcross test and strains often exhibit less biological molecular characterization of the N14 uniformity than F1 hybrids. Journal of locus. Genetica, 104, 33–39. Gerontology, Biological Science, 49, B1–B11. Robine, J. M., & Ritchie, K. (1993). Explaining Pletcher, S. D. (1996). Age-specific mortality fruit fly longevity. Science, 260, 1665. costs of exposure to inbred Drosophila Rogina, B., & Helfand, S. L. (1995). Regulation melanogaster in relation to longevity of gene expression is linked to life span in selection. Experimental Gerontology, 31, adult Drosophila. Genetics, 141, 1043–1048. 605–616. Rogina, B., Reenan, R. A., Nilsen, S. P., & Pletcher, S. D. (2002). Mitigating the tithonus Helfand, S. L. (2000). Extended life-span error: genetic analysis of mortality conferred by cotransporter gene mutations phenotypes. Science Aging Knowledge in Drosophila. Science, 290, 2137–2140. Environment, pe14. Rogina, B., Vaupel, J. W., Partridge, L., & Pletcher, S. D., & Curtsinger, J. W. (1998). Helfand, S. L. (1998). Regulation of gene Mortality plateaus and the evolution of expression is preserved in aging Drosophila senescence: Why are old-age mortality rates melanogaster. Current Biology, 9, 475–478. so low? Evolution, 52, 454–464. Rose, M. R. (1991). Evolutionary biology of Pletcher, S. D., & Curtsinger, J. W. (2000a). Why aging. New York: Oxford University Press. do lifespans differ? Partitioning mean Rose, M. R., & Mueller, L. D. (2000). Ageing longevity differences in terms of age-specific and immortality. Philosophical mortality parameters. Journal of Gerontology, Transactions of the Royal Society of Biological Sciences, 55, B381–B389. London, Series B, 355, 1637–1662. Pletcher, S. D., & Curtsinger, J. W. (2000b). Rose, M. R., Drapeau, M. D., Yazdi, P. G., The influence of environmentally induced Shah, K. H., Moise, D. B., Thakar, R. R., heterogeneity on age-specific genetic Rauser, C. L., and Mueller, L. D. (2002). variance for mortality rates. Genetical Evolution of late-life mortality in Drosophila Research, Cambridge, 75, 321–329. melanogaster. Evolution, 56, 1982–1991. Pletcher, S. D., Houle, D., & Curtsinger, J. W. Sacher, G. A. (1956). On the statistical nature of (1998). Age-specific properties of spontaneous mortality, with special reference to chronic mutations affecting mortality in Drosophila radiation mortality. Radiology, 67, 250–257. melanogaster. Genetics, 148, 287–303. Sacher, G. A. (1966). The Gompertz Pletcher, S. D., Houle, D., & Curtsinger, J. W. transformation in the study of the injury- (1999). The evolution of age-specific mortality relationship: Application to late mortality rates in Drosophila melanogaster: radiation effects and ageing. In P. J. Lindop Genetic divergence among unselected & G. A. Sacher (Eds.), Radiation and ageing strains. Genetics, 153, 813–823. (pp. 411–441). London: Taylor and Francis. Pletcher, S. D., Macdonald, S. J., Marguerie, Semenchenko, G. V., Khazaeli, A. A., R., Certa, U., Stearns, S. C., Goldstein, D. Curtsinger, J. W., & Yashin, A. L. (2004). B., & Partridge, L. (2002). Genome-wide Stress resistance declines with age: analysis transcript profiles in aging and calorically of data from a survival experiment with restricted Drosophila melanogaster. Drosophila melanogaster. Biogerontology, Current Biology, 12, 712–723. 5, 17–30. P088387-Ch10.qxd 10/31/05 11:31 AM Page 294

294 J. W. Curtsinger, N. S. Gavrilova, and L. A. Gavrilov

Service, P. M. (2000a). Heterogeneity in longevity and metabolic rate are not individual mortality risk and its importance inversely correlated in Drosophila for evolutionary studies of senescence. melanogaster. Journal of Applied American Naturalist, 156, 1–13. Physiology, 97, 1915–1922. Service, P. M. (2000b). Stress resistance, Vaupel, J. W., & Carey, J. R. (1993). heterogeneity, and mortality plateaus: a Compositional interpretations of medfly comment on Drapeau et al. Experimental mortality. Science, 260, 1666–1667. Gerontology, 35, 1085–1087. Vaupel, J. W., Carey, J. R., Christiansen, K., Service, P. M. (2004). Demographic Johnson, T. E., Yashin, A. I., Holm, N. V., heterogeneity explains age-specific patterns Iachine, L. A., Khazaeli, A. A., Liedo, P., of genetic variance in mortality rates. Longo, V. D., Yi, Z. Y., Manton, K. G., & Experimental Gerontology, 39, 25–30. Curtsinger, J. W. (1998). Biodemographic Service, P. M., Michieli, C. M., & McGill, K. trajectories of longevity. Science, 280, (1998). Experimental evolution of 855–860. senescence: An analysis using a Vaupel, J. W., Johnson, T. E., & Lithgow, G. J. “heterogeneity” model. Evolution, 52, (1994). Rates of mortality in populations of 1844–1850. Caenorhabditis elegans. Science, 263, Steinsaltz, D. (2005). Reevaluating a test of 668–671. the heterogeneity explanation for mortality Vaupel, J. W., Yashin, A. I., & Manton, K. G. plateaus. Experimental Gerontology, 40, (1987). Debilitation’s aftermath: stochastic 101–113. process models of mortality. Mathematical Steinsaltz, D., & Evans, S. N. (2004). Markov Population Studies, 1, 21–48. mortality models: implications of Vieira, C., Pasyukova, E. G., Zeng, S., quasistationarity and varying initial Hackett, J. B., Lyman, R. F. & distributions. Theoretical Population Mackay, T. F. C. (2000). Genotype- Biology, 65, 319–337. environment interaction for quantitative Strehler, B. L., & Mildvan, A. S. (1960). trait loci affecting lifespan in Drosophila General theory of mortality and aging. melanogaster. Genetics, 154, 213–227. Science, 132, 14–21. Wachter, K. W. (1999). Evolutionary Tahoe, N. M., Lande, J., Khazaeli, A. A., & demographic models for mortality plateaus. Curtsinger, J. W. (2005). Genome-wide Proceedings of the National Academy of analysis of age-specific gene expression in Sciences of the USA, 96, 10544–10547. populations of Drosophila melanogaster Wachter, K. W., & Finch, C. E. (Eds.). (1997). artificially selected for long life. Between Zeus and the salmon. Tahoe, N. M, Mokhtarzhadeh, A., & Washington, DC: National Academy Press. Curtsinger, J. W. (2004). Age-related RNA Williams, G. C. (1957). Pleiotropy, natural decline in adult Drosophila melanogaster. selection, and the evolution of senescence. Journal of Gerontology: Biological Evolution, 11, 398–411. Sciences, 59, B896–901. Yampolsky, L. Y, Pearse, L. E., & Promislow, Valenzuela, R. K., Forbes, S. N., Keim, P., & D. E. L. (2001). Age-specific effects of novel Service, P. M. (2004). Quantitative trait loci mutations in Drosophila melanogaster. I. affecting life span in replicated populations Mortality. Genetica, 110, 11–29. of Drosophila melanogaster. II. Response to Yashin. A., Vaupel, J. W., & Iachine, I. A. selection. Genetics, 168, 313–324. (1994). A duality in aging: the equivalence Van Voorhies, W., Khazaeli, A. A., & of mortality models based on radically Curtsinger, J. W. (2003). Selected different concepts. Mechanisms of Ageing contribution: long-lived Drosophila and Development, 74, 1–14. melanogaster exhibit normal metabolic Zelterman, D., & Curtsinger, J. W. (1995). rates. Journal of Applied Physiology, 95, Survival curves subjected to occasional 2605–2613. insults. Biometrics, 51, 1140–1146. Van Voorhies, W.W., Khazaeli, A. A., & Zelterman, D., Li, C., & Curtsinger, J. W. Curtsinger, J. W. (2004). Testing the “rate of (1994). Piecewise exponential survival curves. living” model: Further evidence that Mathematical Biosciences, 120, 233–250. P088387-Ch11.qxd 10/31/05 11:33 AM Page 295

Chapter 11

Microarray Analysis of Gene Expression Changes in Aging

F. Noel Hudson, Matt Kaeberlein, Nancy Linford, David Pritchard, Richard Beyer, and Peter S. Rabinovitch

I. Introduction these organisms remain as appreciable challenges. Because global analysis of The promise of genome-wide platforms gene expression offers great potential for for biological discovery has been received the elucidation of common mechanisms, by biological scientists with great enthu- pathways, and biomarkers of aging, it is siasm. Of the global discovery tech- safe to predict continued enthusiasm for nologies, the increasing accessibility of application of this technology toward microarrays for analysis of gene expression these goals. has perhaps stirred the greatest interest, certainly within the field of gerontology. II. Technical Issues As this chapter will discuss, recent literature includes applications of these arrays A. Design of Aging Studies to studies of aging in yeast, invertebrates, 1. Biological Aspects of Design rodents, and humans. However, the very nature of the technology—the measure- One of the most important (and often ment of thousands or tens of thousands neglected) considerations in the success- of variables at once—presents new chal- ful design of microarray experiments is lenges for the analysis and interpretation the unique properties of the biological of the data, requiring the development and question being explored. In the case of the application of new statistical and infor- biology of aging, with pleiotropic matics tools. We have begun to see initial and often subtle phenotypes, careful fruition of this work, especially in yeast examination of the precise biological and invertebrate models. Studies in mice features under scrutiny is paramount. and humans are well underway; however, Unfortunately, important considera- the greater inter-individual heterogeneity tions, such as how to define terms like and tissue and genomic complexity of “old” and “premature aging,” are often

295 P088387-Ch11.qxd 10/31/05 11:33 AM Page 296

296 F. N. Hudson et al.

neglected. This section attempts to respect to aging or age-associated pheno- deal with these issues in relationship to types are to be made. Whenever possi- the experimental design of microarray ble, it is strongly recommended that studies. In particular, we will address how multiple timepoints be used rather than experimental design determines the types only “young” and “old.” of gene expression biomarkers obtained. One important consideration in the Several classes of aging-related microarray design of a study comparing young and old experiments are considered, including organisms is how to define the “young” analysis of gene expression at different and “old” populations. For young pop- ages, analysis of gene expression in ulations, the primary criteria should short-and long-lived models, and analysis be organisms that are reproductively of individual longevity. and developmentally mature. In order a. Gene Expression Studies at Different to define “old” populations, the most Age. Perhaps the most common applica- straightforward definition is derived from tion of microarray technology to the study statistical parameters of the life-span dis- of aging is the search for gene expression tribution, such as population median changes that correlate with organismal and maximum life span. For example, one age. Studies of this type typically employ definition of “old” could be individuals a design in which RNA is obtained that have achieved at least 75 percent of from “young” individuals and compared the population maximum life span. The against RNA obtained from “old” individ- potential for degenerative changes present uals. Microarray analysis is then carried in very old animals must be considered, out and a comparison is made between however, as these secondary gene expres- “young” and “old,” with lists of genes pre- sion changes may complicate the observa- sented that either increase or decrease in tion of those more intimately tied to the expression as a function of age. These biology of aging. An alternative definition types of studies have been carried out in of what age constitutes “old” might be all of the model systems commonly used based on the appearance of one or more to study aging, including mammals, flies, phenotypes associated with old age. Such worms, and yeast, as described in subse- a definition, however, is complicated by quent sections. the fact that a majority of aging pheno- In the vast majority of “young” versus types show incomplete penetrance, with “old” studies, only two timepoints have large individual variation in age of onset been used. This two-timepoint design is and severity. On the other hand, an appro- fraught with danger because no informa- priate set of phenotypic markers might tion is gained regarding the kinetics of reflect biological age more accurately gene expression change. In some cases, than does chronological age (see section intermediate age timepoints have been II.A.1.e). In some cases, the age of the collected in addition to “young” and “old” population is determined by the “old.” This type of design has the availability of donor samples. Human advantage that it may be possible to studies in particular are often constrained identify genes that show trends in by sample availability and must make use expression correlated across multiple of tissues or cells obtained from donors of age groups. In addition, it may be possi- a variety of different ages. For microarray ble to identify and classify genes based studies, as with other types of analyses, on the kinetics with which expression there is no clear right answer as to changes occur. This is particularly what the definition of “old” should be. important if any inferences regarding What is clear, however, is that the parame- the causality of observed changes with ters used to define the old population P088387-Ch11.qxd 10/31/05 11:33 AM Page 297

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 297

should be carefully considered and explic- two or more ages. The primary advantage itly presented during both the experi- associated with this type of design is that mental design and data interpretation. In a single experiment can be used to iden- addition, the statistical analyses used tify gene expression changes correlated should reflect the experimental design in with aging, the model of longevity and the this regard. As stated above, a multiple interaction of aging and the model of timepoint design in which samples are longevity. Figure 11.1 shows the possible obtained at several different ages is pre- gene expression changes in a “four-way ferred, as this provides additional informa- design” experiment where young and tion about the kinetics of gene expression old animals are used with and without change with age and a measure of flexibil- a modification that affects longevity ity in choosing appropriate donor ages. (Mod). Biomarkers of aging (light bars in b. Gene Expression Analysis of the Old WT column of Figure 11.1, rows Long-Lived Models. In addition to studies B, D, G, and H), which are unaffected comparing young organisms with aged in old members of the group modified organisms, microarrays can be used to for longevity (Figure 11.1, row B), can compare individuals or populations of sim- be distinguished from biomarkers of ilar age but with different aging potentials. aging, which are attenuated in old mem- For example, many studies have examined bers of the group modified for longevity the gene expression profile of young mice (Figure 11.1, row D). The latter may fed a control diet relative to young mice include age-associated gene expression fed a calorie restricted (CR) diet (see sec- changes that are functionally important tion III.C.2). When age-matched animals for longevity; these are predicted to be are compared across the two dietary regi- attenuated in long-lived animals relative ments, differences reflect gene expression changes associated with caloric intake rather than age. Because CR animals have a longer life span than control-fed animals, observed gene expression changes also have been correlated with increased longevity, at least within the experiment. This distinction is an important one. In “young” versus “old” experiments, the potential exists to discover gene expression changes that correlate with chronological age (biomarkers of age or aging). In experi- Figure 11.1 Possible expression patterns of genes ments carried out on young control versus regulated by aging and a modification associated long-lived individuals, it is possible to with longevity (Mod). Grey bars represent expres- identify gene expression changes that cor- sion changes relative to the black bars; black bars related with longevity (biomarkers of represent genes with unchanged expression relative longevity) and/or the rate of aging. Such to young wildtype (WT) animals. For experimental designs that include intermediate ages and so on, studies may permit the discovery of genes this basic set of possible experimental results can with the potential to extend life span that serve as a guideline for predicting the biological may not be altered in normal aging. implications of potential outcomes of a gene c. Studies of Both Longevity and Age. expression study. This simplified set of results can In several studies, these two types of be expanded to include experimental designs with intermediate ages, the likelihood intermediate gene experimental design have been combined expression patterns, and different interpretations such that microarray analysis is carried depending on the direction of gene expression out on control and long-lived animals at change relative to baseline. P088387-Ch11.qxd 10/31/05 11:33 AM Page 298

298 F. N. Hudson et al.

to controls. The classic example of this models of increased longevity. It should be phenomenon is the observation that noted, however, that none of the “prema- many gene expression changes associated ture aging” models proposed to date reca- with age in control mice are reduced or pitulate all of the phenotypic changes absent in old mice subjected to CR (Cao associated with aging in normal animals, et al., 2001; Dhahbi et al., 2004). Of note, and most have additional pathologies that many of the studies of CR are missing the do not occur during the normal aging young treated (short-term CR) control process. Differentiating gene expression group, and this may complicate the inter- changes associated with accelerated aging pretation of results. In addition, biomark- (if present at all) from those associated ers of longevity in the modified group with non-aging related pathologies is a can be examined in the young animals daunting task. The most effective approach (Figure 11.1, row F), old animals would most likely be a comparative analy- (Figure 11.1, row E), or both (Figure 11.1, sis of gene expression profiles across mul- row C), and effects of the longevity modi- tiple long-and short-lived mutants. In this fication occurring early and late in way, gene expression biomarkers that the aging process can be distinguished. reflect the rate of aging could potentially Furthermore, complex changes associated be identified with higher confidence. with aging and altered by the longevity e. Biomarkers that Predict Individual modification can be identified (Figure Longevity. Another potential use of 11.1, rows G and H). In these cases, bio- microarrays applied to aging research is markers of aging may also be affected in the large-scale identification of gene the young modified group. This might expression biomarkers that predict indi- represent a more complex phenomenon, vidual longevity. Among humans, it is such as a stress induced by the longevity clear that different individuals age at dif- modification that leads to a stress ferent rates, due to both genetic and envi- response similar to that seen in aging. A ronmental factors. For a particular person, related design, typically used for dietary chronological age may not be an accurate modification studies such as caloric predictor of remaining life expectancy. restriction, is a “three-way design” in In fact, several phenotypes have already which young animals from an age prior to been suggested as potential biomarkers of the start of caloric restriction are com- biological age in humans, including body pared to older animals that have been sub- temperature (Roth et al., 2002), serum jected either to caloric restriction or nor- insulin levels, age-related rate of decline mal feeding (Lee et al., 1999). This design in serum dehydroepiandrosterone sulfate is effective for the study of modifications (DHEAS) (Roth et al., 2002), and telomere that are started after development. length in blood cells (Cawthon et al., d. Microarray Studies on Short-Lived 2003). To date, however, it has not been Mutants. Several mutations that result demonstrated that these biomarkers can in shortened life span have been suggested be accurately used to predict survival. as models of premature aging in mammals Microarrays offer the opportunity to (Warner and Sierra, 2003). Microarray detect a group of gene expression bio- analysis of tissues or cells derived from markers that more accurately reflect short-lived mutants offers the opportunity biological age (and hence life expectancy) to identify gene expression changes corre- than currently possible. Hundreds of lated with short life span and, potentially, potential biomarkers can be assayed accelerated aging. In principle, such simultaneously in a single array experi- studies are identical in design to those ment. Appropriate experimental design examining gene expression changes in for the identification of individual P088387-Ch11.qxd 10/31/05 11:33 AM Page 299

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 299

biomarkers of longevity using microar- life expectancy based on these gene expres- rays, however, is nontrivial. sion markers. One approach might be to compare gene From a clinical perspective, the use of expression profiles from a specific tissue microarrays has even greater promise than type using samples from centenarians ver- just identifying candidate genes for poten- sus samples from the general population. tial therapeutic intervention. Microarray A problem with studies of this type, how- analysis to identify biomarkers of aging ever, is the confounding effect of gene and longevity could strongly influence pre- expression changes due to age-associated ventative care and risk management–based disease and degenerative changes that are decisions about screening for diseases of likely to be present in the centenarian the elderly. It is possible that a specialized population simply due to the extreme version of a microarray analysis for poten- chronological age of these individuals. tial biomarkers of aging could be used in One way to get around this complication routine assessment of aging adults. would be to compare samples from age- matched siblings or offspring of centenari- 2. Sources of Variability in Microarray ans versus age-matched individuals from Experiments the general population. Because the probability of achieving extreme longevity is When considering the design of a microar- quantifiably higher in close relatives of ray experiment, the sources of biological centenarians (Perls et al., 2000; Perls et al., and technical variability must be consid- 2002), it might be possible to extract the ered, as they affect the ability to detect genetic component of gene expression true gene expression changes. Microarray changes associated with a predisposition experiments can most clearly detect gene for longevity in this manner. expression changes with low biological Alternatively, a mammalian model sys- variability. However, changes in genes tem, such as mice, could be used to with high biological variability may be identify gene expression biomarkers pre- of interest to the researcher for several dictive of individual longevity. Similar reasons. Some of the sources of biologi- to the human study mentioned above cal and technical variability as well as (Roth et al., 2002), reduced body tempera- methods for addressing them are discussed ture and serum insulin are associated with below. longevity in mice (Weindruch and Walford, a. Cellular Heterogeneity. One impor- 1988). Recent work has also suggested that tant factor to consider in the analysis of body weight and levels of T-cell subsets variability in microarray experiments is and thyroxin can be used to predict indi- the cellular heterogeneity of the samples vidual longevity in animals as young as being analyzed. In most experiments con- 8 months old (Harper et al., 2004). One ducted on multicellular species, a single approach for using microarrays to identify sample is derived from whole animals these types of biomarkers would involve or organs. Therefore, changes in gene analysis of global gene expression in par- expression due to changes in underlying ticular tissue, such as blood, from individ- cell type distribution are indistinguish- ual animals at multiple age points (e.g., able from gene expression changes due to every 6 months). Following the death of all transcriptional events within a cell. This animals in the cohort, it should be possible is a particularly important cautionary to identify gene expression patterns that note when considering aging experiments correlate with individual longevity relative where the time-dependent structural to the population. A computational algo- changes associated with atrophy or hyper- rithm could then be generated to predict trophy are well known and when tissues P088387-Ch11.qxd 10/31/05 11:33 AM Page 300

300 F. N. Hudson et al.

are rarely cleared of blood before RNA due to these effects has not been suffi- preparation. Gene expression differences ciently characterized to date. arising from a population of cells may still b. Temporal Heterogeneity. In addi- provide insight into changes in the function to cellular heterogeneity, effects of tion of an organ or organism with age and temporal differences may also introduce are certainly interesting. However, when variability into a data set. Gene expres- attempting to infer the cause of gene sion differences can be associated with expression changes, it is important to con- time of day or level of activity, and sider that cell type differences may be because a typical gene expression study at the root of the observed differences is a temporal cross-section, it is difficult and that assumptions about alterations to determine whether the differences in intracellular signaling pathways may seen between individuals are due to a be unfounded when whole tissues are true stable heterogeneity or are the result analyzed. of transient gene expression in certain Techniques for measuring gene expres- animals. Care should be taken when sion from single cells in an organism preparing for a microarray experiment are being actively developed in order to to minimize temporal factors that may address this concern. Amplification proce- affect apparent gene expression. dures allow for the labeling of RNA from c. Age-Associated Biological Variability. samples as small as 50 nanograms (ng). There are several potential sources of bio- Most of these amplification procedures are logical variability inherent in the aging based on the methods developed by Van process that can affect the outcome of a Gelder and colleagues (1990). Concerns microarray experiment. For example, age- about bias introduced by amplification associated dysregulation at the level of the appear to differ depending on the microar- genes (e.g., alterations in gene silencing), ray platform. For Affymetrix arrays, a sys- at the level of cellular signaling (e.g., tematic bias introduced by amplification alterations in functionality of signaling has been noted (Wilson et al., 2004). pathways), at the level of tissue (e.g., alter- However, this bias does not affect the ations in cell type composition), or at the genes identified as differentially expressed level of the organism (e.g., alterations in when data is only compared directly circulating hormone levels) may cause an between samples that have undergone the increase in the variability of gene expres- same amplification procedures. For cDNA sion with age. These changes are clearly arrays, Feldman and colleagues (2002) associated with the aging process but may reported that the bias introduced by ampli- not affect the same genes in each individ- fication is negligible. Additionally, there ual to the same extent, thus appearing as are several reports suggesting that amplifi- an increase in the variance of samples cation may reduce the noise inherent in from older individuals. Similarly, discard- rare transcripts for cDNA arrays and pro- ing as uninteresting genes that show duce data that is more likely to be verified significant expression changes in only a by reverse transcriptase polymerase chain subset of individuals may overlook reaction (RT-PCR) (Feldman et al., 2002; biologically important candidates. One Gomes et al., 2003). In addition to poten- positive aspect to the increase in gene tial bias due to amplification, analysis of expression variance with age is that it RNA from single cells may introduce sto- should be possible to obtain information chastic and micro-environmental variabil- regarding the pathways most affected by ity that would normally be averaged out in age-associated transcriptional dysregula- a sample composed of thousands of cells. tion. A thorough study looking at the The magnitude of gene expression differ- gene-specific variation as a function of age ences between neighboring cells in a tissue would be of interest. P088387-Ch11.qxd 10/31/05 11:33 AM Page 301

