Molecular Classification of Multiple Myeloma: A Distinct Transcriptional Profile Characterizes Patients Expressing CCND1 and Negative for 14
http://www.100md.com
《临床肿瘤学》
the Unita Operativa (UO) Ematologia 2 and UO Ematologia 1—Centro Trapianti di Midollo, Dipartimento di Scienze Mediche, Ospedale Maggiore Istituto di Ricovero e Cura a Carattere Scientifico (IRCCS), Università degli Studi di Milano, Milano
Dipartimento di Processi Chimici dell'Ingegneria, Università degli Studi di Padova, Padova
the Divisione di Ematologia e Centro Trapianto di Midollo, Azienda Ospedaliera Bianchi-Melacrinò-Morelli, Reggio Calabria, Italy
ABSTRACT
PURPOSE: The deregulation of CCND1, CCND2 and CCND3 genes represents a common event in multiple myeloma (MM). A recently proposed classification grouped MM patients into five classes on the basis of their cyclin D expression profiles and the presence of the main translocations involving the immunoglobulin heavy chain locus (IGH) at 14q32. In this study, we provide a molecular characterization of the identified translocations/cyclins (TC) groups.
MATERIALS AND METHODS: The gene expression profiles of purified plasma cells from 50 MM cases were used to stratify the samples into the five TC classes and identify their transcriptional fingerprints. The cyclin D expression data were validated by means of real-time quantitative polymerase chain reaction analysis; fluorescence in situ hybridization was used to investigate the cyclin D loci arrangements, and to detect the main IGH translocations and the chromosome 13q deletion.
RESULTS: Class-prediction analysis identified 112 probe sets as characterizing the TC1, TC2, TC4 and TC5 groups, whereas the TC3 samples showed heterogeneous phenotypes and no marker genes. The TC2 group, which showed extra copies of the CCND1 locus and no IGH translocations or the chromosome 13q deletion, was characterized by the overexpression of genes involved in protein biosynthesis at the translational level. A meta-analysis of published data sets validated the identified gene expression signatures.
CONCLUSION: Our data contribute to the understanding of the molecular and biologic features of distinct MM subtypes. The identification of a distinctive gene expression pattern in TC2 patients may improve risk stratification and indicate novel therapeutic targets.
INTRODUCTION
Cytogenetic and molecular studies have provided evidence that multiple myeloma (MM) is characterized by markedly heterogeneous chromosomal aberrations.1 In particular, it has been demonstrated that translocations involving the immunoglobulin heavy chain (IGH) locus at 14q32 and different chromosomal partners occur in approximately 60% of MM cases.2 The most recurrent IGH translocations in MM patients include t(11;14)(q13;q32), t(4;14)(p16.3;q32), t(6;14)(p21;q32), t(14;16)(q32;q23), and t(14;20)(q32;q11) which respectively deregulate the CCND1,3,4 FGFR3 and MMSET/WHSC1,5-7 CCND3,8 MAF,9 and MAFB10 genes.
Together with CCND2, CCND1 and CCND3 are members of the cyclin D family involved in a complex pathway that closely regulates physiologic cell cycle progression from G1 to S phase.11 Various interphase fluorescence in situ hybridization (FISH) studies have indicated that t(11;14) and t(6;14) are found in 15% to 20% and less than 5% of MM patients, respectively.2 However, CCND1 overexpression investigated by real-time quantitative polymerase chain reaction (Q-RT-PCR) and immunohistochemistry has been found in 25% to 50% of MM patients, suggesting that mechanisms other than t(11;14) translocation, such as gene amplification or polysomy, are involved.12,13 In addition, it has been reported that CCND2 is frequently deregulated in MM cases with t(14;16) and t(14;20), and recent experimental evidence indicates that CCND2 is a transcriptional target of the MAF protein.14
It has been recently proposed that the deregulation of cyclin D genes may represent a common event in MM because at least one of them is overexpressed in almost all cases. This has allowed the proposal of a molecular classification of MM based on the presence of the recurrent IGH chromosomal translocations and cyclins D expression, named TC classification.15 MM patients were stratified into five different groups: TC1, characterized by the t(11;14) or t(6;14) translocation, with the consequent overexpression of CCND1 or CCND3, and a nonhyperdiploid status; TC2, showing low to moderate levels of the CCND1 gene in the absence of any primary IGH translocation but associated with a hyperdiploid status; TC3, including tumors that do not fall into any of the other groups, most of which express CCND2; TC4, showing high CCND2 levels and the presence of the t(4;14) translocation; and TC5, expressing the highest levels of CCND2 in association with either the t(14;16) or t(14;20) translocation.
