Endophenotypes in Schizophrenia: A Selective Review

Abstract

Background

Given the wealth of data in the literature on schizophrenia endophenotypes, it is useful to have one source to reference their frequency data. We reviewed the literature on disease-liability associated variants in structural and functional magnetic resonance images (MRI), sensory processing measures, neuromotor abilities, neuropsychological measures, and physical characteristics in schizophrenia patients (SCZ), their first-degree relatives (REL), and healthy controls (HC). The purpose of this review was to provide a summary of the existing data on the most extensively published endophenotypes for schizophrenia.

Methods

We searched PubMed and MedLine for all studies on schizophrenia endophenotypes comparing SCZ to HC and/or REL to HC groups. Percent abnormal values, generally defined as > 2 SD from the mean (in the direction of abnormality) and/or associated effect sizes (Cohen’s d) were calculated foreach study.

Results

Combined, the articles reported an average 39.4% (SD=20.7%; range=2.2-100%) of abnormal values in SCZ, 28.1% (SD=16.6%; range=1.6-67.0%) abnormal values in REL, and 10.2% (SD=6.7%; range=0.0-34.6%) in HC groups.

Conclusions

These findings are reviewed in the context of emerging hypotheses on schizophrenia endophenotypes, as well as a discussion of clustering trends among the various intermediate phenotypes. In addition, programs for future research are discussed, as instantiated in a few recent large-scale studies on multiple endophenotypes across patients, relatives, and healthy controls.

Introduction

Schizophrenia is an inherited, likely complex genetic disorder that “runs in families” and the single best predictor for developing the illness is having an affected first-degree relative (Waddington et al., 2007). However, most affected individuals lack a family history, leaving open the question of how risk is acquired in such cases. Therefore, it is important, while studying prevalence rates for endophenotypes in patients and first-degree relatives, to also be aware of prevalence rates within the general population.

Because the pathophysiology of schizophrenia remains unknown, there are presently no laboratory tests or biological markers (biomarkers) related to the central etiopathology of the illness. Biomarkers are objectively measured characteristics that are “indicators of normal biologic processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention (Atkinson, 2001).” They are disease-specific indicators of the presence or severity of the biological process directly linked to the clinical manifestations and outcome of a particular disorder (Ritsner & Gottesman, 2009). For example, hemoglobin A1c (HbA1c or glycosylated hemoglobin) is a minor component of hemoglobin which binds glucose and whose levels are proportional to average recent blood glucose concentrations. HbA1c is thus a useful indicator of adequacy of blood glucose control in patients with type II diabetes, as well as being related to the pathophysiology of this disorder of carbohydrate metabolism and in detecting an important disease feature, i.e. pathologically elevated blood glucose.

In contrast, Endophenotypes, or “intermediate phenotypes,” are best considered as quantifiable biological variations or deficits that are types of stable trait markers or indicators of presumed inherited vulnerability or liability to a disease (Ritsner & Gottesman, 2009). Because the pathophysiology of schizophrenia remains obscure, and thus biomarkers are lacking, genetic research into the disorder has generally focused on the clinical phenomenology of this complex and likely multi-determined, multi-path, inherited disorder as the relevant phenotype. Endophenotypes are associated with the illness, state-independent, co-segregate within families and are found in some unaffected relatives of individuals with the disorder (because they represent vulnerability for the disorder, not the disorder itself), although at a higher prevalence than in the general population (Gottesman & Gould, 2005). They are not visible to the naked eye and are assessed by experimental, laboratory-based methods rather than by clinical observation. Because schizophrenia is likely to fall into the category of common, multi-genetic disorders (analogous to hypertension or type II diabetes; Pearlson and Folley, 2008a,b) endophenotype strategies are increasingly employed by researchers, based on the presumption that endophenotypes have more straightforward inheritance patterns and are coded for by smaller numbers of genes than are complex, heterogeneous phenomenological entities such as Diagnostic and Statistical Manual-IV-TR (DSM-IV-TR; American Psychiatric Association, 2000) diagnostic categories. Seen in this light, the endophenotype is “intermediate” between a clinical entity and the associated disease vulnerability genes. The hope is that by employing endophenotypes, the search for the etiopathology, including genetic determinants, of schizophrenia is made more straightforward (Chan and Gottesman, 2008; Pearlson and Folley, 2008a,b).

Although new mutations, deletions, or copy number variants may account for some cases (Walsh et al., 2008), other affected individuals are believed to acquire their liability for the disorder through inheritance of several common single nucleotide polymorphism (SNP) based variants, likely acting multiplicatively (Gangestad and Yeo, 2006). At a genetic level, collections of smaller numbers of SNPs may manifest as endophenotypic abnormalities (Campbell et al., 2006).

