The primary cohort consisted of 22 subjects, including seven healthy controls, six patients with bipolar disorder type I (BPD), and nine patients with MDD (Cohort A). The MDD results were validated with an additional independent cohort, including seven controls and five MDD patients (Cohort B). Toxicology data and/or recent clinical data were examined for the BPD and MDD subjects and detailed clinical information is summarized in Table 1. To minimize potential confounding effects of agonal events, we used three criteria in subject selection: brain tissue pH of >6.6, ribosomal RNA 28S/18S ratio of >1.5, and sudden death with the agonal period lasting less than 1 h. We have previously found large differences in gene expression values when using longer agonal periods, >1 h and as long as 1–2 days, and would have found large effects due only to mixing samples with long and short agonal duration (Li et al., 2004; Tomita et al., 2004; Vawter et al., 2006; Atz et al., 2007).

Brain tissue and clinical information were obtained by the UCI Brain Bank staff with the informed consent of decedents' next-of-kin, and were processed through a standard protocol approved by the local UCI Institutional Review Board (Tomita et al., 2004). None of the subjects included in this study had specific agonal conditions including cancer, hypoxia, coma, pyrexia, seizure, dehydration, hypoglycemia, multi-organ failure, skull fracture, ingestion of neurotoxic substances or prolonged agonal duration, which are known to affect tissue pH, RNA integrity and gene expression profiles in post-mortem brain (Li et al., 2004; Tomita et al., 2004). The variables for gender, age, post-mortem interval (PMI), tissue pH, and 28S/18S ratio were balanced between diagnostic and control groups. Patients who were prescribed lithium and selective serotonin reuptake inhibitor (SSRI) at the time of death are also indicated (Table 1).

The experimental procedures are described elsewhere (Evans et al., 2004; Tomita et al., 2004). Total RNA was extracted from ACC and DLPFC of Cohort A (Table 1), and purified with silica-based mini-spin columns (Qiagen, Valencia, California). The oligonucleotide microarray experiments were conducted utilizing Affymetrix U133A Gene Chips following the manufacturer's protocol (Affymetrix, Santa Clara, CA). To detect gene expression differences between diagnostic groups reliably, we replicated the experiments as follows: Cohort A used 22 subjects' total RNA from ACC which were processed at two laboratories (University of California, Irvine and University of California, Davis) and DLPFC that were processed independently (University of California, Davis and University of Michigan, Ann Arbor). For Cohort B 12 total RNA samples from ACC and DLPFC (additional 12 subjects) were processed at the same independent laboratories as cohort A. In summary, for each BPD subject there were duplicate arrays run, and for MDD subjects there were duplicate arrays run, plus an additional cohort of MDD subjects were run also with duplicate arrays.

Since a considerable number of probe sequences in the Affymetrix original chip definition file (CDF) were found to incorrectly BLAST to other transcripts, a refined CDF based on an UniGene cluster was developed (Evans et al., 2004; Dai et al., 2005), and was used for data analyses presented in this paper. The refined CDF file is available on http://brainarray.mhri.med.umich.edu/brainarray/.

The focus of this study was 445 GPLS genes found on the Affymetrix arrays used in this study (Table S1). Signal intensities were extracted with Robust Multi-array Average (RMA) for each probe set and each subject. Gene-wise Pearson's correlation coefficients between experimental duplicates were calculated. Only the genes that were significantly correlated (p < 0.05) between experimental duplicates were considered to be reliably measured genes, and were subjected to downstream analyses. These genes were analyzed in a mixed-model ANOVA utilizing Partek Pro 6.0 (Partek, MO) for the main effect of the diagnostic classification (BPD, MDD, control) and for sex (Evans et al., 2004). We calculated the false discovery rate (FDR) in the microarray study by Benjamini-Hochberg step-down procedure using Partek Genomics setting the FDR to 0.05. Since the three diagnostic groups were well matched on age and pH (Table 1), the gene expression values were not tested by ANCOVA. However, in post-hoc analysis, we determined whether the tissue pH was correlated with any differentially expressed GPLS genes. The post-hoc analysis showed that there were no significant correlations found with tissue pH and differentially expressed GPLS genes. A post-hoc group comparison of subjects with SSRI treatment and without treatment was conducted in MDD subjects, and similar analysis for lithium was conducted for BPD subjects.

The hypergeometric distribution was used for calculating probabilities that the observed number of genes in each gene category for cAMP or PI signaling pathways (listed in Table S2) were detected as differentially expressed genes in the comparisons between patients and control. False discovery rate were evaluated by QVALUE (Storey and Tibshirani, 2003). Calculations for hypergeometric probabilities were performed using SISA online statistics package (http://www.quantitativeskills.com/sisa/distributions/hypergeo.htm).

