NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Research Council (US) Committee on Population; Finch CE, Vaupel JW, Kinsella K, editors. Cells and Surveys: Should Biological Measures Be Included in Social Science Research? Washington (DC): National Academies Press (US); 2001.

Cover of Cells and Surveys

Cells and Surveys: Should Biological Measures Be Included in Social Science Research?

Show details

9The Relevance of Animal Models for Human Populations


To set the stage for considering the relevance of animal models to the ventures that are the focus of this volume, it may be useful to contemplate briefly the various meanings that have become attached to the term “model.” In general discourse, among many definitions, a model can be regarded to be “a tentative ideational structure used as a testing device” (The American Heritage Dictionary of the English Language), or “a description, a collection of statistical data, or an analogy used to help visualize often in a simplified way something that cannot be directly observed” (Webster's Third New International Dictionary).

Philosophers of science have provided more technical definitions, of which the following examples are illustrative. Kaplan (1964:263) observes that “any system A is a model of a system B if the study of A is useful for the understanding of B without regard to any direct or indirect causal connection between A and B.” Rapaport (1954:206) identifies models as “scientific metaphors” and comments thusly on their usage: “Like every other aspect of scientific procedure, the scientific metaphor is a pragmatic device, to be used freely as long as it serves its purpose to be discarded without regrets when it fails to do so.” In the philosophical literature a distinction between conceptual models and physical models is often encountered, and further taxonomic distinctions can be found among analogical, descriptive, explanatory, formal, heuristic, iconic, inferential, interpretive, measuring, predictive, semantical, statistical, symbolic, and syntatical models. Many of these categories are nonexclusive, so any particular model system can be a blend of several types. Indeed, Sattler (1986) observes that the term is increasingly used to denote biological generalizations across the entire range from traditionally labeled theories to laws, rules, and hypotheses.

So, models can be of various kinds for various purposes, ranging from an abstract mathematical expression to a practical experimental or observational system. It is the latter general type of model with which this note will be concerned.


Common to all these definitions of models is the understanding that the essence of models is a simplification or abstraction of the phenomenon being modeled. The establishment of an experimental model is an attempt to isolate a part of a more complex system in order to study a particular element or subset of elements of that system. This process can be characterized in the familiar terms of the prototypic experimental design: one or a few elements are selected to be the manipulated, independent variables; other variables are subjected to control by fixation or randomization; and one or more others are identified as the outcome or dependent variable(s) to be measured. Similarly, in terms of associational research, elements can be predictor or predicted variables. In the broad sense used here, model systems are ubiquitous in the scientific enterprise: every questionnaire item, every score on a cognitive task, every demographic category, invokes a model of some sort.

Among this abundance of model systems is a subset that utilizes other living beings to address questions about some aspect of human life. Although plants can also be informative in this type of research (see Finch, 1990), it has mostly involved other animals, so the generic title is usually given as “animal model research.”

Species Choice

The fundamental rationale for the use of animal models derives from the phyletic relatedness of living things. These phyletic relationships constitute general themes and variations on these themes. When we study some species other than our own, there is a hope that we share with that species enough of the pertinent theme that information from the model system will illuminate something about ourselves. As the molecular exploration of the human genome and genomes of a select group of other species has proceeded, there has been an increasing realization of the extent of syntenic relationhips between these models and humankind. This realization has strengthened enormously the logical base for expectations of successful application of these animal models to complex human phenotypes. A striking example of this potential is provided by Rubin et al. (2000), who describe orthologs in Drosophila for 177 of 289 human disease genes examined.

There is a general presumption that the closer the phyletic relationship, the more informative will be the model. Other things being equal, then, a species from the same family as H. sapiens might be preferred to another species from a different family but from the same order, and so on. However, some themes might be so pervasive and fundamental that almost any species from the same phylum (or even kingdom) will be informative. Species closely related to us are rare and expensive, however, so species choice is usually a matter of trade-off between family resemblance and cost-efficiency. The optimal choice for a particular human phenomenon may be quite different from the best selection for some other phenomenon. So we can't expect that there will be a single gold-standard reference species, suitable for all purposes.