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 301

d. Sources of Technical Variability. intact). Although measuring bulk degrada- There are multiple sources of variability tion in total RNA does not directly address in measurements of gene expression that mRNA quality, intact total RNA profiles can mask biologically relevant signals are consistently associated with measure- (Parmigiani et al., 2003). These sources ments of RNA quality on the arrays, such of variability can be grouped into two as the similarity of the signal intensity his- main categories: systematic and stochastic togram across samples. Affymetrix arrays, (Huber, 2004). Systematic sources include for example, provide indicators of 3’ to the amount and quality of the labeled 5’ ratio for two genes using probes that RNA in the sample, dye-specific effects, hybridize along the sequence. Additionally, and proper calibration of the instrumen- contaminant-free preparation, proper stor- tation used for array manufacture, age at 70 °C, and minimization of freeze- hybridization, and scanning. Stochastic thaw cycles all contribute to RNA quality. sources include variability in the quality Microarray applications are highly sensi- of the arrays themselves, particularly the tive to RNA quality and require extreme DNA on the arrays, nonspecific hybridiza- attention at this step. tion, stray signals, inherent variability in Another important and correctable labeling and extraction of RNA, and day- source of systematic variability is in specific effects. For spotted arrays, spot- the methods used for RNA labeling. ting efficiency and spot size and shape Direct comparison of array results contribute to stochastic variability, and obtained through different labeling for arrays built in situ, efficiency of incor- protocols should not be attempted, poration of each base is included in sto- although differential expression (ratios) chastic variability. Systematic errors can may be compared. Because even small be greatly reduced by careful methodology deviations in a protocol can produce and appropriate instrument calibration. differences in the resulting signal intensi- However, stochastic errors are inherent to ties, simultaneous labeling of all samples the microarray system being used and are compared in a study is preferred. When handled by an experimental design that is this is not possible, simultaneous labeling balanced across the sources of error and of samples that are balanced across the the use of an appropriate statistical error experimental conditions will minimize model that factors in the multiple sources the bias introduced by labeling. For exper- of variation. iments where two or more samples are The most important and correctable labeled with different dyes and compared source of systematic variability in any directly on the same chip, systematic dif- microarray experiment derives from the ferences in the incorporation of dyes dur- quality of the input RNA. Differences ing RNA labeling can also lead to large in RNA degradation between two sam- artifacts. One method to control for this ples can lead to false positives upon variability is to implement a dye flip con- comparison of the gene expression profiles trol for each sample (see section II.A.3.c). that are indistinguishable from true positives even after statistical analysis. Use of 3. Managing Variability in Microarray a fluidics system such as the Agilent Experiments Bioanalyzer for determination of RNA quality is recommended. This system uses a. Replication and Sample Size nanogram amounts of the total RNA Considerations. Like any other scientific preparation and can quantify degradation experiment, the need for replication in in the extracted mRNA or in the 18s and microarray gene expression studies is 28s rRNA bands of total RNA (a 28s:18s well established (Lee et al., 2000b). ratio of 1.3–2:1 is typically considered Broadly speaking, there are two types of P088387-Ch11.qxd 10/31/05 11:33 AM Page 302

302 F. N. Hudson et al.

replication: biological replicates and tech- technical variability. Even so, estimates of nical replicates. In addition, there are two the technical variability will always main types of technical replicates: inde- under-represent the true variability of the pendent oligonucleotides or cDNA prod- system, which includes large biological ucts representing the same gene present components. in multiple locations on the array, and Because the goal of a microarray experi- the use of multiple RNA samples from ment is to identify differentially expressed the same source to hybridize onto multi- genes, it is important to have enough ple arrays. Only the biological replicates replicates to keep the probability of hav- allow for making some inference about ing false positives as low as possible. the population from which the individu- Having sufficient replicates to examine als are drawn. Technical replicates, on the the probability of random variability other hand, allow for the determination of accounting of “positive” results is an measurement error or “noise.” When intrinsic aspect of algorithms for estimat- measuring technical variability in a plat- ing false discovery rates (see below). There form, it is common to measure the same are several standard approaches for cal- labeled RNA sample on multiple arrays. culating how many replicates are needed However, there is a stochastic component when the variances of differential gene to the RNA extraction and labeling, even expression are known. For example, when systematic variability is minimized, see Cui and Churchill (2004), Lee and and use of independently prepared RNA Whitmore (2002), or Parmigiani and col- samples will incorporate this aspect of leagues (2003). Figure 11.2, from Cui and

Figure 11.2 Power for detecting a two-fold change between two treatments at various combinations of number of mice per treatment (biological replicates) and number of arrays per mouse (technical replicates). Circles, triangles, squares, and diamonds represent 2, 4, 6, and 8 mice per treatment, respectively. Dotted lines represent the same number of array pairs (8 or 12) for each treatment. Significance level is 0.05 after Bonferroni correction. Biological and technical variance components are estimated from Project Normal data http://www.camda.duke.edu/camda02/datasets. The plotted data is derived from Cui and Churchill (2004). P088387-Ch11.qxd 10/31/05 11:33 AM Page 303

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 303

Churchill (2004), shows the power for rate estimate of the population mean detecting a two-fold change for various can be determined when independent combinations of biological and technical pools are used in an experiment with a variations. fixed number of arrays. This is because Sample size considerations are espe- more individuals from the population cially important in aging studies because can be assayed than if an experiment the variability between individuals and used a single individual per array. between tissues can be many times However, this technique is highly sensi- larger than the changes due to aging; tive to outliers because a single outlier that is, individual variability may over- may skew the perceived gene expression whelm changes due to aging alone. In for an entire group. Furthermore, general, biological variability is greater caution must but used when pooling in higher eukaryotes, and greater in out- across animals, organs, or even a hetero- bred, rather than inbred, organisms geneous single organ because small (although it is possible that F1 hybrids changes present in a region-specific may have reduced variability compared manner will be undetectable. It has been to the parental strains; Phelan & Austad, demonstrated that when using a pooling 1994). It has been suggested, for exam- strategy, apportioning individual sample, that a minimum of six individual ples into multiple smaller independent mice are required to reduce sampling pools (a given sample is added to only errors to satisfactory levels in cDNA one pool) can provide useful information array experiments with mice (Cui & on the biological variability, which can- Churchill, 2004) and seven individual not be obtained if all samples are pooled mice to detect a 1.5-fold difference in together and only technical replicates 95 percent of genes at the 0.01 level are performed (Kendziorski et al., 2004). of significance with 90 percent power Technical replicates also do not provide (Han et al., 2004). However, experiments satisfactory estimates of interassay vari- examining expression differences in ance, which are used to determine sta- human tissues show that at least 36 tistical significance and false discovery individuals would be required to obtain rates; this challenge has not yet been similar results in human experiments addressed in pooling experiments. due to the increased variance present in Pooling is a standard strategy when the human population (Han et al., 2004). samples are from very small organisms b. Microarrays on Individuals Versus such as yeast and invertebrates Sample Pools. While increasing the (Drosophila and C. elegans), where number of arrays used in a microar- acquisition of RNA from a single organ- ray experiment increases the statistical ism is difficult and leads to an extremely power for detecting population differ- low yield. This is discussed further in ences, the cost of processing a microar- section II.A.2. ray, along with the need to validate c. Design of Two-Channel Arrays. results by an independent technique, has Array platforms such as the Affymetrix driven researchers to look for methods to GeneChip® use a single fluorescent sig- achieve similar data from a smaller num- nal, and all replicates must be hybridized ber of arrays. to separate chips. Two-channel spotted Pooling samples has been proposed as cDNA microarrays, however, allow a strategy to identify genes displaying investigators to perform direct compar- differences in mean expression between isons of two samples on the same array; groups. Though pooling masks underly- these also pose more of a challenge for ing biological variability, a more accu- determining ideal experimental design. P088387-Ch11.qxd 10/31/05 11:33 AM Page 304

304 F. N. Hudson et al.

For more than two samples, a widely (close to 1 kb), leading to inconsis- used approach is the reference design, tent hybridization. Commercial oligonu- which compares all samples to a com- cleotide microarrays have seen dramatic mon reference. However, as pointed out price reductions over the last two by Kerr and Churchill in a series of years, which makes them more attractive papers, there are more sophisticated for academic research. There are three alternatives, such as the loop design, that main commercial oligonucleotide plat- are more efficient for certain types of forms: Affymetrix GeneChip® microar- experiments, such as those with fewer rays, Agilent Oligo microarrays, and than 10 samples. Because there are differ- Amersham Biosciences (GE Healthcare) ences in the incorporation of the two CodeLink™ Bioarrays. Excellent reviews fluorophores, leading to systematic dye- of the issues involved in both oligonu- specific bias, a dye flip design (replicates cleotide and cDNA microarray platforms balanced across both fluorescent chan- can be found in Li and colleagues (2003) nels) is recommended. The importance of and Parmigiani and colleagues (2003). a good design cannot be overemphasized b. Validation of RNA Expression because it is a major factor in the estima- Levels. In order to be convinced that tion of precision as well as the power to changes in gene expression associated detect differential gene expression. with aging are biologically relevant, val- Reviews of two-channel design issues idation is required. The first step is to can be found in Churchill (2002), Kerr ensure that the RNA levels measured (2003), Kerr and Churchill (2001a), Kerr by the microarray experiment can et al. (2000), Lee et al. (2000b), be validated by an independent tech- Parmigiani et al. (2003), and Yang and nique. Real-time quantitative RT-PCR Speed (2002). and northern blotting are the primary methods used to validate candidate gene expression changes. Quantitative 4. Available Technologies for RT-PCR is considered the “gold stan- Microarrays and Validation dard” due to its sensitivity, repro- a. Microarray Technologies. There are ducibility, and large dynamic range. two main microarray technologies in cur- The development of high throughput rent use: cDNA spotted glass arrays and technologies for quantitative RT-PCR oligonucleotide arrays (both in situ syn- including the Taqman Low Density thesized and spotted) (Holloway et al., Arrays by Applied Biosystems has led to 2002). Each type has its advantages and the possibility of independent valida- disadvantages. Spotted cDNA arrays tion of many candidates from initial usually offer the advantage of lower screening experiments. cost, whereas oligonucleotide arrays have c. Biological Validation. Once can- much higher specificity (Hughes et al., didate genes and pathways have been 2000). There are three main factors identified and validated, it is important underlying these differences: (1) cDNA to determine whether the observed gene products may be recombined or contami- expression changes are biologically nated such that sequences from multiple relevant. This has been accomplished genes may be present in a single spot; (2) to varying degrees. Confirmation of oligonucleotide sequences can be chosen changes at the protein level is cru- to distinguish from among related gene cial, particularly in aging experiments, family members, which cDNA sequences because of changes in RNA and protein frequently do not; and (3) cDNA products stability and changes in translational are often both variable and large in size efficiency with age (Brewer, 2002; P088387-Ch11.qxd 10/31/05 11:33 AM Page 305

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 305

Ekstrom et al., 1980), and thus gene role in the extent of phenotypic vali- expression changes may not correspond dation of candidate longevity-associ- to protein levels. Furthermore, meas- ated genes. As biological validation urement of biochemical activity of any of microarray results from aging candidate genes is also extremely experiments becomes more common in important because of potential alter- C. elegans and other model systems, it ations in post-translational modifi- will be possible to more accurately cation, inactivation, and degradation of assess the gene expression changes that proteins with aging. These caveats sug- represent important aging-related bio- gest that microarray studies are useful chemical pathways conserved through for the generation of hypotheses, but, evolution. particularly in aging, investigators must test whether the changes observed at the RNA level result in functional B. Informatics Approaches to Gene changes before clear biological conclu- Expression Data in Aging sions can be drawn. It is worth noting, The analysis of microarray data involves however, that although biological rele- several sequential and parallel steps, as vance is desirable, it is not a prerequi- shown in Figure 11.3. The first three site for the identification of biomarkers steps have been covered in some detail in of aging or longevity. In order for a gene previous sections. In the following sec- expression biomarker to be useful, all tions, we focus on the pre-processing and that is required is a high degree of analysis phases. correlation and reproducibility. An additional level of biological validation is using the gene candidates 1. Preprocessing: Diagnostics and identified by microarrays to test Normalization hypotheses about the aging process. This is most ideally done through genetic After a set of microarrays has been manipulation of the model organism. hybridized and scanned, the images from For example, genes that are upregulated the scanner need to be preprocessed in long-lived organisms should increase before performing any statistical analy- life span when overexpressed by genetic sis. The preprocessing involves visual manipulation, if they are functionally inspection of the scanner images (often relevant. Likewise, functionally relevant TIFF files), spot quantification, slide genes that are down-regulated in long- diagnostics and quality control, and lived organisms should increase life normalization. span when expression or function is Spot quantification software depends on decreased (e.g., by deletion or RNAi). the type of microarray. For commercial This type of phenotypic validation has oligonucleotide arrays, the manufacturers been carried out with much success offer their own quantification software. in C. elegans and represents the most Amersham Codelink slides require the convincing demonstration of microar- Codelink Expression Analysis program. rays as a tool to study the aging pro- Agilent slides use Agilent’s Feature cess to date (Murphy et al., 2003). Extractor program. Affymetrix GeneChips Particularly in more advanced organ- require using GCOS. For two-color cDNA isms, the timing, tissue specificity, and arrays, there are a variety of open-source quantitative level of upregulation and programs as well as commercial programs downregulation of identified key (please refer to The Institute of Genomic regulatory genes will likely play a Research’s Web site for open-source P088387-Ch11.qxd 10/31/05 11:33 AM Page 306

306 F. N. Hudson et al.

Figure 11.3 Microarray Analysis Workflow (after Dudoit et al., 2002)

offerings at http://www.tigr.org). For of calculations, poor quality slides can be commercial programs, choices include identified and excluded. Axon’s GenePix, BioDiscovery’s Imagene, Normalization, the process of remov- or CISRO’s Spot, to name a few. Most of ing the uninteresting variability within these programs allow visual inspection of the quantified images, also requires the slide images to check for defects or specialized software. There are many damage, and many produce results of programs available that will take a spot similar quality. quantification file as input, and perform After the scanner images have been normalization as well as other statistical quantified, the next main step in prepro- tasks discussed in the next section. The cessing involves some type of diagnostics specialized programs mentioned above and/or quality control. Most of the have these statistical capabilities in spot quantification programs mentioned addition to their image-analysis features. previously also perform quality control However, there are also open-source tasks, such as calculating mean signal programs available that allow for the strength, background thresholds, and custom analysis of data. For example, Bio- control spot statistics. From these types conductor (http://www.bioconductor.org) P088387-Ch11.qxd 10/31/05 11:33 AM Page 307

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 307

allows the input of raw quantifica- (Irizarry et al., 2003a). GC-RMA, which tion data. There are many normalization uses information about the GC content of routines available for any type of the oligonucleotide sequences, is able to microarray platform, from Affymetrix to improve on the RMA algorithm to main- two-channel cDNA arrays. Bioconductor tain precision while improving accuracy can also perform more sophisticated (Wu and Irizarry, 2004). multichip normalizations, such as robust multiarray analysis, RMA, GC- 2. Statistical Methods for Identifying RMA, variance stabilization, VSN, and Differential Gene Expression dChip. Each of these methods takes a different approach to adjusting signal Generally speaking, there are four steps intensities to account for nonspecific in the identification of differentially hybridization, optical noise, and between- expressed genes. First is the choice of array variations. RMA is explained in the appropriate statistical model, which more detail by Irizarry and colleagues is used to calculate the average intensi- (2003b), where it is shown that for ties of gene expression for each gene Affymetrix arrays, subtracting the mis- across replicates and the sample variance matched probes from the perfect match for each gene. Appropriate models will probes results in an exaggerated vari- be suggested by the experimental design ance. Their RMA method does a back- and can include mixed effects models if ground adjustment that ignores the higher-order structure is present in the mismatch probes. GC-RMA is explained data, such as groupings of covariates in more detail by Wu and colleagues (Churchill, 2002). Second is the calcula- (2003), where it is shown that using the tion of the test statistic. If only two GC content of the mismatch probes groups are to be compared, then t-tests improves the background adjustment. are useful. For more groups, some type VSN is explained in more detail by of analysis of variance approach is more Warner and colleagues (2002), where appropriate. See for example Churchill variance stabilizing transformations are (2002), Kerr and Churchill (2001a,b), used to normalize the microarray data. Kerr and colleagues (2000), Lee and col- dChip is explained in more detail by Li leagues (2000b), Parmigiani and col- and colleagues (2003), where a model- leagues (2003), and Yang and Speed based expression analysis is used that (2002). normalizes Affymetrix array data based For data with more structure, such as on an invariant set of genes. Each of balanced and unbalanced data, or when these methods performs well on a vari- the within-group correlation is important ety of performance metrics. For further in grouped data, mixed effects models comparisons, see http://affycomp.bio- are important (see Pinheiro and Bates, stat.jhsph.edu for details about a list of 2000). Several investigators prefer a modi- benchmarks for Affymetrix GeneChip fied version of the standard t statistic expression measures. that uses information borrowed from Preprocessing of microarrays is dis- all the genes on the array to estimate cussed in more detail in publications the individual gene variance (Efron and by Parmigiani (2003) and Yang and Tibshirani, 2002; Smyth, 2004; Storey Speed (2002). For the normalization of and Tibshirani, 2003a). This modified test Affymetrix arrays, RMA has been demon- statistic approach is useful for prevention strated to be more robust than other avail- of calculating spuriously significant able programs for identification of spike- genes. As previously reported (Qin and in RNA samples of known concentrations Kerr, 2004), in the analysis of spike-in P088387-Ch11.qxd 10/31/05 11:33 AM Page 308

308 F. N. Hudson et al.

experiments on two-color cDNA arrays, these issues can be found in Storey and it has been found that t-tests performed Tibshirani (2003a). the worst among test statistics for cor- Software programs that do most or all of rectly identifying the rankings of the the steps described above are now present spike-in genes, whereas modified versions in many commercial analysis programs. of the t test were much more robust. One popular program is SAM (Statistical Furthermore, it was found that for some Analysis of Microarrays; see Storey and data sets, the simple median of the log Tibshirani, 2003a,b). However, identifica- ratio across arrays performed best at cor- tion of differential gene expression is a rectly ranking the spike-in genes (Qin and very active research field. To stay most Kerr, 2004). current with rapid changes in the field, we The calculation of unadjusted p-values also recommend using the microarray sta- is the third step. This involves the calcu- tistical research tool Bioconductor, where lation of the null distribution for the test many new algorithms first appear. statistics and the selection of rejection regions (symmetric or one-sided). The 3. Data Visualization fourth step requires some reasonable approach to controlling the number of The visualization of large data sets in order falsely positive genes. When testing a to discovery underlying patterns is an hypothesis, one can make either a Type I important aspect of microarray analysis. error (calling the gene significant when it The goal is to use a dimension reduction is not, a false positive) or a Type II error algorithm such as Principle Component (calling the gene not significant when it Analysis, Singular Value Decomposition, is, a false negative). However, in microar- or multidimensional scaling to capture ray analysis, multiple hypotheses are the essential variations in the data set in being tested, so it is not clear how best to just two or three dimensions. However, specify the overall error rate. As pointed an inherent danger is present in such a out by Storey and Tibshirani (2003b), reduction of dimensionality, as valuable there are a spectrum of choices. At one information may be lost (Slonim, 2002). end are unadjusted p-values, which result Another important goal in microarray in far too many false positives (if you analysis is the assignment of biologi- have an array with 50,000 genes and a cal samples into groups based on their p-value cutoff of 0. 01, there are possibly expression patterns. This process of 500 genes that are false positives). At the assignment is broken out into two other end is the standard Bonferroni approaches: unsupervised and supervised correction to control the family-wise learning. In unsupervised learning, also error rate, which is far too conservative known as clustering, the groups or classes and results in large numbers of false neg- are discovered from the data, as they are atives. What has been found to be most not known beforehand. In supervised useful in the microarray context is the learning, also known as classification, the False Discovery Rate (FDR), or the posi- groups or classes are already known or are tive False Discovery Rate (pFDR) or q- predefined. The task is then to predict the value. With this approach, the investiga- class of a new set of experiments. tor can control the number of false Classification methods fall into three positives in the number of genes called main groups: class comparison, class significant, rather than controlling the prediction, and class discovery. A discus- number of false positives out of all the sion of statistical issues of classifiers genes present, by examining the distribu- such as types of classifiers, usefulness tion of p values. A detailed discussion of and limitations, classifier accuracy P088387-Ch11.qxd 10/31/05 11:33 AM Page 309

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 309

improvement, and performance of five problem, and, therefore, each is useful in main classifiers (k-nearest neighbor, its own way. naïve Bayes, logitBoost, random forests, Many open-source software packages and support vector machines) is given such as Bioconductor, TIGR MultiExp- by Dudoit and Fridlyand (2003a,b). In eriment Viewer, Eisen Lab’s Cluster Dudoit and Fridlyand (2003a), these clas- (Eisen et al., 1998), as well as commercial sifiers are applied to several real data sets, programs such as Silicon Genetics and their performance is assessed. They GeneSpring®, Iobion’s GeneTraffic®, showed that the simpler methods, such SAS® Microarray, Insightful’s S as k-nearest neighbor and naïve Bayes, ArrayAnalyzer™, Rosetta Resolver®, and were competitive with the more complex VizX Lab’s GeneSifter™, to name a methods and are advisable for the more few, have a wide variety of clustering inexperienced user. However, they also capabilities. showed that for larger data sets, the more complex methods performed best. 4. Gene Ontology Mining, Pathways Clustering methods, which can be Analysis, and Systems Biology more difficult than classification, are used to group genes into clusters with It is now possible to use microarray gene similar behaviors, and usually begin with expression data to identify groups of not knowing how many groupings there genes in common gene ontology cate- are in the data. Methods include hier- gories and thereby uncover biological archical clustering, k-means clustering, processes and pathways. This visualiza- self-organizing maps, and model-based tion and analysis ability requires special- approaches. Often it is useful to filter the ized software, such as the open-source data so as to cluster only a useful subset programs GoMiner and GenMAPP (see of genes. Such filtering can include sim- Dahlquist et al., 2002, and Zeeberg et al., ple nonspecific filtering such as removing 2003). The Gene Ontology (GO) consor- low-intensity genes or more sophisticated tium (http://www.geneontology.org) has Principle Component Analysis that is attempted to construct a standardized used to reduce the high-dimensionality of structure for functional categorization of the data set to generate gene lists that genes. While there is a strong need for account for most the differential changes. such a standardized categorization, the (Reviews of relevant clustering issues, assignments are still changing as more algorithms, and software can be found in research is conducted on gene function. Chipman et al., 2003, Do et al., 2003, and In addition to the open-source software, Sebastiani et al., 2003). When repeated there are also commercial software pro- measurements are taken into account, it grams that allow for the visualization is possible to achieve more accurate and of biological pathways, gene regulation stable clusters (Yeung et al., 2003). Keep networks, and protein–protein interac- in mind that clusters will always be tions, such as Iobion’s Pathway Assist and found by the algorithm, even if there are Ingenuity System’s Pathways Analysis. no true clusters in the underlying data. These programs are based on scans of the Choosing the most appropriate clustering biomedical literature, and this is a change- method will depend on the data, design, able, active area of current research. and goals of the research (Slonim, 2002). At the systems biology level of analy- A useful way to think about clustering sis, there is the open-source software methods is that they are different ways of program Cytoscape, which is used in looking at a data set. Each one gives a conjunction with protein–protein, pro- somewhat different view of a complex tein–DNA, and genetic interactions P088387-Ch11.qxd 10/31/05 11:33 AM Page 310

310 F. N. Hudson et al.

databases for investigating biomolecular (http://www.tigr.org/ software/ tm4/), as interaction networks (see Shannon et al., well as others. 2003, for more details). Commercial systems include Silicon Genetics’ SpringCore™, Iobion’s GeneTraffic®, SAS® Microarray, Insightful’s 5. Expression Databases and S ArrayAnalyzer™, Rosetta Resolver®, Meta-Analyses BioDiscovery’s GeneDirector™ and VizX High-throughput technologies such as Lab’s GeneSifter™, to name a few. DNA microarrays generate enormous Public databases for the storage and amounts of data. This necessitates the retrieval of microarray data are use of data management systems for the available. These include the NIH’s storage and retrieval of experiments. Gene Expression Omnibus GEO at Both open-source programs as well as http://www.ncbi.nlm.nih.gov/geo/ and commercial products are available. the European Bioinformatics Institute’s Most, if not all, of these data manage- ArrayExpress at http://www.ebi.ac.uk/ ment systems also have a data analysis arrayexpress/. These public repositories capability. These data management are becoming more widely used as more and data analysis systems come in scientific journals require microarray client/server form or as standalone, Web- data to be deposited in publicly accessi- based, Windows, Macintosh, or various ble databases. flavors of Unix/Linux. Of particular A new area of opportunity for microar- importance in choosing data manage- ray data analysis is the integration of ment and data analysis systems is the microarray data generated by different overall cost of the software. Open-source research groups on different array plat- systems have low to zero initial cost, forms (Moreau et al., 2003). Furthermore, but maintenance and support usually access to microarray databases also affords require more experienced bioinformatics opportunities for meta-analyses of cross- and/or programming staff. Commercial species comparisons of expression profiles systems usually have a much higher that allows the study of biological initial cost, but that cost usually covers processes and global properties of expres- user training and a support help desk. sion networks (Shah et al., 2004). An Furthermore, commercial vendors can ongoing effort at the NIA DNA Array also customize their products for some Unit is the development of a Web-based additional cost, whereas customization database of biological pathways (http:// of open-source software is a job for bbid.grc.nia.nih.gov), which is used to your programming staff. Commercial relate gene expression studies to complex products tend to be much more user biological processes. In addition, a second friendly, coming equipped with easy-to- project includes a Web-based database use graphical user interfaces. Open- of the genetics of common complex source products, especially ones such as diseases. Bioconductor, have a much larger availability of statistical algorithms for use on complex experimental designs. III. Biological Studies Open-source systems include BASE A. Gene Expression in Yeast (http://base.thep.lu.se/) (Saal et al., 2002), The National Center for Genome The budding yeast Saccharomyces cere- Resources’ GeneX-Lite (http:// visiae has been used as a model system www.ncgr.org/ genex/), The Institute for to investigate two fundamentally differ- Genomic Research’s TM4 suite ent types of cellular aging processes P088387-Ch11.qxd 10/31/05 11:33 AM Page 311

