Genes as gerontological variables: uniform genotypes☆

Neurobiology of Aging 20 (1999) 95–104

Genes as gerontological variables: uniform genotypes夞 Gerald E. McClearn*, Scott M. Hofer Center for Developmental and Health Genetics, The Pennsylvania State University, 101 Amy Gardner House, University Park, PA 16802, USA Received 23 October 1998; received in revised form 27 April 1999; accepted 23 June 1999

Abstract Genetic conceptualizations and procedures have become integral to the conduct of research across the spectrum of life sciences, including gerontology, even when genetics is not the focus of inquiry. Among the research tools thus provided, one of the most basic is that of inbred strains. A close approximation to genetic uniformity is achieved by a sufficient number of successive generations of matings of relatives, and, once this near-uniformity is attained, the members of an inbred strain constitute a reference group relatively stable over time and available to diverse investigators. Different inbred strains possess different genotypes, so that numerous distinctive reference groups are available. The stability of these groups enhances prospects of replication-testing, and makes possible the focused accumulation of pertinent data. Phenotypic differences among strains identify particular groups that can be most appropriate for particular subsequent research objectives (and also provide ipso facto evidence of genetic influence on the phenotype). The very substantial advantages of the uniform genotypes provided by inbred strains (and by their F1 offspring) are purchased at the cost of limited generalizability of results and constraints on assessment of co-variation among variables. Uniform genotypes are, thus, not a tool for all purposes but must be seen as a powerful basic tool within an abundant genetic tool-kit. Particular research purposes will require use of more than one tool from the kit. © 1999 Elsevier Science Inc. All rights reserved. Keywords: Inbred strains; Animal models; Aging

1. Introduction As genetics has assumed an increasingly central position in the biological sciences, its principles and methods have become integral features of the research enterprise of most biological disciplines. Of particular relevance to the present theme, substantial bodies of empirical data demonstrate genetic influence on individual differences in a wide variety of phenotypes both in aging and in nervous system structure and function. As one consequence, animal model research in each area is relying increasingly on genetic specification of animal subjects, as are investigations at the interface of the neurobiology of aging. Genetic specification can take a variety of forms, and the animal groups so designated constitute research tools with 夞 The cited examples from the authors’ own work were supported variously by the MacArthur Foundation Research Network on Successful Aging, NIA grants AG04948 and AG09333, and by funds from the Center for Developmental and Health Genetics and the Penn State Agricultural Experiment Station. Other cited examples are from research supported by the Biomarker Initiative of the National Institute on Aging. * Corresponding author. Tel.: ⫹1-814-865-1717; fax: ⫹1-814-8634768.

different strengths and limitations [25,26]. This paper will be concerned primarily with the logic underlying use of one of the basic types of genetic specification—that of uniform genotypes. Application of this logic will be illustrated by results on behavioral biomarkers of aging.

2. The basic quantitative genetic model The phenotypes of central interest to neurosciences and to gerontology generally constitute complex systems and processes. The body of genetic theory that is particularly apposite to such complex systems is referred to as quantitative genetics. In this theoretical perspective, two broad, general domains of influence—the genetic and the environmental—are identified, and the methodologies pertinent to, and derived from, the theory seek to estimate the proportions of the measured phenotypic variance attributable to these domains, and to various sub-domains within them. The basic element in quantitative genetic theory is the genetic locus, understood to be the chromosomal location of a gene, and the basic premise is that a number of such genetic loci may influence the phenotype of interest. At each

0197-4580/99/$ – see front matter © 1999 Elsevier Science Inc. All rights reserved. PII: S 0 1 9 7 - 4 5 8 0 ( 9 9 ) 0 0 0 3 0 - 5

96

G.E. McClearn, S.M. Hofer / Neurobiology of Aging 20 (1999) 95–104

locus, there may be more than one alternative form, or allele, of the genetic material, and these different alleles may have different phenotypic consequences. For a given single locus, if one allele is symbolized by A and the other by a, then there are three possible genotypes (aa, Aa, and AA) that can be derived from the union of two parental gametes. Those genotypes with two copies of the same allele (aa and AA) are described as homozygous and that with different alleles (Aa) is termed heterozygous. It is possible to characterize these genotypes in terms of “allelic dosage,” with, for example, 0, 1, and 2 describing the number of A alleles present. In naturally occurring populations of sexually reproducing species, there may be many more than two alternative alleles at each locus. As will be shown, however, in experimental contexts the two-allele situation is readily attained, and the analyses accordingly simplified. These concepts can be defined in contemporary molecular terms, of course. Thus, we may consider a genetic locus to be a region of a chromosome containing the DNA that constitutes a functional gene, and an allele as one particular configuration of the base pairs of that particular stretch of DNA. A variety of configurations is possible, and these different alleles may result in altered protein products, which may affect processes in the biochemical pathways in which they are involved. For the most part, until quite recently, the polygenes of complex phenotypes have been anonymous, with neither their chromosomal locations nor their molecular configurations known. The elaboration of quantitative genetic theory, beginning early in this century, necessarily proceeded largely by treating the loci and alleles simply as abstract entities. In developing the relevant statistics, it was convenient to assume equal effect sizes of the many polygenic loci affecting the phenotype. It has long been appreciated that the impact of these loci might, in reality, vary considerably, but empirical studies on the issue have been arduous. Advances in gene mapping have now made possible the individuation of those polygenes (called quantitative trait loci or QTLs) with the largest effect sizes. One of the most exciting present prospects of genetic research is the localization and molecular characterization of these individuated loci of complex genetic systems. Among the varied tools available for exploiting these new opportunities are groups with uniform genotypes, which have displayed immense utility in a variety of other contexts.

