Proteomic and Metabolomic Profiling of a Trait Anxiety Mouse Model Implicate Affected Pathways

Abstract

Depression and anxiety disorders affect a great number of people worldwide. Whereas singular factors have been associated with the pathogenesis of psychiatric disorders, growing evidence emphasizes the significance of dysfunctional neural circuits and signaling pathways. Hence, a systems biology approach is required to get a better understanding of psychiatric phenotypes such as depression and anxiety. Furthermore, the availability of biomarkers for these disorders is critical for improved diagnosis and monitoring treatment response. In the present study, a mouse model presenting with robust high versus low anxiety phenotypes was subjected to thorough molecular biomarker and pathway discovery analyses. Reference animals were metabolically labeled with the stable ¹⁵N isotope allowing an accurate comparison of protein expression levels between the high anxiety-related behavior versus low anxiety-related behavior mouse lines using quantitative mass spectrometry. Plasma metabolomic analyses identified a number of small molecule biomarkers characteristic for the anxiety phenotype with particular focus on myo-inositol and glutamate as well as the intermediates involved in the tricarboxylic acid cycle. In silico analyses suggested pathways and subnetworks as relevant for the anxiety phenotype. Our data demonstrate that the high anxiety-related behavior and low anxiety-related behavior mouse model is a valuable tool for anxiety disorder drug discovery efforts.

For an improved understanding of the etiology of complex diseases such as psychiatric disorders the elucidation of molecular pathways is critical. In this regard biomarker information can deliver valuable data not only on individual molecular entities but at the same time on pathways critical for disease pathobiology, thus yielding important information for the development of therapeutic agents.

Animal models have the capability to mimic certain aspects of complex disorders and thereby untangle complicated phenotypes such as anxiety, which can be measured in the mouse with the help of the elevated plus maze (EPM)¹ and other anxiety tests (1). In earlier studies we have identified proteome differences in a mouse model of extremes in trait anxiety that are qualitative and quantitative in nature. Whereas the enzyme enolase phosphatase was found as a different isoform in high (HAB) versus low (LAB) anxiety-related behavior mice, another enzyme, glyoxalase-1 (Glx1), showed altered expression levels between the two lines (1, 2). Our approach thus considers the two poles of the continuum “anxiety”: vulnerability of individuals with high risk scores as well as resilience of individuals with low risk scores at the often neglected “other end” of the continuum of polygenic liability (3).

In order to analyze the proteomes of the HAB and LAB mouse lines in greater detail, we have used a comprehensive and accurate proteomics platform that involves metabolic labeling of live animals with stable isotopes followed by quantitative mass spectrometry (4–6). Complementary metabolomic studies provide additional information on pathways affected in disease pathobiology. Here we present results from our proteomic, metabolomic, and pathway analyses of HAB versus LAB mice and discuss their significance with regard to the anxiety phenotype. Our results demonstrate that the mouse model under investigation reflects several critical aspects of human anxiety pathobiology, making it a valuable tool for guiding drug discovery efforts.

EXPERIMENTAL PROCEDURES

Materials

Standard rodent diet (Harlan Laboratories, Inc. Indianapolis, IN) and bacterial protein-based rodent diets (Silantes GmbH, Munich, Germany) were used. Two isotopic forms of bacterial diets were employed: natural isotopic (denoted as ¹⁴N) and ¹⁵N-enriched. All other chemicals were from Sigma-Aldrich (St. Louis, MO), Merck (Darmstadt, Germany), and BioRad (Hercules, CA).

Animal Experiments

All the animal experiments were conducted in accordance with the “Guide for the Care and Use of Laboratory Animals of the Government of Bavaria.” The ¹⁵N or ¹⁴N feeding was started in utero as described previously (5). Feeding the mice with bacterial protein-based diets did not result in any discernible health effects compared with animals fed a standard diet. The mice gained weight similarly to those fed by normal diet. On postnatal day (PND) 56, all male mice were sacrificed and brain sections including the hippocampus were removed for subsequent analyses. Blood was taken by cardiac puncture, and plasma was obtained by centrifuging the blood in an EDTA and protease inhibitor mixture tablet (F. Hoffmann-La Roche Ltd. Basel, Switzerland) pre-added tube at 1300 × g for 10 min. The pellets representing blood cells were saved. The remaining body blood was removed by 0.9% saline perfusion. Hippocampus and plasma were snap-frozen in liquid nitrogen and stored at −80 °C.

