Meta-analysis of 12 genomic studies in bipolar disorder

Journal of Molecular Neuroscience Copyright © 2007 Humana Press Inc. All rights of any nature whatsoever are reserved. ISSN0895-8696/07/31:221–244/$30.00 JMN (Online)ISSN 1558-6804 DOI 10.1385/JMN/31:03:221 ORIGINAL ARTICLE Meta-Analysis of 12 Genomic Studies in Bipolar Disorder Michael Elashoff,1 Brandon W. Higgs,*,1 Robert H. Yolken,2 Michael B. Knable,3 Serge Weis,4 Maree J. Webster,4 Beata M. Barci,3 and E. Fuller Torrey3 1Elashoff Consulting, Germantown, MD 20876; 2Stanley Laboratory of Developmental Neurovirology, Johns Hopkins University, School of Medicine, Baltimore, MD 21287; 3The Stanley Medical Research Institute, Bethesda, MD 20814; and 4Stanley Laboratory of Brain Research, Uniformed Services University of the Health Sciences, Department of Psychiatry, Bethesda, MD 20814 Received July 7, 2006; Accepted August 6, 2006 Abstract Multiple genome-wide expression studies of bipolar disorder have been published. However, a unified picture of the genomic basis for the disease has not yet emerged. Genes identified in one study often fail to be identified in other studies, prompting the question of whether microarray studies in the brain are inherently unreliable. To answer this question, we performed a meta-analysis of 12 microarray studies of bipolar disorder. These studies included >500 individual array samples, on a range of microarray platforms and brain regions. Although we confirmed that individual studies showed some differences in results, clear and striking regulation patterns emerged across the studies. These patterns were found at the individual gene level, at the functional level, and at the broader pathway level. The patterns were generally found to be reproducible across platform and region, and were highly statistically significant. We show that the seeming discordance between the studies was primarily a result of the following factors, which are also typical for other brain array studies: (1) Sample sizes were, in retrospect, too small; (2) criteria were at once too restrictive (generally focusing on fold changes >1.5) and too broad (generally using p < 0.05 or p < 0.01 as criteria for significance); and (3) statistical adjustments were not consistently applied for confounders. In addition to these general conclusions, we also summarize the primary biological findings of the meta-analysis, focusing on areas that confirm previous research and also on novel findings. DOI 10.1385/JMN/31:03:221 Index Entries: Gene expression; bipolar disorder; meta-analysis; confounders; energy production; metallothionein. Introduction Despite multiple gene expression and linkage studies of bipolar disorder, a clear understanding of the genomic basis of the disease is still elusive (Blair et al., 2002; Ogden et al., 2004). Although genes or pathways have been identified in specific studies, the findings are not consistently observed from study to study. The most common finding across published studies is an association of oligodendrocyte/myelinrelated genes down-regulated in both bipolar disorder and schizophrenia (Ogden et al., 2004; Konradi, 2005). Researchers have also identified mitochondrial/energy processing dysfunction in bipolar disorder (Konradi et al., 2004; Munakata et al., 2005), although not consistently with other work (Altar et al., 2005). *Author to whom all correspondence and reprint requests should be addressed. E-mail: [email protected] Journal of Molecular Neuroscience 221 Volume 31, 2007 222 Biological findings that have been implicated in bipolar disorder in at least one study include neurotransmitters (Tkachev et al., 2003; Kapczinski et al., 2004), protein turnover (Konradi, 2005), endogenous retroviral sequences (Kan et al., 2004), apoptosis (Benes et al., 2006), and stress response (Webster et al., 2002; Iwamoto et al., 2004). Furthermore, at least 40 additional genes have been reported (see below, Table 6), adding to the inconsistency in results at the pathway and functional level across studies. Several explanations have been suggested for the lack of consistent findings across bipolar studies. First is the complex and heterogeneous nature of the disease (Baron, 2002; Kelsoe and Niculescu, 2002). Bipolar disorder is thought to be associated with multiple genetic, genomic, post-translational, and environmental factors. Furthermore, patients might have varying disease severity, with some having psychotic features, as well as exposure to a variety of medications and dosage levels to control their illness. Second is the microarray technology itself, with multiple platforms of varying designs, sensitivity, and versions. One study (Jurata et al., 2004) examined the consistency of regulation between Affymetrix (Affy), Agilent, and qPCR results on a common set of brain samples and found that fewer than one in four significant results on one platform were seen on another platform. Third are the potential confounding variables inherent in the use of postmortem brain samples (Iwamoto et al., 2005), such as brain pH, postmortem interval (PMI), gender, age, hemispheric side, and agonal state. In most cases, these confounders are addressed by matching for the factors in the study groups. In other cases, some of these variables can be adjusted within statistical models. Finally, there are the implicit challenges in analyzing the data, with tens of thousands of genes but relatively few samples. Microarray studies in bipolar disorder typically have between 10 and 35 subjects per group (bipolar and control) for a total of 20–70 samples. The criteria for declaring significance varies from study to study, though common thresholds are p < 0.01/p < 0.05 and fold change (FC) FC > 1.3/FC > 1.5. To address these issues, we performed a metaanalysis of 12 microarray studies of bipolar disorder, using the same two brain collections from the Stanley Medical Research Institute (SMRI) brain bank. These studies included >500 individual array samples, across a range of microarray platforms (Affy, Agilent, Codelink, and cDNA) and brain regions (dorsolateral Journal of Molecular Neuroscience Elashoff et al. prefrontal cortex and cerebellum). Our goals were to further understand the genomic basis of bipolar disorder and, perhaps just as important, to understand how the disease should best be studied in the future. Materials and Methods The SMRI has two brain collections that have been made available to researchers. To use the samples for research purposes, researchers must agree to return to SMRI the gene expression data that result from the use of the brain samples. This meta-analysis includes all genome-wide expression studies that were completed and provided to SMRI as of February 2005. Some of the studies included in this analysis have been published previously by the respective investigators, whereas others have not yet been submitted and/or published. For this reason, the studies have been coded as studies 1–12 for the purposes of this analysis. This is in keeping with the goal of the metaanalysis, to focus on the overall results and findings of the larger investigation, rather than on the specific results from any one particular study. Samples are derived from SMRI’s two brain collections. The first, termed the Neuropathology Consortium collection, has 60 individual subjects, with multiple brain regions per subject. The details of the sample collection procedures have been described previously (Torrey et al., 2000). Notable exclusion criteria included: age >65 yr, poor quality mRNA, and significant structural brain pathology on postmortem examination. These samples were matched for age, gender, race, pH, PMI, side of brain, and mRNA quality. For studies using this collection, tissue samples were provided to investigators, who then performed the RNA extraction. The second brain collection, termed the Array collection, consists of 105 individual subjects, with multiple brain regions per subject. Exclusion criteria were similar to those for Neuropathology Consortium. For this collection, only dorsolateral prefrontal cortex (Brodmann area 46) was used for the microarrays. In contrast to the Consortium collection, SMRI performed the RNA extraction for the Array collection. Tissue was homogenized in Trizol, and nucleic acid was separated with chloroform at high-speed centrifugation; RNA was then precipitated with isopropyl alcohol and washed with 70% alcohol. Pellets of RNA were resuspended in DEPC water. The quality of the RNA was assessed using the Agilent bioanalyzer. RNA processing and microarray data generation were performed by the individual investigators at Volume 31, 2007 Meta-Analysis in Bipolar Disorder 223 Table 1 Summary of Subject Characteristics in Meta-Analysis Studies No. of Subject Samples Age Gender Race pH PMI Smoking at TOD Heavy drug use Heavy alcohol use Suicide Controls Bipolar 331 45.3 ± 8.8 70% Male 98% White 6.5 ± 0.3 27.7 ± 12.2 24% 0% 4% 0% 284 44.4 ± 10.9 52% Male 94% White 6.4 ± 0.3 36.3 ± 17.7 46% 28% 36% 46% their own facilities. RNA processing protocols were generally those recommended by the respective array manufacturer. Because the studies were conducted at different times, using different platforms and different laboratories, no attempt was made to standardize RNA processing. Table 1 summarizes the clinical characteristics of the samples across the 12 studies. As would be expected, the bipolar cohort had a higher incidence of smoking, drug use, alcohol use, and suicide, as well as a somewhat longer PMI. The studies included in the meta-analysis are summarized in Table 2. It is worth reiterating that studies based on a common brain collection will have subjects in common; thus, the studies in this meta-analysis are not completely independent. Studies were reanalyzed using raw data files from the individual studies. For Affy studies, we used the probe-level raw data files (.cel files) generated from Affy Microarray Suite 5.0 (MAS 5). For cDNA studies, we used the .gal and .gpx files from GenePix version 3. For Agilent and Codelink studies, text files generated from the manufacturers’ software were used. All analysis was performed in the statistical software R, with libraries used from Bioconductor. Expression calculation and normalization methods