Animal Models of Psychiatric Disorders

Mark A. Geyer and Athina Markou

This chapter critically discusses the process of developing, validating, and working with animal models relevant to psychiatric disorders. A model is defined as any experimental preparation developed for the purpose of studying a condition in the same or different species. Typically, models are animal preparations that attempt to mimic a human condition, including human psychopathology. In developing and assessing an animal model, it is critical to consider the explicit purpose intended for the model, because the intended purpose determines the criteria that the model must satisfy to establish its validity. Hence, before discussing the criteria by which the validity of an animal model may be assessed, the variety of purposes for which an animal model may be used to increase our understanding of a psychiatric disorder or its treatment will be considered.

PURPOSES OF AN ANIMAL MODEL

At one extreme, one can attempt to develop an animal model that mimics a psychiatric syndrome in its entirety. To do so, one must establish homology between the behavior of the affected animal and the syndrome being modeled. Typically, the signs and symptoms which are characteristic of the particular syndrome in humans are identified and enumerated. The ability of the experimental manipulation to induce homologous changes in the behavior of the test animal is then determined. In the early years of neuropsychopharmacology, the term animal model often denoted such an attempt to reproduce a psychiatric disorder in a laboratory animal. As will be illustrated, this approach is fraught with difficulty, in part because it commonly relies upon arguments of apparent similarities (i.e. face validity) that are difficult to defend and in part because the symptoms that define a given psychiatric disorder are often defined in subjective terms. Furthermore, the defining symptoms of psychiatric disorders and even the diagnostic categories themselves are being revised and re-defined continuously. It is not uncommon for diagnostic categories to be split into multiple new entities or combined into a new category as clinical experience reveals the limitations of previous categorizations.

At the other extreme, one more limited purpose for an animal model is to provide a way to systematically study the effects of potential therapeutic treatments. In such a case, the model may or may not mimic the actual psychiatric disorder. Rather, the model is only intended to reflect the efficacy of known therapeutic agents, and thus lead to the discovery of new pharmacotherapies. Because the explicit purpose of the model is to predict treatment efficacy, the principle guiding this approach has been termed "pharmacological isomorphism" (53). As illustrated below and discussed elsewhere (53, 69), the fact that such models are developed and validated by reference to the effects of known therapeutic drugs frequently limits their ability to identify new drugs having different chemical structures or novel mechanisms of action. Accordingly, an inherent limitation of this approach is that it is not designed to identify new therapeutics that might treat the symptoms of the disorder which are refractory to current treatments. An example of such a limitation is found in the use of drug-discrimination paradigms used to identify new treatment compounds. In these paradigms, the animal is trained to recognize the drug state induced by a prototypical drug. Typically, the animal is required to press either the right or left lever, depending upon whether it had been treated with saline or the training drug. Potential new therapeutics are then identified by their ability to substitute for the prototypical drug on which the animal was trained. Because these paradigms rely only on the subjective drug-induced cue to which each animal responds and not on any endpoint that can be validated by reference to other behaviors in animals or humans, such procedures can only identify drugs having a similar effect on some unknowable dimension.

Because of the complexity and evolving nature of diagnostic categories in psychiatry, another approach to the development of animal models relies on mimicking only specific signs or symptoms associated with psychopathologic conditions, rather than mimicking an entire syndrome. In such cases, specific observables that have been identified in psychiatric populations provide a focus for study in experimental animals. The particular observable behavior being studied may or may not be pathognomonic for or even symptomatic of the disorder, but must be defined objectively and measured reliably. Thus, even phenomena that are found in more than one diagnostic entity may be studied in animals using this approach. It is important to emphasize that the reliance of such a model on specific observables minimizes a fundamental problem plaguing animal models of psychopathology designed to model diagnostic syndromes. Specifically, the number of definitive clinical findings with which one can validate an animal model have been limited by the difficulties inherent in conducting experimental studies of psychiatric populations. As illustrated below, the validation of any animal model can be only as sound as the information available in the relevant clinical literature (69). With this approach, the investigator may generate more definitive information related to a more circumscribed domain of psychopathology. By limiting the purpose of the animal model, one can increase the confidence in the cross-species validity of the model. The narrow focus of this approach generally leads to pragmatic advantages in the conduct of mechanistic studies addressing the neurobiological substrates of the specific behavior in question. By contrast, in models intended to reproduce entire syndromes, the need for multiple simultaneous endpoints makes it relatively difficult to apply the invasive experimental manipulations that are typically required to establish the underlying mechanisms for all the symptomatic behaviors.

Another approach to the development of animal models is more theoretically based upon psychological constructs which are thought to be affected by the psychiatric disorder under investigation. Such an identification of what has been termed psychological processes or behavioral dimensions (53, 69) involves the definition of a hypothetical construct and subsequent establishment of operational definitions suitable for the experimental testing of the validity of the construct. This approach is most fruitful when conceptually related experiments are undertaken in both the relevant patient population and in the putative animal model. That is, studies of appropriate patients are needed to establish the operational definitions of the hypothetical construct, and the construct's relevance to the particular disorder. In concert, parallel studies of the theoretically homologous construct, process, or dimension are required to determine the similarity of the animal model to the human phenomena. An important and advantageous aspect of this approach is that the validation of the hypothetical construct and its cross-species homology can be established by studies of normal humans and animals, in addition to psychiatrically disordered patients or experimentally manipulated animals. Thus, this approach adds to and benefits from the psychological and neuroscientific literature relevant to the hypothetical construct upon which the model is based. In a sense, this approach explicitly recognizes that the experimental study of the disorder in humans involves as much of a modelling process as does the study of the disorder in an animal model.

CRITERIA FOR EVALUATING ANIMAL MODELS

In this section, the criteria for evaluating animal models will be explored. The ultimate need for reliance on predictive validity will be emphasized. Validation criteria are general standards that are relevant to the evaluation of any model. Over the years, several reviews have discussed criteria to be used in the evaluation of animal models (e.g. 34, 51, 53, 54, 69). Most of these discussions are based on an assumption that is not always made explicit. Namely, it is commonly assumed that there is homology, or at least analogy, among the physiological and behavioral characteristics of various species; hence, extrapolations can be made from animals to humans (66). Having accepted this postulate, the question remains as to the necessary and sufficient validation criteria for any animal model. In general, as mentioned above, the validity of a model refers to the extent to which a model is useful for a given purpose. In neurobiological research, the purpose of a model is, most commonly, the elucidation of the mechanisms underlying the human condition. In the following sections, it will be argued that there are only two criteria that a model must satisfy to establish its value in basic neurobiological research: reliability and predictive validity.

