| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||
1 Laboratory of Genetics, University of Wisconsin, Madison.
2 Institute for Behavioral Genetics and Department of Integrative Physiology, University of Colorado at Boulder.
Address correspondence to James F. Crow, Laboratory of Genetics, 445 Henry Mall, University of Wisconsin, Madison, WI 53706. E-mail: jfcrow{at}wisc.edu
"Do not do unto others as you would that they should do unto you. Their tastes may not be the same." G. B. Shaw
AS SO neatly epitomized by Shaw, an environment that is good for one person may not be so for the next one. Genotype–environment interaction (G x E) is a recurring theme throughout this special issue. The subject is in the air. Right now, the National Institutes of Health are considering launching an enormous population-based study to involve genotypic and environmental measures and their interactions. Of course, the subject is not new. We give violin lessons to musically talented children, realizing that they will put to good use training that would be wasted on the tone-deaf. Phenylketonuria (PKU) is an inherited metabolic disease in which the amino acid phenylalanine accumulates to harmful levels, leading, among other things, to mental retardation. For several decades, many states have tested newborns for PKU and withheld phenylalanine from their diet, thus permitting normal mental development. The number of other genetic traits that can be identified early and cured or ameliorated is now a dozen or more and growing (Khoury, McCabe, & McCabe, 2003
). Some of the best examples of interaction come from pharmacogenetics. One example is the drug Iressa (gefitinib; AstraZeneca Pharaceuticals, Wilmington, DE), which inhibits a normal growth factor and is highly effective as treatment for lung cancer in the small minority who carry a gene (more precisely, an allele, one of the alternative states of the gene) for the right protein, but not for others (Marx, 2004). The long-term goal is individualized medicine and appropriately chosen environmental manipulations. Despite promising beginnings, the wide-scale realization is for the future, but such treatments are on the way. So, it is in this light that a workshop on G x E interaction in aging was convened and is emphasized in these articles. Striking in the accompanying articles is the absence of very much from the social sciences about genetically informative studies on aging, the ostensible purpose of the workshop. This is not the fault of the authors; rather, it is a reflection of the current state of the science. Several of the authors, while discussing genetic aspects of other traits, allude to the possibility of using similar techniques to study aging, and, indeed, this is a promising beginning.
The idea that genetics affects longevity is not new. Seventy years ago, Raymond Pearl (Pearl and Pearl, 1934) showed that long-lived persons tended to have long-lived ancestors, and many others have also been interested in the heritability of longevity, which is the subject of recent genetic studies (e.g., Longo & Finch, 2003; Tatar, Bartke, & Antebi, 2003). Although the Pearls could not distinguish genetic from inherited environmental influences, they clearly established a strong intergenerational correlation. More revealing was another early study on the longevity of twins (Kallmann & Sander, 1948). These authors studied 933 twin pairs, 237 one egg (identical) and 548 two egg (fraternal), for which there were suitable birth and death records. The mean difference in age at death was 36.9 months for the former and 78.3 months for the latter, clearly suggesting genetic influences. Although these works are mainly of historical interest, they show that today's subject has been with us for a good part of a century. The articles in this special issue are intended to reflect the current status of studies on environmental influences on aging; however, aging encompasses much more than longevity alone, and we will take up the issue of longevity again later. Surely with powerful molecular techniques, computers, and better record-keeping systems, we can now do better than these early pioneers.
The following articles are sequenced roughly in the order of attention to aging and genetics. The first report (Ryff & Singer) emphasizes the effect of positive social relationships in preventing or delaying disease onset and aging. The authors point out many instances in which a trait does not develop despite an individual's having the requisite genotype, that is, the trait is incompletely penetrant. Especially relevant to aging is a striking G x E interaction between the allele APOE4 and head injury in the development of Alzheimer's disease: no increase with head injury alone, a twofold increase with APOE4 alone, and a 10-fold increase with the combination. Another striking and promising observation is that APOE alleles have less influence on Alzheimer's disease after age 85; this is not because very old individuals are somehow protected, but rather that everyone who is susceptible has already been affected by Alzheimer's disease and only those who are somehow resistant are left in the population. The authors end with a plea for more research on centenarians.
