A centromere-associated gene score for rapid determination of risk in multiple myeloma

Abstract

Risk stratification in patients with multiple myeloma (MM) remains a challenge. As clinicopathological characteristics have been proven deficient for accurately defining risk stratification, molecular markers have gradually become the focus of interests. This study investigated the expressions of centromere-associated genes in MM patients, their potential as prognostic markers, and their roles in disease progression. Several cohorts of 2301 MM patients were enrolled and gene expression profiling (GEP) was used to screen for CENP-A through CENP-W. Correlations between centromere-associated genes and clinicopathological characteristics, proliferative activity and recurrence of MM patients were analyzed. Clinically, CENP-E/H/K/L/N/U/W expressions were present at high-risk MM, which were even stronger elevated in patients with high tumor burden and recurrence. Mechanistically, CENP-E/H/K/L/N/U/W and FOXM1 were positively expressed in MM patients, which play synergistic or additive effects in clinical outcome. Furthermore, CENP-E/H/K/L/N/U/W were used to construct a centromere-associated gene score (CGS) model, which proved to be strongly prognostic values in several independent cohorts compared to usual clinical prognostic parameters using multivariate Cox analysis. Patients in the CGS low-risk group were significantly related to better clinical outcome than those in high-risk group. In this study, we provided proof-of-concept that CENP-E/H/K/L/N/U/W have critical roles in MM patients’ progression and prognosis. The CGS model validated in different datasets clearly indicated novel risk stratification for personalized anti-MM treatments.

Introduction

Multiple myeloma (MM) is plasma cell malignancy that proliferates in the bone marrow, and the progressive plasma cell is characterized with recurrent gene translocations, deletions or gains and changes [1-3]. Gene expression signatures made it possible to identify gene expression in myeloma cell linked with progression free and/or overall survival (PFS/OS) of MM patients. Zhan et al. identified 8 genetic subgroups of MM [4]; Subsequently, Shaughnessy et al. established a 70-gene risk scoring system able to divide 13% of MM cases into high-risk group [5]; Later, Decaux et al. developed an IFM15 risk stratification, which classify 25% of MM cases as high-risk [6]. These risk scoring systems included abundant genes coding for proteins involved in multistep processes of universal aneuploidy and recurrent chromosomal aberrations, which is considered to be associated with chromosomal instability (CIN). CIN can allow the rapid accumulation of changes that promote myeloma progression, growth and heterogeneity, and contribute to intrinsic and acquired drug resistance [7,8]. Therefore, CIN-related biomarkers that can predict the incidence of progression and/or recurrence are clinical priority for MM risk stratification.

The exact causes of CIN in most cancers remain unclear. Proposed mechanisms include oncogene-induced replication stress, breakage-fusion-bridge cycles induced by telomere translocations, dysfunction and aberrant mitosis [9]. Another important mechanism involves centromeres and their associated kinetochores [10]. In particular, the constitutive centromere associated network (CCAN) are required for proper spindle attachment, chromosome congression, mitotic checkpoint activity and separation of sister chromatids during mitosis, leading to the assembly of a functional kinetochore [11]. The CCAN network is comprised of CENP-A/B/C/F/I/J/M/O/P/Q/T/V/E/H/K/L/N/U/W [12]. Previous studies showed that centromere-associated genes are detected in a variety of solid tumors and myeloma. In solid tumors, Zhang et al. demonstrated that centromere and kinetochore gene misexpression predicts cancer patient survival and response to radiotherapy and chemotherapy [13]; In MM, Kryukov et al. focused on centrosome-related genes (CAGP model: CENP-A, CENP-E and so on) that reveal the molecular heterogeneity characteristics and survival for MM patients [14,15]. Nevertheless, in spite of advances in centromere, most other centromere-associated genes undefined abnormalities forming genetic complexity in MM may still exist, and none of these studies focused on MM risk stratification for all centromere-associated genes.

On these bases, we investigated centromere-associated gene signature able to distinguish the different stages of myeloma progression, and constructed a risk stratification model based on centromere-associated gene score (CGS) in MM. As a result, CGS was demonstrated to be an efficient model in prediction of clinical outcome, and enhanced our understanding of CIN in MM risk stratification.

Materials and methods

Gene expression profiling (GEP) and data analysis

Gene Expression Omnibus (GEO) database was carried out to measure the expressions of CENP-E/H/K/L/N/U/W in 2301 MM patients (GSE5900 [16], GSE2658 [4], GSE24080 [17], GSE31161 [18] and GSE9782 [19]). Data acquisition and normalization methods in above datasets have been described previously [17]. The expressions of CENP-E/H/K/L/N/U/W in plasma cells were determined using the Affymetrix U133Plus2.0 microarray (Affymetrix, USA), which was performed as previously described [4].