In considering the criteria relevant to an animal model, it is important to recognize that a model consists of both the independent variable, i.e. the manipulaton used to induce the abnormalilty, and the dependent measure, i.e. the measure(s) used to assess the effects of the manipulation. The criteria relevant to the independent variable include face, etiological, and construct validity, while those relevant to the dependent variable include construct, convergent, discriminant, and face validity. The predictive validity and reliability of the model system are relevant to both the independent variable and the dependent measures. These considerations are discussed below.

Reliability

Reliability refers to the consistency and stability with which the variables of interest are observed, and is relevant to both the independent and the dependent variables. This consistency should be evident at the following levels: a) ability to manipulate the independent variable with a high degree of precision; b) ability to measure the dependent variable objectively; c) small within-subject variability of the dependent variable; d) small between-subject variability of the dependent variable; e) reproducibility of the phenomenon under maximally similar conditions; and f) reproducibility of the effects of manipulations (51). Having a reliable and reproducible experimental system is essential to scientific study. Nevertheless, although small within- and between-subject variability is usually desirable, it should be emphasized that there are cases in which the study of the variability of the model system could lead to a better understanding of the phenomenon. Variability cannot always be considered as error.

Types of Validity

Many different types of validity have been described and defined (12, 14,57), including: predictive; construct; concurrent or convergent; discriminant; etiological; and face validity. Depending on the desired purpose of the test that one wishes to validate, different types of validity are relevant. Because the primary purpose of animal models is to enhance our understanding of a human phenomenon, the question of which criteria should be used for evaluating the usefulness of models is equivalent to the philosophical question of what scientific understanding is. As formalized by the logical positivists of the Vienna circle, scientific understanding is the ability to predict (9). That is, a phenomenon is defined by its correlations with antecedent or consequent phenomena (4). Logical positivism takes the position that the only valid scientific observables are correlations (i.e., one event preceding the other always in the same order), while causality is an inference that is beyond the evidence and can never be shown to be true or false (36). Hence, the only meaningful evaluating criterion for an animal model is the model's ability to lead to accurate predictions, that is, whether the model has predictive validity. Even though other types of validity, such as construct, etiological, convergent, discriminant, and face validity, are relevant to animal models, and the establishment of such types of validity has heuristic value, predictive validity and reliability are the only necessary and sufficient criteria for the initial evaluation of any animal model that will justify the model’s use in research. Undoubtedly, the more types of validity a model satisfies, the greater its value, utility and relevance to the human condition. The establishment of multiple forms of validation provides convergent evidence in support of the postulate of cross-species homology. In addition, the processes of construct, etiological, convergent, and discriminant validation of an animal model are integral components of scientific theory development and testing. Thus, it could be considered circular logic if a model were required to satisfy all types of validity before being considered useful. Consequently, the initial and continued reliance on the satisfaction of the reliability and predictive validity criteria has to be emphasized. The arguments for this position will be made in the following sections as the other types of validity are defined and discussed in relation to animal models.

Predictive Validity

Predictive validity is generally defined as the ability of a test to predict a criterion that is of interest to the investigator (14). For animal models, the criterion is the human phenomenon. Thus, an animal model has predictive validity to the extent that it allows one to make predictions about the human phenomenon based on the performance of the model. Therefore, development of animal models requires parallel development of clinical measures that allow meaningful comparisons. In reference to animal models of human psychopathology, the term predictive validity is often used in a narrow sense to refer to the model's ability to identify drugs with potential therapeutic value in humans (i.e. pharmacological isomorphism, 53,87). Although this use of the term is not incorrect, it is limited because it ignores other important ways in which a model can lead to successful predictions (69). For example, the identification of any variables that influence both the experimental preparation and the modeled phenomenon in similar ways or correlations with other measures can enhance one's understanding of the phenomenon.

Construct Validity

Construct validity of a test is most commonly defined as the accuracy with which the test measures that which it is intended to measure (14). Although construct validity is considered by investigators in a variety of fields as the most important property of a test or measure (14, 87), it can be established only rarely. The process of construct validation of a test is not different in any essential way from the general scientific procedure for developing and testing theories (14), and thus for developing animal models. Conceptions about what a test is supposed to measure or a model is supposed to mimic are constantly changing as scientific theories and theoretical constructs are modified. Thus, a model's usefulness, and hence its overall validity, cannot be determined simply by the degree of construct validation that it has. Nevertheless, the process of construct validation is valuable in the never-ending process of further development and refinement of the model. As new experimental and observational evidence accrues from both the animal model and the clinical conditions, the model is refined and therefore enables more accurate predictions.

Etiological Validity

The concept of etiological validity is closely related to the concept of construct validity. A model has etiological validity if the etiologies of the phenomenon in the animal model and the human condition are identical. As such, assessing etiological validity involves an evaluation of the inducing conditions, and the implicit or explicit hypotheses about the etiology of the disease. Accordingly, like construct validation, the process of etiological validation is a fundamental component of scientific investigation. When etiological validity can be established, the model can become extremely useful in the development of treatments. The limitations of treatment-oriented models based on pharmacological isomorphism, alluded to above, can be overcome if an etiologically based model is found. Unfortunately, the etiologies of psychiatric disorders are seldom known. Hence, etiological validity in this context is generally limited to hypotheses regarding a possible etiology. Indeed, the purpose for the development of animal models is often to enable the identification of or to test a hypothesis about the etiology of the disease. In such an instance, one could not require that a model has etiological validity for the model to be deemed useful. The recent surge of interest in genetic contributions to psychiatric disorders promises to enable the development of a class of animal models based upon hypothesized etiological validity. As specific genes and gene products become linked to specific disorders, molecular biologists will be able to generate mutant or transgenic animals having potentially homologous genetic abnormalities. Behavioral and pharmacologic studies of these genetically engineered animals will then be important in identifying the phenotypic changes associated with the mutation, testing hypotheses about the etiology of the disease, and exploring potential therapeutic treatments. The combination of genetic and molecular biological approaches with behavioral and pharmacological approaches may well revitalize interest in etiologically based models of psychiatric disorders.

Convergent and Discriminant Validity

A model or a test has convergent (or concurrent) and discriminant validity only in relation to other models, tests, or measures. Convergent validity is the degree to which a test correlates with other tests that attempt to measure the same construct (12). Discriminant validity is the degree to which a test measures aspects of a phenomenon that are different from other aspects of the phenomenon that other tests assess (12). Discriminant validity is indicated by low correlations between the measures provided by the various models. With respect to this form of validity, a model is invalid to the extent that it is highly correlated with other models from which it was intended to differ. As such, convergent and discriminant validity primarily refer to the evaluation of the dependent measures. Both convergent and discriminant validity are required for the establishment of construct validity. It has been argued that convergent operationalism (also called converging measures) should replace single operationalism in behavioral sciences (12). Convergent operationalism refers to the process of defining a construct by a multiplicity of methods or operations. The concern is that a single operation might not measure what one wishes to measure, and therefore lead to erroneous conclusions. With convergent operationalism, one could approximate better the measurement of the construct of interest. The evaluation of convergent validity is central to the approach described above in which a model is based theoretically upon a psychological process or behavioral dimension that is hypothesized to be affected in a psychiatric disorder.

