Impact of Genes Highly Correlated with MMSET Myeloma on the Survival of Non-MMSET Myeloma Patients

Multiple myeloma has extremely heterogeneous outcomes. Among many prognostic factors utilized in myeloma, translocation t(4;14)(p16.3;q32.3) is an oncogenic event associated with poor prognosis (1). The key molecular target of t(4;14) is multiple myeloma SET domain (MMSET) at chromosomal band 4p16.3 (2–5). The detection of MMSET overexpression with gene expression profiling (GEP) consistently identifies a high-risk subgroup in multiple myeloma (6). Although the prognostic significance of MMSET is well established, the underlying mechanism of its excess risk is poorly understood. Given that MMSET encodes histone methyltransferase, its overexpression has been attributed to alter epigenetic regulation of genes involved in cell-cycle progression and DNA damage repair (7). However, downstream gene targets and molecular pathways regulated by MMSET remain unclear.

What is also unknown in myeloma is the presence of biologic homology shared between high-risk and non–high-risk subgroups. This question comes within the context of the recent advancement of genetic sequencing, which identified diverse spectrum of disease biology that, at times, redefined conventional risk stratification and management. For instance, “BRCA-ness” was identified in up to 14% of non–small cell lung cancer and 15% of head and neck cancer patients due to epigenetic inactivation of genes responsible for DNA damage repair, such as BRCA1 and FNACF (8). In breast and ovarian cancers, next-generation sequencing demonstrating the presence of certain genes beyond BRCA1/2, such as PALB2, ATM, or CHEK2, was strongly associated with an increased risk of cancer diagnosis and early death (9–11). Recent discoveries in solid tumor suggest a substantial proportion of cancer patients harbor molecular signatures similar to those of high-risk subtypes.

We hypothesize there is an overlap of disease biology between the established high-risk myeloma and its non–high-risk counterpart. Specifically, the same genes involved in the pathogenesis and adverse outcomes of MMSET myeloma (6) could also be relevant to a subset of non-MMSET patients with poor clinical outcomes (hereby referred to as “MMSET-like myeloma”). To characterize genes and molecular pathways influencing survival across different myeloma subtypes, we assessed expression levels of 54,675 genes in 1,188 newly diagnosed multiple myeloma patients. Among 71 genes significantly altered in MMSET myeloma, 10 genes most strongly associated with survival were selected and combined into a GEP risk score. Patients who did not have detectable MMSET but were at the top quartile of the 10-gene risk score were categorized as MMSET-like myeloma. Five-year survivals were similar between patients with MMSET myeloma and MMSET-like myeloma. Pathway analysis identified MYC and TP53 transcriptional regulators as lead candidates targeted by the observed genes within the risk score. Our findings suggest there is a homology of aggressive disease biology and clinical outcomes shared between MMSET myeloma and a subset of non-MMSET myeloma.

From the NCBI Gene Expression Omnibus (GEO), we downloaded unprocessed CEL files from the following datasets: Total Therapy (TT) 2 (N = 345, accession number GSE2658, NCT00083551); TT 3 (N = 214, accession number GSE2658, NCT00081939); HOVON65/GMMG-HD4, (N = 320, accession number GSE19784, ISRCTN64455289); Myeloma IX (N = 247, accession number GSE15695, ISRCTN68454111); and Multiple Myeloma Resource Collection (MMRC; N = 288, accession number GSE26760). The sample size of each dataset was determined after excluding 8 profiles (accession number GSE19784) that were normal plasma cells and 16 patients (accession number GSE26760) who were smoldering myeloma (n = 11), MGUS (n = 2), or plasma cell leukemia (n = 3). Anonymized patient characteristics of TT trials were obtained from GEO and were identified with the same accession numbers. Anonymized patient characteristics of Myeloma IX and HOVON65/GMMG-HD4 trials were obtained through personal correspondence with M. van Duin and Ping Wu (Section of Haemato-Oncology, The Institute of Cancer Research, London, UK), respectively. Anonymized patient characteristics of MMRC were obtained from Multiple Myeloma Genome Portal (12). Selected characteristics of patients from the five studies are shown in Table 1. 

All gene expression data were derived from CD138⁺ purified plasma cells of newly diagnosed myeloma patients, which were hybridized to Affymetrix Human Genome U133 Plus 2.0 cDNA microarray. All raw CEL files were processed using the justMAS function in the R statistical programming language, and gene expression levels were log2 transformed. The final dataset included GEPs of 1,188 myeloma patients with complete data for age, sex, β2-microglobulin, and albumin. For the HOVON65/GMMG-HD4 trial, FISH data regarding MMSET status were available for 241 patients; MMSET status by FISH versus gene expression revealed a correlation of 0.81 (Spearman ρ). For the analysis of survival outcomes, we excluded 156 patients from MMRC, as it was not a clinical trial, and only used the remaining data from 1,032 patients.