3. Uniform genotypes Great interest has been generated by the recent spectacular feats of cloning that have generated animals of identical genotypes both in sheep and in mice. Future research prospects utilizing these techniques are dazzling, but they are not entirely without precedent. Identical genotypes have long featured in human genetic research employing

monozygotic human twins, for example, and identical genotypes (or, at least, a very close approximation) in the form of inbred strains have played a major role in animal model research in a wide variety of biomedical research domains. In the area of aging research, for a pertinent illustration, a recent Medline search of the terms “inbred strains” and “aging” for the period 1966 to 1998 produced 8419 reference hits. 3.1. Inbreeding Technically, inbreeding is the mating of individuals that are more closely related than individuals randomly chosen from the population. In laboratory application, the very close inbreeding regimen of sibling mating is usually employed. Siblings are more likely to be alike genetically at any particular locus than will be randomly chosen mates. Therefore, the chance of offspring from a sibling mating becoming homozygous for a particular allelic state is considerably enhanced relative to that of offspring from randomly mated parents. When succeeding generations are each derived from a single sibling mating pair, the chances of homozygosity at any one locus, and the proportion of loci that have attained a homozygous state, will increase regularly and approach 1.0 asymptotically. (To provide a sufficient number of animals for research purposes, many matings among strain members may be made to generate “production” animals; the nucleus lineage, however, always is through a single pair.) After approximately 20 consecutive generations, all animals within a strain are (nearly) genetically identical. (Males will differ from females, of course, in having one Y and only one X chromosome, but all females will be alike, all males will be alike, and the allelic constitution of the X chromosome will be the same, whether one or two are present.) Obviously, once homozygosity is achieved at a locus, that locus is “locked up,” and, except for the rare event of mutation, will remain stable across generations. 3.2. Representativeness of inbred genotypes Strictly defined, inbreeding is not directional, and an inbred strain is not generated for any particular phenotypic outcome; it simply becomes genetically homogeneous as a consequence of the mating scheme. The particular genotypic configuration of any inbred strain is a consequence of stochastic processes associated with the segregation of chromosomes during the production of eggs and sperm, and the phenomenon of crossing-over which shuffles genetic material between members of chromosome pairs. As a first approximation, and excepting loci that are closely linked on the same chromosome, these processes can be regarded as operating independently for the different loci. It is thus tempting to regard an inbred strain to be an assembly of loci randomly drawn from the mouse gene pool and fixed in homozygous configuration. However, this view must be

G.E. McClearn, S.M. Hofer / Neurobiology of Aging 20 (1999) 95–104

strongly qualified. For example, natural selection plays a strong role during inbreeding, eliminating many incipient strains. It is often the case that homozygosity for one allele results in a reduction of fitness relative to homozygosity for another allele at the same locus. The number of loci becoming fixed in a detrimental allelic state will differ, of course, from strain to strain during the inbreeding process. Although the reduction in fitness arising from a single locus may be slight, many incipient strains will become homozygous for deleterious alleles at a sufficient number of loci that they are eliminated through infertility, lack of viability of offspring, parental neglect, or a variety of other causes. Indeed, even those inbred strains that survive are characterized by some level of “inbreeding depression.” The development of inbred strains of mammals has been most energetically pursued in mice and rats. In each of these species, a large number of strains exists, and strain differences have been documented in a huge range of phenotypes. This literature obviously cannot be summarized here. Among the many studies utilizing inbred strains in the gerontological context, the C57BL/6 and DBA/2 mouse strains and the F344 and Brown Norway rats have been particularly widely employed (e.g., [6,7,9,10,11,12,14,23]). The examples used in this paper will refer to these groups, among others. It is important to note that the extant strains were not all independent in origin. Some strains were derived from stocks generated by crossing existing strains, and so on (for a review of existing mouse inbred strains and their origins, see Festing [5] and Aitchley and Fitch [1,2]. Finally, it may be observed that no animal in a randomly mating, genetically segregating, population (either in the wild or in the laboratory) would ever have a genotype homozygous at all loci. For these reasons, it is clear that the inbred strains available to the researcher cannot, in any sense, be regarded as random samples of the murine genome but must be seen as a highly select sample of survivors. 3.3. Genetic differences among strains Even from the same starting breeding pair, a different choice of sibs at some stage of the inbreeding process could give rise to a different “final” genotype of the strain. With a different initial pair the differences would be expected to be more pronounced. We can assume, then, that any two inbred strains of different origin will have different genotypes. In the course of research utilizing inbred strains, allelic configurations at some major loci do become known, of course. But there is nothing in the inbreeding process per se that permits prediction of what these differences will be, and the specific array of loci pertinent to complex, polygenically influenced phenotypes is usually completely unknown. Because we don’t know the specifics of the genotypes of either of two compared strains, we can only be confident that they are each (nearly) genetically uniform and that they differ in respect to some loci. These loci are the