Population geneticists often assume the Hardy-Weinberg equilibrium (Hardy, 1908) when predicting genetic outcomes in subsequent populations, stating that genes and their phenotypes remain constant barring changes. However, changes including mutations (particularly caused by duplications), selection, migration, and other consequences of population and individual mating choices can cause disorders to be introduced and propagated by these forces, disrupting the Hardy-Weinberg equilibrium. Complex inherited disorders are examples of such disequilibrium including schizophrenia (Sullivan et al., 2003), bipolar disorder (Smoller and Finn, 2003), multiple sclerosis (Oksenberg and Barcellos, 2005), and type II diabetes (Permutt, et al., 2005), which affect multiple loci and have been related to population drift. Thus, through multiplicative and additive models, these relatively prevalent disorders can persist in the population. This complicates the known inheritance of these disorders, but it also makes it possible to observe these multiple loci pooling in certain individuals, which are likely to be more affected by the clinical phenotypes.

It is thus likely not uncommon for healthy individuals in the general population to possess one or a few schizophrenia-associated endophenotypes, although actual prevalence rates are poorly documented. Theoretically, these endophenotypes could be neutral or even beneficial singly, if not combined with other intermediate phenotypes (Keller and Miller, 2006, Pearlson and Folley, 2008a,b). Like the hypothesized “thrifty genes” associated with type II diabetes, they may confer selective advantage under particular circumstances (Neel, 1962). These multiple genetic loci (polygenes) of small relative effect are likely additive or epistatic (interactive) with regard to cumulative schizophrenia risk; only in combination are they deleterious and likely then often in conjunction with environmental events.

There is a wealth of data in the literature on disease-related endophenotypes in schizophrenia patients (SCZ) and their first-degree relatives (REL), yet very few reviews of prevalence rates within all three categories [SCZ, REL, & healthy controls (HC)], despite the theoretical importance of such information. Heinrichs (2001) provides a thorough review of endophenotypes, but concentrates mainly on SCZ and REL, with little data on HC. Recent articles, such as the “Just the Facts” series in this journal (Tandon et al., 2008 a, b; Keshavan et al. 2008), have brought to light the importance of evaluating endophenotypes for schizophrenia in order to assess research progress in this area thus far. As a prelude to further study of the co-occurrence of multiple schizophrenia biomarkers in a large, representative community sample, including all three categories, we surveyed the existing literature on the most widely published endophenotypes (Heinrichs, 2001) in order to continue our examination of the prevalence of endophenotypic abnormalities in the general population (Pearlson and Folley, 2008, a,b). Articles that compared SCZ to HC or REL to HC (or both) within six different groups of endophenotypes (structural and functional brain abnormalities, sensory processing measures, neuromotor abnormalities, neuropsychological measures, physiologic abnormalities and minor physical anomalies) were included in this review. A conservative definition of abnormality was utilized in this review based on a model of statistical infrequency. As such, depending on available data, percent of abnormal findings, generally defined as greater than two standard deviations (SD) from the mean (in the direction of abnormality) and/or effect sizes (Cohen’s d; Cohen, 1988) were extracted from each article. Under our summaries of each endophenotype, the total number of articles reviewed is reported; however, not all articles reviewed reported data on SCZ, REL, and HC samples. The number of articles reported within tables for each endophenotype reflect the number of articles that report unique data contributing to the calculations of each summary statistic, which may be different that the overall total within each endophenotype.

The purpose of this review was to assess the frequency of these established endophenotypes in all three categories, in order to provide a source of reference, as well as a beginning point for discussion on prevalence rates within SCZ, REL, and HC.

Structural and Functional Brain Abnormalities

Ventricular Volume

A Medline search was completed with the search terms of “schizophrenia” combined with “ventricular volume or lateral ventricles.” Studies that assessed volumetric measurements in cubic centimeters or milliliters of the right and left lateral ventricles in SCZ, REL, and HC were included. Effect sizes were calculated based on measurements of absolute volume rather than ventricle-brain ratios since they provided greater discrimination between groups. A total of 13 articles met our criteria and reported data on both left and right lateral ventricle volumes (Buchsbaum et al., 1997; Chua et al., 2007; Degreef et al., 1992; DeLisi et al., 2004; DeLisi et al., 2006; Dickey et al., 2000; Kelsoe et al., 1988; Lawrie et al., 1999b; Marsh et al., 1997; McDonald et al., 2006; Shenton et al., 1991; Suddath et al., 1989; Whitworth et al., 2005). Effect sizes are summarized in Table 1.

Table 1.

Mean effect size and 95% CI for structural and functional brain abnormalities.

Endophenotype: Ventricular Volume
	Effect Size	95% CI	N
Left Lateral Ventricle
SCZ	0.7	0.48-0.92	12
REL	0.4	-1.21-1.96	2
HC	N/A	0	0
Right Lateral Ventricle
SCZ	0.7	0.42-0.97	12
REL	0.3	-0.30-1.0	2
HC	N/A	0	0
Endophenotype: planum Temporale Volume
Left Planum Temporale
SCZ	0.5	0.19-0.80	18
REL	0.1	N/A	1
HC	N/A	0	0
Right Planum Tenmporale
SCZ	0.3	0.16-0.50	19
REL	0.2	N/A	1
HC	N/A	0	0
Endophenotype: Superior Temporal Gyrus
Left Superior Temporal Gyrus
SCZ	0.6	0.38-0.80	17
REL	0.3	N/A	1
HC	N/A	0	0
Right Superior Temporal Gyrus
SCZ	0.5	0.24-0.70	16
REL	0.2	N/A	1
HC	N/A	0	0
Endophenotype: fMRI Activation During 2-Back Task
SCZ	0.9	0.19-1.67	4
REL	N/A	N/A	0
HC	N/A	0	0

