What Is Aging?

Does aging start at conception, at birth, at the age of lowest mortality risk (around puberty), or only when the effects of aging become obvious to clinicians and pathologists? Is it a single process or a group of processes that occur in parallel throughout life? Does it occur in tissue culture, in individual cells, or only in whole organisms? Is the “senescence” that causes oak leaves to turn red and die a form of aging? When a person gives up smoking and chocolate nut sundaes and takes up broccoli and jogging, thereby adding 10 years of healthy life to middle age, has “aging” been reversed? Terms in colloquial use bring with them dangerous baggage when they sneak into scientific discussions, and gerontologists quickly discover that most of their honorable and respected colleagues use the term “aging” in different, idiosyncratic, misleading, tricky, indefensible ways. To simplify my presentation, I will start with a formal definition, one defended in more detail elsewhere (Miller, 1997a, 1999). When I talk about aging, I mean the process that progressively converts healthy young adults (whole mice or whole people, for example) into frailer adults with increased risks of death and disability. It is important to note that this process does not occur only in old age: although 90-year olds are more aged than 20-year olds, there is no reason to think that 90-year olds are aging any more rapidly than they did when they were themselves young adults. Studies of senior citizens are needed to tell us what aging does, but studies of young and middle-aged adults may be at least as informative about how aging creates old people from young ones. Indeed, contrary to some versions of intuition, studies of young adults are likely to tell us more about aging than studies of old individuals, because the former are less affected by two important confounders: (a) the effects of specific serious diseases that often afflict the elderly, and (b) differential survival, which removes individuals from their age cohort and does so nonrandomly with respect to age-dependent physiological effects.

Is There An “Aging Process” Per Se?

Or does the conversion of young adults to noticeably different, more vulnerable, old adults reflect the simultaneous, more-or-less well-synchronized progression of multiple processes, some of which alter muscles and others bone marrow, some of which operate within cells and others on extra-cellular material, some of which alter DNA and others membranes or proteins or cell numbers or circuitry? The latter theory, conveniently termed the “multiple-clocks” model, is by far the most popular among theoretically inclined gerontologists, from John Maynard Smith (1962) through Edward Masoro (1995) and many others. It is difficult to imagine how a single cellular or biochemical mechanism might lead to alterations in so many apparently disparate manifestations of aging. In my view, however, the notion that aging is indeed regulated by a single clock—or a very small collection of clocks—is supported by strong circumstantial evidence. The argument includes four elements:

• Caloric restriction, i.e., the imposition of diets containing about 40 percent fewer calories than a rodent would voluntarily consume, seems to retard nearly all aspects of aging, including changes in protein cross-linking, patterns of gene expression, loss of function in many organ systems, and risks of multiple forms of death (Weindruch and Walford, 1988). Restricted mice and rats stay active and vigorous at ages at which nearly all control animals have died (McCarter et al.,1997). This mass of evidence is very difficult to reconcile with models in which these aspects of aging are independently timed.
• Mutations at single genes can extend life span by about 50 percent in df/df and dw/dw dwarf mice (Brown-Borg et al., 1996; Miller, 1999), and by 200 percent or more in nematode worms (Kenyon et al., 1993). The dw/dw mice, at least, retain youthful levels of immune function into old age (Flurkey et al., unpublished results). Single gene changes that produce dramatic increases in longevity are again tough to reconcile with multiple-clock models of aging.
• Natural selection can rapidly generate subpopulations of mammals that live much longer than the founding population. The clearest example of this comes from Austad's study of an opossum population, stranded for a few thousand years on a relatively risk-free island. Selective pressures in this new ecological niche favored slow-aging genotypes, and the resulting population survives much longer than their mainland cousins. Initial mortality rates do not differ greatly between the two populations, suggesting that differences in predation pressure per se contribute relatively little to these survival statistics, but changes in the slope of the mortality risk curve support the idea that the island opossums age more slowly than their mainland cousins by a factor of nearly two. Furthermore, island opossums not only exhibit life-span extension, but also show decelerated changes in collagen cross-linking and reproductive performance (Austad, 1993). Under a multiple-clock system, production of a new, longer-lived species or subspecies would require multiple adjustments in genes that time each of the multiple systems affected by aging, and would have to do so in a population of animals whose other protective systems continue to decay at the original, rapid rate. The single-clock model, in contrast, does not depend upon a rare concatenation of low-impact mutations, but hypothesizes a fairly small number of genes in which mutation can alter multiple downstream processes in parallel. Natural selection has produced long-lived species and subspecies from shorter-lived ones dozens of times (e.g., turtles, tuna, elephants, bats, parrots, and chimps do not share a recent common ancestor!). Evolution of extended longevity seems to emerge over and over again from selective conditions that promote large body size, slowed development, and relatively small litter sizes.
• The single-clock model predicts that those individuals within a species who appear particularly “youthful” in one respect (e.g., immune function) are likely to appear relatively youthful in others (e.g., muscle function, collagen biochemistry, or mortality risk). Compelling evidence on this point is not yet available, but some recent data from our own laboratory, presented below, have begun to address this issue.

