Differences in Lymphocyte Electron Transport Gene Expression Levels Between Subjects With Bipolar Disorder and Normal Controls in Response to Glucose Deprivation Stress

Context  Bipolar disorder (BPD) is among the top 10 causes of disability worldwide. Recent findings on the etiology of the disease point to a disturbed mitochondrial energy metabolism in the brain of subjects with BPD.

Objective  To test whether gene transcripts for proteins of the mitochondrial respiratory chain have altered levels in glucose-deprived lymphocytes from patients with BPD.

Design  Microarrays were used to measure gene expression levels in fresh lymphocytes and in lymphocytes cultured for 5 days in regular or low-glucose medium.

Setting  Subjects with BPD were recruited through the Schizophrenia and Bipolar Disorders Program, McLean Hospital, Belmont, Mass. Controls were recruited through advertising.

Patients  A total of 21 patients with BPD (inpatients and outpatients) and 21 control subjects.

Main Outcome Measure  Expression levels for genes of proteins involved in mitochondrial respiration.

Results  We found an opposite molecular response of control and BPD lymphocytes to glucose deprivation. Whereas lymphocytes of normal controls responded to glucose deprivation with an up-regulation of nuclear transcripts for proteins of the electron transfer chain, subjects with BPD had a tendency to down-regulate these transcripts.

Conclusions  The results suggest that the normal molecular adaptation to energy stress is deficient in lymphocytes from patients with BPD.

Bipolar disorder (BPD) affects between 1% and 3% of the US population,¹ is associated with a high risk of suicide,² and places an annual economic burden on the nation in excess of $40 billion (estimated in 1991).³ Diagnosis of BPD is based on the nature and course of symptoms, but the etiology of the disease remains elusive and no diagnostic test is available. Recent studies showed decreased hippocampal and dorsolateral prefrontal cortex levels of creatine kinase messenger RNA (mRNA)⁴ as well as decreased levels of high-energy phosphates^5,6 or increased lactate levels⁷ in the frontal and temporal lobes of patients with BPD, introducing the hypothesis that mitochondrial energy metabolism plays an important role in the etiology of the disease.⁸ Previously, we described a down-regulation in nuclear mRNA coding for mitochondrial electron transfer proteins in postmortem hippocampal tissue from patients with BPD.⁹ To follow up on this study, we examined gene expression levels in lymphocytes of patients with BPD.

Gene expression patterns in lymphocytes have been analyzed recently in BPD and schizophrenia.^10-12 Lymphoblastoid cell lines of patients with BPD have been used to study the expression of nuclear genes coding for mitochondrial proteins.¹³ Lymphocytes are easier to collect than brain tissue, and it has been shown that peripheral gene expression can be informative about gene expression in the central nervous system. Whole blood has significant gene expression similarities with multiple central nervous system tissues, and the expression levels of many classes of biologically relevant processes are not significantly different between whole blood and prefrontal cortex.¹⁴ In patients with schizophrenia, a comparison of gene expression profiles from brain tissue with profiles from peripheral blood cells identified disease-associated genes that were common to both tissues, confirming the validity of gene expression profiling of blood for detecting schizophrenia biomarkers.¹⁰

Lymphocytes not only are easily accessible but, unlike postmortem brain tissue, can also be subjected to experimental manipulations. We decided to go beyond gene expression profiling from freshly drawn blood and to culture lymphocytes of patients with BPD and normal controls (NCs) in medium with normal or low glucose levels for 5 days to examine gene expression levels after low-glucose stress.

