Molecular Psychiatry (2007) 12, 1089–1102 & 2007 Nature Publishing Group All rights reserved 1359-4184/07 $30.00 www.nature.com/mp ORIGINAL ARTICLE Stress-induced changes in primate prefrontal profiles of expression AM Karssen1,6, S Her1,6,7,JZLi2, PD Patel3, F Meng3, WE Bunney Jr4, EG Jones5, SJ Watson3, H Akil3, RM Myers2, AF Schatzberg1 and DM Lyons1 1Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA; 2Stanford Genome Center and the Department of Genetics, Stanford University, Stanford, CA, USA; 3Molecular and Behavioral Neuroscience Institute and the Department of Psychiatry, University of Michigan, MI, USA; 4Department of Psychiatry and Human Behavior, University of California, Irvine, CA, USA and 5Center for Neuroscience, University of California, Davis, CA, USA

Stressful experiences that consistently increase cortisol levels appear to alter the expression of hundreds of in prefrontal limbic brain regions. Here, we investigate this hypothesis in monkeys exposed to intermittent social stress-induced episodes of hypercortisolism or a no-stress control condition. Prefrontal profiles of compiled from Affymetrix microarray data for monkeys randomized to the no-stress condition were consistent with microarray results published for healthy . In monkeys exposed to intermittent social stress, more genes than expected by chance appeared to be differentially expressed in ventromedial prefrontal cortex compared to monkeys not exposed to adult social stress. Most of these stress responsive candidate genes were modestly downregulated, including conjugation and ligases involved in synaptic plasticity, cell cycle progression and nuclear receptor signaling. Social stress did not affect gene expression beyond that expected by chance in dorsolateral prefrontal cortex or prefrontal white matter. Thirty four of 48 comparisons chosen for verification by quantitative real-time polymerase chain reaction (qPCR) were consistent with the microarray-predicted result. Furthermore, qPCR and microarray data were highly correlated. These results provide new insights on the regulation of gene expression in a prefrontal corticolimbic region involved in the pathophysiology of stress and major depression. Comparisons between these data from monkeys and those for ventromedial prefrontal cortex in humans with a history of major depression may help to distinguish the molecular signature of stress from other confounding factors in human postmortem brain research. Molecular Psychiatry (2007) 12, 1089–1102; doi:10.1038/sj.mp.4002095; published online 25 September 2007 Keywords: mood disorders; cortisol; hypothalamic-pituitary-adrenal axis; oligonucleotide micro- array; squirrel monkey

Introduction postmortem analysis of whole-genome microarray data.11–14 These findings offer potential new insights In major depression, dysregulation of the hypothala- for the discovery of stress-related genes involved in mic–pituitary–adrenal (HPA) axis is evinced in 40– the pathophysiology of depression,15,16 but must be 80% of patients by an increase in circulating levels of considered with caution. the stress hormone cortisol.1–5 Receptors for cortisol In studies of human postmortem tissue, terminal are densely expressed in prefrontal cortex6,7 where medical conditions affect gene expression throughout they function as transcription factors that regulate the brain17–20 and antidepressant medications alter gene expression.8–10 Hundreds of genes in prefrontal gene expression profiles in prefrontal cortical tissue.21 cortex appear to be differentially expressed in hu- Animal models that utilize rodents are informative22 mans with a history of major depression based on but not sufficient because prefrontal cortex and white matter differ in rodents compared to humans and 23–25 Correspondence: Dr DM Lyons, Psychiatry Neuroscience, Stan- other primates. Relative to rodents, receptors for ford University, 1201 Welch Rd, MSLS P104—Mail Code 5485, cortisol occur at higher levels in primate neocortex6,26 Stanford, CA 94305-5485, USA. and remarkably low levels of glucocorticoid receptor E-mail: [email protected] 27 6 are found in rhesus monkey hippocampus. Neuroi- These authors contributed equally to this work. maging evidence likewise implicates ventromedial 7Current address: Korea Basic Science Institute, Chuncheon, prefrontal cortex as a key region for understanding the South Korea. 28–32 Received 22 August 2006; revised 9 August 2007; accepted 10 biology of stress and depression. Here, we August 2007; published online 25 September 2007 investigate prefrontal profiles of gene expression in Social stress affects gene expression AM Karssen et al 1090 squirrel monkeys exposed to intermittent social no-stress condition. Half of the monkeys in each adult stress-induced episodes of hypercortisolism using condition were previously exposed to postnatal stress microarray technology designed for the human tran- or the postnatal no-stress control. Monkeys sampled scriptome. for cortisol determinations were not used to study gene expression to avoid confounding blood sample collection and adult social stress effects. A total of 14 Materials and methods serial blood samples were collected from each Experimental design monkey according to the following schedule: 1 day A total of 22 male squirrel monkeys (Saimiri sciureus) prior to the first and fifth social separations; 1, 3 and that were born and raised at the Stanford University 17 days during the first and fifth separation sessions Animal Research Facility were randomized to the and 1, 3 and 17 days after completion of the first and following two treatment conditions in adulthood at fifth separations during new pair formations. B9 years of age (range 7.2–10.6 years). In one Matched samples were collected at simultaneous condition, monkeys were exposed to six intermittent time points in the adult no-stress condition. All 38 social separations that each lasted 3 weeks in samples were collected as described elsewhere duration. During each social separation session, between 1530 and 1630 hours to control for diurnal 39 monkeys were individually housed and could see, variation. Cortisol levels were subsequently mea- hear, smell, but not touch other unfamiliar monkeys. sured in duplicate using a radioimmunoassay estab- 40 After each intermittent separation, new male pairs lished for squirrel monkey research. were subsequently formed and maintained for 9 weeks. New pair formations33 and social separations34 Brain tissue collection are known to increase cortisol levels in adult male Brain tissues were collected from six monkeys in each squirrel monkeys. In the no-stress control condition, of the two adult treatment conditions 12 weeks after adult monkeys were housed with the same male the end of the final social separation while all of the companion in stable same-sex pairs. monkeys were housed in same-sex pairs. In each of As part of other studies, half of the monkeys in each the two adult treatment conditions, 2–4 monkeys adult treatment condition were previously exposed to were previously exposed to either postnatal stress or postnatal stress or postnatal no-stress conditions the no-stress postnatal control. Adult monkeys were described elsewhere in detail.35 Randomization of anesthetized with an intramuscular injection of À1 monkeys to the adult social stress versus no-stress 10 mg kg ketamine, followed by euthanasia with an À1 conditions was stratified by prior postnatal condi- intravenous overdose of 120 mg kg pentobarbital. tions to provide similar size samples in the 2 Â 2 Craniotomies were performed, brains were removed, factorial design (Figure 1). Based on evidence that and the left and right cerebral hemispheres were postnatal experiences alter gene expression in the separated by a mid-sagittal incision. Left cerebral forebrain of rodents,36,37 postnatal and adult stress hemispheres were placed in a custom-designed effects were examined in the analysis of data from acrylic brain matrix and cut coronally into blocks monkeys. All procedures were conducted in accor- that were frozen in isopentane on dry ice at À40 1C. dance with the Animal Welfare Act, and were One block contained all prefrontal tissue from the approved by Stanford University’s Administrative anterior commissure to the frontal pole. All brain Panel on Laboratory Animal Care. tissue collections occurred between 0800 and 0930 hours to control for diurnal variation in gene expres- Longitudinal measures of cortisol sion profiles. Serial tissue sections cut 20 micrometers thick were Blood samples for cortisol determinations were thaw-mounted onto Superfrost Plus glass slides and collected from 6 of 12 monkeys exposed to adult stored at À80 1C. From each animal, a 1-in-10 series of social stress and 4 of 10 monkeys from the adult sections that contained white matter tissue extending from the anterior striatum to the frontal pole was randomly selected for dissection at 0 1C under a stereo-zoom microscope. Prefrontal tissue from three defined regions on each section was scraped into separate RNAase-free tubes for subsequent storage at À80 1C. The boundaries used to identify each region (Figure 2) were identical to those previously shown in squirrel monkey neuroimaging research to have high interobserver reliabilities, that is, intraclass correla- tion coefficients greater than 0.90.41

Microarray hybridization A total of 36 Affymetrix (Mountain View, CA, USA) Figure 1 Schematic representation of the 2 Â 2 factorial U133A 2.0. GeneChips were used to design. analyze gene expression profiles in ventromedial