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 311

(Bitterman et al., 2003; Kaeberlein et al., 1. Gene Expression Changes Associated 2001). The study of chronological aging with Replicative Age in Yeast involves maintaining cells in a meta- As in other model systems, microarray bolically active, nondividing state and analysis has been used as a tool to identify monitoring the decrease in viability with gene expression changes associated with time, perhaps akin to the aging of post- old age in yeast. In theory, the design of mitotic cells in mammals (Fabrizio and such an experiment is identical to similar Longo, 2003; MacLean et al., 2001). studies in worms, flies, and mice: RNA is Replicative aging, in contrast, is defined obtained from old organisms and com- by the number of mitotic cycles compared against RNA obtained from young pleted by a mother cell prior to senes- organisms. Attempting to obtain a pure cence (Mortimer and Johnston, 1959). population of replicatively aged yeast, Several genes have been identified that however, presents a unique challenge not determine chronological or replicative present in other models. The median life life span; however, the relationship, if span of a typical lab strain is approxi- any, between these two aging processes mately 25 generations (Jazwinski, 1993). remains unclear. Thus, in order to obtain a single mother Yeast represents an attractive model cell aged to the population median, that for using microarrays to study the aging cell must be physically separated away process. With a doubling time of less from her 225 “progeny” cells. For life- than two hours, yeast is amenable to span analysis, micromanipulation is used both classical genetic as well as high- to separate daughter cells from 40 to throughput approaches. Much of the 50 mother cells per strain; however, it is microarray technology development was not feasible to use micromanipulation carried out in studies of this organism as a method to obtain enough cells for (e.g., by Chu et al., 1998; DeRisi et al., even one microarray experiment without 1997; Eisen et al., 1998; Lashkari et al., extensive, and possibly biased, RNA 1997; Shalon et al., 1996; Spellman amplification. Two technologies are cur- et al., 1998), resulting in a large body of rently available to obtain large populations knowledge regarding appropriate design of aged cells: magnetic sorting and elutria- and an abundance of publicly available tion. In the magnetic sorting procedure, an expression data (Horak and Snyder, unsorted population of cells is treated 2002). In addition, metabolic and protein with biotin then allowed to grow for interaction pathways are relatively well several generations prior to addition characterized compared to other model of streptavidin-coated magnetic beads systems (Barr, 2003). Also, unlike the (Park et al., 2002; Smeal et al., 1996). case in multicellular eukaryotes, the Subsequent magnetic separation allows potential complications arising from tis- enrichment of aged mother cells, which sue heterogeneity and cell specificity specifically retain biotin on their cell do not apply to studies in yeast (see walls. Elutriation is a centrifugal tech- section II.A.2). nique that separates cells based on cell It is worth noting that one significant size (Helmstetter, 1991). Generally, G difference between microarray studies of 0 daughter cells are the smallest cells in a aging in yeast compared to other sys- population, and older cells are the biggest. tems is that almost all have been carried Thus, a population enriched for bigger cells out using spotted cDNA arrays. The rel- by elutriation also tends to be enriched for ative advantages and disadvantages of aged cells (Woldringh et al., 1995). Both this type of platform are discussed in magnetic sorting and elutriation, however, section II.A.4.a. P088387-Ch11.qxd 10/31/05 11:33 AM Page 312

312 F. N. Hudson et al.

suffer from the drawback of significant be elevated in aged cells (Lesur and contamination by daughter cells, a poten- Campbell, 2004). Although the average age tial source of experimental noise that of the “old” cells is a significant improve- must be considered in any gene expression ment over the first study of this type, it is study of “old” versus “young” yeast. still well below the population median. Two studies have described the tran- Further, both studies use an arbitrary scriptome of replicatively aged yeast. two-fold cutoff to identify differentially In one study, microarray analysis was expressed genes and suffer from a lack of carried out on “young” (0–1 generation) rigorous statistical analysis. Future studies or “old” (7–8 generation) wildtype cells of this type should attempt to address obtained by magnetic sorting (Lin et al., both of these weaknesses. The fact that 2001). In addition, young and old cells both studies suggest an age-associated from a 20 percent shorter-lived sip2 shift toward glucose storage at ages below mutant or a 15 percent longer-lived the median life span of the population, snf4 mutant (Ashrafi et al., 2000) were however, is suggestive that major meta- examined. From this analysis, it was con- bolic changes occur early in the yeast life cluded that gluconeogenesis and glucose span. It will be of interest to determine storage increase as cells age, suggesting a whether these changes are retained, or metabolic shift away from glycolysis and perhaps enhanced, at later replicative ages. toward gluconeogenesis. Microanalytic biochemical assays were used to verify 2. Gene Expression Profiles changes in enzyme activity consistent of Long-Lived Yeast Strains with the microarray results. In addition to questions regarding the purity of the In addition to age-associated gene expres- aged cell population, however, the valid- sion studies, microarrays have also been ity of defining 7–8 generation cells as used to examine gene expression changes “old” must be questioned. The median that correlate with extreme longevity in and maximum life spans of the strain yeast. These types of experiments are background used in this study (S288C) simpler to perform than those described are approximately 25 and 50 generations, above. Because the proportion of aged respectively (Kaeberlein et al., 2004). cells present in a logarithmically growing Thus, the use of 7–8–generation mother yeast culture is exceedingly small (less cells to identify gene expression changes than 1 per 2n cells, where n equals replica- associated with aging in yeast is analo- tive age), RNA can be harvested from an gous to using 7-month mice or 20-year- unsorted population of long-lived cells old humans as the aged population for and compared against RNA from an similar studies in these organisms. unsorted population of wildtype cells. In In the second study of this type, elutria- theory, this approach provides an opportu- tion was used to obtain an aged population nity to identify molecular mechanisms in which 75 percent of the cells were at of enhanced longevity in individual long- least 15 generations old and 90 percent of lived mutants. In addition, comparison the cells were more than 8 generations old of multiple long-lived mutants has the (Lesur and Campbell, 2004). Microarray potential to identify gene expression bio- analysis of aged cells relative to young markers of longevity. cells suggested an increase in expression This type of analysis has been success- of enzymes associated with glucose stor- fully performed using genetic models of age and gluconeogenesis, consistent with longevity as well as environmental the previous study (Lin et al., 2001). In models of longevity, such as CR by addition, certain stress- and damage- growth on low glucose. In one study, responsive genes were also reported to microarray analysis was carried out on P088387-Ch11.qxd 10/31/05 11:33 AM Page 313

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 313

two models of CR in yeast: growth on across a subset of genes to place long- low glucose and deletion of the gene cod- lived mutants into genetic pathways, ing for hexokinase, HXK2 (Lin et al., however, is an approach with the poten- 2002). Based on the observed gene expres- tial for broad applicability as additional sion changes, it was suggested that CR of data sets are obtained. yeast cells results in a transcriptional shift from fermentation to respiration. 3. Chronological Aging in Yeast These findings were verified by follow-up experiments showing that oxygen con- To date, the use of microarrays to investi- sumption is elevated by CR. In addition, gate yeast aging has been largely confined based on the overlap between gene to studies of replicative aging. The reasons expression changes observed in cells for this dichotomy are unclear, as chrono- lacking HXK2 and cells grown on low logical aging would seem to present a sys- glucose, 124 putative gene expression tem amenable to microarray technology. biomarkers of CR were reported. It Chronological life span is determined by should be noted, however, that no evi- culturing cells into stationary phase and dence has been presented to suggest that monitoring the percentage of cells that the observed gene expression changes retain viability over time (Fabrizio and play a causal role in CR-mediated life- Longo, 2003; MacLean et al., 2001). span extension. Unlike the case with replicative age, stud- To date, a large-scale comparative ies examining gene expression changes microarray analysis of multiple long-lived associated with increased chronological mutants is lacking. At least two addi- age are trivial in design. In fact, one of the tional studies have compared individual pioneering microarray studies examined gene expression data sets from cells with the gene expression changes associated enhanced life span against the CR data with the yeast diauxic shift, a transition sets described above. In one case, the from logarithmic growth to stationary transcriptional changes associated with phase (DeRisi et al., 1997). A similar high external osmolarity showed signifi- design, but with timepoints spaced over cant overlap with the CR data set several weeks, could be used to identify (Kaeberlein and Guarente, 2002). In the gene expression biomarkers of chronologi- other study, gene expression changes cal age. As with replicative aging, microar- associated with addition of SSD1-V, ray studies comparing chronologically which increases mean replicative life long-lived mutants to wildtype cells could span by approximately 75 percent, also be used to potentially identify gene showed no significant overlap with CR expression biomarkers of chronological (Kaeberlein and Guarente, 2002). These longevity. In this regard, it would be of results were interpreted to suggest that particular interest to determine whether osmotic stress response, but not SSD1-V, there are significant similarities between is likely to promote longevity by a mech- the transcriptional changes associated anism similar to CR. In both of these with enhanced chronological life span and studies, as well as the CR study described those associated with enhanced replicative above, only two data sets for each geno- life span. type were obtained, and an arbitrary two- fold cutoff was used to classify genes 4. Mutation of Orthologous Genes with significantly altered expression. in Yeast Rigorous statistical analysis and additional data sets would improve confi- In addition to directly studying the func- dence in the individual genes reported. tion of genes that regulate chronological The use of gene expression changes or replicative life span, an alternative P088387-Ch11.qxd 10/31/05 11:33 AM Page 314

314 F. N. Hudson et al.

approach is to use microarrays to study B. Gene Expression in Invertebrate the function of yeast orthologs of Models proteins that affect aging in higher A number of studies have employed eukaryotes. For example, increased microarrays to study aging in the expression of the heat shock transcrip- invertebrate model systems Drosophila tion factor Hsf1 has been found to melanogaster and Caenorhabditis elegans. increase life span in worms (Hsu et al., As is the case with yeast and mammalian 2003; Morley and Morimoto, 2004). model systems, two general strategies Although, no chronological or replicative have been employed to identify genes life span phenotype has been reported important for the aging process. In the in response to Hsf1 overexpression in first, microarrays have been used to look yeast, a recent study employed microar- for genes that are differentially expressed ray technology to identify a majority of in aged animals relative to young animals. the direct transcriptional targets of yeast In the second type of study, microarrays Hsf1 (Hahn et al., 2004). Analysis of have been used to identify genes differ- this data to determine which Hsf1 tar- entially expressed in long-lived versus gets have worm orthologs is likely to short-lived animals. The merit of each of provide information regarding potential these approaches is discussed in detail in downstream effectors of the enhanced section II.A.1. longevity conferred by HSF-1 overex- Although there is still substantial dis- pression. This type of approach would agreement as to which specific genes are be particularly amenable to highly con- differentially expressed, and even how served pathways or genes that are likely many genes are differentially expressed, to behave similarly in higher eukaryotes. there is an emerging consensus that certain types of genes show specific changes 5. Conclusions from Yeast Studies in gene expression with age. In particular, several studies have suggested that expres- Microarray analysis has provided valu- sion of stress-response genes is elevated in able insight into certain aspects of the old animals, and that these types of genes aging process in yeast, particularly the are also upregulated in long-lived mutants. identification of global metabolic There is also evidence that metabolic changes associated with calorie restric- genes are downregulated during aging. In tion and replicative age. However, this this section, we review selected microar- technology has not been used to its ray studies to demonstrate these themes. fullest potential. The use of arbitrary fold We also highlight discrepancies between cutoffs and lack of rigorous statistical various studies to demonstrate the limita- analysis, in particular, have been limitations of current technology. tions of studies to date. Future studies should strive to improve in these areas. Given the relative ease with which a 1. Overexpression of Stress Response compendium of gene expression data sets Genes During Aging in Drosophila could be generated for all of the mutants reported to increase yeast life span, this Several studies in Drosophila have should become a priority for future work. reached the conclusion that stress- In this way, it will be possible to rapidly response genes are overexpressed during identify gene expression changes corre- aging (Zou et al., 2000). Zou and col- lated with longevity on a genome-wide leagues used dual-channel cDNA arrays scale, something for which yeast is and probes prepared from pooled male uniquely suited. Drosophila samples of diverse ages to P088387-Ch11.qxd 10/31/05 11:33 AM Page 315

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 315

identify genes differentially expressed protein metabolism, and protein degrada- during aging. The data analysis for this tion were downregulated. Fifty percent of early study was nonstatistical and based age-dependent changes in gene expression on ratio-dependent selection of genes fol- were ameliorated by long-term CR. lowed by hierarchical clustering. From Landis and colleagues (2004) performed this analysis, it was reported that expres- a microarray study with the aim of iden- sion of certain key metabolic genes, tifying gene expression biomarkers of including glucose-3-phosphate dehydro- aging in Drosophila. As with Pletcher and genase and cytochrome c, are decreased colleagues (2002), Landis and colleagues with age. An age-associated upregulation (2004) used Affymetrix arrays coupled of some stress-related genes, includ- with a sophisticated statistical analysis to ing glutathione-s-transferase 1, was also identify differentially expressed genes. observed. To test whether this repre- RNA was obtained from male flies and sented a general response to oxidative pooled at a variety of aging timepoints stress, the gene expression profile of between 10 and 61 days (at the 61-day young animals treated with the super- timepoint, 50 percent of the cohort was oxide-generating drug paraquat was surviving). Using Significance Analysis of obtained and compared to that of aged Microarrays (SAM), 7 percent of genes on flies. Intriguingly, approximately one- the array were differentially expressed at third of genes differentially expressed in one or more timepoints, in good agree- response to the oxidative challenge were ment with the Pletcher study. They also also differentially expressed with normal observed upregulation during aging of aging. stress-response genes, including antioxi- More recently, several other studies have dant genes, antibacterial genes, and some reached similar conclusions. Pletcher and heat shock proteins (hsp 22). In addition, colleagues (2002) used Affymetrix arrays genes coding for enzymes in the purine containing the majority of Drosophila biosynthetic pathway were upregulated, open reading frames to look for differen- whereas protease, proteasome, and meta- tially expressed genes. For this study, RNA bolic genes were downregulated. was collected only from female flies and Landis and colleagues (2004) also analyzed in a pooled fashion. Samples from exposed young flies to 100 percent oxy- both control-fed and calorically restricted gen to induce antioxidant genes. They animals were collected at multiple time observed that there was a 33 percent points. A sophisticated statistical analysis overlap between genes differentially with false discovery rate adjustment was expressed in response to oxygen and used to identify differentially expressed aging. In agreement with the fact that genes. Nine percent of genes on the Drosophila life span scales with the tem- array were differentially expressed at perature at which they are raised, the one or more timepoints during aging. age-associated upregulation of antibacter- Differentially expressed genes were then ial genes also scaled with temperature, mapped against the GO gene functional suggesting that these genes may be good ontology (http://www.geneontology.org/). aging biomarkers in Drosophila. Based on this analysis, stress response, antibacterial, and serine protease inhibitor 2. Overexpression of Stress Response functional categories were upregulated Genes in Long-Lived Invertebrates during aging, and oogenesis genes were downregulated during aging. In calorically The studies described thus far indicate restricted flies, genes involved in the that stress-response genes, such as antioxi- cell cycle, DNA repair, DNA replication, dant genes and antibacterial genes, are P088387-Ch11.qxd 10/31/05 11:33 AM Page 316

316 F. N. Hudson et al.

induced during aging in flies. Kang and study, they collected pooled samples of colleagues (2002) observed that treating sterile worms from a variety of aging flies with the histone deacetylase inhibitor timepoints as well as day 1 DAF-2 versus PBA increased both mean and maxi- DAF-2/DAF-16 worms. These samples mum life span in two strain backgrounds. were hybridized to spotted cDNA arrays They used EST spotted nylon membrane containing the majority of the C. elegans microarrays to identify genes upregu- genome. A nonstatistical analysis plus lated in PBA-treated flies. Genes upreg- hierarchical clustering was used to identify ulated in the long-lived animals included genes that showed differential patterns of stress-response genes such as MnSOD, expression. In particular, genes upregulated glutathione-s-transferase, and several in DAF-2 versus DAF-2/DAF-16 were chaperonins. selected as potential longevity-enhancing Several important studies using genes. Such genes included stress-response C. elegans have also tried to identify genes such as gst-4, sod-3, catalase genes, genes differentially expressed in short- small heat shock proteins, and antibacter- versus long-lived animals (McElwee ial defense genes. Importantly, using RNAi et al., 2003; Murphy et al., 2003). These to inhibit the activity of these genes in studies are based on the observation DAF-2 mutants resulted in a decrease in that an IGF/insulin-like signaling path- life span, suggesting that their upregula- way functions in early adult C. elegans tion is functionally related to enhanced to regulate life span. longevity. Hypomorphic mutations in the DAF-2 These studies also identified the ins-7 insulin-like receptor double life span and gene product as the likely ligand of increase the stress resistance of mutant the IGF/insulin-signaling pathway. ins-7 worms. Mutations in the downstream expression is decreased in the long-lived DAF-16 FOXO transcription factor block DAF-2 mutant but increased in DAF-2/ the effects of DAF-2 mutations. McElwee DAF-16 double mutants. RNAi against and colleagues (2003) used cDNA microar- ins-7 in a DAF-2/DAF-16 background rays, spotted with the majority of C. ele- resulted in an increase in life span. gans genes, to identify genes differentially The study of Murphy and colleagues expressed between young (1 day) DAF-2 marks an important advance in the field, and DAF-2/DAF-16 pooled worm samples. as it demonstrated convincingly that gene In both cases, the worms had additional expression changes observed by microarray mutations to render them sterile. In the can lead to the identification of function- long-lived DAF-2 worms, upregulation of ally relevant regulators of longevity. In mitochondrial superoxide dismutase sod-3, addition, this study enhanced a biological as well as several heat shock proteins, was model for the mechanism by which hor- observed. In addition, substantial overlap monal signals produced in specific cells between the differentially expressed genes can act globally to regulate life span in and genes with DAF-16 binding sites in C. elegans. their promoters was reported. In order to verify the relevance of observed gene 3. Bioinformatic Analysis of expression changes for increased life span Cross-Species Gene Expression in DAF-2 animals, RNAi was used to Changes During Aging knock down the activities of the differentially expressed ins-7 gene, and this length- Collectively, these microarray studies ened life span. argue that, in general, stress-response A similar strategy was employed by genes are upregulated in old animals and Murphy and colleagues (2003). In this that long-lived animals have elevated P088387-Ch11.qxd 10/31/05 11:33 AM Page 317

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 317

expression of stress-response genes. Also, microarray studies have generally pooled in several of these studies, metabolic a large number of whole animals to study genes were noted to be downregulated gene expression at each timepoint. This during aging. In an attempt to synthesize has advantages and disadvantages, as the data from flies and worms, McCarroll described in section II.A.3.b. A major dis- and colleagues (2004) used a bioinformatic advantage of this approach in invertebrate approach to compare genes differentially systems is that changes in gene expres- expressed during aging in both organisms. sion in individual tissues are averaged McCarroll first identified orthologous out over all of the tissues in the body. For gene pairs across the two species and then example, if a gene were strongly differen- used microarrays to analyze gene expres- tially expressed in one small organ but sion during aging in pooled Drosophila not differentially expressed in other tis- heads and in pooled worms at various sues, it is likely that this gene would be ages. Orthologous gene pairs showed a missed in current studies. limited (r .144), but highly statistically Another disadvantage of pooling ani- significant (P 1011), correlation during mals is that gene expression differences in aging. Perhaps more importantly, individual animals are lost. To eliminate McCarroll then mapped the orthologous this problem, Golden and Melov (2004) gene pairs to GO functional categories. analyzed gene expression changes during GO term mapping of age-associated gene aging in individual C. elegans. expression changes in worms and flies also demonstrated a significant overlap in 5. How Many Genes Are Differentially functional categories. Although it is Expressed During Aging in important to note that the majority of Invertebrate Model Systems? differentially expressed aging genes were unique to each species, statistically We have cited evidence that there significant enrichment for 14 GO cate- are broad similarities in the gene expres- gories was observed. This is many more sion changes associated with aging in significant GO categories than would be Drosophila and C. elegans, as well as expected by chance, as permuting the multiple conserved gene expression underlying data and redoing the GO changes among orthologous gene pairs. analysis showed a false positive rate of 1.4 However, there are still substantial / .91 GO categories. Many of these disagreements between studies, even GO categories were involved in general within the same species. For example, metabolism and showed a coordinate there is relatively poor agreement on repression in early adulthood when the which genes are differentially expressed IGF/insulin pathway is known to begin during aging. There is also disagreement regulating life span in C. elegans. This on the magnitude of differential gene study was the first to attempt to demon- expression during aging. For example, strate the existence of broad phylogeneti- using modern Affymetrix arrays and cally conserved similarity in the types of sophisticated statistical analysis, Landis genes showing altered transcription with and colleagues (2004) and Pletcher age as well as specific orthologous gene and colleagues (2002) observed 7 to 9 pairs with similar age-associated profiles. percent of genes as being differentially expressed during aging across multiple timepoints. In contrast, rigorous sta- 4. To Pool or Not to Pool Invertebrates? tistical analysis of the data from Jin Due to the small size of the experimental and colleagues (2001) suggests that the animal, both Drosophila and C. elegans majority of variance in their expression P088387-Ch11.qxd 10/31/05 11:33 AM Page 318

318 F. N. Hudson et al.

data was due first to strain differences, decreases in both synthesis and degrada- then to sex differences, and only last to tion of mRNA and protein. For example, age-specific differences. This observation protein synthesis decreases by as much implies that the magnitude of gene as 60 to 90 percent during aging in expression changes during aging is rela- Drosophila (Arking, 1998). Such large tively modest. Only 1 percent (1.0 percent) changes may be highly significant for the of genes in the Jin study showed age- aging phenotype and yet are not detected dependent statistically significant differ- by microarrays, at least in part because ences in expression, compared to 7 to current microarray studies examine only 9 percent in the other studies. relative, not absolute, changes in gene Some of the differences between the expression. studies can be attributed to the differences in experimental design. However, C. Gene Expression in Rodent Models by plotting P-value versus ratio change for the genes using a volcano plot, Jin and Although studies of yeast, nematodes, colleagues (2001) demonstrated an impor- and fruit flies yield important insight tant aspect of their result: none of the into evolutionarily conserved pathways genes identified as showing a statistically of aging, the rodent model better repre- significant change in gene expression dur- sents aging in a complex mammalian ing aging showed even a two-fold ratio system, and still offers time, space, and change, and many ratio changes were as economic benefits compared to larger little as 1.2-fold. mammals. Additionally, the availability It is important to note that modern of long-lived genetic mouse models (such statistical array analysis tools identify as dwarf mice deficient in growth hor- genes with very modest ratio changes mone signaling), CR, and transgenic and as differentially expressed. Furthermore, allele replacement mice have allowed for many of the studies we have reviewed comparisons of gene expression patterns do not even show the distribution of associated with life-span extension in the observed ratios for their data. This is mammal. unfortunate, as the biological significance Use of a mammalian model, however, of, for example, a 1.2-fold ratio change does present additional concerns. Unlike in gene expression is presently unclear. studies of lower eukaryotes, mammalian Furthermore, many array data normal- studies of gene expression changes with ization procedures can cause small con- age are intrinsically linked to tissue type, sistent shifts in the ratio data. This could as gene expression variability between tis- lead to affected genes being falsely identi- sues overwhelms the changes observed fied as differentially expressed. Thus, the with aging. For most laboratories, it is not significance of statistically significant but economically feasible to pursue microar- tiny changes in gene expression must be ray studies on all tissues, or even several questioned. tissues of interest, and, thus, most of the It can also be asked whether microar- reported studies focus on transcriptional rays are missing key aspects of changes profiles in one or two tissues that may or in gene function during aging in both may not be representative of global gene D. melanogaster and C. elegans. As expression changes. Tissues examined to noted above, in most studies, relatively date include brain (Blalock et al., 2003; small gene expression changes have been Lee et al., 2000a; Preisser et al., 2004; observed in aging. However, biosynthetic Prolla, 2002; Prolla and Mattson, 2001; activity in both invertebrate and verte- Weindruch and Prolla, 2002; Weindruch brate models plunges during aging, with et al., 2002), skeletal muscle (Lee P088387-Ch11.qxd 10/31/05 11:33 AM Page 319

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 319

et al., 1999; Tollet-Egnell et al., 2004; in a series of papers (Lee et al., 1999; Lee Weindruch et al., 2001, 2002; Welle et al., et al., 2000a; Weindruch et al., 2001, 2001), liver (Cao et al., 2001; Dozmorov 2002; Prolla, 2002). (The CR component et al., 2001; Dozmorov et al., 2002; of these studies will be discussed later.) Meydani et al., 1998; Miller et al., 2002; For these studies, the authors compared Tollet-Egnell et al., 2004; Tsuchiya et al., three animals per group using 6347-gene 2004), heart (Bronikowski et al., 2003; oligonucleotide arrays. The data were ana- Csiszar et al., 2003; Edwards et al., 2003, lyzed using pairwise comparisons (nine 2004; Lee et al., 2002; Meydani et al., total comparisons) with Pearson correla- 1998), kidney (Preisser et al., 2004), tion coefficients calculated for individual duodenum and colon (Lee et al., 2001), animals and a fold-change cutoff was used adipose tissue (Higami et al., 2004; Tollet- to classify genes as significantly upregu- Egnell et al., 2004), and submandibular lated or downregulated. gland (Hiratsuka et al., 2002). The principal finding of these studies The current body of literature is that only a small percentage of genes employing microarrays to examine in each tissue (approximately 1 percent) gene expression associated with aging were upregulated or downregulated by in mammals is relatively young and at least two-fold (1.7 fold in the neocor- largely descriptive. However, Helmberg tex), indicating that aging is unlikely (2001) and Weindruch and colleagues to result from widespread gene expres- (2002) discuss the enormous potential sion changes of large magnitude. utility of this approach. The possible Furthermore, the authors observed little outcomes from this work include: overlap among the individual genes (1) insights into the fundamental causes altered in the tissues examined, of aging by, for example, the identifica- although they did observe coordinate tion of biochemical pathways altered induction of complement cascade mem- with age; (2) development of tools and bers and cathepsins in the neocortex biomarkers useful in the evaluation and cerebellum with age. Cathepsins of aging interventions; and (3) perhaps may be of particular interest as they are even to define “individual genomic risk involved in the processing of amyloid constellations” useful in the treatment precursor protein and are upregulated and management of aging-associated in Alzheimer’s diseased brains. Of the conditions. This review highlights genes reported to be upregulated with some of the seminal papers to date age in the tissues examined, the great- that employ microarray technology to est proportion fell into the functional advance our understanding of aging in categories of inflammation and stress mammals. response, whereas decreased expression was observed for genes involved in metabolism and biosynthesis. 1. Gene Expression Changes Associated Although these early studies are some- with Age what limited by their lack of rigorous Lee and associates were among the first to statistical analyses and validation, they report changes in gene expression profiles were among the first to illuminate trends with age in mouse tissues, using microar- that seem to be recurrent in subsequent rays to study gene expression patterns microarray studies—namely, that rela- in brain (neocortex and cerebellum) and tively few transcriptional changes are skeletal muscle from adult (5-month) of great magnitude, and that transcrip- and old (30-month) C57Bl/6NHsd mice tional profiles across tissues and even (average life span 30 months), as described across subregions of a particular tissue P088387-Ch11.qxd 10/31/05 11:33 AM Page 320