The aim of this study was to provide insights into the potential role of D-type cyclins and IGH translocations in the molecular classification of MM. To this end, a panel of 50 MM cases was investigated using combined DNA microarray, Q-RT-PCR, and FISH analyses, and the identified transcriptional fingerprints were validated using two independent gene expression datasets.16,17
MATERIALS AND METHODS
Patients and Samples Preparation
Bone marrow specimens from four healthy donors and pathologic samples from 50 untreated MM patients (39 of whom are described in our previous reports)18,19 were obtained during standard diagnostic procedures after informed consent. Twenty-eight patients were male; the median age was 66 years (range, 39 to 85 years). Thirty-two patients had an immunoglobulin (Ig-) G protein monoclonal component, eleven IgA, two IgG/IgA, and one IgD protein. Four patients had the light chain . The / ratio was 1.4. The diagnosis and clinical staging of MM were made according to previously described criteria.20 Seventeen patients were in stage IA, 16 in stage IIA/B and 17 in stage IIIA/B. No conventional cytogenetic (G-banding) analyses were available (Appendix A, online only).
Plasma cells (PCs) were purified from bone marrow samples using CD138 immunomagnetic microbeads (MidiMACS system, Miltenyi Biotec, Auburn, CA) as previously described.18,19 The purity of the positively selected PCs was assessed by means of morphology and flow cytometry and was 90% in all cases.
Gene Expression Profiling
Total RNA was extracted and purified, and biotin-labeled cRNA was synthesized as previously described.19 In accordance with the Affymetrix protocols, 15 μg of fragmented cRNA was hybridized on HG-U133A Probe Arrays (Affymetrix Inc, Santa Clara, CA), and the oligonucleotide arrays were scanned using an Agilent GeneArray Scanner G2500A (Agilent Technologies, Waldbronn, Germany).
Microarray Data Analysis
Probe level data were converted to expression values using the Bioconductor function for Robust Multi-Array average (RMA) procedure,21 in which perfect match values are background adjusted, normalized using quantile-quantile normalization, and log transformed. Detection calls were calculated using the default parameters of Affymetrix MAS 5.0 software package. Data with absent calls in all of the arrays were filtered out; no filtering procedure was applied to the intensity levels. The filtered data led to 50 samples and 15,764 probe sets. The RMA intensity values of six Affymetrix probes specific for CCND1 (208711_s_at, 208712_at), CCND2 (200951_s_at, 200952_s_at, 200953_s_at), and CCND3 (201700_at) were compared in MM cases versus healthy donors. The cyclin genes were considered as being overexpressed only if the intensity of all their representative probes in the MM samples was greater than the mean value plus 3 standard deviations of the corresponding expression in healthy donors (Appendix B). The differentially expressed genes discriminating the various TC classes were identified using Prediction Analysis of Microarrays (PAM; Excel front-end publicly available at http://www-stat.stanford.edu/tibs/PAM), a statistical method for ranking genes and performing multi-class classification on the basis of gene expression data.22 The method is based on the nearest shrunken centroids, and shrinkage is obtained through soft thresholding (ie, reducing the centroid statistic by an amount in order to obtain the minimum cross-validation error). The optimal value of was chosen using a 10-fold cross-validation process. Full details of the statistic analysis are included in Appendix B (online only).
To perform the hierarchical agglomerative clustering of the selected probe lists, Pearson's correlation coefficient and average linkage were respectively used as distance and linkage methods in DNA-Chip Analyzer (dChip) software,23,24 as previously described.19
The transcriptional fingerprints identified in the proprietary database were validated using the publicly available MM gene expression dataset described in Zhan et al,16 which contains 74 MM cases and 31 healthy individuals hybridized on Affymetrix HuGeneFL microarrays (http://lambertlab.uams.edu). As for the proprietary cases, the MM samples were stratified into the five TC groups on the basis of their cytogenetic characteristics and the cyclin D expression. The intensity of the three Affymetrix probes specific for CCND1 (X59798_at), CCND2 (D13639_at), and CCND3 (M92287_at) in the MM samples were tested against the mean values plus 3 standard deviations of the corresponding expression in normal PCs.
To compare the data of Zhan et al with the proprietary expression sets, Affymetrix HuGeneFL and HG-U133A GeneChip designs were matched using Affymetrix pair-wise comparison spreadsheets (http://www.affymetrix.com/support/), thus allowing the identification of a total of 4,919 unique probe sets (Appendix C, online only). The same meta-analysis approach was also used on an additional database (Appendix D, online only).17
Wilcoxon's rank sum test was used for the statistical analyses when non-parametric tests were required; the contingency tables were analyzed using Fisher's exact test. The data discussed in this article have been deposited in National Center for Biotechnology Information's Gene Expression Omnibus (GEO; http://www.ncbi.nlm.mih.gov/geo) and are accessible through GEO Series accession number GSE2912.
