Psychiatric endophenotypes and the development of valid animal models


  • T. D. Gould,

    Corresponding author
    1. Laboratory of Molecular Pathophysiology, Mood and Anxiety Disorders Program, National Institute of Mental Health, NIH, HHS, Bethesda, MD, USA, and
      T. D. Gould, Laboratory of Molecular Pathophysiology, BLDG 35, Room 1C-912, NIMH, NIH, Bethesda, MD 20892-3711, USA. E-mail:
    Search for more papers by this author
  • I. I. Gottesman

    1. Departments of Psychiatry, University of Minnesota Medical School, and Psychology, University of Minnesota, Minneapolis, MN, USA
    Search for more papers by this author

T. D. Gould, Laboratory of Molecular Pathophysiology, BLDG 35, Room 1C-912, NIMH, NIH, Bethesda, MD 20892-3711, USA. E-mail:


Endophenotypes are quantifiable components in the genes-to-behaviors pathways, distinct from psychiatric symptoms, which make genetic and biological studies of etiologies for disease categories more manageable. The endophenotype concept has emerged as a strategic tool in neuropsychiatric research. This emergence is due to many factors, including the modest reproducibility of results from studies directed toward etiologies and appreciation for the complex relationships between genes and behavior. Disease heterogeneity is often guaranteed, rather than simplified, through the current diagnostic system; inherent benefits of endophenotypes include more specific disease concepts and process definitions. Endophenotypes can be neurophysiological, biochemical, endocrine, neuroanatomical, cognitive or neuropsychological. Heritability and stability (state independence) represent key components of any useful endophenotype. Importantly, they characterize an approach that reduces the complexity of symptoms and multifaceted behaviors, resulting in units of analysis that are more amenable to being modeled in animals. We discuss the benefits of more direct interpretation of clinical endophenotypes by basic behavioral scientists. With the advent of important findings regarding the genes that predispose to psychiatric illness, we are at an important crossroads where, without anthropomorphizing, animal models may provide homologous components of psychiatric illness, rather than simply equating to similar (loosely analogized) behaviors, validators of the efficacy of current medications or models of symptoms. We conclude that there exists a need for increased collaboration between clinicians and basic scientists, the result of which should be to improve diagnosis, classification and treatment on one end and to increase the construct relevance of model organisms on the other.

Psychiatry endures a diagnostic and classification system that is not based upon etiology, neurobiology, epidemiology, genetics or response to medications but rather on gross behaviors that have imprecise similarity and/or correlation with each other within and between individuals [see (Charney et al. 2002; Krishnan 2005; Sadler 2004) for extensive discussions]. A harsh critique of the current system could elicit comparisons between fictitious and real entities, the former being difficult to define experimentally and having limited construct validity (Cronbach & Meehl 1955; Maher & Gottesman in press). Diagnostic and Statistical Manual (DSM) approaches provide a partially validated mechanism whereby physicians can provide reliable diagnoses, communicate amongst themselves and report their findings to insurance providers. However, disease heterogeneity implicit in the current classification schema, and the imprecise quantification of the behaviors being described, makes it difficult to even partially deconstruct such ‘diseases’ within model organisms. This heterogeneity arises not only because of lack of foresight in the diagnostic system but also, in part, from true deficits in basic knowledge. With limited input from scientific progress, it is not surprising that there has been inconsistency in clinical studies (albeit much less than before DSM-III, which preceded the current version IV), at the level of both neurobiology and genetics in the study of psychiatric illness.

There is an imperfect relationship between genes and behaviors so that different combinations of genes (and resultant changes in neurobiology) contribute to any complex behavior (normal or abnormal). Similarly, while the pathways, beginning with genes that are later expressed through biological processes, do not necessarily have a single quantifiable endpoint (i.e. behavior), it may be possible to assay the result of aberrant genes through more biologically ‘simple’ approaches. Simpler neurological processes reached through optimizing reductionism (at a level lower than behavior) are – in most cases – controlled by similar biological processes as behavior (just not as many) and hence lend themselves to study.

In psychiatry, reducing complex behaviors into components, whether they are neurophysiological, biochemical, endocrine, neuroanatomical, cognitive or neuropsychological, is described as an endophenotype strategy or approach (Gottesman & Shields 1972; Gottesman & Shields 1973). Symptoms and clinical subtyping (i.e. depression with or without psychosis) generally are not considered endophenotypes. Subtyping in this manner amounts to little more than altering the defining observations of a complex behavior. While it may result in some constraints on heterogeneity, the difference is marginal: decades of applying this approach have resulted in only slightly greater reproducibility than with the broad definition disorders themselves. In other fields of medicine, ‘endophenotyping’ is considered routine; for example, such disease predictors as blood glucose assays are used for diabetes and stress-electrocardiogram (EKG) interpretations for heart disease.