N = number of studies included; N/A = not applicable; unable to calculate based on available published data

Planum Temporale Volume or Surface Area

A Medline search with the search terms of “schizophrenia” combined with “planum temporale” was conducted. Studies that assessed volumetric measurements in cubic centimeters or millimeters, as well as surface area in centimeters squared, in the regions of the right and left planum temporale were included. A total of 20 articles met our criteria (19 left: Barta et al., 1997; Crespo-Facorro et al., 2004; DeLisi et al., 1994; Dickey et al., 2002; Falkai et al., 1995; Frangou et al., 1997; Hirayasu et al., 2000; Kulynych et al., 1995; Kulynych et al., 1996; Kwon et al., 1999; McCarley et al., 2002; Meisenzahl et al., 2002; Petty et al., 1995; Rossi et al., 1994; Rossi et al., 1992; Shapleske et al., 2001; Sumich et al., 2002; Takahashi, Suzuki, Zhou, Tanino, Hagino, Niu et al., 2006; Yamasue et al., 2004; and 20 right: Barta et al., 1997; Crespo-Facorro et al., 2004; DeLisi et al., 1994; Dickey et al., 2002; Falkai et al., 1995; Frangou et al., 1997; Goldstein et al., 2002; Hirayasu et al., 2000; Kulynych et al., 1995; Kulynych et al., 1996; Kwon et al., 1999; McCarley et al., 2002; Meisenzahl et al., 2002; Petty et al., 1995; Rossi et al., 1994; Rossi et al., 1992; Shapleske et al., 2001; Sumich et al., 2002; Takahashi, Suzuki, Zhou, Tanino, Hagino, Niu et al., 2006; Yamasue et al., 2004)) and effect sizes were calculated for each (see Table 1).

Superior Temporal Gyrus Volume

A Medline search with the terms of “schizophrenia” combined with “superior temporal gyrus” was completed. Studies that assessed volumetric measurements in cubic centimeters and milliliters of the right and left superior temporal gyrus in SCZ, REL, and HC were included. Effect size calculations were completed on a total of 17 articles that reported data on both left and right superior temporal gyrus (Anderson et al., 2002; Barta et al., 1990; Bryant et al., 1999; DeLisi and Hoff, 2005; DeLisi et al., 1994; Dickey et al., 1999; Frangou et al., 1997; Holinger et al., 1999; Kim et al., 2003; Kulynych et al., 1996; Marsh et al., 1997; McCarley et al., 1993; Meisenzahl et al., 2004; Onitsuka et al., 2004; Rajarethinam et al., 2000; Takahashi, Suzuki, Zhou, Tanino, Hagino, Kawasaki et al., 2006; Vita et al., 1995). Effect sizes are summarized in Table 1.

fMRI Activation during 2-back Task

Functional magnetic resonance imaging (fMRI) BOLD signal activation during performance of the 2-back task was examined. A total of 4 articles reported data on activation in the dorsolateral prefrontal cortex (Callicott et al., 2000; Jansma et al., 2004; Meisenzahl et al., 2006; Thermenos et al., 2005) and effect sizes were calculated based on differences in activation between groups. Effect sizes are summarized in Table 1.

Sensory Processing and Event-Related Potential Measures

Prepulse Inhibition

All studies that assessed sensory gating deficits measured by prepulse inhibition in SCZ, REL, and HC were included. A Medline search with the terms “schizophrenia” combined with “prepulse inhibition” was completed. Due to multiple variations in paradigm conditions, we specified two from which effect sizes were calculated: interstimulus interval and sound intensity. We chose an interstimulus interval of 60-120 milliseconds between the prepulse and pulse stimuli and sound intensities ranging from 40-90 decibels. The selection of these criteria increased group differences, resulting in higher effect sizes. In total, 15 records were selected that fit our selection criteria (Braff et al., 2001; Braff et al., 1999; K. Cadenhead et al., 1996; Cadenhead et al., 1993; Cadenhead et al., 2000; Dawson et al., 2000; Hong et al., 2007; Kumari et al., 2000; Kumari et al., 1999; Mackeprang et al., 2002; McDowd et al., 1993; Oranje et al., 2002; Parwani et al., 2000; Perry et al., 2002; Weike et al., 2000). Percent abnormal calculations and effect sizes are summarized in Table 2.

Table 2.

Mean percent abnormal, effect size, and 95% CI for sensory processing and event-related potential measures.

