We examined the gene expression profiles of a postmortem brain sample well matched by gender, age, postmortem interval, pH, side of the brain, and race (18
). We identified 2,191 unique genes (FDR ≤ 0.005) differentially expressed in the premotor cortex of BD individuals not exposed to antipsychotics at time of death or during their lifetime compared to exposed brain and controls. We observed that in exposed brains the expression levels of these genes were similar to that of controls (diminished statistical significance). This suggests that antipsychotics may ‘normalize’ (or at least suppress differences in) the expression levels of several genes relevant to the molecular pathophysiology of BD. This broad effect on gene expression of the antipsychotics may be one of several factors that reflect the significant heterogeneity of the illness and includes pharmacogenomics and pharmacodynamics
The normalization effect of antipsychotics is supported by a recent report using induced pluripotent stem cell (iPSC)-derived neurons to model schizophrenia (35
). This study identified signaling pathways that were abnormal in cells derived from individuals with schizophrenia, that could be ameliorated by treatment with a typical antipsychotic medication, loxapine (35
). While the outcome of gene expression modulation through antipsychotic medication is common to the current study and an iPSC-derived model of schizophrenia, there are clearly many confounds from the environment that affect the adult postmortem brain. Medication is one of many tools to alter the environment at defined stages, the results of which may have causal relevance to either the illness or the neuropharmacology of the medication.
Using their method of convergent functional genomics
, Le-Niculescu and colleagues (36
) generated a list of 1,529 genes implicated in BD or depression. In a study of genomic variation for schizophrenia, Lee and colleagues (37
) reported 2,725 genes containing common SNPs (single nucleotide polymorphisms) that captured a significant proportion of the variance in liability to schizophrenia. Other common conditions have many genes implicated: e.g., thousands of genes were found to change expression levels in response to viral infection in peripheral blood mononuclear cells using integrative personal ‘omics’ profiling (38
In comparing our significant genes with the 375 significant genes identified in the meta-analysis of 12 microarray data sets (17
), 185 genes were common to both lists (Fisher’s exact test p ≤ 0.015), and 99% of these common changes were in the same direction (mostly down-regulated) (Supplementary Table 4
). The significant overlap suggests high concordance between this current study and the meta-analysis. It also implies there may be common aspects of pathophysiology in some brain regions, since the current study was performed on tissue from the BA6 region, while the meta-analysis (17
) combined profiles from different brain regions (mainly BA6, 9, 10, 46) from 12 studies including our microarray raw data.
A recent microarray analysis of the thalamic transcriptome performed on material from the SMRI Neuropathology Consortium identified 72 genes with highly significantly altered expression levels in bipolar brains (15
). Of these, we found that 22 genes were common to our study (Fisher’s exact test p = 2.14 × 10−6
) and all common changes were in the same directions (two increased and 20 decreased) (Supplementary Table 5
). This comparison is reasonably sound since both experiments were performed on the U133P2 Genechips and tested similar number of probe sets (around 22,000).
It remains a challenge to identify causative alterations in gene expression using microarray data of postmortem tissues. We observed 86 genes that showed two-fold or greater difference in mean expression levels between BD brains and the combined group, for example, RGS4, BAG3,
showed two-fold or greater changes (see ). RGS4
(regulator of G protein signaling 4) has been associated with schizophrenia (39
) and implicated in BD (40
). Though BAG3
has not been previously implicated in BD, it has been identified with relevant expression changes in response to antipsychotic treatment in animal brain (29
, a gene involved in synaptic function, has been associated with BD in a functional analysis of extant data (41
). Given the assumption that larger fold changes have stronger functional impact, priorities may be given to the genes with higher magnitude changes for further detailed analysis.
Functional GO term enrichment analysis identified significant clusters of genes under several biological process, including protein metabolic process, protein transport, macromolecule localization, protein localization, intracellular protein transport, establishment of protein localization, post-translational protein modification, protein folding, molecular function, vesicle-mediated transport, synaptic transmission, and intracellular transport; or GO terms of molecular function involve protein binding, nucleotide/ribonucleotide including ATP binding, ligase and catalytic activities. These data and neuropathological studies of BD brain suggest that as in neurodegenerative conditions, protein processing and transport may be significantly altered.