These two viewpoints—aging as a single process or as a collection of processes—are not difficult to combine into a unified perspective. The electrical network in a residence can be viewed as a set of circuits, one running the toaster, another running the radio, a third running the light bulbs, each of which is subject to its own regulatory pathways (switches, rheostats), each engineered and fixable by specialists who need know little about the other appliances beyond their likely demands on the current. Each of these parallel processes, however, can be influenced by a reduction in the available voltage in a summer heatwave. The obvious fact that many systems break down, in different ways, in old people and old rodents does not by itself disprove the hypothesis that these multiple endpoints of aging might be timed, in part, by a shared mechanism. The question is an empirical one, with the evidence not yet fully assembled.

Can Aging Be Measured?

Alas, not yet. Researchers who wish to test whether a particular diet, exercise program, drug, or gene alters aging often use age at death as an indirect measure of the effects of aging itself. In these situations—for example, in studies of caloric restriction (Yu et al., 1985)—they determine if the experimental condition extends the mean life span of a test population, with particular attention to the longevity of longest-lived individuals (the “maximum life span,” conveniently estimated as the mean life span of the longest-lived 5 or 10 percent of the group), and conclude that aging has been slowed if the proportion achieving very old age increases. A related measure, the time over which mortality rate doubles, provides an index of actuarial aging useful when population size is high enough (Finch et al., 1990). These approaches have some face validity; because the risk of many forms of lethal injury or illness increases so dramatically at advanced ages, interventions that influence only one cause of death have relatively small influence on overall mortality rates (Olshansky et al., 1990), and dramatic reduction in overall mortality at very old ages implies effects on many types of diseases—good grounds for concluding that aging itself has been retarded.

The actuarial approach to quantification of aging presents several problems, however. For one thing, age at death is influenced by multiple factors, among which biological aging will in some cases play an insignificant role. An individual unlucky enough to inherit the genes for cystic fibrosis, be infected by HIV, or hit by a bus may die at an age at which aging has yet to cause much increase in vulnerability. Although changes in the mean longevity of a test population may be altered by interventions with little effect on aging, the slope of the mortality curve (or the mortality rate doubling time) and the age at death of the final few percent of the population seem more likely to be useful indicators of interpopulation differences in aging rate. These indices may be less valuable for invertebrate species, in which large population studies have documented a small subset of extremely long-lived individuals among which the mortality rate may stabilize in very old age (Carey et al., 1992; Curtsinger et al., 1992). It is worth noting, however, that longevity rate deceleration in mammals, if it exists at all, is much less dramatic than in the insect populations in which it was first noted. In Mediterranean fruit flies, for example, one fly per million lives to an age that is fivefold higher than the age at which 90 percent of the cohort has died (Carey et al., 1992). A similar life table in humans would create approximately 300 U.S. citizens aged 400 years or older; this discrepancy is thus good news for the Social Security Administration, but bad news for those who seek to develop and test general principles about mammalian aging from demographic studies of invertebrates.

