Elsevier

Translational Research

Volume 185, July 2017, Pages 47-57
Translational Research

Original Article
Identification of pathway-based prognostic gene signatures in patients with multiple myeloma

https://doi.org/10.1016/j.trsl.2017.05.001Get rights and content

Molecular profiling is used to extract prognostic gene signatures in different cancers such as multiple myeloma (MM), which is the second most common hematological malignancy. In this study, we utilized gene expression profiles to find biological pathways that could efficiently predict survival time in patients with MM. Four data sets—namely GSE2658 (559 samples), GSE9782 (264 samples), GSE6477 (147 samples), and GSE57317 (55 samples)—were employed. GSE2658 was used as a training data set and the others as validation data sets. The genes significantly associated with survival were identified using the univariate Cox proportional hazards analysis, and their roles in the biological pathways were explored using the Gene-Set Enrichment Analysis (GSEA) in the training data set. Next, the significant genes and their corresponding pathways were used to reconstruct pathway-based prognostic signatures. Thereafter, the significant gene signatures were externally validated in 3 independent cohorts—namely GSE9782, GSE6477, and GSE57317. Our results revealed that 9 pathway-based prognostic signatures were able to efficiently predict survival time in the training data set (Ps < 0.01). The testing of these signatures in the validation data sets demonstrated that 3 signatures—namely MYC targets, spliceosome, and metabolism of RNA—were able to strongly predict the clinical outcome in the 3 cohorts at P values < 0.01. In addition, in the multivariate Cox analysis, the 3 gene signatures remained as independent prognostic factors compared with the routine prognostic variables in MM—namely serum albumin, serum β2-microglobulin, and age. These signatures were by far the most powerful independent prognostic factors (MYC targets: P = 0.009, spliceosome: P = 0.024, and metabolism of RNA: P < 0.001).

Introduction

Multiple myeloma (MM) is the second most common hematological malignancy, with its worldwide incidence rates varying from 0.4 to 5 per 100,000 person-years.1, 2

The most routine prognostic system in patients with MM is the International Staging System (ISS), which uses serum β2-microglobulin (β2M) and albumin as prognostic elements.3 It has been stated that the ISS does not have enough efficacy to detect highest risk patients.3 In addition, cytogenetic aberrations such as translocations t(4; 14) and t(14; 16) and the deletion of 17p [del(17p)] have been proposed as prognostic factors capable of identifying highest risk patients.4 Combination of these systems (ie, ISS and genetics aberrations) can define more robust prognostic subtypes.4

With the advent of high-throughput technologies such as microarray and next-generation sequencing, the molecular classification of various cancers—including MM—has been extensively developed with a view to providing more precise subtypes with distinct prognosis. Through gene expression profiling, some prognostic gene expression signatures (GSs) have been proposed for MM.5, 6, 7, 8, 9, 10 For example, a 70-gene signature, developed by researchers at the University of Arkansas for Medical Sciences, indicated that the signature was associated with an increased risk of relapse and poor overall survival.6 As another model, a 15-gene signature was developed by Decaux et al (2008) which could accurately predict survival in patients with newly diagnosed MM treated with high-dose therapy. In general, in these studies, a top-down approach was employed to create prognostic signatures, where the constituting elements (genes) in each signature were merely detected using computational analyses and proposed as a prognostic signature without considering the potential relationships of the genes in each signature which manifest in biological pathways. Indeed, many biological pathways containing various genes are dysregulated (activated or suppressed) in cancers such as MM. For example, IL-6R/STAT3, Ras/MAPK, phosphatidyl/inositol3-kinase/Akt (PI3K/Akt), notch, wingless (WNT), and nuclear factor-kappa B are examples of the activated pathways in MM.11 When a pathway is activated or suppressed in a cancer, relevant genes in that pathway will majorly show similar expression patterns (upregulated or downregulated).12, 13, 14 Hence, these series of the genes that act in the same cancer-associated pathway can be used as a prognostic gene signature.12, 13, 14, 15 Ignoring biological relationships between the genes in a prognostic gene signature can negatively affect the interpretation of the findings insofar as these prognostic gene signatures lack biological meaning.16, 17 To overcome these challenges and to reduce the difficulty in biological interpretation, researchers widely draw upon some tests such as gene set analysis that combine gene expression analysis and biological knowledge.15, 18 In the present study, we sought to fill this gap and find biological pathways which could robustly predict survival time in patients with MM. Hence, we explored gene expression profiling in 559 patients with MM (training set) to find pathways that were significantly associated with the patients' survival. We found 9 pathways that efficiently predicted prognosis in patients with MM. Thereafter, we tested the significant pathway-based gene signatures in 466 patients, comprising 3 cohorts (validation data sets). Our analysis revealed that 3 pathway-based gene signatures—namely MYC targets, spliceosome, and metabolism of RNA—were able to strongly predict the patients' outcome in the validation patients at a P value < 0.01.