320 F. N. Hudson et al.

are markedly different. Furthermore, the 2004). These authors measured expression initial findings in these studies, that levels of 9,977 genes using high-density genes involved in stress response and oligonucleotide arrays and found an inflammation seem to increase with age age-dependent decrease in the cardiac while genes involved in protein turnover transcriptional response to the paraquat and structural maintenance decline, are challenge, particularly a decrease in the also echoed in later studies. stress-response pathways signaling through Other studies have also analyzed MAPKKK and JNK, and a decreased induc- age-associated gene expression changes tion of the DNA damage-induced gene (young versus old) in conjunction with GADD45. The authors also note a shift in gene expression changes associated the spectrum of oxidative stress genes with enhanced longevity. For example, induced at different ages, with induction of Dozmorov and colleagues (2001) meas- glutathione-S-transferase A3 specific to ured gene expression changes with age in young mice, glutathione peroxidase 1 and livers of 5-, 13-, and 22-month-old mice peroxiredoxin 4 specific to middle-aged (n 3 to 4 mice per group). This study mice, and superoxide dismutase 1 specific utilized both control and Ames dw/dw to old mice. dwarf mice, but presently this discussion A recent study by Blalock and col- will focus on the analysis of the subset of leagues (2003) has attempted to link genes altered with age. The authors age-dependent expression changes with measured gene expression levels using measurable functional consequences. Atlas 588-gene cDNA membranes, 323 They describe age-dependent transcrip- genes of which were removed from tional profiles associated with cog- analysis due to low levels of expression nitive impairment in the rat hippocam- or proximity on the array to highly pus, in particular the hippocampal CA1 expressed genes. Although the authors region. The authors trained young did observe large changes in the expres- (4-month), middle-aged (14-month), and sion ratios of many of the remaining 265 old (24-month) male rats on two memory genes, a closer examination revealed that tasks, the Morris spatial water maze and high fold-change values were correlated the object memory task, and the hip- with high variability and were likely pocampal CA1 region of each animal to be false positives. Thus, of the 265 was subsequently harvested for expres- remaining genes, the authors report only sion analysis on individual Affymetrix four genes altered in the wildtype mice oligonucleotide arrays (one chip per ani- between 5 and 22 months, and only three mal, N 10 animals per group). The genes altered between 5 and 13 months. expression data were analyzed for both Two of these seven genes, however, were aging effects (ANOVA) and for cognition directly involved in insulin signaling, effects (Pearson’s test), and although IGFBP1 and IGF receptor 2, notable in most of the gene expression changes particular due to the repeated implica- were initially evident in the middle-aged tion of insulin signaling in longevity group, impaired cognition was not studies of lower invertebrates. clearly manifest until late life. The Edwards and colleagues compared car- aging- and cognition-related genes identi- diac gene expression profiles in young fied represent familiar categories such as (5-month), middle-aged (15-month) and oxidative stress, inflammation, decreased old (25-month) mice at 0, 1, 3, 5, and mitochondrial function, and altered pro- 7 hours after a single intraperitoneal tein processing, but also include genes injection of paraquat (N 3 for each age involved in downregulated early response and timepoint) (Edwards et al., 2003, signaling, cholesterol synthesis, lipid P088387-Ch11.qxd 10/31/05 11:33 AM Page 321

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 321

and monoamine metabolism, and other of the age-related gene expression likely brain-specific categories such as changes observed in 30-month-old mouse activity-regulated synaptogenesis, upreg- hearts (Lee et al., 2002). Cao and col- ulated myelin turnover, and structural leagues employed a long-lived F1 hybrid reorganization genes. This study clearly strain of mice, comparing liver gene benefits from the statistical power of a expression profiles of young (7-month) relatively large number of samples versus old (27-month) animals fed (N 10 for each group), but more impor- ad libidum or on a CR diet, and also tantly, the inclusion of an intermediate included a short-term CR cohort consist- timepoint and the correlation with a ing of 34-month-old control mice placed functional outcome in the study design on a CR regimen for 4 weeks (Cao et al., allow the authors to put forth an integra- 2001). Gene expression changes observed tive model of brain aging in which gene with age were consistent with previous expression changes observed in early studies, reflecting increased inflamma- adulthood trigger subtle changes result- tion, stress, and fibrosis with reduced ing in cumulative cognitive deficits not expression of genes involved in apopto- evident until a much later date. sis, xenobiotic metabolism, and DNA replication and cell-cycle. However, the observed changes in gene expression 2. Attenuation of Age-Related were attenuated by both long-term CR Expression Changes by Caloric and 4-week short-term CR in old mice. Restriction in Rodents These results imply that CR begun late The early studies by Lee and coworkers in life and instituted for a short time can comparing genomic expression profiles shift gene expression patterns for a sub- in young and old brain (Lee et al., 1999; set of genes to a more youthful profile. Prolla, 2002) and skeletal muscle (Lee However, it may be that the short-term et al., 1999; Weindruch et al., 2001) effects of CR on gene expression are tissue also examined the effect of CR on the and/or age-dependent. Higami and col- expression profiles of tissues from old leagues (2004) investigated the influences (30-month) animals. In general, the of short-term and long-term CR on gene authors noted that CR (initiated at expression in white adipose tissue and 2 months) selectively prevented many of report gene expression changes associated the age-related increases in inflamma- only with long-term CR. The authors tory and stress-response genes while compared four groups of 10- to 11-month- having little effect on expression of old male C57Bl6 mice (N 5 per group): genes involved in neuronal growth and non-fasted controls, fasted for 18 hours plasticity in the neocortex and cerebel- before death, short-term caloric restriction lum. In skeletal muscle, CR shifted for 23 days, or long-term caloric restriction the expression profile toward increased for 9 months. For this study, the authors energy metabolism, increased biosyn- employed high-density oligonucleotide thesis, and increased protein turnover, arrays with over 11,000 genes. Compared more similar to that of younger animals. to the control mice, only a few transcripts The reversal of age-dependent expres- were differentially expressed in the fasted sion changes by CR seems to occur and short-term CR groups, whereas 345 whether the CR is of short or long dura- transcripts were found to be significantly tion. The same group (Lee et al., 2002) altered by long-term CR, the majority of reported in a separate paper that CR which were directly involved in metabo- initiated at middle-age (14 months) lism or insulin signaling. The discrepancy resulted in a 19 percent global inhibition between results of the Higami study and P088387-Ch11.qxd 10/31/05 11:33 AM Page 322

322 F. N. Hudson et al.

the Cao study likely reflects the different in mice given dietary supplementation tissues investigated as well as the timing with alpha-lipoic acid (LA) or coenzyme of the short-term CR—in the Cao study, Q(10) (CQ) started at middle age the short-term CR was initiated in much (14 months) has been examined and com- older mice than in the Higami study pared to CR as a positive control (Lee (34 months versus 9 months). et al., 2004). In contrast to CR, supplemen- Indeed, Dhahbi and colleagues (2004) tation with LA or CQ had no impact on demonstrate that CR begun relatively late longevity or on the spectrum of observed in life (at 19 months) begins to increase tumors when compared with control mice the mean time to death as well as mean on an isocaloric diet. Global analysis of and maximum life spans within 2 months 9977 genes demonstrated that LA, CQ, and of initiation and is accompanied by a rapid CR mitigated age-dependent gene expres- and progressive shift toward a hepatic sion changes related to cellular and extra- transcriptional profile associated with cellular structure and protein turnover, but long-term CR. (Other studies, however, CR was the only intervention to affect have failed to demonstrate a beneficial gene expression related to energy metabo- effect of late-onset CR, for example, lism. The authors conclude that, although Lipman et al., 1995, 1998). For the Dhahbi supplementation with alpha-lipoic acid or study, the authors used high-density coenzyme Q(10) induces a gene expression oligonucleotide arrays to compare hepatic profile indicative of reduced cardiac oxida- gene expression in control mice with that tive stress, CR is much more effective at of mice on a CR regimen for 2, 4, or inhibiting the aging process in the heart, 8 weeks (N 3 to 4 animals per group). likely due to the observed changes in Additionally, a cohort of long-term CR energy metabolism. mice was returned to a control diet for 8 weeks. Microarray analysis of livers 3. Gene Expression Changes in from 19-month-old control mice switched Long-Lived Dwarf Mouse Models to CR for 2, 4, or 8 weeks revealed a pattern of early and sustained gene expression Genetic mutations in the growth hormone changes in response to CR, with early signaling pathway and targeted disrup- gene expression changes occurring within tion of the growth hormone receptor in 2 weeks, intermediate gene expression the mouse have given rise to various dwarf changes occurring between 4 and 8 weeks, mice that display remarkable increases in and late gene expression changes occur- life span. As the primary effector of ring after 8 weeks of CR. Furthermore, growth hormone signaling is insulin analysis of the hepatic gene expression growth factor 1 (IGF-1), there is consider- profile of mice shifted from long-term able interest in whether the longevity CR to a control diet demonstrated that observed in these mice results from a 90 percent of the gene expression effects of mechanism fundamentally similar to that long-term CR were reversed within of CR. If true, similar gene expression 8 weeks. Thus, the authors’ findings imply patterns associated with CR and dwarf a temporal and phenotypic link between models may reveal a transcriptional profile CR-induced longevity and genomic of longevity. expression changes, but these changes Miller and colleagues (2002) examined may be limited to late onset CR. the influence of CR on gene expression in The impact of other late-onset dietary the livers of both 9-month-old control interventions on global gene expression and growth-hormone receptor knockout has also been evaluated. In particular, the mice (GHR-KO) (n 8 per group) using effects on cardiac gene expression profiles 2352-gene cDNA arrays. In this analysis, P088387-Ch11.qxd 10/31/05 11:33 AM Page 323

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 323

CR was reported to significantly alter studies suggest that these observed gene mRNA levels for 352 genes. Although the expression changes are modulated to GHR-KO genotype had little impact on some extent by interventions that retard gene expression in control-fed animals, aging, such as CR and/or dwarf pheno- the gene expression changes associated type, and thus may provide insight into with CR were significantly diminished in the molecular mechanisms associated the GHR-KO mice, pointing toward an with aging. Furthermore, that the gene interaction between the GHR-KO geno- expression changes are similar in nature type and CR. The genes with altered to those observed in nonmammalian expression patterns in this study were models potentially reflects a universal compared to those identified in a study aging process at the transcriptional level, of another long-lived dwarf mouse and the availability of long-lived rodent model, the Snell dwarf, and expression of models will prove essential for eluci- 29 genes was found to be similarly altered dating these mechanisms in complex in both studies. These findings lend mammalian systems. strength to the hypothesis that various models of increased longevity share similar underlying mechanisms. D. Primates As previously mentioned, Dozmorov 1. Primate Literature—In Vivo and colleagues (2001) conducted an aging study of hepatic gene expression in con- To date there are few microarray studies trol and Ames dwarf mice, comparing 5-, reporting gene expression changes with 13-, and 22-month-old animals using the age in primates, particularly in vivo 588-gene Atlas cDNA arrays . Although studies. This undoubtedly is due to the very few genes (7) survived the statistical relatively recent advent of this technol- significance testing to demonstrate an ogy coupled with the time and logistical age-dependent change, of these, none requirements in obtaining primate or were found to be attenuated by the Ames human tissue samples for analysis, espe- dwarf genotype between 5 and 13 months cially for longitudinal aging studies. or between 13 and 22 months of age. There are, however, several studies in A separate study by the same group primates of note, with more recent stud- investigated hepatic gene expression in ies reflecting an increased sophistication 6-month-old Snell dwarf mice using in the use of microarray profiling to the Atlas 2352-gene cDNA array set derive and test hypotheses in the study (Dozmorov et al., 2002). From this analy- of aging. sis, the authors report several gene An early attempt to detect age-related expression changes associated with the gene expression changes in human dwarf genotype; however, it remains to be samples made use of a previously exist- seen which of these early-age (6-month) ing database of gene expression profiles changes are functionally associated with from colon adenocarcinoma and normal longevity. colon samples (Kirschner et al., 2002). In summary, microarray analyses of Although the original data were not part gene expression changes in aging rodent of an aging study, the samples (n 16) tissues have revealed a trend of increased were derived from donors ranging in expression of genes involved in inflam- age from 35 to 85 years. The samples mation and stress response with age with had been hybridized onto Affymetrix a corresponding decrease in expression high-density oligonucleotide arrays, of genes involved in protein turnover with approximately 6,800 genes repre- and structural maintenance. Several sented, and initial data assessment was P088387-Ch11.qxd 10/31/05 11:33 AM Page 324

324 F. N. Hudson et al.

conducted using Affymetrix software and CR . In a comparison of young (7- to before loading into the public database. 11-year-old) and old (25- to 27-year-old) To determine the effect of age—if any— monkeys (N 3 each group), aging on gene expression, the authors calcu- resulted in a selective induction of tran- lated a correlation coefficient between scripts involved in inflammation, stress mRNA expression and age of donor for response and neuronal factors, with a each gene, the significance of which was downregulation of genes involved in computed using a t-test. In this manner, mitochondrial electron transport and only nine genes were found to be signifi- oxidative phosphorylation. The CR com- cantly altered with age in the normal tis- ponent of the study examined gene sue samples, of which one-third were expression in middle-aged monkeys (age likely to be false positives as determined 19 to 21 years, N 3 each), fed normally by random permutation. In the tumor or on a CR regimen for 9 years at the time samples, 12 genes were found to be of biopsy. CR induced an upregulation of altered with age, again with one-third genes largely representing structural com- likely false, and the overlap between the ponents and growth regulation, and the normal and tumor samples was only authors also observed a downregulation of three genes. genes involved in mitochondrial bioener- Another early study measured gene getics. Interestingly, the authors found expression in young (age 13 and 14, n 2) little or no evidence for an inhibitory versus old (age 62 to 74, n 3) effect of adult-onset CR on the age- human retinas (Yoshida et al., 2002). The dependent changes in gene expression; of RNA extracted from these samples were the 34 genes upregulated or downregu- hybridized to glass cDNA microarray lated with age in the middle-age and old slides with 2,400 genes, 80 percent of monkeys, only three were found to be sig- which were of neuronal origin. Most genes nificantly altered by CR. Thus, it may be were unchanged with age; only a small that, in primates, the benefit of late-onset number of (24) genes displayed differen- CR is limited. The full impact of adult- tial expression, with these representing onset CR on the life span of these animals energy metabolism, stress response, cell is as yet unknown. growth, and neuronal transmission/signal- A recent study focusing on gene expres- ing. Although this study is limited by a sion in Alzheimer’s diseased brains is very small and unequal sample size as nevertheless noteworthy to the aging field, well as by the small number of genes on not only because of the increased inci- the slide, it does represent one of the earli- dence of Alzheimer’s disease (AD) with est microarray studies to investigate aging age, but also due to findings of a correla- in human tissues. tion of genomic profiles with disease The National Primate Research Center severity, particularly with early or incipi- at the University of Wisconsin, Madison, ent AD (Blalock et al., 2004). The authors has maintained a colony of rhesus mon- correlated hippocampal gene expression keys (Macaca mulatta) for conducting with severity of AD based on ante- longitudinal studies of aging in primates, mortem MiniMental Stage Exam (MMSE) and a cohort of these monkeys are part score and post-mortem measurements of of a long-term CR study in primates. Kayo neurofibrillary tangles (NFTs) and Braak and colleagues (2001) reported an expres- stage scoring. Subjects were assigned sion analysis of 7,070 genes from the to four groups: control (N 9), incipient vastus lateralis muscle of these rhesus AD (N 7), moderate AD (N 8), and monkeys, analyzing the data by pairwise severe AD (N 7). Extracted RNA sam- comparisons between young and old, AL ples were analyzed using high-density P088387-Ch11.qxd 10/31/05 11:33 AM Page 325

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 325

oligonucleotide arrays (14,000 genes), (by Pearson correlation) reveal clusters of and after initial data analysis identifying relatively homogenous expression in indi- AD-related genes by Pearson’s correlation viduals under 42 and over 73 years of with MMSE and NFT, the authors age, with genes upregulated in young employed expression analysis systematic (42) being downregulated in old (73) explorer (EASE, a modified Fisher’s test) to and vice versa. The data also clearly test statistically for co-regulation of genes reflect considerable heterogeneity of gene in a common pathway or process. In this expression in the middle years. The great- manner, the authors identified 3,413 genes est number of genes upregulated with age as AD-related genes across all 31 subjects, were stress-response and DNA repair with genes correlating more strongly with genes, whereas downregulated genes MMSE than with NFT. The authors then included synaptic transmission and vesic- examined expression of these genes in ular transport. This coordinated downreg- only the control and incipient AD groups ulation of a defined cluster of genes and to identify early markers of AD disease, or subsequent induction of antioxidant and incipient AD-related genes, and these repair genes led the authors to hypothe- genes were analyzed with EASE to deter- size that the observed downregulation mine over- or under-represented cate- results from oxidative damage to the pro- gories. Using this approach, the authors moters of certain genes that are prone to report that early or incipient AD is charac- such damage. To test this hypothesis, the terized by transcriptional reprogramming authors developed a real-time PCR assay and cell growth in the hippocampus, measuring DNA damage in specific DNA with an unexpected upregulation of sequences based on the resistance of tumor-suppressor genes, and also by a DNA cleaved at apurinic sites to amplifi- downregulation of bioenergetic pathways. cation, thereby allowing quantitation of Although these findings, possible only DNA damage from a ratio of PCR prod- with a global analysis of gene expression, ucts in cleaved versus uncleaved DNA lead to a better understanding of the early templates. The authors assayed the pro- etiology of AD, they also reflect the pow- moters of 30 selected genes in individual erful utility of microarrays in gaining brain samples and demonstrated an information on fundamental changes asso- age-dependent increase in damage to the ciated with any age-related or progressive promoters of downregulated genes such disease, which in turn directs research as mitochondrial ATP synthase alpha, efforts toward better-targeted therapeutics calmodulin 1, sortilin, and calbindins 1 and interventions. and 2. As proof of principal, the authors Lastly, a study illustrating the power of demonstrated that oxidative damage microarray profiling to generate testable (via FeCl2 and H2O2 treatment) reduced hypotheses is reported by Lu and col- expression of the tau gene in cultured leagues in a survey of human brain gene neuronal cells, and that this damage expression profiles from individuals rang- and subsequent reduction in mRNA ing in age from 26 to 106 years (Lu et al., expression was prevented by concomitant 2004). The authors analyzed RNA tran- overexpression of a human base-repair scripts of post-mortem prefrontal cortex excision enzyme. Furthermore, the genes samples (N 30) using high-density downregulated with age in the human oligonucleotide arrays (11,000 genes), brain samples displayed an increased vul- and age-related genes were determined by nerability to oxidative DNA damage in Spearman rank correlation. Hierarchical cultured cells when compared to genes clustering of these age-related genes and which are stable or upregulated with pairwise comparisons of all the samples age, as determined by quantification of P088387-Ch11.qxd 10/31/05 11:33 AM Page 326

326 F. N. Hudson et al.

promoter damage and luciferase reporter the gene expression changes observed in assays. Taken together, these data support organismal aging. However, these associa- the authors’ model that DNA damage in tions have all been established without the promoters of certain susceptible genes quantitative methods for identifying rela- precipitate their downregulation with tive appearance of functional categories as age, with the hypothesis formulated by compared to the gene expression platform clever interpretation by the authors of the used. As methods for categorizing genes global expression profiles generated with and identifying over- or under-represented microarray technology. categories are refined, we will be able to determine whether these patterns hold true across heterogeneous cultured cell 2. Primate Literature—In Vitro populations. Cell culture systems offer several advan- In addition to seeking candidate genes tages in studies of the cellular physiology associated with cellular senescence that aging process, while also being subject to may be biomarkers of aging, chromoso- the well-known controversy as to the mal position of gene expression changes relatedness of in vitro to in vivo senes- has also been investigated. It has been cence. Although cellular senescence, hypothesized that the genes associated which is highly correlated with telomere with replicative senescence may localize shortening, has not been directly linked to to telomere-proximal regions. However, a mammalian aging, it has been argued that lack of preferential expression or repres- cellular senescence may underlie the func- sion of telomere-proximal genes has been tional decrements present in aging tissues reported (Allen et al., 2004; Chen et al., (Bird et al., 2003). Gene expression in cel- 2004; Minagawa et al., 2004). This result lular senescence has primarily been suggests that in vitro senescence is not addressed using cDNA arrays. The cell due to dysregulation of gene expression culture system has the great advantage genes caused by proximity to shortened that gene expression changes can be telomeres. assayed on a relatively homogeneous cellular population. Therefore, it is notable that a high degree of heterogeneity has IV. Conclusions, Future been identified in the senescence-associ- Directions, and Challenges ated gene expression patterns between cell types even assayed on the same array plat- The technologies and tools to support the form (Bortoli et al., 2003; Shelton et al., use of microarrays for global analysis of 1999). Differences between platforms gene expression have perhaps matured further complicate the comparison of from infancy, but clearly remain in rapid results across experimental systems. development. Especially in applications to Additionally, there has been little statisti- yeast and invertebrate models, we have cal analysis and almost no validation of begun to see the promise of this approach candidate genes with a quantitative inde- for elucidating expression profiles associ- pendent technique to determine exact ated with aging. More importantly, studies genes involved in senescence. However, such as those of McCarroll and col- despite these problems, some patterns are leagues (2004) have demonstrated that emerging. Genes associated with the inter-species comparisons of gene expres- extracellular matrix, cell–cell signaling are sion can discover common (“public”) reported as categories of changed genes transcriptional changes in aging, if only (Minagawa et al., 2004; Shelton et al., at the level of functional classifica- 1999), and these are remarkably close to tion. Similarly, inter-comparisons between P088387-Ch11.qxd 10/31/05 11:33 AM Page 327

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 327

different models of life-span extension, Ashrafi, K., Lin, S. S., Manchester, J. K., & such as caloric restriction and the growth Gordon, J. I. (2000). Sip2p and its partner hormone/IGF-1 axis, are likely to eluci- snf1p kinase affect aging in S. cerevisiae. date the differences and commonalities Genes & Development, 14, 1872–1885. between mechanisms of longevity in a Barr, M. M. (2003). Super models. Physiological Genomics, 13, 15–24. more comprehensive fashion than might Bird, J., Ostler, E. L., & Faragher, R. G. (2003). otherwise be possible. Gene expression Can we say that senescent cells cause ageing? profiles can define sets of genes that serve Experimental Gerontology, 38, 1319–1326. as biomarkers of aging and predictors Bitterman, K. J., Medvedik, O., & of longevity, both of which are greatly Sinclair, D. A. (2003). Longevity regulation needed for characterization of aging in Saccharomyces cerevisiae: linking phenotypes. As published databases of metabolism, genome stability, and expression results increase in number heterochromatin. Microbiology and and quality, the power of meta-analyses Molecular Biology Reviews, 67, 376–399. increases, so that discovery of complex Blalock, E. M., Chen, K. C., Sharrow, K., transcriptional relationships, or coexpres- Herman, J. P., Porter, N. M., Foster, T. C., & Landfield, P. W. (2003). Gene microarrays sion networks, becomes ever more feasible in hippocampal aging: statistical profiling (Stuart et al., 2003). Discovery and quanti- identifies novel processes correlated with tation of these networks will be very cognitive impairment. Journal of important to increase the sophistication Neuroscience, 23, 3807–3819. and reduce the complexity of gene expres- Blalock, E. M., Geddes, J. W., Chen, K. C., sion data analysis, as smaller numbers Porter, N. M., Markesbery, W. R., & of networks can substitute for long lists Landfield, P. W. (2004). Incipient Alzheimer’s of individual genes. Quantitation of tran- disease: microarray correlation analyses scriptional pathways also brings transcrip- reveal major transcriptional and tumor tomics to an interface with metabolomics, suppressor responses. Proceedings of the improving the understanding of changes National Academy of Sciences of the USA, 101, 2173–2178. in the programming of cell function. Bortoli, S., Renault, V., Eveno, E., Auffray, C., The limitations of transcriptomics must Butler-Browne, G., & Pietu, G. (2003). Gene always be kept in sight, particularly that expression profiling of human satellite cells messenger expression may not reflect during muscular aging using cDNA arrays. protein level and post-translational modi- Gene, 321, 145–154. fication of proteins. For a more complete Brewer, G. (2002). Messenger RNA decay understanding, proteomics must be stud- during aging and development. Aging ied. Ultimately, using a full suite of tools Research Reviews, 1, 607–625. for global analysis of “systems biology,” Bronikowski, A. M., Carter, P. A., Morgan, T. J., we can look forward to a deeper and more Garland, T., Jr., Ung, N., Pugh, T. D., integrated understanding of the molecular Weindruch, R., & Prolla, T. A. (2003). Lifelong voluntary exercise in the mouse biology of aging. prevents age-related alterations in gene expression in the heart. Physiological References Genomics, 12, 129–138. Cao, S. X., Dhahbi, J. M., Mote, P. L., & Allen, G. C., West, J. R., Chen, W. J., & Spindler, S. R. (2001). Genomic profiling Earnest, D. J. (2004). Developmental of short- and long-term caloric restriction alcohol exposure disrupts circadian effects in the liver of aging mice. Proceedings regulation of BDNF in the rat of the National Academy of Sciences of the suprachiasmatic nucleus. Neurotoxicology USA, 98, 10630–10635. Teratology, 26, 353–358. Cawthon, R. M., Smith, K. R., O’Brien, E., Arking, R. (1998). The biology of aging Sivatchenko, A., & Kerber, R. A. (2003). (p. 436). Sunderland, MA: Sinauer. Association between telomere length P088387-Ch11.qxd 10/31/05 11:33 AM Page 328