Q-RT-PCR
Human CCND1, CCND2, CCND3 and GAPDH expression was analyzed in purified PCs according to a published protocol,25 using an ABI PRISM 7700 Sequence Detection System (Applied Biosystems, Foster City, CA). The following cDNA-specific primers and probes were designed against the GenBank published sequences using Primer Express (Applied Biosystems): CCND1 (NM_053056), Forward (Fw) 5'-ACCTGAGGAGCCCCAACAA-3', Reverse (Rv) 5'-TCTGCTCCTGGCAGGCC-3', Probe FAM-TCCTCATACCGCCTCACACGCTTCCT–TAMRA; CCND2 (NM_001759), Fw 5'-ACCCTACATGCGCAGCAGAATGGT-3', Rv 5'-GACCTCTTCTTCGCACTTCTGTTC-3', Probe FAM-CACCTGGATGCTGGA-MGB; CCND3 (NM_001760), Fw 5'-CAGGCCTTGGTCAAAAAGCA-3', Rv 5'-GGCGGGTACATGGCAAAG-3', Probe FAM-TCTGTGCTACAGAATATA–MGB. GAPDH-specific Pre-Developed Assay Reagent (PDAR; Applied Biosystems) was used as the internal control. RT-PCR amplifications were performed in a final volume of 25 μL containing Universal Master Mix (Applied Biosystems), 50 ng of cDNA equivalents, primers (300 nmol/L each), and 200 nmol/L of the TaqMan Probe (Applied Biosystems, Foster City, CA). Cycling was as follows: 10 minutes at 95°C; 50 bi-phase cycles for 15 seconds at 95°C, and 1 minute at 60°C. The amount of CCND1, CCND2, CCND3 and GAPDH transcripts in each sample was calculated from a standard curve relative to the MM cell lines KMS12, CMA-03 (an interleukin-6 [IL-6] -dependent human myeloma cell line established in our laboratory; Neri et al, submitted for publication) and KMM1, respectively. All of the determinations were made in triplicate. The cyclin/GAPDH ratios of each cyclin D gene in the MM cases and healthy donors were compared following the same procedure used for the microarray data. The Q-RT-PCR data and full details of the comparisons are reported in Appendix B.
FISH
The FISH procedure and the specific probes for the detection of IGH translocations involving the CCND1, CCND3, FGFR3/WHSC1, MAF and MAFB genes have been described previously.18 Abnormalities in the CCND2 locus were investigated using a set of probes located centromerically (BAC 631E11) or telomerically (BAC 828B14) to the gene, spanning a genomic region of approximately 600 Kb. The polysomy of chromosomes 11, 12 and 6 was investigated by means of co-hybridization with the specific chromosome alpha satellite probes (Appligene Oncor, Illkirch, France), and probes specific for the CCND1 (BAC300I6), CCND2 (74J21), and CCND3 (973N23) loci. For the 13q14 deletion analyses, we used two BAC clones, 320G21 and 34F20, the latter located within the minimally deleted region.26 All of the clones were selected by browsing the University of California, Santa Cruz (Santa Cruz, CA) Genome Database (http://genome.ucsc.edu/). Two hundred nuclei taken from six healthy controls were analyzed for each type of co-hybridization. The cutoff values for CCND1, CCND2 and CCND3 extra-signals in interphase nuclei were, respectively, 3.4, 2.8 and 3.2 (mean value plus three standard deviations).
RESULTS
Samples Stratification Into TC Groups
The 50 MM samples were stratified into TC groups15 using molecular characterization and the differential levels of cyclin D expression in the microarray data (see Materials and Methods). The results are summarized in Table 1.
