Abstract

Previous data sets have defined deregulation of the RAS, RAF and MAPK pathway as being critical in Myeloma pathogenesis. More recently, the NFκB pathway has been identified as commonly deregulated. Combined with cytogenetic data these findings suggest that multiple events affecting common pathways lead to the initiation and progression of Myeloma. We employed a novel strategy to identify other deregulated pathways contributing to Myeloma progression. We hypothesize that gain of chromosome segments correlates with over-expression of genes contained within and losses result in a corresponding decrease in expression. For copy number (CN) 0, i.e. homozygous deletions, expression should be lost. Expression may not always be altered by other CN changes, e.g. hemizygous deletions (CN1) where a single copy of the gene may be sufficient to maintain the level of expression. Furthermore, the relationship between CN and expression is not expected to be linear, e.g. CN4 having twice the expression of the normal CN2, as steric interactions and feedback mechanisms play important roles in the expression of multiple copies. Our aim was to identify CN changes in Myeloma plasma cells and identify networks of gene expression altered as a consequence of CN changes. Dysregulation of one gene within a pathway can perturb the overall balance of the system, causing changes in the entire network that could contribute to tumour pathogenesis. Expression changes correlating with genes with CN changes should reflect the interactions of those genes and identify more genes that are co-regulated or directly influenced within the same networks. The CN0 and CN4 plus genes should have the largest downstream effects, i.e. complete ablation or large over-expression of a gene would perturb its networks the most. We analysed DNA and RNA from CD138+ plasma cells from 84 patients with Myeloma. The DNA was analysed using the Affymetrix 500k SNP GeneChip arrays and CN changes were assessed using DNA from paired samples of peripheral blood and inferred using dChip (2007 build). Gene expression was measured using RNA and Affymetrix HG-U133 +2 GeneChips with dChip to generate expression values. Correlation of expression was measured using the Pearson correlation coefficient and significance was assessed using permutation tests and a corrected p<0.05. Analysis of the 500k SNP data set and 54,000 expression probesets was computationally intensive. To minimize compute time the SNP results were filtered at two CN thresholds, this also reduced the overall noise in network identification. The first set was 61,887 SNP probesets that had a CN <1 (homozygous deletions) in any sample. The second was 4,570 SNPs that had a CN of 4+ in any sample. The expression results were filtered to remove probesets that were not highly expressed, log2 ratio<100 in all samples, leaving 25,420 for the analysis. Once significant correlations of CN and expression were identified, networks were constructed linking SNPs and expression by correlation; the weight of the link is inversely proportional to the magnitude of the correlation, i.e. expression and SNPs with a high correlation are closer together than lower correlations. The networks were then classified by gene; physical location, to identify co-regulations that occur in close proximity; and by pathway, to characterise known networks. The analysis identified 2 known pathways including the MAPK pathway, confirming its validity. In addition, a further 8 novel networks containing interactions of up to 37 genes were discovered.

Author notes

Disclosure: No relevant conflicts of interest to declare.