Face Validity

Face validity refers to the phenomenological similarity between the behavior (i.e., dependent variable) exhibited by the animal model and the specific symptoms of the human condition (57). Although face validity appears to be a desirable and is certainly an intuitively appealing criterion with which to validate models (18,88), such a criterion is actually not necessary, can be misleading, and is difficult to defend rigorously. Because the majority of models involve a species other than the one whose condition one tries to mimic, it is unrealistic to expect the two species to exhibit similar symptoms or phenomenology, even in cases where the etiology of the condition is known (53,69). In contrast, finding similarities between certain aspects of the behavior or physiology of animals and humans does not necessarily implicate similar etiology (1,34). Moreover, establishing the face validity of a model objectively is impossible, because claims for face validity of a model almost invariably involve subjective arbitrary arguments that are not necessarily accepted by all investigators in the field (see 87 for examples of models with proposed face validity). While face validity can provide a heuristic starting point for the development of an animal model, it cannot be used to establish the validity of the model. In summary, the face validity of most models of human psychopathology is difficult to establish objectively and is irrelevant to the potential usefulness of the model in understanding the disease. It should be understood that face validity refers to the superficial similarity in symptomatology between the model and the disorder (e.g. changes in appetite or activity levels which are dependent variables) and can be distinguished from construct validity, which relies on similarities in underlying processes or mechanisms (e.g. hypothetical constructs specifying the mechanisms that lead to changes in appetite or activity levels). Thus, while face validity does not detract from an animal model, it simply does not provide scientific support for a model.

ANIMAL MODELS

In the following sections, selected examples of animal models used in the contexts of investigating depression and schizophrenia will be discussed. These discussions will offer neither a compendium of current animal models, nor thorough reviews of the selected models. Rather, selected aspects of these models will be discussed to illustrate the principles, approaches, and concerns that should be contemplated in developing and validating a model.

Animal Models Of Depression

The Learned Helplessness Model

This model is based on the observation that exposure to uncontrollable stress produces performance deficits in subsequent learning tasks that are not seen in subjects exposed to identical stressors that are under the subjects' control (48). This paradigm is an example of a model intended to reproduce a psychiatric syndrome in its entirety, in which claims of face validity and pharmacological isomorphism are offered as the principal support. It has been argued that the model has good face validity because there is similarity between the behavioral characteristics of learned-helpless animals and signs of depression in humans (for critical reviews, see 87, 88). For example, learned-helpless animals exhibit loss of appetite and weight, decreased locomotor activity, and poor performance in both appetitively and aversively motivated tasks. These behavioral characteristics of learned-helpless animals are considered equivalent to loss of appetite and weight, psychomotor retardation, and anhedonia, respectively, demonstrated by depressed humans (DSM-IV). As discussed above, however, symptom similarity (i.e. face validity) does not necessarily imply homology (1,34). Thus, these claims for the model's face validity do not address the question of whether the model has any utility. The predictive validity in terms of pharmacological isomorphism of the model is indicated by the fact that pharmacological treatments clinically effective in depression, such as tricyclics, MAO inhibitors, atypical antidepressants, and electroconvulsive shock therapy are effective in reducing the behavioral and "physical" abnormalities seen in animals exposed to uncontrollable stress (87, 88). Thus, the model appears to have good predictive validity in terms of identifying potentially useful pharmacotherapies for depression. Nevertheless, the model's predictive and discriminant validity have some limitations in that false positives have also been reported (87). The model's predictive and construct validity, and thus its potential to further our understanding of the neurobiology of depression, depends on the validity of two assumptions, which have been debated extensively (88). The first assumption is related to the process that is assumed to occur while the animals are exposed to uncontrollable stressors, namely, that animals learn that they have no control and therefore become helpless. Alternatively, subjects might learn to become inactive (29). Evidence indicates that both learned helplessness and inactivity might be learned by animals in these paradigms (47). Because these two processes are confounded in most learned helplessness experiments (88) (i.e. lack of construct and discriminant validity), this paradigm might not be the best procedure to investigate either of these two processes. The second assumption of the model is that learned helplessness is exhibited by depressed individuals. Very little evidence supports this assumption, however (8). Clinical observations describing depressed patients as exhibiting behavioral despair or negative cognitive sets do not constitute experimental evidence for the existence of the specific psychological process of learned helplessness, on which the model is based. The absence of explicit support for the occurrence of this process in depressed humans seriously questions the construct validity of the paradigm as a model of depression. The discussion up to this point has been in regards to the dependent variables measured in the learned helplessness model and the constructs that these measures are purported to assess. When evaluating the inducing conditions for the model, limitations of the model are also evident. For instance, presumably "normal" humans do not unambiguously develop learned helplessness under conditions of uncontrollable stress (10), and only 10-50% of rats develop this syndrome (17). These latter findings may dispute the reliability of the phenomenon and its generality across species or alternatively may indicate genetic predispositions to depression that would be worth investigating further. In summary, the learned-helplessness paradigm's usefulness as a model of depression is limited severely by the lack of evidence that learned helplessness is a behavioral process characterizing depressed individuals (i.e. lack of predictive and construct validity). The advantage of the model is its pharmacological isomorphism that suggests some limited predictive validity of the model.

The Behavioral Despair Model

A variant of the learned helplessness model is the behavioral despair paradigm. In this model, mice or rats are forced to swim in a confined environment. The animals initially swim around and attempt to escape, and eventually assume an immobile posture. On subsequent tests, the latency to immobility is decreased. In a modification of this paradigm, animals are first exposed to uncontrollable stress before the swim test (84). These paradigms are conceptually similar to the learned helplessness paradigm in assuming that after uncontrollable stress, animals have learned to "despair" (i.e., learned helplessness). As such, the behavioral despair model involves conceptually similar inducing conditions and dependent variables, and thus has the potential of providing convergent support for the construct of learned helplessness. Unfortunately, studies of subjects tested in both paradigms indicated a lack of cross-predictability (i.e. lack of convergent validity), which suggests that the two models may assess different processes that are mediated by separate mechanisms (17,59). Furthermore, the interpretation of failure to cope attributed to the immobility observed during forced swimming has been questioned. It has been suggested instead that the immobility may reflect a successful strategy that conserves energy and allows animals to float for prolonged periods of time, and hence survive longer (85). Further, some evidence suggests that the effect of tricyclic antidepressants on reducing the immobility might be, at least partly, attributable to altered learning processes, and not to reduced "despair" (85). Nevertheless, the behavioral despair model has one of the highest degrees of pharmacological isomorphism (i.e. predictive validity) in terms of identifying antidepressants (16,85). As with learned helplessness, however, the degree of pharmacological isomorphism is limited by false positives. Under certain circumstances, even saline acts as a false positive (85), which constitutes a failure to demonstrate good discriminant validity. More generally, the ability of acute antidepressant treatments to reverse the immobility when the same compounds are effective in humans only after chronic treatment may represent a partial failure to demonstrate good predictive validity. Nevertheless, considerable predictive validity in terms of pharmacological isomorphism has been demonstrated.