328 F. N. Hudson et al.

in blood and mortality in people aged Dozmorov, I., Bartke, A., & Miller, R. A. 60 years or older. Lancet, 361, 393–395. (2001). Array-based expression analysis of Chen, H.-L., Lu, C.-Y., Hsu, Y.-H., & Lin, J.-J. mouse liver genes: effect of age and of the (2004). Chromosome positional effects of longevity mutant Prop1df. Journals of gene expressions after cellular senescence. Gerontology. Series A, Biological Sciences Biochemical and Biophysical Research and Medical Sciences, 56, B72–B80. Communications, 313(3), 576–586. Dozmorov, I., Galecki, A., Chang, Y., Chipman, H., Hastie, T. J., & Tibshirani, R. Krzesicki, R., Vergara, M., & Miller, R. A. (2003). Clustering microarray data. In (2002). Gene expression profile of T. Speed (Ed.), Statistical analysis of gene long-lived snell dwarf mice. Journals of expression microarray data, pp. 159–200. Gerontology. Series A, Biological Sciences Boca Raton, FL: Chapman & Hall/CRC. and Medical Sciences, 57, B99–B108. Chu, S., DeRisi, J., Eisen, M., Mulholland, J., Dudoit, S., & Fridlyand, J. (2003a). Bagging Botstein, D., Brown, P. O., & Herskowitz, I. to improve the accuracy of a clustering (1998). The transcriptional program of procedure. Bioinformatics, 19, 1090–1099. sporulation in budding yeast. Science, Dudoit, S., & Fridlyand, J. (2003b). 282, 699–705. Classification in microarray experiments. Churchill, G. A. (2002). Fundamentals of In T. Speed (Ed.), Statistical analysis of gene experimental design for cDNA microarrays. expression microarray data, pp. 93–158. Nature Genetics, 32, 490–495. Boca Raton, FL: Chapman & Hall/CRC. Csiszar, A., Ungvari, Z., Koller, A., Dudoit, S., Gentleman, R., Irizarry, R. A., & Edwards, J. G., & Kaley, G. (2003). Yang, Y. H. (2002). Pre-processing DNA Aging-induced proinflammatory shift in Microarray Data Bioconductor Short cytokine expression profile in coronary Course Winter 2002. Available at http:// arteries. FASEB Journal, 17, 1183–1185. www.bioconductor.org/workshops/Wyeth Cui, X., & Churchill, G. A. (2003). How many Course101702/PreProc/PreProc4.pdf. mice and how many arrays? Replication Edwards, M. G., Sarkar, D., Klopp, R., in mouse cDNA microarray experiments. Morrow, J. D., Weindruch, R., & Prolla, T. In K. F. Johnson, & S. M. Lin (Eds.), A. (2003). Age-related impairment of the Methods of Microarray Data Analysis III, transcriptional responses to oxidative stress pp. 139–154. Kluwer Academic Publishers: in the mouse heart. Physiological Norwell, MA. Genomics, 13, 119–127. Dahlquist, K. D., Salomonis, N., Vranizan, K., Edwards, M. G., Sarkar, D., Klopp, R., Lawlor, S. C., & Conklin, B. R. (2002). Morrow, J. D., Weindruch, R., & GenMAPP, a new tool for viewing and Prolla, T. A. (2004). Impairment of analyzing microarray data on biological the transcriptional responses to oxidative pathways. Nature Genetics, 31, 19–20. stress in the heart of aged C57BL/6 mice. DeRisi, J. L., Iyer, V. R., & Brown, P. O. (1997). Annals of the New York Academy of Exploring the metabolic and genetic control Sciences, 1019, 85–95. of gene expression on a genomic scale. Efron, B., & Tibshirani, R. (2002). Empirical Science, 278, 680–686. Bayes methods and false discovery rates for Dhahbi, J. M., Kim, H. J., Mote, P. L., microarrays. Genetic Epidemiology, 23, Beaver, R. J., & Spindler, S. R. (2004). 70–86. Temporal linkage between the phenotypic Eisen, M. B., Spellman, P. T., Brown, P. O., & and genomic responses to caloric restriction. Botstein, D. (1998). Cluster analysis and Proceedings of the National Academy of display of genome-wide expression Sciences of the USA, 101, 5524–5529. patterns. Proceedings of the National Do, K., Broom, B., & Wen, S. (2003). Academy of Sciences of the USA, 95, GeneClust. In G. Parmigiani, E. S. Garrett, 14863–14868. R. A. Irizarry, & S. L. Zeger (Eds.), The Ekstrom, R., Liu, D. S., & Richardson, A. analysis of gene expression data: methods (1980). Changes in brain protein synthesis and software, pp. 342–361. New York: during the life span of male Fischer rats. Springer. Gerontology, 26, 121–128. P088387-Ch11.qxd 10/31/05 11:33 AM Page 329

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 329

Fabrizio, P., & Longo, V. D. (2003). The Abiko, Y. (2002). Microarray analysis of chronological life span of Saccharomyces gene expression changes in aging in mouse cerevisiae. Aging Cell, 2, 73–81. submandibular gland. Journal of Dental Feldman, A. L., Costouros, N. G., Wang, E., Research, 81, 679–682. Qian, M., Marincola, F. M., Alexander, H. R., Holloway, A. J., van Laar, R. K., Tothill, R. W., & Libutti, S. K. (2002). Advantages of mRNA & Bowtell, D. D. L. (2002). Options amplification for microarray analysis. available— from start to finish—for Biotechniques, 33, 906–912, 914. obtaining data from DNA microarrays. Golden, T. R., & Melov, S. (2004). Microarray Nature Genetics, 32, 481–489. analysis of gene expression with age in Horak, C. E., & Snyder, M. (2002). Global individual nematodes. Aging Cell, 3, analysis of gene expression in yeast. 111–124. Functional & Integrative Genomics, 2, Gomes, L. I., Silva, R. L., Stolf, B. S., 171–180. Cristo, E. B., Hirata, R., Soares, F. A., Hsu, A. L., Murphy, C. T., & Kenyon, C. Reis, L. F., Neves, E. J., & Carvalho, A. F. (2003). Regulation of aging and age-related (2003). Comparative analysis of amplified disease by DAF-16 and heat-shock factor. and nonamplified RNA for hybridization Science, 300, 1142–1145. in cDNA microarray. Analytical Huber, W. (2004). Courses in Practical Biochemistry, 321, 244–251. DNA Microarray Analysis 2004, Error Hahn, J. S., Hu, Z., Thiele, D. J., & Iyer, V. R. Models and Normalization: The (2004). Genome-wide analysis of the Bioconductor Project. Available at http:// biology of stress responses through heat www.bioconductor.org/workshops/ shock transcription factor. Molecular and Bressanone/Monday/L2/L2.pdf. Cellular Biology, 24, 5249–5256. Hughes, T. R., Roberts, C. J., Dai, H. Y., Han, E. S., Wu, Y., McCarter, R., Nelson, J. F., Jones, A. R., Meyer, M. R., Slade, D., Richardson, A., & Hilsenbeck, S. G. (2004). Burchard, J., Dow, S., Ward, T. R., Reproducibility, sources of variability, Kidd, M. J., Friend, S. H., & Marton, M. J. pooling, and sample size: important (2000). Widespread aneuploidy revealed considerations for the design of high-density by DNA microarray expression profiling. oligonucleotide array experiments. Journals Nature Genetics, 25, 333–337. of Gerontology. Series A, Biological Sciences Irizarry, R. A., Bolstad, B. M., Collin, F., and Medical Sciences, 59, 306–315. Cope, L. M., Hobbs, B., & Speed, T. P. (2003a). Harper, J. M., Galecki, A. T., Burke, D. T., & Summaries of Affymetrix GeneChip probe Miller, R. A. (2004). Body weight, level data. Nucleic Acids Research, 31, E15. hormones and T cell subsets as predictors Irizarry, R. A., Hobbs, B., Collin, F., of life span in genetically heterogeneous Beazer-Barclay, Y. D., Antonellis, K. J., mice. Mechanisms of Ageing and Scherf, U., & Speed, T. P. (2003b). Development, 125, 381–390. Exploration, normalization, and summaries Helmberg, A. (2001). DNA-microarrays: of high density oligonucleotide array probe novel techniques to study aging and guide level data. Biostatistics, 4, 249–264. gerontologic medicine. Experimental Jazwinski, S. M. (1993). The genetics of aging Gerontology, 36, 1189–1198. in the yeast Saccharomyces cerevisiae. Helmstetter, C. E. (1991). Description of a Genetica, 91, 35–51. baby machine for Saccharomyces cerevisiae. Jin, W., Riley, R. M., Wolfinger, R. D., New Biologist, 3, 1089–1096. White, K. P., Passador-Gurgel, G., & Higami, Y., Pugh, T. D., Page, G. P., Gibson, G. (2001). The contributions of Allison, D. B., Prolla, T. A., & sex, genotype and age to transcriptional Weindruch, R. (2004). Adipose tissue energy variance in Drosophila melanogaster. metabolism: altered gene expression profile Nature Genetics, 29, 389–395. of mice subjected to long-term caloric Kaeberlein, M., & Guarente, L. (2002). restriction. FASEB Journal, 18, 415–417. Saccharomyces cerevisiae MPT5 and SSD1 Hiratsuka, K., Kamino, Y., Nagata, T., function in parallel pathways to promote Takahashi, Y., Asai, S., Ishikawa, K., & cell wall integrity. Genetics, 160, 83–95. P088387-Ch11.qxd 10/31/05 11:33 AM Page 330

330 F. N. Hudson et al.

Kaeberlein, M., Kirkland, K. T., Fields, S., & Proceedings of the National Academy of Kennedy, B. K. (2004). Sir2-independent life Sciences of the USA, 101, 7663–7668. span extension by calorie restriction in yeast. Lashkari, D. A., DeRisi, J. L., McCusker, J. H., Public Library of Science Biology, 9, E296. Namath, A. F., Gentile, C., Hwang, S. Y., Kaeberlein, M., McVey, M., & Guarente, L. Brown, P. O., & Davis, R. W. (1997). Yeast (2001). Using yeast to discover the fountain microarrays for genome wide parallel of youth. Science of Aging Knowledge genetic and gene expression analysis. Environment [SAGE KE] 2001[1], e1. Proceedings of the National Academy Kang, H. L., Benzer, S., & Min, K. T. (2002). of Sciences of the USA, 94, 13057–62. Life extension in Drosophila by feeding a Lee, C. K., Allison, D. B., Brand, J., drug. Proceedings of the National Academy Weindruch, R., & Prolla, T. A. (2002). of Sciences of the USA, 99, 838–843. Transcriptional profiles associated with Kayo, T., Allison, D. B., Weindruch, R., & aging and middle age-onset caloric Prolla, T. A. (2001). Influences of aging and restriction in mouse hearts. Proceedings caloric restriction on the transcriptional of the National Academy of Sciences of profile of skeletal muscle from rhesus the USA, 99, 14988–14993. monkeys. Proceedings of the National Lee, C. K., Klopp, R. G., Weindruch, R., & Academy of Sciences of the USA, 98, Prolla, T. A. (1999). Gene expression profile 5093–5098. of aging and its retardation by caloric Kendziorski, C. M., Irizarry, R. A., Chen, K, restriction. Science, 285, 1390–1393. Haag, J. D., & Gould, M. N. (2004). To pool Lee, C. K., Pugh, T. D., Klopp, R. G., Edwards, or not to pool: a question of microarray J., Allison, D. B., Weindruch, R., & Prolla, experimental design. Johns Hopkins T. A. (2004). The impact of alpha-lipoic University, Dept. of Biostatistics Working acid, coenzyme Q10 and caloric restriction Papers [46]: BE Press. Available at http:// on life span and gene expression patterns in www.bepress.com/jhubiostat/paper46. mice. Free Radical Biology & Medicine, 36, Kerr, M. K. (2003). Design considerations for 1043–1057. efficient and effective microarray studies. Lee, C. K., Weindruch, R., & Prolla, T. A. Biometrics, 59, 822–828. (2000a). Gene-expression profile of the Kerr, M. K., & Churchill, G. A. (2001a). ageing brain in mice. Nature Genetics, 25, Statistical design and the analysis of gene 294–297. expression microarray data. Genetical Lee, H. M., Greeley, G. H. Jr., & Research, 77, 123–128. Englander, E. W. (2001). Age-associated Kerr, M. K., & Churchill, G. A. (2001b). changes in gene expression patterns in Bootstrapping cluster analysis: assessing the the duodenum and colon of rats. reliability of conclusions from microarray Mechanisms of Ageing and Development, experiments. Proceedings of the National 122, 355–371. Academy of Sciences of the USA, 98, Lee, M. L. T., Kuo, F. C., Whitmore, G. A., & 8961–8965. Sklar, J. (2000b). Importance of replication Kerr, M. K., Martin, M., & Churchill, G. A. in microarray gene expression studies: (2000). Analysis of variance for gene Statistical methods and evidence expression microarray data. Journal of from repetitive cDNA hybridizations. Computational Biology, 7, 819–837. Proceedings of the National Academy Kirschner, M., Pujol, G., & Radu, A. (2002). of Sciences of the USA, 97, 9834–9839. Oligonucleotide microarray data mining: Lee, M. L. T., & Whitmore, G. A. (2002). search for age-dependent gene expression. Power and sample size for DNA Biochemical and Biophysical Research microarray studies. Statistics in Medicine, Communications, 298, 772–778. 21, 3543–3570. Landis, G. N., Abdueva, D., Skvortsov, D., Lesur, I., & Campbell, J. L. (2004). The Yang, J., Rabin, B. E., Carrick, J., Tavare, S., transcriptome of prematurely aging yeast & Tower, J. (2004). Similar gene expression cells is similar to that of telomerase-deficient patterns characterize aging and oxidative cells. Molecular Biology of the Cell, 15, stress in Drosophila melanogaster. 1297–1312. P088387-Ch11.qxd 10/31/05 11:33 AM Page 331

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 331

Li, C., Tseng, G. C., & Wong, W. H. (2003). long-term dietary supplementation with Model-based analysis of oligonucleotide antioxidants. Annals of the New York arrays and issues in cDNA microarray Academy of Sciences, 854, 352–360. analysis. In T. Speed (Ed.), Statistical Miller, R. A., Chang, Y., Galecki, A. T., analysis of gene expression microarray data, Al Regaiey, K., Kopchick, J. J., & Bartke, A. pp. 1–34. Boca Raton, FL: Chapman & (2002). Gene expression patterns in Hall/CRC. calorically restricted mice: partial overlap Lin, S. J., Kaeberlein, M., Andalis, A. A., with long-lived mutant mice. Molecular Sturtz, L. A., Defossez, P. A., Culotta, V. C., Endocrinology (Baltimore, Md.), 16, Fink, G., & Guarente, L. (2002). Calorie 2657–2666. restriction extends Saccharomyces Minagawa, S., Nakabayashi, K., Fujii, M., cerevisiae life span by increasing Scherer, S. W., & Ayusawa, D. (2004). respiration. Nature, 418, 344–348. Functional and chromosomal clustering of Lin, S. S., Manchester, J. K., & Gordon, J. I. genes responsive to 5-bromodeoxyuridine (2001). Enhanced gluconeogenesis and in human cells. Experimental Gerontology, increased energy storage as hallmarks of 39, 1069–1078. aging in Saccharomyces cerevisiae. Journal Moreau, Y., Aerts, S., De Moor, B., De Strooper, of Biological Chemistry, 276, 36000–36007. B., & Dabrowski, M. (2003). Comparison and Lipman, R. D., Smith D. E., Bronson, R. T., & meta-analysis of microarray data: from the Blumberg, J. B. (1995). Is late-life caloric bench to the computer desk. Trends in restriction beneficial? Aging: Clinical and Genetics, 19, 570–577. Experimental Research, 7, 136–139. Morley, J. F., & Morimoto, R. I. (2004). Lipman, R. D., Smith D. E., Blumberg, J. B., & Regulation of longevity in Caenorhabditis Bronson, R. T. (1998). Effects of caloric elegans by heat shock factor and molecular restriction or augmentation in adult rats: chaperones. Molecular Biology of the Cell, longevity and lesion biomarkers of aging. 15, 657–664. Aging: Clinical and Experimental Mortimer, R. K., & Johnston, J. R. (1959). Life Research, 10, 463–470. span of individual yeast cells. Nature, 183, Lu, T., Pan, Y., Kao, S. Y., Li, C., Kohane, I., 1751–1752. Chan, J., & Yankner, B. A. (2004). Gene Murphy, C. T., McCarroll, S. A., regulation and DNA damage in the ageing Bargmann, C. I., Fraser, A., Kamath, R. S., human brain. Nature, 429, 883–891. Ahringer, J., Li, H., & Kenyon, C. (2003). MacLean, M., Harris, N., & Piper, P. W. Genes that act downstream of DAF-16 to (2001). Chronological lifespan of stationary influence the lifespan of Caenorhabditis phase yeast cells: a model for investigating elegans. Nature, 424, 277–283. the factors that might influence the ageing Park, P. U., McVey, M., & Guarente, L. (2002). of postmitotic tissues in higher organisms. Separation of mother and daughter cells. Yeast, 18, 499–509. Methods in Enzymology, 351, 468–477. McCarroll, S. A., Murphy, C. T., Zou, S. G., Parmigiani, G., Garrett, E. S., Irizarry, R. A., & Pletcher, S. D., Chin, C. S., Jan, Y. N., Zeger, S. L. (2003). The analysis of gene Kenyon, C., Bargmann, C. I., & Li, H. expression data: an overview of methods (2004). Comparing genomic expression and software. In G. Parmigiani, E. S. patterns across species identifies shared Garrett, R. A. Irizarry, & S. L. Zeger (Eds.), transcriptional profile in aging. Nature The analysis of gene expression data: Genetics, 36, 197–204. methods and software, pp. 1–45. New York: McElwee, J., Bubb, K., & Thomas, J. H. Springer. (2003). Transcriptional outputs of the Perls, T., Shea-Drinkwater, M., Bowen-Flynn, J., Caenorhabditis elegans forkhead protein Ridge, S. B., Kang, S., Joyce, E., Daly, M., DAF-16. Aging Cell, 2, 111–121. Brewster, S. J., Kunkel, L., & Puca, A. A. Meydani, M., Lipman, R. D., Han, (2000). Exceptional familial clustering for S. N., Wu, D., Beharka, A., Martin, extreme longevity in humans. Journal of K. R., Bronson, R., Cao, G., Smith, D., & the American Geriatrics Society, 48, Meydani, S. N. (1998). The effect of 1483–1485. P088387-Ch11.qxd 10/31/05 11:33 AM Page 332

332 F. N. Hudson et al.

Perls, T. T., Wilmoth, J., Levenson, R., dynamics. In G. Parmigiani, E. S. Garrett, R. Drinkwater, M., Cohen, M., Bogan, H., A. Irizarry, & S. L. Zeger (Eds.), The analysis Joyce, E., Brewster, S., Kunkel, L., & Puca, A. of gene expression data: methods and (2002). Life-long sustained mortality software, pp. 409–427. New York: Springer. advantage of siblings of centenarians. Shah, G., Azizian, M., Bruch, D., Mehta, R., & Proceedings of the National Academy of Kittur, D. (2004). Cross-species comparison Sciences of the USA, 99, 8442–8447. of gene expression between human and Phelan, J. P., & Austad, S. N. (1994). Selecting porcine tissue, using single microarray animal models of human aging: inbred platform—preliminary results. Clinical strains often exhibit less biological Transplantation, 18 Suppl 12, 76–80. uniformity than F1 hybrids. Journal of Shalon, D., Smith, S. J., & Brown, P. O. (1996). Gerontology, 49, B1–B11. A DNA microarray system for analyzing Pinheiro, J. C., & Bates, D. M. (2000). Mixed- complex DNA samples using two-color effects models in S and S-PLUS. New York: fluorescent probe hybridization. Genome Springer. Research, 6, 639–45. Pletcher, S. D., Macdonald, S. J., Marguerie, R., Shannon, P., Markiel, A., Ozier, O., Baliga, N. S., Certa, U., Stearns, S. C., Goldstein, D. B., & Wang, J. T., Ramage, D., Amin, N., Partridge, L. (2002). Genome-wide transcript Schwikowski, B., & Ideker, T. (2003). profiles in aging and calorically restricted Cytoscape: a software environment for Drosophila melanogaster. Current Biology, integrated models of biomolecular interaction 12, 712–723. networks. Genome Research, 13, 2498–2504. Preisser, L., Houot, L., Teillet, L., Shelton, D. N., Chang, E., Whittier, P. S., Kortulewski, T., Morel, A., Choi, D., & Funk, W. D. (1999). Microarray Tronik-Le Roux, D., & Corman, B. analysis of replicative senescence. Current (2004). Gene expression in aging kidney Biology, 9, 939–945. and pituitary. Biogerontology, 5, 39–47. Slonim, D. K. (2002). From patterns to Prolla, T. A. (2002). DNA microarray analysis pathways: gene expression data analysis of the aging brain. Chemical Senses, 27, comes of age. Nature Genetics, 32 Suppl, 299–306. 502–508. Prolla, T. A., & Mattson, M. P. (2001). Smeal, T., Claus, J., Kennedy, B., Cole, F., & Molecular mechanisms of brain aging and Guarente, L. (1996). Loss of transcriptional neurodegenerative disorders: lessons from silencing causes sterility in old mother dietary restriction. Trends in cells of S. cerevisiae. Cell, 84, 633–642. Neurosciences, 24, S21–S31. Smyth, G. K. (2004). Linear models and Qin, L. X., & Kerr, K. F. (2004). Contributing empirical Bayes methods for assessing Members of the Toxicogenomics Research differential expression in microarray Consortium. Empirical evaluation of data experiments. Statistical Applications transformations and ranking statistics for in Genetics and Molecular Biology, 3(3), 1. microarray analysis. Nucleic Acids Spellman, P. T., Sherlock, G., Zhang, M. Q., Research, 32(18), 5471–9 Iyer, V. R., Anders, K., Eisen, M. B., Roth, G. S., Lane, M. A., Ingram, D. K., Brown, P. O., Botstein, D., & Futcher, B. Mattison, J. A., Elahi, D., Tobin, J. D., (1998). Comprehensive identification of Muller, D., & Metter, E. J. (2002). cell cycle-regulated genes of the yeast Biomarkers of caloric restriction may predict Saccharomyces cerevisiae by microarray longevity in humans. Science, 297, 811. hybridization. Molecular Biology of the Saal, L. H., Troein, C., Vallon-Christersson, J., Cell, 9, 3273–3297. Gruvberger, S., Borg, A., & Peterson, C. Storey, J. D., & Tibshirani, R. (2003a). (2002). BioArray Software Environment Statistical significance for genomewide (BASE): a platform for comprehensive studies. Proceedings of the National management and analysis of microarray Academy of Sciences of the USA, 100, data. Genome Biology, 3, SOFTWARE0003. 9440–9445. Sebastiani, P., Ramoni, M., & Kohane, I. S. Storey, J. D., & Tibshirani, R. (2003b). SAM (2003). Bayesian clustering of gene expression thresholding and false discovery rates for P088387-Ch11.qxd 10/31/05 11:33 AM Page 333