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1091 dually analyzed and reassigned to new probe sets to ensure that all probes within a given set detect sequences within a single UniGene cluster from build 183. The corresponding UniGene cluster then served as the probe set identification name. In contrast to the default Affymetrix chip description file, which inter- rogates 12 757 unique Unigene clusters, the custom chip description file contained 11 839 unique Uni- gene probe sets. Scanned chip data were initially processed by MicroArray Suite 5 (MAS5, Affymetrix) to convert raw image files (.DAT) into probe signal value files (.CEL). Probe signal value files were subsequently processed using two different computational proto- Figure 2 Prefrontal regions on serial sections from the left cols: the robust multiarray average (RMA) protocol44 cerebral hemisphere of a hemisected squirrel monkey brain. available at http://www.bioconductor.org, and the dChip perfect match-only protocol45 available at http://biosun1.harvard.edu/complab/dchip. Log2 trans- prefrontal cortex, dorsolateral prefrontal cortex and formed expression values determined separately by prefrontal white matter samples from each of the 12 RMA and by dChip were imported into R software adult monkeys. No pooling of samples across differ- (http://www.r-project.org), normalized using quantile ent animals was necessary. To accommodate the normalization,46 and median-centered to correct for workload, tissue samples were processed in two technical batch-related variation.42 separate batches that each contained an equal number of randomly selected animals from each treatment condition. Tissue samples were homogenized with a Statistical analysis motorized pellet pestle, and total RNA from each To identify similarities and differences between the sample was extracted using TRIzol Reagent (Invitrogen, prefrontal regions of interest, unsupervised hierarch- Carlsbad, CA, USA) following the manufacturer’s ical cluster analysis and paired t-tests were performed instructions. Quantification was carried out by spec- with data collected from monkeys randomized to the trophotometric analysis, and gel electrophoresis was adult no-stress condition. Experience-dependent dif- used to verify the integrity of each RNA sample. All ferences in gene expression within each region were samples were amplified with the two-cycle method, then assessed using two-way analysis of variance reverse transcribed into cDNA, labeled with biotiny- (ANOVAs) with adult social stress versus no-stress lated nucleotides and hybridized to HG U133A 2.0 conditions stratified by prior postnatal stress versus microarrays at the Stanford University Genomic postnatal no-stress conditions in a 2 Â 2 factorial Facility according to the manufacturer’s instructions. design. Genes with significant postnatal or adult All 36 RNA samples were of high and comparable social stress main effects (P < 0.05) in both RMA and quality as determined by (i) the ratio of 28S:18S dChip data sets were considered to be differentially ribosomal RNAs, (ii) signal intensity ratios from expressed. Postnatal-by-adult social stress interac- probes for the 30 and 50 ends of glyceraldehyde-3- tions were also examined but our power to detect the phosphate dehydrogenase (GAPDH) and b-actin tran- interactions was less than that for either of the two scripts used as quality controls on Affymetrix micro- main effects. arrays and (iii) the absence of outlier microarrays In addition to analyzing individual genes, Gene Set determined using a color-coded correlation matrix Enrichment Analysis (GSEA) was used to evaluate described elsewhere.42 RMA and dChip data at the level of functional gene sets.47 Established gene sets were downloaded in December 2005 from the (GO) Microarray data processing database (ftp://ftp.geneontology.org/pub/go/godatabase/ The original design of the HG U133A 2.0 microarray archive/latest-termdb/). GO terms are nested cate- was based on information from UniGene build 133 gories that summarize known biological processes (April 2001 release). To account for recent advances and molecular functions associated with a given gene, in genomics, signal intensity data were interpreted by as well as its cellular location. GO terms for gene sets use of a custom chip description file (filename with fewer than 8 and more than 250 genes in our data U133Av2_HS_UG_5) based on UniGene build 183 were excluded from further consideration, leaving (April 2005 release) and freely available at http:// 900 gene sets for biological processes, 449 gene sets brainarray.mhri.med.umich.edu/Brainarray/Database/ for molecular functions and 207 gene sets for cellular CustomCDF/genomic_curated_CDF.asp. Design of location. For each of these gene sets, GSEA calculates the custom chip description file is described else- a weighted score that reflects the extent to which where in detail.43 Briefly, every probe within each members of the gene set occur toward the extremes of probe set on the U133A 2.0 microarray was indivi- the gene list rank-ordered by differential expression

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1092 between the conditions of interest. The rank-ordered gene list consisted of all probe sets detected as present on one or more of the microarrays that were used to interrogate each region of interest, and not just probe sets above an arbitrary cutoff in terms of differential expression. Statistical significance of the enrichment score for each gene set on the list rank- ordered by differential expression was determined by condition-based permutation tests with 1000 random permutations used to build the null distribution. This statistical approach preserves gene–gene correlations and provides a more accurate null model compared to other approaches developed to analyze GO informa- Figure 3 Venn diagram depicts the unique and over- tion.47 To minimize the frequency of false positive lapping subsets of genes called present in ventromedial results, only gene sets identified as significant prefrontal cortex (vmPFC), dorsolateral prefrontal cortex (P < 0.05) in both RMA and dChip data were con- (dlPFC) and prefrontal white matter (WM). Approximately sidered enriched. All test statistics from GSEA and half of the 11 839 screened genes were called present on at least one of the microarrays used to interrogate each region. the ANOVAs described above for the microarray data analysis were evaluated with two-tail probabilities.

Results Quantitative real-time polymerase chain reaction (qPCR) confirmation of microarray expression data A total of 36 microarrays were used to compile gene qPCR was used to confirm differential expression expression profiles for three prefrontal regions in 12 using the same RNA samples that were assayed by adult squirrel monkeys. Approximately half of the microarray. Samples were prepared using RNAase- 11 839 screened genes were called present by MAS5 free DNAase (Promega, Madison, WI, USA), and software on one or more of the microarrays used to control reactions with not reverse transcribed interrogate each region, that is, 5539 genes were RNA confirmed the absence of genomic DNA. Primers called present in ventromedial prefrontal cortex, 5501 were selected within or near the Affymetrix target genes in dorsolateral prefrontal cortex and 5625 genes sequence, and designed using Primer Express 2.0 in prefrontal white matter. Many of these genes were software (Applied Biosystems, Foster City, CA, expressed in all three regions of interest (Figure 3). USA). Primers that did not reliably produce single A total of 5389 genes were not called present on any products as indicated by the dissociation curve were array, and these were excluded from further consid- discarded. Random hexamer primed cDNA was made eration. Both RMA and dChip data sets for all from 500 ng total RNA using SuperScript II reverse genes called present in each region are available for transcriptase (Invitrogen) following the manufac- public use at http://www.pritzkerneuropsych.org/data/ turer’s instructions. The reaction was diluted 10-fold data.htm. with water and stored at À20 1C. At the second step, 2 ml diluted cDNA (from approximately 5 ng of Genes enriched in white matter compared to prefrontal starting RNA) was added to a 10 ml PCR reaction cortex containing 10 pmol each of forward and reverse Of the genes called present in prefrontal white matter, primers for each gene tested (Supplementary Table 792 were expressed at higher levels (P < 0.05) in white 1), and 5 mlof2Â SYBER GREEN PCR Master Mix matter compared to cortex in both RMA and dChip (Applied Biosystems, Warrington, UK), in a 384-well data sets for monkeys from the adult no-stress plate. Samples were amplified on an ABI 7900 condition. The percentage of genes differentially (Applied Biosystems, UK) for 40 cycles at 94 1C for expressed in RMA and confirmed by dChip was 15 s followed by fluorescence capture at 60 1C for 80%. Expression levels were, on average, 73% higher 1 min. The critical threshold or point at which signal in white matter compared to cortex for the 792 genes fluorescence exceeds background for each gene differentially expressed in both RMA and dChip data sample was compared against a 10-fold serial dilution sets (Supplementary Table 2). Many of the genes standard curve of cDNA control sample amplified in enriched in white matter are associated with myelina- parallel. The expression level for each gene was tion as determined by GSEA (Supplementary Table 3). subsequently normalized to the mean expression level of GAPDH and b-actin. To validate the micro- Genes enriched in cortex compared to prefrontal white array results, predicted directional differences in gene matter expression (up or downregulation) were assessed Of the genes called present in ventromedial and/or with the qPCR data using directional one-tail t-tests. dorsolateral prefrontal cortex, 1481 were expressed at The relationship between qPCR and microarray data higher levels (P < 0.05) in one or both regions of cortex was also evaluated with a Pearson’s product moment compared to prefrontal white matter in both RMA and correlation. dChip data sets for monkeys from the adult no-stress

