Kraepelin's 1896 dichotomous classification of functional psychoses into dementia praecox (schizophrenia) and manic-depressive insanity (bipolar disorder) is well known and has dominated most diagnostic systems in psychiatry. Less well known is the fact that two decades later, Kraepelin himself questioned the validity of his classificatory system, noting that "it is becoming increasingly clear that we cannot distinguish satisfactorily between these two illnesses and this brings home the suspicion that our formulation of the problem may be incorrect." [1, 2] In recent years, increasing number of observers have shared Kraepelin's doubts, with some proposing instead a continuum hypothesis [3, 4]. Clinically, there is much overlap of symptoms, and the intermediate category of schizoaffective disorder may imply an artificial preservation of the dichotomy between these two diagnoses [5]. The clinical gulf between individuals with bipolar II and those with bipolar I with psychotic features is marked, whereas the latter often resemble individuals with schizophrenia quite closely. Epidemiologically, there is also some overlap, especially in so far as both have a winter-spring seasonal birth excess [6, 7].

Genetically, it has been said that "accumulating evidence supports the existence of an overlap in genetic susceptibility across the traditional Kraepelinian divide" [8]; this has been reported for family studies, linkage studies, and candidate genes such as G72, NRG1, dysbindin, COMT, BDNF, and DISC1 [9–12]. An overlap for some neuropsychological dysfunctions, such as working memory [13], has also been reported, although individuals with bipolar disorder are less severely impaired [14]. This suggests that working memory dysfunction may be closely related to psychotic features in individuals with schizophrenia and with bipolar disorder.

Neuropathologically, an overlap in findings between schizophrenia and bipolar disorder has also been noted, both neurochemically [15] and structurally [16]. As was true for the neuropsychology findings, individuals with bipolar disorder are structurally less severely impaired. Neuropathological studies that have divided bipolar patients into those with and without psychotic features have reported that the former are more closely related to schizophrenia than the latter [17]. This last observation suggests that an important variable in categorizing seriously mentally ill individuals may be the presence or absence of psychotic features. To date, most studies of gene expression in schizophrenia and bipolar disorder have used the traditional Kraepelinian classification. We, therefore, decided to reanalyze a set of microarray studies, dividing the individuals within each study into those with and without psychotic features.

Although genome-wide expression studies of psychiatric disorders have been published, there have been inconsistent findings among previous studies possibly due to several factors including: relatively small sample sizes, inappropriate p-value and fold change criteria, and failure to adjust for potential confounding variables. In order to address these issues, we performed a cross-study analysis of 7 microarray datasets using the subjects with and without psychotic features from the two post-mortem brain collections of the Stanley Medical Research Institute (SMRI). The microarray datasets of the prefrontal cortex with three Affymetrix platforms (hgu133A, hgu133 2.0+ and hgu95Av2) were used for this cross-study analysis.

In the current study, we have found the genes that are differentially regulated in the prefrontal cortex of individuals with psychosis. These findings are important because we performed a cross-study analysis of seven microarray studies that used three different Affymetrix platforms with two different post-mortem brain collections. Therefore, these gene expression changes are "consensus" fold changes across multiple microarray studies following adjustment of confounding variables in individual studies.

It is well known that gene expression studies of psychiatric disorders using post-mortem brains are more challenging than others [25–28]. Possible reasons for this complexity include pre- and post-mortem factors such as RNA quality, brain pH, PMI and agonal state that affect gene expression patterns in the brains of psychiatric patients [23]. Moreover, biological and disease-specific effects are often hindered by several issues such as relatively small sample sizes, small effect sizes and heterogeneity of psychiatric phenotypes in the general population. The clinical information available for the patients is typically sparse, so that unknown clinical covariates may either confound or confuse many of the gene expression findings in psychiatric disorders [20]. Therefore, appropriate statistical adjustment using such demographic and clinical information is critical in order to improve inferences in determining putative genes and biological pathways in psychiatric disorders.