Section snippets

Methods

A schematic diagram depicting the analysis pipeline in our study is presented in Fig 1.

Detection of the genes associated with overall survival and their associated pathways in the training data set

The genes associated with overall survival in GSE2658 were first extracted using the univariate Cox analysis. The results revealed that 2835 out of 21,134 submitted genes were significantly associated with overall survival (P < 0.05). More details such as the corresponding z score, hazard ratio (mean and 95% confidence interval), and P values of the statistically significant genes are provided in Supplementary Table 1.

This gene list was evaluated for the involvement of the genes in the

Discussion

Previous investigations have indicated that although the ISS presents important prognostic information, it has limited efficacy to predict individual patient outcomes.10 In the present study, using a comprehensive computational and bioinformatics approach, we attempted to find the biological pathways that could efficiently predict the clinical outcome in patients with MM. Three pathway-based signatures—namely MYC targets, spliceosome, and metabolism of RNA—were modeled: They were able to

Acknowledgments

Conflicts of Interest: All authors have read the journal's policy on disclosure of potential conflicts of interest and have none to declare The manuscript has been reviewed by and approved by all named authors.

The authors wish to thank Prof Rafael Fonseca and Dr Esteban Braggio for providing clinical metadata on the GSE6477 data set.

References (55)

  • H. Makishima et al.

    Mutations in the spliceosome machinery, a novel and ubiquitous pathway in leukemogenesis

    Blood

    (2012)
  • T.N.T. Dai et al.

    No recurrent mutation of SF3B1, U2AF1 and SRSF2 spliceosomal gene in multiple myeloma but polymorphisms of these genes are associated with the prediction of worse prognosis

    Blood

    (2012)
  • P.R. Greipp et al.

    International staging system for multiple myeloma

    J Clin Oncol

    (2005)
  • K. Neben et al.

    Combining information regarding chromosomal aberrations t(4;14) and del(17p13) with the International Staging System classification allows stratification of myeloma patients undergoing autologous stem cell transplantation

    Haematologica

    (2010)
  • R. Kuiper et al.

    A gene expression signature for high-risk multiple myeloma

    Leukemia

    (2012)
  • T.-H. Chung et al.

    A novel measure of chromosome instability can account for prognostic difference in multiple myeloma

    PLoS One

    (2013)
  • J. Moreaux et al.

    A high-risk signature for patients with multiple myeloma established from the molecular classification of human myeloma cell lines

    Haematologica

    (2011)
  • O. Decaux et al.

    Prediction of survival in multiple myeloma based on gene expression profiles reveals cell cycle and chromosomal instability signatures in high-risk patients and hyperdiploid signatures in low-risk patients: a study of the Intergroupe Francophone du Myélome

    J Clin Oncol

    (2008)
  • H. Itadani et al.

    Can systems biology understand pathway activation? gene expression signatures as surrogate markers for understanding the complexity of pathway activation

    Curr Genomics

    (2008)
  • K.A. Hoadley et al.

    EGFR associated expression profiles vary with breast tumor subtype

    BMC Genomics

    (2007)
  • L.H. Saal et al.

    Poor prognosis in carcinoma is associated with a gene expression signature of aberrant PTEN tumor suppressor pathway activity

    Proc Natl Acad Sci USA

    (2007)
  • Y.-H. Chang et al.

    Pathway-based gene signatures predicting clinical outcome of lung adenocarcinoma

    Sci Rep

    (2015)
  • A. Subramanian et al.

    Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles

    Proc Natl Acad Sci USA

    (2005)
  • J.-H. Hung

    Gene Set/Pathway enrichment analysis

    Methods Mol Biol

    (2013)
  • D.B. Allison et al.

    Microarray data analysis: from disarray to consolidation and consensus

    Nat Rev Genet

    (2006)
  • A. Floratos et al.

    geWorkbench: an open source platform for integrative genomics

    Bioinformatics

    (2010)
  • S.-K. Kim et al.

    APPEX: analysis platform for the identification of prognostic gene expression signatures in cancer

    Bioinformatics

    (2014)
  • Cited by (6)

    All authors have read the journal's policy on conflicts of interest, and all authors declared no conflicts of interest.

    View full text