Endophenotype: Pre-Pulse Inhibition
	%Abnormal	N	Effect Size	95% CI	N
SCZ	38.0	2	0.8	0.07-0.09	14
REL	47.0	1	0.8	N/A	1
HC	21.8	2	N/A	N/A	0
Endophenotype: P50
SCZ	69.6	8	1.5	0.99-1.98	22
REL	59.4	2	1.7	-0.12-3.52	5
HC	15.5	6	N/A	N/A	0
Endophenotype: P300
SCZ	45.8	7	0.8	0.46-1.10	12
REL	18.5	3	0.7	-0.16-1.64	4
HC	6.5	6	N/A	N/A	0
Endophenotype: N400
SCZ	45.7	2	0.8	0.60-0.91	6
REL	N/A	0	N/A	N/A	0
HC	0.0	1	N/A	N/A	0

N = number of studies included; N/A = not applicable; unable to calculate based on available published data

P50

A Medline search with the terms of “schizophrenia” combined with “P50” was completed. All studies that assessed sensory gating deficits measured by the P50 ratio (amplitude of the testing stimulus/amplitude of the conditioning stimulus) in SCZ, REL, and HC were included. We chose the P50 ratio rather than the P50 suppression ratio due to greater discrimination between groups, resulting in greater effect sizes. In total, 21 records were selected that fit our criteria (Adler et al., 1985; Adler et al., 1990; Boutros et al., 1999; Clementz et al., 1997; Cullum et al., 1993; de Wilde et al., 2007; Freedman et al., 1987; Freedman R, 1996; Hong et al., 2007; Jin et al., 1997; Johannesen et al., 2005; Kathmann and Engel, 1990; Louchart-de la Chapelle et al., 2005; Myles-Worsley, 2002; Nagamoto et al., 1989; Olincy et al., 2000; Price et al., 2006; M. C. Waldo et al., 1988; Merilyne C. Waldo et al., 1991; Ward et al., 1996; Yee et al., 1998). Percent abnormal calculations and effect sizes are presented in Table 2.

P300

For studies that report prevalence rates of abnormal P300 response, a Medline search was performed using the terms “schizophrenia” and “P300” or “P3”. Two methods for determining individuals with abnormal P300s were used. In some studies, the prevalence rates were determined by individual subject data presented in scatter plots, where abnormality was defined as the portion of the distribution beyond 2 standard deviations of the HC mean. In addition, we calculated effect sizes from the reported peak latency. Though schizophrenia patients presented event-related potentials with diminished amplitudes as well as increased latencies, we chose to examine only peak latencies due to the temporal precision of electroencephalography. All of the studies reviewed used the auditory oddball task with 1000 Hz and 1500 Hz tones. In total, 12 studies were selected that fit our criteria (Blackwood et al., 1991; Bobes et al., 1996; Bramon et al., 2005; Faux et al., 1990; Ford et al., 1999; Frangou et al., 1997; Mathalon et al., 2000; McCarley et al., 1993; Price et al., 2006; Saitoh et al., 1984; Salisbury et al., 1996; Souza et al., 1995). Percent abnormal calculations and effect sizes are summarized and presented in Table 2.

N400

For the N400, a Medline search was performed with the terms “schizophrenia” and “N400”. The prevalence rates and effect sizes were determined similarly to P300. A potential confound for this endophenotype is the variability in tasks used to elicit the N400. Mostly, studies utilized an incongruent sentence completion task with visual presentation. However, some used an auditory presentation, or a different task, such as the semantic priming task or semantic matching task. In total, 8 studies fit our selection criteria (Adams et al., 1993; Bobes et al., 1996; Condray et al., 1999; Grillon et al., 1991; Koyama et al., 1991; Nestor et al., 1997; Niznikiewicz et al., 2002; Niznikiewicz et al., 1999). Percent abnormal calculations and effect sizes are presented in Table 2.

Neuromotor Abnormalities

Smooth Pursuit Eye Movement

For abnormal smooth pursuit eye movements, a Medline search was completed using the terms “schizophrenia” combined with “smooth pursuit eye movement”. Typically referred to as eye tracking dysfunction (ETD), prevalence rates were determined primarily from qualitative observations. However, some studies reported prevalence rates based on quantitative measures, including the natural log of signal-to-noise ratio (ln S/N), or the root mean square error (RMSE). In addition, we calculated effect sizes from the reported frequency of catch-up saccades (Table 3). This measure was chosen because increased numbers of catch-up saccades are a fundamental characteristic of the smooth pursuit eye movement impairments in schizophrenia. Most of the studies reviewed used a 0.4 Hz pendulum to examine smooth pursuit. However, some studies varied the pendulum frequency, and other studies used a sine wave to examine smooth pursuit. Most of the studied used electro-oculography (EOG) to record eye movement, others used infrared reflectometry (IR), or both. A total of 26 articles fulfilled our selection criteria (Acker and Toone, 1978; Allen et al., 1990; Altman et al., 1990; Amador et al., 1991; Blackwood et al., 1991; Boudet et al., 2005; Clementz et al., 1992; Holahan and O’Driscoll, 2005; Holzman et al., 1973; Iacono et al., 1992; Jones and Pivik, 1985; Keefe et al., 1989; Kinney et al., 1998; Levin et al., 1981; Levin et al., 1988; Levy et al., 1992; Levy et al., 2000; Louchart-de la Chapelle et al., 2005; Matthysse et al., 1986; Ross et al., 1998; Saletu et al., 1986; Scarone et al., 1987; Sereno and Holzman, 1995; Siever et al., 1990; Smeraldi et al., 1987; Thaker et al., 1996). Effect sizes and percent abnormal calculations are presented in Table 3.

Table 3.

Mean percent abnormal, effect size, and 95% CI for neuromotor abnormalities.

