RNA Isolation and Purification. Total RNA was extracted once from each anterior cingulate cortex (ACC) specimen by using TRIzol extraction (GIBCO/BRL) followed by Qiagen RNeasy kits (Qiagen, Valencia, CA). RNA was ethanol precipitated and resuspended in RNase-free sodium citrate, pH 6.5. After digestion of 100 m g of total RNA with DNase to eliminate contamination by genomic DNA, the amount of RNA was quantified using a fluorescence-based assay (Molecular Probes). Sample quality was assessed by running an RNA denaturing gel, by testing the A_{260 nm}/A_{280 nm} absorption ratio in water and by determining the 28S/18S ratio by using Bioanalyser 2100 (Agilent Technologies, Palo Alto, CA). RNA was typically intact, suggesting postmortem damage was negligible, and all samples showed A_{260 nm}/A₂80 nm ratios of at least 1.9 and 28S/18S ratios of at least 1.3.

RNA Target Preparation and Hybridization. RNA targets (biotin-labeled RNA fragments) were produced from 6-8 m g of total RNA by double-stranded cDNA synthesis followed by an in vitro transcription reaction and a fragmentation reaction. Hybridization cocktails containing fragmented cRNA, array controls (Affymetrix, Santa Clara, CA), BSA, and herring sperm DNA were prepared and hybridized to the array at 45°C for 16 h. The hybridized arrays were washed, and bound biotin-labeled cRNA was detected with a streptavidin--phycoerythrin conjugate. A subsequent signal amplification step was performed with a biotinylated antistreptavidin antibody. Washing and staining procedures were automated using the Affymetrix Fluidics Station (model 400). Each array was scanned twice by using the Gene Array Scanner 2500 (Agilent), the images overlaid, and the average intensities of each probe cell compiled.

Construction of Data Matrices. For Data Matrix A, detection calls for each species were generated by assigning as present or absent only those probe sets consistently called either present or absent among all representatives of that species. Probe sets with variable detection calls within a species, as well as those consistently detected as marginal within a species, were assigned a call of marginal. This strategy produced a four-taxon (human, chimpanzee, gorilla, and macaque) data matrix.

To construct Data Matrix B, we started with 81,420 data points, the signal values of the 10 samples for the 8,142 commonly detected probe sets. After ordering these from lowest to highest, the ordered data points were divided into bins of four percentile increments. Signal values representing the low and high point of each four-percentile increment were log₁₀ transformed. This transformation defined 25 bin categories containing equivalent numbers of data points. We then log₁₀ transformed all signal values and assigned each value to its appropriate bin category. This resulted in a data matrix comprised of 10 samples and 8,142 characters, with 25 possible character states per character. A list of the 25 bin categories and their accompanying signal values is presented in Table 5.

For Data Matrix C, we combined Data Matrices A and B to represent all 44,928 probe sets on the HG-U133 array set. The character states for the 8,142 characters (probe sets) from Data Matrix B were retained and added to a matrix with the character states for the remaining 36,786 characters (probe sets) from Data Matrix A. Character states that had been assigned in Data Matrix A on a per-species basis were further assigned to each individual (or replicate of an individual) in that species.