For myriad reasons, including both availability and demonstrated utility, but also with an element of historical accident, certain species have become traditional or standard model systems in different research domains. As literature has accumulated about these animals, as matters of husbandry and testing procedures have become incorporated into the scientific lore as appropriate and state-of-the-art, and as review groups and editorial boards have developed expectations, the motivation to continue the employment of the familiar animals has become very powerful. Thus, the animal-model-derived knowledge that has been accumulated in each substantive domain has been filtered through these standard animal groups and perhaps constrained by the sparse sampling we have made from the enormous phyletic array that is available to us.

It might be argued that the particular choices have served us well. Genetic science, for example, is certainly thriving from data generated from fruit flies, round worms, yeast, and mice, and it is clearly wise to invest more effort where there has been so handsome a payoff. What we don't know, of course, is what we could have learned from wallabies, octopi, kinkajous, or dragonflies. Austad (1993) has commented pungently on this issue with special reference to gerontological research and has urged a broadening of our horizons. It is to be expected that the model systems for the immediate future will be constructed with the traditional species, but it would be desirable to develop a strategy that encourages exploration of new species while continuing to exploit the old.

Environmental Variables in the Total Model System

As central as the species issue is, there's more to a model system than the species that is chosen to serve as a surrogate for humanity. The “animal model” consists not only of the species employed, but also the totality of the assessment situation, with all of its manipulated, controlled, and measured variables (McClearn and Vandenbergh, 2000). With respect to most complex phenotypes, the investigator has numerous choices concerning all of these types of variables. Consider, as an example, the decisions to be made in the construction of a mouse model of human cognition. The critically important issue is whether the performance of the animal in whatever test situation we devise will be homologous, or at least usefully analogous, to some human cognitive process. But, leaving aside this issue of trans-specific generality or validity for the moment, there is the nitty-gritty issue of the specific test situation from which the measurement will be obtained. Should it be a maze, an operant conditioning apparatus, or classical Pavlovian conditioning? How should the animals be motivated to perform: hunger, thirst, escape from shock, escape from water, or curiosity? If hunger, should they be maintained at a standard percentage of initial body weight or subjected to 20-hour food deprivation? What sensory cues should be provided: visual, auditory, or spatial? If visual, what light intensity and what pattern? If auditory, what frequency and intensity? What time of day should be used for testing? Should it be at night, given the nocturnal proclivities of the animals? What response should be required? What interval should be allowed to elapse between successive trials? Should performance be measured in speed of response or in number of errors? It is well understood that the performance of an animal in any situation is a function of all of these variables as well as its cognitive functioning. Is the performance of an animal swimming for its life (as far as it knows, presumably) in a deep water maze displaying its best cognition, its relative ability to detect pertinent extra-maze cues, or its state of panic? In what proportion do these sources contribute to the observed performance?

General husbandry conditions are pertinent as well: temperature, humidity, light/dark cycle. Other possibly influential variables may not be susceptible to control by deciding on some fixed level. All animals cannot be the first tested, so order of testing may need to be randomized. The position of a home cage on the cage rack may expose the animal to different illumination levels, sound levels, or air flow, so it might be desirable to distribute animals of different experimental groups randomly on the racks, and so on for many similar variables.

It is obvious that the particular assemblage of controlled and randomized environmental features may yield idiosyncratic results relative to another assemblage differing in some respect, and it follows that no single constellation of factors constituting the model can be regarded as definitive with respect to the target phenomenon. There is not, and cannot be, a “gold standard”—an utterly valid single model that reflects all of the pertinent attributes of a complex system. Each model system constitutes an operational definition of the target phenomenon, and humility is called for in assumptions about the extent of “coverage” of that target provided by that model. Indeed, in complex systems (and most everything of interest in the present context certainly qualifies as complex), the very definition of the target phenomenon usually emerges pragmatically from converging evidence from numerous models. Again, practical considerations clearly militate against the indefinite proliferation of models in any given area of research, but an area is probably best served if there are several operational definitions in deployment on its research scene.

These considerations pertain to the setting of the model. An essential ingredient is the animal to be placed in the setting, and genotype is a major defining feature of the animal.

Genetic Variables in the Total Model System

The basic selection of the surrogate species for an animal model study is, of course, a selection based on genotype—the differences among species in their gene pools. The genome of a species is fundamental in establishing the “theme” mentioned earlier, but among animals within the same species there exists enormous genetic variation that provides the variation on the theme. Our present focus is on the exploitation and control of this intraspecific genetic variation in the construction of animal model systems. Just as in the case of the situational and measurement variables in the model, the genotype can usefully be considered in terms of the role it plays as a variable (or conglomerate of variables) that can be controlled by fixation or by randomization, and that can be manipulated.

