Introduction

DNA flow cytometry detects aneuploidy in 70–80% of patients with multiple myeloma (MM).1 Hypo-diploidy has been associated with poor prognosis in patients treated with VAD (vincristine, doxorubicin and dexamethasone)2 that was overcome by the use of high-dose melphalan.3 In contrast, hyperdiploidy has been associated with more favorable outcomes.4, 5 Here we have investigated, as part of Total Therapy 3b,6 the prognostic implications of two-color flow cytometry of nuclear DNA and cytoplasmic immunoglobulin (DNA/CIG) parameters in the context of all standard prognostic variables and plasma cell-based gene expression profiling (GEP).

Patients and methods

Treatment, staging and clinical endpoints

The details of the TT3b trial and clinical outcomes have been reported previously.6 Briefly, 177 eligible patients with newly diagnosed MM fulfilling CRAB criteria7 were enrolled, including 26 with one cycle of prior therapy. The protocol consisted of two induction cycles with VTD-PACE (bortezomib, thalidomide, dexamethasone and 4-day continuous infusions of cisplatin, doxorubicin, cyclophosphamide and etoposide) with hematopoietic progenitor cell collection upon recovery from the first cycle. Melphalan 200 mg/m2 was applied with each of the planned two transplants, with dose adjustments for age and renal function.8 Consolidation employed dose-reduced VTD-PACE for two cycles. Maintenance with VRD (bortezomib, lenalidomide and dexamethasone) was planned for 3 years. In compliance with the institutional, federal and Helsinki declaration guidelines, all patients provided written informed consent before enrollment into the protocol that had been approved by the institutional review board.

All patients underwent a standardized staging workup. Bone marrow examinations included DNA/CIG, metaphase karyotyping to document the presence of cytogenetic abnormalities and GEP of purified plasma cells to assign molecular subclass,9 risk according to 70 (ref. 10) and 80 gene models,11 GEP-defined 1q21 amplification (amp1q21) as well as proliferation index9 and centrosome index.12 Clinical endpoints included the frequency of complete response13 and its duration counted from complete response onset to progression or death from any cause. Overall survival (OS) and progression-free survival (PFS) were measured from start of protocol therapy until progression or death from any cause for PFS and death from any cause for OS. Outcome data were updated as of 21 February 2014.

DNA/CIG assay

As part of the diagnostic workup, DNA/CIG was performed in all Total Therapy (TT) protocols with continuous updates on hardware and methodology. A modification introduced in August 2006 on the doublet discrimination method14 increased accuracy and reproducibility of results and has been uniformly applied with the start of TT3b enrollment. Details of the DNA/CIG method have been published.1 Briefly, bone marrow aspirates were separated by Hypaque-Ficoll (Sigma Aldrich, St Louis, MO, USA) gradient centrifugation, erythrocytes lysed with ammonium chloride and samples submitted to overnight ethanol fixation. Single-cell suspensions were exposed to anti-light chain reagents (Dako Kappa and Lambda light chain (Agilent Technologies/Dako, Glostrup, Denmark) F(AB)2/FITC conjugated) and then counterstained for DNA with propidium iodide with the addition of RNase. Acquisition and analysis of the flow cytometric signals for the derived parameters were done through a BD FACScan Flow Cytometer (Beckton, Dickinson and Company, Franklin Lakes, NJ, USA) and the CellQuest/CellFit software (Beckton, Dickinson and Company). Routinely, a total of at least 10 000 events were recorded and analyzed. Assays with fewer than 500 events were rejected. To ensure maximum reproducibility of results, the same instrument was used for all measurements. The instrument was standardized daily with DNA Check Beads (Beckman Coulter, Inc., Brea, CA, USA) for consistent channel settings and coefficient of variation requirements of <3%. A known positive patient specimen for each light chain was run daily and percent positive and light chain intensity were recorded. Titrated polyclonal F(AB’)2 antibodies for light chains were used for low nonspecific binding, and excellent lot-to-lot reproducibility was documented. To quantitate the cellular DNA content, the DNA index (DI)15 was determined and calculated as the ratio of the mean for each light chain-positive G0/1 DNA peak divided by the mean of the light chain-negative diploid G0/1 peak on the x-axis.