are described in Supplemental Data (below). NCBI’s Database for Annotation,Visualization, and Integrated Discovery (DAVID) (Dennis et al., 2003) was used as the source for gene annotation information. The primary fields extracted from DAVID include LocusLink, gene symbol, and gene summary. Additional annotations include gene product mappings to the Kyoto Encyclopedia of Genes and Genomes and Gene Ontology Consortium (GO) for pathway and GO terms/classes, respectively. For Journal of Molecular Neuroscience Affy arrays, queries were based on the Affy probe ID (AFFYID). For other arrays, GenBank accession (GenBank) was used. After the bioinformatic merging of the annotation information across the 12 studies, 19,502 unique genes were identified. For each study, quality control (QC) analysis was performed to determine if any samples should be excluded for reasons of poor quality data. The Supplemental Data describe the QC procedures. Briefly, based on the QC analysis, 19 of 331 control samples (5.7%) and 17 of 284 bipolar samples (6.0%) were excluded from disease comparison. Within each study, each demographic factor (see Table 3) was assessed on a gene-by-gene basis using regression models. We identified which genes were significantly correlated with which demographic factors, where significance was defined as p < 0.01, FC > 1.3. For comparison of effect sizes, all demographics were analyzed using two levels. Continuous variables and ordered categorical variables were cut at values as close as possible to the median (e.g., PMI > 30 vs PMI < 30). Demographic factors were assessed using both controls and bipolar subjects, whereas bipolar-specific variables were analyzed within the disease group to avoid confounding the demographic effect and the disease effect. The following list shows the demographic analyses that were performed: • For all subjects: Age, Sex, PMI, Brain pH, Brain Side, Smoking at Time of Death (TOD), and Sudden Death; • For bipolar disorder: Disease Severity, Heavy Alcohol Use, Heavy Drug Use, Psychotic Features, Suicide Status, Antipsychotic Use, Antidepressant Use, and Mood Stabilizer Use. The disease analysis was performed using a regression analysis on a gene-by-gene basis, adjusting for the demographic terms that were significant for that given gene. The regression analyses yielded an adjusted fold change, S.E., and p value for each gene within each study (Table 3). To compute p values from the meta-analysis FCs and S.E.s, the t-distribution was used. However, as mentioned above, the studies had some subjects in common and thus were not fully independent. The potential impact of the shared samples is that the naïve degrees of freedom, based on the total number of studies minus one, would actually be an overestimate of the actual degrees of freedom computed using the actual correlation between studies. A permutation study was conducted to estimate the actual degrees of freedom for the meta-analysis t-tests. This analysis found that using naïve degrees of freedom Volume 31, 2007 224 Elashoff et al. Table 2 Summary of Study Characteristics in Meta-Analysis Studies Study ID 1 2 3 4 5 6 7 8 9 10 11 12 a Samples Controls Bipolar Collectiona Region Array type Probe sets 66 40 67 28 56 52 67 58 21 71 70 23 34 29 35 14 26 28 34 40 10 36 35 12 32 11 32 14 30 24 33 18 11 35 35 11 A C A C A C A C C A A C Frontal BA46 Frontal BA46/10 Frontal BA46 Frontal BA6 Frontal BA46 Cerebellum Frontal BA46 Frontal BA46/10 Cerebellum Frontal BA46 Frontal BA46 Frontal BA8/9 Affy hgu133A Affy hgu133A Affy hgu133A Affy hgu133 2.0+ Affy hgu133 2.0+ Affy hgu95Av2 Affy hgu133A Agilent Affy hgu95av2 Codelink human 20K cDNA Affy Hgu95Av2 22283 22283 22283 54681 54681 12453 22283 12373 12453 19907 14369 12453 A = Array collection; C = Neuropathology Consortium collection. Table 3 Percentage of Genes Regulated by Clinical Factors Within Individual Studies Regulated in Individual Studies Factor Bipolar disorder Smoking Gender PMI Brain side Brain pH Age Heavy alcohol use Heavy drug use Suicide Agonal state (sudden death) Psychotic feature Disease severity Antidepressant use Antipsychotic use Mood stabilizer use Lithium use Valproate use Median Quartiles 0.20% 0.07% 0.15% 0.11% 0.04% 1.36% 0.04% 0.23% 0.14% 0.07% 0.21% 0.09% 0.07% 0.04% 0.28% 0.09% 0.22% 0.07% 0.10%–1.11% 0.01%–0.51% 0.10%–0.25% 0.01%–0.70% 0.01%–0.11% 0.97%–3.43% 0.01%–0.43% 0.02%–0.36% 0.05%–0.27% 0.05%–0.23% 0.02%–0.97% 0.05%–0.24% 0.05%–0.16% 0.02%–0.22% 0.09%–0.96% 0.03%–0.36% 0.03%–0.59% 0.03%–0.16% The meta-analysis consensus FC was calculated for each gene based on a weighted combination of the individual FCs and S.E.s for the probe sets that mapped to each gene across the platforms/studies. The weights were equal to 1/S.E.i, where S.E.i is the S.E. of the ith probe set for the gene across all of the studies. was an overestimate, but only by a small amount. Thus, a small degrees of freedom correction factor could be applied to the t-tests to correct for the study interdependencies. For each study, pathway/GO associations to disease were tested using Fisher’s exact test. This test was based on the number of significantly regulated genes Journal of Molecular Neuroscience within each pathway/GO term and within each study. Pathway/GO terms that had a value of p < 0.01 within a study were declared significant for that study. For the meta-analysis, pathway/GO association to disease was also tested using Fisher’s exact test, based on the number of significantly regulated genes from the cross-study analysis within each pathway/GO term. Volume 31, 2007 Meta-Analysis in Bipolar Disorder 225 Table 4 Number of Studies in Which Genes Met Significance Criteria Within Individual Studies No. of significant studies No. of genes (p < 0.01, FC > 1.3) No. of genes (p < 0.01) No. of genes (p < 0.05) Pathway/GO terms 0 0 0 1 4 105 1083 7188 0 0 0 1 18 237 1633 6492 6 21 82 277 839 1928 2951 2277 0 1 9 14 37 96 296 905 7 6 5 4 3 2 1 0 The table is restricted to genes appearing in at least 10 studies. The rows for studies 8–12 were not included for visual ease, as each cell contained zero values across all fields. Results Individual Study Analysis Although the focus of this investigation is on the meta-analysis results, we will also briefly summarize the results of the individual studies. To compare across individual studies, we used common criteria for significance that were representative of those used in published brain array research: p < 0.01, FC > 1.3, gene percent present, >33%. We revisit this definition of significance in the discussion below. Based on the definition of significance, the percentage of genes that met the definition within each study was tabulated. The median percentage of significant genes was 0.2% (CI 0.1%–1.1%). Individual Study Demographic Analysis The percentage of genes that met the definition for the analysis of each demographic factor was also assessed (see Table 3). Brain pH was the most influential factor for gene expression (median percentage significant regulation, 1.4%). Other factors such as alcohol use, drug use, gender, agonal state, and medication usage affected comparable numbers of genes as did the disease itself. Adjusting for such confounding effects is very important, as these effects can induce bias, increase variability, or both. Simply matching on or balancing these factors can reduce the likelihood of bias, but in practice one cannot always match on multiple factors at once without substantially restricting the available sample size. Furthermore, matching or balancing has no impact on the increased variability that the factors contribute. Not adjusting for these factors reduces the overall power of the analysis to detect disease differences and leads to an increase in the false discovery rate. Even for Journal of Molecular Neuroscience factors such as suicide status or age, which affected relatively small numbers of genes, the magnitude of their regulation effect might be large for those genes and thus should be adjusted. Overfitting, although a theoretical concern with 17 possible confounding variables assessed, was not an issue in practice, owing to the variable selection process we employed (described in Materials and Methods) that resulted in gene-specific disease models generally having no more than one or two confounders included (and those confounders would differ from gene to gene). The studies are shown to be sensitive to the effects of confounding variables, whose effects were comparable in magnitude to the disease effect. Concordance of Results Looking for correspondence of regulation across studies, we tabulated for each gene the number of studies where the gene met the criteria for significance (p < 0.01, FC > 1.3). Because not every gene was common to every study, we restricted the analysis to genes that appeared in 10 or more studies (8381 genes). Of these 8381 genes, no gene met the criteria for significance in 5 or more studies, and only 4 genes met the criteria in 3 or more of the 12 studies (see Table 4). Looking at the results of Table 4, one can see that of the 8381 genes, 1193 genes were found to be regulated in at least one study. However, only 110 of those 1193 were regulated in more than one study. Put another way, the likelihood of a significantly regulated gene having a repeat finding was only 110/ 1193 = 9%. We examined the impact of p value and FC rules on this apparent lack of agreement by using different criteria. With p < 0.01 significance level but no FC filter, the number of genes meeting Volume 31, 2007 226 Elashoff et al. Table 5 Meta-Analysis FC and p-Value for All Genes Bipolar combined analysis p value FC 1–1.1 1.1–1.2 1.2–1.3 1.3–1.4 1.4–1.5 >1.5 Total Chance Cum. total Cum. chance Cum. FDR <0.0001 25 81 11 0 0 0 116 2 0.0001–0.0005 51 97 13 0 0 0 161 7 0.0005–0.001 43 53 5 0 0 0 98 9 0.001–0.005 213 264 19 0 0 0 496 72 0.005–0.01 218 172 19 1 0 0 410 90 0.01–0.05 1316 629 53 3 0 0 2001 725 >0.05 15147 999 58 8 3 1 16216 18527 116 2 1.7% 277 9 3.1% 375 18 4.6% 871 90 9.4% 1281 180 12.3% 3282 906 21.6% 19502 19502 — the significance criteria in at least two studies increased (from 109 to 256); but the number of single-study hits increased as well (from 1083 to 1633), and again no genes met the criteria in more