SupplementaryMethods

Behavioral testing. Behavioral testing was done during the light cycle of the day (between 14.00-18.00 h) using nine to twelve mice from each inbred strain. Testing of the 129S6/SvEvTac, A/J, C3H/HeJ, C57BL/6J, DBA/2J, and FVB/NJ strains was conducted in two phases. In total ten animals per strain were tested. First, five animals per strain were tested over a period of one week. One month later another five animals were tested over a period of one week to make sure there were no differences in the behavior of mice from separate batches ordered from the vendors. Testing of the AB6F1 (n=15), B6AF1 (n=12), and BALB/cByJ (n=10) animals was done several months later together with control animals from the initial strains. Behavior of these control animals was in accordance with the earlier results. Behavioral testing of the lentivirus injected animals was done over a period of six days for the over-expression and four days for the RNAi experiment.

Two assays developed to measure anxiety-related behaviors were employed. Both the light-dark box test 1 and the open-field test 2,3 were performed using the Med Associates OFA-515 behavioral chambers (42 cm x 42 cm) that have 16 infrared emitters and photo detectors on each side of the box. On two sides, there are emitters and detectors in two rows enabling measurement of rearings. A high level of photo emitter/detector coverage in the apparatus provides redundancy to maximize reliability of monitoring. For the light-dark box test, a light/dark insert was used which divides the chamber into two equally sized compartments – a light one and a dark one with a hole in between that allows mice to move freely between the compartments. In both tests, an additional light source was placed above the test chamber to make it more anxiogenic for the mice. Movement of mice was detected by beam breaks, and the whole system was computerized which enabled reliable measurement of several different parameters, such as time spent in various parts of the chamber, distance traveled, ambulatory time, resting time, number and duration of rearings, and average velocity. The data was analyzed using the Activity Monitor software(Med Associates).

On the first day mice were tested using the light-dark box test. Mice are nocturnal animals and like to spend time in the dark compartment. They are, however, very curious and like to explore new surroundings even though the brightly lit compartment is very anxiogenic to them. Mice were placed in the dark compartment and their movement was monitored for 5 minutes.Anxious animals generally take longer to emerge to the light side of the chamber for the first time. They also spend more time in the dark compartment compared to the non-anxious animals.The latency to emerge from the dark side to light side was measured as the time from the beginning of the test to the time when the mouse had both hind paws in the light compartment. This was measured by the person who conducted the test as it is a more exact measure than we could obtain by the computerized system. All other aspects of the test were measured by monitoring beam breaks. When the results of the light-dark box test were analyzed, the chamber area was divided into three zones instead of two (light and dark) because the mice often peeked out from the dark side to the light side but did not fully emerge to the light side. The “gray area” between the light and dark zones was relatively narrow and consisted of a zone that spanned the middle 1/8 vertical section of the chamber. The time spent in the dark, gray, and light zones, as well as the distance traveled in each zone were calculated separately.

In the beginning of the open-field test a mouse was placed in a corner of the test chamber and it was allowed to move freely for five minutes. Anxious animals spend more time close to the walls of the chamber while non-anxious animals spend time in the middle area as well.When analyzing the results, the chamber was divided into two zones - center area and periphery. The periphery zone consisted of the area within the two outermost beams (5 cm). The test chamber was cleaned with Clidox solution, wiped dry, and allowed to air dry between animals.

Data analysis. Array results were analyzed using several different methods.First, .cel files were generated using Affymetrix software, imported into our expression database, and then processed within the TeraGenomics analysis system (Information Management Consultants) 4. To identify the subset of genes that were consistently differentially expressed between the anxious and non-anxious mice, comparisons were generated between the strains as follows. The two most anxious strains (A/J and DBA/2J which were selected from the three equally anxious strains) were directly compared to the two least anxious strains (C57BL/6J and FVB/NJ). All A/J replicates were compared to all C57BL/6J and FVB/NJ replicates and all DBA/2J replicates were compared to all C57BL/6J and FVB/NJ replicates resulting in 16 pairwise comparison files for each tissue. Genes were considered differentially expressed if they met a set of criteria designed to maximize sensitivity to subtle changes while minimizing the false positive rate: i) a call of increase or marginal increase, or decrease or marginal decrease in at least 75% of the comparisons, ii) a fold change of at least 1.5 in at least 75% of the comparisons, iii) a normalized signal difference of at least 50 in at least 75% of the comparisons, and iv) a call of present in at least one file. More detailed information on the analysis statistics and the TeraGenomics platform can be found at the TeraGenomics home page ( The annotated gene expression data along with data mining tools are also publicly available at