Behavioral Tests

Ultrasonic vocalization tests, EPM, and tail suspension test were performed on PND 5, 49, and 51, respectively, as described previously (5). Briefly, ultrasonic vocalization calls were detected and recorded for 5 min with a bat detector (Mini 3 bat-detector, Ultra Sound Advice, London, U.K.) at 70 kHz. EPM test recorded the number of entries into the closed and open arms and the percentage of time spent on the open arms. The animals' behavior in the tail suspension test was videotaped for 6 min and the duration of total immobility scored by a trained observer blind to the treatment.

Protein Sample Preparation

The hippocampus and plasma of male mice from PND56 were used for quantitative proteomics. The ¹⁴N-HAB and ¹⁴N-LAB specimens were mixed with the respective ¹⁵N-labeled reference material from “normal” anxiety-related behavior (NAB) mice at a 1:1 ratio. HAB/LAB protein expression level differences were deduced by comparing the results from the HAB/NAB and LAB/NAB analyses. This way any potential dietary and/or isotopic effects on protein expression can be avoided a priori. Three biological replicate analyses were conducted for each comparison.

¹⁴N-LAB and ¹⁴N-HAB plasma samples were mixed with ¹⁵N-NAB plasma at a ratio of 1:1. After diluting 1:50 with 10 mm Tris-HCl, pH 7.4, 150 mm NaCl, provided as part of an IgY-M7 Spin column kit (GenWay Biotech, Inc., San Diego, CA), protein concentrations were estimated by Bradford assay. The protein mixtures were subjected to an IgY-M7 spin column to remove the seven most abundant plasma proteins (serum albumin, IgG, fibrinogen, transferrin, IgM, haptoglobin, and α1-antitrypsin), according to the manufacturer's guidelines (GenWay Biotech, Inc.). Briefly, the mixed proteins were first incubated with IgY micro beads, which bound the seven proteins. The remaining unbound proteins were spun down and collected in the flow-through fraction. The flow-through fractions were concentrated by ultrafiltration with a centrifugal YM-3, 3kDa cutoff filter (Millipore, MA). The hippocampal proteins were extracted by using published methods that were slightly modified (7). The mouse hippocampus was put into an ice-cold buffer of 250 mm sucrose, 50 mm Tris-HCl (pH 7.4), 5 mm MgCl₂, 1 mm dithiothreitol, 25 μg/ml spermine, 25 μg/ml spermidine, and a protease inhibitor mixture tablet (F. Hoffmann-La Roche Ltd.) and then homogenized with a Teflon-glass dounce homogenizer and an electric drill at 1200 rpm for 3 min. The homogenates were then centrifuged at 6000 × g for 15 min at 4 °C to pellet the nuclei and mitochondria. The supernatants were collected, and the protein concentrations were measured by Bradford assay. The two samples being compared (¹⁴N-HAB versus ¹⁵N-NAB, ¹⁴N-LAB versus ¹⁵N-NAB) were mixed at a 1:1 ratio, based on total protein content. The protein mixtures were then subjected to ultracentrifugation for 1 h at 100,000 × g in a swing bucket rotor at 4 °C. The supernatants were collected as the cytosol proteins. The pellets were resuspended with 0.5 ml of ME buffer (20 mm Tris-HCl, pH 7.8, 0.4 m NaCl, 15% glycerol, 1 mm dithiothreitol, protease inhibitor mixture tablet, and 1.5% Triton-X-100) and incubated for 1 h with gentle rocking. The supernatants were collected as microsome proteins after centrifugation at 9000 × g, 4 °C for 30 min. Cytosol and microsome fraction protein concentrations were measured by Bradford assay.

The brain and plasma protein mixtures were resolved by Criterion XT Bis-Tris precast gels (BioRad), and the gels stained with Coomassie Brilliant Blue (BioRad). The gel lane containing the separated proteins was cut into 2-mm wide pieces, and the resulting pieces subjected to in-gel tryptic digestion. Gel pieces were destained twice with 100 μl 50 mm NH₄HCO₃/acetonitrile (1:1, v/v) for 30 min, and disulfide bonds reduced with 10 mm dithiotreitol in 50 mm NH₄HCO₃ at 56 °C for 30 min, and then alkylated with 55 mm iodoacetamide in 50 mm NH₄HCO₃ in the dark for 30 min. Subsequently, 12.5 ng/μl trypsin in 25 mm NH₄HCO₃ was added to saturate and cover gel slices. The enzymatic reaction was carried out overnight at 37 °C. After digestion, the peptides were extracted from the gel pieces by adding 5% formic acid at 37 °C for 30 min. The gel pieces were spun down and the liquid collected. The extraction was repeated twice. Finally, the extracted peptide mixture was lyophilized to dryness and dissolved in 10 μl 0.1% formic acid.