than four studies. At p < 0.05 significance level, 109 (1.3%) genes were regulated in half or more studies, but 74% of all genes would have been declared significant in at least one study. We did not find that altering the p value or FC criteria in individual studies could meaningfully improve the agreement for significance at the gene level. In summary, by examining only top hits from individual studies, it appears that the studies are not yielding consistent, reproducible associations. This is particularly noteworthy given the overlap in subjects across the studies. Individual Study Pathway/GO Analysis A total of 3878 pathway and GO terms were assessed within each study; we focused on the terms that had contained mappings to at least 4 significant genes (1358 terms). These studies had a median of 45.5 significant (p < 0.01) hits each. Although the cross-study correspondence between pathway/GO terms was much better than for individual genes, no terms showed up in all of the studies and only one term showed up in at least half the studies. Of the 453 terms that appeared significant in at least one study (term count: 296 + 96 + 37 + 14 + 9 + 1), the majority appeared in only one study. Just as for the gene level analysis, based on examining top hits from individual studies, it appears that the studies are not yielding consistent, reproducible associations. Journal of Molecular Neuroscience Meta-Analysis Results Table 5 shows the results of the meta-analysis at the gene level. The false discovery rate analysis indicated that a p value cutoff of 0.001 would maximize the number of genes while keeping the false discovery rate (FDR) <5%. At p < 0.001, a total of 375 genes were identified as significant (see Supplemental Data). The 375 genes represent 2.0% of the total number of unique genes. This contrasts with the individual studies, which found a median of 0.2% of genes as significant, using a literature criteria of p < 0.01, FC > 1.3. What happens when the p < 0.001 criteria is applied to the individual studies? In that case, a median of only seven genes per study is identified as significant. This is a direct result of the sample sizes of the individual experiments. The studies are simply not powered to detect significance at the p < 0.001 level. Table 5 is noteworthy not only for the large number of highly significant genes but also for the low FC values of those genes. Of the 375 significant (p < 0.001) genes for bipolar disorder, none had a FC > 1.3. If a gene was reported in an individual study with such a small effect size, one might understandably be skeptical of its reliability. But in this analysis, to be significant a regulation effect would need to be present across most of the range of platforms and brain regions. Why, then, did the results of the individual studies appear to miss these genes? The answer can be illustrated using an example. The gene reelin (RELN) was significant at p < 0.01 in only 2 of the 12 studies. But across the 12 studies, Volume 31, 2007 Meta-Analysis in Bipolar Disorder the direction of regulation was down for 11 of the 12 studies (12 of 13 probe sets). The combined FC was –1.22 (p = 0.001, CI = –1.35 to –1.10), and in all cases the individual study confidence intervals contained a FC of –1.22. Thus, the results across the studies are consistent with a true down-regulation of RELN in bipolar disorder of approx –1.2 fold. However, the individual subject variability is such that even a study of 70 subjects is not enough to accurately determine the significance or estimate the magnitude of the association. Previous studies not included in this meta-analysis have found RELN to be associated with schizophrenia and bipolar disorder at both the mRNA level and the protein level (Costa et al., 2001; Grayson et al., 2005). Examination of the other significantly regulated genes in bipolar disorder reveals patterns similar to RELN, where subsignificant but consistent upor down-regulation is seen in multiple studies. In retrospect, the studies were not powered to detect these differences as being significant within an individual experiment. But this does not mean that one can look at a finding in an individual study and presume that just because the other studies missed it, they were underpowered. For every example like RELN, there are numerous other examples like caspase 8. Caspase 8 was significant at p < 0.01 in one study (FC = –1.45), with the remainder of the studies showing no significant regulation. With regard to the direction of the regulation, up-regulation is 13/26 probes and down-regulation is 13/26 probes (see Fig. 1). The combined FC was 1.01 (p = 0.938, CI = –1.03 to 1.04). It seems likely that the significant result in the individual study was a false-positive finding. Interestingly, there is one published study (Bezchlibnyk et al., 2001) on the relationship between caspase 8 and bipolar disorder, where a significant relationship was claimed (frontal cortex, n = 10/group). The examples of RELN and caspase 8 illustrate the two consequences of studies with a relatively small number of brains: false negatives and false positives. Although one can adjust the analysis parameters of an individual study to favor one over the other, the studies were not large enough to simultaneously yield small rates for both types of errors. At a stringent p < 0.001, individual studies found a median of 7 significant genes, meaning that with 375 metaanalysis genes and p < 0.001 as a standard, the median false-negative rate was >95% at the gene level. In contrast, at a nonstringent p < 0.05, about 1250 genes per study would be called significant. The false-positive Journal of Molecular Neuroscience 227 rate in that case would be at least 60%. As noted previously, the discordance between the studies’ significant genes is primarily an artifact of their sample size. The small sample sizes have meant that the studies have not been powered to detect the small FCs associated with bipolar disorder. Unfortunately, the individual studies cannot be reanalyzed on their own to take advantage of this information, as using less strict analysis filters would lead to a greatly increased false discovery rate. Table 6 lists genes reported previously to be associated with bipolar disorder, and the results for those genes in our analysis. Overall, we find confirmatory evidence for approximately one-third of the genes. This relatively low number is to be expected on the basis of the limitations of individual bipolar studies discussed previously. Meta-Analysis of Pathway/GO As noted previously, the pathway/GO analysis of the individual studies was more reproducible than the gene level analysis. We will show in this section that previously reported findings at the pathway/GO level were more often confirmed in our analysis than the gene level analysis, although there were some important exceptions. For pathway/GO terms, we examined the 1358 pathway/GO terms that had mappings to at least four genes. Atotal of 96 terms were significant at p < 0.005. This level of significance was chosen to yield an approximate FDR of 5%. These findings are in contrast to the individual studies, where at p < 0.005 a median of 18.5 terms/study were significant. The metaanalysis of the pathway/GO terms clearly benefits from the increased power of the larger sample size. Looking in more detail at the top pathway/GO terms, we find two main reasons, both associated with sample size, for the apparent inconsistency of the individual study results. First, just as there were nearly significant but consistently regulated genes, the same phenomenon applies to pathways. For example, one of the top scoring pathways was ATP synthesis. This pathway yielded p < 0.01 in 4 of the 12 studies. In an additional 3 studies, the p value was close to but missed that cutoff (p = 0.01, p = 0.07, p = 0.09). A second reason was that the genes themselves were more powerfully and accurately detected as significant in the overall analysis, thus driving the pathway associations. The metallothionein genes are a good example of this. The various metallothionein isoforms were only occasionally significant in the individual studies (see Figure 2), yet Volume 31, 2007 228 Fig. 1. FC values with 95% confidence interval for Reelin and caspase 8 across all mapping probes within each study. Consensus FC values are plotted as the bottom three points for each graph. Meta-Analysis in Bipolar Disorder 229 Table 6 Genes Reported To Be Associated with Bipolar Disorder and Their Regulation Profiles in the Meta-Analysis Gene Reference Meta-Analysis ↓ = down, p < 0.01; ↓↓ = down, p < 0.001 ↑ = up, p < 0.01; ↑↑ = up, p < 0.001 – = no significant regulation Genetic markers DARPP32 PENK TAC1 XBP1 GRK3 DRD4 TPH1 BDNF COMT HSPA5 DISC1 DYSB AKT1 GRIN2A HTR4 IMPA2 GABRA1 G72 LARS2 APO-L AMPA2 HINT1 UBE2N SCA7 GTF2H2 LIM HSPF1 TPH2 Serotonin Spinophyillin PDYN GRIN1 Complexin I,II GFAP NPY PRKAR2A TBR1 NCS1 RELN CASP8 ERBB2 TGFB1 DNMT1 NFKB SLC6A4 Ogden et al., 2004 Ogden et al., 2004 Ogden et al., 2004 Barrett et al., 2003 Barrett et al., 2003 Aguirre-Samudio and Nicolini, 2005 De Luca et al., 2005 Kato et al., 2005 Maier et al., 2005 Kakiuchi et al., 2003 Thomson et al., 2005 Raybould et al., 2005 Kato et al., 2005 Kato et al., 2005 Kato et al., 2005 Kato et al., 2005 Kato et al., 2005 Kato et al., 2005 Genomic markers Munakata et al., 2005 Konradi, 2005 Konradi, 2005 Konradi, 2005 Konradi, 2005 Jurata et al., 2004 Jurata et al., 2004 Iwamoto et al., 2004 Iwamoto et al., 2004 De Luca et al., 2005 Kapczinski et al., 2004 Law et al., 2004 Hurd, 2002 Law and Deakin, 2001 Eastwood and Harrison, 2000 Fatemi et al., 2004 Kuromitsu et al., 2001 Molnar et al., 2003 Molnar et al., 2003 Koh et al., 2003 Grayson et al., 2005 Bezchlibnyk et al., 2001 Bezchlibnyk et al., 2001 Bezchlibnyk et al., 2001 Veldic et al., 2005 Sun et al., 2001 Sun et al., 2001 – ↓ – – ↓ – – ↓ – ↓ – – – – – – – – ↓ ↓, ↓↓, – – ↓↓ ↓↓ – – – – – – – – – – – – – – – ↓ – – – – – – Where multiple codes are given, multiple forms of the gene showed different results. Journal of Molecular Neuroscience Volume 31, 2007 230 the regulation profiles were consistent from study to study, yielding a highly significant result in the metaanalysis. This, in turn, drove the metallothionein GO terms (e.g., metal ion binding) to be significant in the meta-analysis but not in the individual studies. The top pathway/GO terms can be further grouped into conceptual categories (see Table 7). 