Statistical analysis

Various statistical analyses were utilized to assess the roles of CENP-E/H/K/L/N/U/W on clinical features and prognosis of MM patients. Two-tailed Student’s t-test and One-way analysis of variance were adopted to compare two or multiple experimental groups. The Fisher’s test was used to compare clinicopathological features between the high/low expressions of CENP-E/H/K/L/N/U/W. Survival curves were plotted according to the Kaplan-Meier method, and the log-rank test was employed to analyze statistical differences between survival curves. The effect of CENP-E/H/K/L/N/U/W on outcome was analyzed using univariate and multivariate Cox proportional hazard models. GraphPad Prism 6 software was employed for our analyses and *P < 0.05 was considered significant.

Results

CENP-E/H/K/L/N/U/W were high-risk myeloma genes

To evaluate the possibility that centromere-associated genes are crucial for myeloma, we examined centromere-associated gene expression in normal plasma (NP), smoldering multiple myeloma (SMM), monoclonal gammopathy of undetermined significance (MGUS) and myeloma cells using GEP database. Notably, CENP-E/H/K/L/N/U/W expressions increased significantly from NP, MGUS, SMM to MM samples (Figure 1, Supplementary Figure 1). In detail, we asked whether heightened CENP-E/H/K/L/N/U/W expressions might be related to a particular molecular subgroup of myeloma. Figure 1 presented the CENP-E/H/K/L/N/U/W expressions in 8 widely recognized subgroups, showing that elevated CENP-E/H/K/L/N/U/W expressions are particularly prevalent in 3 known to confer high-risk in terms of clinical course and prognosis: MAF/MAFB (MF), MMSET/FGGR3 (MS) and Proliferation (PR). These findings led us to conclude that CENP-E/H/K/L/N/U/W are high-risk myeloma genes.

Figure 1.

CENP-E/H/K/L/N/U/W were high-risk myeloma genes. (Upper row) CENP-E/H/K/L/N/U/W expressions of NP (n = 22), MGUS (n = 44), SMM (n = 12) and MM (n = 559) in GSE5900 and GSE2658 datasets. (Lower row) scatter-plots showed CENP-E/H/K/L/N/U/W expressions in eight MM subgroups (CD1 and CD2 subgroups with spiked expression of CCND1 and CCND3; PR, Proliferation; LB, Low-bone disease; HY, Hyperdiploid; MS, MMSET; MF, MAFB; MY, Myeloid) (*P < 0.05, **P < 0.01, ***P < 0.001).

Correlations between CENP-E/H/K/L/N/U/W expressions and clinicopathological characteristics

To evaluate CENP-E/H/K/L/N/U/W expressions in MM patients, we divided MM patients into two categories according to their CENP-E/H/K/L/N/U/W expressions (low/high expression, using the 50th percentile as cut-offs). The clinicopathological characteristics according to CENP-E/H/K/L/N/U/W expressions were listed in Supplementary Tables 1, 2, 3, 4, 5, 6, 7. No significant correlations were observed between CENP-E/H/K/L/N/U/W expressions and other clinicopathological features such as sex, age, aspirate plasma cells (ASPC) and bone marrow biopsy plasma cells (BMPC). High CENP-E/H/K/L/N/U/W expressions were significantly associated with low serum albumin (ALB) and serum haemoglobin (HB) levels. On the contrary, high CENP-E/H/K/L/N/U/W expressions were also significantly associated with high β2-Microglobulin (β2-MG), C-reactive protein (CRP), creatinine (Creat), lactate dehydrogenase (LDH) and MRI focal lesions levels (Figure 2 and Supplementary Tables 1, 2, 3, 4, 5, 6, 7).

Figure 2.

Correlations between CENP-E/H/K/L/N/U/W expressions and clinicopathological characteristics. A-E. High CENP-E/H/K/L/N/U/W expressions were significantly associated with high β2-MG, CRP, LDH, Creat and MRI focal lesions levels. F. High CENP-E/H/K/L/N/U/W expressions were significantly associated with low serum ALB level (*P < 0.05, **P < 0.01, ***P < 0.001).