The Bullfrog software tool 5, developed in our lab, was then used to identify a subset of these genes with expression patterns that appeared to be correlated with the anxiety-like behavior of the six inbred mouse strains. For this purpose, the normalized signal intensities of the replicate samples were averaged. The “shape vector” tool of Bullfrog was used to identify probe sets with expression patterns (signals across the six strains) that were most highly correlated with the “ideal” pattern defined by the phenotypic variation across the six strains. The “ideal shape” phenotype vector used in this analysis was based on the anxiety-like phenotype of the six inbred mouse strains (1, 1, 1, 0.65, 0.25, 0) for the strains A/J, DBA/2J, 129S6/SvEvTac, C3H/HeJ, C57BL/6J, and FVB/NJ, respectively, with a “1” meaning highly anxious and a “0” not anxious. The behavioral score was determined by using the results from both the open-field test and the light-dark box test. The three equally anxious strains A/J, DBA/2J, and 129S6/SvEvTac all received a score of “1”. The least anxious strain FVB/NJ received a score of “0”. The behavior of the C3H/HeJ and C57BL/6J mice was intermediate. Their scores (“0.65” for C3H/HeJ and “0.25” for C57BL/6J) were determined by their behavior relative to the extreme phenotypic groups taking into account both the results from the open-field test and the light-dark box test. The probe sets with a pattern best correlated with the phenotype vector (absolute Pearson correlation coefficient across all strains greater than or equal to 0.75 (R ≥ 0.75 or R ≤ -0.75 for correlated and anti-correlated patterns, respectively) were selected for further analysis. These probe sets have either high levels of expression in the anxious strains and low levels in the non-anxious strains (positive correlation coefficient) or low levels of expression in the anxious strains and high levels of expression in the non-anxious strains (negative correlation coefficient).

We also used a standard implementation of a linear mixed effects model to assess the association between expression and anxiety trait. Animals were classified on the basis of the open-field test as low, intermediate, or high anxiety. Identification of genes that were significantly differentially expressed between the high and low anxiety mice was then carried out using a mixed effects model:

where is a vector of observed expression responses, is the design matrix of fixed effects (strain), is a vector of fixed-effects parameters, is the design matrix of random effects (anxiety category), and is a vector of random-effects parameters. The analyses were carried out using the lme function in SPLUS v6.0, which uses the formulation described in Laird and Ware6, but also allows for nested random effects. The significance of the association between anxiety and expression involves extensive multiple testing that must be taken into account in assessing overall significance. Given the limited number of strains and animals per strain, empirically estimating the null distribution was not feasible. Therefore, multiple testing was taken into account using the false-discovery rate procedure detailed by Benjamini and Hochberg 7. Of the 12,488 genes represented on the microarray, M were selected based on expression over the different strains over the different tissues. P-values for the correlation between the M genes and anxiety trait were computed and ordered such that . The first tests are declared significant where is the largest such that . This procedure controls the FDR (false discovery rate) at level (e.g., for the FDR was 36%).

The quantitative results for the 19 probe sets that correlated with the anxiety-related phenotype and that were scored as differentially expressed between the most anxious and the least anxious strains were subjected to a cluster analysis using the GeneSpring software (Silicon Genetics). The normalized signal intensities from the replicate samples were averaged and the order of the samples was fixed. The Pearson correlation coefficient was used to cluster the 19 probe sets using the “Gene Tree” function of GeneSpring.

Determination of reproducibility between replicates. The reproducibility of the results was estimated using two different methods (supplementary table 1). First, the comparisons of the replicate samples were created and the number of differentially expressed genes were determined using the criteria that mimic the criteria used in the real analysis: i) a call of increase/marginal increase or decrease/marginal decrease, ii) a fold change of at least 1.5, iii) a signal difference of at least 50, and iv) a present call in at least one file (supplementary table 1A). Similarly, comparison files were created for all other strain comparisons within a tissue and differentially expressed genes were identified using the same criteria as described above. To obtain an FDR, the number of differentially expressed genes between replicates was compared to the number of differentially expressed genes between non-replicate samples. The median FDR was estimated to be 23.1% based on this analysis (supplementary table 1).