Endophenotype: Smooth Pursuit Eye Movement
	%Abnormal	N	Effect Size	CI 95%	N
SCZ	47.8	22	1.0	0.74-1.28	24
REL	18.2	6	0.5	0.76-0.95	7
HC	8.9	24	N/A	N/A	0
Endophenotype: Saccadic Eye Movement
SCZ	44.8	8	1.1	0.71-1.49	15
REL	22.9	4	0.6	0.25-0.85	9
HC	7.7	12	N/A	N/A	0
Endophenotype: Handedness
Non-Right
SCZ	23.9	3	N/A	N/A	0
HC	11.8	3	N/A	N/A	0
Left
SCZ	8.2	11	N/A	N/A	0
REL	5.7	3	N/A	N/A	0
HC	6.2	10	N/A	N/A	0
Mixed
SCZ	32.5	10	N/A	N/A	0
REL	29.2	2	N/A	N/A	0
HC	14.1	10	N/A	N/A	0
Endophenotype: Neuromotor Deviations
SCZ	80.8	3	1.6	1.27-1.89	12
REL	N/A	0	1.6	0.30-2.82	5
HC	9.8	3	N/A	N/A	0

N = number of studies included; N/A = not applicable; unable to calculate based on available published data

Saccadic Eye Movement

For the saccadic eye movement biomarker, a Medline search was completed using the search terms “schizophrenia” combined with “saccadic eye movement”. Data were reviewed similarly to those for smooth pursuit eye movement, except that in addition to using varied pendulum frequency and sine wave, some studies used step-ramp or constant velocity to examine saccadic eye movement. We calculated effect sizes from the reported percent accuracy on the antisaccade task. This measure was chosen because it is independent of the recording apparatus, is easily quantified, and remains consistent across studies. A total of 19 articles fulfilled our selection criteria (Amador et al., 1991; Blackwood et al., 1991; Boudet et al., 2005; Clementz et al., 1992; Clementz, McDowell et al., 1994; Crawford et al., 1998; Ettinger et al., 2006; Fukushima et al., 1988; Holahan and O’Driscoll, 2005; Levin et al., 1988; Levy et al., 2000; Louchart-de la Chapelle et al., 2005; Maccabe et al., 2005; Matthysse et al., 1986; McDowell et al., 1999; O’Driscoll et al., 1998; Price et al., 2006; Sereno and Holzman, 1995; Thaker et al., 2000). Percent abnormal calculations and effect sizes are summarized in Table 3.

Handedness

One of the oldest markers for schizophrenia, mixed- or left-handedness has been purported to be correlated with the development of the disease. Initially, the literature focused on left and non-right handedness. More recently, there has been an interest in mixed-handedness. In order to present a complete review, we included all studies that assessed non-right-, left-, or mixed-handedness in SCZ, REL, and HC. We performed a Medline search with the terms “schizophrenia” combined with “handedness and (left or mixed or dextral or sinistral or non-right)”. Articles were included if they presented a clear description of their handedness assessment. The majority of articles used self-report scales (e.g. Annett Hand Preference, Edinburgh Handedness Inventory) or hand performance tests (e.g. Hand Preference Demonstration Test). While the scoring of handedness varied in strictness between articles, each article presented frequencies for abnormal (all versions of non-right) handedness, which allowed for the comparison of percent abnormal between diagnostic groups between studies. A total of 16 articles fulfilled our selection criteria, with some articles reporting data for multiple handedness categories (3 non-right: O’Callaghan et al., 1995; Sperling et al., 1999; Yan et al., 1985; and 11 left: Clementz, Iacono et al., 1994; Dragovic and Hammond, 2005; Egan, Hyde et al., 2001; Green, Satz, Smith et al., 1989; Lawrie et al., 1999a; Malesu et al., 1996; Nelson et al., 1993; Reilly et al., 2001; Shapleske et al., 2001; Taylor and Abrams, 1984; Upadhyay et al., 2004; and 10 mixed: Dragovic and Hammond, 2005; Egan, Hyde et al., 2001; Giotakos, 2001; Green, Satz, Smith et al., 1989; Gureje, 1988; Malesu et al., 1996; Nelson et al., 1993; Reilly et al., 2001; Taylor and Abrams, 1984; Upadhyay et al., 2004). Percent abnormal calculations are presented in Table 3.

Neuromotor Deviations

For neuromotor deviations, we included all studies that assessed neurological deficits in SCZ, REL, and HC. We performed a Medline search with the search terms “schizophrenia” combined with “neuromotor or neurol* sign, hard sign, or soft sign or NSS or psychomotor”. With the heterogeneity of the neurological assessments used in the literature, any article that provided a general measure of hard signs, soft signs, or global neurological deficit through standardized scales or clinical examination was included in the review. When more than one measure was available, the global deficit was chosen, followed by the soft signs, and lastly the hard signs. A total of 12 articles fulfilled our criteria (Arango et al., 1999; Buchanan and Heinrichs, 1989; Chen et al., 2000; Egan, Hyde et al., 2001; Gourion et al., 2004; Ismail et al., 1998b; Manschreck et al., 1981; Rossi et al., 1990; Sachdev et al., 1999; Taylor and Abrams, 1984; Walker and Green, 1982; Woods et al., 1986). Percent abnormal calculations and effect sizes are presented in Table 3.