Twenty-one healthy NCs and 21 patients diagnosed with BPD according to the criteria of DSM-IV¹⁵ (Table 1) provided informed consent as approved by the institutional review board at McLean Hospital, Belmont, Mass. Samples from patients with BPD and NCs were collected over the course of 6 months, and batches of BPD samples were matched with NC samples. Lymphocytes from 10 to 30 mL of freshly drawn blood were separated by centrifugation using Histopaque columns (Sigma-Aldrich, St Louis, Mo), washed 3 times, and split into 3 batches. One batch, containing two thirds of all cells, was frozen at −80°C and later subjected to gene expression microarray analysis, whereas 1 smaller batch each was cultured in either regular RPMI-1640 medium or low-glucose RPMI-1640 medium (25% normal glucose content; 0.5 g/L) for a period of 5 days, after which cells were frozen at −80°C. After sample collection was concluded, RNA was extracted from each batch (RNagents kit; Promega, Madison, Wis), complementary DNA (cDNA) was synthesized from 4 μg of RNA from fresh lymphocytes (SuperScript double-stranded cDNA synthesis kit; Invitrogen Corp, Carlsbad, Calif) or 1 μg of RNA from cultured lymphocytes (MessageAmp II-96 kit; Ambion, Austin, Tex), and biotinylated RNA was synthesized from cDNA (for fresh lymphocytes, Enzo IVT kit; Enzo Biochem, Farmingdale, NY; for cultured lymphocytes, MessageAmp II-96 kit). Biotinylated RNA was fragmented and hybridized to the HG-U133A 2.0 array (Affymetrix, Santa Clara, Calif) overnight at 45°C and stained on a washing station with 2 rounds of streptavidin-phycoerythrin (Molecular Probes, Eugene, Ore) separated by a round of biotinylated antistreptavidin antibody (Vector Laboratories, Burlingame, Calif) as described previously.^9,16 All of the fresh-frozen lymphocytes were worked up in 1 batch for gene array experiments. All of the cultured lymphocytes were worked up together in a separate batch with an improved protocol developed during the course of this project, for which the amount of input RNA could be lowered from 4 μg to 1 μg. Because of the small sample sizes and the variable amount of lymphocytes yielded from individual probands, a number of samples did not yield enough mRNA for gene array analysis (Table 1). The number of samples per group ranged from 10 to 17.

Gene expression levels were calculated with the robust multichip analysis algorithm (RMAExpress; http://rmaexpress.bmbolstad.com) and compared using the comparison analysis of the dChip program (http://biosun1.harvard.edu/complab/dchip), which computes P values based on the t distribution, with the degrees of freedom set according to the Welch-modified 2-sample t test.^17,18 Only samples that met quality-control criteria provided by the GeneChip Operating Software (Affymetrix) and DNA-Chip Analyzer (dChip 2006)¹⁹ were incorporated into the analysis (Table 1) (mean ± SD noise, 0.9 ± 0.1; mean ± SD percentage present call, 56.2% ± 1.7%; mean ± SD 3′-5′ glyceraldehyde-3-phosphate dehydrogenase ratio, 1.4 ± 0.4; mean ± SD 3′-5′ β-actin ratio, 1.6 ± 0.8; mean ± SD percentage of array outliers, 0.16% ± 0.22%; mean ± SD percentage of single outliers, 0.046% ± 0.043%; no significant differences were observed between groups).

All of the genes differently expressed between 2 groups (P≤.05; ≥50% present call; 4 groups: BPD over NC for low glucose, BPD over NC for normal glucose, low over normal glucose for NC, and low over normal glucose for BPD) were subjected to a classification analysis using the Gene Ontology database gene product attributes (http://www.geneontology.org) calculated with the dChip software. Multiples of same transcripts were masked for classification analyses. Similar results were obtained with log₂-transformed and natural scale data. Analysis of variance filtering was carried out in dChip. Permuted and adjusted P values for mitochondrial genes were obtained with the MAPPFinder program (http://www.genmapp.org).²⁰ We used 271 groupings (MAPPs) of individual genes for this analysis, grouped in a manner that avoided duplication of the same genes in independent groups. MAPPFinder calculates a nonparametric statistic based on 2000 permutations of the data, randomizing the gene associations for each sample to generate a distribution of z scores for each MAPP, which are then used to assign permuted P values. In addition, the Westfall-Young adjustment, which calculates the familywise error rate for each sample and accounts for multiple testing, is used for multiple testing. This adjustment gives the adjusted P value. Fisher exact test was used to examine the statistical difference between the percentage of regulation of mitochondrial transcripts vs the percentage of regulation of all of the transcripts.

