Gene expression data were analyzed with DNA-Chip Analyzer (dChip) (www.dchip.org). Briefly, unsupervised clustering of samples and genes was performed on a filtered gene list. Filtering consisted of selecting those genes with Coefficient of Variation (CV) greater than 1.0 and with expression values greater than 20 in at least 20 percent or more of the total microarrays. This filtering generated a list of 1,747 probesets corresponding to 1,316 unique genes which defined 3 differentially expressed groups by unsupervised hierarchical agglomerative clustering. The meningiomas within the first two of three sample branches revealed highly cohesive and intense expression clusters, while the third cluster of samples contained tumors which exhibited an independent clustering signature with less intense but distinct expression across the 1,747 probesets. To identify the underlying genetic signature of this third sample cluster, we applied an ad hoc secondary filtering Coefficient of Variation greater than 0.8 to these samples and identified 3,237 probesets (2,299 genes) identifying three child subgroups within this third branch by supervised hierarchical agglomerative clustering. These methods identified a total of five gene expression subgroups based on distinct gene expression profiles. Analysis settings were set at default unless otherwise specified. Distances were set to be “Pre-calculated”, and a “1-Correlation” distance metric was employed using “Centroid” as the linkage method. Expression data from samples which demonstrated chromosomal loss or retention over minimal loss regions according to Nexus 3.0 analyses were subsequently analyzed by dChip software using 2 group comparisons over the genomic areas of interest. Common losses observed only in a specific grade of tumors were determined, subsequently the expression of the genes within these regions were compared between tumors of matched grade with and without the deletion of the specified region. The identical analysis method was repeated for common losses observed only in grade III and grade II tumors, and finally repeated for common losses observed only in grade II and grade I tumors. “Lower expressing genes” were identified as those demonstrating greater than a 1.2 decrease in expression difference with t-test p-value thresholds ≤ 0.05 and FDR ≤ 10% after 100 permutations. Pearson correlation coefficients were calculated using chromosome retention/loss calls generated by Nexus 3.0 in conjunction with the expression values exported from dChip. Gene lists were tested for enriched functional and biological themes using the MetaCore pathway web application by GeneGO, Inc., (www.genego.com/). The most significant biological gene network themes were calculated by identifying node genes present from our differential expression analyses uploaded to Metacore pathway analysis. Returned network themes were rank ordered by their hypergeometric p-values calculated during output list production. The resulting data with significance values and enrichment ratios are included in file Supplementary Information S3.