Overview (Figure 1)
Tissue from the arcuate nucleus of the medial basal hypothalamus (MBH), hippocampus (HPC) and amygdala (AMD) of rhesus macaques Macaca mulatta were collected for examination in two separates studies. One study was designed to assess effects of three hormone replacement therapy (HRT) conditions on gene expression in the three brain regions collected, and this gene expression was examined using the rhesus GeneChip®. In the HRT study qRT-PCR was conducted on all tissues in addition to microarray analysis. In a second study (cycle study), gene expression in the three brain regions during early follicular (EF) late follicular (LF) and mid luteal (ML) phases of the menstrual cycles of normally menstruating macaques was examined using the human GeneChip®.
In the gene microarray analyses for both the HRT and cycle studies, comparisons of the relative expression stabilities of probe sets annotated for independently assessed human housekeeping genes  were conducted using NormFinder  algorithms equipped to process data from microarrays. Because the reliability of the NormFinder algorithm is dependent on the expression stability of the sequences examined , the most widely used normalizers from the list of housekeeping genes http://www.compugen.co.il/supp_info/Housekeeping_genes.html, as well as RPL13A and ALG9 (the latter two considered noteworthy based on our prior results  and considered here as part of the "popular normalizer" group) were examined in a separate analyses. Again, in efforts to compare probe sets best suited for use in the NormFinder analysis, in the cycle study we designated probe sets as "most representative" according to criteria described later in the current methods section and analyzed these probe sets separately as well.
All analyses from both experiments (Figure 1) are summarized as follows:
1a) HRT study - qRT-PCR: GABAergic component genes, ALG9, GAPDH and RPL13A expression variability compared using geNorm, NormFinder and BestKeeper algorithms.
1b) HRT study - gene microarray -general probe set selection: NormFinder relative expression stability assessments of 1444 rhesus GeneChip® probe sets annotated for housekeeping genes, grouped by brain region or hormone treatment.
1c) HRT study - gene microarray - "expressed sequences" from the general probe set selection: NormFinder relative expression stability assessments of 824 probe sets showing above average signal intensity and receiving "present" calls in MAS 5.0 analysis.
1d) HRT study - gene microarray - probe set selection for popular normalizers: NormFinder relative expression stability assessments of all rhesus GeneChip® probe sets annotated for 15 "popular normalizers", grouped by brain region or hormone treatment.
2a) Cycle study - gene microarray - general probe set selection: NormFinder relative expression stability assessments of 1433 human GeneChip® probe sets annotated for housekeeping genes, grouped by brain region or cycle phase.
2b) Cycle study - gene microarray - most representative probe set selection: NormFinder relative expression stability assessments of 544 human GeneChip® probe sets selected for the highest annotation quality among the 1433 probe sets annotated for housekeeping genes, grouped by brain region or cycle phase.
2c) Cycle study - gene microarray - popular normalizer probe set selection from among the most representative probe sets: NormFinder relative expression stability assessments of most representative human GeneChip® probe sets annotated for 15 "popular normalizers", grouped by brain region or cycle phase.
Adult (age range 9-12 years) female rhesus macaques (Macaca mulatta) were cared for in accordance with IACUC regulations and the National Research Council's Guide for the Care and Use of Laboratory Animals by the Division of Animal Resources at the Oregon National Primate Research Center (ONPRC). All procedures were ethically and legally approved by these overseeing bodies. Twelve animals were used per study, and were housed in a temperature-controlled environment where Primate Chow (Purina Mills Inc., St. Louis, MO) was made available to the animals twice daily, at 8:00 h and again at 15:00 h. This diet was supplemented with fresh fruit and vegetables, and the animals had continuous access to drinking water.
The animals were euthanized using sodium pentobarbital (25-30 mg/kg i.v.) and exsanguinated following procedures recommended by the American Veterinary Medical Association's Panel on Euthanasia. In all cases, necropsies were performed within a narrow window of time (1000-1300 h). Various postmortem tissues were collected and made available to other investigators through the ONPRC Tissue Distribution Program. After a transcardial flush with 1 liter of 0.9% saline, the brains were immediately removed, cut into blocks, and preserved in RNAlater® (Ambion, Austin, TX). Areas of interest (MBH, HPC and AMD) were subsequently dissected from tissue blocks, resulting in a total of 36 individual samples per study.
As previously described , all animals in the HRT study were ovariectomized (OVX). Untreated ovariectomized animals served as controls (OXV), whereas others received subcutaneous SILASTIC implants containing either 17β-estradiol (E2), or E2 and progesterone (E2+P4).