97

ones that might be said to be “accessible” for evaluation. Any loci for which the two strains have the same allelic configuration are “inaccessible,” in this sense, even if there exist allelic variants at those loci in the total mouse gene pool that can have a major effect on the phenotype. Clearly, the dimensionality of the genotypic space in which our reference strains are located will differ from pair to pair. That is to say, the accessible loci in comparisons of strains A and B will be different from those accessed in comparisons of A with C or B with C. This means also that only those parts of the aging processes or the neurobiological processes that are causally downstream from these accessible loci can be examined by strain comparisons. 3.4. The genetic uniformity of F1 hybrids The existence of inbred strains permits the generation of another class of uniform genotypes—the F1 hybrids. The mating of animals of two different inbred strains will produce progeny that are heterozygous at all accessible loci, including, of course, any subset of such loci that have an influence on the phenotype under study. This heterozygosity presumably accounts for the frequent observation that F1s display mean phenotypic values higher than those of inbred strains, if the phenotype is related to fitness. This “hybrid vigor” is, essentially, the recovery of fitness lost during inbreeding depression. Despite allelic diversity at all of the heterozygous loci, each F1 animal is genotypically like each other one. F1s share, therefore, the logical advantages of inbred strains. A particular advantage of F1s is that a substantial number of groups with different uniform genotypes can be generated by intercrossing of a relatively small number of inbred strains. For example, five inbred strains intercrossed in all combinations can yield 10 genetically unique F1 groups (20 if the reciprocal crosses are considered separately as must be done for certain purposes involving sex linkage, maternal effects of various kinds, mitochondrial transmission, or the recently discovered phenomenon of imprinting). Unlike inbred strains, F1s do not retain their genetic uniformity in subsequent generations of propagation: mating of F1 to F1 results in a genetically non-uniform F2 generation in which each individual animal receives a unique sample from the parents’ genotypes. The genetic variability of the F2 is constrained variability, of course, being limited to the accessible loci. Another F2, derived from different inbred progenitors, will be variable within the envelope defined by its accessible gene pool.

4. Uniform genotypes in experimental design 4.1. The location of uniform genotypes From a methodological point of view, selection of a single inbred strain for research purposes is similar to the

98

G.E. McClearn, S.M. Hofer / Neurobiology of Aging 20 (1999) 95–104

Fig. 1. Mean activity on a stationary dowel rod (maximum score for three trials) for two age cohorts of mice, separately by age and genotype group, sexes combined. Significant effects are indicated for age, genotype, and age X genotype (*p ⬍ 0.05; ***p ⬍ 0.001).

control that can be exerted over other variables such as time of day, temperature, illumination, dosage, pH of solution, etc., by fixing them at a particular value. But there is also a difference. With a uniform genotype, the degree of genotypic control is very good, but we do not have a metric to specify it. Thus, each inbred strain has a location in genotypic hyperspace, the dimensionality of which is unknown. Nonetheless, these strains provide standard, reliable, and replicable reference points, and constitute foundations for research aimed at understanding the genetic architecture underlying particular phenotypes, as well as research simply seeking to use the uniform genotypes to attain some control over individual differences in their animal subjects [4]. A difference between two inbred strains with respect to some phenotype constitutes ipso facto evidence of genetic influence on that phenotype. This is important information, but it reveals little about the genetic system. As already noted, we know only that the compared strains are at different genotypic locations, but we don’t know where either one of them really is. We can, however, establish a dimensional scale between them. A location genotypically halfway between the two strains is provided by F1 offspring of the strains. Subsequent generations—the F2 derived from intermating of F1 animals, and the backcrosses derived from mating F1s to the parent strains, for example—provide further distinctive points along the scale. Strong theory assures that the average allelic dosage for the F2s will be the same as that of the F1s and that that of the backcrosses will be halfway between the F1 and their respective parent strains. There are likewise strong postulates concerning genotypic variances of these different groups. Comparisons of observed phenotypic means and variances to these theoretical expectations permit estimation of the proportion of phenotypic variance due to genetic and to environmental sources,