To obtain another estimate of the false-positive rate for the array measurements, the following analysis was done that better mimics the filtering approach used to identify differentially expressed genes between the anxious and non-anxious strains (supplementary table 1B). For each tissue the following four pairwise comparison files were created: A/J array 1 vs. A/J array 2, C57BL/6J array 1 vs. C57BL/6J array 2, DBA/2J array 1 vs. DBA/2J array 2, and FVB/NJ array 1 vs. FVB/NJ array 2. Genes were scored as false positives if they met the following criteria (the same criteria used in the anxious vs. non-anxious analysis): i) a call of increase/marginal increase or decrease/marginal decrease in at least 75% of the comparisons, ii) a fold change of at least 1.5 in at least 75% of the comparisons, iii) a signal difference of at least 50 in at least 75% of the comparisons, and iv) a present call in at least one file. Since the number of comparison files compared here was four instead of the 16 used in the experimental comparisons, this approach gives a very conservative estimate of the false positive rate (0.013%; supplementary table 1B).

Quantitative RT-PCR. Total RNA was prepared as described in the methods section of the paper. One microgram of RNA was DNase treated using the DNA-free kit (Ambion) according to the manufacturer’s protocol. 200 ng of RNA was used for reverse transcription using SuperScript II (Invitrogen). PCR-reactions were done using the SYBR Green Master Mix (Applied Biosystems) in an ABI Prism SDS 7900 HT machine (Applied Biosystems) according to the manufacturer’s protocol. Primers for the amplification were designed using the Primer Express software (Applied Biosystems), and when possible primers overlapped an exon-intron boundary. Primer sequences are shown in the table below. The genes of interest were amplified in duplicate in two samples from each inbred mouse strain. A housekeeping gene cyclophilin was amplified from all of the samples on the same plate and used as a control. The results of the qPCR amplification were analyzed with the Sequence Detector and Dissociation Curve software (Applied Biosystems). Relative quantification was performed with the standard curve method according to the manufacturer’s recommended procedure. The amplification level of the gene of interest was normalized by dividing it by the cyclophilin amplification level for each sample. The following tissues were used to determine the mRNA levels: Aminolevulinate, delta-, dehydratase (hypothalamus); Erythroid differentiation factor (hypothalamus); Cleavage and polyadenylation specific factor 4 (hippocampus); Kallikrein 16 (pituitary); Glyoxalase I (periaqueductal gray); Prostaglandin D2 synthase (periaqueductal gray); Cadherin 2 (pituitary); Epoxide hydrolase 1 (hypothalamus); cDNA clone MGC:67258 (hippocampus); Glutathione reductase1 (cingulate cortex); Prosaposin (periaqueductal gray).

Primer sequences for quantitative real-time PCR.

Gene / Forward primer / Reverse primer
Gsr / ATGAAGATGGTTTGTGCCAACA / CCAATCCCCTGCATGTGAA
Cdh2 / TCCACCTCGCTGTAAAAATGG / CATGCAAAAAACCTGAATCCAA
Ptgds / GGTCCTCGCCTCCAACTCA / CTTGCACATATACAATACAGCTTTCTTCT
Alad / AGAGGTGGCACTGGCCTATG / CGTCCGTCCATCATGTCTGA
MGC:67258 / TTTGTGGGTGTGAATTGAAAGTCTA / GGGTCCATAAACAGAGAAACACGTAT
Glo1 / CCTGATGACGGGAAAATGAAAG / GCCGTCAGGGTCTTGAATGA
Cpsf4 / GCTTGCTTGTGGGCACCTT / GCAGACCAACTGGGTCAGCTA
Erdr1 / GCCAGGATGGAGCGATTCT / GGGCATTTCTGTACGCAGTCA
Ephx1 / CCCAGGACATCCGCAAGT / CATGGTTGGTGTGTAGCGTCAT
Klk16 / TTGTGATGGTGTTCTCCAAGGA / TTGAACTTAACAAGGTTGGTGTAGATG
Psap / GCAAAGGTAAAGAGAGGCGTTAGA / CGGGTGGTGGTAATAAAAGAAAA
Cyclophilin / GGAGATGGCACAGGAGGAAA / CCCGTAGTGCTTCAGCTTGAA

Sequence analysis. cDNA was prepared from hypothalamus total RNA as described for the microarray analysis. Glo1 and Gsr cDNA was amplified by PCR and purified using the QIAquick PCR purification kit (Qiagen). A commercial service (from Eton Bioscience, Inc.) was used to obtain sequences from two animals from each six mouse strain.