HPLC and Mass Spectrometry for Proteomics

The peptide mixtures were analyzed by nanoHPLC (Eksigent Technologies, Inc., Dublin, CA) coupled to an LTQ-Orbitrap (Thermo Fisher Scientific, Bremen, Germany) hybrid mass spectrometer. Peptides were separated on a C18 column at a 200 nl/min flow rate by using a gradient of 2–98% solvent B (98% acetonitrile in water, 0.1% formic acid) over 130 min. The eluates were on-line electrosprayed into the mass spectrometer via a nanoelectrospray ion source (Thermo Fisher Scientific, San Jose, CA). The LTQ-Orbitrap was running in positive ion, top five data-dependent acquisition mode. For full scans in the Orbitrap, the target ion value was 1,000,000, and the maximal injection time was 500 ms at a resolution of r = 60,000 at m/z 400. The MS full scan range was 380–1600 m/z.

Peptide and Protein Identification

The Orbitrap raw files were searched against a concatenated forward and reversed IPI-mouse database version 3.46 containing 55,272 protein sequences. Peak picking and searches were performed by Extract_msn and the SEQUEST search engine (8), respectively, both of which included in the Bioworks software (version 3.3.1; Thermo Fisher Scientific). Searches used trypsin as enzyme and allowed for up to one missed cleavage. The ¹⁴N database search was performed by using the following parameters: 20ppm mass tolerance for the MS scan, 1Da for the MS/MS scan, fixed carbamidomethylation for cysteine, and variable oxidation for methionine. The ¹⁵N database search was executed using above parameters plus ¹⁵N amino acid masses and an additional hypothetical –1Da variable modification for arginine and lysine residues (9). The SEQUEST results were filtered by using peptide XCorr >1.9 for 1+ charged ions, >2.7 for 2+ charged ions, >3.5 for 3+ or above charged ions, and DeltaCN >0.08. False discovery rates were calculated using the MAYU software (10) based on the number of peptides matching reversed database entries and the number of peptides matching forward database entries. Redundancy and ambiguity in protein identifications were addressed by grouping proteins, which could not be distinguished based on the identified peptides. Each group consists of proteins with identical sets of identified peptides. Proteins, whose peptides were subsets of other proteins' peptides, were removed. This resulted in a minimal list of proteins for all observed peptides, following “Occam's law of parsimony.” Proteins within a group were treated equally. Thus, in the following when we refer to a “protein” we mean the group it belongs to. Furthermore, proteins were required to have at least two identified peptides with distinct sequences.

Protein Quantification and Significance Analysis

Relative quantification of the peptide pair signals was performed with the ProRata software using default parameters (11). Briefly, for each peptide ion chromatograms were extracted based on peak area for both labeled and unlabeled isotope envelopes according to the amino acid sequence. The peak profile of both chromatograms was used to determine abundance ratio and signal-to-noise ratio of the peptide. ProRata removes peptides with insufficient signal-to-noise ratio and proteins with less than two quantified peptides. ProRata protein abundance ratio estimation is based on a probabilistic model of the peptide ratio distributions. This model was used to calculate HAB/LAB protein ratios and to estimate their statistical significance as outlined in the following. We combined biological replicates and calculated protein ratios across groups based on a separate ProRata tool for downstream analysis (combine.exe, available at http://code.google.com/p/prorata/). Briefly, for each protein, peptide ratios of biological replicates were combined within groups (HAB/NAB and LAB/NAB) to calculate a profile likelihood of the protein abundance ratio for each group. Profile likelihoods of both groups were subsequently combined via cross-correlation to get a probabilistic estimate of the indirect HAB/LAB abundance ratio. For each protein, a p value for the null hypothesis of no differential expression (i.e. log2 ratio = 0) was derived from the profile likelihood by means of a likelihood ratio test. Protein ratios were considered to be statistically significant based on a p value threshold of 0.05, which was corrected for multiple testing by the procedure of Benjamini-Hochberg (12). Additionally, a protein fold change (determined as the maximum likelihood estimate) of at least two was required for true “differential expression.” This eliminates a number of protein hits, which would be significant based on their p value, but have a fold change too small to be biologically meaningful.