CHAPTER 11 / Microarray Analysis of Gene Expression Changes in Aging 333

detecting differential gene expression in Weindruch, R., & Prolla, T. A. (2002). DNA microarrays. In G. Parmigiani, Gene expression profile of the aging brain. E. S. Garrett, R. A. Irizarry, & S. L. Zeger Archives of Neurology, 59, 1712–1714. (Eds.), The analysis of gene expression Weindruch, R. H., & Walford, R. L. (1988). data: methods and software, pp. 272–290. The retardation of aging and disease by New York: Springer. dietary restriction. Springfield, IL: Thomas. Stuart, J. M., Segal, E., Koller, D., & Kim, S. K. Welle, S., Brooks, A., & Thornton, C. A. (2003). A gene-coexpression network for (2001). Senescence-related changes in gene global discovery of conserved genetic expression in muscle: similarities and modules. Science, 302, 249–255. differences between mice and men. Tollet-Egnell, P., Parini, P., Stahlberg, N., Physiological Genomics, 5, 67–73. Lonnstedt, I., Lee, N. H., Rudling, M., Wilson, C. L., Pepper, S. D., Hey, Y., & Flores-Morales, A., & Norstedt, G. (2004). Miller, C. J. (2004). Amplification Growth hormone-mediated alteration of protocols introduce systematic but fuel metabolism in the aged rat as reproducible errors into gene expression determined from transcript profiles. studies. Biotechniques, 36, 498–506. Physiological Genomics, 16, 261–267. Woldringh, C. L., Fluiter, K., & Huls, P. G. Tsuchiya, T., Dhahbi, J. M., Cui, X., Mote, P. L., (1995). Production of senescent cells of Bartke, A., & Spindler, S. R. (2004). Saccharomyces cerevisiae by centrifugal Additive regulation of hepatic gene elutriation. Yeast, 11, 361–369. expression by dwarfism and caloric Wu, Z., & Irizarry, R. A. (2004). Preprocessing restriction. Physiological Genomics, 17, of oligonucleotide array data. Nature 307–315. Biotechnology, 22, 656–658. Van Gelder, R. N., von Zastrow, M. E., Wu, Z., Irizarry, R. A., Gentleman, R., Yool, A., Dement, W. C., Barchas, J. D., & Martinez Murillo, F., & Spencer, F. A. Eberwine, J. H. (1990). Amplified RNA (2003). Model based background adjustment synthesized from limited quantities of for oligonucleotide expression arrays. heterogeneous cDNA. Proceedings of the Yang, Y. H., & Speed, T. (2002). Design issues National Academy of Sciences of the USA, for cDNA microarray experiments. Nature 87, 1663–1667. Reviews Genetics, 3, 579–588. Warner, H. R., & Sierra, F. (2003). Models of Yeung, K. Y., Medvedovic, M., & accelerated ageing can be informative Bumgarner, R. E. (2003). Clustering about the molecular mechanisms of gene-expression data with repeated ageing and/or age-related pathology. measurements. Genome Biology, 4(5), R34. Mechanisms of Ageing and Development, Yoshida, S., Yashar, B. M., Hiriyanna, S., & 124, 581–587. Swaroop, A. (2002). Microarray analysis of Warner, H. R., von Heydebreck, A., gene expression in the aging human retina. Sultmann, H., Poustka, A., & Vingron, M. Investigative Ophthalmology & Visual (2002). Variance stabilization applied Science, 43, 2554–2560. to microarray data calibration and to Zeeberg, B. R., Feng, W. M., Wang, G., Wang, M. quantification of differential expression. D., Fojo, A. T., Sunshine, M., Narasimhan, S., Bioinformatics, 18, S96–S104. Kane, D. W., Reinhold, W. C., Lababidi, S., Weindruch, R., Kayo, T., Lee, C. K., & Bussey, K. J., Riss, J., Barrett, J. C., & Prolla, T. A. (2001). Microarray profiling of Weinstein, J. N. (2003). GoMiner: a resource gene expression in aging and its alteration for biological interpretation of genomic and by caloric restriction in mice. Journal of proteomic data. Genome Biology, 4(4), R28. Nutrition, 131, 918S–923S. Zou, S., Meadows, S., Sharp, L., Jan, L. Y., & Weindruch, R., Kayo, T., Lee, C. K., & Jan, Y. N. (2000). Genome-wide study Prolla, T. A. (2002). Gene expression of aging and oxidative stress response in profiling of aging using DNA microarrays. Drosophila melanogaster. Proceedings of Mechanisms of Ageing and Development, the National Academy of Sciences of the 123, 177–193. USA, 97, 13726–13731. P088387-Ch12.qxd 10/31/05 2:53 PM Page 334

Chapter 12

Computer Modeling in the Study of Aging

Thomas B. L. Kirkwood, Richard J. Boys, Colin S. Gillespie, Carole J. Procter, Daryl P. Shanley, and Darren J. Wilkenson

I. Introduction programs alone. There needs to be some integration of the mass of data and A. The Why, What, and How of insight from study of the detailed mecha- Biological Modeling nisms at the level of the physiological Three intersecting processes are making “system.” the application of mathematical and Third, the sophistication and power of computer modeling increasingly import- desktop computer hardware has increased ant in the biological sciences. First, biol- to a point where the kind of model that ogy itself has become much more of an two decades ago might have required an informational science, as a result prima- overnight run on a large mainframe com- rily of the development of genomic puter can now be done in the individual (based on advances in gene sequence and scientist’s lab or office with a response expression data) and post-genomic (based time that makes possible a much more on advances in proteomic and functional interactive way of working. data) sciences. Our capacity to answer Alongside these changes is the develop- questions ranging from cell and molecu- ment of a different perception of the role lar function through to evolutionary and value of computer modeling in bio- genetics requires an increasing ability medical research. To many scientists who to acquire, store, and manipulate large have trained and worked in environments volumes of raw data. This requirement where modeling has not been a part of the has called upon biologists to develop scientific toolkit, the nature and scope of the necessary computational skills and computer modeling is still unclear. Many understanding. see models as essentially descriptive, beg- Second, there is a realization that ging the question “Why bother?” when complex biological processes cannot be the real answer will be revealed in time understood through the application of by experiment. Others have been indoc- ever-more reductionist experimental trinated with the widespread—but largely

334 P088387-Ch12.qxd 10/31/05 2:53 PM Page 335

CHAPTER 12 / Computer Modeling in the Study of Aging 335

false—idea that as soon as a model has broadly consistent with a hypothesized more than two or three parameters, it can mechanism, but modeling can show that “explain” anything, resulting in a suspen- the magnitude of the effect is too small sion of belief that models, particularly of to explain aging on its own. complex systems, can be of any real use 5. Modeling can result in improved at all. Fortunately, the increasing dialog experimental design, especially where the between modelers and experimentalists is system embodies the potential for beginning to break down these barriers of complex interactions. Complexity is very misunderstanding and giving rise to new hard to deal with experimentally but is interactions that are likely to change the relatively straightforward in a computer way a great deal of science will be done in model. Models are thus ideal for analyzing the coming decades. This new approach complex interactions prior to experimen- is commonly being described as systems tal tests. In extreme cases, modeling may biology. This chapter reviews how com- actually reveal that because of interac- puter modeling is developing within the tions within complex systems, a proposed context of the biology of aging. experiment would be inconclusive. The distinctive advantages of modeling 6. Modeling can provide a low-cost, a biological process with the rigor that is rapid test bed for candidate interventions, needed to build a computer model are as thereby enabling a more predictive follows: approach and effecting significant savings in time and money. 1. Model building requires that verbal hypotheses be made specific and concep- To the non-modeler, the science of bio- tually rigorous. Before a mathematical logical modeling can easily appear to model can be formulated, the investigator involve the application of the same set of must specify each element of the model skills to a very diverse range of problems, and how it interacts with other elements. and in much the same kind of way. 2. Starting to build a computer model A facility with numbers, knowledge of may help to highlight gaps in current computer programming, and some under- knowledge. The process of specifying a standing of the biological system seem mathematical model will highlight any all that is required. In reality, the range important unknowns. Sometimes these of approaches and skills in computer can be represented as variables yet to be modeling is broad, involving a significant estimated or determined. diversity of skills and research subdisci- 3. The process of model development plines. Later sections of this chapter might lead to the recognition of a gap examine the different kinds of computer that needs to be filled by further modeling, how they are performed, and experimental investigation, which may what they can achieve. If the integration be fundamental to understanding a between theoretical and experimental complex system. Thus, modeling can be science is to take place as fast and as useful even if the gap means that a effectively as is needed, researchers from model cannot yet be completed. both communities will need to learn 4. Computer models yield quantita- more about each other’s methods of tive as well as qualitative predictions. working. Curiously, there is more simi- A hypothesis can be tested much more larity between the methods of working of rigorously by a model that permits quan- experimenters and modelers than is usu- titative predictions to be made. In aging, ally recognized. The experimenter has to where multiple mechanisms might be at (1) decide which factors to include and work, it often happens that data are vary in the study in order to address the P088387-Ch12.qxd 10/31/05 2:53 PM Page 336

336 T. B. L. Kirkwood et al.

hypothesis most efficiently and directly; the relatively new field of systems (2) spend a large amount of the total biology, which aims, in part, to bridge effort of the study on controls in order to molecular biology and physiology by reduce the possibility of artifact; (3) be capitalizing on the large amount of post- careful in framing the conclusions from genomic data currently being generated. the study so as not to extrapolate beyond Many biochemical networks involve what the results support; and (4) be con- nonlinear components, which means scious of planning the study within a that relying on intuition is not reliable. constrained budget of time, money, and In a recent essay, Lander (2004) provides human resources. So does the modeler. an excellent example of how modeling A number of texts expand on some of has helped reveal the details of a molecu- the issues raised here (e.g., Hilborn & lar mechanism that was proving difficult Mangel, 1997). to understand from a purely experimen- In parts of this chapter, we describe the tal approach: the role of the segmenta- mathematical, statistical, and computa- tion polarity genes in maintaining the tional approaches that have been brought segmentation pattern during Drosophila to bear on understanding the aging development (von Dassow et al., 2000). process. Because these approaches will Although details differ, a similar process be unfamiliar to some readers, we have is shared by all insects, and to a lesser explained the terms and basic concepts degree in vertebrates. as clearly as possible. It is not feasible, Drosophila embryonic development however, to include all of the explana- can be described as a three-stage process. tion that would be necessary to equip the In the first stage, maternally expressed reader new to these approaches with a mRNA enters the Drosophila oocyte complete knowledge base. We have within the ovary, and following transla- therefore had to find a balance between tion, a polarized protein gradient is explanation and concision. At all rele- established. In the second stage, gap vant points within the text, we give and coordinate genes are expressed in references to texts where the reader can response to the protein gradient, which find more detailed explanation. in turn govern the periodic expression of pair-rule genes. The final stage involves the segment polarity genes. A repetitive B. How Computer Models Have Been pattern of engrailed (en) and wingless Instrumental in Solving Biological (wg) expression is established based on Problems the pair-rule genes. The embryo then In many biological domains, it is difficult undergoes cellularization, and the pat- to see how a clear understanding of key tern of en and wg expression is trans- processes could be gained without math- ferred to an intracellular context involv- ematical modeling. This includes, for ing signals between morphologically example, the study of population dynam- distinct bands. The gap and pair-rule ics in ecology, disease transmission in gene products fade, en and wg expression epidemiology, and population genetics is maintained via a complex network of and life history theory in evolutionary transcription factors and intracellular biology. In other domains, single exam- signals, and the segmented structure is ples exemplify the insights that mathe- retained during substantial morphologi- matical investigation can provide, such cal change. as in the study of cardiac fibrillation Von Dassow and colleagues (2000) con- in physiology. Modeling is making an structed a mathematical model based on increasingly important contribution in current information about the segment P088387-Ch12.qxd 10/31/05 2:53 PM Page 337

CHAPTER 12 / Computer Modeling in the Study of Aging 337

polarity gene network to test whether it distantly related insect species in which could maintain a stable segmented earlier stages of development differ. structure. The structure was represented as a series of connected cells, each cell popu- lated with the focal mRNAs, proteins, and II. Why Aging Particularly protein complexes. The known intracellu- Needs Models lar and intercellular relationship between interacting molecules was then repre- Recent years have seen rapid progress sented as a series of differential equations, in the science of aging. A key factor in with simulation used to provide informa- this progress has been the interaction tion on the temporal variation in concen- between evolutionary (why?) and mecha- tration of all the molecules. Although a nistic (how?) lines of research, which substantial body of information existed to gives shape to the likely genetic basis of construct the network, it was insufficient aging and to the mechanisms that may to fully quantify the model—50 parameters be involved (Kirkwood & Austad, 2000). were unknown. The approach taken was This has helped overcome a situation to run simulations with many randomly where the field was dominated by a chosen parameter sets, with each para- plethora of rival theories, with little meter bounded within realistic limits. effective dialog between them. In partic- Interestingly, no solution could be found ular, the disposable soma theory that satisfactorily reproduced the observed (Kirkwood, 1977; Kirkwood & Austad, stable segmented pattern. 2000) suggests that aging is caused Attention was then turned to the ultimately by evolved limitations in network structure, and it was realized organisms’ investments in somatic main- that modifications were needed to cap- tenance and repair rather than by active ture the known biology, namely the gene programming. This predicts that asymmetry in signaling to neighboring aging is due to the gradual accumulation cells anterior and posterior. With appro- of unrepaired random molecular faults, priate modifications in place, the differ- leading to an increasing fraction of dam- ential equations were updated and the aged cells and eventually to functional process of simulation with randomly impairment of older tissues and organs. chosen sets of parameters was repeated. Genetic effects on the rate of aging are, The observed segmented structure was in this view, mediated primarily through now reproduced with surprising ease. genes that influence somatic mainte- The important result was that it was the nance and repair. fine detail of the network structure that Although the idea of aging as a buildup was key in determining the system of damage is straightforward in principle behavior and not exact parameter values. and supported by a growing range of data, Interestingly, this may reflect an evolu- it presents a number of distinctive chal- tionary adaptation as minor variations, lenges (Kirkwood et al., 2003). First, it pre- such as mutations in components of the dicts that there are multiple mechanisms network or environmental fluctuations that cause aging, instead of just one or a affecting levels of signals, would not few. Second, it predicts that aging is inher- unduly affect the segment structure and ently stochastic—that is, it is modulated to subsequent development. The major an important degree by chance. Extensive message was that the segment polarity evidence points to an important contri- genes represent a “robust developmental bution in aging that arises from chance module” that ensures the formation of variations, which are not explained by an appropriate pattern even across genetic or environmental factors (Finch & P088387-Ch12.qxd 10/31/05 2:53 PM Page 338

338 T. B. L. Kirkwood et al.

35 upstream mechanisms that set a process

30 in train and the end-stage mechanisms that dominate the cellular phenotype 25 at the end of its life. For example, a 20 gradual accumulation of mitochondrial 15 (mt)DNA mutations, occurring over No dying

10 years, might lead to a steady increase in the production of reactive oxygen 5 species (ROS) and a gradual decline in 0 0 1020304050 energy production (Kowald & Kirkwood, Days 1996) . However, although the buildup of Figure 12.1 Life-span distributions for individual mtDNA mutations initiates the process, Caenorhabditis elegans nematodes in isogenic popu- what ultimately destroys the cell is that lations of wildtype (filled bars) and age-1 (open bars) eventually a threshold is reached where strains. Redrawn from Kirkwood and Finch (2002); homeostatic mechanisms collapse. The original data from Johnson (1990). end-stage of the cell’s life span is dominated by dramatic biochemical changes, such as an accumulation of damaged Kirkwood, 2000). A particularly clear protein. Experimental study of the latter example of the role of chance in aging effect, or even of the former cause, in the is the threefold range in life span (see absence of a quantitative model to link Figure 12.1) and the apparently stochastic the two would find it hard to establish age-related cell degeneration of individ- the connection. ual worms in isogenic populations of Another benefit of integrative model Caenorhabditis elegans reared under building is that it is well suited to take uniform laboratory conditions (Herndon account of the fact that many of the key et al., 2002; Kirkwood & Finch, 2002). reactions involved in normal cell mainte- Third, since multiple mechanisms con- nance and metabolism do not act in tribute to aging, a high level of complexity isolation—rather, they belong to a net- is to be expected. For all of these reasons, work of activity. When the activity of there is exceptional need in aging research one enzyme changes, all connected for the use of computer models to help metabolite pools and enzyme activities integrate findings from different lines of may be altered. In some cases, there may experimental work. be redundancy in pathways, which pro- Although the multiplicity of aging vides buffering against damage, whereas mechanisms is now widely acknowl- in other cases, the effect of damage may edged, the reductionist nature of be propagated. experimental techniques means that, in Another important area for modeling is practice, most research is still narrowly to understand the actions of genes that focused on single mechanisms. This is affect the rate of aging. Over the past where computer modeling can make a decade, scores of genes have been identi- major contribution. By allowing for fied that affect aging in yeast, nematodes, interaction and synergism between dif- fruit flies, and mice, and there is growing ferent processes, models reveal that interest in genes affecting human the predicted effects on the system are longevity (Gems & Partridge, 2001; often much greater than when mecha- Jazwinski, 2000; Larsen, 2001; Lithgow, nisms are considered one at a time. 1998; Tan et al., 2004). Experimental Furthermore, models can highlight data are beginning to reveal the interac- important differences between the tions of these genes within pathways P088387-Ch12.qxd 10/31/05 2:53 PM Page 339

CHAPTER 12 / Computer Modeling in the Study of Aging 339

that control the aging rate, and there is low-cost, rapid test bed for candidate evidence that several of the most import- interventions. ant genes are those that affect basic cellular processes, such as insulin and B. Simple Versus Complex insulin-like growth factor (IGF) signaling, which are strongly conserved across the Biological systems are complex and species range (Gems & Partridge, 2001; involve the interrelationships of many Rincon et al., 2004). Nevertheless, we are different “species,” where species can a long way from understanding the inter- refer to molecules, cells, tissues, or actions between these effects. These organisms. A model is a description of studies need also to take account of the the system. The “art” in building a good intrinsic stochastic nature of gene regula- model is to capture the essential details tory networks. of the biology, without burdening the model with nonessential details. Every model is to some extent a simplification III. Different Approaches to of the biology, but it is valuable in taking Modeling Biological Systems an idea that might have been expressed purely verbally and making it more A. Descriptive Versus Predictive explicit. Nevertheless, the question still A descriptive model describes a process remains: what level of complexity should or behavior that has already been be incorporated in the model? observed. A predictive model predicts the At the most basic level, a model must behavior of a system not previously be able to capture the desired inputs and observed. A valid descriptive model is outputs of a system. This is where a clear, often easier to develop but it has less prior specification of the problem to be value than a model with predictive addressed is as essential in modeling as it power. However, descriptive models are is in experimentation. For instance, if we useful for highlighting gaps in our cur- wish to investigate how an increase of rent knowledge. A model may start out ATP will affect the production of ROS by as being descriptive but can then be used mitochondria, then obviously these ele- to predict outcomes when parts of the ments must be included into the model. system are perturbed. For example, a Other elements, such as the dynamics of descriptive model of a metabolic path- a cell cycle, are likely to be excluded. way of proteins under known conditions However, it is at this point that the mod- can be used to predict protein functions eler needs to exercise caution and to bear under different circumstances or in in mind the opportunities that are avail- different species. able to include further factors in the A predictive model provides quantifi- model than could easily be added to an able as well as qualitative predictions. experiment. The choice of which reac- The value of quantifiable predictions is tions should be left out of the model— that a hypothesis can be tested much since the activity of one enzyme may con- more rigorously. In aging, where numer- ceivably affect all connected metabolite ous mechanisms might be at work, data pools—has no easy or universal answer. are often broadly consistent with a Some modelers prefer to start simple and hypothesized mechanism, but modeling add further detail as required; others pre- can show that the magnitude of the fer to recognize greater complexity from effect is too small to explain aging on the outset. Either way, a modeler should its own. Another advantage of predic- always develop a model with as much tive models is that they can provide a direct biological input as possible. P088387-Ch12.qxd 10/31/05 2:53 PM Page 340

340 T. B. L. Kirkwood et al.

Another key factor in the modelers’ GRID—the new generation of hardware/ decision-making process is the time software computer networking that is scale of processes involved. Molecular designed to facilitate the sharing of data processes often need to be modeled on a and compute resources over a network. time scale of seconds or less; outcomes A benefit of the GRID is the harnessing affecting aging develop in months or of idle computer power. For instance, years; evolutionary changes occur over whereas a model may take weeks on a generations. Models that seek to integrate single 500-MHz processor, if 50 machines, across levels present particularly challeng- say in a university computer laboratory, ing problems that need to be addressed in which are idle for 12 hours per day were defining the aims and scope of the project. set to the task, a properly formulated When modeling a process as complex model could take hours. as aging, an unfortunate side-effect is that very quickly, the mathematical C. Discrete Versus Continuous representation can become exceedingly complex. Although the modeler may be When modeling biological processes, it is comfortable with each and every detail of often helpful to treat time as a discrete the model, the reader may be presented quantity divided into a number of inter- with an indecipherable collection of vals. For instance, when dealing with symbols. Conversely, an oversimplifica- the cell cycle of the budding yeast tion of the systems may lead to the Saccharomyces cerevisiae, it may natu- justified claim that the model does not ral to deal in terms of generations represent the structure under considera- (Gillespie et al., 2004; Sinclair, 2002). tion. Thankfully, part of this problem is Although some types of system natu- being overcome with the introduction of rally lend themselves to discrete-time standard methods for describing models modeling, it is important to consider any used throughout the biological commu- distortion that may be introduced. In the nity, such as the Systems Biology Markup yeast cell cycle example, a mother cell Language (SBML, described in more produces on average 24 daughter cells; detail later). When a standard has been however, the time taken to form a decided, this enables generic tools to be daughter cell gradually increases. So if developed that aid the understanding of the events being modeled were directly models. For instance, an SBML-aware affected by the interbudding interval, the visualization tool should accept any model may be of limited validity if only SBML-encoded model and return a graph- discrete generations were considered. ical representation of it. A barrier that limits the amount of D. Deterministic Versus Stochastic complexity that can be included in a model is computational power. Put sim- A model can be generally classed as deter- ply, do we have a computer powerful ministic or stochastic. A deterministic enough to calculate a solution to our model is one that takes no account of model? It is relatively easy to construct a random variation and therefore gives a simple model that when simulated could fixed and precisely reproducible result. take weeks to finish. With the yearly It can be solved by numerical analysis increase in processor power, models that or computer simulation. Deterministic would have taken weeks of computa- models are often mathematically tional time 5 years ago can now be solved described by sets of differential equations. in a matter of minutes. Other exciting Deterministic models are appropriate avenues include the emergence of the when large numbers of individuals of a P088387-Ch12.qxd 10/31/05 2:53 PM Page 341

CHAPTER 12 / Computer Modeling in the Study of Aging 341

species are involved and the importance A stochastic model is formulated in of statistical variations in the average terms of probabilities, so at each time behavior of the system is relatively unim- interval, the degradation of species X has portant. However for many biological sys- an associated probability. Again, since tems, this assumption may not be valid. the model is reasonably simple, the exact To illustrate the concepts, let us con- stochastic solution can be obtained: sider perhaps the simplest of molecular reactions, spontaneous degradation. The X(0) P (t) e k1Xt(1e k1t)X(0) X (3) ordinary differentiation equation for the x X degradation of a species X at rate k1 is given by where X takes the values between 0 and its initial amount X(0). dx k1X . In most biological systems, the number dt (1) of species involved and the interactions between them mean that for stochastic Because this is a simple equation, it models, an analytical solution—that is, can be solved exactly to give the deter- one that can be obtained by purely alge- ministic solution braic formulae without using a com- X(t) X(0)e k1t (2) puter—will not be feasible. In these cases, computer simulations of the stochastic where X(0) is the initial amount of kinetics are used. A simulation keeps species X. track of the number and state of each A stochastic model should be used species over time. Therefore, it is neces- when either the number of a particular sary to carry out repeated simulations and species is small or when there is reason then look at the distribution of results to to expect random events to have an get a picture of the central tendency, the important influence on the behavior of dispersion, and outliers. This process is the system. Often, a stochastic model called Monte Carlo simulation. will be more appropriate when we need Figure 12.2 shows a stochastic real- to take account of species as discrete ization of a spontaneous degradation units rather than as continuous variables, and particularly when the numbers of a particular species may become small. It 100 Deterministic may also be necessary to take account of 90 Stochastic 1 Stochastic 2 events occurring at random times. The 80 Stochastic 3 essential difference between a stochastic 70

and deterministic model is that in a sto- 60

chastic model, different outcomes can 50

result from the same initial conditions. 40

A stochastic model is formulated in No. of molecules 30

terms of probabilities and is constructed 20 by considering the probability that an 10 event occurs during a small time period. 0 Formulating the model in this manner 0 102030405060708090100 Time enables us to calculate the probability that the population is of size X at time t, Figure 12.2 Comparison between a stochastic realization and the deterministic solution for a simple px(t). Because the model is reasonably degradation reaction. In this stochastic realization, simple, the exact stochastic solution can the molecules are degraded quicker than predicted be obtained. by the deterministic solution. P088387-Ch12.qxd 10/31/05 2:53 PM Page 342

342 T. B. L. Kirkwood et al.

reaction and its deterministic counter- model for further investigation generally part. For a given starting condition, a requires a significant degree of comput- single stochastic realization may differ ing skill. considerably from its deterministic coun- As noted above (in section IIIB), there terpart. This is particularly important in has recently been an increasing effort to models when the number of a particular agree on a standard method for represent- species becomes low and the species may ing mathematical models. One standard or may not become extinct. Figure 12.2 that has been widely adopted is the shows a clear example of a species reach- Systems Biology Markup Language ing a very low concentration but never (SBML). SBML provides a computer-read- becoming totally extinct in the deter- able format for representing models of ministic model, whereas in the stochas- biochemical reaction networks. Although tic model, the species becomes extinct SBML is human-readable, it is intended and the time of extinction varies in that it will usually be other software that different realizations. In the modeling of would both read and write any models. epidemic diseases within a host popula- This is analogous with the now wide- tion, where it may matter greatly spread use of HTML for Web documents. whether and when the first or last infec- Although a human can read HTML tive individual dies or recovers, the source documents, these are intended difference between stochastic and deter- primarily for reading by a browser, such ministic models can be very marked. as Internet Explorer, that transforms the Similar considerations can arise in gene HTML code into a more easily read docu- regulatory networks with respect to ment on screen. the random association of transcription Currently there are over 60 groups factor complexes. using the SBML standard (see Hucka Both the deterministic and stochastic et al., 2003). Some tools, such as methods have their respective advan- CellDesigner, have been created that tages and disadvantages. The modeler enable models to be constructed using a should determine which method is more drag-and-drop approach. Using this suitable to the task at hand (it may approach, a user creates species that are sometimes be both) and use that which assigned to graphical nodes. The nodes is appropriate. are then connected up using arrows to denote reactions (see for example Funahashi et al., 2003). E. Software Tools Other tools, such as JigCell, allow the There are many ways to develop a model, user to construct a model using chemical from using traditional programming equations combined with a spreadsheet languages such as C, Fortran, and Java approach (Allen et al., 2003). Using this to mathematical packages such as approach is perhaps more useful when Mathematica, Matlab, and R. Conven- dealing with large and complex models, tional publication of computer models is whereas the graphical approach is partic- generally restricted to the presentation of ularly useful when constructing a model a few key predictions, although it is com- for the first time. mon to allow the reader to download the SBML is not the only standard that has computer program, or source. In order to emerged. For example, the Petri Net replicate the same predictions as were Markup Language (PNML) and CellML published, the program must first be are similar efforts in creating standards downloaded and any necessary algorithm (see Lloyd et al., 2004; Weber & Kindler, libraries installed. To actually use the 2003). Although each standard focuses on P088387-Ch12.qxd 10/31/05 2:53 PM Page 343

CHAPTER 12 / Computer Modeling in the Study of Aging 343

different aspects of the model, they are Model not mutually exclusive. Hence, an effort building is being made to create tools that allow the transformation from one language to another.