Acquisition of the G0/G1 populations was done through the modified doublet discrimination method14 and the CellQuest/CellFit software. A DI between 0.99 and 1.01 was referred to as diploid, whereas hyperdiploid implied DI >1.01 and hypodiploid DI <0.99. The excess of kappa- or lambda-positive cells identified the involved or light chain-restricted (LCR) cell population, the percentage of which was calculated in relation to the total number of gated events. Among the LCR cell population, discrete populations of cells with different nuclear DNA content were identified, which we refer to from here on as DNA stem lines, and their respective percentage could be calculated by referral to the total number of gated events. The involved DNA stem line with the highest percentage was considered dominant. The ploidy status was characterized from the DI of the dominant LCR DNA stem line. To quantitate the cytoplasmic immunoglobulin content of a light chain-positive population, the cytoplasmic immunoglobulin index (CIg) was used and calculated from the ratio of the geometric mean of the y-axis (cytoplasmic immunoglobulin fluorescence intensity) for the light chain-positive G0/1 peak divided by the y-axis geometric mean of the light chain-negative diploid G0/1 population. The CIg of each distinct DNA stem line was calculated as explained above. An example of a kappa-positive hyperdiploid MM with two distinct stem lines along with a case of high and a case of low CIg are shown in Supplementary Figures 1 and 2.

There was absolute concordance between the light chain classification of the LCR population by FDC and the conventional serological methods. In addition, the dominant CIg correlated with the ratio of M-protein to the percentage the dominant stem line (RS=0.621, P<0.001). Although the DNA/CIG method described here does not discriminate between mature B cells and plasma cells, it does include all the myeloma cell subpopulations that have either an aberrant phenotype16 or a dim expression of the selected antigen17, 18 or that even belong to the rare category of nonsecreting and nonproducing myeloma cells.1 When multiparameter flow cytometry was performed to identify LCR non-plasma B cells, their percentage was consistently found to be <1%.19

Statistical Analysis

Kaplan–Meier methods were used to generate survival distribution graphs, and comparisons were made employing the log-rank test. The Pearson χ2-test was used for categorical comparisons, whereas Student’s t-test and Mann–Whitney U-test were used to compare the means or medians, respectively, of two different populations. Spearman’s rank correlation coefficient (RS) was used as a measure of association between the ranks of two variables. For continuous variables, the running log-rank method was applied for the calculation of optimal cutoff points.20 Stepwise selection and Cox proportional hazard regression modeling were applied in multivariate analyses. The R2 statistic was used to evaluate the predictive power of different models.21 For the identification of differentially expressed gene probe sets between dichotomized groups, the Wilcoxon’s rank sum test of significance analysis of microarrays22 was used with an adjustment of a false discovery rate (or q-value) of <10% to be considered significant. The logarithmic base 2 expression levels of the gene probe sets were used in the analyses. Microarray data used in this study have been deposited in the NIH Gene Expression Omnibus under accession number GSE2658. A modified approach to the ComBat method23 was used to transform HOVON gene expression data to the same scale as TT3b while keeping the TT3b gene expression data fixed.

Results

Standard baseline characteristics were available in 173 of 177 patients enrolled; in addition, 166 had GEP and 143 had DNA/CIG data. Herein we report on the 139 patients with complete data sets for both DNA/CIG and GEP analyses (Table 1). Standard variables and GEP data did not differ from the larger patient sets (data not shown) but, compared with earlier TT trials, cytogenetic abnormalities (42%) and GEP-70-defined high risk (23%) were more frequent. Aneuploidy was detected in 88%. DNA stem line frequencies were 1 in 18%, 2 in 70% and >2 in 12%. In case of multiple LCR DNA stem lines, the designations of hyperdiploid applied to 58%, diploid to 38% and hypodiploid to 4%. There was absolute concordance between the light chain classification of the LCR population by FDC and the conventional serological methods. Moreover, there was a substantial correlation (RS=0.621, P<0.001) between the dominant CIg and the ratio of M-protein to the percentage of that stem line.

Table 1 Patient baseline characteristics
Full size table

Clinical outcomes for the 139 patients of the TT3b study are shown in Supplementary Figure 3. In a univariate analysis, 4-year estimates were 73% for OS, 67% for PFS and 69% for complete response duration among the 67% achieving complete response . Both OS and PFS were inferior with low levels of albumin <3.5 g/dl and high levels of beta-2-microglobulin >5.5 mg/l and lactate dehydrogenase ⩾190 U/l (Table 2). Both GEP70 and GEP80 high-risk designations were associated with poor OS and PFS. Other adverse GEP variables included PR subgroup, proliferation index ⩾10, centrosome index ⩾3 and amp1q21. Among DNA/CIG-derived parameters, adverse prognostic implications were linked to cases with >2 DNA stem lines, LCR ⩾50% and low CIg <2.8 (optimal cutoff point derived from running log-rank analysis on PFS), regardless of DNA stem line dominance. Next, we performed several multivariate analyses. In the absence of GEP data (model 1), low albumin, older age and low CIg were associated with shorter OS and PFS. The combined effect of the presence of these variables is depicted in Figure 1. When GEP variables were also considered (model 2), low albumin, low CIg and age maintained their independent prognostic significance. New variables entering the model included GEP70-defined high risk, proliferation index, and—for PFS only—IgA isotype.