1. Energy Metabolism. The strongest finding to emerge from the meta-analysis was a consistent down-regulation of the energy metabolism system in patients with bipolar disorder (see Fig. 2). Association of energy metabolism dysfunction with bipolar disorder has been reported previously (Konradi et al., 2004; Munakata et al., 2005), although some studies have not confirmed this finding (Altar et al., 2005). It has also been suggested that this finding is due to medication usage in these patients (Iwamoto et al., 2005). We addressed this issue by performing the same pathway analysis on medication variables as was done for bipolar disorder, looking at the effects of antipsychotic use, antidepressant use, and mood stabilizer use. We found that the pathways and gene ontology terms regulated by medication use had little overlap with those dysregulated by bipolar disorder. For example, mood-stabilizing agents were associated primarily with dysregulation in signal transduction and transferase activity. The most significant of these pathways was protein amino acid phosphorylation, containing such genes as glycogen synthase kinase-3β (GSK-3β), protein kinase C (PKC), and protein kinase A (PKA), which have each been reported in multiple studies to be associated with mood stabilizer use (Manji et al., 1999; Lenox and Hahn, 2000). 2. Protein Turnover. Categories corresponding to protein turnover, including proteasome and ubiquitination, showed strong down-regulation across numerous genes. Association of protein turnover with bipolar disorder has been reported previously (Konradi et al., 2004). 3. Major Histocompatibiltiy Complex (MHC) Antigen Response. This represents a novel finding. The category was driven by multiple human leukocyte antigen (HLA) genes showing significant down-regulation. This might reflect the potential contributory cause of infectious agents to bipolar disorder or other functions, such as stress response, associated with this family of genes. 4. RNA Processing. Categories corresponding to RNA processing, binding, splicing, etc., were commonly down-regulated. 5. Intracellular Transport Activity. Down-regulation of this system was reported recently to be associated with bipolar disorder (Iwamoto et al., 2004). 6. Stress Response. Stress response genes have been reported previously to be down-regulated in bipolar disorder (Webster et al., 2002; Iwamoto et al., 2004). 7. Metallothionein. This was another novel finding and was driven by the highly significant up-regulation Journal of Molecular Neuroscience Elashoff et al. of multiple isoforms of metallothionein (see Fig. 2). Metallothionein is involved with metal ion binding. Metallothioneins might also function as stress response genes in the brain (Kim et al., 2004; Natale et al., 2004), although little is known about the exact function of metallothionein in the human brain; therefore, the implications of this finding are unknown. A natural question is whether these terms are significant because they are disease associated or because of confounders such as brain pH, medication use, etc. Recall that the gene level analysis was adjusted for these terms on a gene-by-gene basis when they were found to have a significant effect on gene expression. As a result, the gene level analysis should not be strongly confounded by the variables. The pathway/ GO analysis is based on the gene level analysis, by counting the number of significant genes in each term relative to the total number of genes for the term. Thus, the potential effects of confounding on these results should be minimized. We can also establish the lack of confounding effect more directly by performing a pathway/GO analysis on the medication and confounding variables themselves. The top pathway and GO categories described above in the bipolar analysis were generally not significant in the medication analyses (see Supplemental Data). Thus, the findings appear to be related to the disease itself and not attributable to confounding variables. It should be noted, however, that some subjects had incomplete medication information, and it was not possible to verify all of the medication information available for each subject. There are some noteworthy pathway/GO terms that have been reported in the literature but do not appear in the above list: 1. Oligodendrocyte/myelin-related genes (Tkachev et al., 2003; Ogden et al., 2004; Konradi, 2005). The only gene in this category that was significant at p < 0.001 was MOG. Two additional genes were significant at p < 0.01 (MBP, OMG). However, the remainder of the genes in this category were not found to be significantly regulated. Thus, although we can replicate some of the association reported previously between bipolar disorder and oligodendrocyte/myelin-related genes in these brain regions, the results were modest relative to the top sets of genes. 2. Dopamine related genes (Koh et al., 2003; Kapczinski et al., 2004; Aguirre-Samudio and Nicolini, 2005). No differential expression was observed in any of the dopamine-related genes in the meta-analysis. 3. Serotonin-related genes (Sun et al., 2001; Kapczinski et al., 2004). No differential expression was observed in any of the serotonin-related genes in the meta-analysis. Volume 31, 2007 Meta-Analysis in Bipolar Disorder 231 Table 7 Top-Regulated Pathway/GO Terms in the Meta-Analysis Functional grouping Energy metabolism Protein turnover RNA Processing Transport Stress response MHC antigen response Metallothionein Oligodendrocyte/myelin Dopamine related Serotonin related GABA related Synapse related Representative terms Oxidative phosphorylation Proteasome Ubiquitin-conjugating enzyme activity mRNA splicing Intracellular protein transport Heat shock protein activity MHC class-II receptor activity Metallothionein Notable literature gene sets Oligodendrocyte/myelin Dopamine receptor activity Serotonin receptor activity GABA-A receptor activity Synapse 4. GABA-related genes (Woo et al., 2004). GABA-related genes were rarely observed to be regulated (no genes with p < 0.001; one gene with p < 0.01). 5. Synapse-related genes (Eastwood and Harrison, 2000; Ogden et al., 2004). This category just missed the p < 0.005 cutoff for top pathway/GO terms. Several genes were highly significant and others were moderately significant. Notable genes included APBB1, APBB2, SYN2, STY5, STY11, VAMP1, VAMP2, and VAMP3. Conclusions We have established that there is highly significant and reproducible gene regulation in bipolar disorder. Furthermore, this finding is associated with specific pathways and functional categories. In some cases, these genes, pathways, and categories confirm previous results regarding bipolar disorder; in other cases, they refute prior reports. We have tried further to explain why prior studies have failed to reach a consensus; the primary reason for the lack of consistency can be attributed to the fact that individual gene dysregulation in bipolar disorder is small in magnitude, on the order of 10%–20%. This finding has several important implications. Implications of Small FCs Beyond the already addressed issues of increasing sample sizes (can greatly increase FDR) and the importance of correcting disease effects for confounding variables, the independence of each study is also a contributing factor. In Results, we noted that the studies had some subjects in common, and in some Journal of Molecular Neuroscience % Reg. (p < 0.01) ↓ ↑ 109 31 57 62 235 34 15 11 47% 71% 44% 42% 36% 41% 80% 45% 51 22 25 26 84 14 12 0 0 0 0 0 1 0 0 5 19 5 13 22 28 16% 0% 0% 5% 29% 3 0 0 1 8 0 0 0 0 0 No. of genes cases shared the same set of subjects. It might have been expected that this would result in a higher degree of concordance in the significant gene and pathway lists across the set of studies, but that was not seen. For the meta-analysis, the fact that the studies are only partially statistically independent means that the findings carry somewhat less weight than a metaanalysis of 12 completely independent studies. This issue goes beyond this particular study, as relatively few brain banks supply the samples for the multitude of published genomic, genetic, and proteomic studies in bipolar disorder. This meta-analysis highlights the importance of statistical methodology in the analysis of brain microarray studies to take into account the fact that bipolar-dysregulated genes have small FCs. Importance of Statistical Variation The results also highlight the underappreciated role of statistical variation in microarray studies. The within-study coefficient of variation for individual genes was typically in the range of 30%–50%. This means that FCs can generally only be quantified to within ±0.2 for studies of the sizes we examined. One is tempted to search for reasons for disagreements between results by assigning causes such as platform differences, brain region differences, processing differences, etc. In this meta-analysis, however, we found that the differences in results across studies were often consistent with random variation around the underlying consensus FC. Because the true effects in bipolar disorder seem to be small, this variation assumes a larger role than in array studies of cancer, Volume 31, 2007 232 Fig. 2. FC values with 95% confidence interval for all probes that map to the specific pathway/Go term. Example graphs are provided for (A) oxidative phosphorylation, (B) MHC class II receptor activity, (C) proteasome, and (D) Metal ion binding. Meta-Analysis in Bipolar Disorder for example, where FCs might be on the order of 3–5. With large FCs, study differences owing to statistical variation of ±0.2 for FCs would not significantly alter the genes identified in the same way it would for real FCs of 1.1. This is not to say that platform and region differences do not exist. We found many genes with probe set–specific regulation, but cases in which these differences could be attributed to some known cause were the exception and not the rule. The importance of statistical variation in microarray studies also illustrates the