Without question, the term endophenotype represents a current ‘buzz word’ in neuropsychiatric research, and it is envisioned that endophenotypes will assist in clarifying many relevant issues (Gottesman & Gould 2003). Among clinician-scientists in psychiatry, the approach has received immense interest in such diverse disorders such as schizophrenia (Braff & Freedman 2002; Gottesman & Erlenmeyer-Kimling 2001; Hariri & Weinberger 2003; Heinrichs 2005; Lenzenweger 1999; Weinberger et al. 2001), bipolar disorder (Ahearn et al. 2002; Glahn et al. 2004; Hasler et al. in press; Lenox et al. 2002), depression (Berman et al. 1999; Hasler et al. 2004; Niculescu & Akiskal 2001), Alzheimer's disease (Kurz et al. 2002; Neugroschl & Davis 2002), attention-deficit hyperactivity disorder (Castellanos & Tannock 2002; Doyle et al. 2005; Gould et al. 2001; Waldman 2005), obsessive compulsive disorder (Chamberlain et al. 2005; Miguel et al. 2005), autism (Belmonte et al. 2004), alcoholism (Dick et al. in press; Porjesz et al. 2005) and personality disorders (Siever 2005). It has become increasingly obvious that among studies of psychiatric diseases, there exists an overwhelming number of biological markers. However, these often solitary findings frequently have limited reproducibility, both among and within patients, and may represent state-dependent results. This has led to current criteria for an endophenotype, distinguished from markers or biomarkers generally, in the hope of reaching genetically and biologically meaningful conclusions (Gottesman & Gould 2003) (Fig. 1).

Figure 1.

Biological markers (a.k.a. subclinical traits and vulnerability markers) may be primarily environmental, epigenetic or multifactorial in origin. For this reason, criteria useful for the identification of markers to study psychiatric genetics have been proposed, adapted and refined over time (see Gershon & Goldin 1986; Gottesman & Gould 2003; Hasler et al. in press; Leboyer et al. 1998; Lenox et al. 2002; Shields & Gottesman 1973). Current criteria for an endophenotype, to be distinguished from biological markers, are designed to direct clinical research in psychiatry toward genetically and biologically meaningful conclusions. © 2005 I.I. Gottesman, and used with permission.

  • 1An endophenotype is associated with illness, in the population.
  • 2An endophenotype is heritable.
  • 3An endophenotype is state independent (manifests in an individual whether or not illness is active) but age-normed and may need to be elicited by a challenge, e.g. glucose tolerance test in relatives of diabetics.
  • 4Within families, endophenotype and illness cosegregate.
  • 5An endophenotype identified in probands is found in their unaffected relatives at a higher rate than in the general population.

Additional considerations have been suggested, including the importance of good psychometric properties such as test-retest reliability and normal distribution (Waldman 2005). Following confirmation of these criteria, endophenotypes should share specific regions and gene association with the disorder (Waldman 2005). Importantly, not all patients are expected to show endophenotypes, and not all persons with these impairments are expected to demonstrate symptoms of the psychiatric disease (probabilism vs. determinism).

What is the primary goal of utilizing endophenotypes? The central hypothesis begins with what is now wholly accepted – that genetics of psychiatrics disorders is complex and further complicated by epigenetic and stochastic contributors and various gene-by-gene and gene-by-environment interactions/coactions. Many genes interact at many levels, leading to activation of multiple neuronal circuits, which results in behavioral variations (Fig. 2). This is complicated by the knowledge that there can be more than one pathway to a given behavior. Endophenotypes represent more defined and quantifiable measures that are envisioned to involve fewer genes, fewer interacting levels and ultimately activation of a single set of neuronal circuits (Fig. 2). The fewer the pathways that give rise to an endophenotype, the better the chances of efficiently discovering its genetic and neurobiological underpinnings.

Figure 2.