Table 2 Cox proportional hazards regression modeling for OS and PFS
Full size table
Figure 1
figure 1

Kaplan–Meier plots of OS and PFS in TT3b as defined by the multivariate survival analysis model (GEP variables excluded).

Full size image

Given the association of CIg with poor survival in this trial, we examined the variables linked to low CIg (<2.8; Table 3). With the exception of low albumin, low CIg was linked to all adverse standard parameters (beta-2-microglobulin, C-reactive protein, lactate dehydrogenase, hemoglobin, marrow plasmacytosis and cytogenetic abnormalities). Significant associations were also noted between low CIg and GEP-defined high risk (both GEP70 and GEP80), centrosome index and LCR%. The MS molecular subgroup was under-represented in patients with low CIg. High beta-2-microglobulin and C-reactive protein, centrosome index ⩾3 and LCR exceeding 50% were independently and positively linked to low CIg in multivariate analysis.

Table 3 Logistic regression analysis for factors associated with ‘any CIg <2.8’
Full size table

As low CIg was strongly correlated with a multitude of different prognostic variables and retained independent adversity in the multivariate models 1 and 2 of Table 2, a comparative genomic analysis was carried out to identify gene probes distinguishing low from high CIg cases. Such analysis would enable us to validate our approach in trials where DNA/CIG had not been performed. The Wilcoxon’s rank sum test of significance analysis of microarrays of the GEP data for the two groups revealed 12 gene probe sets derived from 11 genes with a P-value <10−4 and a false discovery rate <10% (Table 4). A risk score (GEP12) was computed from the significant probe sets by subtracting the sum of the expressions of the probes over-expressed in patients with low CIg from the sum of the expressions of the probes under-expressed in patients with low CIg, divided by the total number of probes. Using the running log-rank method, adverse prognostic implications were observed in TT3b for patients exhibiting a GEP12 score <5.35. This GEP12 score <5.35 substituted effectively for low CIg in model 3 of Table 2 and, importantly, dispelled GEP70 high risk and proliferation index. We next examined whether the GEP12 score held prognostic implications in other trials where the doublet discrimination method could not be retrospectively applied or DNA/CIG data were unavailable. Indeed, the GEP12 risk score segregated OS and PFS strongly in the bortezomib-containing TT3b training set (Figure 2a), in test sets of TT3a6 (Figure 2b) and in the HOVON65/GMMG-HD424 trials (Figure 2c). In TT2, PFS differed with a strong trend in OS, when both arms were considered combined (Figure 2d).

Table 4 List of differentially expressed gene probes with a q-value less than 0.1 from the Wilcoxon’s rank sum test significance analysis of microarrays of the ‘any CIg <2.8’ and ‘all CIg ⩾2.8’ groups of patients
Full size table
Figure 2
figure 2

Kaplan–Meier plots of OS and PFS in TT3b (a), TT3a (b), ComBat-transformed HOVON (c) and TT2 (d), according to the 12-probeset score for genes associated with CIg.

Full size image

Discussion

We show that the presence of low CIg as detected by DNA/CIG is a major adverse prognostic factor in TT3b, even when other GEP-derived prognostic factors were accounted for (Table 2). Although linked to a multitude of standard adverse prognostic factors (Table 3), low CIg survived the multivariate models even in the presence of GEP data. Factors that were not linked to CIg, such as older age and low albumin, retained independent adverse significance. The CI-linked GEP12 score outperformed GEP70-risk in TT3b (see Table 2) and was validated in TT3a, TT2 and HOVON trials. In this trial with contemporary treatment components, DNA/CIG ploidy status (DI) per se was not prognostic for either OS or PFS, even when an optimal cutoff point approach for the DI value was obtained (data not shown). We believe that this reflects the improvement in prognosis through newer treatments.6, 25