For further technical evaluation of the microarray data, we evaluated mRNA expression levels by quantitative real-time reverse transcriptase PCR (qPCR) for the seven genes listed in Table 3. We followed up on results primarily that met FDR in cohort A for BPD, and both Cohorts A and B for MDD, except for RGS20, which we followed up based upon Cohort A only. These genes selected for qPCR analyses met the criteria of FDR at the level of accepting 5% false positives and percentage change greater than 20% in the ACC microarray results. The threshold cycle (C_t) values for the target transcripts were normalized to averaged C_t values for two reference transcripts, Jagged 1 (JAG1) and solute carrier family 9 isoform 1 (SLC9A1), which showed equivalent expression levels among the three diagnostic groups (BPD, MDD, and control) throughout the 3 brain regions on our normalized microarray data. After DNase digestion and purification, total RNA (1 μg) samples from each of the 34 subjects in Cohorts A and B were used as a template for first-strand cDNA synthesis using poly-dT primer. The mRNA for each transcript was measured using the SybrGreen system with Prism model 7000-sequence detection instrument (Applied Biosystems, Foster City, California) and primer sequences listed in Table S1. The C_t, which correlatedinversely with the target mRNA concentration, was measured as the cycle number at which the SybrGreen fluorescent signal increased abovea pre-set threshold level. The qPCR experiments were performed three times for each transcript, and the C_t triplicate values were averaged.

The gene expression fold changes for GPRC5B and GPR37 showed an increase in BPD and decrease in MDD. We selected those genes for in situ hybridization (ISH) using the ACC regions from six representative subjects to evaluate the cellular localization of GPRC5B and GPR37 mRNA expression. A formal statistical analysis of the ISH data was not applied due to low subject number. ISH was performed following a protocol as described elsewhere (Neal et al., 2001; Lopez-Figueroa et al., 2004). Briefly, frozen ACC tissue blocks were cut into 12-μm sections on a cryostat. After fixation with 4% paraformaldehyde for 1 h, sections were hybridized overnight with a ³⁵S-labeled antisense cRNA probes for GPRC5B and GPR37. Following hybridization, wash, RNase A treatment, dehydration and air-dry, sections were exposed to Kodak XAR-5 X-ray film for 21 days. Technical control studies were performed by (1) using ³⁵S-labeled sense RNA probes for GPRC5B and GPR37, and (2) pre-treating tissue with RNase A before hybridizing with the antisense probes.

The genes involved in cAMP signaling pathways, which were differentially expressed in ACC of BPD patients compared to controls, are listed in Table 2. The cAMP signaling pathway is shown in Figure 1A. The data show that six significant transcripts out of 39 that repress cAMP signaling activity (Table 1) were increased in ACC from BPD patients at the transcriptional level. The group of molecules that act negatively on cAMP signaling activity was significantly over-represented in BPD based on a hypergeometric distribution (p = 0.037, q = 0.36). Thus, there was a nominal enrichment of genes that act negatively on cAMP signaling activity (G_i proteins, phosphodiesterase, protein kinase A inhibitors, and protein phosphatase) in BPD in the ACC.

In contrast, of those genes that repressed cAMP signaling activity in MDD (Figure 1B and Table 2) there was only one gene (PDE8A) differentially expressed, hence this signaling pathway was not over-represented (p > 0.05).

The two GPCRs that showed the most consistent differential expression patterns through both mood disorder analyses were GPCR family C, group 5, member B (GPRC5B) and G protein-coupled receptor 37 (GPR37)as shown in Table 2 and validated by QPCR (Table 3). GPRC5B was significantly increased in ACC of BPD patients and in DLPFC of BPD. GPR37 was also significantly increased in ACC of BPD. GPRC5B was significantly decreased in ACC and DLPFC of MDD patients.

We followed up on results primarily that met FDR in cohort A for BPD, and both Cohorts A and B for MDD, except for RGS20, which we followed up based upon Cohort A meeting FDR while Cohort B met uncorrected p-value significance only. Nine GPLS transcripts were selected for validation, five for BPD and four for MDD. The qPCR results showed 100% concordance between qPCR and microarray fold change for BPD in ACC (Table 3) and 4/5 transcripts were statistically significant. Four GPLS related transcripts showed 100% concordance between qPCR and microarray fold change for MDD in ACC (Table 3) and 4/4 transcripts were statistically significant. NPY, GPRC5B, GPR37, and INPP1 were significantly increased in ACC of BPD, whereas GPRC5B, GPR37, RGS20, and PPP1R3C were significantly decreased in the ACC of MDD.