Neuropsychological Measures

Wisconsin Card Sorting Task

A Medline search with the terms “Wisconsin Card Sorting Task” combined with “schizophrenia” was conducted. Articles reporting number of categories achieved and perseverative errors in SCZ, REL, and HC were included. Effect sizes were calculated and reported on a total of 41 articles (38 categories achieved: Altshuler et al., 2004; Battaglia et al., 1994; Braff et al., 1991; Cadenhead et al., 1999; Condray et al., 1999; Dieci et al., 1997; Drakeford et al., 2006; Egan, Goldberg et al., 2001; Franke et al., 1992; Glahn et al., 2000; Gold et al., 1997; Goldberg et al., 1998; Gooding et al., 1999; Gooding and Tallent, 2002; Gooding et al., 2001; Haut et al., 1996; Hoff et al., 1992; Hoff et al., 1998; Josman and Katz, 2006; Keefe et al., 1994; Keri et al., 2001; Laurent et al., 2000; Laurent et al., 2001; Merrin et al., 2006; Morrens et al., 2006; Perry and Braff, 1998; Rybakowski and Borkowska, 2002; Seidman et al., 1991; Shum et al., 2004; Snitz et al., 1999; Stratta et al., 2003; Stratta, Daneluzzo, Mattei et al., 1997; Stratta, Daneluzzo, Prosperini et al., 1997; Suhr, 1997; Sullivan et al., 1993; Tallent and Gooding, 1999; Toomey et al., 1998; Wolf et al., 2002; and 30 perseverative errors: Altshuler et al., 2004; Battaglia et al., 1994; Condray et al., 1999; Dieci et al., 1997; Drakeford et al., 2006; Egan, Goldberg et al., 2001; Glahn et al., 2000; Gold et al., 1997; Gooding et al., 1999; Gooding and Tallent, 2002; Gooding et al., 2001; Haut et al., 1996; Josman and Katz, 2006; Keefe et al., 1994; Keri et al., 2001; Laurent et al., 2000; Laurent et al., 2001; Merrin et al., 2006; Rybakowski and Borkowska, 2002; Scarone et al., 1993; Shum et al., 2004; Snitz et al., 1999; Stratta, Daneluzzo, Mattei et al., 1997; Stratta, Daneluzzo, Prosperini et al., 1997; Suhr, 1997; Sullivan et al., 1993; Szoke et al., 2006; Tallent and Gooding, 1999; Wolf et al., 2002; Zanello et al., 2006), with some articles reporting data on both types of scores. Effect sizes are summarized in Table 4.

Table 4.

Mean percent abnormal, effect size, and 95% CI for neuropsychological measures.

Endophenotype: Wisconsin Card Sorting Task
	%Abnormal	N	Effect Size	CI 95%	N
Categories Achieved
SCZ	N/A	0	1.15	0.97-1.31	30
REL	N/A	0	0.38	0.15-0.60	10
HC	N/A	0	N/A	N/A	0
Perseverative Errors
SCZ	N/A	0	0.95	0.75-1.10	24
REL	N/A	0	0.37	0.20-0.55	10
HC	N/A	0	N/A	N/A	0
Endophenotype: Continuous Performance Task
Numbers
SCZ	28.6	1	1.19	0.64-1.75	6
REL	26.6	1	0.22	-0.07-0.51	3
HC	11.4	1	N/A	N/A	0
Shapes
SCZ	30.4	1	1.50	0.90-2.18	6
REL	27.9	1	0.50	0.20-0.84	3
HC	13.9	1	N/A	N/A	0
Endophenotype: Delayed Response Task
SCZ	N/A	0	1.26	0.85-1.67	8
HC	N/A	0	N/A	N/A	0

N = number of studies included; N/A = not applicable; unable to calculate based on available published data

Continuous Performance Task

For the Continuous Performance Test, we included studies that only used the Identical Pairs (CPT-IP) version (B. Cornblatt et al., 1988) to assess performance on the numbers and shapes conditions in SCZ, REL, and HC. Effect sizes were calculated for one of the five major performance indices, d’. A Medline search with the search terms “continuous performance task” and “identical pairs version” and “schizophrenia” was conducted. Effect sizes are reported for a total of 6 articles (Cornblatt et al., 1989; Cosway et al., 2002; Franke et al., 1992; Laurent et al., 1999; Obiols et al., 1992; Roitman et al., 1997) across both categories, with percent abnormal performance reported for one. Percent abnormal calculations and effect sizes are presented in Table 4.

Visuospatial Delayed Response

To examine working memory, we included studies that used a visuospatial delayed response task. A Medline search with the terms “delayed response task” and “schizophrenia” and “visuospatial working memory” and “schizophrenia”. Effect sizes were reported for a total of 8 articles (Coleman et al., 2002; Fleming et al., 1997; Gooding and Tallent, 2004; Lencz et al., 2003; Minor and Park, 1999; Park, 1997; Stratta et al., 1999; Stratta et al., 2001), with all studies only looking at SCZ vs. HC. Effect sizes are presented in Table 4.