The clinical significance of CIg in MM may be related to its impact on the pathophysiology of the plasma cell. Immunoglobulin-producing and -secreting cells, normal or malignant, are characterized by a low proteasome capacity26 that puts the cells under endoplasmic reticulum stress27 that is dealt with by the unfolded protein response.28 Failure of the plasma cell to mount an effective unfolded protein response in the presence of the immunoglobulin production stress leads to apoptosis.29, 30 Bortezomib targets the proteasome and increases endoplasmic reticulum stress.31 Consequently, in cases of high CIg signifying high immunoglobulin production, endoplasmic reticulum stress is augmented further by exposure to bortezomib, resulting in accelerated apoptosis regardless of other biologic characteristics of that cell. This hypothesis is supported by the finding that the CIg-derived gene score was significant in the bortezomib-containing TT3a/b and HOVON studies but to a lesser extent in TT2 devoid of a proteasome inhibiting agent. The GEP-defined MS molecular subgroup, corresponding to the t(4; 14) translocation and known to benefit from bortezomib,6, 32 was associated with a high CIg in our series (Table 3), thus providing a potential explanation for the sensitivity of this subgroup to proteasome inhibitors.

Low CIg was associated with aggressive disease characteristics (Table 3). Recently, the identification of a subpopulation of MM cells with a reduction in the immunoglobulin production, pre-plasmablastic morphology and immaturity when examined by multicolor flow cytometry has been linked with proteasome inhibition resistance and reduced PFS.18 The linkage of low CIg to a high GEP-defined centrosome index is novel. Beyond providing support for the successful completion of the anaphase in eukaryotic cells, centrosomes also serve in the orientation of the cellular cilia33 and are hence an integral part of a successful cellular migration,34 perhaps facilitating the generation of extramedullary disease.35 Interestingly, a centrosome inhibitor has shown promising activity in preclinical models of MM,36 thus potentially providing a selective drug for patients with low CIg myeloma.

Of the 12 gene probe sets strongly associated with a low CIg in the Wilcoxon Rank sum test analysis, only 3 were over-expressed (Table 4). Importantly, (204251_s_at) CEP164, encoding a centrosomal protein crucial for cilia formation37 and not amongst the gene probes forming the centrosome index, had the highest expression in the low CIg group, fitting the association of increased centrosome expression with low CIg (Table 4). Another hyperexpressed gene in the low CIg group was (209776_s_at) SLC19A1, which is one of the GEP70-constituting genes. SLC19A1 is a member of the Solute Carrier (SLC) group of membrane transporters, which encode for a membrane protein that functions as a folate carrier implicated in methotrexate cellular accumulation in pediatric acute lymphoblastic leukemia.38 Consequently, under the prism of the recent advances in this class of drugs,39 folate antagonists merit a new look in MM with low CIg. The remaining gene with an inverse relation of expression, (227896_at) BCCIP, is involved in cell cycle regulation and it was recently shown that it promotes tumor progression.40

Among the 9 gene probes under-expressed in the low CIg group, (213187_x_at and 212788_x_at) FTL encodes for the L subunit of the ferritin protein. Recently, the H subunit of the ferritin molecule was linked to predicting sensitivity to bortezomib of myeloma cells in vitro.41 (215949_x_at) IGHM encodes for the constant part of the heavy mu chain and is a marker of B-cell differentiation, as has also been shown by others.42 In a similar fashion, (219117_s_at) FKBP11, a member of the FKBP family of peptidyl-prolyl cis/trans isomerases, has been found to be uniquely highly expressed in MM;43 its downregulation in the low CIg group furthers supports the dedifferentiation of the low immunoglobulin-producing plasma cells. (207408_at) SLC22A14, a member of the SLC group of membrane transporters, encodes for a transmembrane small molecule cation transporter,44 implying that it could be potentially involved in the intracellular transportation of agents in MM. (217622_at) RHBDD3, otherwise known as PTAG (pituitary tumor apoptosis gene), encodes for a protein that has been shown to be involved in cell cycle regulation and promote apoptosis in solid tumors,45 whereas (226286_at) ELMOD encodes for a cytoskeleton protein that recently has been shown to be important in the functionality of stereo-cilia.46 Finally, (215432_at) ACSM1 encodes for a protein with a mitochondrial location that is implicated in the metabolism of fatty acids,47 and (239844_x_at) C1orf228 encodes for a protein of unknown functionality.48

In conclusion, DNA/CIG, a readily applicable, fast and low-cost test, offers valuable prognostic information even in the era of genomic profiling and contemporary therapies. Its incorporation into survival analysis revealed new insights into the disease biology and hitherto unsuspected MM-relevant genes. These genes, when used in a GEP12 risk score, proved to be prognostically powerful in a multitude of MM trials and may provide useful information for the evaluation and establishment of new targeted therapies.