CENP-E/H/K/L/N/U/W were linked to disease progression and relapse in MM

To validate our findings, we also evaluated the efficiency of centromere-associated genes in myeloma cell proliferation (Figure 3A). CENP-E/N/U expressions were positively correlated (r = 0.6643; r = 0.6964; r = 0.6134; P < 0.0001) with cell proliferation in 246 bortezomib-treated MM patients available at GSE9782, using the gene expression-based proliferation index (GPI) of myeloma devised by Mayo Clinic as proxy of effective tumor cell proliferation [20]. In addition, CENP-E/H/K/L/N/U/W expressions significantly increased in the relapsed MM patients from TT2 and TT3 cohorts compared to baseline patients in GSE31161 (Figure 3B). These data strongly suggested that CENP-E/H/K/L/N/U/W expressions could be adopted in the evolution of myeloma progression and relapse.

Figure 3.

CENP-E/H/K/L/N/U/W is linked to disease progression and relapse in MM. A. Scatter plots demonstrating positive correlation of CENP-E/H/K/L/N/U/W expression and myeloma proliferation in 246 bortezomib-treated patients from the Mayo Clinic. Tumor cell proliferation was scored with the assistance of a gene expression-based proliferation index (GPI) developed by Bergsagel et al. B. The expressions of CENP-E/H/K/L/N/U/W were significantly up-regulated in relapsed patients from TT2 and TT3 cohort in comparison with baseline patients (*P < 0.05, **P < 0.01, ***P < 0.001).

Increased CENP-E/H/K/L/N/U/W expressions correlated with poor prognosis in MM

To assess the survival time with CENP-E/H/K/L/N/U/W expressions in MM, we divided all MM patients into two groups based on high/low CENP-E/H/K/L/N/U/W expressions, The high CENP-E/H/K/L/N/U/W expression groups had shorter median OS and PFS time than low expression groups. As shown in Figure 4, MM patients with strong CENP-E/H/K/L/N/U/W expressions had an inferior OS and PFS. Additionally, we used the univariate cox analysis to evaluate CENP-E/H/K/L/N/U/W expressions on clinical outcomes, CENP-E/H/K/L/N/U/W resulted independently associated with survival (Table 1). To further understand the regulatory mechanisms of CENP-E/H/K/L/N/U/W, CENP-E/H/K/L/N/U/W associated with predicted targeted genes were analyzed using GEO database. It was identified that CENP-E/H/K/L/N/U/W expressions were highly correlated with transcription factor FOXM1 expression (r > 0.30, *P < 0.05; Figure 5A, Supplementary Figure 2). To confirm this hypothesis, by combining CENP-E/H/K/L/N/U/W and FOXM1, we found that MM patients with high expression (cutoff: 50%, high vs. low) of CENP-E/H/K/L/N/U/W and FOXM1 simultaneously had the worst prognosis compared to the patients with low expression of two genes together and the rest of the MM patients (medium) for both OS and PFS in GSE24080 (Figure 5B).

Figure 4.

Increased CENP-E/H/K/L/N/U/W expressions correlated with poor prognosis in MM. Kaplan-Meier analyses showed OS (Upper row) and PFS (Lower row) of 559 newly diagnosed MM patients.

Table 1.

Univariate Cox regression analyses for OS and PFS in 559 MM patients

Variables	OS			PFS
	HR	95% CI	p	HR	95% CI	p
CENPEhigh	1.590	1.174-2.153	0.003	1.450	1.137-1.876	0.004
CENPHhigh	1.767	1.297-2.408	0.000	1.378	1.072-1.771	0.012
CENPKhigh	1.499	1.105-2.035	0.009	1.377	1.071-1.770	0.013
CENPLhigh	1.712	1.261-2.326	0.001	1.449	1.128-1.881	0.004
CENPNhigh	1.348	0.998-1.822	0.052	1.362	1.081-1.749	0.015
CENPUhigh	1.159	0.859-1.564	0.334	1.209	0.943-1.551	0.134
CENPWhigh	1.355	1.003-1.832	0.048	1.292	1.007-1.658	0.044

Figure 5.

Relationship between the expressions of CENP-E/H/K/L/N/U/W and FOXM1 in MM. A. Pearson’s correlations between the transcript level patterns of CENP-E/H/K/L/N/U/W and predicted targeted genes (*P < 0.05). B. Survival analyses were performed based on the combination of CENP-E/H/K/L/N/U/W and FOXM1 expressions. Kaplan-Meier showed OS and PFS curves of GSE24080.