Assessment of Interindividual Differences in Aging Rate

Developing measures of aging that can assess interindividual differences is a tougher nut to crack, but in my view deserves a high priority among the issues facing research gerontologists. The concept of biomarkers has earned a useful niche in toxicology: measuring some effect of the toxin may be more convenient than trying to measure the integrated levels of the agent itself. Thus, for example, measures of glycated hemoglobin provide a valuable surrogate marker for exposure of cells to ill-regulated glucose levels, and measures of DNA damage may provide a useful way to judge exposure to specific mutagens (e.g., after Chernobyl) or to whole classes of mutagenic agents (e.g., in those living close to oil refineries).

Attempts to derive analogous indices of “biological age” have been cogently criticized on methodological grounds (Costa and McCrae, 1988, 1995), and indeed those who are convinced philosophically that there is no such thing as the biological age of an individual would argue that biomarker measurement has no likelihood of success and correspondingly little importance. But in my view—from the perspective that multiple aspects of aging are indeed likely to be coordinated in their timing by a single factor—developing a reliable way to measure interindividual differences in aging would have great importance in two respects: it would help to settle the debate as to whether aging can be treated as a unitary process, and it would provide a yardstick with which to measure the effects of genetic or environmental factors thought to have an effect on aging. Development of a replicable, validated set of tests that could discriminate between individuals aging at different rates would have the same effect in biogerontology as the effect of the sphygmomanometer on studies of hypertension or the invention of the clock on studies of celestial motion.

How Would We Know if We Had Successfully Validated One or More Biomarkers of Aging?

It is regrettably safe to agree that “it is too early to constitute a definitive panel of biomarkers for either animal models or humans” (Sprott, 1999:464). Proposed validation rules for candidate biomarkers vary widely among different experts, and sometimes between different formulations from the same experts. Some definitions (Lipman et al., 1999) propose that a biomarker of aging must be able to distinguish physiologically young from old, but such a formulation merely rewords the problem without offering an operational solution. I would propose an alternate definition, specifically, that a biomarker of aging should meet both of the following criteria simultaneously:

1. It should correlate well, among middle-aged adults, with a wide range of other age-sensitive traits in multiple domains, after statistical adjustment for the effects of calendar age per se; and
2. It should be a good predictor of life expectancy for individuals whose death is due to any of a wide range of lethal illnesses.

Each of these criteria deserves some amplification. A biomarker validation program would start with a list of candidate biomarkers, each known to be age-sensitive (by cross-sectional and/or longitudinal analyses) in adults. By hypothesis, some of these traits would reflect interindividual differences in the aging process, but each would also be sensitive to genetic and nongenetic factors that also vary among individuals, statistical “noise” that would interfere with the extraction of the “signal” attributable to aging itself. A correlation between age-sensitive immune parameters—for example, T-cell proliferation and T-cell cytokine production—would be relatively unhelpful in evaluating each of these parameters as potential biomarkers of aging, because the two assays are closely related and likely to be influenced by many factors unrelated to aging (e.g., recent infection, vaccination history, polymorphisms in immune system genes). However, a correlation between T-cell proliferation and, for example, muscle strength, or reflex speed, or lens protein cross-linking, or age at menopause, would be difficult to attribute to any obvious metabolic or pathophysiological mechanism other than linkage to some fundamental aging rate that might by hypothesis retard or accelerate changes in a wide range of age-sensitive traits.

Because each such candidate biomarker is already known to be age-sensitive, merely showing that pairs of these traits are mutually correlated would be unsurprising and uninteresting. However, correlations that remain strong after adjustment for the effects of calendar age would not be as easy to explain away and could provide good prima facie evidence that the components of the test battery measure, even if indirectly and imprecisely, some common factor that contributes to impairment of function in multiple cells and organ systems.

The second proposed criterion—correlation of the candidate biomarker with life expectancy regardless of the specific cause of death—is designed to link the assays to physiological decline. Loss of physiological reserve is not the only outcome of aging, but it is the outcome that causes most concern, and has furthermore been extensively studied in previous intervention studies. Loss of physiological reserve therefore serves, imperfectly, as the “gold standard” for most demonstrations of decelerated aging. An ability to predict subsequent longevity is not by itself sufficient to evaluate biomarker candidates, because risk factors (genetic or non-genetic) for specific forms of death would otherwise qualify: unstable angina, high titers of serum HIV, or a consistent history of cigarette or drug use are all predictors of subsequent longevity because of their association with specific causes of death, but these traits do not provide information about the kind of underlying aging processes that would, by hypothesis, increase risk factors for the wide range of late-life causes of death.