interactions between alleles at a locus (dominance) and between or among different loci (epistasis), and interactions between genes and environment. Details of these genetic analyses are outside the focus of this paper, which is the utilization of uniform genotypes as control features in research oriented to gerontological issues. It is the case, however, that the latter type of application usually generates, in addition to the primary gerontological data, broad stroke information that genetic influence exists. An illustration is shown in Fig. 1. Activity while on a stationary dowel rod (maximum score from three trials) was assessed for animals of the C57BL/6 and DBA/2 strains and their B6D2F1 [15]. One group was measured at about 5 months of age, and another group at about 25 months of age. Significant main effects were found for age and strain, and for the interaction of strain with age. The interpretation of the interaction is straightforward: activity is lower at the older age and the strain differences are evident only at the younger age. This latter effect is particularly noteworthy in the context of developmental genetics, from which field powerful data are emerging concerning the differential regulation of gene activity at different ages. The ensemble of genes affecting a phenotype at one age may differ substantially from that affecting the same phenotype at other ages, and any conclusions about genetic effects, or their interactions with other key variables, must be tempered by appreciation that they may be limited to a particular developmental period. A situation that is at least analogous to this situation of developmental change (and possibly with homologous aspects) is the performance change that can occur over much shorter intervals as a function of practice or adaptation. A good example is provided by Forster and Lal [6] who assessed performance (amount of time) on an accelerating

G.E. McClearn, S.M. Hofer / Neurobiology of Aging 20 (1999) 95–104

Fig. 2. Effect of genotype upon rotorod performance as a function of training sessions. Rotorod time was the mean latency to fall (s ⫾ SE) from a cylinder rotating with constant acceleration (from 0 to 75 rpm in 150 s). Separate groups of 8- to 9-month-old, ad libitum-fed C57BL/6, DBA/2 and B6D2F1 mice received daily training sessions on the accelerating rotorod until a criterion of stability had been attained. The first seven trials are plotted for all mice. With permission from Forster and Lal [6].

motor-driven treadmill (rotorod) over seven sessions in ad libitum-fed 8- to 9-month-old C57BL/6, DBA/2, and B6D2F1 mice. All genotypes performed at approximately the same level in the first session, as seen in Fig. 2. However, whereas all genotypes showed improvements in performance, the C57BL/6 mice exhibited the largest training improvements with the B6D2F1 hybrids achieving intermediate gains. In this example, the maximum performance varied as a function of training and genotype, illustrating the importance of considering fundamental measurement issues, such as the desirability of repeated testing to achieve some performance criteria (e.g., asymptotic performance).

99

A cross-sectional study of the effects of dietary restriction and age on behavioral biomarkers provides an illustration of the usage of uniform genotypes in evaluating the effects of an environmental intervention—in this case, that of caloric restriction (Fig. 3). C57BL/6, DBA/2, and B6D2F1 mice were maintained either on a 60% calorically restricted diet or fed ad libitum. Mean geotaxis speed, measured as the rate of turning 180 degrees from a downwardfacing position on an inclined plane, was assessed as part of a larger battery of behavioral biomarkers. Strain differences give evidence for genetic influence on this biomarker, and the location of the F1, approximately intermediate to the two parental strains, is consistent with an additive genetic model. The interaction of age by diet condition, seen as the decreased geotaxis speed in the ad libitum groups compared to the relatively steady slope in the calorically restricted mice from 17 to 26 months of age, demonstrates that the efficacy of the environmental intervention is related to age. This type of research may also reveal interactions between experimental interventions and genotype. A pertinent example is the work of Fosmire et al. [8] who, in the context of risk factors for Alzheimer’s disease, examined the effects of elevated dietary aluminum in five inbred mouse strains. One strain, the DBA/2, was particularly susceptible, with the brain aluminum levels of treated animals greatly exceeding that of their like-strain controls. C3H animals were less sensitive and C57BL/6, BALB/c, and A/J mice showed no effect at all of the dietary intervention. A possible interpretation of these results is that genetic influences in this case operate through effects on the blood-brain barrier. Most important is the general point that the efficacy of any inter-

Fig. 3. Mean negative geotaxis speed (degrees/sec ⫾ SEM) for DBA/2NNia, B6D2F1, and C57/BL6NNia mice as function of age and diet condition. Geotaxis speed was measured using a 25 degree incline plane over two consecutive trials, each with a 120 s maximum allowable latency. The black squares denote diet-restricted mice whereas the white squares indicate ad libitum fed animals. There were insufficient numbers of diet-restricted DBA/2NNia animals at the third occasion to permit analysis. Results are presented with permission from M. J. Forster.

