There have been groundbreaking efforts to sequence Multiple Myeloma (MM) genomes to better define the landscape of this disease. Similar efforts to integrate multiple different 'omic' data types of clinical relevance remains a challenge. We sought to develop a systematic approach to filter the spectrum of genomic alterations through the integration of multi-platform gene expression, methylation, SNP, and microRNA microarrays with transcriptome and whole genome data, combined with clinical outcome. This approach to define clinically relevant molecular signatures in myeloma was tested using three human myeloma cell lines (HMCLs), RPMI8226, MM.1s and KMS11and compared to publically available datasets.
DNA and RNA were isolated from HMCLs and applied to array-based platforms: Illumina Omni1 Quad, Illumina Human HT12v4.0 Expression BeadChip, Illumina Infinium Human Methylation 27K and 450K, and Affymetrix MicroRNA GeneChip 2.0 following manufacturers protocols. Illumina TruSeq RNA and DNA protocols were used to generate RNA-Seq and DNA-Seq (mate-pair and paired-end) libraries that were analyzed using the Illumina HiSeq2000 instrument. A novel filter was developed to identify changes specific to each HMCL and applied to each platform, separately and combined. Genes were further filtered according to their clinical relevance based on publicly available data (GSE9782) from clinical trials using quantile survival analysis. Genomic changes predicted in the array-generated data were further examined and supported by RNA- and DNA-seq data, and validated by the use of external, publicly available MM cell line (MM genome portal; Keats 2007, Cancer Cell) and patient data (GSE26849; Chapman 2011, Nature). Statistical significance of results was set at 0.01.
We focused our analytical pipeline development on identifying genes specifically altered in KMS11, with MM1s and RPMI8226 specific and common signatures similarly defined. Based on our cell line specific filter statistic, 120 genes were identified as having KMS11-specific significant differences: 48 expression (18 over-expressed-; 23 under-expressed), 19 methylation (10 hypermethylated; 9 hypomethylated), and 53 genes featured in regions of gain (28) or loss (15). Among the 48 genes defined as having KMS11-specific expression, 12 were identified in combination with KMS11-specific methylation changes (3 over-expressed and hypomethylated; 9 under expressed and hypermethylated). Twenty of the 120 genes showed a significant association with overall survival, 5 of which showed a significant association with treatment response. Among these genes, Epidermal growth factor receptor pathway substrate 15 (EPS15) displayed a combination of increased CpG methylation, copy number loss, and low expression. We validated EPS15 expression is lowest in KMS11 using qRT-PCR and at the protein level by western blot analysis. Western blot analysis of 10 HCMLs demonstrates that EPS15 is absent in KMS11 as well as NCI-H929 and is low in OPM-2. Interestingly all three of these lines have the t(4;14). Additionally, based on publicly available aCGH and expression data, we were able to independently validate our results in both MM cell lines and patients, with a prevalence of deletion in this region in 27% of 46 MM cell lines and 16% of 234 MM patient samples. Finally, low EPS15 expression was associated with significantly shorter overall survival (OS) among MM patients treated with Bortezomib, in contrast with significantly longer OS among Dexamethasone treated patients.
We developed an analytical pipeline for integrative analyses to obtain molecular signatures in MM using multiple genomic data types. This approach may be useful for guiding treatment decisions, in addition to the identification of genomic changes associated with MM. We initially focused on EPS15, as one of a few filtered genes that showed clinically relevant expression differences to treatment. While this gene is located on 1p23, a deletion 'hotspot' in MM, the combination of its deletion and hyper-methylation, along with its expression difference association in response to current treatment for MM provide support for studying this gene in myeloma. Taken together, these results provide a template for large-scale, integrative informatics analyses of studies in MM as well as across many tumor types.
No relevant conflicts of interest to declare.
Asterisk with author names denotes non-ASH members.