CD4 Memory T-Cells: A Candidate Biomarker in Aging Mice

In our own work we have attempted to test whether age-sensitive T-cell subsets might be useful biomarkers of aging in mice. There are several varieties of T-lymphocyte in the blood, spleen, and lymph nodes of mice (most of them with human counterparts), and their relative proportions change dramatically with age in adult animals and people. The CD4 lymphocytes—so called because they express the CD4 protein on their surface—act as “helper” cells in immune responses, assisting other cells in making antibodies and protective cytokines. The CD8 T-cells specialize in killing tumor cells and virus-infected cells. These are mutually exclusive classes: to a first approximation, all peripheral T-cells are either CD4+ or CD8+ but not both. The CD4 pool can be further subdivided into two mutually exclusive subsets: the naive CD4 cell (sometimes called the “virgin” CD4 cell, CD4V), and the memory CD4 cell (CD4M). Newborn CD4 cells, fresh from the thymus, are considered naive and can be identified by a specific pattern of surface proteins. Once stimulated by a virus or a bacterial antigen, they can produce clones of CD4M cells that are specific for the antigen in question and can help to protect against later encounters with the same or closely related pathogens. CD8 cells, too, can be divided into naive CD8 cells and the CD8M memory cells to which they give rise. Aging of both rodents and humans leads to a progressive increase in the proportion of CD4 cells in the CD4M subset, with a corresponding fall in the proportion of CD4 cells with the CD4V phenotype. The shift from CD4V to CD4M can be seen in blood and internal lymphoid tissues such as lymph nodes and spleen (Lerner et al., 1989), and is decelerated by calorically restricted diets that postpone aging and death (Miller, 1997b). The ability to measure CD4M levels in blood cells allows for repeated assessment of each individual mouse and could facilitate comparisons with human populations.

For these reasons, we assessed in several ways the possibility that CD4M levels could serve as a biomarker of aging. A group of genetically heterogeneous mice was used for these analyses, to diminish the possibility that our findings would later prove to be applicable only to a single genotype of inbred or F₁ hybrid mouse (Miller et al., 1999). Each mouse in a group of 46 males and 83 females was tested at 8 months of age, and the survivors (93 percent) tested again at 18 months of age. Each animal was then observed until it died or became severely ill (and was euthanized on humane grounds). Regression analysis was then used to examine the dependence of longevity on CD4M values, with gender as a covariate. The results (Figure 8-1) showed a strong association between high CD4M values at 18 months of age and lower longevity; after adjustment for gender, the CD4M effect had p(t) = 0.0003, and 18 percent of the variance in longevity could be accounted for by gender and CD4M alone (R² = 0.18). The direction of the association was consistent with theoretical expectations, in that high levels of CD4M cells are characteristic (in cross-sectional and longitudinal studies) of older mice, and in the biomarker analysis were seen to be associated with diminished longevity. Seven other age-sensitive T-cell subsets were also tested in this initial set of mice (Miller et al., 1997), and four of these were found in univariate analyses to be associated with differences in longevity, but only at marginal significance levels that could have resulted from chance in a series of multiple comparisons.

FIGURE 8-1

Correlation of mouse longevity with the percentage of CD4M cells measured at 18 months of age. The filled circles and darker line represent female mice, and the open circles and lighter line represent males. There is a significant correlation between (more...)

If CD4M levels are indeed useful as biomarkers of an underlying aging process, then their association with longevity should be equally apparent for a wide range of terminal diseases. Indeed, the strength of this association was found to be similar in subgroups of the mice dying of different causes, including fibrosarcoma, lymphoma, mammary carcinoma, and a miscellaneous group of other neoplasms (Miller et al., 1997); the association reached statistical significance in the two subgroups with the largest numbers of cases. The number of deaths attributable to non-neoplastic causes was too low (N = 13) to provide adequate statistical power, although in these mice, too, high CD4M levels were associated with poor survival.