limitations of so-called PCR validation. As an example, consider a study that finds FC = 1.3 and p < 0.05 for caspase 8 instead of the consensus value of approx 1.0 (no difference) that the meta-analysis indicated. We might suppose further that this observed FC of 1.3 was a reflection of the magnitude of random variation in FCs around the true value. PCR testing of the same samples that were used to identify a 1.3 FC for caspase 8 would be expected to also yield a significant FC of 1.3, recapitulating the study results but providing an erroneous validation of the role of caspase 8 in bipolar disorder. Summary of Biological Findings This meta-analysis provides evidence to support the energy processing dysfunction hypothesis in bipolar disorder. Many genes, pathways, and GO terms related to energy processing were highly significantly regulated, even after accounting for the effects of confounding variables, including medication usage. In contrast, support for the oligodendrocyte/ myelin hypothesis was rather modest in this analysis. There was strong evidence for regulation in genes associated with protein turnover, including proteasome and ubiquitination. The metallothionein and MHC regulation findings were also quite robust, but their role in the pathophysiology of the disease process has yet to be elucidated. One cannot hope to summarize 20,000 genes, 4000 pathway/GO terms, 500 subject samples, and 12 studies in a single report. Additional work remains to be done for this large set of data, and these data will be made publicly available on-line in the near future, though much of the gene/pathway-level and meta-analyses results are already available (Higgs et al., 2006). Several of the findings can be followed up with animal studies, protein studies, etc. In addition, although regional and/or platform difference were minimal for many genes and pathways and the analysis focused on these commonalities, a further exploration of instances of disagreement can be conducted. We also plan to study single-nucleotide Journal of Molecular Neuroscience 233 polymorphism profiling for these patients and correlate the genetic information with the gene expression information. These studies also evaluated a schizophrenia cohort, and a similar meta-analysis for schizophrenia, looking for disease similarities and differences in the results, is under way. If an overall lesson can be drawn from this analysis, it is the benefit of data sharing for the elucidation of the genomic basis of psychiatric disorders. Acknowledgments The data of the following SMRI collaborators were used in this meta-analysis: Drs. C. Anthony Altar, Sabine Bahn, Haiming Chen, Seth E. Dobrin, Allen A. Fienberg, Tadafumi Kato, Pamela Sklar, Marquis P. Vawter, and L. Trevor Young. References Aguirre-Samudio A. J. and Nicolini H. (2005) DRD4 polymorphism and the association with mental disorders. Rev. Invest. Clin. 57, 65–75. Altar C. A., Jurata L. W., Charles V., Lemire A., Liu P., Bukhman Y., et al. (2005) Deficient hippocampal neuron expression of proteasome, ubiquitin, and mitochondrial genes in multiple schizophrenia cohorts. Biol. Psychiatry 58, 85–96. Baron M. (2002) Manic-depression genes and the new millennium: poised for discovery. Mol. Psychiatry 7, 342–358. Barrett T. B., Hauger R. L., Kennedy J. L., Sadovnick A. D., Remick R. A., Keck P. E., et al. (2003) Evidence that a single nucleotide polymorphism in the promoter of the G protein receptor kinase 3 gene is associated with bipolar disorder. Mol. Psychiatry 8, 546–557. Benes F. M., Matzilevich D., Burke R. E., and Walsh J. (2006) The expression of proapoptosis genes is increased in bipolar disorder, but not in schizophrenia. Mol. Psychiatry 11(3), 241–251. Bezchlibnyk Y. B., Wang J. F., McQueen G. M., and Young L. T. (2001) Gene expression differences in bipolar disorder revealed by cDNA array analysis of postmortem frontal cortex. J. Neurochem. 79, 826–834. Blair I. P., Adams L. J., Badenhop R. F., Moses M. J., Scimone A., Morris J. A., et al. (2002) A transcript map encompassing a susceptibility locus for bipolar affective disorder on chromosome 4q35. Mol. Psychiatry 7, 867–873. Costa E., Davis J., Grayson D. R., Guidotti A., Pappas G. D., and Pesold C. (2001) Dendritic spine hypoplasticity and downregulation of reelin and GABAergic tone in schizophrenia vulnerability. Neurobiol. Dis. 8, 723–742. De Luca V., Likhodi O., Van Tol H. H., Kennedy J. L., and Wong A. H. (2005) Tryptophan hydroxylase 2 gene expression and promoter polymorphisms in bipolar disorder and schizophrenia. Psychopharmacology (Berl.) 183, 378–382. Volume 31, 2007 234 Dennis G. Jr., Sherman B. T., Hosack D. A., Yang J., Gao W., Lane H. C., and Lempicki R. A. (2003) DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol. 4, P3. Eastwood S. L. and Harrison P. J. (2000) Hippocampal synaptic pathology in schizophrenia, bipolar disorder and major depression: a study of complexin mRNAs. Mol. Psychiatry 5, 425–432. Fatemi S. H., Laurence J. A., Araghi-Niknam M., Stary J. M., Schulz S. C., Lee S., and Gottesman I. I. (2004) Glial fibrillary acidic protein is reduced in cerebellum of subjects with major depression, but not schizophrenia. Schizophr. Res. 69, 317–323. Grayson D. R., Jia X., Chen Y., Sharma R. P., Mitchell C. P., Guidotti A., and Costa E. (2005) Reelin promoter hypermethylation in schizophrenia. Proc. Natl. Acad. Sci. U. S. A. 102, 9341–9346. Higgs B. W., Elashoff M., Richman S, and Barci B. (2006) An online database for brain disease research. BMC Genomics 7, 70. Hurd Y. L. (2002) Subjects with major depression or bipolar disorder show reduction of prodynorphin mRNA expression in discrete nuclei of the amygdaloid complex. Mol. Psychiatry 7, 75–81. Iwamoto K., Bundo M., and Kato T. (2005) Altered expression of mitochondria-related genes in postmortem brains of patients with bipolar disorder or schizophrenia, as revealed by large-scale DNA microarray analysis. Hum. Mol. Genet. 14, 241–253. Iwamoto K., Kakiuchi C., Bundo M., Ikeda K., and Kato T. (2004) Molecular characterization of bipolar disorder by comparing gene expression profiles of postmortem brains of major mental disorders. Mol. Psychiatry 9, 406–416. Jurata L. W., Bukhman Y. V., Charles V., Capriglione F., Bullard J., Lemire A. L., et al. (2004) Comparison of microarray-based mRNA profiling technologies for identification of psychiatric disease and drug signatures. J. Neurosci. Methods 138, 173–188. Kakiuchi C., Iwamoto K., Ishiwata M., Bundo M., Kasahara T., Kusumi I., et al. (2003) Impaired feedback regulation of XBP1 as a genetic risk factor for bipolar disorder. Nat. Genet. 35, 171–175. Kan P. X., Popendikyte V., Kaminsky Z. A., Yolken R. H., and Petronis A. (2004) Epigenetic studies of genomic retroelements in major psychosis. Schizophr. Res. 67, 95–106. Kapczinski F., Frey B. N., and Zannatto V. (2004) Physiopathology of bipolar disorders: what has changed in the last 10 years? Rev. Bras. Psiquiatr. 26, 17–21. Kato T., Kuratomi G., and Kato N. (2005) Genetics of bipolar disorder. Drugs Today (Barc.) 41, 335–344. Kelsoe J. R. and Niculescu A. B. III (2002) Finding genes for bipolar disorder in the functional genomics era: from convergent functional genomics to phenomics and back. CNS Spectr. 7, 215–216, 223–216. Kim E. H., Kim T. S., Sun W., Kim D. S., Chung H. S., Kim D. K., and Park S. H. (2004) Differential regulation of metallothionein-I and metallothionein-II mRNA Journal of Molecular Neuroscience Elashoff et al. expression in the rat brain following traumatic brain injury. Mol. Cells 18, 326–331. Koh P. O., Undie A. S., Kabbani N., Levenson R., GoldmanRakic P. S., and Lidow M. S. (2003) Up-regulation of neuronal calcium sensor-1 (NCS-1) in the prefrontal cortex of schizophrenic and bipolar patients. Proc. Natl. Acad. Sci. U. S. A. 100, 313–317. Konradi C. (2005) Gene expression microarray studies in polygenic psychiatric disorders: applications and data analysis. Brain Res. Brain Res. Rev. 50, 142–155. Konradi C., Eaton M., MacDonald M. L., Walsh J., Benes F. M., and Heckers S. (2004) Molecular evidence for mitochondrial dysfunction in bipolar disorder. Arch. Gen. Psychiatry 61, 300–308. Kuromitsu J., Yokoi A., Kawai T., Nagasu T., Aizawa T., Haga S., and Ikeda K. (2001) Reduced neuropeptide Y mRNAlevels in the frontal cortex of people with schizophrenia and bipolar disorder. Brain Res. Gene Expr. Patterns 1, 17–21. Law A. J. and Deakin J. F. (2001) Asymmetrical reductions of hippocampal NMDAR1 glutamate receptor mRNA in the psychoses. Neuroreport 12, 2971–2974. Law A. J., Weickert C. S., Hyde T. M., Kleinman J. E., and Harrison P. J. (2004) Reduced spinophilin but not microtubule-associated protein 2 expression in the hippocampal formation in schizophrenia and mood disorders: molecular evidence for a pathology of dendritic spines. Am. J. Psychiatry 161, 1848–1855. Lenox R. H. and Hahn C. G. (2000) Overview of the mechanism of action of lithium in the brain: fifty-year update. J. Clin. Psychiatry 61, 5–15. Maier W., Hofgen B., Zobel A., and Rietschel M. (2005) Genetic models of schizophrenia and bipolar disorder: overlapping inheritance or discrete genotypes? Eur. Arch. Psychiatry Clin. Neurosci. 