Deater-Deckard and Mayr emphasize "executive control," those functions that are responsible for organizing basic processes and resources in a goal-directed manner. They point out the difference between shared and nonshared environment and the greater heritability with age. They note the deficiencies of cross-sectional measurements and call for more longitudinal studies. They believe that research should move beyond descriptive heritability studies and aim for a precise specification of G x E mechanisms as well as their interactions; this will prove to be an enormous task. These authors also introduce an interesting distinction between passive and nonpassive environments. The first might be, for example, the influence of a relative. The second occurs when individuals select their environments. Although the data come mainly from other traits, the authors apply them to thinking about aging.
McClintock and colleagues discuss mainly cancer. Their research work is on rats, in which they find that social isolation and hypervigilance increase the incidence of mammary cancers, accelerate aging, and shorten life span. They believe that the rat studies are telling us something about ourselves and are consistent with human findings. They suggest that differences in social isolation may be an important factor in the different rates of breast cancer in Caucasians and African Americans. Again, G x E interaction is emphasized.
Johnson and Krueger note that high income promotes better health yet does not seem to be a major cause. They studied a sample of 719 same-sexed twins from the Midlife Development in the United States (MIDUS) Study, applying Wright's method of path analysis in an attempt to measure the magnitude of different causal paths. They produce a much simpler interpretation by using latent variables. The study could be criticized for using only income as a measure of socioeconomic status, for using data from self-assessment, and for not including cognitive skills. But a sample of 719 like-sexed twins is a gold mine, even if one could suggest other measurements that might have been made. It is possible that some of the seeming paradoxes in their analysis might be eliminated if intelligence were a co-variable. The MIDUS Study promises to include this in the future.
Grigorenko points out the inherent complexities of G x E interactions. The objective is "phenotypic prevention," that is, combining genetic screening with environmental information. She emphasizes phenotypic plasticity and reaction norm (different phenotypes in different environments) and gives a number of examples from the medical literature. She notes that many of the best examples of G x E interaction come from pharmacogenetics. She points out the necessity for proper attention to statistical power and appropriate statistical analysis. Many traits may depend on higher-order interactions, but with current uncertainties, we may have to be content with main effects and first-order interactions. There is no end to complications as more factors are taken into account.
Shanahan and Hofer ask how social context affects gene expression. They note that there are 17 quantitative trait loci (QTLs; chromosomal loci affecting quantitative traits) known to affect Drosophila life span, and they doubt that the human is any simpler. So far, however, searches for specific genes affecting common human behavior have not been very successful, but these are just beginning. These authors, too, emphasize the need for longitudinal studies. They also believe that because of correlations and interactions, it is important not to study context features in isolation. These authors missed the numerous studies on nematode aging, including the first QTL studies in model systems (Johnson & Wood, 1982; Johnson et al., 2002b).
These articles, although representing only modest progress in elucidating genetic and environmental factors in aging and their interactions, point the way to many more and much better studies in the future. Most of the discussion dealt with traits other than aging, with the expectation that the methods used could be applied to aging research, a much more difficult problem. We shall have to wait and see how good a prediction this is. These reports raise a number of questions, and we now discuss some of them.
HERITABILITY AND TWINS
Historically, there has been great emphasis on heritability studies, especially those involving twins. Partitioning the variance (mean squared deviation from the mean) into genetic and environmental components is the first step toward an understanding. Heritability is the fraction of the measured variance that is attributable to genetic differences. Monozygotic (one-egg) and dizygotic (two-egg) twins would seem to be the ideal research subjects. If one assumes that the environmental differences are the same for the two types of twins, the fact that identical twins share 100% of their genetic differences whereas dizygotic twins share 50% permits an estimate of genetic influence. The environmental variance is easily subdivided into shared (between family) and nonshared (within family) components. One of the early surprises was the finding that, except for very young ages, shared environment accounts for only a small fraction of the total variance. This has held true for a variety of cognitive and behavioral measures. A remarkable feature of heritability analysis is that it provides an estimate of an environmental influence without our having to know anything about what this effect is. But the flip side is a weakness, for the studies offer no hint as to where to look for the environmental factors. Discovering them calls for other procedures and insights. Of course, there are a number of necessary simplifying assumptions, and it is a challenge to develop more robust methods.