A stronger test of the potential status of CD4M as a biomarker would be to determine whether mice with high levels of CD4M cells also displayed relatively extreme levels of age-dependent traits outside the immune system. A group of 59 genetically heterogeneous mice was tested for CD4M levels and also on several age-sensitive measures of muscle strength between 9 and 12 months of age, and then the surviving 42 mice were tested again at ages 20-22 months. Because muscle force generation declines with age, the hypothesis was that within each age group, force generation would be inversely proportional to CD4M levels. Figure 8-2 presents the results: mice with high CD4M levels did indeed, as predicted, tend to have significantly lower muscle strength at each of these ages (Faulkner et al., unpublished results). Similar tests conducted on these mice at intermediate ages (14-16 months), however, did not show this relationship (R = −0.06), contrary to hypothesis. Additional work is now in progress to see whether CD4M levels correlate, in these heterogeneous mouse populations, with age-sensitive measures of bone density, cataract formation, protein oxidation, and spontaneous activity, as well as tests of functional immunity.

FIGURE 8-2

Correlation of CD4M cell percentage with muscle force generation in mice. CD4M levels were measured in each mouse at 9 and (in survivors) at 20 months of age, and muscle force measured at 12 and 22 months. There is a significant inverse correlation at (more...)

Effects of Gender and Hormonal History

To see if associations between T-cell subsets and longevity might differ between males and females, we reanalyzed the data separately for animals of both sexes, and found one example of a sex-specific association. A subset of T-cells distinguished by high level expression of the surface membrane pump P-glycoprotein (P-gp) had previously been shown to increase with age in mice in both the CD4M and CD8M populations (Witkowski and Miller 1993), and interestingly the P-gp^hi subset of CD4M had been shown to be refractory to stimulation (Bining and Miller, 1997). The regression analysis showed (Figure 8-3) that high levels of CD4P cells—the subset characteristic of older mice—were indeed associated with shorter life expectancy, but only in male mice (Miller et al., 1997).

FIGURE 8-3

Correlation of mouse longevity with percentage of CD4P cells (i.e., proportion of CD4 cells with high levels of P-glycoprotein) measured at 18 months of age. Female mice (dark line, filled circles) show no association between CD4P levels and life expectancy, (more...)

The strongest associations in these initial studies had involved T-cell subsets measured on 18-month-old mice, i.e., mice that had already completed 70 percent of the median life span (approximately 26 months) of the population, but correlations of longevity and T-cells subsets tested in 8-month-old mice were of only marginal significance (p = 0.06). A second study, however, has produced promising results with mice tested at 8 months of age and provides another example of sex-specific associations. The test population included 121 virgin males, 135 virgin females, and 292 mated females, i.e., females that had been housed together with a stud male until 6 months of age. (The two virgin populations included the mice analyzed in our previous publication—see Figures 8-2 and 8-3—plus additional animals.) Of the 121 males, 33 died of a peculiar mouse urinary syndrome (MUS) involving bite wounds on the genitals, urethral obstruction, and rupture of the bladder (Tuffery, 1966), which is seen only in nondominant males housed with more aggressive males. This lesion, thought to be secondary to adjustments in dominance hierarchy, typically causes death at relatively early ages, and therefore mice dying of MUS are treated as a separate subgroup. None of the T-cell subsets tested at 8 months of age was able to predict subsequent longevity in the virgin males or virgin females, but there was a significant inverse correlation between CD8M cells and longevity in the mated females. Figure 8-4 shows the scatterplots for all four sets of mice. The correlation for mated females (R = −0.22, p < 0.001) is in the predicted direction, that is, with high levels of memory cells associated with lower life expectancy. There is no correlation in virgin females or in the virgin males dying of causes other than MUS. Males dying of MUS, similar to mated females, show an inverse correlation (R = −0.27, p = 0.13), which, however, is not statistically significant.