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1093 condition. The percentage of genes differentially cortex included gene sets involved in neural plasti- expressed in RMA and confirmed by dChip was city, chromatin remodeling and transcriptional re- 83%. Expression levels were, on average, 42% higher pression (Table 1). Of the 19 overrepresented GO in cortex compared to white matter for the 1481 genes terms for dorsolateral prefrontal cortex, 10 included differentially expressed in both RMA and dChip data gene sets related to phosphatase activity (Supplemen- sets (Supplementary Table 4). Biological process and tary Table 8). molecular GO terms identified by GSEA as over- represented in cortex compared to prefrontal white Adult social stress-induced changes in gene expression matter included gene sets involved in energy-depen- In ventromedial prefrontal cortex, a significant adult dent proton transport, glycolysis, homeostasis, synap- social stress main effect (P < 0.05) was discerned for tic transmission and voltage-gated ion channel 440 genes in the data set from RMA, and 985 genes activity (Supplementary Table 5). Overrepresented were identified in the data set from dChip (Figure 5). cellular component GO terms included gene sets The number of genes in each data set expected to localized to the synapse and voltage-gated channels. show a significant main effect by chance is 277, that is, 0.05 Â 5539 genes called present in ventromedial Differences between regions of prefrontal cortex prefrontal cortex. The percentage of genes with a Genes enriched in cortex compared to white matter significant main effect for adult social stress in RMA were further examined with hierarchical clustering of and confirmed by dChip was 49%, that is, 216 genes the data for monkeys from the adult no-stress were differentially expressed in both data sets (Table 2). condition. Samples of prefrontal cortex from the same Far fewer genes in prefrontal white matter (22 genes) anatomical region in different monkeys tended to and dorsolateral prefrontal cortex (23 genes) showed a cluster into regional groupings (Figure 4). Paired t- significant adult social stress main effect in both RMA tests identified 477 genes that were consistently and dChip data sets (Figure 5). expressed in both RMA and dChip data at higher Most of the affected genes in ventromedial pre- levels (P < 0.05) in ventromedial versus dorsolateral frontal cortex (Table 2) were downregulated to a prefrontal cortex (Supplementary Table 6). Con- modest extent by exposure to adult social stress. For versely, 268 genes were expressed at higher levels the 171 genes diminished in both RMA and dChip (P < 0.05) in dorsolateral versus ventromedial prefron- data sets, expression levels were, on average, 16% tal cortex according to both RMA and dChip data lower in monkeys exposed to adult social stress (Supplementary Table 7). The percentage of genes compared to adult no-stress control. Downregulation differentially expressed in RMA and confirmed by of gene expression in ventromedial but not dorsolat- dChip data was 59%. Expression levels differed, on eral prefrontal cortex corresponds with our finding of average, by a modest 16% between ventromedial and regional differences in the capacity for transcriptional dorsolateral prefrontal cortex for the combined list of repression. As indicated above, genes involved in 745 genes differentially expressed in both RMA and transcriptional repression and chromatin remodeling dChip data sets. Biological process and molecular GO are overrepresented in ventromedial compared to terms identified by GSEA as overrepresented in dorsolateral prefrontal cortex for monkeys from the ventromedial compared to dorsolateral prefrontal adult no-stress control condition (Table 1). Fourteen of 21 biological process and molecular GO terms identified by GSEA in ventromedial prefrontal cortex as overrepresented in monkeys exposed to adult social stress compared to the no-stress condi- tion are involved in catabolism and ligase activity (Table 3). Overrepresented cellular component GO terms included gene sets localized to mitochondria. Of the core genes involved in catabolism and ligase activity discerned by GSEA, several ubiquitin con- jugation enzymes (that is, UBE2A, UBE2D2, UBE2E1) and ubiquitin ligases (that is, SMURF2, UBR2, UBE3A) were also identified by the above-mentioned two-way ANOVAs as having a significant adult social stress main effect in both RMA and dChip data sets.

Postnatal effects on gene expression Figure 4 Hierarchical cluster analysis of dorsolateral In ventromedial prefrontal cortex, a significant post- (dlPFC) and ventromedial (vmPFC) prefrontal cortical natal stress main effect (P < 0.05) was discerned for samples from six monkeys in the adult no-stress condition. The dendrogram topography is highly robust, and does not 237 genes in the data set from RMA, and 387 genes differ appreciably with either dChip (shown) or RMA data. were identified in the data set from dChip. As All 1481 genes enriched in prefrontal cortex compared to indicated above, the number of genes in each data prefrontal white matter were included in the cluster set expected to show a significant main effect by analysis. chance alone is 277. The percentage of genes in

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1094 Table 1 Biological, cellular and molecular GO terms identified in both RMA and dChip data as overrepresented in ventromedial compared to dorsolateral prefrontal cortex

Class Family Gene set GO term Number RMA, dChip, of genes P-value P-value

Biological Synapse GO:0043062 Extracellular structure 38 0.004 0.000 process organization and biogenesis GO:0048169 Regulation long-term neuronal 8 0.002 0.000 synaptic plasticity GO:0030198 Extracellular matrix organization 38 0.004 0.000 and biogenesis GO:0048168 Regulation of neuronal synaptic 14 0.000 0.011 plasticity GO:0050803 Regulation of synapse structure 25 0.000 0.038 and function GO:0007416 Synaptogenesis 14 0.046 0.018 GO:0048489 Synaptic vesicle transport 27 0.037 0.018 Axonogenesis GO:0000902 Cellular morphogenesis 223 0.000 0.009 GO:0048468 Cell development 154 0.027 0.014 GO:0048699 Neurogenesis 149 0.000 0.000 GO:0030182 Neuron differentiation 131 0.010 0.002 GO:0048666 Neuron development 105 0.026 0.009 GO:0007417 Central nervous system 102 0.022 0.021 development GO:0048667 Neuron morphogenesis during 85 0.030 0.020 differentiation GO:0031175 Neurite morphogenesis 85 0.030 0.020 GO:0007409 Axonogenesis 72 0.043 0.004 GO:0007420 Brain development 52 0.025 0.020 GO:0048675 Axon extension 8 0.045 0.016 Cell growth GO:0000902 Cellular morphogenesis 223 0.000 0.009 GO:0040007 Growth 110 0.000 0.000 GO:0040008 Regulation of growth 85 0.000 0.000 GO:0008361 Regulation of cell size 79 0.008 0.004 GO:0016049 Cell growth 78 0.008 0.004 GO:0001558 Regulation of cell growth 67 0.000 0.004 Other GO:0007599 Hemostasis 36 0.000 0.020 GO:0050878 Regulation of body fluids 45 0.000 0.000 GO:0016477 Cell migration 120 0.030 0.009 GO:0016568 Chromatin modification 79 0.015 0.029 GO:0006338 Chromatin remodeling 27 0.033 0.047 GO:0007186 G- coupled receptor 225 0.028 0.045 protein signaling GO:0015698 Inorganic anion transport 57 0.033 0.031 GO:0007269 Neurotransmitter secretion 35 0.021 0.019 GO:0046578 Regulation of Ras protein signal 10 0.018 0.029 transduction GO:0051046 Regulation of secretion 20 0.038 0.000 GO:0042060 Wound healing 47 0.011 0.009 Cellular Membrane GO:0019898 Extrinsic to membrane 31 0.025 0.008 component GO:0019897 Extrinsic to plasma membrane 27 0.029 0.022 Other GO:0005834 Heterotrimeric G-protein 18 0.020 0.018 complex Molecular Binding GO:0008092 Cytoskeletal protein binding 188 0.000 0.014 function GO:0042802 Protein self binding 76 0.000 0.007 GO:0003682 Chromatin binding 51 0.021 0.012 GO:0008201 Heparin binding 38 0.040 0.014 GO:0051015 Actin filament binding 16 0.049 0.045 Transcriptional GO:0016564 Transcriptional repressor activity 111 0.026 0.023 repression GO:0003714 Transcription corepressor 64 0.045 0.016 activity

Abbreviations: GO, gene ontology; RMA, robust multiarray average. All 6450 genes called present on at least one array are included in the analysis. The number of genes for each GO term is provided, and GO terms grouped into families are part of the same GO tree.

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1095 Adult social stress-induced episodes of hypercortisolism As expected from previous research, cortisol levels were consistently increased during the first and fifth separations compared to when the same adult monkeys were sampled in previous pairs (Figure 6). Intermittent social separations also appeared to sensitize the HPA-axis response, as cortisol levels were 70% greater 1 day after the fifth compared to the first separation (paired-t = 3.06, P = 0.028). During each adult social separation, cortisol levels returned to baseline then increased for a short time thereafter during the formation of new pairs. Cortisol levels in the adult no-stress condition did not differ signifi- cantly over time, and postnatal effects were not statistically significant.