Experimental Model data validation F. Validation of Models Once we have constructed a model and are satisfied with its behavior, we need to Extending/ Using the test the model against observations from refining the model the biological system that it represents. model This process is called validation. First, it Figure 12.3 Flow diagram to show the steps in is necessary to test the model to see building a model. Once a model has been built, it is whether it fits the data that we already tested against experimental data. If the model does have. Any discrepancies here need to be not agree well, then the modeler goes back to the model-building stage. Otherwise, the model can be addressed. Once the model has been used and then extended further as more knowledge validated in this way, we should then becomes available. test the model against data that were not used to estimate parameters for our model. If the model predictions and the from data, a multiple of the standard error observed data are not in close agreement, is appropriate. For parameter values that then the modeler needs to study the have been guessed, a guess at the percent- model to try and find where the discrep- age reliability is also required. Caution ancies arise. This could mean modifying needs to be taken when parameter esti- the model or adding further detail to the mates are correlated because if one para- model. This is an important step as it meter estimate is changed, some of the may highlight that the current knowl- others might have to be changed too. edge of the system is insufficient and that further experimental work should be carried out. Once modifications to the IV. Currently Available Models model have been made, the model is of Aging tested again, as shown in Figure 12.3. Another aspect of validation is sensitiv- Existing work reflects the variety of cur- ity analysis to assess how varying model rent models in aging research, which parameters affect the model outcomes. range from detailed modeling of individ- There are two reasons why this is useful. ual intracellular mechanisms to higher- First, we may be interested in some partic- level modeling required to address the ular parameters—for example, the rate of fundamental problem of why aging degradation of a protein and how this should occur. We will not attempt to be might affect the buildup of damaged pro- wholly comprehensive in this review but tein. Second, some of the parameter rates will illustrate the breadth of coverage and might not have been accurately deter- the different methodologies employed. mined and so it is important to see how sensitive the model is to small changes in A. Intracellular Mechanisms these. The size of changes to make for each parameter depends on how well the para- There are a large number of models cur- meter was determined initially. In the case rently available that focus on individual of parameters that have been estimated intracellular mechanisms. Currently, the P088387-Ch12.qxd 10/31/05 2:53 PM Page 344

344 T. B. L. Kirkwood et al.

models that relate directly to aging have 2. Somatic Mutations been mainly concerned with telomere The role of somatic mutations in aging shortening, the accumulation of somatic is an area of particularly active research, mutations, and the accumulation of following new methods for measuring defective mitochondria. DNA modification and repair. The somatic mutation theory was first pro- 1. Telomere Models posed several decades ago after experiments showed that irradiation shortened Telomeres are repetitive DNA sequences life span in animal models and induced found at both ends of linear chromo- features of premature aging (Henshaw somes. In telomerase-negative cells, et al., 1947; Lindop & Rotblat, 1961). telomeres shorten with each cell divi- Szilard (1959) proposed a mathemati- sion, and this process eventually causes cal model that assumed that recessive cells to enter a state of replicative senes- mutational “hits” in diploid organisms cence. One cause of telomere shortening would accumulate so that a cell could is the end-replication problem caused by continue to function until one pair of the inability of DNA polymerases to genes had both received a “hit.” replicate a linear DNA molecule to its Holliday and Kirkwood (1981) developed very end. In the 1990s, models were a deterministic model of the accumula- developed to try to explain replicative tion of recessive mutations in human senescence in human fibroblasts based fibroblast populations. A stochastic solely on the end-replication problem model of the same processes was later (Arino et al., 1995; Levy et al., 1992). developed (Kirkwood & Proctor, 2003), Later models included additional mecha- which also considered the possibility nisms of telomere shortening. Rubelj and that there may be synergistic inter- Vondracek (1999) modeled abrupt telom- actions between mutations. ere shortening due to DNA recombina- tion or nuclease digestion. It has been found that an increase in oxidative stress 3. Mitochondria Models accelerates the rate of telomere shortening due to an accumulation of single- The free-radical theory of aging proposes strand breaks in telomeric DNA (von that ROS, which are constantly gener- Zglinicki et al., 1995; von Zglinicki ated through normal cell metabolism in et al., 2000). More recent models have the mitochondria, cause aging by damag- included this additional mechanism ing membranes, proteins, and DNA and found that the models predict that (Harman, 1956). The mitochondrial oxidative stress plays an important role theory of aging proposes that an accumu- (Proctor & Kirkwood, 2002, 2003). For lation of defective mitochondria is a example, simulations showed that major contributor to the cellular deterio- increasing the level of ROS led to fewer ration that underlies the aging process cell divisions on average. Space does not (Harman, 1972). Studies have shown that permit detail of all of the current models defective mitochondria accumulate with of telomere shortening, but the inter- age to a greater extent in post-mitotic tis- ested reader may refer to the references sues (Cortopassi et al., 1992; Lee et al., for further models (Aviv et al., 2003; den 1994), although it has recently been Buijs et al., 2004; Golubev et al., 2003; reported that high levels of mitochondr- Hao & Tan, 2002; Olofsson & Kimmel, ial defects are observed in aged human 1999; Sidorov et al., 2004; Tan, 1999a,b, colon (Taylor et al., 2003). In addition, 2001). several studies have shown that muscle P088387-Ch12.qxd 10/31/05 2:53 PM Page 345

CHAPTER 12 / Computer Modeling in the Study of Aging 345

fibers are taken over by a single form of and somatic mutations in nuclear and mutant mtDNA (Brierley et al., 1998; mitochondrial DNA (Sozou & Kirkwood, Müller-Höcker et al., 1993). 2001). We are currently engaged in a Hypotheses to explain the apparent major effort, the Biology of Ageing “clonal expansion” of mutant mtDNA are e-Science Integration and Simulation a replication advantage for the mutant (BASIS) project, to develop interactive mtDNA; slower degradation of mutant models that can network a variety of mitochondria (de Grey, 1997); and random individual processes together in a flexi- intracellular drift. Mathematical models ble, user-friendly manner (Kirkwood et have been developed to explore quantita- al., 2003). One of the aims of the BASIS tive predictions from these ideas. Kowald project is to allow models of individual and Kirkwood (2000) developed a deter- mechanisms to be linked together to ministic model based on de Grey’s form a “Virtual Aging Cell” (Proctor & hypothesis. Other models are based on Kirkwood, 2003). the idea of random intracellular drift (Chinnery & Samuels, 1999; Elson et al., B. Tissue Models 2001). The functional properties of an aging organ or tissue can become compro- 4. Chaperone Models mised, even if most of the cells are in Molecular chaperones have an important good working order. Mathematical role in helping to maintain protein models are required to help us try to homeostasis within cells. It has been understand how a fraction of damaged observed that the induction of heat cells can lead to altered tissue function. shock proteins, a major class of chaper- Early models were motivated by the fact ones, is impaired with age and that there that cultured human diploid fibroblasts is also a decline in chaperone function. cannot be grown indefinitely in culture Although there are a few mathematical (Hayflick, 1972). These models were models on the role of heat shock pro- based on the commitment theory, the teins in the cell, to date only one model idea that cells become irreversibly com- has looked at the role of chaperones in mitted to senescence while still out- the aging process (Proctor et al., 2005). wardly healthy (Holliday et al., 1977; This model describes how heat shock Kirkwood & Holliday, 1975). proteins are upregulated after an Recently, extensive experimental data increase in intracellular stress and can has been generated on intrinsic age be used to investigate the effect of stress changes that affect the function of intes- on protein homeostasis. tinal stem cells in aging mice (Loeffler et al., 1993; Martin et al., 1998a,b). This has led to a number of mathematical 5. Network Models models (e.g., Gerike et al., 1998; Loeffler A few models exist that show how differ- et al., 1993; Meineke et al., 2001). ent mechanisms interact synergistically Another model based on data from (Kowald & Kirkwood, 1994, 1996; Sozou muscle-derived stem cells has also been & Kirkwood, 2001), examples being the developed (Deasy et al., 2003). interactions of defective mitochondria, Another tissue system that has been aberrant proteins, free radicals, and scav- extensively studied and modeled is the engers in the aging process (Kowald & population of T cells and their role in Kirkwood, 1996); and the interactions of immunosenescence (Luciani et al., 2001; telomere shortening, oxidative stress, Romanyukha & Yashin, 2003). P088387-Ch12.qxd 10/31/05 2:53 PM Page 346

346 T. B. L. Kirkwood et al.

C. Organism Models tion to age have been analyzed using techniques such as dynamic programming, There are only a very limited number of simulated annealing, and Pontryagin’s models dealing with whole-organism maximization principle (Abrams & aging, and these are limited to unicellular Ludwig, 1995; Blarer & Doebeli, 1996; organisms. Budding yeast, Saccharomyces Cichon, 1997; Clark & Mangel, 2000; cerevisiae, is commonly used to study Houston & McNamara 1999; Schaffer, cellular aging. Accumulation of extrachro- 1983; Teriokhin, 1998; Vaupel et al., 2004). mosomal ribosomal DNA circles (ERCs) The following specific issues have appears to be an important contributor to attracted attention using approaches and aging in yeast, and a mathematical model modeling techniques drawn from the has been developed to examine this domains of life-history theory, demogra- process (Gillespie et al., 2004). Another phy, and population genetics. interesting model contrasts regulatory with stochastic processes in genetic segre- gation during division as a mechanism for 1. Dietary Restriction the aging observed in the asexually repro- Dietary restriction is observed to cause ducing ciliate Styloncychia (Duerr et al., slowing of aging and extension of life in 2004). many species. One hypothesis is that animals have evolved a response to temporary fluctuations in resource availability, D. Population Models in which energy is diverted from reproduc- An area of research that has contributed to tion to maintenance functions in periods a fundamental understanding of why aging of food shortage, thereby enhancing sur- occurs is life-history theory—a theory that vival and retaining reproductive potential essentially deals with schedules of growth, for when conditions improve. A detailed survival, and reproduction maximizing quantitative development of this hypothe- Darwinian fitness (Kirkwood & Austad, sis using a dynamic resource allocation 2000; Kirkwood & Rose, 1991; Partridge & model revealed that the effect could be Barton, 1993). Many classic life-history the result of the suggested evolutionary papers included an investigation of senes- process provided that the following condi- cence (Cole, 1954; Fisher, 1930; Hamilton, tions were satisfied: (1) there is a substan- 1966; Williams, 1957). Another modeling tial initial cost to reproduction, and approach is represented by the disposable (2) juveniles are at a disadvantage during soma theory (Kirkwood, 1977; Kirkwood periods of food shortage (Shanley & & Rose, 1991), founded on the principles Kirkwood, 2000). An alternative approach of optimality theory (Parker & Maynard is presented using a dynamic energy Smith, 1990). The disposable soma theory budget (van Leeuwen et al., 2002). suggests that aging arises as part of an Recently, metabolic control analysis has optimal life history due to tradeoffs in been used to help identify an increased resource allocation between investment in proton leak in the mitochondrial inner reproduction and maintenance affecting membrane as one possible mechanism long-term survival. There is much data in whereby ROS production is reduced general support of the existence of such (Lambert & Merry, 2004). tradeoffs, including in humans (Lycett et al., 2000; Westendorp & Kirkwood, 2. Negligible Senescence 1998). More complex general life-history models that have incorporated measures Species that exhibit negligible senescence of an organism’s state such as size in addi- are of particular interest (Finch, 1998) and P088387-Ch12.qxd 10/31/05 2:53 PM Page 347

CHAPTER 12 / Computer Modeling in the Study of Aging 347

are well represented in species that con- Menopause is manifest when this pool is tinue to grow after maturation, so-called near to exhaustion, and mechanistic indeterminate growth. A dynamic model models have focused on the follicular that explicitly included size as a state vari- dynamics (Faddy et al., 1992) and prob- able predicts that this should be the case, lems of fetal loss that increase in fre- and indeed predicts that some species quency as the oocyte pool is depleted should show negative senescence (Vaupel (O’Connor et al., 1998). Other modeling et al., 2004). Interestingly, conditions for has focused on determining whether non-aging can be found in a general life- at some age mothers may increase their history model in an analysis of vitality: a fitness by diverting investment from term that combines the declining fecun- continued reproduction to existing dity and increasing mortality characteris- offspring and grand-offspring (Hawkes tic of senescence (Sozou & Seymour, et al., 1998; Lee, 2003; Peccei, 1995; 2004). Rogers, 1993; Shanley & Kirkwood, 2001). To date the results have not been conclusive, but given the importance of 3. Gompertz, Mortality Plateaus, and intergenerational assistance—for exam- Heterogeneity ple, as seen in the two-fold improvement Most species exhibit an exponential in mortality for infants with a living increase in mortality with age that can be grandmother (Sear et al., 2002) combined described by the Gompertz model or by with the particularly high risk of mortal- the Gompertz-Makeham model that has ity in childbirth for human females—an an extension to include extrinsic sources of evolutionary explanation remains a clear mortality (Golubev, 2004). One problem in possibility and the development and the acceptance of this model is the obser- testing of further models appears likely. vation that in large laboratory populations, mortality rates appear to plateau at later ages (Vaupel et al., 1998). A number of V. Models, Data Collection, and models have been proposed to account for Experimental Design this pattern, such as an evolutionary tradeoff (Mueller & Rose, 1996; Mueller et al., Models are developed based on the 2003), a combination of mutation accumu- collective understanding of the scientific lation and pleiotropy (Charlesworth, 2001), community regarding the underlying a state-based approach (Mangel & Bonsall, mechanisms driving the processes of 2004), and individual heterogeneity within interest. The related activity of designing the population (Pletcher & Curtsinger, experiments to provide data to falsify or 1998). refine the models is essentially just a for- malization of “the scientific method.” However, there are a number of issues 4. Human Menopause that arise in the context of complex The rapid reproductive senescence asso- dynamic models that make this proce- ciated with menopause in human dure far from straightforward in practice. females occurs well in advance of gen- Models are typically concerned with eral somatic senescence and poses an underlying mechanisms that are difficult interesting evolutionary problem. The or impossible to measure directly through reproductive life span of human females conventional experimental procedures. is limited, as in almost all other mam- Consequently, they often contain param- mals, by a finite pool of oocytes estab- eters (such as various kinds of rate lished in the developing fetus. constants) whose values are not known. P088387-Ch12.qxd 10/31/05 2:53 PM Page 348

348 T. B. L. Kirkwood et al.

If a model cannot make predictions it is necessary to collect data that are (or regarding quantities that can (at least in can be) predicted by the model. This may principle) be measured experimentally, require model refinement but is neces- then it is not falsifiable, and in an impor- sary; otherwise, there is nothing to link tant sense is not “scientific.” There is the model and the data that is collected. therefore a requirement to develop mod- Second, one must collect data correspon- els that are predictive, as this then ding to model predictions that are sensi- affords an opportunity to compare the tive to underlying model assumptions model predictions with experimentally (structural and otherwise). This is neces- determined reality. As well as providing sary on the grounds of falsifiability. the opportunity for falsification, this pre- Third, it is necessary to gather measure- dictive behavior also potentially allows ments that help to answer questions of “calibration” of model parameters by key scientific interest. For example, if a finding combinations of parameters that key model parameter of interest is the reduce the discrepancy between the degradation rate of a particular protein, model predictions and reality. then measurements should be taken on These kinds of “inverse” problems data that are sensitive to the choice of have long been recognized in the physical rate rather than data that are relatively and engineering sciences, and there is a robust to this choice. This will ensure large literature concerned with attempts that the data collected provide useful to solve them. The problem can be under- information. Fourth, enough data should stood generally as follows: a complex be gathered to ensure that an adequate system has a range of inputs, and based assessment can be made of inter- and on these, produces a range of outputs. intra-experimental variation; otherwise, The inverse problem is to find a set of there is no way to be sure that the data inputs to the system that closely matches are representative and that the model is a given set of outputs (desired target or not being calibrated to fit an atypical experimental observation). In the context data set. If the model being calibrated is of an attempt to match an experimentally stochastic, more data are probably observed history of a physical system, required, as it is likely to be necessary to attempts to solve the inverse problem are reliably determine a good approximation often referred to as history matching or to the full probability distribution of key calibration, of which more is presented observables. These basic principles are in the next section. Such calibration tech- fairly self-evident, but effectively opera- niques are generally applied post hoc, tionalizing them for a complex dynamic after the experimental data have been simulation model is not necessarily collected. However, in the context of easy. However, there are many excellent biological modeling, there is often an texts on experimental design that can opportunity to go back to the lab to col- provide further guidance (see for example lect appropriate data for model validation Clarke & Kempson, 1997; Cochran & and calibration. Cox, 1992; and Mead, 1988). A question then naturally arises as to exactly what data should be collected, and how much. In the context of com- VI. Parameter Inference plex biological models, such experimen- A. The Calibration Problem tal design questions are difficult to tackle within a formal statistical frame- Many approaches have been applied to work, but it is relatively straightforward the calibration problem within the to justify some guiding principles. First, traditional engineering context. First, a P088387-Ch12.qxd 10/31/05 2:53 PM Page 349

CHAPTER 12 / Computer Modeling in the Study of Aging 349

(deterministic) computer simulator for model parameters given data, and (model) of the physical (or biological) sys- these correspond to different schools of tem under study is built, so that for a statistical thought. Classical frequentist given set of inputs, a corresponding set of approaches seek to construct estimators outputs may be computed. Next, a sim- that are a function of the data and have ple measure of “distance” for how far the desirable properties (such as consistency output is from the desired match is and lack of bias) under repeated sam- defined. The input space is then searched pling. However, even leaving aside the for a set that minimizes the distance serious philosophical objections many measure. This optimization problem can people have to the repeated sampling be approached using a variety of tech- framework at the heart of frequentist niques for multivariate function mini- inference, there are many practical diffi- mization—for example, steepest descent, culties associated with applying such Newton methods, conjugate gradients, techniques in the context of complex simulated annealing, genetic program- dynamic models. Consequently, few stat- ming, and so on (Koza et al., 2001). Such isticians would consider a frequentist minimization techniques can work well framework in this scenario. if the computer simulator is very fast, Approaches based on the likelihood but for a simulator of a large, complex function of the data provide a more system, which may take hours or days powerful and natural way of addressing for a single run, such naïve approaches the simulation model parameter inference generally fail. problem. These can be divided into two Simple search methods are very waste- main camps. The first is the maximum ful of information. Typically, a very small likelihood school, which attempts to be number of runs are used to decide on a “objective” by using only the likelihood new “best guess” for the target input function of the data and seeks combina- parameters, and then all existing infor- tions of parameters, which makes the mation is discarded as the search contin- data as likely as possible conditional on ues from this new input set. In contrast, those parameters. The likelihood (or log- statistical approaches to the calibration likelihood) function is used as a way of problem attempt to use all available runs “scoring” the goodness of fit, which can from the simulator in order to infer a then be optimized. Although this sounds model for the relationship between the straightforward in principle, the likeli- inputs to and outputs from the simulator. hood function is typically not analytically In this context, it is not necessarily opti- tractable for complex models, and this mal to always evaluate the simulator at introduces a variety of complications. the current “best guess” at the optimal The second camp is the Bayesian input set, but instead to evaluate at an school, which corrects the conditioning input set that gives the most information from data on parameters to parameters on regarding the relationship between the data, and thereby seeks parameters that inputs and outputs in the vicinity of the are likely given the data. This is done at predicted optimal set. Thus, such statisti- the expense of introducing prior distribu- cal approaches to calibration need to tions into the problem but has a range of combine both non-parametric statistical benefits as a result. These include the inference techniques and experimental fact that the resulting framework is fully design algorithms in order to effectively probabilistic, and that probabilistic infor- solve the problem (Sacks et al., 1989). mation regarding all parameters can be A range of different approaches can be obtained from the posterior distribution. taken to carrying out statistical inference The use of priors is also valuable, as they P088387-Ch12.qxd 10/31/05 2:53 PM Page 350

350 T. B. L. Kirkwood et al.

help regularize the problem and allow that is independent of the calibration the modeler to incorporate information issue. In addition to the experimental data, regarding realistic parameter ranges there will be data y (y1, y2,...,yN) into the inference algorithm. A further obtained from N runs of the computer benefit of the Bayesian framework is that simulator (where N will typically be larger because it is probabilistic, powerful com- than n, even in the case of an expensive putational algorithms may be naturally simulator), where applied to problems where the likelihood is analytically intractable. Markov yj (x*j, tj) (5) chain Monte Carlo (MCMC) algorithms (Gamerman, 1997) use stochastic simula- is the result of the jth computer experi- tion techniques to obtained realizations ment, xj* is the experimental condition from the (complex) posterior distribution, associated with the jth computer experi- which are then used to draw inferences ment and tj is the set of calibration about model parameters. For more infor- parameters associated with the jth com- mation about Bayesian inference, see puter experiment (Kennedy & O’Hagan, Bernardo and Smith (2000), O’Hagan and 2001). Forster (2004), and references therein. Note that within this framework, the computer simulator of the biological model is represented by a (deterministic) B. Statistical Approaches to Simulation function (.,.), which can be evaluated at Model Calibration any combination of experimental condi- Although non-Bayesian approaches to tions and calibration parameters by run- the calibration problem are possible, the ning the simulator with the specified complexity and dimensionality of the input. If the simulator were very fast, so problem, together with the need to incor- that evaluating (.,.) were cheap, then porate available expert prior information standard Bayesian inference techniques regarding, inter alia, plausible ranges for could be used in order to make direct rate constants and information on data inferences for using (1), generating data quality, mean that a Bayesian approach is of the form (2) as and when required. particularly attractive. Typically, (in the However, due to the expense of evaluat- context of deterministic processes), a ing (.,.) for large complex models, (.,.) is model is specified in the following form: often regarded as an unknown function, modeled using a Gaussian process. Thus zi (xi) i , (xi) (xi, ) (xi) (4) inference may proceed for using only the N computer simulator runs available. Here z (z1, z2,...,zn) represents the Bayesian inference is typically carried available experimental data, obtained out using a mixture of analytic direct from n different experimental conditions matrix computations related to Gaussian x1, x2, ...xn; (xi) is the real behavior of processes together with computationally the biological system under experimental intensive techniques, using Markov condition xi; i is the measurement error chain Monte Carlo (MCMC) methods. associated with the ith experiment; is a Note that in the context of biological bias associated with the computer simula- modeling, choice of the experimental tor of the biochemical system; (xi, ) is conditions for the n “wet lab” experiments the result of running the computer simu- will often (though not always) be predeter- lator under experimental condition xi with mined. However, the choice of conditions the “perfect” set of calibration parameters for the N computer simulator runs will be ; and (xi) represents model inadequacy at least partly under the calibrator’s control P088387-Ch12.qxd 10/31/05 2:53 PM Page 351