Genotypic Constraint

The primary method of fixing a genotype is inbreeding—the mating of relatives. After about 20 consecutive generations of mating of siblings, for example, the animals in a family lineage asymptotically approach the condition of being homozygous in like allelic state for all genetic loci. That is to say, they are nearly genetically uniform. The reservation refers to the fact that the process is asymptotic, that strong selective advantage to heterozygotes might maintain segregation at a very few loci within a strain (Falconer and Mackay, 1996), and that mutation might occur and be propogated in the strain. But, for most purposes, these considerations are quibbles; inbred strains closely approach genetic uniformity. They approximate an indefinitely large number of clones or of monozygotic twins, with an additional uniformity conferred by their homozygosity at all or nearly all loci, and the logic of their use is similar to that which can be used with these latter groups. Inbred strains constitute a major, robust animal model strategy in biomedical research.

Each inbred strain is derived from a single sibling pair. The genes of a pair of animals obviously can be only a partial sample of the gene pool of the population from which they are taken. Because they are siblings, the mates will be alike in homozygous state at some genetic loci, but they will be in unlike state at others and will be heterozygous at still others. The inbreeding process will eliminate the allelic differences at the polymorphic loci in a more-or-less stochastic way. Thus, the resulting inbred strain is only one configuration of homozygous loci derivable from the mini-gene pool represented by the initial mating pair (which, as noted, is only a tiny sample of the species gene pool).

The methodological virtue of an inbred strain is its relative genetic uniformity and stability. These attributes confer on inbred strains a replicability so that the basic biological properties of the animal in the animal model can be assured in different laboratories and at different times (Festing, 1971; McClearn and Hofer, 1999a). In terms of opportunity to conduct cumulative researches on the same basic animal material, not only by one laboratory but by anyone in the world, this is an enormous advantage over animals of unknown origin and unsystematic maintenance. Many different inbred strains exist (Festing, 1971), differing widely in almost every phenotype that has been explored, so a screening is likely to identify one that displays a desired level of a phenotype, and this strain can serve thereafter as a reliable, repeatable element in the model system.

The price that is paid for the stable replicability of inbred strains is that they represent such a sparse genetic sampling from the species and their phenotypic variance is due solely to environmental agencies. The first feature raises questions of representativeness of the chosen strain for the species in general. Remembering that construction of a model involves simplification and abstraction from the totality of the “real” phenomenon, this constraint may be acceptable for many purposes. A particular strain may provide a very nonrepresentative, exaggerated phenotype that makes it particularly valuable. An example is that of the C57BL/6 strain, which is atypical of mouse strains in general in displaying a high preference for a 10 percent alcohol solution when offered a choice between that beverage and plain water. These animals have played a useful role in a variety of research programs which require animals that will voluntarily ingest ethanol. If broader representation is required, the use of multiple strains offers an approach: if a particular finding can be shown in several strains, confidence is increased that the relationship being explored is at least not idiosyncratic to a single genotype. It remains the case that each of the inbred strain genotypes is an “abnormal” one, in the sense that homozygosity at all loci will not be found in a randomly mating population.

The second feature—displaying only environmentally caused variance—constitutes a limitation on the exploration of covariation of variables. It is certainly conceivable that a research program has interest only in the pattern of interrelationships of variables that can be attributed to variation in the environmental circumstances experienced by the individual animals. Most purposes, however, will likely be served better by a variance/covariance structure in which there have also been contributions from genotypic differences among the individuals. Again, one partial solution is to employ a number of inbred strains. The limitation in this case is that the interesting genetic covariance is the covariance among strain means, and statistical power is related to the number of strains rather than individuals (McClearn and Hofer, 1999a).

Obviously, the study of a single inbred strain is uninformative about genetics. All one knows is that the genotype is uniform. Comparisons among strains begin to provide some genetic information, however. The logic of inbred strain comparisons is as follows: the animals of each strain are replicas of the same genotype; variability within strains is attributable to environmental sources; animals of different inbred strains have different genotypes; mean phenotypic differences between strains (reared and tested in the same environment) are attributable to these genetic differences between strains. This logic applies to all loci that affect the phenotype for which the compared strains possess different alleles. Obviously, such a comparison cannot assess the influence of loci for which the strains happen to possess the same alleles.