Western Blot Verification

Relative expression levels for several proteins were further analyzed by Western blot. These proteins were selected based on commercial antibody availability and their relevance for psychiatric disorders. Protein mixtures with equal protein content (30 μg) were first resolved by SDS-PAGE. Subsequently, the separated proteins were transferred onto polyvinylidene fluoride (PVDF) membranes (Millipore, Billerica, MA). After incubation with antigen-specific antibodies (anti-carbonic anhydrase sc-17244, anti-transthyretinsc-13098 [Santa Cruz Inc., Santa Cruz, CA] or anti-serum amyloid p-component ab40882 [Abcam, Cambridge, U.K.]) the membranes were treated with HRP-conjugated secondary antibody. The ECL system and film (GE Healthcare, Chalfont St. Giles, U.K.) were used for protein visualization. Protein signals were quantified by Quantity One software (BioRad).

Metabolite Sample Preparation

Metabolomics analyses were carried out at the UC Davis Metabolomics Core Center (Davis, CA). The plasma samples from six animals from each mouse line were employed in metabolic studies by using the method described previously (13). Briefly, the plasma proteins were precipitated and metabolites extracted. The plasma extract was dried and derivatized by first adding methoxyamine in an aprotic basic solvent and subsequently adding a trimethylsilylating agent.

GC-MS Data Acquisition and Analysis for Metabolomics

The derivatized samples were used for GC/MS profiling. GC-TOF-MS analysis was performed by using an Agilent 6890 N gas chromatograph (Palo Alto, CA) interfaced to a time-of-flight (TOF) Pegasus III mass spectrometer (Leco, St. Joseph, MI). The Agilent injector temperature was held constant at 250 °C while the Gerstel injector was programmed (initial temperature 50 °C, hold 0.1 min, increased at a rate of 10 °C/s to a final temperature of 330 °C, hold time 10 min). Injections of 1 μl were made in split (1:5) mode (purge time 120 s, purge flow 40 ml/min). Chromatography was performed on an Rtx-5Sil MS column (30 m × 0.25 mm i.d., 0.25 μm film thickness) with an Integra-Guard column (Restek, Bellefonte, PA). Helium carrier gas was used at a constant flow of 1 ml/min. The GC oven temperature program had an initial temperature of 50 °C, with a 1 min hold time, and was ramped at 20 °C/min to a final temperature of 330 °C with a 5 min hold time. Both the transfer line and source temperatures were 250 °C. The mass spectrometer ion source operated at −70 kV filament voltage with ion source. After a solvent delay of 350 s, mass spectra were acquired at 20 scans per second with a mass range of 50 to 500 m/z. Significant differences in metabolite concentrations between HAB and LAB mouse specimens were assessed by an unpaired, two-sided t test using six animals per group.

Pathway Analyses

Hippocampal proteins were sorted according to their HAB/LAB log2 relative expression ratio in ascending order and divided into five bins with log2 ratio borders of –1.0, –0.5, 0.5, 1.0. The Gene Ontology (GO) analysis (14) was performed as described previously (15) by using R (16) and the GOstats (17) package. Briefly, for each bin the p values for every GO category were calculated with the conditional hypergeometric test by using the quantitative proteome as a background. GO categories were then filtered on the basis of their p values. Categories with no significant enrichment (p value < 0.05) in any bin were removed. Those categories that were filtered out received a conservative p value of 1. Finally, the p values were transformed with the equation x = –log10 (p value), and the z-scores were calculated by [x – mean(x)]/sd(x). For Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis (18), the mouse proteins were mapped to the KEGG ortholog level. This allows an interspecies comparison for further investigations. Afterward, the hypergeometric test was calculated by using R. The background of the test was set to all KEGG mouse proteins that have at least one occurrence in a KEGG pathway. The p values were then transformed into z-scores.

Subnetwork Analysis

The differentially expressed hippocampal protein and plasma metabolite information was used to generate enriched interaction and subnetwork maps with the Pathway Studio software (Ariadne Genomics, Inc., Rockville, MD). p values (Fisher Exact Test) were used to select the most significant results.

RESULTS

Proteomic Analyses

For the sensitive quantitative proteomic analysis of the HAB/LAB mouse model we have used tissue specimens from NAB animals metabolically labeled with the stable isotope ¹⁵N as reference. Brain and plasma proteins obtained at PND 56 showed a ¹⁵N incorporation that was greater than 90% (5). The EPM behavioral test demonstrated phenotypes that were consistent for all three animal lines, HAB, LAB, and NAB, used in this study (5).

In previous proteomic analyses we had identified Glx1 as a biomarker for trait anxiety by two dimensional polyacrylamide gel electrophoresis (2-DE) in both brain tissue and blood cells (1). In order to confirm the differential Glx1 expression between HAB and LAB mice and validate the metabolic labeling-based proteomic method employed in the present study, relative Glx1 expression levels were assessed between HAB and LAB mice (Fig. 1). In hippocampus, Glx1 shows a fivefold higher expression level in LAB compared with HAB mice. These results confirm our earlier findings obtained by 2-DE (1).