One common aspect of the two models discussed above that furnishes them with construct, etiological, and predictive validity is the fact that neurochemical as well as behavioral changes are produced by exposing subjects to chronic unpredictable stress. These neurochemical changes involve dysfunctions of the noradrenergic (84) and serotonergic (42) systems on which most classic antidepressants have their primary pharmacological actions (52). Extensive neuropharmacological investigations indicated that in chronically stressed animals the locus coeruleus, from which most noradrenergic projections to the forebrain originate, is disinhibited (84). Therapeutic doses of antidepressants (except selective serotonin re-uptake inhibitors) or electroconvulsive shock therapy downregulate beta-adrenergic receptors, presumably normalizing the dysregulated function of noradrenergic neurotransmission that has been hypothesized to be one of the neurotransmitter imbalances mediating depression (52). Similarly, forced swim, the stressor used in the behavioral despair paradigm, has been shown to alter dialysate serotonin levels in several brain regions (42), while chronic treatment with antidepressants has been consistently shown to increase serotonergic neurotransmission (52). Even though the exact parameters of stress (e.g., frequency, severity, duration, repetitiveness) necessary to induce the behavioral and neurochemical syndromes of interest have not been investigated systematically, the above reviewed findings taken together suggest that the etiology of the syndrome in animals may be the same as the etiology of some types of depression in humans. It has been hypothesized that stress predisposes humans to depression (33). Even though some human data support this hypothesis, well-controlled clinical investigations are needed to more extensively and unambiguously define the role of stress in depression. Such clinical investigations would assist in the development and validation of animal models related to human depression. In summary, the degree of overall construct validity of the behavioral despair paradigm as a model of depression is rather limited because neither the process of learned helplessness nor the process of behavioral despair have been demonstrated satisfactorily in depressed individuals. Nevertheless, the stress-induced alterations in the monoamine neurotransmitters believed to be involved in the actions of antidepressants indicates some etiological and construct validity for this model. Finally, this model has been demonstrated to have a high degree of predictive validity in terms of pharmacological isomorphism.

Reward Models: Intracranial self-stimulation, sucrose preference, and place preference paradigms

These paradigms have been proposed to model a specific psychological process (i.e., reward ) that appears to be altered in depressed individuals, and is also included as a negative symptom of schizophrenia (DSM-IV). Thus, these paradigms are not considered models of an entire syndrome, but rather provide operational measures of anhedonia, a core feature of depression and a negative symptom of schizophrenia. Anhedonia in humans is defined as "the markedly diminished interest or pleasure in all, or almost all, activities most of the day, nearly every day" (DSM-IV). The discussion of the intracranial self-stimulation (ICSS), sucrose consumption, and place preference paradigms will demonstrate the importance of evaluating separately the independent variable (i.e., inducing conditions) from the dependent variable that provides the operational measure of the symptom of interest. The ICSS paradigm involves brief electrical self-stimulation of specific brain sites, which is very reinforcing as indicated by the fact that animals will work for it. Both the rate of responding for ICSS and the psychophysically defined threshold(s) for ICSS have been used as measures of the reward value of the stimulation (50). Substantial evidence confirms that ICSS thresholds provide reliable measures of reward (50) that reflect a broad continuum from hedonia to anhedonia. Lowering of thresholds is interpreted as an increase in the reward value of the stimulation, whereas elevation of thresholds is interpreted as a decrease in the reward value of the stimulation. Because it appears that ICSS acts directly on some of the same neuronal substrates that mediate the rewarding effects of natural reinforcers such as food and water, it is considered a valuable tool for the investigation of the function of brain reward systems. Further, by directly stimulating the putative reward circuit(s), one presumably bypasses the input side of the system, thus eliminating non-specific effects associated with consummatory behaviors that can complicate data interpretation. Also, the threshold measures provided by ICSS procedures are easily quantifiable and extremely reliable in that ICSS threshold estimates are stable over periods of several months.

The study of the neurosubstrates of ICSS behavior after experiential or pharmacological manipulations promises to enhance our understanding of reward mechanisms that seem to be altered in several psychiatric disorders, including depression and schizophrenia (DSM-IV). Two manipulations have been used to produce an "anhedonic" state in animals, as operationally defined by decreases in ICSS response rates or elevations of thresholds: a) exposure to uncontrollable stress (56,91); and b) withdrawal from long-term exposure to drugs of abuse (e.g.20,46,49,52). In the first of these procedures, rats or mice exposed to either swim stress or uncontrollable, but not controllable, footshock or chronic mild stress exhibit protracted decreases in the reward value of ICSS (56, 91). These stress-induced alterations in ICSS behavior were reversed by repeated antidepressant treatment, (e.g., 56, 91), indicating predictive validity of the model in terms of pharmacological isomorphism. The phenomenon appears to be reliable because it has been replicated several times (56). This model also appears to have considerable construct validity insofar as it addresses a specific process that appears to be impaired in depressed individuals and enables the investigation of the neurobiology of this phenomenon. Potentially, the paradigm may have etiological validity by inducing an "anhedonic" state through a mechanism (i.e., exposure to uncontrollable stress) that has been hypothesized to play a critical role in inducing depression in humans (33).