The level of genetic information yielded by strain mean differences is a modest one, in and of itself. Data of this sort, however, set the stage for further analyses by other mating schemes (McClearn, 1991).

Genetically Heterogeneous Stocks

There are research questions that are best answered by samples of animals from genetically heterogeneous populations. Such samples are a counterpoint to the genotypic fixation of inbreeding and can be attained by several routes. Some colonies are maintained with no particular mating plan. These are almost certain to contain genetically variable animals. Live trapping from the wild will certainly provide genetic heterogeneity. Intercrossing of two or more inbred strains will produce animals of differing genotypes.

The first situation, the unsystematic colony, suffers from the replicability difficulties mentioned earlier. Such stocks are often of unknown origin and of unknown degree of inbreeding or heterogeneity. Clearly, results obtained from these groups are not uninformative—they show that representatives of the species can provide the outcome described. With respect to replicability, an investigator with such a colony can obviously return for further samples, but with unsystematic mating protocols, the stability of the gene pool over time is problematic, and the prospects of investigators from other laboratories making use of the material is usually limited. The shortcomings of unspecified groups are increasingly appreciated by the scientific community, and, in some research areas, results from genetically unspecified or unspecifiable animals are simply unpublishable.

Live trapping from the wild is sometimes recommended to overcome a perceived shortcoming of the typically used laboratory stocks (Miller et al., 1991). It is contended that generations of existence in the laboratory have made them nonrepresentative of the species in the wild state. Wild-trapped animals can be of particular value to research with a strong evolutionary or comparative orientation. They will also provide the basis for the desirable broadening of the phyletic sampling used in gerontological research (Austad, 1993). There are, however, some associated problems. For example, with respect to representativeness of the sampled species, selective trapability can introduce a substantial bias in the sample actually obtained, and quickly acting selective survival and reproduction once the wild-caught specimens have entered into laboratory existence will rapidly reduce the genetic variability represented in the original trapped sample (McClearn, 1998). With regard to specifiability in this type of research, some degree of replicability of sampling can be obtained by careful repeating of the trapping procedures in the same ecological area. For some purposes this may suffice, but problems of availability to other investigators may be significant. Until a sufficient body of data concerning live-trapped groups has accumulated and become a standard research resource, it is likely that genetically heterogeneous stocks will be obtained by mating of stocks and strains already abundantly available in laboratory colonies.

Assembling heterogeneous stocks from the existing inbred strains has the problems of derivation from the clearly “abnormal” starting points but has the substantial advantage of a high degree of replicability. That is, although the genotypes of individual animals in a genetically heterogeneous stock are not repeated, the operations for constructing the stock are clearly specifiable and repeatable. Although the ways in which matings can generate heterogeneity from inbred beginnings are many, there are several more-or-less standard outcomes: F2s, backcrosses, four-way crosses, and advanced intercrosses.

The initial step in most of the “recipes” is to produce an F1 generation. Mating of individuals of two different strains will generate offspring which are heterozygous for all loci for which their parental strains differed in allelic configuration. This heterozygosity is a dramatic difference from their homozygous parents, and F1s figure prominently in basic genetic studies of the phenomena of dominance, hybrid vigor, fitness, developmental canalization, and so on (see Lerner, 1954, for an early review; Mitton, 1997, for a recent one). But it is important to note that all of the animals within an F1 generation are genetically alike. They are each heterozygous at the same loci and homozygous at all the loci for which their parents did not differ.

It has long been observed that F1 hybrids between two strains may be less variable phenotypically than their parent strains. One interpretation is that the heterozygosity of the F1 animals confers a higher degree of developmental homeostasis (see Phelan and Austad, 1994; McClearn and Hofer, 1999a, 1999b; Miller et al., 1999). As a consequence, F1 animals are sometimes recommended instead of inbred strains in studies requiring genetic uniformity of the subjects.

F1s can thus be used with advantage in many circumstances, but it should be noted that they cannot be maintained in the genetically uniform state simply by mating inter se. Instead, each F1 sample must be generated anew by strain crossing.