Fig. 1.

Quantification of glyoxalase-I in HAB and LAB hippocampus. Mass spectrometry total ion current data for the hippocampal Glx1-derived tryptic peptide GLAFIQDPDGYWIEILNPNK. ¹⁵N-labeled NAB hippocampal specimens were used as a reference and either mixed with unlabeled HAB or LAB material. An indirect comparison of the HAB/NAB (left) and LAB/NAB (right) datasets revealed that Glx1 expression in LAB mice is ∼fivefold higher than in HAB mice.

In the following, we present data for the in-depth quantitative proteomic analyses of hippocampal and plasma proteins from the trait anxiety mouse model. The hippocampus was chosen because it represents a brain region that has been implicated in the pathobiology of a number of psychiatric disorders including anxiety and is therefore an important source for gaining insights into dysfunctional molecular pathways. Because of its availability the specimen of choice for biomarker analyses in the clinical laboratory is plasma.

From the hippocampal cytosolic fraction 18,594 distinct peptides and 2956 protein groups, from the hippocampal microsome fraction 19,749 peptides and 3289 groups, and from plasma 5049 peptides and 1297 groups were identified by tandem mass spectrometry after filtering the data using the criteria mentioned above. We used the MAYU (10) software (version 1.06) to estimate false discovery rates (FDR) on the peptide-spectrum match level and the protein level. Because of the relatively stringent filtering criteria, peptide-spectrum match FDRs were estimated as 0.6%, 0.7%, and 1.2% for cytosol, microsome, and plasma, respectively. We required proteins to have at least two peptide identifications with distinct sequences, which removes the majority of decoy proteins. Thus, corresponding protein level FDRs were similarly low in case of cytosol and microsome and amounted to 0.8% and 0.6%, respectively. For plasma, however, we observed an increased protein FDR of 8.2%. Closer inspection of the data revealed that the majority of decoy proteins had been identified in the ¹⁵N database search only. Because of the higher complexity of the ¹⁵N isotopologue patterns, a higher number of false positives is to be expected in the ¹⁵N search compared with the ¹⁴N search. To estimate the impact of this bias, we recalculated the protein FDR of plasma excluding proteins, which had been identified in the ¹⁵N search only. The FDR dropped to an acceptable level of 3%, demonstrating that the majority of dubious protein hits is caused by ¹⁵N-only identifications. Because the plasma data is not used in any downstream analysis, we did not adjust the filtering criteria. Instead, to allow for a better judgment of the reliability of protein identifications, we provide for all supplementary data in addition to the total number of unique identifications the number of ¹⁴N peptide identifications with unique sequences per protein. Since almost all of the reported proteins have at least one ¹⁴N identification we expect the impact of increased false positives resulting from ¹⁵N searches to be negligible. In total, 1576 proteins were quantified, of which 312 were found to be differentially expressed (≥twofold change, ≥2 unique peptides, corrected p value < 0.05) in the hippocampal cytosolic fraction between HAB and LAB mice (supplemental Table S1). For the hippocampal microsome fraction, a total of 1349 proteins were quantified, of which 206 were found to be differentially expressed (supplemental Table S2). In plasma 383 proteins were quantified, of which 58 were found to be differentially expressed (supplemental Table S3). These numbers only reflect proteins identified and quantified in both ¹⁴N-HAB/¹⁵N-NAB and ¹⁴N-LAB/¹⁵N-NAB comparisons.

Expression level differences for three biomarker candidates, carbonic anhydrase 2 (CA2), transthyretin (TTR), and serum amyloid P-component (SAP), were further verified by Western blot analyses (Fig. 2). Based on previous reports these proteins are of particular interest because they have been implicated to play a role in psychiatric disorder pathobiology (19–22).

Fig. 2.

Relative protein quantification in HAB/LAB mice. A, D, G, Eluted chromatographic profiles for carbonic anhydrase 2 (CA2, IPI00121534) tryptic peptide AVQQPDGLAVLGIFLK, transthyretin (TTR, IPI00127560) tryptic peptide TAESGELHGLTTDEK, serum amyloid P-component (SAP, IPI00309214) tryptic peptide GRDNELLIYKEK, respectively. The peak areas are used for the ¹⁴N/¹⁵N signal quantification. In all cases ¹⁵N-labeled NAB proteins were used as a reference and either mixed with unlabeled HAB or LAB material; B, E, H, Western blot analyses of CA2, TTR and SAP protein levels; C, F, I, Western blot protein band density quantification for CA2 (p = 0.0025), TTR (p < 0.0001) and SAP (p = 0.0008), respectively. In all cases the Western blot analyses confirm the relative protein expression level data obtained by mass spectrometry.