Relevant overexpression of CCND1 gene was found in 22 of the 50 MMs, of CCND2 in 23, and of CCND3 in five. Although only five MM cases showed CCND3 overexpression, it is worth noting that moderate expression levels could be detected in healthy individuals and all of the MM tumors, whereas the normal PCs showed no CCND1 or CCND2 expression. An appreciable combined expression of two cyclin D genes was observed in two cases (MM-067, MM-069), and only MM-089 showed the combined expression of all three. Overall, a significant overexpression of at least one type of cyclin D was found in 46 of 50 MM patients (92%). The samples were stratified as follows (Fig 1A): 11 patients in TC1, including 10 carrying t(11;14) and expressing high levels of CCND1 and one harboring a t(6;14) strongly deregulating CCND3; 11 patients in TC2, all of whom were negative for IGH translocations and overexpressing CCND1 at lower levels than TC1 (P = 1.3 x 10–3 for both probes), but not CCND2 or CCND3; 15 patients in TC3, all of whom were negative for the IGH translocation and CCND1 expression, most of whom (11 of 15) expressed CCND2 at variable levels, and one patient positive for CCND3; 10 patients in TC4, all harboring t(4;14); and three patients in TC5, all carrying deregulated MAF or MAFB genes. The patients in TC4 and TC5 were characterized by a consistent expression of CCND2, with the TC5 patients generally showing the highest levels.
The samples classification determined by the microarray data was validated using Q-RT-PCR in all four healthy donors, and 42 of the MM patients (Fig 1B; Table 1). The correspondence between the results of the two analyses was evaluated by assessing the correlation coefficients of cyclin D expression levels determined by microarray and Q-RT-PCR: the coefficients were 0.90 and 0.94 for the CCND1 probes; 0.74, 0.71 and 0.82 for the CCND2 probes; and 0.98 for the CCND3 probes, thus indicating an almost complete concordance for CCND1 and CCND3, and a satisfactory correspondence for the CCND2 gene (Appendix B). As shown in Figure 2A, Q-RT-PCR data indicated that TC1 cases showed significantly higher CCND1 expression levels than TC2 cases (P = 3.9 x 10–3). Moreover, the TC2 group showed significantly higher CCND1 expression levels than the TC3, TC4 and TC5 (P = 1.39 x 10–4, P = 1.05 x 10–4, and P = .022, respectively). The TC3 patients defined two different subgroups, one characterized by the absence of CCND2 expression and the other by heterogeneous CCND2 expression levels (Fig 2B). Similarly to microarray data, all patients in TC4 and TC5 were characterized by a consistent expression of CCND2, with the TC5 cases reaching the highest levels (Fig 2B).
Finally, concerning the clinical parameters (Appendix A), in our data set the only identified correlation with the determined TC classes was the prevalence of focal bone lesions in the patients overexpressing CCND1 (the TC1 and TC2 groups) in comparison with all the other MMs (P = .0231), thus confirming previously reported data.27
FISH Analysis
The results of the FISH analysis are reported in Table 1. Among the patients with t(11;14) included in the TC1 group, increased CCND2 and CCND3 copy numbers were only found in one of nine and two of nine studied cases, whereas the chromosome 13 deletion was found in four of 10 (40%).
The analysis of the cyclin D loci arrangements in the TC2 group showed that 9 of 11 patients (82%) had extra copies of the CCND1 locus, whereas none were positive for CCND2 polysomy and only 1 of 11 (9%) was positive for extra copies of CCND3. Notably, all but one of the TC2 patients were negative for the chromosome 13q deletion.
Twelve cases in the TC3 group were analyzed: extra copies of CCND1, CCND2 and CCND3 were found in, respectively, four (33%), three (25%) and eight patients (66%), whereas chromosome 13q deletion was found in eight (66%). The FISH analyses of the patients in the TC4 group showed extra copies of only CCND3 in 3 of 9 cases (33%), and 13q deletion in 8 of 9 (89%). Extra copies of all the cyclin D genes were found in both of the TC5 patients for whom material was available; these patients also had a chromosome 13q deletion. It is interesting that none of the 50 patients showed any structural alterations in the CCND2 locus, such as split centromeric and telomeric probes (see Materials and Methods).
A contingency analysis made to determine whether the number of cases with increased copies of cyclin D genes correlated with the assigned classes indicated that the samples with extra copies of CCND1 were more significantly represented in the TC2 than the TC3 (P = 6.3 x 10–3) or TC4 groups (P = 3 x 10–4), whereas extra copies of CCND3 were more significantly represented in the TC3 than the TC1 (P = .0281) or TC2 groups (P = .094). Globally, extra copies of CCND1 more strictly correlated with the TC2 group (9 of 11 with extra copies) when this was compared with all of the other available MM samples (11 of 34 with extra copies, P = 5.9 x 10–3). Only the TC2 group showed polysomy associated with the CCND1 locus but not the CCND2 or CCND3 loci; in the other TC groups, the significant presence of polysomy (eg, CCND3 in TC3 group; P = .0106) was not related to any single type of cyclin D genes.
Finally, chromosome 13q deletion was more represented in the TC4 group (eight of nine) than in all of the other MM samples (15 of 36; P = .0220), whereas its absence significantly correlated with the TC2 group (10 of 11) in comparison with all of the other MM cases (12 of 34; P = 1.7 x 10–3).