In the generation of gametes, crossing over regularly occurs, and genetic information is swapped between members of a chromosome pair. That doesn't matter within inbred animals, because the swapped parts are identical. In an F1 animal, however, the chromosomes of a particular pair are genetically different, one each having come from each parent. Each gamete produced will be unique, as will be each F2 zygote formed by uniting of the gametes from two F1 parents. An F2 group thus provides for expression of some genetic variability. This variability is limited to the allelic differences existing between the parent strains of the F1s, so that another F2, derived from different inbred strains, will express different genetic differences.

Even greater genetic heterogeneity can be achieved by mating of F1s that were derived from different parent strains. The progeny of such a cross are “four-way cross” animals, which, relative to F2s, offer scope for more loci at which allelic differences can exist and for more than two allelic alternatives at these segregating loci. Four-way crosses are clearly specifiable, and can be constituted in relatively short order as required. The general principle can be extended, of course, but eight-way crosses appear to be something of a limit as a useful compromise between genetic diversity and manageability (McClearn et al., 1970).

From a statistical point of view, heterogeneous stocks offer a more substantial model system for the evaluation of associations between and among variables than that provided by inbred strains. The opinion expressed earlier—that the characterization of complex systems will require multiple measurements—carries with it an implicit need for multivariate descriptive and analytical procedures. The heterogeneous stocks will provide a favorable vehicle for generating the variance/covariance matrices and other statistical bases for these purposes.

From a genetic point of view, the segregating of multiple loci makes possible the manifestation of various forms of epistasis, or interactions among genes, which are hidden within inbred strains. Although not yet well explored for the phenotypic domains of particular interest in this volume, the general evidence for epistasis in complex systems is substantial and growing. Two early examples from mouse research will make the point. Coleman and Hummel (1975) showed that the pathophysiological expression of a diabetes gene differed dramatically depending upon the genetic background. Fowler and Edwards (1961) found that a gene severely affecting growth behaved as a model Mendelian gene in the population in which it initially was found but yielded non-Mendelian ratios in matings in another population. Molecular analyses of epistasis are proliferating, adding reductionist underpinning to observations at the whole-organism level. Theoretical perspectives (e.g., Bonner, 1988; Kauffman, 1993) also make the case for the importance of epistatic relationships in complex genetic systems.

As the avalanche of genome-mapping information continues, heterogeneous stocks of mice are proving of great value in the quest for quantitative trait loci (QTL). Until recently, the genes of a polygenic set affecting a complex phenotype were anonymous, and their influence was explored by statistical description. Very recently, it has become possible because of the mapping information now available for human beings, as well as the model organisms of mice, yeast, fruit flies, and nematodes, to individuate many of these loci by identifying a chromosomal region in which they reside. Such identifications open the way to subsequent molecular analysis, and the result will undoubtedly be an enormous enrichment of understanding of the dynamics of polygenic systems.

In the first generation of the heterogeneous stocks, linkage relationships persist, so that some observed associations among phenotypes may be spurious. This possibility can be evaluated by observations in an “advanced intercross”—groups derived by subsequent generations of mating within the heterogeneous stock. The linkage relationships will break down systematically as a function of number of generations, and obtained correlations among phenotypes will more assuredly be due to shared mechanisms causally downstream from the same gene or genes rather than fortuitous location of independent genes on the same chromosomes. In the QTL and mapping endeavors, different purposes are served by heterogeneous stocks in different stages of linkage disequilibrium. Initial detection of a QTL, for example, is more readily accomplished in a group with extensive linkage; finer localization is facilitated by a more advanced intercross.

Genetic variability can be explored usefully by yet another classical procedure—the production of backcross groups. F1 animals are mated with one of the parental types, yielding offspring that are intermediate in allelic frequencies to those of the F1 and the parent. Thus, while genetically heterogeneous, that heterogeneity is reduced relative to the F2. Backcrosses can be obtained in either direction, and their means and variances are interpretable in the context of quantitative genetic theory in estimating parameters of the polygenic system. Continued successive backcrossing will result in increasing genetic uniformity, and, coupled with selection for possession of a particular allele, generates useful test-beds for examining the effect of allelic differences at that locus.