Endophenotypes are characterized by simpler neurobiological and genetic antecedents than exophenotypic psychiatric disorders, thereby employing optimal reductionism. The bipolar disorder phenotype, as an example, is associated with a number of candidate genes and chromosomal regions, the influence of which can be observed at either the levels of behavior or endophenotypes. Endophenotypes, located closer to genes in the pathway from genes to behaviors, have fewer genes associated and thus are more amenable to genetic investigations and studies in model systems. This skeleton (genes to endophenotypes to behaviors), which allows for epigenetic, ‘environmental’ and purely stochastic influences upon clinical observations, can be applied to other diseases of complex genetics with the input of disease-specific candidate genes/regions and endophenotypes (Gottesman 1997; Gottesman & Gould 2003; Hasler et al. in press; Manji et al. 2003; Sing et al. 1994; Sing et al. 1996). © 2005 I. I. Gottesman and used by permission.

It is hoped that clinical research on the level of genetics, neurobiology or gross behavioral analysis will benefit from an endophenotype approach. This certainly has been the case thus far (Gottesman & Gould 2003). Ultimately, endophenotypes are envisioned to aid in diagnosis, classification, treatment, clinical research and the development of preclinical models. Indeed, this reductionist approach implicit in endophenotypes has clear parallels with the general mechanisms employed in preclinical research, where the concept of studying more than one variable simultaneously would rarely be warranted. In fact, animal modeling in psychiatry has relied almost exclusively upon simpler phenotypes. Future major advancements in biological modeling of psychiatric disease will likely be derived extensively from clinical endophenotypes, where acknowledgment, enthusiasm and development of valid endophenotypes are increasing. Novel clinically validated endophenotypes, with distinct gene-endophenotype relationships, have major implications for influencing preclinically oriented research in psychiatry, with ramifications for both in vitro and in vivo investigations.

It is well established that the development of better validated, and more appropriate, animal models is a task representing major importance for psychiatry. For most psychiatric disorders, this deficiency of suitable models for in-depth biochemical, histological and behavioral analysis has greatly hindered progress in understanding neurobiology and in developing novel medications (Einat in press; Flint et al. 2005; Gould & Manji 2004; Nestler et al. 2002; Seong et al. 2002; Spedding et al. 2005; Tecott 2003). While it is often considered (and certainly a view we support) an impossible task to fully model the diverse constellation of behavioral observations observed in psychiatric diseases in laboratory animals, endophenotypes are proving more amenable to the task (Petryshen et al. 2005; Seong et al. 2002).

Animal models of human symptoms do not necessarily have simpler genetics or biological mechanisms than the human disorders. This is not to say that genetic and biological studies of human symptoms, such as models of anxiety, do not advance our understanding of the neurobiology of behavior tremendously and may in fact provide important leads (Crusio 2001; Henderson et al. 2004; Yalcin et al. 2004). However, multiple genes and multiple pathways contribute – even in the mouse (Phillips et al. 2002; Seong et al. 2002). Thus, we may not be studying specific genes or biological pathways that are relevant to human anxiety but instead those most relevant to the mouse behavior in question. Symptom-based models rely upon observable signs and symptoms of the disorders often represented by DSM-IV criteria and can be useful in deciphering the general genetics and neurobiology of behavior. However, we believe that the future development of animal models for psychiatric disorders (not necessarily for the actions of medications) will require a greater focus on validated endophenotypes rather than on symptom-based models. This is especially true when attempting to find genes and validate neurobiological mechanisms in model organisms, which are derived from studies of the human condition (Harrison & Weinberger 2005; Kirov et al. 2005). Therefore, a major advantage of an endophenotype approach to understanding psychiatric illnesses, given their complexity, is that animal models using endophenotypes are generally more straightforward and congruent with the human condition both on the level of biology and genetics, thus facilitating the use of such approaches.

Although progress in defining endophenotypes for some disorders (such as mood disorders) has been slow in both clinical and preclinical research, the endophenotype concept has had success in schizophrenia, where endophenotypes including prepulse inhibition (PPI, a measure of sensory motor gating deficits), eye-tacking dysfunction and working memory deficits are already connected to genetic polymorphisms and have enough support to be considered true endophenotypes of the disorder. There are extensive efforts underway to study, in model organisms, the neurobiological antecedents of various schizophrenia endophenotypes including PPI deficits (Geyer et al. 2002; Joober et al. 2002; Petryshen et al. 2005; Robbins 2005; Seong et al. 2002; Swerdlow et al. 2001). The cause of putative neuroanatomical and neurodevelopmental endophenotypes in schizophrenia (Harrison & Weinberger 2005; Weinberger 1995) may soon be explained as the functional consequences of susceptibility genes are elucidated (e.g. Brandon et al. 2004; Schurov et al. 2004). Furthermore, multiple cognitive domains deficient in schizophrenia, in addition to working memory, represent possible endophenotypes (Green et al. 2004). Many of these are quite amenable to rodent studies. Neuropsychological findings have been correlated with in vivo imaging measures (Weinberger et al. 2001). However, similar imaging technology in rodents is far from ideal and will require significant technical advances to become a feasible approach.