Analysis of Gene Expression Profiling Data
The stratified gene expression database was analyzed using the PAM software22 in order to identify the transcriptional fingerprints characterizing the five TC groups. The classifier was first trained on the entire MM data set in order to select the set of genes with the smallest estimated misclassification error by means of a cross-validation process. The minimum error corresponded to 51 differentially expressed probes ( = 5) specific for 36 genes, which globally characterized four of the TC groups (Appendix B). Figure 3 shows the hierarchical clustering of the selected probe sets.
The TC1, TC4 and TC5 groups were characterized mainly by the expression of genes previously identified by us (on a smaller data set and using different algorithms) as part of the specific transcriptional signatures respectively related to t(11;14), t(4;14), and t(14;16)/t(14;20).19
The TC2 group showed the specific expression of five marker transcripts, including one codifying for a novel protein containing a POZ domain (BTBD3) characteristic of transcriptional repressor factors, and two codifying for the L13a and L36 ribosomal proteins involved in ribosome biogenesis (RPL13A and RPL36).
No marker was differentially expressed in the TC3 group, thus suggesting the relative heterogeneity of the TC3 samples. As indicated by the dendrogram in Figure 3, the TC3 patients were divided in two main subgroups that respectively clustered with the TC2 and TC4/TC5 patients. Interestingly, the expression of the CCND2 probe sets was significantly higher in the subgroup clustering with TC4/TC5 (P < .05 for all the three probes).
Because no differentially expressed gene characterized the TC3 group, the 15 TC3 samples were excluded from the analysis in order to evaluate whether the TC1, TC4, TC5 and, particularly, the TC2 group could be better characterized by reducing the heterogeneity of the data set. PAM training and cross-validation results indicated retaining 112 probe sets ( = 4.6), distinguishing the TC1, TC2, TC4 and TC5 groups (Fig 4, left panel; Appendix B). In particular, the TC2 group was characterized by 30 genes, mainly involved in protein biosynthesis at translational level. Twelve ribosomal protein genes related to large (L13a, L36, L28, L18a, L35, L4 and L7a) and small (S19, S9, S15 S16 and S27) ribosome subunits, the translation initiation factor 3 (eIF3k), and the translation elongation factor 2 (EEF2) were overexpressed in the TC2 cases, as well as other genes involved in cell cycle regulation (PIK3R3, ISG20, KCNS3), Wnt signaling (FRZB1) or nuclear-cytoplasmic transport (NPM1). In particular, the putative oncogene NPM1 is involved in chromosomal translocations typical of large cell lymphoma, and in acute promyelocytic leukemia.28 In line with the results obtained by analyzing the five groups, the most differentially expressed probe in TC2 was that specific for the BTBD3 gene (Figs 3 and 4).
The right panel of Figure 4 shows the expression profiles in the TC3 cases of the identified probes, whose heterogeneous expression confirmed the previous observation (Fig 3). Furthermore, when the PAM model was used to classify the TC3 samples (as a test set), four patients were predicted as belonging to TC2, ten to TC4 and one to TC5 class (Appendix B). These data prompted us to verify whether the distribution of TC3 specimens in the various groups could be related to the expression levels of CCND1 or CCND2. A nonparametric test applied to the probes specific for the two genes showed that only CCND2 expression was statistically higher in the TC3 samples predicted as belonging to TC4 than in those predicted as belonging to TC2 (P = 2 x 10–3; Appendix B). Furthermore, the only TC3 sample predicted as belonging to TC5 (MM-036) had CCND2 levels comparable with those of the samples with deregulated MAF genes.
Validation of Gene Expression Signatures by Meta-Analysis of Independent Samples
The expression signatures identified in the proprietary database as characterizing the TC1, TC2, TC4 and TC5 groups were validated by means of a meta-analysis of an independent set of 74 MM cases (validation set) profiled on Affymetrix HuGeneFL microarrays.16
The validation set was stratified as follows: the TC1 samples (n = 15) were identified by the presence of t(11;14) or the ectopic overexpression of the CCND3 gene; the TC2 samples (n = 25) by their relevant CCND1 expression (higher than the mean value plus three standard deviations of the corresponding expression in normal PCs) and the absence of t(11;14); the TC4 samples (n = 9) by their ectopic expression of the FGFR3 gene; the TC5 samples (n = 7) by their combined expression of CCND2, ITG?7 and CX3CR1,29 which were significantly different from the normal PC samples; and the TC3 samples (n = 15) by their CCND2 expression and the absence of CCND1, together with the absence of the genes characterizing the TC4 and TC5 groups.