The combination of genetic fixation and variability provided by recombinant inbred strains is proving itself of particular value in the current search for quantitative trait loci. Recombinant inbred (RI) strains are derived by consecutive sib mating from a common F2 group. Thus, the genetic differences between the original strains that produced the F1 that produced the F2 are reshuffled (by virtue of genetic recombination) into different lines and rehomogenized by inbreeding. In the use of this type of population, the advantages of genetic uniformity are possessed by the individual RI strains, each of which is a different combination of the allelic differences between the progenitor strains, thus providing genetic heterogeneity. Genetic influence is represented by differences among the RI means, and environmental influence is revealed by the within-RI strain variance. The power for correlational analysis is, of course, limited to the number of strains. Because many of the panels of RI strains have been characterized with respect to chromosomal markers, they are highly cost-efficient material for initial nomination of QTL (though they require confirmation because of inevitable false positive indications due to the large number of statistical tests required).

In view of the general advantages and the special applicability of heterogeneous stocks, their relatively modest record of employment in animal epidemiological models is regrettable. Given the genetic heterogeneity of human populations, cogent arguments can be advanced for the particular appropriateness of such stocks for this type of research, and there have been several recent reviews encouraging their more widespread deployment in future research (McClearn, 1999; McClearn and Hofer, 1999b; Miller et al., 1991, 1999).


The manipulation of genes—through mating procedures—has been stock in trade of the science of heredity since Mendel. In the case of major genes, the genotypes of individual animals can be deduced from their phenotype or from the phenotypes of their offspring or other relatives. Appropriate matings can generate offspring groups of known or strongly inferred genotypes. This procedure can test hypotheses about mode of inheritance of a particular phenotype, or, in the case of a well-established genetic condition, can be useful for providing different genotypes for the study of mechanism of gene action, and so on.

It is also possible to manipulate genes even when nothing is known about their number, effect size on the phenotype in question, chromosomal location, or mode of action. This general procedure, phenotypic selective breeding, also relies on inferring something about the genotype from the phenotype. Thus, if any of the variance of a phenotype is due to genetic differences among the individuals in the population (i.e., the heritability is nonzero), then individuals from the low and high extremes of the distribution will possess, on average, fewer and more “increasing” alleles, respectively, than the population in general. If members of the like extremes are mated, their offspring will thus have reduced and increased allelic frequencies, respectively, relative to those of their parents' generation. Further selection of mates from the appropriate extremes will lead, ultimately, to two groups of animals: one containing all, or most, of the increasing alleles; and another with all, or most, of the decreasing alleles at all of the unknown, and possibly multitudinous, loci in the gene pool provided by the foundation stock that can influence the phenotype in question. From the rigor of the selection and the rate of divergence of phenotypes of the two lines of animals, deductions can be made about certain parameters of the genetic “architecture” of the phenotype. Perhaps more importantly, once generated, these selectively bred animals can serve as prime research material for the investigation of the mechanisms through which the genetic influences are mediated.

Phenotypic selective breeding is an extremely effective method for manipulating entire blocks of anonymous genes, operating on all loci affecting the phenotype that are segregating in the foundation (necessarily heterogeneous) population, and sorting those alleles associated with increased expression into one group and those for decreased expression into another group. These groups then constitute prime research material for the exploration of mechanisms underlying the phenotypic differences. In short, phenotypic selective breeding is a method for the systematic generation of animal models. Research fields differ widely in their utilization of this tool-making procedure. Examples of areas in which selected lines have contributed fundamentally include body growth (see Falconer and Mackay, 1996, for review) and alcohol-related processes (see Crabbe et al., 1994; McClearn et al., 1981). In gerontological science, selection studies in Drosophila (reviewed by Shmookler Reis and Ebert, 1996) have provided extremely valuable model systems, and similar projects in mice are contemplated (Harrison and Roderick, 1997).

Advances in molecular genetics have made it possible to manipulate the genotype in more specific ways. In contrast to the genetic collectivity involved in phenotypic selective breeding, these methods permit the manipulation of specified individual genes. By various techniques, it is possible to introduce new genes into an organism or to negate expression of a particular gene. Use of these transgenic and knock-out preparations is expanding at a spectacular rate, and they will undoubtedly be major components of animal model systems of the future.

These methods will be particularly valuable in identifying mechanisms of expression of single genes identified by other methods. They are likely to provide valuable opportunities to explore gene-gene interactions, given the frequent observations of dependence of outcome on the genetic background of the host animal. A noteworthy example of the prospects for this type of research is the work of Sullivan et al. (1997). These authors replaced the mouse apolipoprotein E (APOE) gene with the human APOE e3 allele, providing a model system for the study of diseases associated with the human APOE isoforms.