Discussion

Figure 5 Venn diagrams depict the unique and over- In monkeys exposed to adult social stress-induced lapping subsets of genes discerned in robust multiarray episodes of hypercortisolism, more genes than ex- average (RMA) and dChip data as differentially expressed pected by chance appeared to be differentially (P < 0.05) after exposure to adult social stress compared to the expressed in ventromedial prefrontal cortex com- adult no-stress condition for three prefrontal regions. The pared to monkeys not exposed to adult social stress. numbers of genes differentially expressed after exposure to Most of these stress responsive candidate genes were adult social stress exceeds chance levels for ventromedial modestly downregulated, including ubiquitin conju- prefrontal cortex (vmPFC), but not dorsolateral prefrontal gation enzymes and ligases involved in synaptic cortex (dlPFC) or prefrontal white matter (WM). plasticity, cell cycle progression and nuclear receptor signaling. Adult social stress did not affect gene ventromedial prefrontal cortex with a significant expression beyond that expected by chance in postnatal stress main effect in RMA and confirmed dorsolateral prefrontal cortex or prefrontal white by dChip was 27%, that is, 65 genes were differen- matter. Prefrontal profiles of gene expression in tially expressed in both data sets. Only 15 genes in monkeys from the adult no-stress condition were prefrontal white matter and 59 genes in dorsolateral consistent, however, with microarray data from recent prefrontal cortex had a significant postnatal stress studies of healthy humans as discussed below. These main effect (P < 0.05) in both RMA and dChip data results demonstrate that the application of human sets. Postnatal stress did not robustly modify the arrays to nonhuman primates recovers information subsequent effects of adult social stress. Less than 5% from thousands of genes, and may represent a useful of the genes with a consistent adult stress main effect new strategy for understanding the biology of stress in showed a postnatal-by-adult stress interaction in both humans with major depression. RMA and dChip data sets. Human arrays used in monkey research do of course suffer from a loss of sensitivity and coverage qPCR validation of microarray data of the transcriptome because hybridization is im- To test the accuracy of the microarray data for genes paired by inter-species gene sequence differences. For identified by RMA and confirmed by dChip, 19 example, 46% of all probe sets were called present in regional pairwise comparisons (Table 4) and 13 adult squirrel monkey dorsolateral prefrontal cortex (Figure social stress main effects (Table 5) were re-assessed 3) compared to 59% in a similar size sample of with qPCR using the same RNA samples that were humans (unpublished data). Comparable differences assayed by the microarray. Twenty three of the 32 have been reported in humans and rhesus monkeys qPCR comparisons were consistent with the micro- for different tissue types.48,49 Observed species differ- array-predicted direction for a validation rate of 72%. ences may reflect genuine differences in gene expres- A total of 16 additional pairwise comparisons that sion, or methodological artifacts related to gene were not significantly different according to the sequence differences. However, when groups of microarray were also examined by qPCR, and 5 were samples are compared within a single species, gene found to differ significantly according to qPCR for a sequence differences consistently carry the same microarray false negative rate of 31%. Microarray data effect across all samples and no longer represent a often underestimated fold-difference scores (Tables 4 confounding factor.42 and 5), but gene expression differences measured by In regional comparisons of squirrel monkeys - the microarray were highly correlated with those that domized to the adult no-stress condition, 792 genes were measured by qPCR (r = 0.92, n = 48 comparisons, were expressed in both RMA and dChip data sets at P < 0.001). higher levels in prefrontal white matter compared to

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1096 Table 2 Genes identified in both RMA and dChip data as differentially expressed in ventromedial prefrontal cortex for monkeys exposed to adult social stress compared to the no-stress control

Downregulated genes Upregulated genes

PSMB3 À1.54 HSPA1A À1.18 UAP1 À1.14 MRPS31 À1.10 CXCL5 1.09 HRSP12 À1.52 DNAJB1 À1.18 HSD17B6 À1.14 B4GALT5 À1.10 KIAA0276 1.09 CHGA À1.30 COPS4 À1.18 TIMM9 À1.14 TOMM70A À1.10 HMGA2 1.09 REV1L À1.29 UBE2E1 À1.18 DNAJC13 À1.14 TXNL2 À1.10 FLJ14249 1.12 CRYAB À1.28 CASP7 À1.18 KIAA0261 À1.14 C20orf111 À1.10 MID2 1.12 CGI-116 À1.28 BPY2IP1 À1.18 ZCWCC1 À1.14 GLRX2 À1.10 PEMT 1.13 PCDH8 À1.27 PHLPP À1.18 RNF130 À1.14 ATP2C1 À1.10 FOXH1 1.13 KIAA0907 À1.26 NPY5R À1.18 ZNF281 À1.14 TSEN34 À1.10 LMO3 1.13 MCTS1 À1.26 SRPRB À1.18 RCHY1 À1.14 RANBP9 À1.10 FOXF2 1.13 SH2BP1 À1.26 MRPL48 À1.18 SRP9 À1.14 UCRC À1.09 SMARCD1 1.13 BCOR À1.25 SCG3 À1.18 B3GALT2 À1.14 TXNDC4 À1.09 DNM1 1.14 SURB7 À1.25 SPRY2 À1.18 ABHD6 À1.14 SRP54 À1.09 CIDEC 1.15 TMEM23 À1.24 INPP5F À1.17 GTL3 À1.13 HTATSF1 À1.09 COCH 1.15 EDG1 À1.24 OSBPL8 À1.17 DPM3 À1.13 LIN7C À1.09 SDPR 1.15 ZNF638 À1.24 RAI2 À1.17 C12orf29 À1.13 YES1 À1.09 SFRS5 1.17 PER2 À1.24 RB1CC1 À1.17 PAPOLA À1.13 EIF2AK3 À1.09 RPS6KA4 1.17 CGI-12 À1.24 C6orf80 À1.17 NMT1 À1.13 TNFRSF11B À1.09 FLJ12355 1.20 TOMM20 À1.23 GLRB À1.17 COPB À1.13 TRAPPC4 À1.09 CUL2 1.20 RBP4 À1.23 UBE3C À1.17 SDCCAG1 À1.13 CREBL2 À1.09 PSME1 1.20 GUK1 À1.23 NR3C2 À1.17 SLC6A14 À1.13 SCYL3 À1.08 FLJ10579 1.22 PRPS2 À1.22 CBX1 À1.17 C20orf24 À1.13 UBE2D2 À1.08 DPP6 1.22 UQCR À1.22 KCNK1 À1.17 RAB7L1 À1.13 LAMB1 À1.08 COG2 1.22 ATP5O À1.22 SPRED2 À1.17 PHGDH À1.13 GABRA5 À1.08 TARBP1 1.23 CCK À1.22 RANBP6 À1.17 CCS À1.13 TXNDC9 À1.08 UBE2H 1.24 EIF3S12 À1.22 RCN2 À1.16 RNF6 À1.12 SERPINI1 À1.08 HNRPL 1.25 DDA3 À1.22 RIN2 À1.16 GABARAPL2 À1.12 CYP2B6 À1.07 ABI2 1.25 GGH À1.21 ATP10D À1.16 MRPL18 À1.12 UBE3A À1.07 POLD3 1.26 PLCL2 À1.21 FABP7 À1.16 UPF2 À1.12 NAB1 À1.07 HNT 1.26 PKIG À1.21 CREBBP À1.16 TMEM4 À1.12 MBD4 À1.07 STRAP 1.28 DBI À1.21 APG5L À1.16 NDUFB3 À1.11 CAMSAP1L1 À1.06 TPRT 1.28 GARNL4 À1.21 GOLGA4 À1.16 ABCD3 À1.11 ITSN1 À1.06 FLJ13576 1.29 SAP30 À1.20 CTCF À1.16 PPP1R16B À1.11 FLJ23560 À1.06 RAB3IL1 1.29 CAPZA1 À1.20 KIAA0103 À1.16 FNDC3A À1.11 TNFAIP1 À1.06 CXorf9 1.31 CGI-30 À1.20 PSMA2 À1.16 ZNF410 À1.11 PSMD14 À1.05 SLPI 1.32 DNAJA1 À1.20 C18orf9 À1.16 RIF1 À1.11 UBR2 À1.04 PHEMX 1.32 SLC22A17 À1.20 JAK1 À1.16 C18orf8 À1.11 MOBK1B À1.04 MEN1 1.33 SCRG1 À1.19 SPARC À1.15 PSMB1 À1.11 CD53 1.36 SNN À1.19 ATP5E À1.15 CCDC6 À1.11 ASH2L 1.37 PTPN20 À1.19 APG3L À1.15 ATRN À1.11 GSTP1 1.37 FAIM2 À1.19 ARHGAP12 À1.15 KIAA0431 À1.11 ZBTB5 1.40 CENTB2 À1.19 GAS2L1 À1.15 MDS009 À1.11 C3orf18 1.47 COPS2 À1.19 COPB2 À1.15 PTGS2 À1.11 VPS52 1.47 RSU1 À1.19 NT5C2 À1.15 ARFGAP1 À1.11 HOXA6 1.50 SMURF2 À1.18 PTPNS1 À1.15 MRPS22 À1.11 GPX3 1.54 MRPL17 À1.18 NIFUN À1.14 UBE2A À1.11 ZNF175 1.74

Abbreviations: GO, gene ontology; RMA, robust multiarray average. Fold-difference scores averaged across RMA and dChip data are presented for each gene.

prefrontal cortex (Supplementary Table 2). More than Additional evidence that human arrays recover 550 of these genes were similarly classified as white valid information from monkeys was obtained with matter genes in an independent microarray study of GSEA, which instead of analyzing individual genes, human prefrontal tissue.50 Conversely, we identified evaluates transcriptional profiles at the level of 1481 genes expressed at higher levels in one or both functional gene sets.47 As expected, 7 of 8 biological regions of prefrontal cortex compared to prefrontal process gene sets identified by GSEA as over- white matter (Supplementary Table 4). Of these genes, represented in prefrontal white matter compared 1251 were previously examined in healthy humans,50 to prefrontal cortex are involved in myelination and 80% were classified as enriched in human and (Supplementary Table 3). Gene sets involved in monkey prefrontal cortex. phosphatase activity are overrepresented in dorsolat-