100

G.E. McClearn, S.M. Hofer / Neurobiology of Aging 20 (1999) 95–104

vention may be expected to differ from genotype to genotype. A variant of the usage of uniform genotypes, that of recombinant inbred (RI) strains, has assumed an increasingly important role in recent animal model genetic research. These strains of animals are derived by inbreeding from an F2 or other intercross. Basically, the accessible loci of the original parent strains can recombine in the F2, and alleles will be differentially distributed in the new, recombined, inbred (thus, recombinant inbred) strains. Each RI thus represents a different configuration of the same accessible loci. These animals have proven extremely valuable in research identifying chromosomal regions in which there exist loci (quantitative trait loci or QTLs) that influence some particular phenotype. They also provide a basis for a limited exploration of association between variables. 4.2. The phenotypic variance of uniform genotypes Because all members of an inbred strain or an F1 have the same genotype, any phenotypic differences within these groups are interpreted to be due to environmental sources and to measurement error. Empirically, the phenotypic variance within strains is often very substantial, even under rigidly controlled colony and experimental conditions. These observations reveal an exquisite sensitivity of some phenotypes to small nuances of environment and should engender some modesty about the degree to which we have, in fact, controlled, or even identified, all the relevant variables. Measurement error is conventionally regarded to arise from random influences that result in measured values that are normally distributed around “true” values. The measured value for any individual is therefore a single sample of the values that could have been obtained on that measurement occasion. In recent years, increasing attention has been drawn to the possible importance of intraindividual variability [20,21] as a lawful source of variance that can sometimes masquerade as measurement error. Briefly, the argument is as follows. Labile phenotypes are highly likely to be under some sort of dynamic control, such as the negative feedback processes that typify many homeostatic systems. When some influence is encountered that causes displacement of the phenotypic value from a set-point, a control process is engaged to return the phenotype toward that set-point. In effect, the “true score” will fluctuate from time to time. The temporal parameters of this fluctuation will depend upon the dynamics of the control system in terms of magnitude of displacement by a given stimulus, delay in initiation of recovery, rate of the recovery response, degree of damping in the response, steady state error, etc. When both fluctuance and genuine measurement error are considered, it becomes clear that the phenotypic value of an individual is a moving target. Figure 4 provides two illustrations of the distribution generated by taking 8 repeated measurements of a behav-

Fig. 4. Intraindividual distributions of locomotor activity over eight occasions of measurement (2–3 day intervals) for two individual mice from a genetically heterogeneous stock.

ioral measure—locomotor activity—at 2–3 day intervals. Data from two mice (in this case, genetically heterogeneous ones) are shown [16]. Generally speaking, when such a measure is taken, the investigator assumes that it will be characteristic of the animal for a considerable period of time—much longer than the 17-day period involved here. Yet in this example each animal displays a range of values during this brief period that the investigator could have obtained, depending on the day of measurement. These particular animals were chosen for display because they illustrate that substantial interindividual differences may exist in intraindividual variability. Animal A has a relatively narrow, positively skewed distribution, whereas animal B has a flat distribution spanning the entire range of scores. This example does not permit disentanglement of measurement error from fluctuance, of course, but it does suggest that, for many phenotypes, it may be highly desirable, or, indeed, necessary to sample more than once to characterize an organism at a particular age. In addition to this methodological implication, there is an interesting theoretical prospect that changes in the parameters of the dynamic control systems affecting intraindividual variability may be a fundamental feature of aging process(es), and may warrant intensive study in their own right. A substantial literature suggests that heterozygosity confers superior homeostatic control or buffering capacity to that of highly homozygous individuals in a wide array of phenotypes. Much of this research has concerned developmental processes, but the principles may apply to the shorter-term control processes of intraindividual variability, and thus on the assessed variance at any single measurement occasion. Indeed, it is frequently, though not invariably, observed that the within-group variance of F1s is less than that of the parent inbred strains [22]. Locomotor activity in

G.E. McClearn, S.M. Hofer / Neurobiology of Aging 20 (1999) 95–104

101

Fig. 5. Standard deviation and mean locomotor activity by strain and diet condition. Forward locomotor activity (cm) over three sessions was assessed using a standardized optical activity monitoring system. Mice were approximately 10 months of age at time of testing with 20 –25 animals per strain/diet group. Results are presented with permission from M. J. Forster.