To allow classification, the probes' IDs in different arrays were correlated using Affymetrix probe match spreadsheets (Appendix C); 50 probes were the best-match HuGeneFL counterparts for the 112 HG-U133A predictor transcripts. A PAM model was designed using all 50 matching probes trained on the proprietary samples, and was then used to classify the validation set samples into the TC1, TC2, TC4, and TC5 groups (because no marker probe characterized the TC3 group, the TC3 cases were excluded from the classification session). The model correctly recognized the "unseen" samples of the validation set, classifying 12 of 15 specimens into TC1 (the three not correctly predicted samples showed CCND3 but not CCND1 gene ectopic expression), 20 of 25 into TC2, nine of nine into TC4 and seven of seven into TC5 (Table 2). The classification accuracy measures (a global classification rate of 85.71%; mean specificity, 85.71%; and mean sensitivity, 90%) suggest that the identified TC expression signature is a highly conserved characteristic of the MM samples that is not affected by cohort- or lab-specific biases. Full details are provided in Appendix C.
In addition, we further verified the robustness of the identified gene expression signatures by directly analyzing the validation set through the same multi-class classification approach as that used for the proprietary data set, and by comparing the lists of transcripts obtained in the two analyses (Appendix C). In particular, also in Zhan's database we identified the specific upregulation in the TC2 group of several genes belonging to the translational machinery. Among the genes found to be particularly well preserved among independent data sets, we identified in TC2 RPS9, RPL35, eIF3k, FBL and GNB2L1, as well as FRZB, ISG20, COX5A, ATP5D and NPM1. Unfortunately, there was no specific probe for BTBD3 (the most differentially expressed transcript in the TC2 group in the multiclass analysis of the proprietary data set) on the HuGeneFL array.
Finally, the same interplatform comparisons were used to validate our results on an additional publicly available database.17 It is worth noting that, in addition to the verification of the translational machinery genes, also the BTBD3 transcript resulted to be significantly upregulated in the TC2 group (Appendix D).
DISCUSSION
This study extends previous evidence that cyclin D overexpression is an almost universal event in MM, and provides new insights into the molecular definition of MM based on cyclin D deregulation and IGH translocations.
Approximately 95% of the cases investigated by means of DNA microarrays overexpressed at least one of the three cyclin D genes, with a globally very good concordance with the Q-RT-PCR data. The TC1 and TC2 groups were characterized by the overexpression of CCND1, which was almost undetectable in the normal controls. In particular, low to moderate levels of CCND1 expression were found in 22% of the samples (TC2 group), which did not show either overexpression of CCND2 and CCND3 or known chromosomal translocations. A remarkable finding in this group was the significant occurrence of chromosome 11 polysomy. Similarly, Specht et al13 have recently reported such a correlation determined by Q-RT-PCR analysis in a series of 14 cases. However, whether polysomy 11 per se directly or indirectly causes increased CCND1 expression remains to be clarified; since in other patients included in our database chromosome 11 polysomy is not associated with the overexpression of the gene, we may suggest that other mechanisms may contribute to CCND1 deregulation.
The gene expression profiling analysis showed that the TC2 group is a well-defined entity at transcriptional level: in particular, the TC2 samples were characterized by the differential expression of 30 genes, most of which are involved in protein biosynthesis at translational level. This result was supported by meta-analyses on two independent MM databases, also showing that the TC2 group is characterized by the specific upregulation of the translational machinery pathway. A number of studies have indicated that the constitutive activation of signal transduction pathways and the inactivation of tumor suppressor genes lead to an abnormal activation of translational components.30 In addition, there are experimental evidences that the constitutive expression of some of these factors (such as eIF-2, eIF-3 and eIF-4) may induce transformation in vitro, as well as protect from apoptosis.31-34 Previous studies have shown that CCND1 overexpression is regulated not only at posttranscriptional, but also at translational level: in particular, under specific experimental conditions, translation factors may regulate the synthesis of factors, such as nuclear factor kappa B (NF-B), known to regulate the transcription of CCND1 positively.35,36 Although the mechanisms by which the translational machinery influences CCND1 expression remain to be fully elucidated, these previous observations may explain the moderate CCND1 expression levels in TC2 patients. Interestingly, the highly coordinated expression of ribosomal and translation-associated genes has been recently reported in human cancers investigated by means of gene expression profiling: particularly, they have been found to be upregulated in well-differentiated ovarian tumors37 and in a subset of chronic lymphocytic leukemia patients characterized by a more favorable clinical course.38 Recent studies have shown that CCND1 overexpression in patients carrying either the t(11;14) or chromosome 11 polysomy is a favorable prognostic factor associated with a prolonged remission and event-free survival after autologous transplantation.2,39 In addition, we have shown that the TC2 group is associated with the absence of the chromosome 13q deletion (known to be a negative prognostic factor).40 Further studies of larger cohorts with an adequate follow-up are needed to determine whether TC2 patients are characterized by a distinct prognosis, a finding that may contribute to the risk stratification of MM patients.