Converging evidence for stress-induced dysfunctions in reward processes is provided by two other measures of the hedonic value of stimuli. In the first case, rats are exposed chronically to a series of mild stressors and their consumption of a sweet saccharin or sucrose solution is monitored. Stressed animals tend to consume less sweet solution than controls (although the phenomenon has been unreliable; 90), suggesting an induction of a mild "anhedonic" state by stress. This effect of stress was reversed by antidepressant treatment, but not by antipsychotic, anxiolytic, amphetamine, or morphine treatment (90), indicating good predictive validity in terms of pharmacological isomorphism. Nevertheless, the validity of volummetric measures of sucrose consumption as a hedonic measure has been seriously questioned. Specifically, the amount of sucrose consumption is highly correlated with body weight, which is often lower in stressed animals compared to controls, especially when food deprivation is one of the "stress" manipulations. When correction for body weight changes were made or food restriction was not included as one of the "stress" manipulations (22), there were no differences in sucrose consumption between stressed and control animals. Accordingly, sucrose preference, in contrast to total sucrose consumption, may be a more appropriate reward measure. Nevertheless, the multiple failures to consistently demonstate the reductions in sucrose intake under similar experimental conditions (90) indicate the poor reliability of the phenomenon. Taken together, the above reviewed data severely limit the usefulness of the model when chronic mild stress is the inducing condition and sucrose consumption is the dependent variable, demonstrating the need to separately evaluate dependent and independent variables of a model. In the second case, the effects of the same series of mild stressors on conditioned place preference measures of reward are investigated. This paradigm involves repeated pairing of a rewarding stimulus, such as food or drug, with a previously neutral environment. During testing, control animals prefer the environment previously paired with reward over another neutral environment. The degree of place preference is considered a measure of reward. Exposure to chronic mild stress prevented or attenuated the formation of an association between rewarding stimuli and a previously neutral environment, indicating decreased sensitivity to rewards after stress manipulations (90). In conlusion, the experimental data from the ICSS, sucrose preference, and place preference paradigms provide converging evidence that stress induces abnormalities in reward processes. As discussed above, for each paradigm the inducing conditions and the dependent measure(s) have to be evaluated separately. For instance, the relevance of such stress-induced abnormalities to depressive illnesses in humans needs to be established to etiologically validate these effects as models of depression.

The second inducing condition that results in elevations in ICSS thresholds is withdrawal from prolonged exposure to several drugs of abuse, such as cocaine, amphetamine, nicotine, morphine, or ethanol (20,46,49,52). For example, when rats were withdrawn after being allowed to self-administer cocaine for prolonged periods of time, their thresholds for ICSS reward were elevated in proportion to the amount of cocaine consumed during the cocaine "binge" (49). Because the procedure allowed rats to self-administer cocaine at doses and patterns that were reinforcing, rather than aversive or stressful, the subsequent anhedonia cannot be attributed to the effects of uncontrollable stress but instead appears attributable to the pharmacological effects of cocaine. Similar elevations in ICSS thresholds have also been observed after repeated treatment with experimenter-administered cocaine (43) or amphetamine (46), confirming the reliability of the phenomenon. Evidence for the predictive validity of this model derives from the finding that the post-cocaine elevations in thresholds were reversed by administration of either the dopamine agonist bromocriptine or the tricyclic desmethylimipramine (52). These compounds have been used in the treatment of stimulant-induced withdrawal and dependence. Even though the relationship between drug- and non-drug-induced depressions is not yet clear, the evidence that clinically proven antidepressant treatments reverse reward deficits associated with drug withdrawal supports the hypothesis that common neurobiological abnormalities mediate these two types of depression (52), and suggests potential etiological validity of the drug-withdrawal-induced elevations in thresholds as a model of non-drug-induced depressions. In humans, withdrawal from several drugs of abuse is characterized by the symptom of anhedonia, which is also one of the core symptoms of non-drug induced depression (52). Thus, although many more studies are needed, the evidence to date indicates that the model involving drug withdrawal as the inducing condition and ICSS as the dependent measure has good construct and etiological validity and exhibits some pharmacological isomorphism (i.e. predictive validity) as a model for drug- and potentially non-drug-induced anhedonia.

Genetic Models

Lines of rats were selectively bred for hypo- or hyper-sensitivity of specific receptor subtypes whose altered function has been hypothesized to be involved in the etiology of depression. Hence, these models allow the testing of hypotheses about the etiology of depression, or specific depressive symptomatology, assuming these animals show abnormalities that are also seen in depressed humans. Based on the hypothesis that depression involves a cholinergic-noradrenergic neurotransmitter imbalance, a putative genetic animal model of depression has been developed by selectively breeding rats (Flinders Sensitive Line rats) to exhibit increased thermic responses to a muscarinic cholinergic agonist (60). Among common characteristics of these rats with depressed humans is their decreased sensitivity to rewarding stimuli, as indicated by decreased intake of a sucrose solution after exposure to chronic mild stress. Other interesting characteristics of these rats that are also seen in depressed individuals are short REM sleep latency and fast REM-REM cycle. Thus, supersensitivity of cholinergic muscarinic receptors may be involved in the reduced sensitivity to reward and the REM sleep abnormalities seen in depressed individuals. In summary, this model has potential etiological validity and some construct validity as indicated by the anhedonia and REM sleep abnormalities seen in these animals.

Another recently introduced animal model of depression based on selective breeding involves hyposensitivity of the thermic response to a 5-HT_1A agonist (61). The development of this line of rats is based on findings that several antidepressant drugs decrease 5-HT_1A autoreceptor function in the raphe nuclei and that some partial 5-HT_1A agonists are effective antidepressants (52). As with the Flinders Sensitive Line rats and chronically stressed animals, these 5-HT_1A hyposensitive rats may also exhibit decreased sensitivity to reward since they have been shown to consume less of a sweet solution in a two-bottle choice paradigm than controls, without showing changes in anxiety-like behaviors (31). Nevertheless, the 5-HT_1A hyposensitivity has been attributed to a decreased number of 5-HT_1A receptors in cortical areas, while no differences were found in the raphe region. Because this model has been introduced only recently, additional studies are needed to further examine potential changes in 5-HT_1A receptor function in these animals and test the hypothesis that 5-HT_1A receptors are involved, not only in the mode of action of antidepressants, but also in the etiology of depression. Such investigations would assist in beginning to evaluate the overall validity of this relatively new model.

Animal Models Of Schizophrenia

The vast majority of animal models related to schizophrenia have been based upon the induction of abnormal behaviors by drugs considered to be psychotomimetics, including both psychostimulants and hallucinogens. It is important to recognize, however, that the drug-induced states in humans have similarities with the early stages of a range of psychotic disorders and not necessarily with the diagnostic syndrome of schizophrenia (39,69). More recently, research efforts have emphasized parallel studies of specific and potentially homologous behavioral phenomena that can be examined both in animals and in patients with schizophrenia. This emphasis on specific behaviors has proven to be more fruitful than attempts to model complex syndromes in their entirety. The other major emerging trend is that experiential, developmental, and genetic manipulations are beginning to replace the more traditional drug and lesion approaches to the induction of the relevant behavioral abnormalities.