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1097 Table 3 Biological, cellular and molecular GO terms identified in ventromedial prefrontal cortex as overrepresented in monkeys exposed to adult social stress compared to the no-stress control

Class Family Gene set GO term Number RMA, dChip, of genes P-value P-value

Biological Catabolism GO:0044248 Cellular catabolism 252 0.000 0.013 process GO:0006512 Ubiquitin cycle 223 0.026 0.021 GO:0009057 Macromolecule catabolism 186 0.005 0.022 GO:0044265 Cellular macromolecule 170 0.002 0.009 catabolism GO:0043285 Biopolymer catabolism 131 0.030 0.037 GO:0030163 Protein catabolism 127 0.029 0.036 GO:0044257 Cellular protein catabolism 111 0.028 0.023 GO:0051603 Proteolysis in cellular protein 110 0.029 0.017 catabolism GO:0019941 Modification-dependent protein 93 0.022 0.005 catabolism GO:0006511 Ubiquitin-dependent protein 91 0.024 0.005 catabolism Other GO:0006457 112 0.000 0.028 GO:0006631 Fatty acid metabolism 78 0.000 0.024 GO:0009260 Ribonucleotide biosynthesis 49 0.028 0.034 GO:0006986 Response to unfolded protein 32 0.025 0.044 GO:0030323 Respiratory tube development 17 0.010 0.019 Cellular Mitochondrium GO:0031975 Envelope 217 0.003 0.016 component GO:0031967 Organelle envelope 216 0.003 0.016 GO:0005740 Mitochondrial envelope 151 0.016 0.037 GO:0031966 Mitochondrial membrane 145 0.012 0.037 GO:0031980 Mitochondrial lumen 43 0.031 0.047 GO:0005759 Mitochondrial matrix 43 0.031 0.047 Other GO:0005794 Golgi apparatus 246 0.015 0.040 GO:0012505 Endomembrane system 180 0.011 0.000 GO:0005795 Golgi stack 170 0.006 0.023 GO:0048500 Signal recognition particle 8 0.015 0.000 GO:0005786 Signal recognition particle 8 0.015 0.000 GO:0000123 Histone acetyltransferase 8 0.026 0.044 complex Molecular Ligase activity GO:0016874 Ligase activity 211 0.000 0.017 function GO:0016879 Ligase activity, forming C-N 156 0.038 0.019 bonds GO:0016881 Acid-amino acid ligase activity 140 0.012 0.010 GO:0004842 Ubiquitin-protein ligase activity 135 0.020 0.011 Other GO:0019899 binding 110 0.048 0.046 GO:0051082 Unfolded protein binding 82 0.021 0.031

Abbreviations: GO, gene ontology; RMA, robust multiarray average. All 6450 genes called present on at least one array were included in the analysis. The number of genes for each GO term is provided, and GO terms grouped into families are part of the same GO tree. eral compared to ventromedial prefrontal cortex (Figure 5). Exogenous glucocorticoids likewise down- (Supplementary Table 8). Phosphatase activity has regulate the majority of glucocorticoid responsive been linked to a core cognitive function of dorsolat- genes in rodent hippocampus.8,55,56 eral prefrontal cortex, that is, working memory.51 In monkey ventromedial prefrontal cortex, a subset Ventromedial prefrontal cortex mediates emotional of related genes differentially expressed after expo- and autonomic responses to psychological stress.52–54 sure to adult social stress was reliably identified in Gene sets involved in transcriptional repression are dChip and RMA data. These genes are ubiquitin overrepresented in ventromedial compared to dorso- conjugation enzymes (UBE2A, UBE2D2, UBE2E1) and lateral prefrontal cortex (Table 1). This regional ligases (SMURF2, UBR2, UBE3A) that target ubiquitin difference in the capacity for transcriptional repres- to protein substrates for selective degradation by the sion corresponds with our finding that adult social proteasome. Ubiquitin proteasome pathways regulate stress-induced hypercortisolism downregulates brain involved in synaptic plasticity,57 cell the expression of hundreds of genes in ventromedial, cycle progression58 and diverse aspects of nuclear but not dorsolateral, prefrontal cortex in monkeys receptor signaling.59,60 In addition to proteolysis of

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1098 Table 4 Microarray and qPCR fold-difference scores for pairwise comparisons of vmPFC, dlPFC and prefrontal WM

Gene name (symbol) vmPFC versus WM dlPFC versus WM dlPFC versus vmPFC

Array qPCR Array qPCR Array qPCR

Apolipoprotein H (APOH) 12.98* 24.28w 9.68* 18.35w À1.34* À1.32w Cholecystokinin (CCK) À3.31* À2.30w À3.01* À1.53w 1.10* 1.50w 11-b-Hydroxysteroid dehydrogenase1 (HSD11B1) 5.86* 8.50w 3.93* 5.31w À1.49* À1.60 Neurofilament (NEFH) À2.37* 2.25 À5.61* 1.34 À2.35* À1.68w Protocadherin 8 (PCDH8) À7.44* À8.17w À4.58* À3.71w 1.61* 2.20w Neurofascin (NFASC) 2.95* 3.03w 3.07* 6.78w ND Glucocorticoid receptor (NR3C1) ND ND À1.17* 2.97 Suppression of tumorigenicity 18 (ST18) ND 9.22* 19.27w ND

Abbreviations: dlPFC, dorsolateral prefrontal cortex; qPCR, quantitative real-time polymerase chain reaction; vmPFC, ventromedial prefrontal cortex; WM, white matter. Regional differences in gene expression initially identified by microarray with two-tail tests (*P < 0.05) were subsequently assessed by qPCR comparisons with directional one-tail t-tests (wP < 0.05). ND indicates a regional comparison that was not significantly different as determined by the microarray.

Table 5 Microarray and qPCR fold-difference scores for to play a role in the pathophysiology of stress and adult social stress-induced downregulation of gene express- depression.62,63 sion in ventromedial prefrontal cortex Despite indications that ventromedial prefrontal cortex is involved in depression,28–30 microarray Gene name (symbol) Fold difference studies of humans have not examined this prefrontal region. In humans with a history of major depres- Array qPCR sion, whole-genome expression analysis has focused on dorsolateral prefrontal cortex (Brodmann’s area Chromogranin A (CHGA) À1.30* À2.09 11,12,14,64 64 w 8/9), ventrolateral prefrontal cortex (BA47), Protocadherin 8 (PCDH8) À1.27* À1.87 orbital frontal cortex (BA11),14 anterior cingulate Malignant T-cell amplified sequence 1 À1.26* À1.75 11,12 13 (MCTS1) (BA24) and the frontal pole (BA10). The ven- Period homolog 2 (PER2) À1.24* À1.37 tromedial prefrontal region examined in monkeys is Translocase of outer mitochondrial À1.23* À1.38 primarily comprised of agranular cortex lacking a membrane 20 homolog (TOMM20) well-developed granular layer IV,65,66 and roughly Cholecystokinin (CCK) À1.22* À1.51w corresponds to BA25, BA12 and medial portions of Fas apoptotic inhibitory molecule 2 À1.19* À1.64 BA11 in human prefrontal cortex based on neuro- (FAIM2) w psychological functions and anatomical connec- COP9 constitutive photomorphogenic À1.18* À1.71 tions.23,67–71 homolog subunit 4 (COPS4) Microarray studies of prefrontal cortex in humans Retinoic acid-induced 2 (RAI2) À1.17* À2.00w w with a history of major depression have identified Ubiquitin protein ligase E3C (UBE3C) À1.17* À1.68 w abnormalities in gamma-aminobutyric acid (GABA) Mineralocorticoid receptor (NR3C2) À1.17* À1.84 11 14 UDP-N-acteylglucosamine À1.14* À1.77w signaling and catabolism of polyamines. In pyrophosphorylase 1 (UAP1) monkeys exposed to adult social stress, three Mitochondrial ribosomal protein S22 À1.11* À1.78w GABA-related genes are diminished in ventromedial (MRPS22) prefrontal cortex compared to monkeys from the no- stress condition (Table 6). The spermidine/spermine Abbreviation: qPCR, quantitative real-time polymerase N1-acetyltransferase (SSAT) candidate gene for suici- chain reaction. dal depression14 is likewise diminished in monkeys Stress-induced downregulation of gene expression initially exposed to adult social stress, based on directional determined by microarray with two-tail tests (*P < 0.05) one-tail tests of RMA and dChip data. SSAT is the were subsequently assessed by qPCR using directional one- w rate-limiting enzyme in the catabolism of polyamines, tail tests ( P < 0.05). and is involved in the polyamine stress response evoked by emotional stimuli or exogenous gluco- nuclear receptors, ubiquitin conjugation enzymes and corticoids.72 Also in keeping with our findings in ligases also function as co-regulatory factors in monkeys (Table 6), diminished expression of ubiqui- nuclear receptor signaling and chromatin remodel- tin-related genes has been observed in orbital frontal ing.59,61 These findings are of interest because cortisol and not dorsolateral prefrontal cortex in humans with effects are mediated by two nuclear receptors and alternating episodes of mania and depression in dysregulation of this dual receptor system is thought bipolar disorder.73 Differential expression of prefron-

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1099

Figure 6 Longitudinal measures of cortisol in adult monkeys exposed to intermittent social stress. Plasma cortisol levels are presented for six monkeys sampled during the first (left panel) and fifth (right panel) recurrent sessions of social separation and new pair formation (mean7s.e.m.). Asterisks signify cortisol levels that are greater than the levels in previous pairs (*P < 0.05; **P < 0.01).