Primers for sequencing Glo1 and Gsr:

Sequence name / Accession number / Target bp / Forward primer sequence / Reverse primer sequence
Glo1-A / XM_128546 / 18-771 / gactgcaaaggcgtcccc / tggagcagagagacccatc
Glo1-B / XM_128546 / 500-1216 / gtgcctgtaagagatttgaag / attagtatttctatctgtgtcc
Gsr-A / NM_010344 / 61-692 / agggccgagaggatgcctgcga / cgcgcgctcacctgcaccat
Gsr-B / NM_010344 / 504-1252 / gtttaccgctccacacatcc / cattgtctttcccatacttatg
Gsr-C / NM_010344 / 1103-1646 / gccggaaacttgcccataga / cgtgtgtgtgtttagctgga

Glo1 was sequenced in two overlapping parts (A and B) and Gsr in three overlapping parts (A, B, and C).

Enzyme activity assays. Eight week old mice were killed by decapitation and their cortex, hippocampus, and striatum dissected under a dissection microscope, frozen on dry ice, and stored at -80 oC. The samples were homogenized in 1.5 ml of ice cold homogenization buffer for 1 min using a Polytron homogenizer (Kinematica). The buffer used for the glutathione reductase assay consisted of 50 mM potassium phosphate, pH 7.5, 1 mM EDTA, and a proteinase inhibitor tablet (Roche) in 50 ml total volume. The buffer for the glyoxalase 1 assay consisted of 0.1 M sodium phosphate, pH 7.5, 1 mM EDTA, and a proteinase inhibitor tablet (Roche) in 50 ml total volume. The buffer for the aminolevulinate delta dehydratase assay consisted of 100 mM sodium phosphate, pH 7.4 and a proteinase inhibitor tablet (Roche) in 50 ml total volume. Homogenates were centrifuged at 8600 g for 10 min at 4 oC and the supernatant was used for the further assays. The total protein concentration was determined using the BCA Protein Assay Kit (Pierce) according to the manufacturer’s instructions.

Glutathione reductase activity was determined from 500 µg of total protein using the Glutathione Reductase Assay Kit (Calbiochem) according to the kit instructions. Glyoxalase 1 activity was determined from 50 µg of total protein as described 8. Briefly, the activity of glyoxalase 1 was assayed by measuring the initial rate of formation of S-D-lactoylglutathione from the hemithioacetal formed non-enzymatically from methylglyoxal and reduced glutathione. Hemithioacetal was prepared by pre-incubation of methylglyoxal (2 µmol, 10 µl) with glutathione (2 µmol, 10 µl) at 37 oC for 10 min in sodium phosphate buffer (50 mM, pH 6.6, 990 µl). Brain homogenate dilution containing 50 µg of total protein in 10 µl was added and the absorbance at 240 nm monitored over 4 min. The 240 of 2.86 mM-1cm-1 was used to calculate the glyoxalase 1 activity rate.

Aminolevulinate delta dehydratase activity was measured as described 9 with minor modifications. Fifty microliters of brain homogenate containing 300 µg of total protein was mixed with 115 µl assay buffer (8 mM ALA, 20 mM DTT, 50 mM sodium phosphate buffer, pH 5.8) and incubated for 2.5 h at 42 oC. 450 µl of 6 % TCA/0.1 M HgCl2 was added and the mixture was centrifuged for 5 min at 1000 g. 450 µl of the supernatant was incubated with 450 µl modified Ehrlich’s reagent for 15 min and the absorbance at 553 nm measured. The Ehrlich-PBG color salt has a molar absorption coefficient of 6.1 x 104 at 553 nm.

Regression analysis of open-field behavior and Glo1 and Gsr enzyme activities. The relationship between open-field behavior and enzyme activity levels of Glo1 and Gsr across BALBc/ByJ, A/J, C57BL/6J, and AB6F1/B6AF1 strains was assessed using a procedure that takes into account both intrinsic and measurement error associated with the assigned strain values for these variables 10. Thus, the mean behavior and enzyme levels for the strains was calculated, and double weighted regression methods that account for within-strain variances for these measures were applied, as described by Akritas and Bershady 10. A t-test (1-tailed, three degrees of freedom) was performed to assess the significance of the correlation between the open-field behavior and enzyme activity. The p-value for the correlation between the behavior and Glo1 activity was 0.0005, and between the behavior and Gsr activity 0.0090.