Multiple myeloma is a genetically complex and heterogeneous neoplasm in which the concurrency of multiple genomic events results in tumor development and progression. MM exists as hyperdiploid and nonhyperdiploid forms, with different karyotype ^[15][16][26,27]. Its most important oncogenic mechanisms are believed to be oncogene activation by IgH translocations and oncogene mutations ^[17][28]. IgH translocations are present in up to 50% of patients, and mainly involve five chromosomal loci, 11q13, 6p21, 4p16, 16q23, and 20q11, which contain the CCND1, CCND3, FGFR3/NSD2, MAF, and MAFB oncogenes ^[18][29].

The transcriptome of multiple myeloma has been evaluated in different patient cohorts ^{[19][20][21][22]}[30,31,32,33]. Studies based on GEP have been widely used to better understand the biology of MM by identifying the genes involved in the molecular pathogenesis of the disease and their clinical significance, to predict survival in multiple myeloma, and to identify patients who will benefit from particular types of therapy. Some groups have even made an attempt to compare the transcriptome of MM and primary plasma cell leukemia: a more aggressive form of plasma cell dyscrasia ^[23][34]. Expression profiles of differentially expressed genes are of critical importance and have provided insights into MM biology. These genes may relate to cell cycle, cell death, autophagy, kinome, stemness, cytogenetic abnormalities, chromosome 1, homozygous deletions, and immune subnetworks ^{[22][24][25][26][27][28][29][30][31]}[33,35,36,37,38,39,40,41,42].

GEP studies have led to the identification of Cyclin D family deregulation in MM and MGUS ^[19][32][33][30,43,44]. Deregulation of the cyclin D family (CCND1, CCND2, and CCND3) appears to be one of the key molecular events in the pathogenesis of MM ^[34][45]. It can result from the translocation of CCND1 or CCND3 with the IgH gene in the t(11;14) and the t(6;14), specific cyclin D amplification, trisomies, and other cytogenetic events. CCND2 is particularly overexpressed in t(4;14) and t(14;16) patients ^[19][20][30,31]. A proposed classification based on CCND1 gene expression status and 14q32 translocations divides MM patients into eight different subgroups ^[33][44].

Another attempt to use gene expression profiling in order to develop a prognostically relevant molecular classification of MM was made by Zhan et al. ^[21][32] The findings indicated the presence of seven disease subtypes that were strongly influenced by known genetic lesions including c-MAF– and MAFB-, CCND1- and CCND3-, MMSET-activating translocations and hyperdiploidy, these being CD1 [(t(11;14)], CD2 [t(11;14) and t(11;16)], MS [t(4;14)], MF [t(14;16) and t(14;20)], hyperdiploid cluster (HY), low bone disease (LB), and proliferation-associated genes (PR). Zhan et al. also identified myeloid gene expression signatures but excluded them from profiling analyses ^[21][32]. Broyl et al. confirmed the findings made by Zhan et al. and identified three novel subsets of MM: the nuclear factor kappa light chain-enhancer (NF-kB) subgroup, the cancer/testis antigen (CTA) subgroup characterized by high proliferation index, and the PRL3 subgroup characterized by up-regulation of protein tyrosine phosphatases PRL-3 and PTPRZ1 ^[20][31].

A review by Szalat et al. indicated the existence of 11 different molecular subgroups of MM based on transcriptomic studies ^[35][46]. A summary of this classification correlated with the clinical outcome is given in Table 1. Liu et al. combined data from whole-genome gene expression profiling microarrays and CytoScan HD high-resolution genomic arrays to integrate GEP with copy number variations (CNV); the findings highlighted certain molecular alterations in MM that were important for disease initiation, progression, and poor clinical outcome. In particular, eight cytogenetic driver lesions essential to the development and progression of myeloma were highlighted by the amplification of chromosome 1q: they suggest that 1q gains and the upregulated ANP32E, DTL, IFI16, UBE2Q1, and UBE2T gene expression could be responsible for MM aggressiveness ^[36][47]. These findings support those of Shaughnessy et al., who found that most of the up-regulated genes mapped to chromosome 1q, and the down-regulated genes mapped to chromosome 1p; this suggests that disease progression may be influenced by changes in the transcriptional regulation of genes mapping to chromosome 1 ^[29][40]. However, studies based on different molecular methods have yielded conflicting findings considering 1q gain as an adverse prognostic factor. Some early studies suggest it has no prognostic value ^[37][38][48,49], while some latest reports suggest it may be associated with an inferior outcome ^{[39][40][41][42]}[50,51,52,53].