Families of genes, such as genes of the mitochondrial respiratory chain or genes specific for B or T cells, were compared between NC and BPD samples with 2-tailed, paired t tests using the natural expression values. For example, for the mitochondrial respiratory chain, the expression level of each of the 114 individual transcripts in an experimental group was divided by the average expression level of each transcript in all of the groups. False discovery rates were calculated in the dChip program by estimating the empirical false discovery rate for a group of genes (ie, the 114 mitochondrial transcripts) using 2000 random permutations.

Real-time quantitative polymerase chain reaction (qPCR) was used for data verification and carried out as previously described.^9,16 Briefly, cDNA was synthesized from 1 μg of total RNA (SuperScript First-Strand Synthesis System for real-time qPCR; Invitrogen Corp) and oligonucleotide deoxythymidine primers. Primer sets for each gene were designed with the Primer3 software (http://www.genome.wi.mit.edu/cgi-bin/primer/primer3.cgi) for amplicons of 100 to 200 base pairs. Melt curve analysis and polyacrylamide gel electrophoresis were used to confirm the specificity of each primer pair. The iQ SYBR Green Supermix (Bio-Rad, Hercules, Calif) was used for the experiment carried out with a MyiQ real-time PCR detection system (Bio-Rad) in a volume of 20 μL, with 4 μL of 1:10 diluted cDNA samples and 0.3 μM primers. The PCR cycling conditions were initially 95°C for 5 minutes followed by 39 cycles of 94°C for 10 seconds, 55°C for 15 seconds, and 72°C for 20 seconds. Data were collected between 72°C and 82°C depending on amplicon melt temperature. A melt curve analysis was performed at the end of each qPCR experiment, from 60°C to 95°C. Dilution curves were generated for each primer pair in every experiment by diluting cDNA from a vehicle sample to a final concentration of 1.00, 0.20, 0.04, and 0.008. The logarithms of the dilution values were plotted against the cycle values for the standard curve. Blanks were run with each dilution curve to control for cross contamination. Dilution curves, blanks, and samples were run in duplicate. Reported values were normalized to the average of 3 internal standards (Table 2), which were not regulated in the gene array analysis. Using an average of multiple internal standards for normalization leads to increased accuracy, as the conventional use of a single gene for normalization leads to relatively large errors.²¹ Genes for qPCR were chosen based on gene array results (Figure 1). Sequences and Entrez GeneID numbers are provided in Table 2.