Table 6 Microarray findings from monkeys and relevant postmortem studies of prefrontal cortex in humans with mood disorders

Stress responsive genes in Findings from studies of humans Reference monkey prefrontal cortex

GABA signaling Dysregulated prefrontal expression of GABA-related Choudary et al.11 GABAR5, GABARAPL2, DBI genes in depression Polyamine catabolism Diminished prefrontal SSAT in depressed suicide Sequeira et al.14 SSAT completers Ubiquitin enzymes Diminished prefrontal expression of ubiquitin-related Ryan et al.73 UBE2A, UBE2D2, UBE2E1, genes in bipolar disorder SMURF2, UBR2, UBE3A Neuropeptides Dysregulation of CCK, NPY and related receptors in Harro;74 Caberlotto and Hurd;75 CCK, NPY5R bipolar disorder, depression and suicide Kuromitsu et al.76 Transcription factors Dysregulation of transcription factors in Xing et al.;77 Iwamoto et al.;13 CREBBP, CREBL2, SAP30, depression and bipolar disorder Yamada et al.78 NR3C2

Abbreviations: CCK, cholecystokinin; GABA, gamma-aminobutyric acid; NPY, neuropeptide; SSAT, spermidine/spermine N1-acetyltransferase. tal cholecystokinin (CCK), neuropeptide Y, and regions in humans and chimpanzees.50,80–83 In ro- related receptors in bipolar disorder, depression and dents, broad regional differences in gene expression suicide74–76 corresponds with our finding that stress have been reported in carefully controlled and affects these same neuropeptide systems in primate adequately powered studies of the hippocampus.84–88 prefrontal cortex (Table 6). Stress-induced regulation Our study examined prefrontal tissue of the highest of transcription factors in monkey ventromedial possible quality, collected at the same time of day, prefrontal cortex (Table 6) likewise concurs with within a small age range from adult monkeys born findings from human postmortem brain re- and raised in the same research facility. Cluster search,13,77,78 and with new advances in understand- analysis clearly identified differences between mon- ing the molecular mechanisms of antidepressant key dorsolateral and ventromedial prefrontal profiles drugs.79 of gene expression in both RMA and dChip data sets These results should be interpreted in the context (Figure 4). of several potential limitations. The sample of Cluster analysis also identified adult social stress monkeys is entirely male, and the reported gene effects in ventromedial prefrontal cortex for dChip but expression profiles may or may not hold true for not RMA data (dendrograms not presented). This females. Moreover, the regional differences discerned discrepancy reflects our finding that fewer genes are between monkey dorsolateral and ventromedial pre- differentially expressed in both RMA and dChip for frontal profiles of gene expression (Supplementary adult social stress effects compared to the regional Tables 6–8) are at odds with microarray surveys that comparisons. That regional comparisons are more failed to find broad differences between neocortical robust than those for adult stress effects is consistent

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1100 with the fold-difference scores we determined for these data from monkeys and those for ventromedial stress responsive genes (Table 2). Many of the genes prefrontal cortex in humans with a history of major expressed in brain tissue occur at low levels55 or are depression and related psychiatric disorders may restricted to subpopulations of cells.88 Fold-difference help to distinguish the molecular signature of stress scores are therefore ‘diluted’ in microarray studies of from other confounding factors in human postmortem bulk brain tissue collected by supra-cellular dissec- brain research. tions.42 Furthermore, microarrays tend to underesti- mate fold-difference scores relative to qPCR based on the 48 pairwise comparisons that we examined in Acknowledgments monkeys. Small fold-difference scores commonly occur in microarray studies of human brain tissue This work was funded by Public Health Service Grant and peripheral blood cells in which psychiatric case- MH47573 and the Pritzker Neuropsychiatric Dis- control comparisons rarely exceed 2-fold thresh- orders Research Consortium, which is supported by olds.11–15,64,73,89 the Pritzker Neuropsychiatric Disorders Research Microarrays screen thousands of genes and many Fund LLC. A shared intellectual property agreement false positive type I errors are expected by chance. exists between the Pritzker Neuropsychiatric Disor- Procedures for controlling False Discovery Rates ders Research Fund LLC and the University of (FDR) are increasingly used in microarray studies Michigan, the University of California and Stanford designed to test one hypothesis per gene, but FDR University to encourage the development of appro- procedures have not been well characterized for priate findings for research and clinical applications. factorial microarray experiments that test multiple effects. An alternative approach was therefore applied based on evidence that different methods of calculat- References 42 ing gene expression may provide different results. 1 Heuser I, Yassouridis A, Holsboer F. The combined dexametha- To minimize false positives, only genes or functional sone/CRH test: a refined laboratory test for psychiatric disorders. gene sets identified as nominally significant (P < 0.05) J Psychiatr Res 1994; 28: 341–356. in both RMA and dChip data were considered as 2 Carroll BJ, Feinberg M, Greden JF, Tarika J, Albala AA, Haskett RF et al. A specific laboratory test for the diagnosis of melancholia. having potential biological relevance. To test the Standardization, validation, and clinical utility. Arch Gen Psy- accuracy of this approach, 48 pairwise comparisons chiatry 1981; 38: 15–22. were re-assessed by qPCR using the same RNA 3 Rubin RT, Poland RE, Lesser IM, Winston RA, Blodgett AL. samples assayed by microarray. Thirty four of the 48 Neuroendocrine aspects of primary endogenous depression. I. Cortisol secretory dynamics in patients and matched controls. qPCR comparisons were consistent with the micro- Arch Gen Psychiatry 1987; 44: 328–336. array-predicted result, including CCK, NR3C2 and the 4 Plotsky PM, Owens MJ, Nemeroff CB. Psychoneuroendocrinology recently discovered90 UBE3C (Tables of depression. Hypothalamic-pituitary-adrenal axis. Psychiatr Clin 4 and 5). Microarray and qPCR data were also highly North Am 1998; 21: 293–307. correlated. Permutation analysis of our microarray 5 Murphy BE. Steroids and depression. J Steroid Biochem Mol Biol 91 1991; 38: 537–559. data using procedures described elsewhere suggests 6 Patel PD, Lopez JF, Lyons DM, Burke S, Wallace M, Schatzberg AF. that more than 85% of the 216 stress responsive genes Glucocorticoid and mineralocorticoid receptor mRNA expression identified by dChip and confirmed by RMA in in squirrel monkey brain. J Psychiatr Res 2000; 34: 383–392. ventromedial prefrontal cortex (Table 2) are not false 7 Webster MJ, Knable MB, O’Grady J, Orthmann J, Weickert CS. Regional specificity of brain glucocorticoid receptor mRNA positive type I errors due to chance alone. Never- alterations in subjects with schizophrenia and mood disorders. theless, microarrays primarily provide exploratory Mol Psychiatry 2002; 7: 985–994. data and many specific candidate genes must be 8 Morsink MC, Steenbergen PJ, Vos JB, Karst H, Joels M, De Kloet ER independently confirmed. et al. Acute activation of hippocampal glucocorticoid receptors In summary, we emphasize the value of this study results in different waves of gene expression throughout time. J Neuroendocrinol 2006; 18: 239–252. as a model of stress rather than hypercortisolism or 9 Meijer OC, van der Laan S, Lachize S, Steenbergen PJ, de Kloet ER. depression per se. Certain key features of depression Steroid receptor coregulator diversity: what can it mean for the are probably unique to humans, for example, low self- stressed brain? Neuroscience 2006; 138: 891–899. esteem, feelings of worthlessness, and suicidal idea- 10 Chrousos GP, Kino T. Intracellular glucocorticoid signaling: a formerly simple system turns stochastic. Sci STKE 2005; 2005: tion. Furthermore, despite indications that hypercor- pe48. tisolism may be a marker of severity or depressive 11 Choudary PV, Molnar M, Evans SJ, Tomita H, Li JZ, Vawter MP subtypes,1,2,92 dysregulation of the HPA-axis is not a et al. Altered cortical glutamatergic and GABAergic signal formal feature of depression in DSM-IV. Although the transmission with glial involvement in depression. Proc Natl effects of adult social stress in monkeys are consistent Acad Sci USA 2005; 102: 15653–15658. 12 Evans SJ, Choudary PV, Neal CR, Li JZ, Vawter MP, Tomita H et al. with evidence that cortisol regulates hundreds of Dysregulation of the fibroblast growth factor system in major 8–10 genes in certain brain tissue types, our study does depression. Proc Natl Acad Sci USA 2004; 101: 15506–15511. not establish that gene expression changes are 13 Iwamoto K, Kakiuchi C, Bundo M, Ikeda K, Kato T. Molecular directly caused by hypercortisolism. Other physiolo- characterization of bipolar disorder by comparing gene expression profiles of postmortem brains of major mental disorders. Mol gical adaptations induced by adult stress may have Psychiatry 2004; 9: 406–416. contributed to the changes discerned in ventromedial 14 Sequeira A, Gwadry FG, Ffrench-Mullen JM, Canetti L, Gingras Y, prefrontal gene expression. Comparisons between Casero Jr RA et al. Implication of SSAT by gene expression and