FIGURE 8-4

Correlations between CD8M cells at 8 months of age and life span in four groups of genetically heterogeneous mice. High levels of CD8M cells are associated with diminished longevity in mated females (left panel; p < 0.001), but not in virgin females (more...)

These data thus support the idea that tests of age-sensitive traits, measured at ages as early as the first third of the life span, may be able to predict subsequent longevity, but raise the concern that the associations may vary with gender and either hormonal exposure or reproductive history. Levels of CD4M and CD8M cells are strongly and positively correlated at all ages (R = 0.70, 0.65, and 0.40 at 8, 14, and 20 months, respectively, all p < 0.005) (Miller, 1997b), and there is no a priori reason to expect that the former subset would be associated with longevity only in virgin animals and the latter only in mated females. We have now initiated a number of collaborations to see if these subsets correlate in expected directions with indices of age-sensitive change in cells and tissues outside the immune system, as well as with life span and protective immune function in these heterogeneous mice.

Coordinated Genetic Influences on Life Span Regardless of Cause of Death

The QTL data also provide analytical tools for addressing hypotheses about the relationship of longevity to aging, disease, and age-sensitive changes in cell and organ function. In particular, we have looked to see whether genetic alleles associated with differences in longevity have equivalent effects in subgroups of mice dying of different lethal illnesses. We reasoned that if a specific allele or allele combination had an influence on aging itself, then it might have detectable effects on many different consequences of aging, including both cancer and non-neoplastic diseases. Figure 8-5 shows the results of such an analysis (Miller et al., unpublished results). Female mice that inherit the C3H allele at D2Mit58 plus the BALB allele at D16Mit182 live, on average, 173 days longer than their sisters with the DBA/2 plus C57BL/6 combination (pointwise p < 0.0001, adjusted p = 0.05). Of the 132 females for which both genotype and necropsy data were available, 97 died of one or another form of cancer, and among these the favorable allele combination was associated with a 180-day increment in mean life span (p < 0.0001) compared to the least favorable combination. Twenty-two of the mice died of congestive heart failure, inanition, amyloidosis, endometritis, or another non-neoplastic illness, and among these the favorable allele combination was associated with 165 days of additional life span. The bottom two bars in the figure document the effects of the C3H-plus-BALB alleles on the 39 mice dying of lymphoma (effect size 186 days, p = 0.02) or in the 24 mice dying of fibrosarcoma (effect size 277 days, p = 0.01).

FIGURE 8-5

Genetic regulation of longevity in mice stratified by cause of death. Female mice that inherit the C3H allele at D2Mit58 plus the BALB allele at D16Mit182 (light gray bars) have significantly higher longevity than their sisters (dark gray bars) with the (more...)

The data in Figure 8-5 thus show that genetic combinations that postpone death in mice dying of cancer also improve longevity, to a similar extent, in mice dying of non-neoplastic illnesses. In three other cases—D4Mit171 in males, D9Mit110 in males over 657 days of age, and D12Mit167 in males and females over 657 days—the favorable alleles had significant effects in mice dying of neoplastic causes and in those dying of non-neoplastic disease. We interpret these data as evidence that multiple forms of lethal illness can be regulated, in their age of incidence or rate of progression, by shared elements that are themselves under genetic control. If it can be shown that these loci, individually or in combination, also influence differences among mice in other age-sensitive traits—T-cell subsets, muscle strength, cataract progression, etc.—we will take this as evidence that they may affect longevity through an effect on the aging process per se.

The available dataset also provides examples in which genetic variants seem to influence the risk of specific late-life diseases. Figure 8-6, for example, shows longevity results for mice stratified by their inheritance at the 12th chromosome locus D12Mit167. This is a locus associated with differential longevity in both male and female mice, with the strongest effect (adjusted p < 0.01) seen in those mice living more than 657 days (Jackson et al., unpublished results). The longest-lived mice are those that inherit both the C57BL/6 allele from their mother and the C3H allele from their father; on average, they survive 93 days longer than siblings with the BALB plus C3H combination. Figure 8-6 shows that the D12Mit167, like the pair of loci illustrated in Figure 8-5, has significant and similar effects in mice dying of cancer (85 days) and in mice dying of non-neoplastic diseases (126 days). A more detailed analysis of the cancers, however, suggests that while lymphoma and hepatoma victims are equally protected by the favorable alleles (effect sizes of 93 and 167 days, respectively), the effect on mice dying of fibrosarcoma is negligible (10 days). A similar analysis (not shown) revealed that the D4Mit171 allele was associated with exceptional longevity in male mice dying of non-neoplastic diseases (177 days, p = 0.03) and of pulmonary adenocarcinoma (155 days, p = 0.03), but not in males dying of other forms of neoplasia (26 days).

