In all the samples a CD138-positive PC isolation using the AutoMACs separation system (Miltenyi-Biotec) was carried out. Final purity was >95% in all MM and SMM cases, and >90% in MGUS patients and healthy donors. Total RNA was extracted from normal and tumor plasma cells using miRNEasy Mini Kit (Qiagen, Valencia, USA) following manufacturer's protocol. The RNA integrity was assessed using Agilent 2100 Bioanalyzer (Agilent, Palo Alto, CA).
Labelling and hybridizations were performed according to protocols from Affymetrix. Briefly, 100-300 ng of total RNA were amplified and labeled using the WT Sense Target labelling and control reagents kit (Affymetrix Inc., Santa Clara, CA, USA)
Labeled RNA was hybridized to Human Gene 1.0 ST Array
Washing and scanning were performed using GeneChip System of Affymetrix (GeneChip Hybridization Oven 640, GeneChip Fluidics Station 450 and GeneChip Scanner 7G).
Normalization was carried out by using the expression console (Affymetrix) with RMA algorithm which includes background correction, normalization and calculation of expression values (log2). Since myeloid contamination signature can be detected even in samples with high purity, those probes identifying genes exclusive of myeloid lineage were subtracted from the analysis from the out set. Unsupervised analysis: In order to classify the samples, multidimensional scaling method (MDS) implemented with SIMFIT statistical package (version 6.4.1, available at http://www.simfit.manchester.ac.uk) was performed using Euclidean distance. In addition, an unsupervised hierarchical clustering with Euclidean distance as the distance measure, and group average as clustering method was carried out. Supervised analysis: Significant Analysis of Microarrays (SAM) algorithm (http://www-stat.standford.edu/-tibs/SAM) was used to identify genes with statistically significant changes in expression between different classes(25). All data were permutated over 1,000 cycles by using the two-class and multiclass response format, without considering equal variances. Significant genes were selected based on the lowest q-value (q-value < 10-5). Gene function analysis: The probe sets were functionally annotated and grouped according to their biological function using Gene Ontology biological process descriptions. The functional analysis to identify the most relevant biological mechanisms, pathways and functional categories in the data sets of genes selected by statistical analysis, was generated through the use of IPA (Ingenuity Systems, www.ingenuity.com). probe group file: HuGene-1_0-st-v1.r4.pgf meta-probeset file: HuGene-1_0-st-v1.r4.mps