The Psychostimulant Model

The most widely studied class of drug-induced models of schizophrenia is based on the behavioral effects of psychostimulant drugs such as amphetamine. Much of the impetus for this group of models derives from the similarities between the effects of amphetamine in humans and the symptoms of schizophrenia (face validity) or at least the early stages of psychotic disorders, broadly defined (39). The prevailing view that schizophrenia reflects a hyperactivity of brain dopaminergic systems (potential etiological validity) has also encouraged the use of this group of models. A wide range of behaviors has been studied in this context, using a variety of species. Typically, rodent studies have focussed on increases in locomotor activity or stereotypy, while studies of infrahuman primates have examined both motor effects and alterations in social interactions induced by amphetamine-like drugs (18). As many reviews have evaluated these models in great detail (e.g. 13), they will not be reviewed again here. Suffice it to say that these models have a considerable degree of predictive validity, particularly with respect to their having pharmacological isomorphism in predicting established dopaminergic treatments for schizophrenia. Cogent attempts have also been made to argue that some of these models exhibit construct validity as well (see below). One aspect of predictive validity which has received considerable attention involves the dosage regimens required for amphetamine or related drugs to produce psychotic-like behavior in psychiatrically healthy humans. Because of the widespread belief that amphetamine-induced psychosis is produced only by repeated exposure to the drug (19), many preclinical researchers have directed their attention toward the behavioral effects of amphetamine that are augmented or sensitized by repeated drug administrations. A review of the available clinical literature, however, strongly indicates that chronic exposure is not required and that psychotic episodes can be produced by acute administrations of amphetamine or related drugs (70). The complex and limited nature of the clinical data seems to have led to mistaken interpretations which have inordinately influenced a large proportion of the basic research in this area. While it is clear that tolerance does not occur to the psychosis-inducing effects of amphetamine in humans, it is not clear that sensitization is required for these effects. Hence, while an animal model based on the effects of chronic amphetamine could be invalidated if tolerance were observed, the development of sensitization does not provide evidence supporting a model's relevance to schizophrenia. Indeed, it appears that the animal models having the greatest amount of predictive validity are those based on the effects of the psychostimulant which are evident after acute administrations (13,70).

Arguments for the construct and etiologic validity of the psychostimulant models derive largely from the DA hypothesis of schizophrenia, as reviewed elsewhere (13,69). These models arose because of the apparent similarity (i.e. face validity) between the effects of high doses of amphetamine in presumably normal humans and the symptoms of schizophrenia (19). Cross-species studies in animals treated with psychostimulants revealed striking stereotyped or perseverative behaviors, which were seen as having face validity for the stereotyped behavior induced by amphetamine in humans (13,18,19). Despite the fact that stereotyped behavior was not seen as a key feature of schizophrenia, subsequent studies in rodents have focussed on the stereotypies produced by psychostimulants. Although the models that evolved from this approach have demonstrated considerable predictive validity in terms of pharmacological isomorphism, current thinking now indicates that the original appearance of face validity was actually somewhat misleading. In recent years, the DA hypothesis of schizophrenia has evolved into the narrower hypothesis that the mesolimbic DA system, as distinct from the nigrostriatal DA system, is most relevant to schizophrenia. The nigrostriatal DA system is now seen as most relevant to the dyskinetic side effects of antipsychotic treatments. The mesolimbic system appears to mediate the locomotor activating effects of low doses of amphetamine, while the nigrostriatal system mediates the stereotypies that predominate at higher doses (13). Thus, the stereotypies originally proposed to have the most face validity for the human condition now appear to be more closely linked neurobiologically to phenomena that are considered side effects of the clinical treatments. Because schizophrenic patients are not generally considered to be motorically hyperactive, the amphetamine-induced hyperactivity that is mediated by what is believed to be the most relevant neurobiological substrate has seldom been considered to have face validity as a model of the human disorder. Note that the failure of the model to have face validity has in no way weakened its utility in neurobiological research, which is based on its etiologic and predictive validity. Nevertheless, virtually any of the behavioral effects of amphetamine in rodents, including either locomotor hyperactivity or stereotypy, have high degrees of pharmacological isomorphism, a form of predictive validity, as models for the efficacy of DA antagonist treatments for schizophrenia (13,70); thus the predictive validity of these models appears limited to dopaminergic treatments.

The Hallucinogen Model

Soon after the discovery of the psychoactive properties of lysergic acid diethylamide (LSD) over 50 years ago, researchers began to explore the idea that the class of drugs represented by LSD might appropriately be called psychotomimetics or even psychotogens. This initial effort was engendered by the apparent face validity of the effects of LSD on perception and affective lability as being similar to the symptoms of the early stages of psychoses such as schizophrenia (5). Although many similarities were noted between the effects of the drugs in humans and the symptoms of schizophrenia (35,82), two major differences prompted the widely accepted, but not neccessarily justified, conclusion that this class of drugs does not provide a useful model of schizophrenia. First, tolerance was found to develop rapidly to the subjective effects of LSD-like drugs, while the symptoms of schizophrenia persist for a lifetime (82). Second, the hallucinations produced by LSD and related drugs are typically in the visual modality, while those characteristic of schizophrenia are in the auditory modality (35). These two observations weaken the predictive validity of the hallucinogen model as a model of the syndrome of schizophrenia.

Initial interest in hallucinogens was spurred by the possibility that abnormalities of normal biochemistry might lead to the endogenous production of such compounds and hence be responsible for some psychotic symptomatology. The transmethylation hypothesis in particular posited that serotonin could provide a substrate for the endogenous production of hallucinogens similar to N,N-dimethyltryptamine (DMT) (82). Initially, this etiologically based model was dismissed because of the rapid tolerance associated with traditional hallucinogens such as LSD and mescaline. More recent studies indicating that tolerance may not occur to the subjective and behavioral effects of DMT in humans (73) suggest that DMT may differ from other hallucinogens and that this model may still be viable. Indeed, different mechanisms may be involved in the various actions of the different hallucinogenic drugs, as suggested by the lack of cross-tolerance to DMT in human subjects made tolerant to LSD (65). Hence, further studies are warranted to provide the objective evidence needed to adequately evaluate the model of psychosis based on the hypothesis of an endogenous psychotogen.

