Schizophrenia may have etiological heterogeneity, and may reflect common symptomatology caused by many genetic and environmental factors. In this review, we show the potential existence of heterogeneity in schizophrenia based on the results of our previous studies. In our study of the NOTCH4 gene, there were no significant associations between any single nucleotide polymorphisms (SNPs) of NOTCH4 and schizophrenia. However, exploratory analyses suggested that the SNP, rs3134928 may be associated with early-onset schizophrenia, and that rs387071 may be associated with schizophrenia characterized by negative symptoms. In our highly familial schizophrenia study, the African-American cohort without environmental exposure showed a possible linkage at marker 8p23.1 in the dominant model and in the European-American cohort, a marker at 22q13.32 showed a probable linkage in the recessive model. In the less familial schizophrenia families, these linkages were not shown. Based on our eye movement study, a putative subtype of schizophrenia with severe symptoms related to excitement/hostility, negative symptoms and disorganization may be associated with chromosome 22q11. We consider that a sample stratification approach may clarify the heterogeneity of schizophrenia. Therefore, this approach may lead to a more straightforward way of identifying susceptibility genes of schizophrenia. © 2013 Wiley Periodicals, Inc.
Schizophrenia is a severe psychiatric illness with a prevalence of approximately 1%. Epidemiological studies of schizophrenia have shown that genetic components play an important role in the development of schizophrenia [Tsuang and Faraone, 1995; Tsuang, 2000]. Its heritability was estimated to be ∼80% [Sullivan et al., 2003]. However, the genetic etiology of schizophrenia is still unclear. One of the most important sources of conflict may be the potential etiological heterogeneity of schizophrenia. Many investigators have suggested that schizophrenia is not a single disease entity, but may reflect common symptomatology caused by many genetic and environmental factors [Tsuang et al., 1990; Tsuang and Faraone, 1995; Tsuang, 2000; Sawa and Snyder, 2002; Takahashi et al., 2005]. Since human diseases have a noticeable genetic heterogeneity [McClellan and King, 2010], when studying the genetic etiology of schizophrenia, heterogeneity should be part of the study design [Sebat et al., 2009].
Recent copy member variant (CNV) studies also indicate the heterogeneity of schizophrenia. Based on the findings of CNV studies, an understanding of the genetic heterogeneity of complex disorders, such as schizophrenia and autism, has begun [Bassett et al., 2010]. Because results have shown a rare mutation with a large CNV effect and a common mutation with small/modest effect by single nucleotide polymorphism (SNP), schizophrenia may be characterized by much more genetic heterogeneity than was formerly thought [Sebat et al., 2009]. A microdeletion at 22q11.2 was the first de novo CNV shown in schizophrenia [Karayiorgou et al., 1995]. Earlier findings of the 22q11.2 deletion and recent CNV studies have highlighted the 22q11.2 deletion as a major factor in the genetic architecture of schizophrenia [Rodriguez-Murillo et al., 2012]. Patients with the 22q11.2 deletion (22q11.2 deletion syndrome: 22q11.2DS; also called Digorge/velocardiofacial syndrome) may be a subtype of schizophrenia [Bassett et al., 2010]. More than 20% of patients with 22q11.2DS develop schizophrenia [Bassett et al., 2005].
The detection of potential mutations for pathogenesis may be inhibited by the heterogeneity of schizophrenia. In this review, we show proof of heterogeneity in schizophrenia based on the existing literature. Moreover, from recent CNV studies, we discuss the etiological heterogeneity of schizophrenia.
THE NOTCH4 GENE
Wei and Hemmings  firstly reported that schizophrenia was significantly associated with the NOTCH4 gene. However, Sklar et al.  found no significant association between NOTCH4 and schizophrenia. McGinnis et al.  also showed no significant association between NOTCH4 and schizophrenia.
We also performed a family-based association study of NOTCH4 and schizophrenia [Takahashi et al., 2003a]. In this study using Sanger sequencing, we sequenced genomic DNA around rs387071 that was a significant SNP in the original study [Wei and Hemmings, 2000]. We identified four new SNPs near rs387071. There were no significant associations between any of the five SNPs and schizophrenia.