Metabolomic Analyses

The great complexity of the brain and cellular heterogeneity even within a defined brain section like the hippocampus makes the identification of metabolite biomarkers in this tissue difficult. We therefore restricted our metabolomic analyses to HAB and LAB mouse plasma specimens. From plasma 265 metabolites were detected by GC-MS analysis of which 86 have known chemical structures. The concentrations of 15 plasma metabolites of known identity differed significantly (p value<0.05) between the HAB and LAB lines (supplemental Table S4). Among these are two inositol isomers. Whereas HAB mice showed a higher level for allo-inositol (p value = 0.011), LAB mice had higher levels of its isomer myo-inositol (p value = 0.002) (Fig. 3). Myo-inositol has been shown to have antidepressant and anxiolytic activities in both humans and animals (23–28). Also of interest is the finding that several energy metabolism related metabolites, including amino acids, cholesterol, fumarate, and malate, were found at different levels between the two lines. Moreover, the key excitatory neurotransmitter, glutamate, was expressed at higher levels in HAB compared with LAB mice (p value = 0.000002) (Fig. 3).

Fig. 3.

Plasma allo-inostitol, myo-inositol, and glutamate levels in HAB and LAB mice. The intensity mean and standard error of mean for the metabolites are indicated.

KEGG and GO Analyses

The correlations between differential hippocampal protein expression and KEGG and “GO cellular component” are shown in Fig. 4. The pathways and categories enriched with proteins from the noncentral bins are of greatest interest because they indicate a significant protein expression level difference between HAB and LAB animals. In the following we highlight those identified pathways that have particular relevance for psychiatric phenotypes.

Fig. 4.

Correlation between hippocampal protein regulation versus KEGG pathways (left), and protein regulation versus GO cellular component (right). The blue boxes at the top indicate relative HAB/LAB protein expression levels as log2 ratios. Proteins were divided into five bins and analyzed with respect to KEGG pathways and GO cellular component. p values were transformed to z-scores, indicating bin-specific enrichments. (because of the limited image resolution the pathway names are not legible; a higher resolution figure is available in the electronic file.)

Based on the KEGG and GO analyses, proteins responsible for inositol phosphate metabolism (p value = 0.0016), phosphatidylinositol signaling (p value = 3.93E-5), and phosphatidylinositol binding (p value = 0.038) were expressed at higher levels in LAB compared with HAB mice. Inositol has been associated with psychiatric disorders, especially bipolar disorder pathobiology in a number of studies. Based on our results the “phosphatidylinositol signaling system” is apparently dysregulated in HAB mice exemplified by an alteration of a number of protein and metabolite levels.

Ubiquitin Mediated Proteolysis (p value = 2.62E-4), ubiquitin-protein ligase activity (p value = 0.001), proteasome (p value = 6.22E-5), and “ubiquitin-specific protease activity” (p value = 0.04) proteins were found at lower levels in HAB compared with LAB mice. This finding is in accordance with previous reports indicating that ubiquitin ligase may act as an anxiety suppressor (29). Further supporting the significance of proteasome-ubiquitin mediated protein degradation for the anxiety phenotype is the fact that “long-term depression” pathway associated proteins were enriched in the first bin (p value = 0.02). An association of the two pathways has been previously reported (30). On the other hand, proteins relevant for “long-term potentiation” (p value = 0.04) were expressed at higher levels in HAB mice, which supports previous electrophysiology data performed in our laboratory (31).

The KEGG analyses also revealed proteins involved in pathways related to energy metabolism, including glycolysis (p value = 6.31E-13), pyruvate metabolism (p value = 4.4E-9), and the TCA cycle (p value = 2.02E-8). In addition, GO analyses demonstrated that electron carrier activity (p value = 0.04), aerobic respiration (p value = 0.024), electron transport chain (p value = 0.04), “acetyl-CoA metabolic process” (p value = 0.006), and the mitochondria (p value = 3.65E-10), are affected in HAB mice. Fumarate and malate, two major intermediates of the TCA cycle, were found at higher levels in HAB mice. The conversion from fumarate to malate is catalyzed by fumarate hydratase, which, in agreement with the metabolite data, was found at an elevated expression level in HAB mice. The same trend was observed for all the other major enzymes that are part of the TCA cycle, including citrate synthase, aconitase, isocitrate dehydrogenase and malate dehydrogenase (Fig. 5A), indicating a major alteration of this pathway. Furthermore, GO analyses revealed that proteins involved in peroxiredoxin activity (p value = 0.005), oxidoreductase activity (p value = 6.62E-4) and oxidation reduction (p value = 1.89E-7) were more abundant in HAB mice, suggesting an important role for oxidative stress in anxiety etiology.