Institutional Review Boards of respective institutions approved all studies. All subjects provided written informed consents approving the use of their samples for research purposes.

To classify patients into MMSET or non-MMSET myeloma, we used the previously reported microarray model using 700 gene probes to assign subjects into one of seven molecular subtypes (13). We assessed the association of MMSET myeloma with individual expression levels of 54,675 available probes. By using linear regression models for each probe and for each study, gene expression levels were dependent variables of MMSET status, age (divided by 50 years or less, 51–60 years, 61–70 years, 71 years or older), sex, and International Staging System (ISS) stage (14). For each probe, study-specific linear regression coefficients for MMSET myeloma were then combined across studies using a random-effects meta-analysis (15). Prior to finalizing the probes that were significantly associated with MMSET myeloma, all 700 gene probes used in the Arkansas model (13) were removed. We performed a random-effects meta-analysis after Bonferroni correction for multiple testing (P < 0.05/54,675 = 9.14 × 10⁻⁷). The absolute value of the random-effects slope parameter for MMSET myeloma was 2 or greater, indicating MMSET myeloma had 2-fold or greater changes in log expression of a given gene.

To identify probes relevant to survival of non-MMSET patients, results from the aforementioned meta-analysis were analyzed by a stepwise variable selection (proc phreg, SAS 9.3) in a Cox proportional hazards model. Duration of follow-up was defined by the start of treatment until death or censoring. Censoring occurred when a subject reached 5 years or was lost to follow-up. For the initial selection of probes, we included probes that passed Bonferroni correction for multiple testing, and those showed log 2 or greater changes of expression. For each probe, a minimal P < 0.1 in a marginal Cox proportional hazards model was set for initial inclusion, and a criteria of P < 0.05 was set to retain a probe in the model. All models were adjusted for age, sex, ISS stage (14), and treatment (16–18). The risk score was calculated on the basis of the adjusted Cox regression model (Appendix 1).

To assess the unbiased association of the risk score and survival, we conducted a 5-fold cross-validation (18). Briefly, the original dataset was divided into five equal parts, with equal numbers of patients from individual studies in each part. Four of the five parts were used to develop a gene signature following the aforementioned procedures (training set). The remaining fifth part was used to compute the association of the risk score and survival using Cox regression models (test set). Validation was performed five times with each part serving as a test set once. Risk scores from five test sets were median centered and combined to form an independently scored measure of risk.

To assess the stability of our results, we conducted three separate sensitivity analyses (Supplementary Tables S1 and S2). First, we used survival outcomes throughout the full follow-up time of up to 98 months instead of censoring at 60 months. Second, we excluded patients who were treated with proteasome inhibitor–based regimens, such as VTD-PACE and PAD, from the analysis. Third, patients on Myeloma IX trial were coded separately if they encountered death or censoring before the second randomization for thalidomide maintenance. In all three sensitivity analyses, the main results remained unchanged.

To determine biologic functions of the identified gene probes, pathway analysis was performed using Ingenuity Pathway Analysis (IPA) software package and the molecular signaling database from the Broad Institute (Cambridge, MA; MsigDB; ref.19). Gene networks were constructed using the upstream regulator analysis to identify transcription factors with the most interactions with selected genes (Fig. 1).

Among 1,032 myeloma patients included in this study, 139 (13.4%) had MMSET myeloma defined by GEP (Table 1; ref.13). In MMSET myeloma, the median age was 59 years (range 24–89), and 68% were males. Distributions of ISS stage I, II, and III were 48%, 30%, and 22%, respectively. Similar to prior reports (6), MMSET myeloma was associated with a higher mortality after adjusting for age, sex, ISS stage, and treatment (HR = 1.7; P < 0.001).

To determine if the same genes involved in MMSET myeloma were also relevant to the survival of non-MMSET myeloma patients, we took the following analytic approach: first, as described in Materials and Methods, we obtained GEPs of 1,188 newly diagnosed myeloma patients and defined 71 gene probes correlated with MMSET myeloma (Supplementary Table S3). From these probes, we further identified those associated with 5-year survival in non-MMSET patients and created a 10-gene risk score predictive of survival. Finally, we conducted a functional pathway analysis.