10-month-old mice from a study of effects of caloric restriction on behavioral biomarkers of age provides an example of this. Greater within-group variability in inbred strains relative to the F1s is illustrated in the top two graphs of Fig. 5. Regardless of dietary condition, the F1s displayed less variability in activity than either of the parental strains. Also of note is the fact that variability of locomotor activity was greater under restriction in both of the inbred strains than in the B6D2F1. Indeed, under dietary restriction, variance in the F1 remained the same whereas the variability increased in both of the inbred strains. This result suggests caution when utilizing caloric regulation with inbred strains to control variability in certain phenotypes, as is sometimes attempted. The comparative variability of F1 and parent strains has long been a topic of considerable theoretical interest (e.g., [13]), but remains intriguing and incompletely understood (e.g., [19]). There is a broad consensus that the effect is a manifestation of heterosis, ascribable to increased allelic diversity of the heterozygous loci, affecting not only the mean value of fitness-related traits, but also their developmental buffering. It is an attractive hypothesis that this

buffering operates, at least in part, by affecting the parameters controlling the fluctuance or intraindividual variability. 4.3. Covariances in uniform genotypes Just as variances can be decomposed into components, covariances occur basically because a gene or set of genes influences two phenotypes, because genes influence one phenotype that influences another, or because environmental factors do. Within a single inbred strain or F1, of course, only the covariance arising from the environmental domain can be assessed. There may be research purposes for which this limitation is desirable, generating evidence as to the potence of environmental factors for creating correlations (under the constraint of the particular strain’s genotype). Ordinarily, however, the purposes of correlational assessment are best served when both genetic and environmental influences on covariances are included [17]. However, there are circumstances when phenotypic correlations in heterogeneous stocks are expected to be lower than that of isogenic groups [18].

102

G.E. McClearn, S.M. Hofer / Neurobiology of Aging 20 (1999) 95–104

The comparison of two strains on two variables might seem to be a useful way to incorporate genetic variance in the search for correlated variables. Thus, strains A and B might differ in respect to, say, longevity; it is tempting to determine if they differ also in antioxidant defenses. These two-strain, two-variable comparisons are, however, very limited, and possibly deceptive. Consider two strains that differ substantially with respect to one phenotype. The inbreeding process will have led to homozygosity at all of the loci pertinent to all phenotypes. Assuming that the second phenotype is not related to fitness, the expectation would be that, for any given strain, an equal number of the loci affecting this phenotype will be fixed in the “increasing” allelic configuration and in the “decreasing” configuration. When comparing two strains, however, sampling variation will regularly result in one strain having a higher value than the other, even if the first and second phenotypes have no common causal elements. Conventional alpha-level caution is required; note that each strain is one sample genotype, and the correlation being evaluated by observations on two strains should be thought of as a scatterplot with only two points. The degrees of freedom are (n-2) ⫽ [2–2] ⫽ 0, providing little basis for confidence in the outcome. The conceptually strongest outcome of a two-strain, twophenotype comparison is a negative one, in which the strains do not differ. An apparently positive result, however, with the strains differing according to hypothesis, is weak, being only suggestive or permissive in the sense that it permits one to continue to think that there might be a relationship between the biomarker and longevity. Definitive evidence must come from other designs. One ameliorative approach is to add other strains to the comparison. As the number of strains increases, confidence in the portrayed relationship increases, but, fundamentally, statistical power is related to the number of strains, not the number of animals. It is clear that many strains would be required by such a design to obtain an adequate estimate of covariance relationships. In this and other situations requiring assessment of associations between or among variables, a genetically heterogeneous stock has many advantages. An argument can be made that the most productive use of the general design of inbred strain comparison on multiple phenotypes is in hypothesis generation rather than hypothesis testing. Such an approach is often disdained as a “fishing expedition,” but as has sometimes been observed, such expeditions are effective ways of catching fish.

5. Strengths and advantages of uniform genotypes The foregoing description of uniform genotypes in experimental designs has accented their limitations and made clear that they do not constitute a panacea for methodological problems arising from the brute facts of genotypically influenced individual differences. It is important for these