Physiologic Abnormalities

Niacin Flushing

A small amount of research has addressed the prevalence of presumed prostaglandin deficiency in schizophrenia with deficient niacin flushing as a potential test for disease liability. For this biomarker, we included all relevant studies on oral or topical niacin administration that compared SCZ, REL, and HC. A Medline search with the term “schizophrenia” combined with “niacin or flush” was completed. After excluding articles that did not present percent abnormal scores (absolute measures of the flush response were not reported in a comparable way), 6 (2 oral: Hudson et al., 1999; Hudson et al., 1997; and 4 topical: Lin et al., 2007; Puri et al., 2001; Puri et al., 2002; Ward et al., 1998) articles fulfilled our criteria. The percent abnormal calculations and effect sizes are presented in Table 5.

Table 5.

Mean percent abnormal, effect size, and 95% CI for physiological abnormalities.

Endophenotype: Niacin Flush
	%Abnormal	N	Effect Size	95% CI	N
Oral Administration
SCZ	43.2	2	0.6	0.64-0.64	2
HC	1.7	2	N/A	N/A	0
Topical Administration
SCZ	60.7	4	0.7	0.26-1.17	5
REL	28.0	2	N/A	N/A	0
HC	24.4	5	N/A	N/A	0

N = number of studies included; N/A = not applicable; unable to calculate based on available published data

Minor Physical Anomalies

Dysmorphology

A well-documented correlate for schizophrenia, minor physical anomalies (MPA) are theorized to represent evidence of a genetic variant and/or prenatal insult, resulting in abnormal physical development, that may serve as a marker for schizophrenia, and that may additionally be a proxy for disturbed neurodevelopment. For dysmorphology, a Medline search with the search terms “schizophrenia” combined with “MPA or physical anomalies or dysmorphology” was completed. Two measures were subsequently chosen that best represented the literature. Waldrop or modified Waldrop scores greater than or equal to 3 were regularly cited as the distinction of abnormal physical anomaly. Minor physical anomaly scores that were not rated through the Waldrop scale were considered abnormal when ≥ 6. The majority of the literature supplied data on at least one of these measures and allowed for the broadest assessment of prevalence. A total of 11 articles (5 Waldrop: Green et al., 1994; Green, Satz, Gaier et al., 1989; Griffiths et al., 1998; Ismail et al., 1998a; Lohr and Flynn, 1993; and 6 non-Waldrop: Gourion et al., 2004; Gualtieri et al., 1982; Ismail et al., 1998a; Lohr and Flynn, 1993; Sivkov and Akabaliev, 2004; Trixler et al., 2001) fulfilled our selection criteria. Effect sizes and percent abnormal calculations are presented in Table 6.

Table 6.

Mean percent abnormal, effect size, and 95% CI for minor physical anomalies.

Endophenotype: Dysmorphology
	%Abnormal	N	Effect Size	95% CI	N
Waldrop > 3
SCZ	22.8	4	0.7	0.35-1.06	5
REL	5.0	3	0.2	-0.27-0.58	4
HC	5.3	5	N/A	N/A	0
MPA > 6
SCZ	34.9	6	1.0	0.73-1.28	6
REL	49.0	2	0.9	-2.05-3.92	2
HC	5.1	6	N/A	N/A	0

N = number of studies included; N/A = not applicable; unable to calculate based on available published data

Summary

Summary statistics (mean and standard deviations) across all endophenotypes reviewed are presented in Table 7. Median effect size and 95% CI for each endophenotype are presented in Figure 1. Also, see Figure 2 for summary of percent abnormals for each category (with the exception of “structural and functional MRI” due to not having percent abnormals available for this category).

Table 7.

Mean percent abnormal, effect size, and 95% CI for all endophenotypes examined.

	%Abnormal	N	Effect Size	95% CI	N
SCZ	39.4	11	0.9	0.85-1.03	15
REL	28.1	8	0.6	0.44-0.77	11
HC	10.2	11	N/A	N/A	0

N = number of endophenotypes included; N/A = not applicable; unable to calculate based on available published data

Figure 1.

Median effect sizes for each endophenotype are plotted with 95% CI. Outliers (denoted by “o”) and extreme cases (denoted by “*”) are labeled by author and year. Note: Outliers were included in main effects summarization tables.

Figure 2.

Average percent abnormal for each endophenotype category. The yellow line represents the expected percent abnormal (2.5%). Note: MRI results are not included in this graph, as there were no percent abnormal values for this category.

Discussion

To examine the prevalence of known endophenotypes for schizophrenia in SCZ, REL, and HC, we reviewed the literature to assess the percent of participants in each category with endophenotypic abnormalities. The intention of this review was to provide a summary of prevalence data for endophenotypes, as well as to begin a discussion on these prevalence rates, especially those of the general population who present with endophenotypes at a rate higher than would be expected based on a statistically normal bell curve (10.2 %, with a range of 0 to 34.6%, compared to the expected 2.5%; Keller and Miller, 2006; Pearlson and Folley, 2008, a,b).