In the cycle study, daily menstrual records were established for each female based on close inspection of the monkey's perineum and cage pan for signs of menstrual bleeding. All females were determined to have regular monthly cycles of approximately 28 days. Only data obtained from healthy monkeys, as determined by the ONPRC veterinary staff, were included in the statistical analyses. Serum E2 levels and serum P4 levels were analyzed to assist in establishing an animal's menstrual cycle phase. Using menstrual cycle monitoring, tissue was timed for collection during the early follicular stage ("EF" = low E2 and P4), late follicular ("LF" = declining E2 after ovulation, low P4) and mid luteal ("ML" = low E2 high P4). Four animals were identified for each of the three phases of the cycle (total = 12). Average values for EF were E2 = 64.8 pg/ml, P4 = 0.18 ng/ml; for LF, E2 = 71 pg/ml, P4 = 0.52 ng/ml; for ML, E2 = 39 pg/ml, P4 = 4.92 ng/ml. At necropsy, menstrual stage was also verified using histological examination of the endometrium, with EF = thin endometrium, ciliated oviduct; LF = thin, proliferative endometrium with ciliated and fully secretory oviduct; ML = thickened endometrium.
Quantitative real-time RT-PCR (qRT-PCR) and data collection were conducted using the 7900HT Fast Real-Time PCR thermal cycler and sequence detection systems (Software version 2.2.1) from Applied Biosystems (Foster City, CA). No-template (negative) controls were included for each gene analysis. Four-point standard curves were constructed using cDNA pooled equally from all brain regions and all animals. The curves were constructed from serial 5- fold dilutions and covered a cDNA dilution range of 0.2 to 0.0016 (larger fold dilutions produced unacceptably high cycle times for some genes in the study).
qRT-PCR reactions for unknowns were conducted in sealed 384-well optical plates in a total volume of 5 μl per well, using 1.0 μl of 1:5-diluted cDNA, and final concentrations of 0.25 μM Taqman® TAMRA probe and 0.3 μM each of forward and reverse primers. Reactions were conducted using thermal cycler conditions of: 2 min at 50°C, 10 min at 95°C and 50 cycles of 15 s at 95°C (DNA melting) and 1 min at 60°C (primer annealing/extension). Baseline and threshold levels for amplification plots were determined automatically using the ABI sequence detection systems software version 2.2.1.
We examined expression of prospective internal reference genes (ALG9, GAPDH and RPL13A), as well as genes encoding GABA receptor subunits (GABRA1, GABRA4, GABRG2, GABRD and GABRE) and a GABA-synthesizing enzyme (GAD1). We compared expression stability across three conditions of ovarian steroid exposure in the three brain regions examined (MBH, HPC and AMD). These comparisons were initially made in an investigation of hormone effects on expression of GABAergic components , but we describe the qRT-PCR results in more detail here for further comparison against results from gene microarray analyses.
qRT-PCR primers and probes
Primers and probes for prospective internal reference genes ALG9, GAPDH and RPL13A as well as GABA subunit receptors GABRA1, GABRA4, GABRG2, GABRD and GABRE and the enzyme GAD1 were designed using Primer Express 2.0 software (Applied Biosystems, Foster City, CA) from sequence sources listed in Table 1. Rhesus macaque sequences for GAPDH and RPL13A were confirmed in-house using RT-PCR primers against source sequences to amplify rhesus macaque hippocampal cDNA pooled from several individuals. These PCR products were sequenced using an ABI 3730×l DNA sequencer (Applied Biosystems) according to manufacturer's instructions.
1a. qRT-PCR GeNorm analysis
Quantity values calculated using the standard curve method  were entered into the geNorm version 3.5 VBA applet for Windows  and used to calculate the gene-stability measure (M) defined as the average pairwise variation of a given gene with all other considered genes. The least stable gene was then eliminated and stability based on pairwise comparisons recalculated to produce a new ranking. This sequential elimination process  was continued until no stability differences were detected between genes. Because one of the MBH OVX samples introduced a high level of variation in gene expression, the analysis was repeated with this sample omitted, in order to test the effects of variation in this group on relative gene expression stability output.
1a. qRT-PCR NormFinder analysis
Quantity values calculated using the standard curve method  were evaluated using the model-based variance estimation approach facilitated by the "NormFinder" Visual Basic application  for Microsoft Excel, available from the Molecular Diagnostics Laboratory (Aarhus University, Denmark). Samples were grouped according to brain region and stability was estimated with and without group identifiers. For each of the three brain regions, intra-group and inter-group variation  was calculated for each gene. Three rounds of analysis were conducted as follows: 1) All nine genes were included in order to test the stability of GABA receptor subunit genes; 2) The five most stable genes (ALG9, GAPDH, RPL13A GAD1 and GABRA4) were analyzed in attempts to maximize the quality of variation estimation methodology based on sample size; 3) Only the three predicted normalizers, ALG9, GAPDH and RPL13A, were analyzed to maximize estimation quality based on removal of systematic variation . As described for the geNorm analysis, all tests were repeated with and without the single MBH OVX sample responsible for most of the MBH variation omitted. High gene expression variation in the MBH also prompted additional stability analysis where the MBH was considered separately from the HPC and AMD.