However, exploratory analyses revealed significant results. To perform the family-based association analysis while taking into account the clinical heterogeneity of the disorder, we examined the age at onset of schizophrenia and the type of schizophrenic symptoms as assessed by the Positive and Negative Syndrome Scale (PANSS) [Kay et al., 1991]. Schizophrenic offspring with age at onset younger than 19 were defined as early-onset schizophrenia (EOS). Also, when we classified samples into subtypes according to PANSS criteria, both negative type and mixed type had many negative symptoms; thus, the negative and mixed types were merged into a group called “schizophrenia with many negative symptoms” (SMNS). The A allele of rs3134928 was transmitted significantly more often to EOS offspring (allele by allele: P = 0.0127; global allelic: P = 0.0076). Furthermore, in the SNP rs387071, the A allele was transmitted significantly more frequently to the SMNS offspring (allele by allele: P = 0.0074; global allelic: P = 0.0045).
HIGHLY FAMILIAL SCHIZOPHRENIA WITHOUT ENVIRONMENTAL EXPOSURES
The National Institute of Mental Health Human Genetics Initiative (NIMH-HGI) was created as a national repository of DNA for genetic studies. Faraone et al.  performed a linkage analysis of the NIMH-HGI schizophrenic sample in European-Americans (EA). They reported that two markers on chromosome 10p showed statistical significance suggestive of linkage. Kaufmann et al.  did a linkage analysis of the NIMH-HGI sample in African-American (AA). They presented several regions, including chromosomes 6q16–6q24, 8pter–8q12, 9q32–9q34, and 15p13–15q12 that showed evidence consistent with linkage. In light of the potential etiological heterogeneity of schizophrenia, we reanalyzed the NIMH-HGI linkage data for schizophrenia [Cloninger et al., 1998; Faraone et al., 1998; Kaufmann et al., 1998]. In reanalyzing this data, we tried to reduce etiological heterogeneity [Takahashi et al., 2005]. It is plausible that the use of more homogeneous subsets may lead to gene identification. We considered that cases without environmental exposures known to influence schizophrenia might form a schizophrenia groupin which the causal pathway to disease is more likely to be affected by genes and not the environment. Furthermore, cases with many affected family members are also likely to have a strong genetic loading for schizophrenia (we denote this group as “highly familial”). Therefore, we reanalyzed data from the NIMH-HGI for schizophrenia dividing into sub-groups based on the number of affected individuals in the pedigree and removing all families with environmental exposures.
Highly familial schizophrenia pedigrees were defined as those pedigrees having at least one nuclear family with either (1) schizophrenia or schizoaffective disorder in at least one parent of the affected children or (2) at least three siblings with schizophrenia or schizoaffective disorder.
To further generate homogeneity in the highly familial schizophrenia sample, we excluded cases of schizophrenia possibly caused by non-genetic factors. Research suggests that obstetric complications (OC) and viral infections during neurodevelopment (VIN) can lead to schizophrenia [Tsuang and Faraone, 1995; Tsuang, 2000]. Therefore, we excluded schizophrenic or schizoaffective disorder patients with either of these conditions from the highly familial schizophrenia sample using information provided in the Diagnostic Interview for Genetic Studies (DIGS) version 2.0 [Nurnberger et al., 1994; Faraone et al., 1996]. The DIGS asks the following questions: (1) “Was your own birth or early development abnormal in any way?”; (2) “Were there any problems with your mother's health while she was pregnant with you, or with your birth, such as prematurity or birth complications?”; and (3) “Was your development abnormal in any way, for example, did you walk or talk later than other children?” Schizophrenic or schizoaffective disorder patients with a clearly positive answer to at least one of these questions were defined as having either OC or VIN and were therefore removed from the highly familial subgroup.
Using this procedure, we created the following schizophrenia subgroups: (1) the entire ethnic group with no individuals removed; (2) a highly familial subgroup (HFS), but still includes individuals with environmental influences (OC or VIN); and (3) a HFS without OC or VIN. Due to sample size limitations, we only compared the first two groups in the EA sample.
Results showed that in the AA HFS without OC or VIN group, a marker at 8p23.1 showed a single point LOD score (SLOD) of 2.90 and a multipoint LOD score (MLOD) of 2.11 in the dominant model. Allowing for locus heterogeneity, this marker showed a single point heterogeneity LOD (SHLOD) of 3.04 (alpha = 0.88) and a multipoint heterogeneity LOD (MHLOD) of 2.11 in the dominant model.
Moreover, results in the EA HFS sample, a marker at 22q13.32 had a SLOD of 0.91 and a MLOD of 1.85 in the recessive model. Allowing for locus heterogeneity, this marker showed a SHLOD of 1.64 (alpha = 0.61) and a MHLOD of 1.97 (alpha = 0.65) in the recessive model. When analyzing the ethnic groups (AA or AE) in their entirety, no linkages were found.