There are other difficulties. With a conventional variance analysis, accidents of development and errors of measurement are lumped with environment. Several writers have suggested that we should have three components: heredity, environment, and chance (Finch & Kirkwood, 2000). That stochastic effects occur at the most basic level is vividly illustrated by recent yeast studies (Raser & O'Shea, 2004) and by studies in nematodes (S. Rea, D. Wu, J. W. Vaupel, & T. E. Johnson, unpublished data). To the extent that these are important, heritability studies overestimate the environmental contribution. In contrast, environmental correlations between identical twins cause an opposite bias. Furthermore, there has often been some doubt as to the correct identification of the two types of twins. Molecular methods can remove this difficulty.
Another problem with twin studies is that a substantial fraction of identical twins develop in the same chorion or even the same amnion. These twins usually have cross-connections in their blood supply, and usually these lead to inequalities in the blood distribution. Such twins can be considerably different, as is strikingly apparent in most Siamese twins. A substantial improvement in the accuracy of twin studies and probably an increase in estimated heritability would be found if only twins with separate chorions were used. Then their intrauterine environments would be fully comparable with those of dizygotic twins. The importance of this distinction was made more than a half-century ago (Price, 1950), but that work seems to have had a negligible influence on subsequent research. There is considerable variability as to whether information on extraembryonic membranes is included in hospital records of twin births, but if it were feasible to separate monochorionic from dichorionic twins, this would substantially improve twin studies.
Animal and plant breeders distinguish between heritability in the broad sense and in the narrow sense. The broad-sense heritability includes variance components due to dominance and gene interactions (epistasis). Narrow-sense heritability is that part of the genetic variance that is transmitted to the offspring, so dominance and most of epistasis are not included. These are important distinctions for breeders, but for human studies, we are interested mainly in the broad-sense heritability, which is what the twin studies show.
A greater difficulty, calling for considerable statistical expertise and invention, is disentangling G x E covariance and interaction. G x E covariance arises when genotypes are not allotted to environments randomly. As humans, to a considerable extent, choose their environments or have environments chosen for them, this means that for traits that are at all complicated, there will be G x E covariance, and this may be substantial. In experimental organisms, covariance is easily controlled by randomization, but human studies do not have that luxury. G x E interactions, in contrast, are caused by the differential response to the environment of different genotypes. As we have already said, this is a subject of great interest now, but in all but the simplest cases, distinguishing between covariance and interaction will not be easy.
For research purposes, it would be better if the twins could be raised in randomized environments. This is impossible, of course, but studying twins who were separated in infancy and reared in different homes offers an approach. So does the study of unrelated children adopted into the same family. All of these studies suffer from a restricted environmental range, as adopting parents tend to come from the upper socioeconomic groups. A strong argument that this may make a large difference is offered by a recent study of heritability of intelligence (Turkheimer, Haley, Waldron, D'Onofrio, & Gottesman, 2003). These workers did a large twin study with a broader range of socioeconomic environments than most previous studies. A standard twin analysis showed a heritability of 0.75 in the upper socioeconomic status group and only 0.10 in the lower. Even though the cautions that we have raised throw some doubt on these absolute values, there can be little doubt that the difference is large. This needs to be confirmed, but, assuming this, there is a need for a revision in popular thinking about heritability and the recognition that heritability estimates apply only to the population being studied. Also suggestive of future research is a twin study (Kim-Cohen, Moffitt, Caspi, & Taylor, 2004) showing the importance of social warmth and stimulating activities in producing resilience to socioeconomic status deprivation. Heritability analysis may well be the beginning, not the end, of a genetic–environmental analysis. It could point the way to more specific studies to identify specific factors, both genetic and environmental.
One other aspect of twins deserves much more exploration. Discordant pairs of identical twins must surely be very promising research material, as suggested by several authors, including Ryff and Singer in this special issue. Clearly, these twins have the same genetic potential, so any differences are environmental (or chance). A thorough investigation of the history of older twins that differ, for example, in affective disorder might supply leads to possible environmental interventions. An examination of all aspects of the lives of twins whose longevity differed has so far failed to be very revealing owing to the effects of so many variables differentiating two twins. Some of the articles in this volume have argued that traditional heritability studies have outgrown their usefulness. This may well be true, especially for human aging, where appropriate data are hard to come by. Indeed, in this era after the human genome project, heritability studies are most often used to show that heritability accounts (usually) for a smaller fraction of the trait variance than does the environment. In other words, genes are not destiny.