Fig. 5.

TCA cycle pathways and biological subnetworks enriched with proteins and metabolites. A, TCA cycle; enzymes and metabolites labeled in red are expressed at higher levels in HAB compared with LAB mice. B, Subnetwork related to dexamethasone; C, subnetworks related to glycinergic synaptic transmission and myelin maintenance. (because of the limited image resolution the protein and metabolite names are not legible; a higher resolution figure is available in the electronic file.)

GO analyses further demonstrated that proteins relevant to the synapse (p value = 0.01), stress fiber (p value = 3.19E-4), and “neuron projection” (p value = 0.01) were expressed at higher levels in LAB mice. Proteins relevant to “neurotransmitter catabolic process” (p value = 0.02), were found expressed at higher levels in HAB mice, indicating a crucial role of neurotransmission in anxiety.

Subnetwork Enrichment

The above in silico analyses focused on predefined KEGG pathways. To identify general network hotspots, we also conducted a subnetwork enrichment analysis. For this purpose we grouped differentially expressed proteins into small interaction maps, allowing the identification of small nonstatic pathways sharing a high correlation with the phenotype. These analyses indicated the involvement of a number of networks highly enriched with proteins and metabolites differentially expressed between HAB and LAB mice. A significant number of subnetworks were related to the Ras/Raf/MEK/ERK pathway. The central entities in these subnetworks either inhibit the Ras/Raf/MEK/ERK pathway, including PD 98059 (p value = 1.05484E-22) (32), genistein (p value = 3.32622E-17) (33), and wortmannin (p value = 5.64E-13) (34), or are themselves part of the pathway, including MAP2K1 (p value = 3.49E-6), MAPK1 (p value = 2.3E-5), MAPK3 (p value = 1.8E-4), and MAPK8 (p value = 0.001). The identification of both subnetwork types suggests an involvement of this pathway in anxiety-related behavior. In addition, the GO analyses (see above) also demonstrated that the proteins involved in protein serine/threonine kinase activity (p value = 3.98E-4) and small GTPase regulator activity (p value = 0.02), which are quite relevant for the Ras/Raf/MEK/ERK pathway, were expressed at higher levels in LAB mice. The Ras/Raf/MEK/ERK pathway has been previously reported to play a role in neuronal modulation in the context of psychiatric disorders (35–41).

Another important finding of the subnetwork analysis is that dexamethasone (p value = 1.11942E-18) and its interacting network entities are highly enriched with proteins and metabolites differentially expressed between HAB and LAB mice (Fig. 5B). Dexamethasone is a synthetic cortisol and a modulator of the hypothalamus-pituitary-adrenal (HPA) axis whose dysfunction has been implicated to play a major role in depression (42–46).

Glycinergic synaptic transmission (p value = 1.84E-9) is also of particular importance for psychiatric disorders (Fig. 5C). Glycine is an inhibitory neurotransmitter in the spinal cord and brainstem and has been shown to have a key function in the regulation of locomotor behavior (47–50) and beneficial effects in the treatment of depression (51, 52). In addition, proteins relevant for myelin maintenance (p value = 4.73E-8) (Fig. 5C) were also enriched.

DISCUSSION

The pathogenesis of psychiatric disorders remains elusive, and there is growing evidence that several neural circuits and brain pathways are affected. For the characterization of these pathways we have used a systems biology analysis based on both proteomic and metabolic data from a robust trait anxiety mouse model (1). A quantitative proteomic approach that involves metabolic labeling of mice with stable isotopes (4) has enabled us to identify and quantify a large number of proteins in a high throughput manner. Differentially expressed proteins were further interrogated with regard to pathways they are involved in. The resulting protein and metabolite interaction maps suggest several biological processes and pathways to be affected in the genetic predisposition to extremes in trait anxiety. Whether these pathways are causative or the result of distinct psychiatric endophenotypes is unknown at the present time. Be that as it may, the HAB/LAB mouse model used in the present study faithfully represents a number of endophenotypic aspects also found in patients afflicted with anxiety disorders.