After the random-effects meta-analysis, we identified that 71 gene probes (0.13%) correlated with MMSET myeloma. The selected genes showed 2-fold or greater changes in log expressions (range 2.0 to 3.7 or −2.0 to −3.7) in MMSET patients compared with non-MMSET patients, and meta-analytic P values ranged from 1.9 × 10⁻¹¹ to 5.2 × 10⁻³⁶ (Supplementary Table S3). Genes highly correlated with MMSET myeloma included cyclin D1 (CCND1), cyclin D2 (CCND2), a transcription factor Kruppel-like factor 4 (KLF4), ubiquitin carboxyl-terminal esterase L1 (UCHL1), and α2-glycoprotein (AZGP1; Supplementary Table S3).

From the identified 71 gene probes, 10 genes were strongly associated with 5-year survival of non-MMSET patients (Table 2). AZGP1 and CCND1 were most significantly associated with survival (probe-specific HRs: 0.89–0.91 for CCND1 and 1.07–1.14 for AZGP1; P < 0.001). To define risk scores relevant to the survival of non-MMSET myeloma patients, a multivariable Cox proportional hazards model was applied to the 10 genes. Risk score groups of the first quartile (low-risk) and the fourth quartile (high-risk) were compared with non-MMSET patients in cross-validation. High-risk non-MMSET patients (hereby referred as “MMSET-like myeloma”) had a similarly increased risk of mortality [HR = 2.0; 95% confidence interval (CI), 1.5–2.8; P < 0.001) comparable with MMSET patients (HR = 2.3; 95% CI, 1.6–3.3; P < 0.001).

To characterize genes and molecular pathways influencing survival across different myeloma subtypes, we conducted analysis of the 10 genes associated with 5-year survival in MMSET-like myeloma by using IPA (Ingenuity Systems). Pathway analysis identified MYC and TP53 transcriptional regulators as lead candidates for the observed gene expression changes within the gene signature risk score (Fig. 1). TP53 was identified as a transcriptional regulator of four genes (CCND1, PTP4A3, MYBL1, and ROBO1; P = 1.9 × 10⁻³), and MYC was a transcriptional regulator of five genes (CCND1, AZGP1, PTP4A3, MYBL1, and RNF130; P = 3.1 × 10⁻⁴).

To characterize genes and molecular pathways influencing survival across myeloma subtypes, we assessed expression levels of more than 55,000 gene probes from tumor cells obtained from 1,188 newly diagnosed myeloma patients. Seventy-one genes were significantly altered in patients with the MMSET molecular subtype. Selecting from these genes, 10-gene risk score demonstrated similar 5-year survivals between MMSET myeloma and non-MMSET patients categorized as the top quartile risk score (MMSET-like myeloma). A 5-fold cross-validation was conducted to determine the unbiased association of the risk score and survival. Pathway analysis identified that MYC and TP53 transcriptional regulators were associated with the observed gene expression changes of 10 genes.

Of clinical relevance, our findings suggest an overlap of disease biology between conventionally divided groups of high-risk and non–high-risk myelomas. The study findings should be interpreted within the context of recent advancement of genetic sequencing, which refined tumor subtypes based on recurrent genetic alterations. In ovarian cancer, approximately half of the patients were found to have homologous recombination deficiency (HRD) mimicking the genetic phenotype of BRCA mutation (i.e., “BRCA-ness”; ref.20). Intriguingly, the presence of HRD in BRCA wild-type patients predicted striking sensitivity to PARP inhibition in a prospective trial (overall response rate: 32% with HRD vs. 11% without HRD), albeit less than the true BRCA-mutated group (66%). Evolving knowledge in biologic homology across different tumor subtypes proposes that a new therapeutic strategy is required to improve the outcome of patients with MMSET-like gene signature. As seen in differential responsiveness to PARP inhibition in cancers with BRCA-ness, MMSET-like subgroup may also benefit from established or investigational regimens developed for high-risk myeloma, rather than those developed for standard-risk population. Such regimens tested in high-risk myeloma include proteasome inhibitors (21–23) and other investigational agents aimed at novel targets, such as FGFR3 (5), CD38 (24), and MEK pathway (25). Further research needs to validate the role of genomic risk stratification tools to capture high-risk population and to prospectively assess clinical outcome to potential treatment options within the identified subgroup.