Construction of a centromere-associated gene score model

We added a score to CENP-E/H/K/L/N/U/W (high expression = 1 and low expression = 0) and then constructed centromere-associated gene score (CGS) model as follows: CENPE + CENPH + CENPK + CENPL + CENPN + CENPU + CENPW. The CGS model could assume 8 different values and according to 50th percentile, patients were divided into three groups: low-risk (LR) = CGS 0-1, intermediate-risk (IR) = CGS 2-5 and high-risk (HR) = CGS 6-7. Then, we calculated the CGS of each MM patient in GSE24080. All patients were divided into CGS^LR group (n = 150), CGS^IR group (n = 272) and CGS^HR group (n = 137) according to their risk fraction (Figure 6A). As a result, CGS model was strongly related to survival, with patients in CGS^LR group having better OS and PFS compared to CGS^HR group in GSE24080 (Figure 6B and 6C). Additionally, we used the univariate and multivariate cox analysis to evaluate CGS and clinicopathological characteristics on clinical outcomes (Tables 2 and 3). The OS was decreased for CGS^HR group versus CGS^LR+IR groups (HR = 1.401, 95% CI: 1.008-1.948, P = 0.045), as well as PFS (HR = 1.379, 95% CI: 1.041-1.825, P = 0.025). To confirm the robustness of the CGS, we tested CGS model in predicted clinicopathological parameters distribution. Using 6 clinicopathological parameters, we identified different distribution among risk subgroups in 559 patients. Respectively, the levels of β2-MG, CRP, LDH, Creat and bone lesions were significantly increased in CGS^HR group compared to CGS^LR group. In contrast, ALB was obviously decreased in CGS^HR group (Figure 6D).

Figure 6.

The correlations between CGS model and disease progression. A. Heat map (upper row) reporting probe fluorescence intensity of 7 selected genes for each patient evaluated in accordance with its survival, CGS risk score (lower row). B, C. The CGS^HR group identified MM patients with the lowest OS and EFS in GSE24080. D. The CGS^HR group was significantly associated with high β2-MG, CRP, LDH, Creat and MRI focal lesions levels. In contrast, CGS^LR group was significantly associated with high serum ALB level.

Table 2.

Univariate and Multivariate Cox regression analyses for OS in 559 MM patients

Variables	Univariate model			Multivariate model
	HR	95% CI	p	HR	95% CI	p
Age ≥ 65 yr	1.206	0.855-1.700	0.286
Male sex	0.968	0.714-1.313	0.835
β2-MG ≥ 3.5 mg/L	2.185	1.613-2.958	0.000	1.867	1.330-2.647	0.000
Creat ≥ 1.2 mg/dL	1.731	1.278-2.345	0.000	1.210	0.862-1.699	0.271
CRP ≥ 4 mg/L	1.539	1.132-2.092	0.006	1.353	0.985-1.859	0.062
ALB ≥ 3.5 g/dL	0.521	0.360-0.756	0.001	0.704	0.478-1.035	0.074
CGSHR	1.583	1.147-2.185	0.005	1.401	1.008-1.948	0.045

Table 3.

Univariate and Multivariate Cox regression analyses for PFS in 559 MM patients

Variables	Univariate model			Multivariate model
	HR	95% CI	p	HR	95% CI	p
Age ≥ 65 yr	1.138	0.853-1.518	0.379
Male sex	0.990	0.768-1.275	0.936
β2-MG ≥ 3.5 mg/L	1.903	1.482-2.445	0.000	1.773	1.329-2.364	0.000
Creat ≥ 1.2 mg/dL	1.469	1.136-1.889	0.003	1.087	0.813-1.455	0.573
CRP ≥ 4 mg/L	1.290	1.002-1.659	0.048	1.154	0.890-1.496	0.280
ALB ≥ 3.5 g/dL	0.665	0.477-0.927	0.016	0.827	0.586-1.165	0.277
CGSHR	1.493	1.135-1.963	0.004	1.379	1.048-1.825	0.025

Evaluation of the CGS model in different datasets

We validated CGS model in other independent datasets, Figure 7A presented the values of CGS in 8 genetic subgroups of MM, showing that increased CGS is particularly distributed in high-risk subgroups. In sync with that, we observed a significantly increasing between the Zhan et al. defined two risk categories (low-risk groups: CD1 + CD2 + LB + HY + MY vs. high-risk groups: MF + MS + PR; 3.175 ± 0.014 vs. 4.351 ± 0.256, P < 0.0001). In addition, we utilized the Kaplan-Meier analysis to validated CGS model in two independent datasets, and the Kaplan-Meier survival analysis indicated that CGS^LR group had better OS compared to CGS^HR in TT2 (induction therapy: D(T)-PACE, Dex with or without thalidomide) and TT6 (autologous hematopoietic stem cell transplant) cohorts (Figure 7B and 7C).

Figure 7.

Validation of the CGS model in independent datasets. A. A scatter-plot showed CGS in eight MM subgroups. B. The CGS^HR group identified MM patients with the lowest OS in GSE2658. C. The CGS^HR group identified MM patients with the lowest OS in GSE57317.