255, 159–166. Manji H. K., Bebchuk J. M., Moore G. J., Glitz D., Hasanat K. A., and Chen G. (1999) Modulation of CNS signal transduction pathways and gene expression by moodstabilizing agents: therapeutic implications. J. Clin. Psychiatry 60, 27–39, 40–21, 113–116. Molnar M., Potkin S. G., Bunney W. E., and Jones E. G. (2003) MRNA expression patterns and distribution of white matter neurons in dorsolateral prefrontal cortex of depressed patients differ from those in schizophrenia patients. Biol. Psychiatry 53, 39–47. Munakata K., Iwamoto K., Bundo M., and Kato T. (2005) Mitochondrial DNA 3243A>G mutation and increased expression of LARS2 gene in the brains of patients with bipolar disorder and schizophrenia. Biol. Psychiatry 57, 525–532. Natale J. E., Knight J. B., Cheng Y., Rome J. E., and Gallo V. (2004) Metallothionein I and II mitigate agedependent secondary brain injury. J. Neurosci. Res. 78, 303–314. Ogden C. A., Rich M. E., Schork N. J., Paulus M. P., Geyer M. A., Lohr J. B., et al. (2004) Candidate genes, pathways and mechanisms for bipolar (manic-depressive) and related disorders: an expanded convergent Volume 31, 2007 Meta-Analysis in Bipolar Disorder 235 functional genomics approach. Mol. Psychiatry 9, 1007–1029. Raybould R., Green E. K., MacGregor S., Gordon-Smith K., Heron J., Hyde S., et al. (2005) Bipolar disorder and polymorphisms in the dysbindin gene (DTNBP1). Biol. Psychiatry 57, 696–701. Sun Y., Zhang L., Johnston N. L., Torrey E. F., and Yolken R. H. (2001) Serial analysis of gene expression in the frontal cortex of patients with bipolar disorder. Br. J. Psychiatry 41, s137–s141. Thomson P. A., Wray N. R., Millar J. K., Evans K. L., Hellard S. L., Condie A., et al. (2005) Association between the TRAX/DISC locus and both bipolar disorder and schizophrenia in the Scottish population. Mol. Psychiatry 10, 616, 657–668. Tkachev D., Mimmack M. L., Ryan M. M., Wayland M., Freeman T., Jones P. B., et al. (2003) Oligodendrocyte dysfunction in schizophrenia and bipolar disorder. Lancet 362, 798–805. Torrey E. F., Webster M., Knable M., Johnston N., and Yolken R. H. (2000) The Stanley Foundation Brain Collection and Neuropathology Consortium. Schizophr. Res. 44, 151–155. Veldic M., Guidotti A., Maloku E., Davis J. M., and Costa E. (2005) In psychosis, cortical interneurons overexpress DNA-methyltransferase 1. Proc. Natl. Acad. Sci. U. S. A. 102, 2152–2157. Webster M. J., Knable M. B., O’Grady J., Orthmann J., and Weickert C. S. (2002) Regional specificity of brain glucocorticoid receptor mRNA alterations in subjects with schizophrenia and mood disorders. Mol. Psychiatry 7, 924, 985–994. Woo T. U., Walsh J. P., and Benes F. M. (2004) Density of glutamic acid decarboxylase 67 messenger RNAcontaining neurons that express the N-methyl-Daspartate receptor subunit NR2A in the anterior cingulate cortex in schizophrenia and bipolar disorder. Arch. Gen. Psychiatry 61, 649–657. Supplemental Data Table 1 Top genes Symbol Locus ID Gene name FC p value LST1 PSME3 NFYC SCAP2 K-α-1 C6orf68 UBE2N CLCN4 KDELR2 7940 10197 4802 8935 10376 116150 7334 1183 11014 Leukocyte-specific transcript 1 Proteasome (prosome, macropain) activator subunit 3 (PA28 γ; χ) Nuclear transcription factor Y, γ Src family-associated phosphoprotein 2 Tubulin, α, ubiquitous Chromosome 6 open reading frame 68 Ubiquitin-conjugating enzyme E2N (UBC13 homolog, yeast) Chloride channel 4 KDEL (Lys-Asp-Glu-Leu) endoplasmic reticulum protein retention receptor 2 Heat shock 70-kDa protein 8 MHC, class II, DRα Sorting nexin 3 Splicing factor, arginine/serine-rich 2 Heterogeneous nuclear ribonucleoprotein D-like Zinc finger protein 363 v-crk sarcoma virus CT10 oncogene homolog (avian) MHC, class I, B β-Amyloid (A4) precursor-like protein 2 Protein tyrosine phosphatase type IVA, member 1 Four-and-a-half LIM domains 1 COP9 constitutive photomorphogenic homolog subunit 8 (Arabidopsis) Acid phosphatase 1, soluble Mitofusin 1 Dual-specificity phosphatase 6 Ubiquitination factor E4B (UFD2 homolog, yeast) Phosphoglycerate kinase 1 S-phase kinase-associated protein 1A (p19A) WD repeat endosomal protein –1.11 –1.20 –1.12 –1.18 –1.12 –1.08 –1.23 –1.15 –1.15 5.79E-10 2.26E-08 1.14E-07 1.17E-07 1.57E-07 1.83E-07 2.64E-07 4.67E-07 5.79E-07 –1.25 –1.29 –1.22 –1.17 –1.13 –1.19 –1.13 –1.13 –1.16 –1.12 –1.16 –1.17 6.52E-07 8.26E-07 9.10E-07 9.69E-07 1.09E-06 1.89E-06 2.43E-06 2.62E-06 3.06E-06 3.16E-06 3.47E-06 3.84E-06 –1.17 –1.12 –1.22 –1.12 –1.20 –1.19 –1.14 4.12E-06 4.31E-06 4.40E-06 5.35E-06 5.44E-06 6.26E-06 6.55E-06 HSPA8 HLA-DRA SNX3 SFRS2 HNRPDL ZNF363 CRK HLA-B APLP2 PTP4A1 FHL1 COPS8 3312 3122 8724 6427 9987 25898 1398 3106 334 7803 2273 10920 ACP1 MFN1 DUSP6 UBE4B PGK1 SKP1A KIAA1449 52 55669 1848 10277 5230 6500 57599 (Continued) Journal of Molecular Neuroscience Volume 31, 2007 236 Elashoff et al. Table 1 (Continued) Symbol EIF5 ILF3 DR1 Gene name FC p value Eukaryotic translation initiation factor 5 Interleukin enhancer binding factor 3, 90 kDa Down-regulator of transcription 1, TBP-binding (negative cofactor 2) Cadherin 11, type 2, OB-cadherin (osteoblast) Karyopherin (importin) β3 Solute carrier family 30 (zinc transporter), member 5 APG5 autophagy 5-like (Saccharomyces cerevisiae) flightless I homolog (Drosophila) CGI-01 protein Ribulose-5-phosphate-3-epimerase FUS-interacting protein (serine-arginine rich) 1 parvin, β HESB-like domain containing 2 DnaJ (Hsp40) homolog, subfamily B, member 6 Tyrosine 3-monooxygenase/tryptophan 5- monooxygenase activation protein, ζ polypeptide DEAD (Asp-Glu-Ala-Asp) box polypeptide 18 Cytoplasmic linker-associated protein 2 Hypothetical protein FLJ90005 Melanoma cell-adhesion molecule Fucosidase, α-L-2, plasma ATPase, Ca2+ transporting, plasma membrane 2 Solute carrier family 33 (acetyl-CoA transporter), member 1 Ubiquitin-conjugating enzyme E2D 3 (UBC4/5 homolog, yeast) Putative translation initiation factor Potassium voltage-gated channel, shaker-related subfamily, β member 1 HIV-1 Tat-interactive protein 2, 30 kDa Peptidylprolyl isomerase A (cyclophilin A) deltex 3 homolog (Drosophila) Reticulon 4 Methyltransferase-like 3 Endothelial-derived gene 1 Hypothetical protein MGC29875 Splicing factor 3a, subunit 1, 120 kDa Mitogen-activated protein kinase kinase kinase 7 Platelet-activating factor acetylhydrolase, isoform Ib, α-subunit, 45 kDa Propionyl coenzyme A carboxylase, β polypeptide Microtubule-associated protein 7 Synovial sarcoma, X breakpoint 2-interacting protein Chemokine (C-X-C motif) receptor 4 Adenosylmethionine decarboxylase 1 Sparc/osteonectin, cwcv, and kazal-like domain proteoglycan (testican) 3 Signal transducer and activator of transcription 1, 91 kDa Ras homolog enriched in brain Synovial sarcoma translocation, chromosome 18 Zinc finger protein 576 Tumor protein D52 Splicing factor, arginine/serine-rich 5 Coated vesicle membrane protein Calcium/calmodulin-dependent protein kinase (CaM kinase) IIβ cAMP responsive element binding protein-like 2 –1.17 –1.12 –1.10 7.05E-06 7.14E-06 7.71E-06 –1.10 –1.16 –1.09 –1.13 –1.14 –1.17 –1.10 –1.10 –1.07 –1.21 –1.17 –1.17 7.97E-06 9.66E-06 1.01E-05 1.06E-05 1.11E-05 1.15E-05 1.17E-05 1.17E-05 1.25E-05 1.26E-05 1.34E-05 1.41E-05 –1.13 –1.21 –1.23 –1.09 –1.09 –1.15 –1.07 –1.14 –1.13 –1.14 1.51E-05 1.57E-05 1.58E-05 1.59E-05 1.66E-05 1.73E-05 1.97E-05 2.26E-05 2.31E-05 2.32E-05 –1.12 –1.09 –1.09 –1.17 –1.11 –1.07 –1.07 –1.11 –1.09 –1.15 2.45E-05 2.47E-05 2.54E-05 2.55E-05 2.57E-05 2.65E-05 2.71E-05 2.78E-05 2.88E-05 2.93E-05 –1.10 –1.17 –1.13 –1.08 –1.15 –1.13 2.96E-05 3.08E-05 3.08E-05 3.13E-05 3.21E-05 3.25E-05 –1.08 –1.10 –1.06 –1.14 –1.13 –1.11 –1.22 –1.15 –1.17 3.25E-05 3.27E-05 3.60E-05 3.71E-05 3.71E-05 3.81E-05 3.84E-05 3.87E-05 4.07E-05 Locus ID 1983 3609 1810 CDH11 KPNB3 SLC30A5 APG5L FLII CGI-01 RPE FUSIP1 PARVB HBLD2 DNAJB6 YWHAZ 1009 3843 64924 9474 2314 51603 6120 10772 29780 81689 10049 7534 DDX18 CLASP2 FLJ90005 MCAM FUCA2 ATP2B2 SLC33A1 UBE2D3 SUI1 KCNAB1 8886 23122 127544 4162 2519 491 9197 7323 10209 7881 HTATIP2 PPIA DTX3 RTN4 METTL3 EG1 MGC29875 SF3A1 MAP3K7 PAFAH1B1 10553 5478 196403 57142 56339 80306 27042 10291 6885 5048 PCCB MAP7 SSX2IP CXCR4 AMD1 SPOCK3 5096 9053 117178 7852 262 50859 STAT1 RHEB SS18 ZNF576 TPD52 SFRS5 RNP24 CAMK2B CREBL2 6772 6009 6760 79177 7163 6430 10959 816 1389 Journal of Molecular Neuroscience Volume 31, 2007 Meta-Analysis in Bipolar Disorder 237 Table 1 (Continued) Symbol NR3C1 2908 YME1L1 GGCX ACTG1 BTN2A1 SH3GLB1 AGA ST13 10730 2677 71 11120 51100 175 6767 EPB41L3 LOC54499 OGT 23136 54499 8473 ADPRTL2 10038 HTATIP PSMA1 IDH3B PRKAR1A 10524 5682 3420 5573 XLKD1 MAPK1 PRO1853 LAPTM5 NAP1L1 SRP72 DKFZp762C186 FLJ21940 CAPN3 TNFSF10 IVNS1ABP GAPD PCTK1 HSPA4 DAB2 Gene name FC p value Nuclear receptor subfamily 3, group C, member 1 (glucocorticoid receptor) YME1-like 1 (S. cerevisiae) γ-Glutamyl carboxylase Actin, γ1 Butyrophilin, subfamily 2, member A1 SH3-domain GRB2-like endophilin B1 Aspartylglucosaminidase Suppression of tumorigenicity 13 (colon carcinoma) (Hsp70-interacting protein) Erythrocyte membrane protein band 4.1-like 3 Putative membrane protein O-linked N-acetylglucosamine (GlcNAc) transferase (UDP-N-acetylglucosamine:polypeptide-N-acetylglucosaminyl transferase) ADP-ribosyltransferase (NAD+; poly[ADP-ribose] polymerase)-like 2 HIV-1 Tat-interactive protein, 60 kDa Proteasome (prosome, macropain) subunit, α type, 1 Isocitrate dehydrogenase 3 (NAD+) β Protein kinase, cAMP-dependent, regulatory, type I, α (tissue-specific extinguisher 1) Extracellular link domain containing 1 Mitogen-activated protein kinase 1 Hypothetical protein PRO1853 Lysosomal-associated multispanning membrane protein-5 Nucleosome assembly protein 1-like 1 Signal recognition particle, 72 kDa Tangerin FLJ21940 protein Calpain 3, (p94) Tumor necrosis factor (ligand) superfamily, member 10 Influenza virus NS1A-binding protein Glyceraldehyde-3-phosphate dehydrogenase PCTAIRE protein kinase 1 Heat shock, 70-kDa protein 4 disabled homolog 2, mitogen-responsive phosphoprotein (Drosophila) Sec23 homolog A (S. cerevisiae) Cleavage and polyadenylation specific factor 5, 25 kDa Golgi SNAP receptor complex member 2 Growth arrest and DNA-damage-inducible, β Nuclear receptor subfamily 4, group A, member 2 Discs, large homolog 3 (neuroendocrine-dlg, Drosophila) Nucleolar protein 7, 27 kDa Hypothetical protein KIAA1164 Small inducible cytokine subfamily E, member 1 (endothelial monocyte-activating) Putative dimethyladenosine transferase Ariadne homolog 2 (Drosophila) Allograft inflammatory factor 1 Phosphatidylinositol binding clathrin assembly protein Malic enzyme 2, NAD(+)-dependent, mitochondrial –1.12 4.10E-05 –1.14 –1.08 –1.11 –1.13 –1.10 –1.09 –1.19 4.37E-05 4.65E-05 4.70E-05 4.88E-05 4.95E-05 5.01E-05 5.32E-05 –1.23 –1.12 –1.12 5.36E-05 5.48E-05 5.52E-05 –1.13 5.64E-05 –1.17 –1.22 –1.18 –1.19 5.69E-05 5.81E-05 5.83E-05 5.97E-05 –1.10 –1.15 –1.11 –1.24 –1.10 –1.12 –1.11 –1.09 –1.15 –1.16 –1.17 –1.10 –1.13 –1.08 –1.09 6.84E-05 6.85E-05 6.85E-05 6.89E-05 6.99E-05 7.00E-05 7.24E-05 7.35E-05 7.47E-05 7.50E-05 7.64E-05 7.88E-05 7.94E-05 7.96E-05 8.41E-05 –1.22 –1.10 –1.08 1.12 –1.14 –1.13 –1.11 –1.08 –1.09 8.61E-05 8.86E-05 9.02E-05 9.05E-05 9.29E-05 9.31E-05 9.34E-05 9.44E-05 9.52E-05 –1.06 –1.08 –1.08 –1.10 –1.12 0.000101 0.000102 0.000103 0.000104 0.00011 Locus ID 10894 5594 55471 7805 4673 6731 254102 64848 825 8743 10625 2597 5127 3308 1601 