Gottesman and Gould5
offered a set of five defining criteria for endophenotypes. In order to be called an endophenotype, a biological marker should (a) be associated with the illness in the relevant population; (b) be state independent, ie, present both during periods of illness and wellness; (c) be heritable; (d) cosegregate with the illness within families; and (e) be present in unaffected family members more frequently than in the general population. The issue of state independence of a marker has been modified to include the epigenetic influence on a given marker. In the absence of a disease or before the onset of a disease, these markers should be able to be detected following a challenge, such as a cognitive challenge in Alzheimer's disease or schizophrenia or glucose challenge in diabetes mellitus.6
In addition to the above criteria, other investigators have suggested different but overlapping “criteria” sets.7–12
They are as follows: (1) Endophenotypes need to be related to the cause rather than the consequence of the illness (in other words, the marker should at least be involved in a biologically plausible mechanism of pathogenesis13,14
), (2) they should have reliable psychometric properties or be reliably measured, (3) the markers should be stable over time, and (4) they should be related to the symptoms/features of the illness. In addition, some researchers propose that an endophenotype should be continuously quantifiable, should probabilistically predict the disease, and should be closer to the primary etiology than to the diagnostic category.12
Thus, the definition of endophenotype can vary in its stringency. At this stage, it is not clear how many of these criteria a marker should meet in order to consider it an endophenotype.
Certain caveats in the utility of endophenotypes merit discussion here. It is often questioned whether the endophenotypes are indeed internal (endo-) as opposed to being exophenotypes, eg, neurocognitive variations. For the latter, the term “intermediate phenotypes” is suggested, implying that these traits may be closer to the syndromal expression. The term “endophenotype” may be reserved for quantifiable, heritable traits that are not manifestly expressed, eg, the brain structural variations and brain oxygenation level–dependent (BOLD) responses measured using functional magnetic resonance imaging. These traits also may be localized within the pathophysiologic pathway whose impact may be externally manifested as an intermediate phenotype.
Several potential endophenotypic/intermediate markers have been proposed in schizophrenia, including neurocognitive impairments15–20
and electrophysiological abnormalities.21,22
Because the endophenotypes may be relatively closer to the genetic etiology of a disease, it is assumed that these traits may be associated with fewer interactions, determined by more discrete networks, and the effect size of genetic variations on the endophenotypes are presumed to be larger than on the clinical phenotype. This may not be applicable to all traits in the pathophysiologic pathway. For example, neurocognitive traits may be farther away from the genetic variations but closer to the clinical manifestation. Environmental influences and nongenetic factors, such as subject motivation, test administration, may influence the measurement process affecting the heritability estimates. Nevertheless, a recent genome-wide linkage study on a large sample of nuclear families with multiple affected members showed stronger association with neurocognitive endophenotypes than for the clinical diagnosis.23
For example, a locus on 4q13-25 within a 30cM region accounted for 33%, 33%, and 32% of the variation in delayed memory, semantic clustering, and verbal learning, respectively, using variance component analysis. When these investigators included the schizophrenia spectrum disorder as dichotomous variable, it greatly reduced the linkage signal. On the contrary, a large meta-analysis of published studies found that the effect size of genetic variations was not necessarily larger for the endophenotypes compared with the phenotype itself.24
These authors examined neurocognitive performance on Wisconsin card-sorting test (WCST) that measures executive function, N-back tasks for working memory, and the P300 event-related potential variables with one exonic single-nucleotide polymorphism (SNP) on catechol-O-methyl transferase gene (COMT
; rs4680; Val/Met). All three examined measures showed small effect sizes; for perseverative errors on WCST and the N-back task performance, an estimated effect size was 0.5% of the phenotypic variance (VP
), whereas it was 0.01% for the event-related potential P300. These small effects may be because, as authors point out, WCST and N-back structure may be complex. Further, performance on such tasks may be influenced by measurement and performance biases, testing conditions, and other variables not directly related to the pathology in question and may not be related to the genetic effects. In contrast to these measures, BOLD responses related to serotonin transporter gene polymorphism (5-HTT
) and P50, an event-related potential related to 7-nicotinic acid receptor, were found to have larger effect sizes. These studies demonstrate that selection of endophenotypes needs closer examination of the trait using sufficiently powered samples, replicate studies, and meta-analyses.
While it remains unclear whether these putative endophenotypes help in linkage or association studies, the utility of these markers as quantitative traits in the QTL approach may provide better power to detect bigger effect sizes. In this approach, traits of interest are continuous variables compared with the dichotomous trait of diagnosis. Quantitative trait locus is a region of the chromosome that would be correlated with the quantitative trait. Such QTLs may be clustered together or scattered throughout the genome. Mouse genetic studies show that an overwhelming majority of loci contributed small proportion of variance to several quantitative traits, suggesting that even endophenotypes may pose a challenge in elucidating their genetic architecture.25
Another equally daunting challenge could be poor spatial localization of these markers in the brain. Therefore, more stable measures are recommended to be of greater utility as endophenotypes, such as brain structural measures. It must be noted that schizophrenia is a clinically and etiologically heterogeneous disorder. It is possible that some endophenotypes may be associated with certain distinct but not yet teased out biological subsyndromes, giving rise to a possibility of lower heritability or even lower effect sizes. For this reason, such observations need not deter one from examining the usefulness of endophenotypes.
One potential way forward is to improve the utility of endophenotypes by combining functionally related endophenotypes, eg, merging the brain structural changes with cognitive variations regulated by the same regions. We propose the term “extended endophenotypes” to a network of endophenotypes that are linked on the basis of putative or documented functional basis. Constructing such “extended endophenotypes” may improve our chances of delineating the pathway from the genetic variations to the behavioral phenotype and could help in deconstructing the schizophrenia phenotype into biologically more meaningful clinical phenotypes that may be amenable to developing rational pharmacotherapy. In addition, such a construct could help identify potential “branch points” for other related psychotic disorders.
Recent years have witnessed an expansion of neuroimaging research in psychiatric disorders, notably schizophrenia.26
Measurements of brain structure are highly reliable and, in some cases, correlate robustly with phenomenologic features of psychiatric disorders; this makes brain structural alterations potentially attractive endophenotypes. However, few studies have critically examined whether various brain structure abnormalities in schizophrenia meet the criteria for endophenotypes.
In this paper, we examine whether brain structural alterations as measured by magnetic resonance imaging (MRI) could be considered as endophenotypes. We searched the literature databases for meta-analyses and systematic reviews using terms “endophenotype,” “intermediate phenotype,” “neuroanatomy,” “brain structure,” “MRI,” “schizophrenia,” “genetics,” “psychosis,” “white matter,” “at-risk,” and “high-risk.” Additionally, we searched the literature for individual studies and included relevant studies. Wherever relevant, we also searched the literature including other disorders such as bipolar disorders. We briefly review the results from consistently replicated studies, systematic reviews, and meta-analyses of the relevant literature. We consider structural imaging studies in both patients with schizophrenia and the affective psychoses and studies on at-risk populations. Further, we discuss the value of bringing together more than one endophenotype to construct pathophysiologically meaningful “extended endophenotypes.” Such an approach enables integration of genetic and neuroimaging paradigms in our efforts to clarify whether MRI morphometric measures meet the criteria for endophenotypes of psychotic and related disorders.