1a. qRT-PCR BestKeeper analysis
Crossing point values  were examined using the BestKeeper Excel-based tool . Standard curves were used to calculate amplification efficiency (EA) according to the convention EA = 10(-1/slope). All samples were analyzed together, and in light of NormFinder analysis results highlighting GAPDH expression variation in the MBH, the MBH was analyzed separately from the AMD and HPC.
RNA extraction and preparation of cDNA from MBH, HPC and AMD was conducted as previously described [26, 52]. Both RMA and Affymetrix® Microarray Suite version 5.0 (MAS 5.0)  analyses were conducted using methods previously described [26, 52]. RMA-normalized results were analyzed using the NormFinder Visual Basic application  where identifiers were used to conduct two analyses (according to grouping) in each study. In the HRT study, expression values were grouped according to 1) brain region and 2) hormone treatment. For the cycle study, expression values were grouped according to 1) brain region and 2) to phase of the menstrual cycle.
1b. Microarray general probe set selection from the rhesus GeneChip®
A publicly available list of 575 human genes expressed under all conditions tested, and derived from publicly available microarray results, http://www.compugen.co.il/supp_info/Housekeeping_genes.html,
was used to identify genes with high likelihood of meeting criteria required to be considered as appropriate housekeeping genes . NetAffx build number 28 (March 11, 2009) for the Affymetrix® GeneChip® rhesus Macaque Genome Array was used to select rhesus macaque probe sets with RefSeq mRNA transcript IDs matching genes from the publicly available list of 575 proposed human housekeeping genes . Where no matches were found, probe sets were selected if gene symbols for housekeeping genes of interest were included in their annotation. If RefSeq mRNA transcript and gene symbol annotations for probe sets could not be found, the annotation build was searched by gene name.
1c. Microarray selection of "expressed" probe sets from the rhesus GeneChip®
Because many of the sequences obtained from the general probe set selection were poorly annotated or showed low expression, a second pair of NormFinder relative expression stability assessments, grouped by brain region or hormone treatment, was made using only probe sets with average expression levels meeting detection criteria using MAS 5.0 analysis. Probe sets meeting "expressed" criteria (824 probe sets in total) showed average signal detection of 200 (global scaling target intensity) or higher, and had present calls  in more than 50% of the animals.
2a. Microarray general probe set selection from the human GeneChip®
Using annotations from HG-U133 Plus 2.0 NetAffx build number 28 (March 11, 2009), RefSeq record numbers for mature mRNA transcripts, indicated by the prefix "NM_", from the housekeeping gene list were used to identify probe sets on the Affymetrix® U133 Plus 2.0 GeneChip®  representative of the genes in question.
2b. Microarray most representative probe set selection from the Human GeneChip®
Probe set selection using this approach was designed to maximize accuracy while minimizing redundancy in the data used to compare gene expression stability. To maximize annotation consistency between the probe set sequences and the genes listed in the public database of 575 housekeeping genes, probe sets with representative public IDs matching those in the public database list were prioritized. However, if no public ID match was found, RefSeq Transcript ID listings were used. If multiple transcript IDs were listed, or multiple probe sets had the same transcript ID, where possible, we eliminated probe sets with "x" and "s" suffixes  where probes in the probe set may have matched transcripts from different genes. In general we selected final probe sets with the highest annotation grade (A), however, to ensure consistency with original mRNA record numbers, probe sets with B-grade annotations were selected in rare instances. Where probe sets appeared equivalent, we selected the set with the most complete mRNA coding sequence (cds), or more recent submission. If complete cds sequences overlapped then sets with the embedded sequence were selected. Where annotation was unclear, the gene was not included.
1d and 2c. Microarray probe sets from "popular normalizers"
Thirteen genes from the publicly available list of housekeeping genes were designated "in popular use as reference in real-time PCR" http://www.compugen.co.il/supp_info/Housekeeping_genes.html. These 13 genes together with ALG9 and RPL13A are referred to as the "popular normalizers" in this manuscript. Because the relative expression stability values calculated by the NormFinder algorithm can be argued to be more accurate when more stably expressed genes are used, we conducted separate analyses using only the popular normalizers for comparison with the larger publicly available group. Probe sets used in analyses of the popular normalizers were selected using the "general" and "representative" methods described above.