SEC23A CPSF5 GOSR2 GADD45B NR4A2 DLG3 NOL7 KIAA1164 SCYE1 10484 11051 9570 4616 4929 1741 51406 54629 9255 HSA9761 ARIH2 AIF1 PICALM ME2 27292 10425 199 8301 4200 (Continued) Journal of Molecular Neuroscience Volume 31, 2007 238 Elashoff et al. Table 1 (Continued) Symbol Locus ID DLGAP1 SYT5 CRH UPF3A C11ORF4 MAFG 9229 6861 1392 65110 56834 4097 KIAA1102 SLC38A2 13CDNA73 HLA-DPA1 KPNA1 DDX3X H41 CPD ABCG1 SIP KPNB1 SLC29A1 CAST APBB2 22998 54407 10129 3113 3836 1654 55573 1362 9619 27101 3837 2030 831 323 RANGAP1 DCTD LAPTM4B CAMKK2 DJ1042K10.2 TDE1 OK/SW-cl.56 ACLY HLA-C PTPRN2 TPT1 BHLHB2 ALDH3A2 HFE CD24 SULF1 EIF4E DCN PRMT3 5905 1635 55353 10645 27352 10955 203068 47 3107 5799 7178 8553 224 3077 934 23213 1977 1634 10196 ANGPTL2 MS4A6A PCTAIRE2BP MADH2 SNX2 G3BP2 CUL3 PAI-RBP1 RFP KCNK1 FLJ39616 CCT2 GRIK1 23452 64231 23424 4087 6643 9908 8452 26135 5987 3775 51275 10576 2897 Gene name Discs, large (Drosophila) homolog-associated protein 1 Synaptotagmin V Corticotropin-releasing hormone UPF3 regulator of nonsense transcripts homolog A (yeast) Chromosome 11 hypothetical protein ORF4 v-maf musculoaponeurotic fibrosarcoma oncogene homolog G (avian) KIAA1102 protein Solute carrier family 38, member 2 Hypothetical protein CG003 MHC, class II, DPα1 Karyopherin α1 (importin α5) DEAD (Asp-Glu-Ala-Asp) box polypeptide 3, X-linked Hypothetical protein H41 Carboxypeptidase D ATP-binding cassette, subfamily G (WHITE), member 1 Siah-interacting protein Karyopherin (importin)β1 Solute carrier family 29 (nucleoside transporters), member 1 Calpastatin β-amyloid (A4) precursor protein-binding, family B, member 2 (Fe65-like) Ran GTPase-activating protein 1 dCMP deaminase Lysosomal-associated protein transmembrane 4β Calcium/calmodulin-dependent protein kinase kinase 2, β Hypothetical protein DJ1042K10.2 Tumor differentially expressed 1 β5-tubulin ATP citrate lyase MHC, class I, C Protein tyrosine phosphatase, receptor type, N polypeptide 2 Tumor protein, translationally controlled 1 Basic helix-loop-helix domain containing, class B, 2 Aldehyde dehydrogenase 3 family, member A2 Hemochromatosis CD24 antigen (small cell lung carcinoma cluster 4 antigen) Sulfatase 1 Eukaryotic translation initiation factor 4E Decorin Protein arginine N-methyltransferase 3 (hnRNP methyltransferase S. cerevisiae)-like 3 Angiopoietin-like 2 Membrane-spanning 4 domains, subfamily A, member 6A Tudor repeat associator with PCTAIRE 2 MAD (mothers against decapentaplegic) homolog 2 (Drosophila) Sorting nexin 2 Ras-GTPase activating protein SH3 domain-binding protein 2 Cullin 3 PAI-1 mRNA-binding protein Ret finger protein Potassium channel, subfamily K, member 1 Apoptosis-related protein PNAS-1 Chaperonin-containing TCP1, subunit 2 (β) Glutamate receptor, ionotropic, kainate 1 Journal of Molecular Neuroscience FC p value –1.08 –1.16 –1.22 –1.16 –1.10 –1.11 0.000114 0.000114 0.000116 0.000117 0.000119 0.000119 –1.11 –1.27 –1.09 –1.17 –1.10 –1.12 –1.11 –1.09 –1.10 –1.10 –1.07 –1.18 –1.06 –1.06 0.000121 0.000121 0.000121 0.000123 0.000126 0.000131 0.000134 0.000136 0.000136 0.00014 0.000143 0.000146 0.000154 0.000156 –1.20 –1.14 –1.18 –1.18 –1.11 –1.19 –1.22 –1.20 –1.13 –1.16 –1.11 –1.19 –1.09 –1.04 –1.09 –1.08 –1.16 –1.08 –1.09 0.000156 0.000157 0.000157 0.00016 0.000167 0.00017 0.000171 0.000172 0.000172 0.000172 0.000175 0.000175 0.000176 0.000179 0.000185 0.00019 0.00019 0.000191 0.000201 –1.09 –1.11 –1.16 –1.10 –1.13 –1.17 –1.12 –1.08 –1.12 –1.25 –1.12 –1.23 –1.10 0.000202 0.000206 0.000209 0.000209 0.000212 0.000213 0.000215 0.000218 0.000219 0.000221 0.000222 0.000224 0.000224 Volume 31, 2007 Meta-Analysis in Bipolar Disorder 239 Table 1 (Continued) Symbol Locus ID HIST1H1C CANX KLHL12 PAIP1 MRPS6 SDHC 3006 821 59349 10605 64968 6391 GNAQ UBE2D2 GK001 CGI-27 NUDT4 SCAMP1 YWHAQ 2776 7322 57003 51072 11163 9522 10971 PREP TP53AP1 UBE2J1 KIAA0276 LOC283033 SGCB ACTR3 ATP5L 5550 11257 51465 23142 283033 6443 10096 10632 PHTF1 TRA1 BBP MTMR2 SIP1 DKFZP564G2022 PDCD4 TRO FLJ11149 MORF4L1 SFRS6 ENSA APACD MRPL10 RAB6A CCNK C14orf65 DKFZP547E1010 PREI3 RNF14 API5 LGALS8 PDCD6IP MRGX1 GOLGA1 HLA-DMB NRCAM ANAPC5 DDX54 10745 7184 83941 8898 8487 25963 27250 7216 55312 10933 6431 2029 10190 124995 5870 8812 317762 26097 25843 9604 8539 3964 10015 259249 2800 3109 4897 51433 79039 Gene name Histone 1, H1c Calnexin Kelch-like 12 (Drosophila) Poly(A)-binding protein interacting protein 1 Mitochondrial ribosomal protein S6 Succinate dehydrogenase complex, subunit C, integral membrane protein, 15 kDa Guanine nucleotide-binding protein (G protein), q polypeptide Ubiquitin-conjugating enzyme E2D 2 (UBC4/5 homolog, yeast) GK001 protein C21orf19-like protein Nudix (nucleoside diphosphate-linked moiety X)-type motif 4 Secretory carrier membrane protein 1 Tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein, θ polypeptide Prolyl endopeptidase TP53-activated protein 1 Ubiquitin-conjugating enzyme E2, J1 (UBC6 homolog, yeast) KIAA0276 protein Hypothetical protein LOC283033 Sarcoglycan, β (43-kDa dystrophin-associated glycoprotein) ARP3 actin-related protein 3 homolog (yeast) ATP synthase, H+ transporting, mitochondrial F0 complex, subunit g Putative homeodomain transcription factor 1 Tumor rejection antigen (gp96) 1 β-Amyloid-binding protein precursor Myotubularin-related protein 2 Survival of motor neuron protein interacting protein 1 DKFZP564G2022 protein Programmed cell death 4 (neoplastic transformation inhibitor) Trophinin Riboflavin kinase Mortality factor 4-like 1 Splicing factor, arginine/serine-rich 6 Endosulfine α ATP-binding protein associated with cell differentiation Mitochondrial ribosomal protein L10 RAB6A, member RAS oncogene family Cyclin K Chromosome 14 open reading frame 65 DKFZP547E1010 protein Preimplantation protein 3 Ring finger protein 14 Apoptosis inhibitor 5 Lectin, galactoside-binding, soluble, 8 (galectin 8) Programmed cell death 6-interacting protein G-protein-coupled receptor MRGX1 Golgi autoantigen, Golgin subfamily a, 1 MHC, class II, DM β Neuronal cell-adhesion molecule Anaphase-promoting complex subunit 5 DEAD (Asp-Glu-Ala-Asp) box polypeptide 54 FC p value 1.09 –1.15 –1.17 –1.11 –1.10 –1.14 0.000227 0.000231 0.000234 0.000239 0.000243 0.000244 –1.10 –1.13 –1.17 –1.13 –1.13 –1.12 –1.18 0.000244 0.000245 0.000253 0.000255 0.000257 0.000258 0.00026 –1.12 –1.10 –1.07 –1.12 –1.09 –1.05 –1.18 –1.17 0.000263 0.000266 0.000268 0.00027 0.00027 0.000271 0.000272 0.000281 –1.09 –1.14 –1.09 –1.14 –1.10 –1.11 –1.07 –1.11 –1.16 –1.16 –1.16 –1.15 –1.18 –1.05 –1.12 1.09 1.09 –1.10 –1.16 –1.18 –1.06 –1.07 –1.15 1.05 –1.09 –1.10 –1.14 –1.14 –1.11 0.000282 0.000285 0.000287 0.000287 0.000295 0.000296 0.000298 0.000309 0.000311 0.000315 0.000319 0.000328 0.000328 0.000329 0.000331 0.000331 0.000331 0.000332 0.000332 0.000332 0.000337 0.000338 0.000346 0.000351 0.000357 0.000359 0.000364 0.000371 0.000377 (Continued) Journal of Molecular Neuroscience Volume 31, 2007 240 Elashoff et al. Table 1 (Continued) Symbol HNRPU GFPT1 GNB1 TMEM1 MTNR1A PRO1855 FBXO9 LOC339290 BUB3 OAZIN RABIF HLA-DRB1 CUL4B CNOT8 EMILIN5 PC4 HNRPH3 CASP1 RABEP1 ZNF45 SYNCRIP KIAA0999 SMT3H1 DEPDC6 BAG2 HES7 ATP2C1 ZNF278 DLAT ATP5C1 RAD1 STS BRE MGC40168 ZNF341 DUSP3 C21orf63 ARHE TIP120A HNRPK AMFR DHPS OAZ2 COQ7 APOL2 GNAS CCNI POLR2C PDHA1 Gene name FC p value Heterogeneous nuclear ribonucleoprotein U (scaffold attachment factor A) Glutamine-fructose-6-phosphate transaminase 1 Guanine nucleotide-binding protein (G protein), β polypeptide 1 Transmembrane protein 1 Melatonin receptor 1A Hypothetical protein PRO1855 F-box only protein 9 Hypothetical protein LOC339290 BUB3 budding uninhibited by benzimidazoles 3 homolog (yeast) Ornithine decarboxylase antizyme inhibitor RAB-interacting factor MHC, class II, DR β1 Cullin 4B CCR4-NOT transcription complex, subunit 8 Elastin microfibril interfacer 5 Activated RNA polymerase II transcription cofactor 4 Heterogeneous nuclear ribonucleoprotein H3 (2H9) Caspase 1, apoptosis-related cysteine protease (Interleukin 1, β, convertase) Rabaptin, RAB GTPase-binding effector protein 1 Zinc finger protein 45 (a Kruppel-associated box (KRAB) domain polypeptide) Synaptotagmin binding, cytoplasmic RNA-interacting protein KIAA0999 protein SMT3 suppressor of mif two 3 homolog 1 (yeast) DEP domain-containing 6 BCL2-associated athanogene 2 hairy and enhancer of split 7 (Drosophila) ATPase, Ca2+ transporting, type 2C, member 1 Zinc finger protein 278 Dihydrolipoamide S-acetyltransferase (E2 component of pyruvate dehydrogenase complex) ATP synthase, H+ transporting, mitochondrial F1 complex, γ polypeptide 1 RAD1 homolog (Schizosaccharomyces pombe) Steroid sulfatase (microsomal), arylsulfatase C, isozyme S Brain and reproductive organ-expressed (TNFRSF1A modulator) Hypothetical protein MGC40168 Zinc finger protein 341 Dual-specificity phosphatase 3 (vaccinia virus phosphatase VH1-related) Chromosome 21 open reading frame 63 ras homolog gene family, member E TBP-interacting protein Heterogeneous nuclear ribonucleoprotein K Autocrine motility factor receptor Deoxyhypusine synthase ornithine decarboxylase antizyme 2 Coenzyme Q7 homolog, ubiquinone (yeast) Apolipoprotein L, 2 GNAS complex locus Cyclin I Polymerase (RNA) II (DNA-directed) polypeptide C, 33 kDa Pyruvate dehydrogenase (lipoamide) α1 –1.07 0.000382 –1.11 –1.14 –1.07 1.04 –1.12 –1.13 –1.07 –1.12 –1.13 –1.13 –1.19 –1.09 –1.11 1.13 –1.15 –1.08 –1.04 0.000384 0.000389 0.000391 0.000393 0.000397 0.000401 0.000404 0.000407 0.000409 0.00041 0.00042 0.000421 0.000423 0.000429 0.000434 0.000437 0.000437 –1.08 –1.07 0.000439 0.000442 –1.07 –1.08 –1.22 –1.15 –1.07 –1.06 –1.10 –1.06 –1.14 0.000442 0.000443 0.000445 0.000446 0.000447 0.000457 0.000461 0.000464 0.000474 –1.13 0.000482 –1.09 –1.10 –1.11 1.01 –1.07 –1.12 0.000486 0.000489 0.000491 0.000495 0.000496 0.000498 1.02 –1.17 –1.12 –1.19 –1.26 –1.21 –1.15 –1.07 –1.07 –1.08 –1.18 –1.11 –1.13 0.000502 0.000505 0.000506 0.000509 0.000512 0.000517 0.000518 0.000521 0.000523 0.000523 0.000524 0.000532 0.000532 Locus ID 3192 2673 2782 7109 4543 55379 26268 339290 9184 51582 5877 3123 8450 9337 90187 10923 3189 834 9135 7596 10492 23387 6612 64798 9532 84667 27032 23598 1737 509 5810 412 9577 148645 84905 1845 59271 390 55832 3190 267 1725 4947 10229 23780 2778 10983 5432 5160 Journal