To test the hypothesis that these wide divergences of correlation coefficients between arrays might reflect agonal factors and RNA integrity, similarity between arrays was summarized as the ACI for each of the brain regions. The ACI was calculated for each of the 40 subjects with the following steps. Step 1: Pearson’s simple correlation of expression profile, composed of 12,625 probe sets, was calculated between all 780 pairs among 40 arrays, as shown as gray-scale grids in . Step 2: in the correlation matrix, the average of the 39 correlations between one chip and the other 39 arrays was calculated for each array. Step 3: an array with the lowest averaged correlation of the 40 was determined, and this lowest averaged correlation was referred to as the ACI for this array. The array with the lowest ACI was removed from the group of 40 for the remaining calculations. Step 4: for each remaining array, the average correlation with the other 38 arrays was recalculated. Step 5: the lowest ACI of the 39 was determined, this array was removed from the group for the remaining calculations, and the averages of the remaining group were recalculated. Step 6: these iterative calculations (steps 2–5) for ACI were carried out until ACIs were obtained for all 40 arrays. All of these steps were performed with Excel software (Microsoft, Redmond, Washington).
The ACIs for the three brain regions of the 40 brains were calculated. The correlation coefficients between ACI and each of the following variables were calculated: AFS, brain tissue pH, RNA integrity indicators [percent(18S+28S), 3′/5′ ratio of GAPDH, percent present call, and degradation slope] and postmortem factors (PMI, FI). As an example, displays the results for anterior cingulate cortex. The ACI was significantly correlated with AFS and all of the RNA integrity indicators analyzed, whereas ACI showed no significant correlation with PMI and FI. These observations were similar in all three brain regions analyzed. displays the average ACI across three brain regions plotted for presence (more than 1) or absence (0) of AFS for each subject. All subjects with no agonal factors had average ACIs greater than .95, whereas of 16 subjects with an agonal factor, 15 had average ACIs of less than .95. These observations were also stable among alternative methods for data analyses. For example, different condensation programs, such as Robust Multi-Array analysis in the Affy package version 2.0 (Irizarry et al 2003
), or Spearman rank correlation.
Figure 2 Agonal factors and ribonucleic acid (RNA) degradation decrease Average Correlation Index (ACI). The ACI was calculated for the anterior cingulate cortex for 40 subjects and plotted on the y axis in the eight dot plots (A–H). The x axis shows the (more ...)
Figure 3 Average Correlation Index (ACI) of the microarray chips is a sensitive and specific indicator of agonal factors. The consistency of the ACI across three brain regions (anterior cingulate cortex, dorsolateral prefrontal cortex, and cerebellum) was examined (more ...)
Interestingly, AFS and brain tissue pH showed higher correlations with ACI than did the other RNA integrity indicators in three brain regions, which suggests that ACI can be a reliable indicator of the effects of agonal factors and brain acidosis on RNA integrity.
Taken together, agonal factors affect gene expression profile divergence more than do postmortem factors and other biological factors, including age, gender, and diagnoses of mood disorders. The large divergences in expression profiles influenced by agonal factors are associated with RNA degradation, which suggests that the divergence might be due to the differential vulnerability of mRNAs to degradation during the agonal phase. The postmortem factors seem to have less effect on integrities of mRNAs. After excluding subjects with specific agonal conditions, such as coma and hypoxia, or prolonged death, correlation coefficients between gene expression profiles of remaining subjects were consistently high. It might be feasible to detect relatively smaller effect size between psychiatric disorder and control groups by excluding subjects with the specific agonal conditions.