limitations to be understood, but it is equally important that their very substantial advantages be comprehended. Both the weaknesses and the strengths derive from the fact that all animals within a strain are representatives of the same genotype. The principal virtue of uniform genotypes becomes apparent in experimental situations where an application of an independent variable may affect the response to subsequent applications of that variable, or when the outcome measures are destructive. The fundamental strength is that the same sampling entity, the genotype, can be phenotypically measured and re-measured as required or desired. The central limit theorem assures us that it is possible therefore to estimate the “true” phenotypic score for the genotype represented in an inbred strain or F1 to any desired degree of accuracy simply by increasing sample size. Similar reasoning applies to the estimation of the effect size of an intervention such as, for example, caloric restriction. The price that is exacted for the advantage of uniform genotypes is that estimates of mean values or of effects of interventions are pertinent to only one or a few genotypes from the thousands upon thousands of possible ones. The extent to which highly accurate information about that genotype generalizes to this enormous population is unknown. Alternative designs that sample from a population of genetically heterogeneous individuals enjoy greater generalizability, but the prospects of matching individuals across groups is limited. The difference in power is essentially the difference between unmatched and matched sample designs. Until cloning becomes a practical laboratory technique, inbred strains and their derived F1s represent the best available matching on the critical variable of genetic constitution. 5.1. Uniform genotypes in cross-sectional and longitudinal designs Special mention should be made of the advantage of uniform genotypes in the methodological intricacies of developmental designs. The strengths and weaknesses of cross-sectional and longitudinal designs are, of course, of central salience in any gerontological research. Major problems with cross-sectional observations are that if individuals of different ages are measured at the same time, they will necessarily be from different birth cohorts, and if from the same cohort, they will necessarily be measured at different times. Thus, either cohort effects or occasion-of-measurement effects may be confounding features. Longitudinal studies share the possibility of occasion effects, though cohort influences can be eliminated if all subjects are of the same birth cohort. Longitudinal studies have an additional potential complication in sequence-of-measurement influences such as practice effects, familiarity effects, and so on. Both longitudinal and cross-sectional designs are complicated by selective loss issues. In longitudinal studies, the loss occurs sometime after initial measurement(s); in cross-

G.E. McClearn, S.M. Hofer / Neurobiology of Aging 20 (1999) 95–104

sectional studies, the loss occurs before measurement. In either case, the older individuals are samples from a different population than are the younger individuals. These and similar considerations eventuated in a major examination of gerontological research design, analysis, and interpretation (e.g., [3,24]), concerned principally with human research. In animal research, the greater capacity for experimental control ameliorates some of the problems, but other problems arise. For example, in the case of research utilizing animals imported to the research colony from another colony, importation stress due to the conditions of shipment and new environmental conditions (light cycle, noise level, density altitude, humidity, food, caretakers, ambient odors, etc.) may influence many biomarkers of aging. Assessment of, say, two simultaneously imported cohorts of different ages before adaptation has occurred to the new environment may reflect age X stress interactions. A cohort-sequential design with repeated measures on these cohorts might reveal as much about age differences in adaptation rate as it does about aging. These particular potential problems are alleviated, of course, if all animals can be born, maintained and tested in the same, highly controlled environment. But they are not eliminated. Even in rigorously controlled circumstances, for example, seasonal changes in a variety of variables are commonly experienced. Thus, any design requiring comparison of animals of different ages measured at different seasons may have a confounded element. The issue of differential loss also cannot be eliminated in the animal laboratory, as has been discussed above in relation to inbreeding depression and strain loss during inbreeding. Even in a barrier operation, where differential susceptibility to infectious disease may be largely eliminated as a biasing factor, other intrinsic or endogenous factors remain. Thus, for many purposes, in animal-model research as in human research, neither longitudinal nor cross-sectional designs are utterly unambiguous, and a combination of methods may be required to give best estimates of parameters affected by aging. Uniform genotypes provide an interesting and pertinent perspective on issues of longitudinal and cross-sectional methods. In essence, the merit of longitudinal observations derives from the capability of identifying the same individuals at different ages. That is to say, to enjoy the interpretational advantages of longitudinal study, there must be a capability of relating a particular measurement on occasion 1 to a particular measurement on occasion 2, etc. In the case of ordinary inbred strains or RIs, the genotypes are thus identifiable even in the case of cross-sectional observations. The cross-sectional observations on groups of uniform genotype will closely approximate the abstractly ideal but realistically impossible longitudinal results. Thus, crosssectional observations on groups with uniform genotypes offer the advantages of matching on successive measurement occasions, but without the problem of successive measurement effects. Indeed, an argument can be made that, for many purposes in the developmental sciences, multiple-

103

occasion, cross-sectional data from uniform genotypes actually enjoy some interpretational advantages over both conventional longitudinal and cross-sectional data from heterogeneous animals.

6. Summary Animal groups of uniform genotypes, either inbred strains, recombinant inbred strains, or F1 hybrids derived from them, are a major resource in the conduct of biological and biomedical research. By virtue of their uniformity with respect to genotype, a fundamental aspect of the animals’ biological status, they constitute standard reference groups that make possible efficient conduct of the essential operation of scientific inquiry—that of replication. In addition they offer focused relevance of information generated about the same genotypes in different laboratories and at different times. The costs of this precision of reference include reduced a priori confidence of generalizability of results and inefficiency in assessing associations among variables. It is important to appreciate that no single research tactic is adequate for all purposes. For comprehensive exploration of a research domain, the information from uniform genotypes must often be complemented by data from other genotypically specifiable groups such as intercrosses, genetically heterogeneous stocks, or phenotypically or genotypically selectively bred lines. These complementary tools are themselves usually derived from inbred strains, so the latter can justifiably be regarded as the foundation for this type of research.