GPR37 mRNA was preferentially expressed in subcortical white matter. The sense probe hybridization signal showed background signal. GPR37 mRNA expression in subcortical white matter was higher in ACC of both of the BPD subjects compared to both of the control subjects, whereas GPR37 mRNA expression in subcortical white matter was lower in ACC of both of the MDD subjects compared to both of the control subjects (Figure 2). This preliminary data suggests that the significant differential microarray and QPCR expressions of GPR37 between BPD/MDD and controls may reflect altered expression levels of GPR37 mRNA in subcortical white matter. mRNA expression of GPRC5B was not specific in ACC samples tested, most likely due to strong background hybridization issues.

Among all GPCRs, the most consistent differential expression patterns were observed for GPRC5B and GPR37. GPRC5B was significantly increased in ACC and DLPFC of BPD, while significantly decreased in ACC and DLPFC of MDD patients. GPR37 was significantly increased in ACC of BPD, and significantly decreased in ACC and DLPFC of MDD patients. The consistent cortical alterations of GPRC5B and GPR37 expression observed in MDD patients suggest a role of these genes in the pathophysiology of MDD. So far, these two orphan GPCRs have not been highlighted as candidate susceptible genes for BPD and MDD, however previous microarray data supports decrease of both GPRC5B and GPR37 expressions in MDD patients (Aston et al., 2005). Our microarray data strongly suggests these genes might be involved in the pathophysiology of the disorders. GPR37 and GPRC5B are both implicated in Parkinson's disease (Marazziti et al., 2009, 2011), which has depression as one premorbid indicator, possibly concomitant with brain neurodegenerative processes.

GPR37 appears to facilitate dopamine neurotransmission, and reductions in brain GPR37 leads to behavioral alterations in conditioned place preferences to reinforcements such as cocaine or amphetamine (Marazziti et al., 2011). GPR37 (Class A receptor) contains a large extracellular domain which led to speculations that the ligand of this receptor may be a protein rather than a peptide (Leng et al., 1999). GPR37 is predominantly expressed in the brain (Zeng et al., 1997; Marazziti et al., 1998; Leng et al., 1999), especially in glial cells of the fiber tracts, Purkinje cells in cerebellum, neuronal cells in hippocampus CA3 region and substantianigra (Zeng et al., 1997; Imai et al., 2001). GPR37 has been shown to be involved in the dopaminergic system (Imai et al., 2001, 2002; Marazziti et al., 2004). GPR37 is also called Pael receptor (PAELR). Unfolded PAELR is a substrate of E3 ubiquitin ligase Parkin, and accumulation of Pael-R in the endoplasmic reticulum (ER) of dopaminergic neurons is considered to induce ER stress leading to neurodegeneration in Parkinson disease (Imai et al., 2002). Gpr37^−/− mice shows reduced striatal dopamine content and enhanced amphetamine sensitivity (Marazziti et al., 2004). Our ISH data suggest that the predominant expression of GPR37 mRNA in ACC occurred mainly in subcortical white matter. GPR37 is a substrate of parkin (PARK2), and its insoluble aggregates accumulate in brain tissue samples of Parkinson's disease patients, AKA PaelR (Marazziti et al., 2009).

GPRC5B is phylogenetically classified into a type C GPCR, however, in contrast to other type C genes, GPRC5B has a short N terminal domain (Brauner-Osborne and Krogsgaard-Larsen, 2000). QPCR based expression analysis shows GPRC5B is predominantly expressed in brain, with most abundant expression in neuronal cells of caudate, putamen, substantia nigra, thalamus and hippocampus and glial cells of the fiber tracts (Brauner-Osborne and Krogsgaard-Larsen, 2000; Robbins et al., 2002). Behavioral abnormalities exist in GPRC5B knock-outs, decreased spontaneous locomotor activity, and decreased responses to a new environment (Sano et al., 2011).

The expression patterns of GPRC5B and GPR37 suggest that both of these orphan GPCRs may be involved in various cellular functional activities in the brain including neuronal and glial cells. Functional characterization of GPRC5B and GPR37 may provide new insights into the pathophysiology of mood disorder, and potential novel strategies for developing methods for diagnoses and treatment of the disorders.

There are limitations to this post-mortem study of GPLS pathway expression. We conducted seven qPCR validations of genes that passed FDR, and although a small number, those genes did represent large enough percent fold changes to validate by qPCR. Other technologies could be used that are less costly, such as Nanostring or Fluidigm for additional validation studies, which are now being conducted on larger sample sizes. In addition, we do not have confirmed toxicological analysis of all drugs of abuse and therapeutic levels to rule out these effects in explaining case—control expression differences. Finally, we are aware that small residual agonal effects might exist that are not corrected in this analysis, however we did confirm that the present data presented in Table 2, was not due to the effects of age or pH on gene expression.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.