Another manipulation that becomes possible with identification of specific loci is that of genotypic selective breeding. Such a procedure, with mate assignments based on measured genotype rather than measured phenotype, will be particularly useful for exploration of the complexities of gene-gene interaction, because the genetic systems can be assembled in any desirable combination. Whereas genotypic selection will be particularly illuminating when conducted with actual genes, the principle can be extended to QTL selection. Thus, genetic complexes of yet-anonymous genes can be built in various configurations based on mate assignment according to QTL genotypes, capable of generating, as in the case of phenotypic selection, animal models purpose-built to investigator specification.


In epidemiological and demographic research arenas, where genes and biological markers can be sought directly in human populations as well as animal models, the utility of research on animals lies in, among other attributes, their shorter lifetimes, shorter generation intervals, and accessibility for study of putative mechanisms. The major concern with the animal model in this context is its validity, which by informal definition is the extent to which it measures what it purports to measure. There are two subissues here: does the particular model represent the target phenomenon adequately in the animal, and does the animal phenomenon have points of similarity with the target phenomenon in humankind? For some phenotypes, there is a reasonableness about inferring from animal data to human traits. The human-validity of genetic observations from various invertebrate and vertebrate species is resounding. Similarly, it appears likely that blood pressure; hematocrit; glucose tolerance; collagen cross-linkage; proliferative response; white cell count; growth hormone; and similar neurological, physiological, endocrinological, and immunological variables in mammalian systems will be informative about the human condition. These face-validity assumptions are not so easily made in connection with behavioral and sociological variables. An earlier example explored the issues in establishing an animal model for human cognition. What does black-white discrimination under hunger motivation and food reward have to do with human general intelligence? It is not obvious on the face of the matter. But it is very unlikely that the functioning of the central nervous system in changing behavioral responses in what we call learning situations is totally unrelated in mouse and man. We lack, unfortunately, criteria that permit confident a priori predictions about the utility of any particular model for any particular complex human attribute. So, the proof of the model is in its application. The matter calls for caution (perhaps cautious optimism) in making the interspecific connection, and, given the nonexistence of gold standard model situations, for seeking converging evidence from different model systems. Our understanding, indeed, our definition of these phenomena, will evolve as information from the different avenues is collated and integrated.

For many demographic or epidemiological purposes, the target phenomenon of the model will have a complex genetic architecture with a large number of genes affecting the phenotype. For these circumstances, the quantitative genetic model is the apposite one. The pertinent analytical procedures yield bottom-line assessments of the relative influence of the genetic system and of various environmental domains. Molecular genetics has made enormous strides in the identification and molecular characterization of single genes that, by themselves, exert detectable influence on a wide array of medical and other phenotypes. Furthermore, the generation of marker genotypes permits the localization of some of the genes (QTL) in polygenic systems to general chromosomal regions. Thus both identified genes and implicated anonymous genes can be investigated.

The genetic tools that are available to the animal researcher are many and varied, and they are differentially useful and powerful for different purposes. The important phenomena addressed in this volume deserve optimal utilization of these tools. To this end, it should be remembered that the logic of animal model research requires efficient communication between the worlds of the animal modelers and the human population researchers. In the real world, of course, these research communities develop their own traditions, literatures, and networks; communication is less than optimal between communities. The situation is reminiscent of the current concern over “translational” research from the bench to the bedside. It is reasonable to expect that a systematic program to improve this two-way communication could enhance the relevance of studies on our phyletic relatives to the understanding of ourselves.