Manasach et al. compared the value of retrospective GEP data with FISH criteria to identify high-risk (HR) patients. They conclude that GEP identified more HR patients than FISH. Patients reclassified from standard-risk FISH to HR GEP presented with 1q amplification of equal to or over four copies ^[43][54]. Elsewhere, a multi-tissue transcriptome-wide association study (TWAS) aimed at exploring MM biology by Went et al. ^[44][55] identified 108 genes at 13 independent regions associated with MM risk; all of these were within 1 Mb of known MM GWAS risk variants ^{[45][46][47][48]}[56,57,58,59].

It should be noted that transcriptomic approaches have rarely been employed in assessments of the risk of multiple myeloma or progression from MGUS. A number of GWAS and SNP studies have been conducted in order to explore this field, including multiple studies by the International Multiple Myeloma Research (IMMEnSE) consortium ^{[45][46][47][48][49][50][51]}[56,57,58,59,60,61,62].

3. Gene Expression Profile and Multiple Myeloma Prognosis

Many different transcriptomic models for prognostication have been identified; however, none of them have been introduced into routine clinical practice. So far, the revised International Staging System (R-ISS) is still the first choice in MM management ^[52][63], and the older Durie-Salmon staging system is still used in some places ^[53][64]. Zhan et al. performed a microarray analysis on tumor cells from 532 newly diagnosed patients with MM in order to identify high-risk disease ^[21][32]. They report that high-risk groups presented a similar gene expression profile to human MM cell lines, whereas low-risk MM groups exhibited patterns identical to MGUS and normal plasma cells. After evaluation of the 70-gene risk model in relapse samples of 51 out of 351 of the training cohort, high-risk scores associated with poor survival were found in 39 patients. Kuiper et al. identified a 92-gene signature (EMC-92) that proved to be an independent prognostic factor of survival ^[54][65]. More recently Decaux et al. proposed a risk stratification model based on 15 different genes and note that patients with high-risk MM were characterized by the overexpression of genes involved in multiple phases of the entire cell cycle ^[22][33]. Dickens et al. limited the prognostication to six genes ^[30][41]. Similarly, Botta et al. proposed a prognostic risk score based on only six genes: IFNG, IL2, LTA, CCL2, VEGFA, and CCL3 ^[31][42]. This list was acquired from a gene expression profiling dataset of MGUS, smoldering MM, and symptomatic-MM, and identified inflammatory and cytokine/chemokine pathways as the most progressively affected during disease evolution. Hose et al. proposed that assessment of proliferation by GEP allows the selection of patients for risk-adapted anti-proliferative treatment ^[55][66]. Liu et al. ^[24][35] constructed a multiple myeloma molecular causal network (M3CN) based on gene expression, copy number variation, and clinical data to better understand MM tumorigenesis, progression, and drug responses. The M3CN-derived prognostic subnetwork achieved demonstrated satisfactory separation between different risk groups ^[24][35]. However, the most complex approach was proposed by Katiyar et al. ^[56][67], who identified unified potential signatures for MM based on a genome-wide meta-analysis of differentially expressed genes (DEGs) and miRNAs (DEMs) in MM cells and normal plasma cells. The authors identified the top five most functionally connected hub genes (UBC, ITGA4, HSP90AB1, VCAM1, VCP) using protein–protein interactions. In addition, transcription factor regulatory networks were determined for five seed DEGs with four or more biomarker applications (CDKN1A, CDKN2A, MMP9, IGF1, MKI67) ^[56][67]. The above studies indicate, that DEGs may influence disease pathogenesis, clinical presentation, and drug sensitivities in MM patients. In recent years, gene expression profiling has been used to establish classifiers for prognostication. Various studies have shown that that GEP classifiers are more robust than FISH markers in identifying risk. For instance, a multivariate analysis by Kuiper et al. found that combinations of GEP with ISS, particularly SKY92 + ISS, proved superior to other combinations for stratifying MM into high-risk and low-risk categories ^[57][68].