The enriched genes under specific GO terms often include multiple members of gene families. For example, five members of the disintegrin and metalloprotease (ADAM) gene family, ADAM10, ADAM11, ADAM15, ADAM17,
, were in enriched under the GO term protein metabolic process
. The ADAM genes encode membrane-anchored proteins that have been shown to play important roles in the development of the nervous system, including regulation of proliferation, migration, differentiation and survival of various cells, as well as axonal growth and myelination (42
). Enriched under the GO term protein transport
were 13 members of adaptor-related protein complex (AP) gene families: AP1M1, AP1S1, AP1S2, AP2A1, AP2A2, AP2B1, AP2M1, AP2S1, AP3B2, AP3M2, AP4B1, AP4M1,
. AP complexes, AP-1, AP-2, AP-3, and AP-4, play important roles in synaptic vesicle formation and endocytosis (33
). The expression levels of these genes are significantly lower in non-exposed brains compared to controls, suggesting a hypothesis of synapse impairment in BD. How these genes may be involved in BD is currently unknown, and awaits analysis in relevant cellular models.
A number of ‘high profile’ genes previously implicated in BD also stand out in our study. For example, the mRNA levels of GSK3B, FKBP5, ANK3
genes were altered in the non-antipsychotic medication brains. Glycogen synthase kinase 3 beta (GSK3β) is a known target of lithium, and has been hypothesized to be the molecular basis of lithium treatment of BD (43
). FK506 binding protein 51, the protein product of the FKBP5
gene, forms part of a complex with the glucocorticoid receptor and can modulate cortisol-binding affinity (44
). Variations in FKBP5 have been reported to be associated with BD (45
gene product, Ankyrin-G, is present at the axonal initial segment and at nodes of Ranvier. Ankyrin-G plays key roles in node formation and function in the central and peripheral nervous systems. Genome wide association analysis identified single nucleotide polymorphisms at the ANK3
locus associated with BD (46
), and cis-acting variations in the ANK3
locus were shown to affect its expression (47
). In addition to ANK3
, several genes encoding key components of node of Ranvier or paranodal region, such as NRCAM
, and ANK2
, are among the list of genes identified in this study (Supplementary Table 2
). These observations suggest node impairment may be a neural mechanism of BD.
The limitations of the current analysis relate primarily to the relatively small number of individuals in the study of postmortem samples. It would be useful to replicate these findings in additional samples and studies; however, there is considerable overlap in the genes identified in the current study with those from previous studies (15
). There are many confounding factors that essentially limit the utility of postmortem gene expression analyses which include sex-dependent expression differences, death and agonal factors including the postmortem interval and tissue pH. The postmortem brain most likely reflects the end-stage organ disease state and may not reflect the initial etiological mechanisms of the disorder. There are several options for study design and include: (i) the ascertainment and procurement of additional postmortem brains for study and (ii) the establishment of cellular models that are derived from individuals with BD from whom there are considerable phenotypic and longitudinal data that may be factored into the analyses. Clearly, the second option is the ideal option. Individuals with BD that have common phenotypic features may be ascertained and samples, the cell lines grown under controlled conditions of exposure to environmental perturbations (medications and other biological variations) and measured consistently. Cellular models range from iPSC-derived neurons to cells derived from B lymphocytes transformed with EBV. The consistent feature being cellular tissue derived from an individual with the disorder. The limitations of the cellular models are the lack of ability to study the complex circuitry of the human brain, however complex biochemical pathways could be studied at a sophisticated biological level to determine interactive correlates between pathways.
In summary, we identified a large number of genes with altered expression in BD brains not exposed to antipsychotics. These changes are normalized to the levels of healthy controls and BD brains treated with antipsychotics. Functional GO terms analysis suggests a theme that these gene products are involved in neuronal communication that may be impaired in BD brains. Although our current results are concordant with previous findings, a caveat exists in interpreting of data generated from postmortem brains that are further confounded by agonal status, tissue pH, postmortem interval, medication, etc. Neuropsychiatric and neurological diseases are increasingly thought to be developmental in nature, and the fact that disease-specific changes are unlikely be distinguished from confounders in postmortem brains, combine to suggest that disease-specific live neurons are critically required to study the molecular basis of BD and to assemble a palette of disease-causing genes. This goal will be achieved in the systematic study of live neurons derived from iPSCs (48
) or from differentiated easily accessible tissue of non-neural origin (35