References [1] Aitchley WR, Fitch W. Gene trees and the origins of inbred strains of mice. Science 1991;254:554 – 8. [2] Aitchley WR, Fitch W. Genetic affinities of inbred mouse strains of uncertain origin. Mol Biol Evol 1993;10(6):1150 – 69. [3] Donaldson G, Horn JL. Age, cohort, and time developmental muddles: easy in practice, hard in theory. Exp Aging Res 1992;18:213–22. [4] Festing MFW. Inbred strains in biomedical research. London: The MacMillan Press Ltd.; 1979. [5] Festing MFW. Inbred strains of mice. In: Lyon MF, Searle AG, editors. Genetic variants and strains of the laboratory mouse, 2nd Edition. Oxford: Oxford University Press, 1989. p. 636 – 48. [6] Forster MJ, Lal H. Neurobehavioral biomarkers of aging: influence of genotype and dietary restriction. Biomed Environ Sci 1991;4:144 – 65. [7] Forster MJ, Lal H. Within-subject behavioral analysis of recent memory in aging mice. Behav Pharmacol 1992;3:337– 49. [8] Fosmire GJ, Focht SJ, McClearn GE. Genetic influences on tissue deposition of aluminum in mice. Biol Trace Element Res 1993;37: 115–21. [9] Ingram DK. Toward the behavioral assessment of biological aging in the laboratory mouse: concepts, terminology, and objectives. Exp Aging Res 1983;9:225–38. [10] Ingram DK. Perspectives on genetic variability in behavioral aging of mice. In: Harrison DE, editors. Genetic effects on aging II. Caldwell, NJ: The Telford Press, 1990. p. 205–31.

104

G.E. McClearn, S.M. Hofer / Neurobiology of Aging 20 (1999) 95–104

[11] Ingram DK, Jucker M, Spangler EL. Behavioral manifestations of aging in laboratory rats. In: Mahr U, Capen C, Dungworth D, editors. Pathology of aging animals, Vol. 1. New York: Springer Verlag, 1991. p. 149 –70. [12] Ingram DK, Reynolds MA. Assessing the predictive validity of psychomotor tests as measures of biological age in mice. Exp Aging Res 1986;12:155– 62. [13] Lerner IM. Genetic homeostasis. New York: John Wiley & Sons, 1954. [14] Lhotellier L, Perez–Diaz F, Cohen–Salmon C. Locomotor and exploratory activity in three inbred strains of mice from young adulthood to senescence. Exp Aging Res 1993;19:177– 87. [15] McClearn GE. Biomarker characteristics and research on the genetics of aging. In: Harrison DE, editor. Genetic effects on aging II. Caldwell, NJ: The Telford Press, Inc, 1990. p. 234 –54. [16] McClearn GE. The reliability and stability of biomarkers of aging. Ann N Y Acad Sci 1992;673:1– 8. [17] McClearn GE, Hofer SM. Genes as gerontological variables: genetically heterogeneous stocks and complex systems. Neurobiol Aging 1999;20:147–56. [18] McClearn GE, Wilson JR, Meredith W. The use of isogenic and heterogenic mouse stocks in behavioral research. In: Lindzey G, Thiessen DD, editors. Contributions to behavior-genetic analysis,. New York: Appleton-Century-Crofts, 1970. p. 3–22.

[19] Mitton JB. Selection in natural populations. Oxford, England: Oxford University Press, 1997. [20] Nesselroade JR. Sampling and generalizability: adult development and aging research issues examined within the general methodological framework of selection. In: Schaie KW, Campbell RT, Meredith WM, Rawlings CS, editors. Methodological issues in aging research. New York: Springer, 1988. p. 13– 42. [21] Nesselroade JR. Interindividual differences in intraindividual change. In: Collins LM, Horn JL, eds. Best methods for analysis of change. Washington, DC: American Psychological Association, 1991. p. 92– 105. [22] Phelan JP, Austad SN. Selecting animal models of human aging: Inbred strains often exhibit less biological uniformity than F1 hybrids. J Gerontol Biol Sci 1994;49:B1–11. [23] Spangler EL, Waggie KS, Hengemihle J, Roberts D, Hess B, Ingram DK. Behavioral assessment of aging in male Fischer 344 and brown Norway rat strains and their F1 hybrid. Neurobiol Aging 1994;15: 319 –28. [24] Schaie KW, Hertzog C. Longitudinal methods. In: Wolman BB, editor. Handbook of developmental psychology. Englewood Cliffs, NJ: Prentice Hall, 1982. p. 91–115. [25] Sprott RL. Development of animal models of aging at the National Institute of Aging. Neurobiol Aging 1991;12:635. [26] Sprott RL. Mouse and rat genotype choices. Exp Gerontol 1997;32: 79 – 86.