Furthermore, it remains possible that these drugs may be psychotomimetic without being psychotogenic. That is, the study of hallucinogen effects in animals may have relevance as models of some aspects of psychotic episodes in humans. Recent suggestions of serotonergic abnormalities in schizophrenia (40) and of 5-HT_2A receptor contributions to the clinical efficacy of antipsychotics (55) have revitalized interest in this possibility. For example, hallucinogens are now believed to produce their characteristic subjective effects by acting as 5-HT_2A agonists (30). Many of the newer atypical antipsychotic drugs appear to be potent antagonists at 5-HT_2A receptors (55). Further, if one seeks to model a specific abnormality exhibited by patients with schizophrenia rather than the syndrome of schizophrenia, evidence indicates that the study of hallucinogen action may provide a fruitful line of investigation. For example, both schizophrenic and schizotypal patients are known to exhibit deficits in the habituation of startle (3,11,26). Habituation, the simplest form of learning, refers to the inhibition of responses to unimportant stimuli or cognitions and is, therefore, essential for the development of selective attention. Hallucinogenic 5-HT_2A agonists such as LSD and mescaline produce similar deficits in the homologous behavior in rats (26). Conversely, opposite behavioral effects are produced by 5-HT_2A antagonists (26,79), some of which are thought to be effective in the treatment of schizophrenia (55). Hence, the effects of hallucinogens in animals have some degree of predictive validity with regard to both specific abnormalities exhibited by patients with schizophrenia and the effects of antipsychotic drugs (25). As discussed further below, such a model has been argued to have construct validity based on the widespread evidence that both the symptoms of schizophrenia and the effects of hallucinogens reflect exaggerated responses to sensory and cognitive stimuli due theoretically to failures in normal processes of sensorimotor gating such as habituation (26,69). Such a model would suggest that 5-HT_2A antagonism by itself might be effective in the treatment of certain forms of schizophrenia. Indeed, a rather selective 5-HT_2A antagonist, M100907, is currently being tested as a putative antipsychotic in phase 3 clinical trials (67). This effort represents a rare test of a non-dopaminergic mechanism for a therapeutic treatment in schizophrenia research and an important test of the predictive validity of the hallucinogen model of psychosis.

Sensorimotor Gating Models

Theories describing the group of schizophrenias often conceptualize the common denominator linking these disorders as involving deficits in one or more of the several mechanisms that enable normal individuals to filter or gate most of the sensory and cognitive stimuli they receive. Collectively, this class of mechanisms is referred to as sensorimotor gating (6,26). Theoretically, impairments in gating lead to sensory overload and cognitive fragmentation. It is also possible that the mechanisms that subserve experimental examples of sensorimotor gating are also responsible for the gating of cognitive information. The hypothetical construct of sensorimotor gating has been operationalized and explored in both human and animal studies. For example, numerous studies have examined habituation deficits in patients with schizophrenia (e.g. 26), which may reflect failures of sensory filtering that could lead to disorders of cognition. As mentioned above, such quantifiable abnormalities in the behavior of patients with schizophrenia are mimicked when the homologous behavior is examined in rats treated with hallucinogenic serotonin agonists or glutamate antagonists such as phencyclidine (PCP) (25,26). The convergent validity of this gating construct has been further assessed using other operational measures, such as prepulse inhibition (PPI) of startle and the gating of P50 event-related-potentials (ERPs). As with the habituation model, both of these paradigms take advantage of the fact that very similar experiments can be conducted in humans and animals when homologous behaviors have been identified. The PPI paradigm is based on the fact that weak prestimuli presented 30-500 msec prior to a startling stimulus reduce, or gate, the amplitude of the startle response. The generality and reliability of this robust phenomenon is clear: PPI is observed in many species; PPI is evident both within and between multiple sensory modalities using a variety of stimulus parameters; PPI does not require learning or comprehension of instructions; and several laboratories have reported PPI deficits in schizophrenic, schizotypal, and presumably psychosis-prone subjects (7,11,77).

In the P50 sensory gating paradigm, two acoustic clicks are presented in rapid succession, usually 500 msec apart. In normal individuals, the P50 ERP to the second click is reduced or gated relative to the ERP to the first click. Schizophrenic patients and their first degree relatives exhibit less of this sensory gating (23). These demonstrations that schizophrenic patients exhibit conceptually linked deficits in habituation, PPI, and P50 gating provide converging support (i.e. convergent validity) for the hypothetical construct that schizophrenia involves disturbances in the filtering of sensory and cognitive information (i.e. construct validity). A recent study of individual differences explicitly tested the convergent validity of the hypothetical construct of gating. In a group of normal subjects, P50 gating was strongly correlated with the amount of startle habituation and only weakly with PPI (68), despite the fact that P50 gating appears to be more similar phenomenologically to PPI than to habituation. Further supporting the convergence of these different operational measures of the gating construct, the amount of habituation was positively correlated with the amount of PPI. Thus, the suggestion that gating deficits characterize schizophrenia has considerable construct validity by virtue of a number of converging lines of evidence. Of critical importance is the relationship, if any, between experimental measures of gating phenomena and the clinical complaints of sensory overload or signs of thought disorder. Surprisingly, within a cohort of schizophrenic patients (38), those with deficient P50 gating reported fewer complaints of sensory overload than did the patients with normal P50 gating, i.e. the opposite of the predicted relationship. With regard to the PPI gating measure, however, significant correlations have been observed among patients with schizophrenia between deficient PPI and both distractibility (41) and explicit measures of thought disorder based on an abstract problem-solving task (63).

Animal models addressing the gating deficits seen in patients with schizophrenia involve studies of very similar behavioral phenomena in rodents. Studies of the N40 ERP in rats have been used successfully as an animal analog of the P50 sensory gating paradigm in humans (23). Pharmacological treatments, such as amphetamine and PCP, which are thought to produce psychosis-like symptoms in humans (19,37), produce deficits in the sensory gating of the N40 ERP that appear similar to the gating deficits in schizophrenic patients (23,24). Similarly, studies of the PPI of startle responses in rats have provided extensive support for the predictive, construct, and convergent validity of sensorimotor gating deficits as an animal model of the demonstrated gating deficits in schizophrenic patients (27,75,78). Specifically, the apparent homology between the human and animal behaviors assessed in the PPI paradigm is supported not only by evidence of parametric similarities of startle responses across species, but also by the many studies confirming the cross-species comparability in the specific features of PPI. Thus, this literature constitutes much of the fundamental predictive validity supporting this animal model as being relevant to the human condition. In rats, disruptions in PPI similar to those observed in schizophrenic patients are produced by either direct or indirect DA agonists and by PCP and other non-competitive NMDA antagonists (27,78). As in studies of schizophrenic patients, the effects are observed with both intra-modal and cross-modal presentations of the prepulse and startle stimuli (7,27). In keeping with the corresponding model psychoses in humans, the DA agonist effects are blocked by neuroleptics and the NMDA antagonist effects are not (27). Indeed, impressive correlations have been found between the clinical efficacy of antipsychotic drugs in patients and their ability to reduce the disruptive effects of the DA agonist apomorphine on PPI (75). Even the ability of the atypical antipsychotic clozapine to block the effects of a DA agonist in this paradigm is predictive of its clinical efficacy (75). Furthermore, with appropriate test parameters, clozapine can improve PPI in otherwise untreated rats (78). This result parallels preliminary indications that clozapine treatment tends to normalize PPI deficits in schizophrenic patients. Thus, the model has considerable predictive validity, in addition to construct and convergent 18,75). In this case, the predictive validity of the model is derived from both parametric similarities and pharmacological isomorphism.