FIGURE 8-6

Genetic regulation of longevity in mice stratified by cause of death. Male or female mice that inherit the C57BL/6 (maternal) and C3H (paternal) alleles at D12Mit167 (light gray bars) are longer lived than their siblings that inherit the BALB plus C3H (more...)

A last example illustrates two additional principles: (a) that genes that delay death for certain lethal illnesses can accelerate it for others; and (b) that risk of a condition and timing of a condition can be regulated by different loci. The top panel of Figure 8-7 depicts age at death in male mice of two subgroups: those dying of the urinary syndrome MUS, and those dying of all other causes. The genetic analysis contrasts mice with both the C57BL/6 allele at D4Mit84 and the C3H allele at D9Mit110 to mice with any of the three other allele combinations. In the males dying of causes other than MUS, this allele pair is associated with a 170-day increment in longevity (post-hoc p < 0.00003). But for males that do die of MUS, the same allele combination is associated with a 187-day decline in mean life span (post-hoc p < 0.03). This effect is thus pleiotropic, in that these alleles accelerate death in mice susceptible to MUS, while postponing death for all other males in the population. Although these loci are associated with differential longevity in mice that do develop MUS, they do not have a significant effect on the chances that MUS will indeed occur (not shown). The risk of developing MUS seems to be under control of a separate locus on chromosome 6. As shown in the bottom panel of Figure 8-7, males that inherit the C3H allele at D6Mit268 are far more likely to develop MUS (28 percent risk) than are their brothers who receive the DBA/2 allele at this locus (7 percent risk; p = 0.012 by two-tailed Fisher's exact test).

FIGURE 8-7

Separate genetic regulation of incidence and timing of a lethal urinary syndrome. The top panel shows that the C57BL/6 allele of D4Mit84 plus the C3H allele of D9Mit110 (black bars) are associated with a dramatic (170-day) postponement of death in male (more...)

It will be worthwhile, once larger numbers of mice have been examined, to determine whether these increases in longevity reflect changes in the rate at which mortality risks increase over the life span, differences in the age at which these risks begin to increase, or age-independent declines in risk without alteration in the slope of the risk/age curve. Large differences in longevity across species are sometimes due to alterations in mortality rate doubling time, but in other cases reflect alterations in initial mortality risks; Finch (1990), for example, notes that the threefold difference in longevity between rhesus monkeys and humans reflects a 100-fold difference in initial mortality rate without any detectable difference in mortality rate doubling time. It will be interesting to determine whether or not the set of genes segregating in crosses among laboratory-adapted inbred stocks includes loci with an influence on the initial mortality rate, the pace of increase in mortality risks, or both. Nonetheless, evidence of genetic loci that retard death regardless of the specific cause of death should provide useful tools for analysis of how aging coordinates multiple late-life processes, including illnesses.