A major finding of our analyses implicates the inositol pathway to be critically involved in the anxiety phenotype with several proteins and metabolites that are part of phosphatidylinositol signaling having altered expression levels. Inositol has been shown in several reports to have anxiolytic effects and lithium, a well known mood stabilizer used for treating bipolar disorder, is believed to exert its therapeutic effects through the inositol pathway by decreasing intracellular myo-inositol concentrations (53). Our findings of different myo- and allo-inositol levels in the trait anxiety mouse model lend further support to the relevance of the inositol pathway for psychiatric phenotypes including anxiety.

Oxidative stress has been found to be involved in the pathogenesis of neurological diseases, including Alzheimer's disease, Parkinson's disease, multiple sclerosis, and stroke (54) as well as psychiatric disorders (55–61). Human studies on panic disorder and obsessive-compulsive disorder also suggest an involvement of oxidative stress in anxiety (62–64). Oxidative stress is caused by altered mitochondrial energy pathways leading to abundant reactive oxidative stress compounds. It is therefore not surprising that TCA cycle enzyme and metabolite levels were found to be significantly different between HAB and LAB mice. In studies by others it was also shown that stress-induced anxiety in mice leads to elevated levels for a number of TCA cycle intermediates (65) that can result in excessive oxidative damage. Results from other animal and patient studies also support these findings (60, 66).

A dysfunctional HPA axis that has lost its ability of negative feedback inhibition is considered a hallmark in depression and anxiety disorders (43, 45, 46). The HPA axis and glucocorticoids regulate neuronal survival, neurogenesis, memory, and emotions (67). Excess glucocorticoids may impair or even damage the hippocampus, which may initiate and maintain a hypercortisolemic state found in certain cases of depression (68). Our data show that a significant number of entities relevant to dexamethasone are altered between HAB and LAB mice, indicating an involvement of the HPA axis in anxiety.

An elevated excitatory or decreased inhibitory neurotransmission is frequently observed in depressive and anxious patients (69–71). Anxiety disorder treatment is targeting neurotransmitter pathways using either benzodiazepines or selective serotonin reuptake inhibitors (SSRIs) (72). In our metabolomic analyses the major excitatory neurotransmitter, glutamate, which binds to the N-methyl-d-aspartate (NMDA) receptor was found at higher levels in HAB compared with LAB mouse plasma. This result is consistent with previous findings in patients with depression where higher levels of glutamate in blood, CSF, and certain brain regions were found (69–71, 73, 74). NMDA receptor antagonists have antidepressant effects and are promising alternatives to monoamine-based agents for the treatment of depression and anxiety (75–78). Our study has also identified a number of candidate proteins associated with glycinergic synaptic transmission (Fig. 5C). Both glycine and GABA are essential inhibitory neurotransmitters in the central nervous system. Although glycine's involvement in psychiatric disorders is less understood than that of GABA's, the co-localization and release of GABA and glycine are widespread in inhibitory neurons of the brain and spinal cord. GABA acts as a co-agonist to modify the response of glycine receptors (79). In this regard studies have found that glycine exerts inhibitory effects in certain brain areas, resulting in significant anxiety relief (80, 81).

The Ras/Raf/MEK/ERK pathway is a signal transduction pathway involved in metazoan development. It controls many biological processes, including metabolic processes, the cell cycle, cell migration, and cell shape as well as cell proliferation and differentiation (82). In contrast to its relevance in cancer, the importance of the Ras/Raf/MEK/ERK pathway in psychiatric disorders is still poorly understood. The corticotropin-releasing hormone receptor, which is part of the HPA axis discussed above, exerts its function through the activation of the Ras/Raf/MEK/ERK pathway (83–85). In the present study, a number of enriched sub-networks were found to be relevant for the Ras/Raf/MEK/ERK pathway. Furthermore, inhibition of the Ras/Raf/MEK/ERK pathway by the MEK inhibitor U0126 was found to decrease the depression-like behavior in both wild-type and mutant mice, indicating an involvement of this pathway in psychiatric disorders (41).

Our data further support the notion of anxiety as a polygenic trait caused by multiple gene products, each providing a minor contribution to the phenotype, which is additionally shaped by environmental influences. Such gene-by-environment interactions can induce persistent functional changes in neuronal pathways that underlie variation in anxiety-related behavior and vulnerability to anxiety. This flow of information from DNA to the anxiety phenotype includes a variety of proteins, metabolites and other molecular biomarkers (86). Which of them are causally related to the phenotype is unknown at the present time.

In summary, our -omics data implicate a number of proteins, metabolites and pathways that corroborate previous findings on psychiatric disorder pathobiology. They lend further support to the validity of the trait anxiety mouse model for further experiments with the goal to verify candidate biomarkers in patients afflicted with anxiety disorders and test new medications for their treatment.