Schizophrenia is a complex genetic disorder, where disease risk for a significant but unknown proportion of individuals is likely epistatic and contributed to by many alleles, individually adding little risk (Waddington et al., 2007), although recent reports also stress that in some cases multiple, individually rare mutations altering genes in neurodevelopmental pathways may also contribute to schizophrenia risk (Walsh et al., 2008). Endophenotypes are presumed to have a simpler genetic structure and to be “closer to the relevant genes” than the overt clinical disorder. When viewed within this context, the results support a hypothesis that random mating of healthy individuals in the general population who have single or small numbers of endophenotypic abnormalities may be more likely to produce offspring with greater numbers of abnormal endophenotypes, thus increasing the risk of their developing schizophrenia (Pearlson and Folley, 2008a,b). At this time, what also remains unknown is how the relevant genes operate under this model (for example how to explain increasing risk in specific terms of genetic recombination/inheritance). In part, this reflects multiple unanswered questions regarding relationships between mechanisms operating in the genetic and in the endophenotype domains.

Because studies examining multiple endophenotypes in the same individuals are only very recent, clustering patterns within patients, relatives, and healthy controls currently remain obscure. Understanding the genetic architecture of such related endophenotypes is likely to prove extremely important in better comprehending what constitutes biological risk for schizophrenia (Ritsner & Gottesman, 2009).

It is important to emphasize that one inevitable limitation of a study such as ours, compared to true population genetic studies, is that we gathered information from multiple, previously published but unconnected studies examining individual endophenotypic abnormalities gathered in separate populations. To compare rates of such abnormalities on a “level playing field” and in a maximally informative manner, data would need to be gathered within a single large study, ideally with multiple assessments in the same individuals.

This highlights the importance of large-scale studies of multiple endophenotypes within and between categories of psychosis, such as Consortium on the Genetics of Schizophrenia (COGS; Calkins et al., 2007) and The Bipolar Schizophrenia Network on Intermediate Phenotypes (B-SNIP [http://www.b-snip.org/]; Thaker, 2008). These recent studies have used such approaches across multiple geographic sites using the same equipment and assessment methods with extensive cross-validation and cross-training. An additional advantage of some of these studies is that they address the question of whether endophenotypic risk is specific to schizophrenia or generalizes to psychosis across various conditions, including bipolar disorder.

A major limitation of this review was the relatively modest number of studies that reported percent of abnormal findings within each group, or event sufficient data to allow for a post-hoc calculation of percent abnormal or effect sizes in healthy controls. We propose, based on the findings from this review, that the inclusion of such data in future research on schizophrenia endophenotypes would be beneficial in terms of providing the basis for discussion on the points raised in this paper, as well as allowing for a standardized comparison of findings across the literature as a whole. This issue may be related to our choice of effect size statistic. We chose to use Cohen’s d in the context of a two-group comparison (analogous to a t-test) between SCZ and HC. Although Hedge’s g is often used when sample sizes are different, we did not find dramatic differences between sample sizes of SCZ and HC in the studies that we had investigated, with the one exception of pre-pulse inhibition (F=11.07, p=0.003).

Another limitation, again not addressable in a review article such as this, is the possibility that that the definition of “healthy controls” (i.e., non-relative vs. non-relative with no other psychiatric illness) varied from study to study. In addition, there is a possibility that healthy controls, usually volunteers, were not stringently screened for psychiatric illnesses or family psychiatry psychiatric history, leading to a potential confound in terms of the prevalence of these endophenotypes in the general population. In fact, these “healthy controls” may not be fair representatives of the general population, as discussed elsewhere (Raz et al., 1988; Schwartz and Link, 1989; Shtasel et al., 1991); Smith et al., 1988). In the studies that we reviewed, only a little over half used a structured clinical interview (e.g. SCID-IV-TR), about a third used self-report of no psychiatric history, and the rest either did not screen their healthy controls or did not report their exclusion methods. In addition, the studies we reviewed seldom performed toxicology screens for alcohol or drugs, or at least did not include this information in their methods section.

This issue also relates to the “file drawer” problem when using published studies; it is important to address the general caveat to our findings, as with any systematic review of published data, may inherently misrepresent true population-wide statistics as negative or equivocal findings tend to be under-reported in the literature (Rosenthal, 1979).

A counterargument to the significance of the larger than expected number of HC with endophenotypes is the risk of assuming normality in endophenotypic traits (i.e., that these traits follow a Gaussian bell curve). Unless data are provided to allow a determination of whether or not a trait is normally distributed, normality is assumed. Some traits have been explored for normal distribution trends with different results depending on the measure. Structural brain volume (Lange et al., 1997), including all of the areas that were examined in this study except left superior temporal gyrus, and dysmorphology (Ward et al., 1998), for instance, have been shown to be normally distributed, while some of the neuromotor traits (e.g. smooth pursuit and saccadic eye movement) have not (O’Driscoll et al., 1998), and still others have not yet been explored.

Despite the limitations of this review, these findings provide a basis for future explorations of the prevalence of schizophrenia endophenotpyes in SCZ, REL, and HC. Especially notable is that these biological markers seem to be present in a larger percentage of individuals than would be expected based on our assumptions of statistical normality, and perhaps have an evolutionary role that has yet to be fully uncovered.