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1101 genetic variation in suicide and major depression. Arch Gen 36 Weaver IC, Meaney MJ, Szyf M. Maternal care effects on the Psychiatry 2006; 63: 35–48. hippocampal transcriptome and anxiety-mediated behaviors in 15 Bunney WE, Bunney BG, Vawter MP, Tomita H, Li J, Evans SJ et al. the offspring that are reversible in adulthood. Proc Natl Acad Sci Microarray technology: a review of new strategies to discover USA 2006; 103: 3480–3485. candidate vulnerability genes in psychiatric disorders. Am J 37 Rampon C, Jiang CH, Dong H, Tang YP, Lockhart DJ, Schultz PG Psychiatry 2003; 160: 657–666. et al. Effects of environmental enrichment on gene expression in 16 Wong ML, Licinio J. Research and treatment approaches to the brain. Proc Natl Acad Sci USA 2000; 97: 12880–12884. depression. Nat Rev Neurosci 2001; 2: 343–351. 38 Lyons DM, Mendoza SP, Mason WA. Psychosocial and hormonal 17 Li JZ, Vawter MP, Walsh DM, Tomita H, Evans SJ, Choudary PV et aspects of hierarchy formation in groups of male squirrel monkeys. al. Systematic changes in gene expression in postmortem human Am J Primatol 1994; 32: 109–122. brains associated with tissue pH and terminal medical conditions. 39 Zeitzer JM, Buckmaster CL, Parker KJ, Hauck CM, Lyons DM, Hum Mol Genet 2004; 13: 609–616. Mignot E. Circadian and homeostatic regulation of hypocretin in a 18 Tomita H, Vawter MP, Walsh DM, Evans SJ, Choudary PV, Li J et al. primate model: implications for the consolidation of wakefulness. Effect of agonal and postmortem factors on gene expression J Neurosci 2003; 23: 3555–3560. profile: quality control in microarray analyses of postmortem 40 Lyons DM, Ha CM, Levine S. Social effects and circadian rhythms human brain. Biol Psychiatry 2004; 55: 346–352. in squirrel monkey pituitary-adrenal activity. Horm Behav 1995; 19 Mexal S, Berger R, Adams CE, Ross RG, Freedman R, Leonard S. 29: 177–190. Brain pH has a significant impact on human postmortem 41 Lyons DM, Afarian H, Schatzberg AF, Sawyer-Glover A, Moseley hippocampal gene expression profiles. Brain Res 2006; 1106: ME. Experience-dependent asymmetric variation in primate 1–11. prefrontal morphology. Behav Brain Res 2002; 136: 51–59. 20 Vawter MP, Tomita H, Meng F, Bolstad B, Li J, Evans S et al. 42 Karssen AM, Li JZ, Her S, Patel PD, Meng F, Evans SJ et al. Mitochondrial-related gene expression changes are sensitive to Application of microarray technology in primate behavioral agonal-pH state: implications for brain disorders. Mol Psychiatry neuroscience research. Methods 2006; 38: 227–234. 2006; 11: 663–679. 43 Dai M, Wang P, Boyd AD, Kostov G, Athey B, Jones EG et al. 21 Palotas M, Palotas A, Puskas LG, Kitajka K, Pakaski M, Janka Z Evolving gene/transcript definitions significantly alter the et al. Gene expression profile analysis of the rat cortex following interpretation of GeneChip data. Nucleic Acids Res 2005; treatment with imipramine and citalopram. Int J Neuropsycho- 33: e175. pharmacol 2004; 7: 401–413. 44 Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B, Speed TP. 22 Panksepp J. Emotional endophenotypes in evolutionary psychia- Summaries of Affymetrix GeneChip probe level data. Nucleic try. Prog Neuropsychopharmacol Biol Psychiatry 2006; 30: Acids Res 2003; 31: e15. 774–784. 45 Li C, Wong WH. Model-based analysis of oligonucleotide arrays: 23 Preuss TM. Do rats have prefrontal cortex? The Rose-Woolsey- expression index computation and outlier detection. Proc Natl Akert program reconsidered. J Cognitive Neurosci 1995; 7: 1–24. Acad Sci USA 2001; 98: 31–36. 24 Uylings HB, Groenewegen HJ, Kolb B. Do rats have a prefrontal 46 Bolstad BM, Irizarry RA, Astrand M, Speed TP. A comparison cortex? Behav Brain Res 2003; 146: 3–17. of normalization methods for high density oligonucleotide 25 Schenker NM, Desgouttes AM, Semendeferi K. Neural connectiv- array data based on variance and bias. Bioinformatics 2003; 19: ity and cortical substrates of cognition in hominoids. J Hum Evol 185–193. 2005; 49: 547–569. 47 Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, 26 Watzka M, Bidlingmaier F, Beyenburg S, Henke RT, Clusmann H, Gillette MA et al. Gene set enrichment analysis: a knowledge- Elger CE et al. Corticosteroid receptor mRNA expression in the based approach for interpreting genome-wide expression profiles. brains of patients with epilepsy. Steroids 2000; 65: 895–901. Proc Natl Acad Sci USA 2005; 102: 15545–15550. 27 Sanchez MM, Young LJ, Plotsky PM, Insel TR. Distribution of 48 Uddin M, Wildman DE, Liu G, Xu W, Johnson RM, Hof PR et al. corticosteroid receptors in the rhesus brain: relative absence of Sister grouping of chimpanzees and humans as revealed by glucocorticoid receptors in the hippocampal formation. J Neurosci genome-wide phylogenetic analysis of brain gene expression 2000; 20: 4657–4668. profiles. Proc Natl Acad Sci USA 2004; 101: 2957–2962. 28 Mayberg HS, Lozano AM, Voon V, McNeely HE, Seminowicz D, 49 Wang Z, Lewis MG, Nau ME, Arnold A, Vahey MT. Identification Hamani C et al. Deep brain stimulation for treatment-resistant and utilization of inter-species conserved (ISC) probesets on depression. Neuron 2005; 45: 651–660. Affymetrix human GeneChip platforms for the optimization of 29 Pizzagalli DA, Oakes TR, Fox AS, Chung MK, Larson CL, the assessment of expression patterns in nonhuman primate (NHP) Abercrombie HC et al. 2004 Functional but not structural samples. BMC Bioinformatics 2004; 5:165. subgenual prefrontal cortex abnormalities in melancholia. Mol 50 Erraji-Benchekroun L, Underwood MD, Arango V, Galfalvy H, Psychiatry 2004; 9: 393–405. Pavlidis P, Smyrniotopoulos P et al. Molecular aging in human 30 Drevets WC, Ongur D, Price JL. Neuroimaging abnormalities in the prefrontal cortex is selective and continuous throughout adult life. subgenual prefrontal cortex: implications for the pathophysiology Biol Psychiatry 2005; 57: 549–558. of familial mood disorders. Mol Psychiatry 1998; 3: 220–226. 51 Runyan JD, Moore AN, Dash PK. A role for prefrontal calcium- 31 Urry HL, van Reekum CM, Johnstone T, Kalin NH, Thurow ME, sensitive protein phosphatase and kinase activities in working Schaefer HS et al. Amygdala and ventromedial prefrontal cortex memory. Learn Mem 2005; 12: 103–110. are inversely coupled during regulation of negative affect and 52 Price JL. Free will versus survival: brain systems that underlie predict the diurnal pattern of cortisol secretion among older intrinsic constraints on behavior. J Comp Neurol 2005; 493: adults. J Neurosci 2006; 26: 4415–4425. 132–139. 32 Wang J, Rao H, Wetmore GS, Furlan PM, Korczykowski M, Dinges 53 Damasio AR. The somatic marker hypothesis and the possible DF et al. Perfusion functional MRI reveals cerebral blood flow functions of the prefrontal cortex. Philos Trans R Soc Lond B Biol pattern under psychological stress. Proc Natl Acad Sci USA 2005; Sci 1996; 351: 1413–1420. 102: 17804–17809. 54 Barbas H, Saha S, Rempel-Clower N, Ghashghaei T. Serial 33 Coe CL, Franklin D, Smith ER, Levine S. Hormonal responses pathways from primate prefrontal cortex to autonomic areas may accompanying fear and agitation in the squirrel monkey. Physiol influence emotional expression. BMC Neurosci 2003; 4: 25. Behav 1982; 29: 1051–1057. 55 Datson NA, van der Perk J, de Kloet ER, Vreugdenhil E. 34 Lyons DM, Wang OJ, Lindley SE, Levine S, Kalin NH, Schatzberg Identification of corticosteroid-responsive genes in rat hippo- AF. Separation induced changes in squirrel monkey hypothala- campus using serial analysis of gene expression. Eur J Neurosci mic-pituitary-adrenal physiology resemble aspects of hypercorti- 2001; 14: 675–689. solism in humans. Psychoneuroendocrinology 1999; 24: 131–142. 56 Vreugdenhil E, de Jong J, Schaaf MJ, Meijer OC, Busscher J, Vuijst 35 Lyons DM, Martel FL, Levine S, Risch NJ, Schatzberg AF. Postnatal C et al. Molecular dissection of corticosteroid action in the rat experiences and genetic effects on squirrel monkey social hippocampus. Application of the differential display techniques. affinities and emotional distress. Horm Behav 1999; 36: 266–275. J Mol Neurosci 1996; 7: 135–146.