Gene expression microarray technology has become a powerful tool for discovery, although this technology had some limitations such as high cost of experiments, limited sample size, and lack of standardization on data analysis. Recently, there have been improvements in the microarray field, and more microarray datasets are available for meta-analysis. Several studies have described important issues related to the meta-analysis of microarray data including standardization, statistical analysis and databasing strategy [29], and the development of new bioinformatic tools such as searchable online archives of genetic/genomic findings in psychiatric disorders [30]. One of the main advantages of meta-analysis as compared to an individual study is the increased statistical power to detect consensus fold changes across multiple studies. It has been reported that gene expression changes in post-mortem brains of psychiatric patients are relatively small. This may explain some of the discrepancies in previous studies that reported putative genes in psychiatric disorders, because most of the studies were not sufficiently powered (due to small sample sizes) to detect small changes in gene expression. Moreover, some studies did not have sufficient clinical information of post-mortem brains to adjust for possible confounding variables.

In the current analysis, we have included seven microarray studies that used the frontal cortex (BA46/10, BA8/9 and BA6) as compared to the previous study [21] that included the cerebellum studies. A recent study demonstrated that different brain regions such as the frontal, cingulate, temporal, parietal, occipital cortices, hippocampus and sub-cortical regions show different patterns of gene expression [31]. The authors also found that gene expression patterns in the prefrontal cortices (BA10 and BA46) are similar to each other as compared to non-frontal regions. To our knowledge, there is little evidence on gene expression differences in the sub-regions of the frontal cortex because most studies have focused on one region of the frontal cortex of schizophrenia patients. Another limitation was an availability of the prefrontal cortex (BA46) tissues from the Neuropathology Consortium when individual microarray studies were conducted. This brain cohort has been used by many investigators for various types of research in the past. Therefore, we included 3 studies that used the frontal cortices (BA46/10, BA8/9 and BA6) from the Neuropathology Consortium because these regions are adjacent to each other and have been implicated in schizophrenia [32–34]. Nevertheless, we included another 4 studies that used the Array Collection with the same dorsolateral prefrontal cortex (BA46).

In the current cross-study analysis, we identified a set of metallothionein genes including MT1X, MT1K, MT1E, MT1H, MT1F, MT2A and MT3 that are significantly up-regulated in psychosis. We used a real-time qPCR to validate differential expression of metallothionein genes in the prefrontal cortex (BA46) of psychosis patients. Real-time qPCR is often referred as the gold standard for validation of gene expression in microarrays due to its advantages in detection sensitivity, large scale dynamic range, high precision and reproducible quantitation compared to other techniques [35, 36]. For the qPCR, we used a complete set of dorsolateral prefrontal cortex (BA46) tissues from the Array Collection (psychosis N = 56, non-psychosis N = 49). It is important to note that most previous microarray studies have used a relatively small number of post-mortem brains for the qPCR validation.

Although metallothioneins were discovered almost 50 years ago, their functional role in the brain has not been well-characterized. Putative physiological functions of metallothioneins in the CNS, including neuroprotection, regeneration and cognitive function, have been described in a recent paper [37]. Other studies reported that metallothioneins are involved in cellular host defense response, stress response, immunoregulation, cell survival and brain repair [38–41]. Astrocytes appear to be the main source of metallothioneins in the brain, even though the primary function of metallothioneins are protecting neurons from a variety of pathology [42]. Increased MT1 and MT2 gene expression have been reported in several types of brain pathology including traumatic and excitotoxic injury, amyotrophic lateral sclerosis, Alzheimer's disease and Parkinson's disease [43–47]. Studies using animal models showed that metallothioneins may be involved in behavior associated with substance dependence [48] and learning and memory [49]. Similar to our findings, one study reported that MT2A gene expression is increased in the PFC of schizophrenia patients [50]. These results suggest that metallothioneins in the brain may play a significant role in neuroprotection and cognitive function. It is possible that neurodegenerative process may disrupt cognitive function in the frontal cortex of psychosis patients and up-regulation of metallothionein genes in the PFC may be a compensatory mechanism against these adverse processes. Interestingly, a recent study reported that metallothionein-related compounds are well tolerated in animal studies [39], and therefore, these compounds may be potential candidates for novel medication development in psychosis-related disorders.

