Background
Bipolar disorder (BP) is a serious mental illness with considerable public health implications. It affects 1-2% of the general population [
1], and costs the United States approximately $78.6 billion dollars annually in direct and indirect costs [
2]. It is clear from family, twin and adoption studies that genetic factors play an important role in BP. Family studies show that compared to the general population the risk of disease is 5–10 times greater in first-degree relatives of a proband with bipolar disorder, and estimates of its heritability from twin studies range from 80-90% [
3]. Yet, despite the overwhelming evidence, the genetic causes of BP remain largely unknown. This is likely due to the fact that the etiology of BP is complex and probably involves multiple independent and interacting genetic factors [
4].
Microarray technology provides a powerful tool for studying the genetic contribution to complex disorders [
5]. It allows for the measurement of gene expression levels genome-wide in a range of tissues and across disease conditions. A number of studies have used this technology to examine expression differences in BP versus unaffected controls with the goal of identifying genes or pathways of genes that are up or down regulated in the disorder [
6,
7]. These studies have typically used RNA samples from either peripheral blood or brain tissue [
8]. The advantage of the former is that it is relatively easy to collect from participants. However, it may not be the relevant tissue for psychiatric disorders, that presumably have origins in the brain, and there may be constitutive differences in gene expression between blood and the brain. By contrast, the brain is the relevant tissue to study for BP. The disadvantage of brain tissue is that it can only be collected after the participant is deceased, which may limit the ability to collect sufficiently large samples. Additionally, because of the relative instability of RNA, post mortem factors (for e.g. brain tissue pH, coma, respiratory arrest, hypoxia, seizures, dehydration, multiple organ failure, and head injury) may confound the relationship between measured expression levels and disease status [
9,
10]. As a result, findings from studies using brain tissue have largely been inconsistent.
In order to synthesize the current findings to increase accuracy, we carried out a systematic review of existing gene expression studies of BP in humans. Motivated by the consideration that studies with brain would be the most informative for the etio-pathogenesis of BP, we conducted a quantitative mega-analysis of those studies carried out with this tissue. By combining data across studies to increase the sample size and using consistent procedures to process and analyze the data, we sought to summarize the findings from these studies and clarify their relevance for BP. The findings from this analysis are made available on Metamoodics (
http://metamoodics.igm.jhmi.edu), a bioinformatics resource that synthesizes the results from genomic experiments in mood disorders and displays them within their genomic context.
Discussion and conclusion
We report here the results of systematic review of gene expression studies in BP. BP is a complex disorder with a considerable genetic component that has been challenging to resolve. Gene expression studies may help to identify genes or sets of genes that are up or down regulated in the disorder and thereby provide clues about its genetic underpinnings. At least 30 studies using modern array-based technology to assay gene expression genome-wide have been published on BP. Most of these have studied expression in either blood or brain tissue samples. Although blood samples are easier to collect, brain samples provide more direct access to changes in the tissue most relevant to psychiatric disorders. We, therefore, conducted a quantitative mega-analysis of the most recent and robust of studies on the brain in BP in order to synthesize the findings and provide a comprehensive overview of what is currently known from these efforts.
The most significant findings were observed in the analysis of the PFC. This may reflect the central role the prefrontal cortex is thought to play in mood disorders, especially bipolar disorder [
66]. However, it may also be due to the fact that the PFC was the focus of more studies than any other brain region. Although the analysis of any brain regions included more studies, these studies covered several different brain regions including the PFC, which may have introduced heterogeneity and diluted the findings. The analysis of the hippocampus only included two studies and was, therefore, relatively underpowered to detect differentially expressed genes.
In the PFC, there were 11 genes with a q-value < 0.05. Among these were two genes of great interest in mood disorders:
FKBP5 and
WFS1. Mutations in
WFS1 are known to cause Wolfram syndrome, a disorder characterized by insulin deficiencies leading to high blood sugar levels and progressive vision loss, and which often co-occurs with psychiatric disturbances such as mood disorders. Several studies have directly implicated
WFS1 in the etio-pathogenesis of bipolar disorder [
67].
FKBP5, on the other hand, encodes for FK506 binding protein 5, a co-chaperone of the glucocorticoid receptor heterocomplex, which mediates downstream effects of cortisol. The role of
FKBP5 and cortisol dynamics have been the focus of intense investigations in mood disorders [
68,
69] and response to antidepressant treatment [
70,
71]. Interestingly,
CRH (corticotrophin releasing hormone) [
72,
73], another key gene underlying cortisol action, was identified as differentially expressed in the analysis of PFC. Several other notable candidate genes for mood disorders were implicated in the current analyses, including
DUSP6 (dual-specificity phosphatase 6) [
74‐
76],
NPY (neuropeptide Y),
NR4A2 (nuclear receptor subfamily 4, group A, member 2),
SST (somatastatin),
GRIK2 (glutamate receptor ionotropic kainate 2 isoform precursor) [
77‐
79],
S100B (S100 calcium binding protein B) [
80,
81] and
CACNA1C (calcium channel, voltage-dependent, L type, alpha 1C subunit). Perhaps of greatest interest among these is
CACNA1C, which has emerged from recent genome-wide association studies as one of the leading candidate genes for bipolar disorder [
82,
83]. The MAPK gene, DUSP6, and neuropeptides,
NYP and
SST, are discussed further below.