CHAPTER 12 / Computer Modeling in the Study of Aging 351

and will be of key importance to the over- stochastic process corresponding to the all effectiveness of the procedure. This is a simulator is modeled directly, and all nontrivial (sequential) experimental design aspects of the process that are not problem, but limited literature already observed are “filled-in” probabilistically exists that will provide guidance in this using appropriate MCMC techniques. area (Craig et al., 1996; Currin et al., 1991; Conditional on complete knowledge of the Kennedy & O’Hagan, 2001; Sacks et al., stochastic process, inference for any rate 1989). parameters driving the system dynamics is Many complex computer codes have straightforward. The difficulty of such the facility to be run at different levels of methods is in the construction of the sophistication, and hence accuracy. The MCMC algorithms to fill in the missing BASIS simulator, for example, may be run aspects of the stochastic process. This is in “exact” mode, where the simulation of very problem-specific and generally the stochastic process used to model a requires a fairly detailed understanding of given biochemical system is “perfect,” the underlying dynamics, including the based on a discrete event simulation strat- likelihood function, as well as experience egy similar to the Gillespie algorithm. in the use of MCMC algorithms. The use Such exact simulation procedures are of these techniques for identification of desirable but are typically very expensive biological models is still in its infancy, but to carry out. On the other hand, the sys- see Boys and colleagues (2004), Gibson & tem may also be run in an approximate Renshaw (2001), Golightly & Wilkinson mode, based on a time-discretization of (2005), and O’Neill (2002) for some suc- the process, where both the accuracy cessful examples. of the procedure and the time taken for a run depend on the size of time-step adopted. In this case, it can often be VII. Conclusions optimal to combine a large number of fast (but less accurate) runs with a small num- This chapter has examined the rationale ber of slow (but accurate) runs in order to for the use of computer models in study- make most efficient use of computer ing the aging process and has reviewed the time. There is already a sizable literature range of models that have been developed. in this area; see for example Higdon and It has also described some of the generic colleagues (2003), Kennedy and O’Hagan issues that need to be addressed in terms (2001), and references therein. of the methodology of modeling. In coming years, it is likely to be essential, if aging research is to realize its potential, C. Direct Statistical Parameter Inference that modeling studies are greatly extended The calibration techniques alluded to and that models are increasingly used to above work well in the deterministic con- link together the pieces of the picture that text but are not completely straightfor- are revealed by reductionist experimental ward to extend to the case of stochastic techniques. These developments are an simulation models. For a stochastic model inherent part of the “new” ways of doing of relatively low dimension, it may be pos- science that are commonly described as sible to make a direct attempt to carry out “systems biology.” Whether systems biol- statistical inference for the parameters of ogy is really new or not is a matter for the system given (for example, time debate, and a spectrum of opinion can be course) experimental data on the system found. What is unquestionably new is the dynamics. Here, rather than regarding the mass of detailed information emerging simulator as an “unknown function,” the at accelerating pace from functional P088387-Ch12.qxd 10/31/05 2:53 PM Page 352

352 T. B. L. Kirkwood et al.

genomic technologies, the rapid expan- Boys, R. J., Wilkinson, D. J., & sion of raw computing power, the devel- Kirkwood, T. B. L. (2004). Bayesian oping connectivity offered by advances in inference for a stochastic kinetic model. Internet (soon to be GRID-enabled) Web Submitted. services, and the recruitment of increas- Brierley, E. J., Johnson, M. A., Lightowlers, R. N., James, O. F. W., & ing numbers of mathematicians, statisti- Turnbull, D. M. (1998). Role of cians, and computer scientists into the mitochondrial DNA mutations in human life sciences. aging: implications for the central nervous Not all areas of biology need to be system and muscle. Annals of Neurology, taken over by the systems approach, but 43, 217–223. few are likely to remain untouched by it. Charlesworth, B. (2001). Patterns of age- The biology of aging is one area where it specific means and genetic variances of is hard to envisage the necessary progress mortality rates predicted by the mutation- being made without embracing the sys- accumulation theory of ageing. Journal of tems approach. There are just too many Theoretical Biology, 210, 47–65. mechanisms, levels of action, and experi- Chinnery, P. F., & Samuels, D. C. (1999). Relaxed replication of mtDNA: a model mental models for it to be realistic to with implications for the expression of anticipate integration without the use of disease. American Journal of Human computer models. Effecting the building Genetics, 64, 1158–1165. of the cross-disciplinary research pro- Cichon, M. (1997). Evolution of longevity grams to bring this about is going to be a through optimal resource allocation. challenge, but it should also prove to be Proceedings of the Royal Society of London intellectually stimulating and fun. Series B, 264, 1383–1388. Clark, C. W., & Mangel, M. (2000). Dynamic state variable models in ecology: methods References and applications. England: Oxford Abrams, P. A., & Ludwig, D. (1995). University Press. Optimality theory, Gompertz’ law, and the Clarke, G. M., & Kempson, R. E. (1997). disposable soma theory of senescence. Introduction to the Design and Analysis of Evolution, 49, 1055–1066. Experiments. London: Arnold. Allen, N. A., Calzone, L., Chen, K. C., Cochran, W., & Cox, G. (1992). Experimental Ciliberto, A., Ramakrishman, N., designs. Chichester, NY: Wiley. Shaffer, C. A., Sible, J. C., Tyson, J. J., Vass, Cole, L. C. (1954). The population M. T., Watson, L. T., & Zwolak, J. W. consequences of life history phenomena. (2003). Modeling regulatory networks at Quarterly Review of Biology, 29, 103–137. Virginia Tech. OMICS, A Journal of Cortopassi, G. A., Shibata, D., Soong, N. W., & Integrative Biology, 7, 285–299. Arnheim, N. (1992). A pattern of Arino, O., Kimmel, M., & Webb, G. F. (1995). accumulation of a somatic deletion of Mathematical modelling of the loss of mitochondrial-DNA in aging human tissues. telomere sequences. Journal of Theoretical Proceedings of the National Academy of Biology, 177, 45–57. Sciences of the USA, 89, 7370–7374. Aviv, A., Levy, D., & Mangel, M. (2003). Craig, P. S., Goldstein, M., Seheult, A. H., & Growth, telomere dynamics and successful Smith, J. A. (1996). Bayes linear strategies for and unsuccessful human aging. Mechanisms matching hydrocarbon reservoir history. In of Ageing & Development, 124, 829–837. A. F. M. Smith (Ed.), Bayesian statistics 5. Bernardo, J. M., & Smith, A. F. M. (2000). (pp. 69–95). Oxford: Oxford Science Bayesian theory. Chichester, NY: Wiley. Publications. Blarer, A., & Doebeli, M. (1996). Heuristic Currin, C., Mitchell, T., Morris, M., & optimization of the general life history Ylvisaker, D. (1991). Bayesian prediction of problem: a novel approach. Evolutionary deterministic functions, with applications Ecology, 10, 81–96. to the design and analysis of computer P088387-Ch12.qxd 10/31/05 2:53 PM Page 353

CHAPTER 12 / Computer Modeling in the Study of Aging 353

experiments. Journal of the American picture. Current Opinions in Genetics & Statistical Association, 86, 953–963. Development, 11, 287–292. de Grey, A. D. N. J. (1997). A proposed Gerike, T. G., Paulus, U., Potten, C. S., & refinement of the mitochondrial free Loeffler, M. (1998). A dynamic model of radical theory of aging. BioEssays, 19, proliferation and differentiation in the 161–166. intestinal crypt based on a hypothetical Deasy, B. M., Jankowski, R. J., Payne, T. R., intraepithelial growth factor. Cell Cao, B., Goff, J. P., Greenberger, J. S., & Proliferation, 31, 93–110. Huard, J. (2003). Modeling stem cell Gibson, G. J., & Renshaw, E. (2001). Likelihood population growth: incorporating terms for estimation for stochastic compartmental proliferative heterogeneity. Stem Cells, 21, models using Markov chain methods. 536–545. Statistics & Computing, 11, 347–358. den Buijs, J. O., van den Bosch, P. P. J., Gillespie, C. S., Proctor, C. J., Boys, R. J., Musters, M., & van Riel, N. A. W. (2004). Shanley, D. P., Wilkinson, D. J., & Mathematical modeling confirms the Kirkwood, T. B. L. (2004). A mathematical length-dependency of telomere shortening. model of ageing in yeast. Journal of Mechanisms of Ageing & Development, Theoretical Biology, 229, 189–196. 125, 437–444. Golightly, A., & Wilkinson, D. J. (2005). Duerr, H. P., Eichner, M., & Ammermann, D. Bayesian inference for stochastic kinetic (2004). Modeling senescence in models using a diffusion approximation. hypotrichous ciliates. Protist, 155, 45–52. Biometrics, 61, 781–788 Elson, J. L., Samuels, D. C., Turnbull, D. M., & Golubev, A. (2004). Does Makeham make Chinnery, P. F. (2001). Random intracellular sense? Biogerontology, 5, 159–167. drift explains the clonal expansion of Golubev, A., Khrustalev, S., & Butov, A. (2003). mitochondrial DNA mutations with age. An in silico investigation into the causes of American Journal of Human Genetics, 68, telomere length heterogeneity and its 802–806. implications for the Hayflick limit. Journal Faddy, M. J., Gosden, R. G., Gougeon, A., of Theoretical Biology, 225, 153–170. Richardson, S. J., & Nelson, J. F. (1992). Hamilton, W. D. (1966). The moulding of Accelerated disappearance of ovarian senescence by natural selection. Journal of follicles in midlife: implications for Theoretical Biology, 12, 12–45. forecasting menopause. Human Hao, Y. H., & Tan, Z. (2002). The generation Reproduction, 7, 1342–1346. of long telomere overhangs in human cells: Finch, C. E. (1998). Variations in senescence a model and its implication. and longevity include the possibility of Bioinformatics, 18, 666–671. negligible senescence. Journals of Harman, D. (1956). A theory based on free Gerontology A: Biological Sciences, 53, radical and radiation chemistry. Journal of B235–B239. Gerontology, 11, 298–300. Finch, C. E., & Kirkwood, T. B. L. (2000). Harman, D. (1972). The biologic clock: the Chance, development & aging. New York: mitochondria? Journal of American Oxford University Press. Geriatics Society, 20, 145–147. Fisher, R. A. (1930). The genetical theory of Hawkes, K., O’Connell, J. F., Jones, N. G. B., natural selection. Oxford: Clarendon Alvarez, H., & Charnov, E. L. (1998). Press. Grandmothering, menopause, and the Funahashi, A., Tanimura, N., Morohashi, M., & evolution of human life histories. Kitano, H. (2003). CellDesigner: a process Proceedings of the National Academy of diagram editor for gene-regulatory and Sciences of the USA, 95, 1336–1339. biochemical networks. Biosilico, 1, 159–162. Hayflick, L. (1972). In H. Bredt, & J. W. Rohen, Gamerman, D. (1997). Markov chain Monte (Eds.), Ageing and Development, Vol. 4, Carlo: stochastic simulation of Bayesian pp. 1–15. Stuttgart: Schattaner Verlag. inference. London: Chapman & Hall. Henshaw, P. S., Riley, E. F., & Stapleton, G. E. Gems, D., & Partridge, L. (2001). Insulin/IGF (1947). The biologic effects of pile radiation. signalling and ageing: seeing the bigger Radiology, 49, 349–364. P088387-Ch12.qxd 10/31/05 2:53 PM Page 354

354 T. B. L. Kirkwood et al.

Herndon, L. A., Schmeissner, P. J., Kirkwood, T. B. L., & Finch, C. E. (2002). The Dudaronek, J. M., Brown, P. A., old worm turns more slowly. Nature, 419, Listner, K. M., Sakano, Y., Paupard, M. C., 794–795. Hall, D. H., & Driscoll, M. (2002). Kirkwood, T. B. L., & Holliday, R. (1975). Stochastic and genetic factors influence Commitment to senescence: a model for tissue-specific decline in ageing C. elegans. the finite and infinite growth of diploid and Nature, 419, 808–814. transformed human fibroblasts in culture. Higdon, D., Lee, H., & Holloman, C. (2003). Journal of Theoretical Biology, 53, 481–496. Markov chain Monte Carlo-based Kirkwood, T. B. L., & Proctor, C. J. (2003). approaches for inference in Somatic mutations and ageing in silico. computationally intensive inverse Mechanisms of Ageing & Development, problems. In M. West (Ed.), Bayesian 124, 85–92. Statistics 7 (pp. 181–197). Oxford: Oxford Kirkwood, T. B. L., & Rose, M. R. (1991). Science Publications. Evolution of senescence—late survival Hilborn, R., & Mangel, M. (1997). The sacrificed for reproduction. Philosophical ecological detective. Confronting models Transactions of the Royal Society of London with data. Princeton, NJ: Princeton Series B-Biological Sciences, 332, 15–24. University Press. Kirkwood, T. B. L., Boys, R. J., Gillespie, C. S., Holliday, R., & Kirkwood, T. B. L. (1981). Proctor, C. J., Shanley, D. P., & Wilkinson, Predictions of the somatic mutation and D. J. (2003). Towards an e-biology of ageing: mortalization theories of cellular aging are integrating theory and data. Nature Reviews contrary to experimental-observations. Molecular Cell Biology, 4, 243–249. Journal of Theoretical Biology, 93, 627–642. Kowald, A., & Kirkwood, T. B. L. (1994). Holliday, R., Huschtscha, L. I., Towards a network theory of ageing: a Tarrant, G. M., & Kirkwood, T. B. L. (1977). model combining the free radical theory Testing the commitment theory of cellular and the protein error theory. Journal of aging. Science, 198, 366–372. Theoretical Biology, 168, 75–94. Houston, A. I., & McNamara, J. M. (1999). Kowald, A., & Kirkwood, T. B. L. (1996). A Models of adaptive behaviour: an approach network theory of ageing: the interactions of based on state. Cambridge: Cambridge defective mitochondria, aberrant proteins, University Press. free radicals and scavengers in the ageing Hucka, M., Finney, A., Sauro, H. M., process. Mutation Research – DNAging Bolouri, H., Doyle, J. C., & Kitano, H. (2003). Genetic Instability and Aging, 316, 209–236. The Systems Biology Markup Language Kowald, A., & Kirkwood, T. B. L. (2000). (SBML): a medium for representation and Accumulation of defective mitochondria exchange of biochemical network models. through delayed degradation of damaged Bioinformatics, 19, 524–531. organelles and its possible role in the ageing Jazwinski, S. M. (2000). Metabolic of post-mitotic and dividing cells. Journal mechanisms of yeast ageing. Experimental of Theoretical Biology, 202, 145–160. Gerontology, 35, 671–676. Koza, J. R., Mydlowec, W., Lanza, G., Yu, J., & Jahnson, T. E. (1990). Increased life span of Keane, M. A. (2001). Automated reverse age-1 mutants in caenorhabditis elegans engineering of metabolic pathways from and lower Gompertz rate of aging. Science observed data by means of genetic 249, 908–912. programming. In H. Kitano (Ed.), Kennedy, M. C., & O’Hagan, A. (2001). Foundations of systems biology (pp. Bayesian calibration of computer models. 95–121). Cambridge, MA: MIT Press. Journal of the Royal Statistical Society Lambert, A. J., & Merry, B. J. (2004). Effect of Series B-Statistical Methodology, 63, caloric restriction on mitochondrial 425–450. reactive oxygen species production and Kirkwood, T. B. L. (1977). Evolution of ageing. bioenergetics: reversal by insulin. Nature, 270, 301–304. American Journal of Physiology- Kirkwood, T. B. L., & Austad, S. N. (2000). Regulatory, Integrative & Comparative Why do we age? Nature, 408, 233–238. Physiology, 286, 71–79. P088387-Ch12.qxd 10/31/05 2:53 PM Page 355

CHAPTER 12 / Computer Modeling in the Study of Aging 355

Lander, A. D. (2004). A calculus of purpose. Martin, K., Kirkwood, T. B. L., & Potten, C. S. Plos Biology, 2, 712–714. (1998a). Age changes in stem cells of Larsen, P. L. (2001). Asking the age-old murine small intestinal crypts. questions. Nature Genetics, 28, 102–104. Experimental Cell Research, 241, 316–323. Lee, C. M., Pang, C. Y., Hsu, H. S., & Wei, Y. H. Martin, K., Potten, C. S., Roberts, S. A., & (1994). Differential accumulation of 4977 bp Kirkwood, T. B. L. (1998b). Altered stem deletion in mitochondrial DNA of various cell regeneration in irradiated intestinal tissues in human ageing. Biochemica Et crypts of senescent mice. Journal of Cell Biophysica Acta, 1226, 37–43. Science, 111, 2297–2303. Lee, R. D. (2003). Rethinking the Mead, R. (1988). The design of experiments. evolutionary theory of aging: transfers, not England: Cambridge University Press. births, shape social species. Proceedings of Meineke, F. A., Potten, C. S., & Loeffler, M. the National Academy of Sciences of the (2001). Cell migration and organization in USA, 100, 9637–9642. the intestinal crypt using a lattice-free Levy, M. Z., Allsopp, R. C., Futcher, A. B., model. Cell Proliferation, 34, 253–266. Greider, C. W., & Harley, C. B. (1992). Mueller, L. D., & Rose, M. R. (1996). Telomere end-replication problem and cell Evolutionary theory predicts late-life aging. Journal of Molecular Biology, 225, mortality plateaus. Proceedings of the 951–960. National Academy of Sciences of the USA, Lindop, P. J., & Rotblat, J. (1961). Shortening 93, 15249–15253. of life and causes of death in mice exposed Mueller, L. D., Drapeau, M. D., Adams, C. S., to single whole-body dose of radiation. Hammerle, C. W., Doyal, K. M., Nature, 189, 645–648. Jazayeri, A. J., Ly, T., Beguwala, S. A., Lithgow, G. J. (1998). Aging mechanisms from Mamidi, A. R., & Rose, M. R. (2003). nematodes to mammals. Nutrition, 14, Statistical tests of demographic 522–524. heterogeneity theories. Experimental Lloyd, C. M., Halstead, M. D. B., & Gerontology, 38, 373–386. Nielsen, P. F. (2004). CeIIML: its future, Müller-Höcker, J., Seibel, P., present and past. Progress in Biophysics & Schneiderbanger, K., & Kadenbach, B. (1993). Molecular Biology, 85, 433–450. Different in situ hybridization patterns of Loeffler, M., Birke, A., Winton, D., & Potten, C. mitochondrial DNA in cytochrome c (1993). Somatic mutation, monoclonality and oxidase-deficient extraocular muscle fibres stochastic-models of stem-cell organization in the elderly. Virchows Arch (A), 422, 7–15. in the intestinal crypt. Journal of Theoretical O’Connor, K. A., Holman, D. J., & Wood, J. W. Biology, 160, 471–491. (1998). Declining fecundity and ovarian Luciani, F., Valensin, S., Vescovini, R., ageing in natural fertility populations. Sansoni, P., Fagnoni, F., Franceschi, C., Maturitas, 30, 127–136. Bonafe, M., & Turchetti, G. (2001). A O’Hagan, A., & Forster, J. J. (2004). Bayesian stochastic model for CD8() T cell dynamics inference. London: Hodder Arnold. in human immunosenescence: Implications Olofsson, P., & Kimmel, M. (1999). Stochastic for survival and longevity. Journal of models of telomere shortening. Theoretical Biology, 213, 587–597. Mathematical Biosciences, 158, 75–92. Lycett, J. E., Dunbar, R. I. M., & Voland, E. O’Neill, P. D. (2002). A tutorial introduction (2000). Longevity and the costs of to Bayesian inference for stochastic reproduction in a historical human epidemic models using Markov chain population. Proceedings of the Royal Monte Carlo methods. Mathematical Society of London Series B-Biological Biosciences, 180, 103–114. Sciences, 267, 31–35. Parker, G. A., & Smith, J. M. (1990). Mangel, M., & Bonsall, M. B. (2004). The Optimality Theory in Evolutionary Biology. shape of things to come: using models with Nature, 348, 27–33. physiological structure to predict mortality Peccei, J. S. (1995). A hypothesis for the origin trajectories. Theoretical Population and evolution of menopause. Maturitas, 21, Biology, 65, 353–359. 83–89. P088387-Ch12.qxd 10/31/05 2:53 PM Page 356

356 T. B. L. Kirkwood et al.

Pletcher, S. D., & Curtsinger, J. W. (1998). shortening in infants and adults. Journal of Mortality plateaus and the evolution of Theoretical Biology, 226, 169–175. senescence: Why are old-age mortality rates Sinclair, D. A. (2002). Paradigms and pitfalls so low? Evolution, 52, 454–464. of yeast longevity research. Mechanisms of Proctor, C. J., & Kirkwood, T. B. L. (2002). Ageing & Development, 123, 857–867. Modelling telomere shortening and the role Sozou, P. D., & Kirkwood, T. B. L. (2001). A of oxidative stress. Mechanisms of Ageing & stochastic model of cell replicative Development, 123, 351–363. senescence based on telomere shortening, Proctor, C. J., & Kirkwood, T. B. L. (2003). oxidative stress, and somatic mutations in Modelling cellular senescence as a result nuclear and mitochondrial DNA. Journal of of telomere state. Aging Cell, Theoretical Biology, 213, 573–586. 2, 151–157. Sozou, P. D., & Seymour, R. M. (2004). To age Proctor, C. J., Söti, C., Boys, R. J., or not to age. Proceedings of the Royal Gillespie, C. S., Shanley, D. P., Wilkinson, Society of London Series B-Biological D. J., & Kirkwood, T. B. L. (2005). Sciences, 271, 457–463. Modelling the actions of chaperones and Szilard, L. (1959). On the nature of the aging their role in ageing. Mechanisms of Ageing process. Proceedings of the National & Development, 126, 119–131. Academy of Sciences of the USA, 45, 35–45. Rincon, M., Muzumdar, R., Atzmon, G., & Tan, Q., Yashin, A. I., Christensen, K., Jeune, Barzilai, N. (2004). The paradox of the B., De Benedictis, G., Kruse, T. A., & Vaupel, insulin/IGF-1 signaling pathway in J. W. (2004). Multidisciplinary approaches in longevity. Mechanisms of Ageing & genetic studies of human aging and Development, 125, 397–403. longevity. Current Genomics, 5, 409–416. Rogers, A. R. (1993). Why menopause. Tan, Z. (1999a). Intramitotic and intraclonal Evolutionary Ecology, 7, 406–420. variation in proliferative potential of Romanyukha, A. A., & Yashin, A. I. (2003). human diploid cells: Explained by telomere Age related changes in population of shortening. Journal of Theoretical Biology, peripheral T cells: towards a model of 198, 259–268. immunosenescence. Mechanisms of Tan, Z. (1999b). Telomere shortening and Ageing & Development, 124, 433–443. the population size-dependency of life Rubelj, I., & Vondracek, Z. (1999). Stochastic span of human cell culture: further mechanism of cellular aging: abrupt implication for two proliferation- telomere shortening as a model for restricting telomeres. Experimental stochastic nature of cellular aging. Journal Gerontology, 34, 831–842. of Theoretical Biology, 197, 425–438. Tan, Z. (2001). Simulated shortening of Sacks, J., Welch, W., Mitchell, T., & Wynn, H. proliferation-restricting telomeres during (1989). Design and analysis of computer clonal proliferation and senescence of human experiments. Statistical Science, 4, 409–435. cells. Experimental Gerontology, 36, 89–97. Schaffer, W. M. (1983). The application of Taylor, R. W., Barron, M. J., Borthwick, G. M., optimal control theory to the general life Gospel, A., Chinnery, P. F., Samuels, D. C., history problem. American Naturalist, 121, Taylor, G. A., Plusa, S. M., Needham, S. J., 418–431. Greaves, L. C., Kirkwood, T. B. L., Sear, R., Steele, F., McGregor, A. A., & Mace, R. Turnbull, D. M. (2003). Mitochondrial (2002). The effects of kin on child mortality DNA mutations in human colonic crypt in rural Gambia. Demography, 39, 43–63. stem cells. Journal of Clinical Shanley, D. P., & Kirkwood, T. B. L. (2000). Investigation, 112(9), 1351–1360. Calorie restriction and aging: a life-history Teriokhin, A. T. (1998). Evolutionary optimal analysis. Evolution, 54, 740–750. age schedule of repair: computer modelling Shanley, D. P., & Kirkwood, T. B. L. (2001). of energy partition between current and Evolution of the human menopause. future survival and reproduction. Bioessays, 23, 282–287. Evolutionary Ecology, 12, 291–307. Sidorov, I. A., Gee, D., & Dimitrov, D. S. van Leeuwen, I. M. M., Kelpin, F. D. L., & (2004). A kinetic model of telomere Kooijman, S. (2002). A mathematical model P088387-Ch12.qxd 10/31/05 2:53 PM Page 357

CHAPTER 12 / Computer Modeling in the Study of Aging 357

that accounts for the effects of caloric breaks is the major cause of telomere restriction on body weight and longevity. shortening in human fibroblasts. Free Biogerontology, 3, 373–381. Radical Biology & Medicine, 28, 64–74. Vaupel, J. W., Baudisch, A., Dolling, M., Roach, von Zglinicki, T., Saretzki, G., Docke, W., & D. A., & Gampe, J. (2004). The case for Lotze, C. (1995). Mild hyperoxia shortens negative senescence. Theoretical Population telomeres and inhibits proliferation of Biology, 65, 339–351. fibroblasts: a model for senescence. Vaupel, J. W., Carey, J. R., Christensen, K., Experimental Cell Research, 220, 186–193. Johnson, T. E., Yashin, A. I., Holm, N. V., Weber, M., & Kindler, E. (2003). Petri net Iachine, I. A., Kannisto, V., Khazaeli, A. A., technology for communication-based Liedo, P., Longo, V. D., Zeng, Y., systems: Advances in Petri nets. In H. Manton, K. G., & Curtsinger, J. W. (1998). Weber (Ed.), The Petri Net Markup Biodemographic trajectories of longevity. Language (pp. 124–144). Heidelberg: Science, 280, 855–860. Springer-Verlag. von Dassow, G., Meir, E., Munro, E. M., & Westendorp, R. G. J., & Kirkwood, T. B. L. Odell, G. M. (2000). The segment polarity (1998). Human longevity at the cost of network is a robust developmental module. reproductive success. Nature, 396, 743–746. Nature, 406, 188–192. Williams, G. C. (1957). Pleiotropy, natural von Zglinicki, T., Pilger, R., & Sitte, N. selection, and the evolution of senescence. (2000). Accumulation of single-strand Evolution, 11.