In the comparison of NC and BPD lymphocytes in low-glucose medium, among the Gene Ontology database categories with the most significant hits for down-regulated genes were mitochondrion (P≤.001), cytochrome c oxidase activity (P<.001), mitochondrial electron transport chain (P = .001), and ubiquinol–cytochrome c reductase activity (P<.001). These groups were second only to ribosomal proteins, a category that was also affected in our previous study in the human hippocampus in BPD.⁹ In the MAPPFinder program, the permuted P value for transcripts of the electron transport chain was less than or equal to .001 and the adjusted P value was .01 (z score, 6.1). Further analyses revealed that the expression of 18 probe sets of electron transfer transcripts, of 114 on the array (for GenBank and Entrez Gene numbers of all 114 transcripts, see eTable 1), was significantly lower in BPD lymphocytes under glucose deprivation (Figure 1A), whereas no probe sets were expressed at higher levels than in NC lymphocytes. The 18 probe sets represented 15 individual mRNA transcripts, composing 19% of all individual electron transfer mRNAs on the array (35/114 probe sets were duplicate probe sets), whereas on average only 8.2% of the probe sets were lower in BPD lymphocytes under glucose deprivation (Figure 2A). This difference was significant in Fisher exact test. Furthermore, the entire group of electron transfer transcripts was shifted significantly toward lower expression levels in BPD (Figure 3A and Table 3), and these trends were also seen in qPCR (Figure 4). No significant shift in expression levels of mitochondrial transcripts was observed between BPD and NC lymphocytes under normal glucose concentrations (Figure 1B, Figure 2B, Figure 3B, and Table 3) or in fresh, uncultured lymphocytes (Figure 5 and Table 3). The pattern of electron transfer transcript expression in subjects with BPD and NCs suggests a different molecular response to glucose deprivation. Whereas NCs showed an up-regulation of these transcripts in response to energy stress (Figure 1C, Figure 2C, Figure 3C, and Table 3), subjects with BPD showed no response (Figure 3D and Table 3). Indeed, under low-glucose stress, lymphocytes of subjects with BPD had a number of individual mitochondrial transcripts that were down-regulated (Figure 1D), although the entire group of genes was not significantly shifted (Figure 3D).

Up-regulated transcripts in NC lymphocytes in low-glucose medium as compared with NC lymphocytes in normal-glucose medium had significant hits in the Gene Ontology database categories of mitochondrion (P = .002) and cytochrome c oxidase activity (P = .002), whereas down-regulated transcripts in BPD lymphocytes in low- glucose medium as compared with BPD lymphocytes in normal-glucose medium had a significant hit in the Gene Ontology database category of mitochondrion (P = .01). However, although the entire group of electron transfer transcripts was significantly shifted toward up-regulation in the NC lymphocytes under glucose deprivation stress (Figure 3C and Table 3), no significant shift was observed in the BPD lymphocytes under energy stress (Figure 3D and Table 3). Regulation trends were verified with qPCR (Figure 4). Four electron transfer transcripts that were used to verify the gene array data replicated the major patterns observed in the gene array analysis (Figure 4), although the levels of difference seen in the gene expression microarray study are at the threshold of detectability for qPCR. When the analysis was limited to paired samples (n = 13 for subjects with BPD, n = 7 for NCs; see Table 1 for pairs), 15 transcripts showed high between-group variability as determined in a factorial analysis of variance (eTable 2). These 15 transcripts were averaged and plotted (Figure 6). In BPD lymphocytes, these transcripts were down-regulated under low-glucose stress (P≤.003, paired t test), whereas in NC lymphocytes, these transcripts were up-regulated (P≤.02, paired t test). In the paired samples, a comparison of NC and BPD lymphocyte mRNA expression levels in low glucose showed that 17 transcripts were expressed significantly lower in BPD lymphocytes, similar to the larger sample. Finally, no significant relationship between electron transfer transcript expression and medication was found when mitochondrial expression levels obtained in the gene arrays were plotted against drug treatment in a hierarchical cluster analysis or when analyses of variance were calculated (each group of drug compared with absence of that drug in low and normal glucose) using qPCR data (data not shown).

To determine whether a shift between B and T cells had taken place in any of the comparisons, the expression levels of 54 B-cell–specific transcripts and 77 T-cell–specific transcripts were examined (Figures 7, 8, and 9; see eTable 3 for transcripts). The percentage of individually regulated genes did not surpass the chance expectations in any of the comparisons (Figure 7; see Figure 2 for chance expectations), and the group of B-cell–specific (Figure 8) and T-cell–specific (Figure 9) transcripts was not significantly shifted. In addition, 5 marker genes for natural killer lymphocytes and 5 marker genes for monocytes were unchanged in all of the comparisons. Sixteen marker genes for granulocytes were examined as well; however, most were under the detection limit and none were affected by any condition.