RELIABILITY THEORY AND HETEROGENEITY
Gavrilov and Gavrilova (2003) have argued that reliability theory provides a useful way to approach the aging problem. They point out that the high 
coefficient in the Gompertz equation argues for a great deal of redundancy in developmental systems. Strong evidence for redundancy is offered by the number of instances in which "knockouts" of presumed essential genes cause little or no overt effect. This theory predicts not only a rise of death rate with age, but also a plateau—and possibly even an absolute reduction—in death rate in the very old. This is consistent with viewing aging as the effect of accumulation of random damage in redundant systems. The plateauing in very old individuals is also noted by Ryff and Singer; this phenomenon can be accounted for by "heterogeneity of frailty" (Yashin et al., 1999) and can also account for plateaus in survival and multiple other phenomena associated with the aging process, such as the aforementioned reduced risk of Alzheimer's disease from APOE4 late in life. We are not all the same.
FEW OR MANY FACTORS?
Immediately after the rediscovery of Mendel's laws, there was a tendency to explain all sorts of traits by simple Mendelian theory. It soon became evident, however, that most traits, especially quantitative ones, were more easily accounted for by assuming a large number of more or less independent factors. The necessary theory was developed, mainly by R. A. Fisher and Sewall Wright. So the genetics world, at least the animal breeding world, became obsessed with analyzing quantitative traits by variance and path analysis. The number of factors was treated as effectively infinite, and it was regarded as not promising to try to isolate single genes.
It is very unlikely, however, that among multiple factors, their effects will be equal. It is much more likely that there is an exponential-like distribution, with one or more factors being much more important than the rest. This has led to QTL mapping. It is now feasible to pick out those loci that are making the largest contributions to quantitative measures. What was previously hopeless is now reasonable because of the abundance of molecular markers, making linkage studies feasible. QTL methods are only beginning to be applied to genetic studies of human aging (e.g., Puca et al., 2001). This is a heroic undertaking because studies of the very old are difficult. We can expect many similar studies in the future. An alternative to QTL mapping in humans is association studies, many of which have been performed on human life span but not on many other aspects of human aging. Numerous positive associations between target genes and longevity have been reported in the literature (Tan et al., 2004). Association studies are extremely difficult to control because of the need to avoid stratification in the ascertainment of proper controls. If controls are not taken from the same genetic population, then any differences in the frequency of alleles between the controls and the affected (which is the predicted outcome if there is a causal association between the disease and the gene) could easily stem from this difference in allele frequency between control and experimental groups. This is very difficult to control, but several methods for controlling this in human aging research have been suggested (Tan et al., 2003, 2004; Yashin et al., 1999).
More promising for the present is the identification of individual loci with major effects. Whether these constitute a large fraction of genetic variability is questionable. But they are the easiest to study, so this is the place to begin. We have already mentioned examples from clinical studies. Another is found in breast cancer, where tamoxifen is more effective in patients with the BRCA2 causative allele than those with BRCA1. Animal studies have found numerous examples of single-gene behavioral traits. For example, pair bonding in voles, foraging in honeybees, and care of offspring in rats have all been shown to depend on single loci, whose effects could be isolated and studied in molecular detail (Robinson, 2004).