of Molecular Neuroscience Volume 31, 2007 Meta-Analysis in Bipolar Disorder 241 Table 1 (Continued) Locus ID Gene name FC p value ARF3 SRP46 UBE2L3 SET IGF1 TIMM23 TA-LRRP SLCO1A2 CYP2E1 B3GALT3 GPX3 TPM1 CRI1 KIAA0368 PCMT1 UXS1 INPP4A DKFZP434C212 PUM1 DKFZp761B1514 FKBP1A XRCC5 377 10929 7332 6418 3479 10431 23507 6579 1571 8706 2878 7168 23741 23392 5110 80146 3631 26130 9698 84248 2280 7520 –1.23 –1.08 –1.13 –1.12 –1.09 –1.17 –1.17 –1.08 –1.07 –1.13 –1.14 –1.08 –1.18 –1.12 –1.19 –1.11 –1.06 –1.10 –1.14 –1.17 –1.11 –1.13 0.000533 0.000534 0.000545 0.000548 0.000549 0.000554 0.000557 0.000558 0.000561 0.000563 0.000567 0.000569 0.00057 0.000571 0.000572 0.00058 0.000582 0.000588 0.000594 0.000594 0.000604 0.000606 MGC2776 FLI1 FLJ10876 UBE2G1 80746 2313 55758 7326 –1.07 –1.07 –1.11 –1.11 0.000619 0.000619 0.000621 0.000622 GPAA1 DNAJA1 STARD13 ADAM28 RGS4 RPL22 SLAMF9 MGC3067 PMS2L9 PTBP1 HLA-DRB3 CX3CR1 LAMP2 PROSC HCCS RBBP4 MOG CHMP1.5 OSBP LARP WARS MAP2K4 MYO1B ENPP4 8733 3301 90627 10863 5999 6146 89886 79139 5387 5725 3125 1524 3920 11212 3052 5928 4340 57132 5007 23367 7453 6416 4430 22875 ADP-ribosylation factor 3 Splicing factor, arginine/serine-rich, 46 kDa Ubiquitin-conjugating enzyme E2L 3 SET translocation (myeloid leukemia-associated) Insulin-like growth factor 1 (somatomedin C) Translocase of inner mitochondrial membrane 23 homolog (yeast) T-cell activation leucine repeat-rich protein Solute carrier organic anion transporter family, member 1A2 Cytochrome P450, family 2, subfamily E, polypeptide 1 UDP-Gal:CGlcNAc V 1,3-galactosyltransferase, polypeptide 3 Glutathione peroxidase 3 (plasma) Tropomyosin 1 (α) CREBBP/EP300 inhibitory protein 1 KIAA0368 Protein-L-isoaspartate (D-aspartate) O-methyltransferase UDP-glucuronate decarboxylase 1 Inositol polyphosphate-4-phosphatase, type I, 107 kDa DKFZP434C212 protein pumilio homolog 1 (Drosophila) Hypothetical protein DKFZp761B1514 FK506-binding protein 1A, 12 kDa X-ray repair complementing defective repair in Chinese hamster cells 5 (double-strand-break rejoining; Ku autoantigen, 80 kDa) Hypothetical protein MGC2776 Friend leukemia virus integration 1 Hypothetical protein FLJ10876 Ubiquitin-conjugating enzyme E2G 1 (UBC7 homolog, Caenorhabditis elegans) GPAA1P anchor attachment protein 1 homolog (yeast) DnaJ (Hsp40) homolog, subfamily A, member 1 START domain-containing 13 A disintegrin and metalloproteinase domain 28 Regulator of G-protein signaling 4 Ribosomal protein L22 SLAM family member 9 Hypothetical protein MGC3067 Postmeiotic segregation increased 2-like 9 Polypyrimidine tract-binding protein 1 MHC, class II, DR β3 Chemokine (C-X3-C motif) receptor 1 Lysosomal-associated membrane protein 2 Proline synthetase cotranscribed homolog (bacterial) Holocytochrome c synthase (cytochrome c heme-lyase) Retinoblastoma-binding protein 4 Myelin oligodendrocyte glycoprotein CHMP1.5 protein Oxysterol-binding protein Likely ortholog of mouse la-related protein Tryptophanyl-tRNA synthetase Mitogen-activated protein kinase kinase 4 Myosin IB Ectonucleotide pyrophosphatase/phosphodiesterase 4 (putative function) –1.13 –1.21 –1.15 –1.04 –1.18 –1.12 –1.05 –1.10 –1.05 –1.06 –1.11 –1.29 –1.14 –1.08 –1.15 –1.14 –1.14 –1.10 –1.12 –1.10 –1.18 –1.17 –1.06 –1.13 0.00063 0.000634 0.000642 0.000657 0.000657 0.000658 0.000659 0.000663 0.00067 0.000673 0.000679 0.000681 0.000687 0.000687 0.000689 0.000692 0.0007 0.0007 0.000702 0.000705 0.000712 0.000712 0.000715 0.000719 Symbol (Continued) Journal of Molecular Neuroscience Volume 31, 2007 242 Elashoff et al. Table 1 (Continued) Symbol Locus ID HMGB1 UBE2A UCP2 MT1X ABCF2 CALU SRPK2 SCARB2 NOLC1 GTPBP4 MED8 3146 7319 7351 4501 10061 813 6733 950 9221 23560 112950 PTGS1 5742 KIAA1040 RNASET2 ADAMTSL1 KIAA0252 CTNND2 23041 8635 92949 23168 1501 MPRG HIVEP2 MAPK8 NCKAP1 TSN RALGPS1A E46L ARFD1 HDGFRP3 ARNTL HSPC056 SNX1 SEC23B DAF 54852 3097 5599 10787 7247 9649 25814 373 50810 406 25852 6642 10483 1604 MGC3262 GHITM DHFR APXL ATP6V1C1 78992 27069 1719 357 528 PHF10 FAM20B HINT1 PAPOLA MTO1 PSMB4 ICA1 D4S234E SIAT1 CREM JTB PITPN PABPN1 CPR8 55274 9917 3094 10914 25821 5692 3382 27065 6480 1390 10899 5306 8106 9236 Gene name High-mobility group box 1 Ubiquitin-conjugating enzyme E2A (RAD6 homolog) Uncoupling protein 2 (mitochondrial, proton carrier) Metallothionein 1X ATP-binding cassette, subfamily F (GCN20), member 2 Calumenin SFRS protein kinase 2 Scavenger receptor class B, member 2 Nucleolar and coiled-body phosphoprotein 1 GTP-binding protein 4 Mediator of RNA polymerase II transcription, subunit 8 homolog (yeast) Prostaglandin-endoperoxide synthase 1 (prostaglandin G/H synthase and cyclooxygenase) KIAA1040 protein Ribonuclease T2 ADAMTS-like 1 KIAA0252 protein Catenin (cadherin-associated protein), δ2 (neural plakophilin-related arm-repeat protein) Membrane progestin receptor γ Human immunodeficiency virus type I enhancer-binding protein 2 Mitogen-activated protein kinase 8 NCK-associated protein 1 Translin Ral guanine nucleotide exchange factor RalGPS1A Like mouse brain protein E46 ADP-ribosylation factor domain protein 1, 64 kDa Hepatoma-derived growth factor, related protein 3 Aryl hydrocarbon receptor nuclear translocator-like HSPC056 protein Sorting nexin 1 SEC23 homolog B (S. cerevisiae) Decay accelerating factor for complement (CD55, Cromer blood group system) Hypothetical protein MGC3262 Growth hormone-inducible transmembrane protein Dihydrofolate reductase Apical protein-like (Xenopus laevis) ATPase, H+ transporting, lysosomal 42 kDa, V1 subunit C, isoform 1 PHD finger protein 10 Family with sequence similarity 20, member B Histidine triad nucleotide-binding protein 1 Poly(A) polymerase α Mitochondrial translation optimization 1 homolog (S. cerevisiae) Proteasome (prosome, macropain) subunit, β type, 4 Islet cell autoantigen 1, 69 kDa DNA segment on chromosome 4 (unique) 234 expressed sequence Sialyltransferase 1 (β-galactoside α-2,6-sialyltransferase) cAMP-responsive element modulator Jumping translocation breakpoint Phosphotidylinositol transfer protein Poly(A)-binding protein, nuclear 1 Cell cycle progression 8 protein Journal of Molecular Neuroscience FC p value –1.11 –1.21 –1.12 1.25 –1.07 –1.07 –1.10 –1.11 –1.11 –1.12 –1.13 0.000722 0.000726 0.000726 0.000731 0.000731 0.000738 0.000739 0.00075 0.000753 0.000755 0.000757 –1.07 0.000768 –1.08 –1.09 –1.09 –1.10 –1.17 0.00077 0.000774 0.000778 0.000796 0.000802 1.03 –1.21 –1.08 –1.16 –1.13 –1.10 –1.18 –1.15 –1.12 –1.12 –1.06 –1.09 –1.09 –1.08 0.000806 0.00081 0.000815 0.000816 0.000821 0.000823 0.000826 0.000828 0.000833 0.000835 0.00084 0.000844 0.000848 0.000852 –1.10 –1.15 –1.05 –1.15 –1.14 0.000852 0.000863 0.000867 0.000868 0.000868 –1.10 –1.15 –1.22 –1.07 –1.12 –1.20 –1.08 –1.14 –1.09 –1.05 –1.12 –1.14 –1.11 –1.09 0.000869 0.000883 0.000887 0.000892 0.000894 0.000896 0.000914 0.000923 0.000927 0.00093 0.000942 0.000943 0.000946 0.000949 Volume 31, 2007 Meta-Analysis in Bipolar Disorder 243 Table 1 (Continued) Symbol FKBP4 SERF1A SYPL GDI2 BIN1 RAB4A MGC12909 P5 PDE4DIP Locus ID 2288 8293 6856 2665 274 5867 147339 10130 9659 Gene name FK506-binding protein 4, 59 kDa Small EDRK-rich factor 1A (telomeric) Synaptophysin-like protein GDP dissociation inhibitor 2 Bridging integrator 1 RAB4A, member RAS oncogene family Hypothetical protein MGC12909 Protein disulfide isomerase-related protein Phosphodiesterase 4D-interacting protein (myomegalin) FC p value –1.19 –1.14 –1.14 –1.15 –1.12 –1.11 –1.06 –1.12 –1.06 0.000951 0.000959 0.000963 0.000963 0.000978 0.000982 0.000985 0.00099 0.000999 Expression Calculation and Normalization values for a given gene so that its expression was comparable across the platforms. Affymetrix: MAS 5.0 expression values were calculated based on scaling to a target intensity of 100, then transformed by log2(x + 20). Calls were computed using the MAS 5.0 present/absent algorithm. Codelink: Expression values were median scaled to a target intensity of 66 (the overall median), then transformed by log2(x + 20). Genes with a G call were called present; others were called absent. cDNA: Cy5 foreground and Cy5 background were computed using loess normalization. Next, Cy5 foreground/Cy5 background was calculated. These values were median scaled to a target intensity of 985 (the overall median), then log2 transformed. Genes were called present if they had no quality flags and if their Cy5/Cy3 ratio was above the 96% quantile of ratios for control genes. Agilent: Cy5 (or Cy3 for dye swaps) expression values were calculated based on loess normalized signals. These values were median scaled to a target intensity of 640 (the overall median) and then averaged across dye-swap replicates. Genes were called present if their Cy5 (or Cy3 for dye swaps) value was above the 96% quantile for control genes. Cross-Platform Normalization: A gene-specific normalization function was computed for each unique gene in the studies. This function used a robust median normalization to scale the expression Quality Control Journal of Molecular Neuroscience For QC analysis, we used four primary QC metrics: scale factor, percent present, 5′3′ ratio, and average correlation. For each metric, we computed the distribution of the metric across the samples within each study. Although no hard cutoffs were applied for each of the QC metrics, we examined the distribution of the metrics to determine if samples appeared to be outliers. We also used principal components analysis and clustering to visualize the relationship between the samples and determine if one or more samples appeared to be outliers. Scale Factor: Scale factor for an Affy chip is the ratio between the trimmed mean of the expression values for that chip and the target intensity (in our case, 100). A similar value was computed for the non-Affy chips (target intensity/median expression value). Percent Present: The number of probes called present on the array divided by the total number of probes. A description of how calls were generated for nonAffy arrays is detailed in Materials and Methods. 5′3′ Ratios: For Affy arrays, the GAPDH and βactin 5′3′ ratios were computed based on MAS 5 expression values. Average Correlation: The average within-study correlation of a particular sample to each other sample in the study. Volume 31, 2007