The most specifically modulated transcript in the TC2 group (BTBD3) contains a BTB/POZ domain, a conserved protein-protein interaction domain found in many zinc finger transcription factors. This domain has been reported to be associated with the transcriptional repression mediated by the promyelocytic leukemia zinc finger protein.41 In addition, it has been demonstrated that the BTB/POZ domain mediates the transcriptional repressor activity of the BCL6 oncogene, which is frequently deregulated in B-cell lymphomas.42 On the basis of these considerations, this transcript could play an important role in the pathogenesis of a subtype of MM patients.
Our results confirm the strong association between CCND2 overexpression and the presence of t(4;14) in TC4 or the deregulation of MAF or MAFB genes in TC5. Consistently with recent reports indicating that CCND2 is a transcriptional target of MAF protein,14 the highest expression levels were observed in the TC5 group. The mechanisms responsible for the upregulation of CCND2 in the TC4 group are not clear, but it is worth noting that our study did not identify chromosome 12 polysomy or CCND2 extra copies in any of the 10 cases with t(4;14).
The TC3 group has been defined as a miscellany of MM tumors that do not fall into any of the other four groups and do not show any of the known major translocations. In our study cohort, 10 of 15 patients included in the TC3 group (67%) expressed moderate or high levels of CCND2, whereas no deregulation of any cyclin D was observed in four cases. Furthermore, CCND2 expression levels could be statistically correlated with the gene expression–based prediction of TC3 cases assigned to the TC2 or TC4 groups. Because previous data have indicated that CCND2 overexpression in different types of lymphoid tumors is not associated with chromosomal aberrations,43 it is possible to hypothesize that CCND2 deregulation may be the result of epigenetic mechanisms in a significant fraction of MM cases. Overall, these data suggest that CCND2 deregulation may play a role in the molecular and biologic phenotype of the majority of patients included in the TC3 group, and raise the question of whether TC3 patients should be considered an independent group or distributed into the TC2, TC4 or TC5 groups.
Finally, no correlation between CCND3 overexpression and TC classification was found in our data set as overexpression was only found in the single t(6;14) patient and four additional patients stratified into different TC classes. A larger number of patients harboring t(6;14) or overexpressing CCND3 is required in order to establish whether a specific signature can be associated with CCND3 deregulation, thus leading to the definition of an additional TC subgroup of MM patients. We have recently identified molecular signatures associated with the presence of the main IGH translocations by means of gene expression profiling.19 The genes specifically expressed in the TC1 group only partially overlapped those previously described as part of the t(11;14) signature, a discrepancy that may be due to the presence of one sample carrying the t(6;14) translocation in the TC1 group.
Our data may provide important contributions to the understanding of the molecular and biologic features of distinct MM subtypes associated with different prognoses and treatment responses. In particular, together with the reported shorter survival of t(4;14) (TC4) and t(14;16) (TC5) patients receiving either standard- or high-dose therapy, and the longer and significantly better survival of t(11;14) MM cases (TC1) on high-dose conventional therapy,2 the identification of distinctive patterns of gene expression associated with the TC2 subtype may improve risk stratification and indicate novel therapeutic targets. In this context, Q-RT-PCR analyses of highly purified PCs populations provide a suitable approach to the characterization and stratification of MM patients. Furthermore, validation of the identified TC transcriptional signatures by means of the meta-analysis on almost 250 samples from two different MM data sets highlighted the robustness of microarray expression profiling as an integrated platform, thus suggesting that disease models can be efficiently analyzed through interplatform comparisons.
Appendix A
Appendix B
Stratification of proprietary samples in TC groups
The 50 MM samples were stratified into TC groups using molecular characterization and the differential levels of cyclins D expression in the microarray data (see Materials and Methods). Specifically, RMA intensity values of Affymetrix probes specific for CCND1 (208711_s_at, 208712_at), CCND2 (200951_s_at, 200952_s_at, 200953_s_at), and CCND3 (201700_at) were compared in healthy and MM populations. The cyclin genes were considered as overexpressed only if all of their representative probes presented an intensity level in MM samples higher than the threshold computed as the mean plus 3 standard deviations of the corresponding expression values in healthy donors (indicated by "+" in Table 2).