EYE MOVEMENT DYSFUNCTION AND CLINICAL SYMPTOMS
We and others [Kojima et al., 1990, 1992, 2001] have studied eye movement dysfunction using the exploratory eye movement (EEM) test and have shown that subjects with schizophrenia show abnormalities. We have also shown that the siblings of schizophrenics show EEM dysfunction [Takahashi et al., 2008]. In addition, those subjects with EEM dysfunction have a significant linkage to chromosome 22q11.28 [Takahashi et al., 2003b]. As mentioned in the Introduction, 22q11 is one of the most interesting regions in the genetic etiology of schizophrenia [Rodriguez-Murillo et al., 2012]. Detailed descriptions of the EEM test method have been presented in previous studies [Kojima et al., 1992; Suzuki et al., 2009].
On the basis of these findings, we considered that the EEM test might be useful for the clinical diagnosis of schizophrenia. Suzuki et al.  carried out a discriminant analysis between schizophrenics and non-schizophrenics in a large sample using EEM test data. EEM performances were recorded in 251 clinically diagnosed schizophrenics and 389 clinically diagnosed non-schizophrenics (111 patients with mood disorders, 28 patients with neurotic disorders, and 250 normal controls). As a result, 184 of the 251 clinically diagnosed schizophrenics were discriminated as having schizophrenia (sensitivity: 73.3%); and 308 of the 389 clinically diagnosed non-schizophrenic subjects were discriminated as non-schizophrenics (specificity: 79.2%).
In the discriminant-analytic study [Suzuki et al., 2009], we were interested in characteristics of the schizophrenic patients who were discriminated as having schizophrenia (SPDS), or those who were discriminated as not having schizophrenia (SPDNS). We considered that EEM parameters could detect different subtypes of schizophrenia. Hence, to clarify features of SPDS and SPDNS by the EEM test, we reanalyzed data of our study [Suzuki et al., 2009] and investigated symptomatic characteristics [Suzuki et al., 2012].
Two hundred fifty-five schizophrenic patients participated in our discriminant-analytic study [Suzuki et al., 2009] and were assessed using the Brief Psychiatric Rating Scale (BPRS) [Overall and Gorham, 1962]. For group comparisons (SPDS vs. SPDNS) of the symptomatic data, scores for factors extracted by factor analysis of BPRS ratings and BPRS total scores were used.
The principal component analysis extracted five factors using 255 schizophrenic samples: Factor 1: excitement/hostility (17.5% of total variance); Factor 2: negative symptoms (17.5%); Factor 3: depression/anxiety (14.1%); Factor 4: positive symptoms (12.6%); and Factor 5: disorganization (8.4%).
As seen in Figure 1, SPDS had significantly higher scores for excitement/hostility (P = 0.031), negative symptoms (P = 0.011), and disorganization (P = 0.040) than SPDNS. Furthermore, the BPRS total score of SPDS was significantly higher than that of SPDNS (P = 0.003).
In our NOTCH4 schizophrenia study [Takahashi et al., 2003a], of the five SNPs analyzed, none were found to be statistically significant. Although our data suggest that NOTCH4 is not a significant susceptibility gene for schizophrenia when clinical heterogeneity is ignored, we found some preliminary evidence for associations with clinically homogenous strata. The SNPs rs3134928 may be associated with early-onset schizophrenia, and rs387071 may be associated with schizophrenia with many negative symptoms.
Several studies did not find significant associations between NOTCH4 and schizophrenia [Glatt et al., 2005] but recent larger genome-wide association studies revealed an association between NOTCH4 and schizophrenia [Stefansson et al., 2009]. These results may be based on strong statistical power due to the use of large samples. At any rate, the NOTCH4 gene is attracting attention once more. Moreover, based on our findings, NOTCH4 may be related to early-onset schizophrenia and schizophrenia with many negative symptoms. Future studies using these subtypes may be more likely to detect an association between NOTCH4 and schizophrenia.
In the highly familial schizophrenia study [Takahashi et al., 2005], we used linkage analysis data to compare findings in more and less homogeneous schizophrenia subgroups. For the AA HFS without environmental exposures, the possible linkage resulted from the dominant model at marker 8p23.1. For the EA HFS, the marker at 22q13.32 showed the probable linkage in the recessive model. In the combined sample, linkage was not found.