Current animal models for bipolar disorder or depression mostly attempt to observe naturalistic behaviors or the results of pharmacological manipulation (Crawley 2000; Crowley & Lucki 2005; Cryan & Holmes 2005; Einat et al. 2003; Einat in press). Evolving theories suggest that endophenotypes for these disorders, based upon genetic and biological contributions, include attention deficits, circadian rhythm instability, dysmodulation of motivation and reward, brain structural changes and increased sensitivity to stress and stimulant medications (Glahn et al. 2004; Hasler et al. 2004; Hasler et al. in press; Lenox et al. 2002) (Fig. 2). Some of these are more amenable to modeling in animals than others. For example, a polymorphism in the putative bipolar disorder susceptibility gene, brain-derived neurotrophic factor (BDNF), has been linked to anatomic variations in the hippocampus and prefrontal cortex (Pezawas et al. 2004), as well as to hippocampal and prefrontal cortex cognitive performance (Egan et al. 2003; Hariri et al. 2003; Rybakowski et al. 2005) in humans. There exists significant preclinical support for the involvement of BDNF in both learning and memory and neuronal growth/plasticity (Egan et al. 2003; Lu 2003). Ultimately, the endophenotypes most valuable for affective disorders may be specific structural, functional and neuropsychological deficits, which at face value have poor correlation with overt phenotype-based models (Glahn et al. 2004; Hasler et al. 2004; Hasler et al. in press; Lenox et al. 2002; McDonald et al. 2004). Limbic-hypothalamo-pituitary-adrenocortical axis malfunction represents another viable putative endophenotype for mood disorders, and extensive animal models are already in existence (Akil 2005; Gould et al. 2003; Hasler et al. 2004; Seong et al. 2002).

As ongoing genetic studies advance, a full complement of validated animal models will be necessary to pinpoint the relevance of each individual single nucleotide polymorphism. Given appropriate caveats, transgenic mouse technology has become commonplace, allowing for relatively rapid modification of gene expression (Crusio 2004; Gerlai 2001; Phillips et al. 2002). Clinical linkage studies continue to provide leads, which will inform the study of syntenic regions on the mouse and human chromosomes. These are mostly established, and recent research has already taken advantage of the available knowledge (Bogani et al. 2005; Wang et al. 2005).

Current attempts to model human disease in model organisms often report multiple deficits in many behaviors. However, the present field of psychiatric genetics was initiated with the understanding that human diseases are complex in nature – needing multiple genes working in disharmony with non-genetic contributors for the human syndrome (Gottesman & Shields 1967). The current standard of 6+ rodent phenotypes to make a high-impact paper is questionable given the nature of the genetics of these disorders. In these instances, a single gene disruption results in widespread ‘human-related’ psychiatric phenotypes, which is in contrast to the predictions of the field. An improved expectation, even in a best case scenario, may be one or two endophenotypes being modulated by a true susceptibility gene acting improperly, as in the parallel human condition. Thus, it is conceivable that an endophenotype approach to genetic studies, when applied to preclinical studies in rodents, will not run the gamut of fulfilling multiple, diverse animal models of bipolar disorder; instead it may singularly affect one biological process that will only be present in tests of that single arena of focus (endophenotype) in laboratory animals. These may have limited (or no) face validity vis-à-vis symptom-based models. Clearly, for the development of improved therapeutics, animal models of medication efficacy (e.g. the tail suspension test and approach-avoidance situations) have tremendous utility; however, as gene exploration studies continue, we may find that current models have limited utility in discerning the relevant contribution of susceptibility genes. Continued improvements in valid animals models will depend on well-defined endophenotypes, downstream of gene expression and upstream of clinical symptoms, that necessitate increased collaborations between clinical and bench scientists.


Research supported by the Intramural Research Program of the National Institute of Mental Heath (TDG), the Foundation for the National Institutes of Health (Neuroscience Research Fellowship) (TDG), the National Association for Research on Schizophrenia and Depression (Young Investigator Award to TDG) and the Drs Irving and Dorothy Bernstein Professorship in Adult Psychiatry, University of Minnesota Medical School (to IIG). We thank Robert Gerlai and Kelley O'Donnell for helpful comments.