Another important observation of this study is the demonstration of TP53 and MYC as downstream targets of MMSET gene signature. t(4;14) accounts for 15% of myeloma population and is linked to universal overexpression of MMSET gene (3, 4). Histone methyltransferase encoded at catalytic SET domain methylates lysine residue of histone, leading to epigenetic regulation of genes involved in cell-cycle progression, p53 pathway, and integrin signaling (7). The role of MMSET as a myeloma oncogene is supported by an experimental knockdown of MMSET in myeloma cell lines, which led to decreased proliferation and increased apoptosis (2, 26, 27). Among many targets altered by MMSET overexpression, c-myc is an important downstream pathway enhanced by MMSET through downregulation of miR-126 (28, 29). The overlap of p53 and MYC pathways has also been described in model systems of other malignancies. In vitro, p53 represses c-myc transcription by deacetylation of histone located at c-myc promoter (30) and by miR-145–mediated gene silencing (31) and arrests cell cycle. These findings support the primary role of MMSET as a regulator of epigenetic machineries, rather than genetic instability, and is corroborated by findings from whole-exome sequencing, which demonstrated only a few mutational changes in the t(4;14) subgroup (32). Taken together, an aggressive clinical phenotype of MMSET overexpression is attributable to the fine-tuning of selected genes. Functional studies are required to assess direct binding or indirect modulation of 10 genes by MSMET and to validate downstream activity of MMSET-like signature converging into selected signaling pathways, such as MYC and p53.

GEP is a mature and robust technology with many validated platforms in multiple myeloma reported to date (33). Compared with previously established platforms, MMSET-like signature has several unique aspects. First, MMSET-like gene signature was developed from a biologically homogeneous population with a single genetically defined abnormality and was applied to the overall population with an aim to select patients influenced by similar pathobiology. This sequence of development is reversed from what had been done in conventional studies, which performed hierarchical gene clustering among biologically heterogeneous population (34). By using the latter method, a given gene expression group can contain several different genetic abnormalities within the subtype (6), which may have led to inconsistent results in predicting therapeutic responses (35). The 10-gene signature proposed by this study was developed from a homogenous subgroup, hence may be more representative of a single biologic entity and can serve a useful risk stratification tool for treatment trials. Second, with the exception of one gene, 10-gene signature did not overlap with previously reported platforms, such as EMC 92-gene (34), UAMS 70-gene (6), and IFM 15-gene signatures (36). This finding further supports that MMSET-like gene signature represents a distinct biologic subtype utilizing a selected set of genes. Interestingly, ROBO1 was the only gene within our 10-gene platform that was previously reported in another gene expression profile (37) and in a sequencing study as a candidate gene in myeloma (32). Downstream of ROBO1 is associated with E-cadherin–mediated regulation of WNT signaling in pancreatic cancer, and its functional role in myeloma remains to be studied.

We demonstrated 10-gene signatures that were significantly altered in MMSET myeloma and associated with inferior survival in non-MMSET myeloma patients. Pathway analysis of the MMSET-like gene signature recapitulated clustering of important signaling pathways in myeloma, specifically TP53 and MYC pathways. MMSET-like gene expression profile was able to capture a distinct biologic subtype underrepresented by conventional platforms and was strongly linked with poor clinical outcome. The proposed gene signature can serve as a reliable screening platform representative of high-risk disease biology, as we move towards personalized therapy for myeloma.

P. Sonneveld reports receiving commercial research grants from Amgen, Celgene, and Janssen and is a consultant/advisory board member for Amgen, Celgene, Janssen, and Skyline Dx. G. Morgan is a consultant/advisory board member for Bristol-Myers Squibb, Celgene, Janssen, Kesios, Novartis, and Takeda. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S.P. Wu, R.M. Pfeiffer, P. Sonneveld, O. Landgren

Development of methodology: S.P. Wu, G. Morgan, O. Landgren

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.P. Wu, P. Sonneveld, B.A. Walker, G. Morgan, O. Landgren

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.P. Wu, R.M. Pfeiffer, I.E. Ahn, S. Mailankody, P. Sonneveld, N.C. Munshi, G. Morgan, O. Landgren

Writing, review, and/or revision of the manuscript: S.P. Wu, R.M. Pfeiffer, I.E. Ahn, S. Mailankody, P. Sonneveld, N.C. Munshi, B.A. Walker, G. Morgan, O. Landgren

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.P. Wu, M. van Duin, O. Landgren

Study supervision: O. Landgren

This work was supported by the intramural research program of the NIH. Research support for S.P. Wu was made possible through the NIH Medical Research Scholars Program, a public–private partnership supported jointly by the NIH and generous contributions to the Foundation for the NIH from Pfizer Inc., The Leona M. and Harry B. Helmsley Charitable Trust, and the Howard Hughes Medical Institute, as well as other private donors. For a complete list, please visit the Foundation website at http://www.fnih.org/work/programs-development/medical-research-scholars-program.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.