Lymphocytes of NCs responded to low-glucose stress with an up-regulation of nuclear transcripts of the mitochondrial respiratory chain. Individual transcripts were significantly up-regulated, and the group of all transcripts for proteins of the mitochondrial respiratory chain was shifted toward higher expression levels. Lymphocytes of patients with BPD did not have the same response. Indeed, a number of individual mitochondrial transcripts were down-regulated in BPD lymphocytes in response to low-glucose stress, a finding mirrored by the pairwise comparison, although the group of mitochondrial respiratory chain transcripts as a whole was not changed much. Because marker genes for B and T cells did not change in a similar manner, it is unlikely that the changes were owing to a shift between B and T cells.

Are lymphocytes a good model for the brain? Obviously, the best model tissue for psychiatric disorders is brain tissue. However, human brain tissue is not easily accessible, it does not permit experimental manipulations, and the results from molecular analyses can be adversely influenced by postmortem handling. Lymphocytes can be readily harvested, they are easier to handle than postmortem brain tissue, and their gene expression levels can be informative about gene expression in the central nervous system.¹⁴ Thus, they could reveal biomarkers of psychiatric disorders.¹⁰ Indeed, the lower expression of nuclear genes of the mitochondrial respiratory chain in lymphocytes in BPD, observed here under low-glucose stress, is similar to our previous findings in hippocampal tissue.⁹ Although the data in the hippocampus originally led us to hypothesize that patients with BPD have lower expression levels of these genes, the lymphocyte data suggest that NCs might have had an up-regulation in response to energy deprivation of the tissue during or immediately after death.

A potential confounding effect on gene expression in the patients with BPD is the exposure to psychotropic drugs. However, no single agent was present in more than 30% of all patients with BPD and medications were diverse, including lithium carbonate, valproic acid, anticonvulsants, antidepressants, and antipsychotics. We also found no significant relationship between electron transfer transcript expression and medication. Three aspects of our finding make a confounding effect of psychotropic drugs doubtful: (1) both the fresh (uncultured) lymphocytes and lymphocytes cultured in normal-glucose medium showed no difference between NCs and subjects with BPD; (2) in the pairwise comparison, the same medication profile is present in low and normal glucose samples; and (3) the lymphocytes in culture were cultured for 5 days in the absence of any drugs and had been washed 3 times before plating. The latter reasoning, however, is not particularly strong as it is well established that many psychotropic medications have a long half-life, at least in the brain.

In conclusion, we have found evidence of a functionally different response to glucose deprivation between subjects with BPD and NCs. This response is reflected in specific expression patterns of genes coding for electron transfer protein transcripts and suggests that the normal molecular adaptation to energy stress, present in NCs, is deficient in patients with BPD. No differences were observed in basal expression levels of electron transfer transcripts. Although we have only examined patients with BPD at this point, the findings raise hope that gene expression analysis in easily accessible tissue might provide helpful insights into molecular abnormalities in psychiatric disorders.

Correspondence: Christine Konradi, PhD, Vanderbilt University, Department of Psychiatry, Medical Research Building 3, Room 7158C, 465 21st Ave S, Nashville, TN 37232-8548 (christine.konradi@vanderbilt.edu).

Submitted for Publication: March 18, 2006; final revision received October 17, 2006; accepted October 18, 2006.

Author Contributions: Dr Konradi had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All of the statistical analyses were carried out by Dr Konradi.

Financial Disclosure: Dr Konradi and McLean Hospital have submitted a patent application on the findings presented here.

Funding/Support: This work was supported by a grant from Jim and Pat Poitras (Dr Konradi).

Additional Information:eTable 1, eTable 2, and eTable 3 are available.

Acknowledgment: We thank Stephan Heckers, MD, MSci, for help, Ian Greenhouse, BA, and Kang Sim, MD, for technical support, and Bruce Cohen, MD, PhD, for helpful discussions. Bing Zhang, PhD, and Cheng Li, PhD, gave advice on biostatistics.