From these analyses, it appears as if the linkages were stronger in the HFS. Moreover, selecting families with clusters of affected individuals and excluding individuals with environmental exposures, such as OC or VIN, may generate more genetically homogenous subtypes that can detect susceptibility genes for schizophrenia.
Our two studies analyzing schizophrenia subcategories suggest that using more genetically homogeneous groups to be a more effective way to identify susceptibility genes [Takahashi et al., 2003a; Takahashi et al., 2005] for schizophrenia. However, there are major limitations that should be considered. The sample size in each of these studies was small and the statistical power low. Therefore, these results should be interpreted with caution. Even though these findings among the schizophrenia subcategories may be false positives, the sample stratification approach we used may be effective for clarifying genetic heterogeneity. Additional studies using similar stratifications with larger samples are needed to confirm our findings.
In our eye movement study, Suzuki et al.  performed the discriminant analysis between schizophrenics and non-schizophrenics using the eye movement disturbances. As a result, 184 of the 251 clinically diagnosed schizophrenics were found to have schizophrenia (73.3%).
In another study [Suzuki et al., 2012], results of a factor analysis of BPRS ratings from the above 251 schizophrenic subjects revealed five symptoms (excitement/hostility, negative symptoms, depression/anxiety, positive symptoms and disorganization). Excitement/hostility, negative symptoms and disorganization were more predominant in the 184 SPDS subjects compared to the SPDNS subjects. Furthermore, the BPRS total score of SPDS was significantly higher than that of SPDNS. Consequently, SPDS may consist of severe schizophrenia; and the severity of symptoms in SPDS is mainly due to excitement/hostility, negative symptoms, and disorganization.
From these findings, SPDS may have three different symptoms (excitement/hostility, negative symptoms, and disorganization); however, it can also be said that SPDS may be a schizophrenia subtype characterized by these three factors. Our findings may indicate that there is a putative subtype of schizophrenia with severe symptoms related to excitement/hostility, negative symptoms and disorganization.
In our previous study, we found a significant linkage to chromosome 22q11.2–12.1 using eye movement impairment as an endophenotype of schizophrenia [Takahashi et al., 2003b]. Thus, if eye movement dysfunction is associated with the schizophrenia subtype with severe symptoms related to excitement/hostility, negative symptoms and disorganization, chromosome 22q11 and the genes of 22q11 may be related to this subtype. In this manner, if we are able to find a new subtype, and clarify the heterogeneity of schizophrenia, an association study for schizophrenia using the subtype may yield additional knowledge regarding the genetic influences on schizophrenia.
The CNV findings indicate that schizophrenics are beginning to do classification based on causal factors [Bassett et al., 2010]. The 22q11.2 microdeletion is one of the strongest risk factors for schizophrenia [Rodriguez-Murillo et al., 2012]. Schizophrenic patients with the 22q11.2 deletion and those without it are clinically indistinguishable [Bassett et al., 2003]. On the basis of this issue, it can be said that schizophrenia may have an etiological heterogeneity by the 22q11.2 deletion. Our eye movement and linkage study may identify a putative subtype associated with 22q11.2. Based on our finding, patients with the 22q11.2 deletion and without it may be distinguishable. While these two findings may have discrepancies, schizophrenia may have an etiological heterogeneity in the 22q11.2 region.
In addition to CNV at the 22q11.2 locus, several CNVs are found to be enriched in schizophrenia, including deletion/duplications at 1q21.1, 3q29, 7q36.3, 15q11.2, 15q13.2–15q13.3, 16p11.2, and 17p12 [Doherty et al., 2012; Rodriguez-Murillo et al., 2012]. Bassett et al.  reviewed six genome-wide CNV studies of schizophrenia published in 2008. They reported that patients with the 1q21.1 and 15q13.3 deletions may also be putative subtypes of schizophrenia, respectively. Stratifying schizophrenics by specific CNVs may enhance power to reduce the etiological heterogeneity.
In conclusion, based on many studies including our findings, we consider that schizophrenia may have etiological heterogeneity. The sample stratification approach, which clarifies the heterogeneity of schizophrenia using clinical and molecular biological methods, may lead to a more straightforward way to identify susceptibility genes of schizophrenia. However, when we perform this approach, the resulting sample size tends to be small. In the future, using this approach with a large sample may prove fruitful for studying the genetics of schizophrenia.
The author appreciates the assistance of Dr. Ming T. Tsuang, University of California, San Diego, and Dr. Stephen V. Faraone, SUNY Upstate Medical University, with the genetic studies.