Will such factors be found for aging? More than 20 years ago, single-gene mutations that could almost double the animal's life span were discovered in roundworms (Caenorhabditis elegans) (Johnson, 1990; Klass, 1983). More recently, both fruit flies (Drosophila melanogaster) and mice have also been shown to carry these so-called "gerontogenes" (Longo and Finch, 2003). An increase in longevity caused by a mutant allele was put forth many years ago as being a "gold standard" for identifying genes that clearly could be implicated in aging. (About 70% of the genes in roundworms shorten life span when eliminated.) About 1 mutant in 50 in C. elegans increases longevity. How can we explain this? One possibility is that there has actually been selection in the past for genes that keep the life span short. There are very good arguments against this possibility that show that aging cannot be selected for because, as a special case of altruism, individuals showing such a rapid aging phenotype would be selected against (Charlesworth, 1980). Another possible clue comes from hybrid luxuriance. An extreme example is Raphanobrassica, the hybrid between radish (Raphanus) and cabbage (Brassica), with the chromosome number doubled in the hybrid, produced by the pioneer Russian geneticist (and early Stalin victim) Karpenchenko. The size was enormous; one plant not only filled a greenhouse but protruded from a skylight. One can only assume that there were factors controlling growth in the parent strains that were suppressed in the hybrid. Could this loss of control, by mutation of controlling genes, possibly be the explanation in insects and nematodes?
The human story may well be different. Human evolution since our separation from the chimpanzee line has been toward a lengthening of the reproductive age and a concomitant change in longevity. There may even have been selection for survival of grandparents, as has been suggested many times. This would argue that most mutations would reduce the life span. All sorts of life-shortening diseases fit this pattern. Will future scientists identify genes causing large increases in life span, such as in flies and nematodes? Only the future will tell.
The one thing we know that certainly increases life span, at least in insects, rodents, and probably monkeys, is dietary restriction. There is not simply an increase in the age of death but a prolongation of active life. There are clearly genes that affect appetite, so these are indirectly affecting life span. Mutations in any one of several genes in C. elegans cause difficulty in eating, thereby increasing longevity. Surely, for caloric restriction, environmental intervention is promising. It has proven remarkably difficult to apply the research from rodents to humans because of the lack of control of food intake, so that even after 40 years of intense research, we really do not know for sure that dietary restriction works in humans. There is a huge research effort to identify appetite suppressors, not only for increased longevity per se but for other problems associated with obesity. Another area of research is to find key molecules that can stimulate or mimic dietary restriction in humans without necessitating actual food reduction. But all this assumes that we are like flies and mice, which may or may not be true.
AGED INDIVIDUALS AS RESEARCH MATERIAL
There could well be more research directly on aged people. For example, statin drugs reduce Alzheimer's disease in some people. A number of experiments on diet and social environment are already under way, and much more could be done. Older people typically have reduced muscle mass. Steroids or, much better, growth hormone, can alleviate this, but unwanted side effects argue against general use. There may also be an unwanted biological trade-off, because the mouse that holds the record for longevity had a gene making it deficient in growth hormone (Harder, 2004). Also, it has long been known that small dogs live longer than large ones. But this consideration is not likely to be important for treatments that start at age 70.
Could we be more adventurous in research on older people? An 80 year old is not likely to worry much about a carcinogen that has a 25-year latent period, so expensive and time-consuming tests of carcinogenicity could be dispensed with for the elderly. Many older people would gladly put up with uncertain or delayed side effects in exchange for a definite lifestyle improvement.
We have no illusion that such a research program would be free of social, legal, and political consequences. It may be that treating elders as different from the rest of the population will turn out to be politically unacceptable. Yet, it seems possible that a different view of risk could well lead to ways to improve the lifestyle of the aged.
Acknowledgments
The Editors acknowledge and highlight the special help of Michael Stallings, PhD, who reviewed all of the submitted manuscripts and made detailed suggestions regarding editorial content. T. E. J. is supported by grants from the National Institutes of Health.
References
This article has been cited by other articles:
 |
|
 |
|
  H. Kendig, P. A. Dykstra, R. I. van Gaalen, and T. Melkas Health of Aging Parents and Childless Individuals Journal of Family Issues, November 1, 2007; 28(11): 1457 - 1486. [Abstract] [PDF]
|
|
|
|
 |
|
 L. A. Wray, D. F. Alwin, R. J. McCammon, T. Manning, and L. E. Best Social status, risky health behaviors, and diabetes in middle-aged and older adults. J. Gerontol. B. Psychol. Sci. Soc. Sci., November 1, 2006; 61(6): S290 - S298. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. L. Angel and R. J. Angel Minority Group Status and Healthful Aging: Social Structure Still Matters Am J Public Health, July 1, 2006; 96(7): 1152 - 1159. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