Discussion

MM remains incurable despite novel treatments, and plenty of prognostic markers that reflect tumor- or host-related factors have failed to explain thoroughly the heterogeneity in clinical outcomes [21]. Therefore, it is important to stratify risk stratification for MM patients. With advances in MM study, several prognostic systems were constructed using previously reported prognostic parameters [22,23]. However, these prognostic factors could not completely reflect the real prognostic condition of MM patients. Thus, evaluating a novel and powerful MM prognostic model is crucial for predicting the prognosis and determining personalized anti-MM treatment.

In the present study, CENP-E/H/K/L/N/U/W were significantly higher expressed in aggressive subgroups of myeloma (MF, MS and PR), which are characterized by high-risk MM and associated with an adverse prognosis [4,24]. We also analyzed the prognostic significance of CENP-E/H/K/L/N/U/W in MM and correlated with markers of myeloma activity, such as lower levels of ALB, higher levels of β2-MG, Creat, LDH CRP and MRI focal lesions. Among them, International Staging System (ISS) has been constructed which combines biomarkers of tumor burden (ALB and β2-MG) with biomarkers of aggressive myeloma biology (bone lesions and LDH) [25,26]. ALB and renal function have been considered easy and good indicators of survival [27]. The serum level of β2-MG is one of the most important independent predictors of survival and considered an indicator of tumor burden [28]. High levels of circulating LDH enhance myeloma cell proliferation and drug resistance under stressed conditions, and correlate with poor prognosis in myeloma [29-31]. Another interesting finding in this study is that the CENP-E/H/K/L/N/U/W expressions appear to correlate with response to dexamethasone or bortezomib-based chemotherapy. High-dose dexamethasone is commonly used for myeloma treatment [32]. Bortezomib, which targets the 26S proteasome subunit β5, has induced a high level of positive response rates [33,34]. However, toxicities associated with global proteasomal inhibition and resistance to bortezomib or dexamethasone in MM are major concerns, prompting the further development of novel target and therapies. A great deal of variance was exhibited in CENP-E/H/K/L/N/U/W, and suggested new potential mechanisms of therapeutic molecules. More importantly, our results supported the fact that CENP-E/H/K/L/N/U/W might have prognostic values, and high expression groups had significantly shorter OS and PFS.

Following bioinformatics analysis, the present study identified that CENP-E/H/K/L/N/U/W had significant correlations with FOXM1 expression, and FOXM1 is also highly expressed in MM [35,36]. In addition, the overall survival rate of patients with high expression of FOXM1 was worse. However, there was no investigation between the survival trend of FOXM1 and the survival trend of CENP-E/H/K/L/N/U/W. FOXM1 is a transcription factor that participates in all stages of biological functions, including cell proliferation and cell cycle, DNA damage repairs and cell self-renewal, which are involved in tumor progression and the response of chemotherapy [37,38]. In regard to different biological functions attributable to FOXM1 in MM, the transcription factor seems to resemble well-established “master” transcription factors, such as IRF4 and MYC [39,40]. Therefore, the present study hypothesized that CENP-E/H/K/L/N/U/W may be involved in FOXM1 regulatory network of MM (Supplementary Figure 2).

The above results provided stable support for the centromere-associated gene signature in the biologic function of myeloma cells. On these bases, we constructed a prognostic risk score with MM patients classified into three risk groups. Firstly, we analyzed the prognostic significance of CGS model in MM, CGS^HR group correlated with markers of myeloma activity, such as lower levels of ALB and HB, higher levels of LDH, CRP, bone lesions and β2-MG. More importantly, CGS^HR group correlated significantly to all the aforementioned parameters of disease activity, which support the fact that CGS model might have prognostic value. Next, the scatter plot showed that the CGS model was similar to 8-subgroup model among all groups. Aggressive subgroups of myeloma also had significantly higher CGS compared to all other molecular subgroups. At last, we analyzed the correlations between gene expression and clinical outcome based on CGS model in independent datasets. Our result showed that there was significant difference in the survival conditions of CGS^HR, CGS^IR and CGS^LR patients. Univariate and multivariate Cox proportional hazard regression analyses were then performed to verify the association of clinicopathological parameters and CGS model with survival. Our results further testified that the CGS^HR is an independent prognostic factor.

In conclusion, our results demonstrated the prognostic and predictive power of the CGS model, suggested a role for centromere misregulation in MM progression. Incorporation of CGS model into risk determination algorithms for newly-diagnosed MM patients will facilitate the development of CIN-targeted treatments.

Disclosure of conflict of interest

None.