Genetic Control of Age-Sensitive Traits

The preceding discussion has concentrated on age of death and cause of death, but the same strategies (and indeed the same animals) can be applied to questions about genetic regulation of other age-sensitive traits, including those that seem likely to serve as biomarkers of aging rate. Our own analyses of the genetic control of age-sensitive T-cell subsets provide an example. Blood samples taken from genotyped UM-HET3 mice at two ages (8 and 18 months) were tested for eight different T-cell subsets, including five that were known to be age-sensitive from cross-sectional and longitudinal data. Figure 8-8 illustrates two of the findings, adapted from the original paper (Jackson et al., 1999). The top panel shows an example of a QTL, linked to D12Mit34 on the 12th chromosome, that influences the proportion of CD4 cells that have the CD4M memory cell phenotype. Mice inheriting the BALB allele of D12Mit34 have higher proportions of CD4M cells; this differential is clear at 8 months of age (about 30 percent of the median longevity), and is equally apparent by age 18 months, an age at which the proportion of CD4M cells clearly shows the typical age-dependent increase. The effect is statistically significant with an experiment-wise p = 0.03. The D12Mit34-associated locus affects this age-sensitive trait, but exerts its influence in the first third of the life span. The bottom panel of Figure 8-8 shows an association between the DBA/2 allele at D9Mit110 and differences in expression of P-glycoprotein on CD4 cells; cells with high levels of P-glycoprotein expression (“CD4P”), shown in other work to be age-sensitive (Witkowski and Miller, 1993) and nonfunctional (Bining and Miller, 1997), accumulate to higher levels in mice inheriting the C3H allele at D9Mit110 (experiment-wise p = 0.02). In contrast to the D12Mit34 effect on CD4M cells just discussed, the D9Mit110 effect is not apparent in mice tested at 8 months of age, but is instead associated with a relatively rapid accumulation of CD4P cells between 8 and 18 months of age. Of the eight significant gene/T-cell phenotype associations documented in the original pilot study (Jackson et al., 1999), three (including D9Mit110) were detectable only in 18-month-old mice; two affected 8-month-old mice but had no influence on the subsets at 18 months of age; two (including D12Mit34) had effects seen in mice of both ages, and one other had a more complex, crossover effect. It is clear that longitudinal datasets, in which phenotypes are assessed periodically over the entire life span of individual genotyped subjects, have great potential for sorting out the ways in which genetic variations might influence health through continuous, delayed, or transient effects.

FIGURE 8-8

Genetic influences on age-sensitive T-cell subsets in mice: two representative examples. The top panel shows the association between the BALB allele at D12Mit34 and relatively high levels of CD4M cells; the association is apparent in mice as young as (more...)

It is worth stressing that this initial study involved analysis of only 145 mice, and thus had very low statistical power for detecting genetic effects that controlled less than 25 percent or so of the phenotypic variance. Similar analyses are now underway in larger samples of the same genetic constitution and should reveal additional examples of gene/subset associations. The study in progress includes tests of a much wider range of age-sensitive traits, and one outcome should be a preliminary list of QTLs that affect traits that are of interest in their own right: QTLs for collagen modifications, for cross-linking of eye lens proteins, QTLs with an effect on bone density, antibody production, spontaneous activity, glucocorticoid levels, etc. The mapping project should thus help to guide the search for human genes that regulate these interesting phenotypes and at the same time spark new investigations, in animal models, for the biochemical differences that mediate the genetic effects we detect.

At the same time, the dataset that emerges should also allow us to test more general questions about the nature of aging and its genetic control. We may, for example, be able to identify QTLs that not only retard the development of one or more age-sensitive T-cell subsets, but also retard age-dependent changes in protein conformation, bone matrix turnover, and brain GFAP levels. Such a finding would imply that these changes are influenced, together, by a common biochemical pathway, and the corresponding QTLs would be excellent candidates for genes that regulate aging per se, rather than merely one among the many more age-sensitive traits. In the same way, it will be of particular interest to determine if QTLs that regulate age-sensitive traits also are associated with differences in life span, and conversely if QTLs identified on the basis of longevity effects modify one (or nearly all?) of the age-sensitive traits in our test battery.

Mice selected on the basis of longevity QTLs may be of particular value for testing hypotheses about the role of specific biochemical and cellular mechanisms in aging and disease. It should be possible to identify mice that inherit combinations of genetic alleles associated with optimal longevity, and to do so within a few weeks of birth. Mice with this particular combination of alleles, together with randomly chosen sibling controls, could then be tested, at any age, for factors—telomere length, fibroblast growth patterns, levels of protein oxidation, immunological vigor—hypothesized to play a role in controlling age-sensitive physiological changes or disease risks. By providing a way to identify siblings destined for longer or shorter lives, the gene-selected mouse populations could provide an approach to sorting the long list of potential aging mechanisms into those that do or do not show the expected association with longevity.