Molecular Psychiatry Social stress affects gene expression AM Karssen et al 1102 57 Patrick GN. Synapse formation and plasticity: recent insights from 77 Xing GQ, Russell S, Webster MJ, Post RM. Decreased expression of the perspective of the ubiquitin proteasome system. Curr Opin mineralocorticoid receptor mRNA in the prefrontal cortex in Neurobiol 2006; 16: 90–94. schizophrenia and bipolar disorder. Int J Neuropsychopharmacol 58 Devoy A, Soane T, Welchman R, Mayer RJ. The ubiquitin- 2004; 7: 143–153. proteasome system and cancer. Essays Biochem 2005; 41: 187–203. 78 Yamada S, Yamamoto M, Ozawa H, Riederer P, Saito T. Reduced 59 Kinyamu HK, Chen J, Archer TK. Linking the ubiquitin-protea- phosphorylation of cyclic AMP-responsive element binding some pathway to chromatin remodeling/modification by nuclear protein in the postmortem orbitofrontal cortex of patients with receptors. J Mol Endocrinol 2005; 34: 281–297. major depressive disorder. J Neural Transm 2003; 110: 671–680. 60 Nawaz Z, O’Malley BW. Urban renewal in the nucleus: is protein 79 Blendy JA. The role of CREB in depression and antidepressant turnover by proteasomes absolutely required for nuclear receptor- treatment. Biol Psychiatry 2006; 59: 1144–1150. regulated transcription? Mol Endocrinol 2004; 18: 493–499. 80 Watakabe A, Sugai T, Nakaya N, Wakabayashi K, Takahashi H, 61 Lonard DM, O’Malley BW. Expanding functional diversity of the Yamamori T et al. Similarity and variation in gene expression among coactivators. Trends Biochem Sci 2005; 30: 126–132. human cerebral cortical subregions revealed by DNA macroarrays: 62 de Kloet ER, Joels M, Holsboer F. Stress and the brain: from technical consideration of RNA expression profiling from post- adaptation to disease. Nat Rev Neurosci 2005; 6: 463–475. mortem samples. Brain Res Mol Brain Res 2001; 88: 74–82. 63 Holsboer F. The corticosteroid receptor hypothesis of depression. 81 Evans SJ, Choudary PV, Vawter MP, Li J, Meador-Woodruff JH, Neuropsychopharmacology 2000; 23: 477–501. Lopez JF et al. DNA microarray analysis of functionally discrete 64 Sibille E, Arango V, Galfalvy HC, Pavlidis P, Erraji-Benchekroun L, human brain regions reveals divergent transcriptional profiles. Ellis SP et al. Gene expression profiling of depression and suicide Neurobiol Dis 2003; 14: 240–250. in human prefrontal cortex. Neuropsychopharmacology 2004; 29: 82 Watakabe A, Komatsu Y, Nawa H, Yamamori T. Gene expression 351–361. profiling of primate neocortex: molecular neuroanatomy of 65 Rosabal F. Cytoarchitecture of the frontal lobe of the squirrel cortical areas. Genes Brain Behav 2006; 5(Suppl 1): 38–43. monkey. J Comp Neurol 1967; 130: 87–108. 83 Khaitovich P, Muetzel B, She X, Lachmann M, Hellmann I, 66 Sanides F. The architecture of the cortical taste nerve areas in Dietzsch J et al. Regional patterns of gene expression in human squirrel monkey (Saimiri sciureus) and their relationships to and chimpanzee brains. Genome Res 2004; 14: 1462–1473. insular, sensorimotor and prefrontal regions. Brain Res 1968; 8: 84 Sandberg R, Yasuda R, Pankratz DG, Carter TA, Del Rio JA, 97–124. Wodicka L et al. Regional and strain-specific gene expression 67 Ongur D, Price JL. The organization of networks within the orbital mapping in the adult mouse brain. Proc Natl Acad Sci USA 2000; and medial prefrontal cortex of rats, monkeys and humans. Cereb 97: 11038–11043. Cortex 2000; 10: 206–219. 85 Lein ES, Zhao X, Gage FH. Defining a molecular atlas of the 68 Reep R. Relationship between prefrontal and limbic cortex: a hippocampus using DNA microarrays and high-throughput in situ comparative anatomical review. Brain Behav Evol 1984; 25: 5–80. hybridization. J Neurosci 2004; 24: 3879–3889. 69 Lewis DA, Foote SL, Goldstein M, Morrison JH. The dopaminergic 86 Zhao X, Lein ES, He A, Smith SC, Aston C, Gage FH. Transcrip- innervation of monkey prefrontal cortex: a tyrosine hydroxylase tional profiling reveals strict boundaries between hippocampal immunohistochemical study. Brain Res 1988; 449: 225–243. subregions. J Comp Neurol 2001; 441: 187–196. 70 MacLean PD, Newman JD. Role of midline frontolimbic cortex in 87 Leonardo ED, Richardson-Jones JW, Sibille E, Kottman A, Hen R. production of the isolation call of squirrel monkeys. Brain Res Molecular heterogeneity along the dorsal-ventral axis of the 1988; 450: 111–123. murine hippocampal CA1 field: a microarray analysis of gene 71 Rolls ET, Burton MJ, Mora F. Neurophysiological analysis of brain- expression. Neuroscience 2006; 137: 177–186. stimulation reward in the monkey. Brain Res 1980; 194: 339–357. 88 Datson NA, Meijer L, Steenbergen PJ, Morsink MC, van der Laan S, 72 Gilad GM, Gilad VH. Overview of the brain polyamine-stress- Meijer OC et al. Expression profiling in laser-microdissected response: regulation, development, and modulation by lithium hippocampal subregions in rat brain reveals large subregion- and role in cell survival. Cell Mol Neurobiol 2003; 23: 637–649. specific differences in expression. Eur J Neurosci 2004; 20: 73 Ryan MM, Lockstone HE, Huffaker SJ, Wayland MT, Webster MJ, 2541–2554. Bahn S. Gene expression analysis of bipolar disorder reveals 89 Segman RH, Shefi N, Goltser-Dubner T, Friedman N, Kaminski N, downregulation of the ubiquitin cycle and alterations in synaptic Shalev AY. Peripheral blood mononuclear cell gene expression genes. Mol Psychiatry 2006; 11: 965–978. profiles identify emergent post-traumatic stress disorder among 74 Harro J. CCK and NPY as anti-anxiety treatment targets: promises, trauma survivors. Mol Psychiatry 2005; 10: 500–513. pitfalls, and strategies. Amino Acids 2006; 31: 215–230. 90 You J, Wang M, Aoki T, Tamura TA, Pickart CM. Proteolytic 75 Caberlotto L, Hurd YL. Neuropeptide Y Y(1) and Y(2) receptor targeting of transcriptional regulator TIP120B by a HECT domain mRNA expression in the prefrontal cortex of psychiatric subjects. E3 ligase. J Biol Chem 2003; 278: 23369–23375. Relationship of Y(2) subtype to suicidal behavior. Neuropsycho- 91 Lazarov O, Robinson J, Tang YP, Hairston IS, Korade-Mirnics Z, pharmacology 2001; 25: 91–97. Lee VM et al. Environmental enrichment reduces Abeta levels and 76 Kuromitsu J, Yokoi A, Kawai T, Nagasu T, Aizawa T, Haga S et al. amyloid deposition in transgenic mice. Cell 2005; 120: 701–713. Reduced neuropeptide Y mRNA levels in the frontal cortex of 92 Schatzberg AF, Rothschild AJ. Psychotic (delusional) major people with schizophrenia and bipolar disorder. Brain Res Gene depression: should it be included as a distinct syndrome in Expr Patterns 2001; 1: 17–21. DSM-IV? Am J Psychiatry 1992; 149: 733–745.

Supplementary Information accompanies the paper on the Molecular Psychiatry website (http:// www.nature.com/mp)

Molecular Psychiatry