There are two respects in which the PPI model could be considered to have failed to exhibit discriminant validity. First, one could require that the dependent measure used in the model discriminate abnormalities in schizophrenia from other psychiatric disorders. Such a requirement would not be met by the model because deficits in PPI have also been found in disorders such as Obsessive Compulsive Disorder (74). Nevertheless, this model is a theoretically derived model that purports to investigate an abnormality in a neuropsychological process which is not necessarily specific to schizophrenia. The lack of discriminant validity in terms of currently accepted psychiatric diagnoses does not invalidate the usefulness of the model for the study of sensorimotor gating. Similarly, with respect to the independent variable used to elicit a PPI deficit, one could ask that the model discriminate the effects of DA agonists and NMDA antagonists from other drugs that do not appear to produce psychotic symptoms in humans. Evidence that 5-HT_1A and other serotonin agonists disrupt PPI in rats (64,71) speaks against such discriminant validity for the model. Furthermore, extensive studies of the neurobiological substrates that modulate PPI in rats have revealed that multiple neurotransmitter systems influence this form of sensorimotor gating, including dopaminergic, serotonergic, glutamatergic, GABAergic, and cholinergic pathways (27,64,76). Similarly, the animal version of the ERP sensory gating model is reportedly sensitive to dopaminergic, noradrenergic, and cholinergic manipulations (23,24). Hence, if one evaluates the gating models strictly with respect to the prevailing DA hypothesis of schizophrenia, the models appear to be over-inclusive. Alternatively, it has been suggested that the multiple determinants of PPI in rats may ultimately be related to the multiple subtypes of schizophrenia (26) and that the observation of gating deficits in non-schizophrenic patients may reflect abnormalities in different parts of the limbic-striatal circuitry that work in concert to enable the appropriate filtering of sensory and cognitive events (76). The viability of such possibilities illustrates why discriminant validity is seldom required of an animal model of psychopathology intended to explore a hypothetical construct.

Non-pharmacological Models

The independent variables used to induce gating disturbances in the animal models discussed above are all pharmacological treatments. Accordingly, they derive much of their support from high degrees of pharmacological isomorphism derived from the fact that agonist-induced behaviors are generally sensitive to the corresponding antagonists. Thus, it is not particularly informative to demonstrate that the effects of amphetamine are blocked by typical antipsychotic drugs which are DA antagonists. Some newer models relevant to schizophrenia have sought to avoid this reliance on behaviors induced by drugs. For example, deficits in PPI of startle have been demonstrated to result from socially isolating rats from weaning until after puberty (28). Social isolation of rats in early stages of development has been used to produce a variety of behavioral abnormalities that have been related to both schizophrenia and depression (44). Recent studies have shown that 6-8 weeks of social isolation during development, but not during adulthood (86), produces deficits in PPI that are at least partially reversible by the administration of neuroleptic DA antagonists (28) or by clinically effective atypical antipsychotics having antagonist activity at multiple receptors (2,80). Furthermore, post-weaning isolation rearing of rats also results in deficits in the gating of the N40 ERP, that are analogous to the deficits in P50 gating observed in schizophrenia (72). Because schizophrenia commonly emerges in early adulthood, developmental factors have provided the basis for some etiological hypotheses (58,83). Hence, further study of the gating deficits produced by isolation rearing of rats may establish a non-pharmacological and developmentally relevant animal model of the gating deficits observed in patients with schizophrenia. Potentially, in contrast to the drug-induced models of gating deficits, such a model might have etiological validity and might be sensitive to antipsychotic drugs having novel mechanisms of action.

Other examples of non-pharmacologically based models relevant to schizophrenia are emerging from the field of molecular biology in which genetic engineering is being used to generate transgenic and knockout animals. Genetic contributions to schizophrenia have been the focus of considerable research, although the application of linkage analyses to schizophrenia have not proven successful, perhaps because schizophrenia does not represent a single phenotype. Nevertheless, it remains possible that genetic approaches could lead to etiologically based models. In the absence of established candidate genes, the use of mutant animals in models of schizophrenia has focussed on the identification of phenotypic differences in behaviors considered relevant to the clinical disorder, i.e. validity has been sought primarily in the dependent measure rather than the independent variable. For example, schizophrenia-like deficits in PPI of startle have been observed in specific strains of mice (62) and in "knockout" mice in which specific genes have been deleted (45). Similarly, some inbred strains of mice are deficient in the gating of the N40 ERP (72), which is the rodent analog of the P50 gating deficit seen in schizophrenia. Indeed, a series of elegant studies have demonstrated a linkage between the P50 gating deficit in patients with schizophrenia and a specific chromosomal marker associated with the gene for the alpha-7 subunit of the nicotinic acetylcholine receptor (24). The potential power of cross-species studies of specific behavioral abnormalities seen in psychiatric disorders is exemplified by the parallel between these human linkage studies and the observation that the strain of mice that is most deficient in gating of the N40 ERP is also the most deficient in alpha-7 nicotinic receptors (72). Such parallel investigations in both patients and animals provide an exemplar for the modern application of molecular biological techniques to the generation and validation of animal models of psychiatric disorders.

CONCLUSIONS

While there may be no perfect animal models, it is clear that each model has strengths and limitations that need to be recognized in order to use the model effectively in the investigation of psychiatric disorders. Therefore, multiple animal models are needed for each psychiatric disorder to allow investigation of the various aspects of the disease and to provide convergent validation of the research findings. Further, each model is useful for a specific purpose, for example, the identification of potential pharmacotherapies or the investigation of a specific hypothetical construct. The validation criteria that each model has to meet to demonstrate its usefulness in neurobiological research are determined by the defined purpose of the model. Ultimately, however, the degree to which the performance of the model allows accurate predictions to be made about the human condition, or predictive validity, is the only necessary and scientifically meaningful criterion for the validation of any animal model. As a result, it is critical that objectively defined abnormalities characteristic of the disorder in question be identified to provide the focus for the development of animal models. Thus, the process of developing and validating animal models must work in concert with the process of identifying reliable measures of the human phenomenology.

ACKNOWLEDGMENTS

This work was supported by National Institute of Health (NIH) Research Scientist Awards to MAG (K05MH01223) and AM (DA00213), NIH Research Grants DA02925 and MH52885 to MAG, and a Novartis Research Grant to AM.

published 2000