The genes down-regulated in psychosis include neuropeptide genes such as somatostatin (SST), neuropeptide Y (NPY) and tachykinin (TAC1). The neuropeptide genes have been implicated in working memory function and in schizophrenia [51]. The NPY gene may be associated with schizophrenia [52], impulsivity [53], aggression and bipolar disorders [54, 55]. Previous studies reported decreased NPY levels in the temporal cortex of schizophrenia [56] and decreased NPY receptor gene expression in lymphocytes of schizophrenia [57]. Gabriel et al reported that both NPY and SST levels are decreased in the cerebral cortex of individuals with schizophrenia [58]. A microarray study reported that SST gene expression is increased in the PFC of bipolar disorder but not with schizophrenia patients [59]. It is possible that this study may not have been sufficiently powered (due to small sample sizes) to detect small gene expression changes in schizophrenia. Consistent to the current results, mRNA levels of SST in the PFC are decreased in schizophrenia patients in a study using in situ hybridization [60] and another microarray study reported that both SST and NPY expression levels are decreased in the PFC of schizophrenia [51]. TAC1-related genes are also implicated in psychiatric disorders [61, 62] and increased protein levels of TAC1 receptor in the PFC of schizophrenia have been reported [63]. Taken together, previous findings and the current results provide strong evidence for involvement of neuropeptide genes in psychosis-related disorders.

One of the potential confounding variables affecting neuropeptide gene expression may be the antipsychotic medication. Previous studies reported that NPY levels in the brain are affected by various drugs including typical and atypical antipsychotics [64, 65]. Using rodents, one study reported that NPY and SST levels are increased in the PFC following haloperidol treatment [66], while another study reported that NPY levels are decreased following chronic antipsychotic treatment [67]. However, GABA-system related genes including SST are not changed in the PFC of monkey brains treated with chronic antipsychotic drugs [51]. In order to minimize the confounding effect of antipsychotic medication from the disease effect, we used multiple regression models to adjust antipsychotic medication in the current analysis.

Another gene down-regulated in psychosis is the transcription factor Nurr1 (NR4A2), an orphan nuclear receptor associated with the development of dopaminergic cells in the mid-brain (p < 0.00001, FC-1.16). This gene has been implicated in attention deficit hyperactivity disorder, alcohol dependence, schizophrenia and bipolar disorder [68–70]. A recent study with NR4A2 knock-out mice suggests that these animals display behavioral endophenotypes that are displayed in other animal models of psychosis/schizophrenia. [71]. Therefore, reduced NR4A2 gene expression in the PFC may be associated with psychotic features.

Although genome-wide gene expression profiling in the postmortem brains may reveal valuable information related to psychiatric disorders, this approach alone is limited in terms of being able to distinguish between changes reflecting the primary disease etiology from those reflecting compensatory mechanisms and many potential confounding influences such as medications and drug use [72]. Therefore, it is necessary to study the relationship between gene expression, genetic and epigenetic variations in individuals with psychiatric disorders. The integration of postmortem gene expression with genetic variations including single nucleotide polymorphisms (SNPs) is the subject of ongoing investigations.

We found a set of genes that are differentially regulated in the prefrontal cortex of individuals with psychosis. Among those genes, metallothionein genes are consistently up-regulated and neuropeptide genes are down-regulated. These genes may have implications in neurodegeneration and working memory function in the prefrontal cortex that are thought to be disrupted in psychosis. With an advantage of cross-study analysis, we were able to detect small but "consensus" changes in gene expression across multiple microarray studies. By combining gene expression analyses across multiple microarray studies and with the qPCR validation of putative genes, we can provide greater confidence in the scientific findings, which include potential genes and biological pathways in psychiatric disorders, as compared to the result obtained from a single study.