  • Austad SN. The comparative perspective and choice of animal models in aging research. Aging and Clinical Experimental Research. 1993;5:259–267. [PubMed: 8297929]
  • Bonner JT. The Evolution of Complexity. Princeton, NJ: Princeton University Press; 1988.
  • Coleman DL, Hummel KP. Influence of genetic background on the expression of mutations at the diabetes locus in the mouse. II. Studies on background modifiers. Israeli Journal of Medical Science. 1975;11:708–713. [PubMed: 1099054]
  • Crabbe JC, Belknap JK, Buck KJ. Genetic animal models of alcohol and drug abuse. Science. 1994;264:1715–1723. [PubMed: 8209252]
  • Falconer DS, Mackay TRC. Introduction to Quantitative Genetics. 4th ed. Essex, England: Longman; 1996.
  • Festing MFW. Inbred Strains in Biomedical Research. London: The Macmillan Press Ltd.; 1971.
  • Finch CE. Longevity, Senescence and the Genome. Chicago: The University of Chicago Press; 1990.
  • Fowler RE, Edwards RG. ‘Midget,’ a new dwarfing gene in the house mouse dependent on a genetic background of small body size for its expression. Genetical Research. 1961;2:272–282.
  • Harrison DE, Roderick TH. Selection for maximum longevity in mice. Experimental Gerontology. 1997;32:65–78. [PubMed: 9088903]
  • Kaplan A. The Conduct of Inquiry: Methodology for Behavioral Science. New York: Chandler, Harper & Row; 1964.
  • Kauffman SA. The Origins of Order. New York: Oxford University Press; 1993.
  • Lerner IM. Genetic Homeostasis. New York: John Wiley & Sons; 1954.
  • McClearn GE. The Genetic Basis of Alcohol and Drug Actions. New York: Plenum Press; 1991. The tools of pharmacogenetics.
  • McClearn GE. Heterogeneous reference populations in animal model research in aging. ILAR Journal. 1998;38:119–123. [PubMed: 11528053]
  • McClearn GE. Commentary. Exotic mice as models for aging research: Polemic and prospectus by R. Miller et al. Neurobiology of Aging. 1999;20:233–236. [PubMed: 10537032]
  • McClearn GE, Deitrich RA, Erwin VG, editors. Development of animal models as pharmacogenetic tools. Washington, DC: U.S. Government Printing Office; 1981. (DHEW Publication No. [ADM] 81-113)
  • McClearn GE, Hofer SM. Genes as gerontological variables: Uniform genotypes. Neurobiology of Aging. 1999;20:95–104. [PubMed: 10537018]
  • McClearn GE, Hofer SM. Genes as gerontological variables: Uses of genetically heterogeneous stocks. Neurobiology of Aging. 1999;20:147–156. [PubMed: 10537024]
  • McClearn GE, Vandenbergh DJ. The structure and limits of animal models: Examples from alcohol research. ILAR Journal. 2000;41(3):144–152. [PubMed: 11406706]
  • McClearn GE, Wilson JR, Meredith W. Contributions to Behavior-Genetic Analysis: The Mouse as a Prototype. New York: Appleton-Century-Crofts; 1970. The use of isogenic and heterogenic mouse stocks in behavioral research.
  • Miller RA, Austad S, Burke D, Chrisp C, Dysko R, Galecki A, Jackson A, Monnier V. Exotic mice as models for aging research: Polemic and prospectus. Neurobiology of Aging. 1991 Mar-Apr;20:217–231. [PubMed: 10537031]
  • Miller RA, Burke D, Nadon N. Announcement: Four-way cross mouse stocks: A new, genetically heterogeneous resource for aging research. Biological Sciences. 1999;54A:B358–B360. [PubMed: 10496542]
  • Mitton JB. Selection in Natural Populations. New York: Oxford University Press; 1997.
  • Phelan JP, Austad SN. Selecting animal models of human aging: Inbred strains often exhibit less biological uniformity than F1 hybrids. Journal of Gerontology. 1994;49(1):B1–11. [PubMed: 8282970]
  • Rapaport A. Operational Philosophy: Integrating Knowledge and Action. New York: Harper and Brothers Publishers; 1954.
  • Rubin GM, Yandell MD, Wortman JR, et al. Comparative genomics of the eukaryotes. Science. 2000;287:2204–2215. [PMC free article: PMC2754258] [PubMed: 10731134]
  • Sattler R. Biophilosophy: Analytic and Holistic Perspectives. Berlin: Springer-Verlag; 1986.
  • Shmookler Reis RJ, Ebert RH II. Genetics of aging: Current models. Experimental Gerontology. 1996;31:69–81. [PubMed: 8706806]
  • Sullivan PM, Mezdour H, Aratani Y, Knouff C, Najib J, Reddick RL, Quarfordt SH, Maeda N. Targeted replacement of the mouse apolipoprotein E gene with the common human APOE3 allele enhances diet-induced hypercholesterolemia and atherosclerosis. The Journal of Biological Chemistry. 1997;272(29):17972–17980. [PubMed: 9218423]
Copyright © 2001, National Academy of Sciences.
Bookshelf ID: NBK110040
PubReader format: click here to try


Related information

  • PMC
    PubMed Central citations
  • PubMed
    Links to pubmed

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...