Among the top findings from our pathway analyses were the up-regulation of metallothionein genes across any brain region and specifically in the PFC. This collection of genes was highlighted as significantly differentially expressed in several previous studies, including a weighted gene co-expression network analysis of BP and schizophrenia [
84] and two previous meta-analyses of BP and psychosis using gene expression studies from SMRI [
6,
85]. The two meta-analyses included several studies that we excluded due to quality control measures, and we included one study on a unique set of brain samples that was not included in theirs. In addition, we used an entirely different approach for processing and analyzing the data. The fact that the results for the metallothionein proteins were sustained in multiple analyses lends support to the conclusion that the findings are real. Interestingly, studies with animal models have suggested the involvement of metallothioneins in neurocognitive function [
86,
87], and particularly in protecting the central nervous system against degeneration caused by various types of brain injury [
88,
89].
Also implicated in the pathway analysis were the mitogen-activated protein (MAP) kinase phosphotases. These are members of the dual specificity phosphatase (DUSP) family, which are known to negatively regulate members of the MAP kinase superfamily. MAP kinases have been shown to play a role in neuronal differentiation, neuronal survival, and long term neuroplasticity, and it has been suggested that lithium and valproate may exert therapeutic effects in BP by activating MAPK/ERK signaling cascades [
90]. One of the key genes in the pathway identified by our current analysis was DUSP6, which was found to be significantly down-regulated in BP. DUSP6 is known to bind to and inactivate ERK1 and ERK2 [
91], and previous studies have suggested a genetic association between DUSP6 and both schizophrenia and BP [
74,
75].
Another notable finding from our pathway analyses suggested there is a down-regulation of neuropeptides such as neuromedin U (
NMU), neuropeptide Y (
NPY), and somatostatin (
SST) in BP.
SST, in particular, was reported as significantly down-regulated in the other meta-analysis referenced earlier as well [
84]. It was also implicated in a combined analysis of gene expression studies of the dorsolateral prefrontal cortex in schizophrenia [
92], and in analysis of studies of the subgenual anterior cingulate cortex in major depression [
93]. Neuropeptides are chemical messengers that are widely distributed throughout the peripheral and central nervous system, and they exert diverse effects in serving as hypothalamic releasing factors, neuromodulators, and/or neurotransmitters. There has been a great deal of interest in the role of neuropeptides such as neuropeptide Y and somatostatin in mood and anxiety disorders and as potential therapeutic targets [
94].
It is noteworthy that the metallothioneins were not found to be significantly differentially expressed in the hippocampus. This may reflect differences in dysregulated gene expression patterns across different brain regions in BP, or it may be due to the fact that there were considerably fewer studies of the hippocampus resulting in relatively less power to detect meaningful differences. Clearly, further expression studies in this important brain region are needed.
The effort to synthesize findings from existing genome-wide expression studies of the brain in BP was complicated by several important challenges. First, there may be concerns about combining results across potentially heterogeneous studies. For example, studies of gene expression in the brain have used a variety of array platforms and examined different regions of the brain, which might contribute to the heterogeneity. In order to minimize such concerns, we included only the most recent and most comprehensive studies that all used a comparable array platform, and we obtained the raw data from each of the studies and analyzed this data using a standardized pipeline. In addition, we conducted separate mega-analyses for key regions of the brain.
Second, multiple studies were carried out using overlapping brain samples. Because of the challenges in collecting post-mortem brain tissue, there are limited such samples. Indeed, available samples have essentially come from 4 brain banks, and these have been studied multiple times by different research groups. Unfortunately, data from two of the existing brain banks were not available. We sought to use whatever data was available, and we used an analytic approach that appropriately handled the correlation induced within studies and within samples used across multiple studies.
Third, there may be many factors that confound the relationship between gene expression levels in post-mortem brain samples and disease status. Pre-mortem exposures and treatment histories, especially pharmacologic, may vary between cases and controls and drive differences in gene expression observed in brain samples. Likewise, post-mortem factors such as the agonal state, post-mortem interval between death and sample extraction, or sample pH may further degrade potential expression signals. Many of these factors may or may not be measured, and thus are difficult to correct [
95]. We used an analytic approach that did not require all of the factors to be measured to account for this as best as possible. In particular, we used surrogate variable analysis which has been shown to be a powerful method for removing unwanted measured and unmeasured sources of heterogeneity [
30]. However, it is possible this approach did not completely correct for all sources of heterogeneity, which may have confounded the findings.
Despite the challenges, our analyses provide an up-to-date summary of results from expression array data in BP. These analyses focused on the highest quality non-redundant data available and provides results by brain region so that similarities and differences can be sought that might be relevant to disease status. The results are available for closer inspection on-line at Metamoodics [
http://metamoodics.igm.jhmi.edu/], a bioinformatics resource that we have created to gather results from genomic experiments in mood disorders. Investigators can look up any genes of interest and view the current results in their genomic context and in relation to leading findings from other genomic experiments in bipolar disorder.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
FS, MP and PPZ participated in the design of the study. FS performed the statistical analysis. FS, MP, FSG, JJ, JBP and PPZ conceived of the study, and participated in its design and coordination and helped to draft the manuscript. All authors read, contributed to and approved the final manuscript.