Similarly, the levels of CCND1, CCND2, and CCND3 over GAPDH ratios obtained from Q-RT-PCR ratios were analyzed in all healthy donors (n = 4) and in 42/50 MM patients, following the same procedure described for microarray data (Table 3).
Comparison of microarray and Q-RT-PCR data for cyclins D probes
The correspondence between TC classification obtained by microarray and Q-RT-PCR was evaluated assessing correlation coefficients of cyclins expression levels determined using the two different methods. Being on different intensity scales, both microarray and Q-RT-PCR data have been scaled in the range 0÷1, subtracting the minimum value from original data and dividing it by the data range (ie, the maximum minus the minimum value). The correlation coefficients of expression signals were 0.90 and 0.94 for CCND1; 0.74, 0.71 and 0.82 for CCND2; and 0.98 for CCND3 probes, respectively. The side-by-side comparison and the linear regression curve of cyclin expression levels obtained by microarray and Q-RT-PCR methods are shown in Figures 1A-C.
Identification of transcriptional fingerprints of TC groups: Details of statistical analysis
The stratified gene expression database was analyzed using Prediction Analysis of Microarrays (PAM). PAM uses the nearest shrunken centroids methods to select differentially expressed genes and perform multi-class classification Tibshirani et al.22 The classifier was first trained on the entire dataset to select relevant and the optimal value of chosen by a 10-fold cross-validation process. The training samples were randomly divided 10 times into 10 approximately equal-size, class-balanced parts, the model fitted on 90% of the samples and used to predict the class labels of the remaining 10%. This procedure was repeated 10 times, with each part playing the role of the test samples, and the errors on all 10 parts added together to compute the overall error as a function of . Details of PAM statistical analysis applied on the five TC groups are reported in Figures 2 and 3.
Details of PAM statistical analysis applied on the TC1, TC2, TC4, and TC5 groups only are reported in Figure 4. The description and class scores of the 112 probe sets retained by PAM model constructed on 4 TC groups are reported in Table 4. In Figure 5 is reported the PAM classification of TC3 samples on the basis of the 112 probes identified by applying the PAM model on four TC groups.
Appendix C
Meta-analysis on independent data set (I)
The 112 probe set expression signature identified in the proprietary database as characterizing TC1, TC2, TC4 and TC5 groups, was validated through a meta-analysis on an independent set of 74 MM cases (validation set) profiled on Affymetrix HuGeneFL microarrays (Zhan et al16). To modify this test dataset into data applicable to the predictor gene list from the HG-U133A arrays, the probe IDs between these different arrays were correlated using Affymetrix probe match spreadsheets. Since no conversion table is currently available for a direct comparison between HuGeneFL and HG-U133A chip designs, HuGeneFL probes were first matched to HG-U95Av2 using the HuGeneFL to Human Genome U95A comparison file, and then linked to the HG-U133A annotations through the HG-U95 to HG-U133 Best Match spreadsheet. This procedure generated a conversion query matching a total of 4,919 unique probe sets out of the original 7,070 HuGeneFL non-control probes and led to 50 best match HuGeneFL counterparts for the 112 HG-U133A predictor transcripts (Table 5).
The distribution of the validation set cases in different TC classes was performed comparing the expression levels of cyclins D and the genes targeted by IGH translocations in MM and in 30 normal samples as described for the proprietary database (ie, molecular characterization and cyclin D genes with an intensity level in MM samples higher than the threshold computed as the mean plus 3 standard deviations of the corresponding expression values in healthy donors). Table 6 reports original expression values (as in Zhan et al) for 30 healthy donors and 74 MM samples and TC assignments. TC1 samples (n = 15) were identified for the presence of t(11;14) or the ectopic overexpression of the CCND3 gene; TC2 (n = 25) samples were identified by relevant level of CCND1 expression and the absence of t(11;14); TC4 (n = 9) samples were identified for the expression of FGFR3 gene; TC5 (n = 7) samples were identified for the combined expression of CCND2, ITGB7 and CX3CR1, significantly different from healthy donor samples; TC3 (n = 15) samples were identified for the CCND2 expression and the absence of CCND1, together with the absence of genes characterizing TC4 and TC5 groups. Three samples could not be assigned due to the simultaneous overexpression of CCND1 and CCND2 genes (Fig 6).
A PAM model was designed using all the 50 matching probes (ie, always selecting a threshold parameter =0), trained on the proprietary samples, and used to classify the validation set expression signals into TC1, TC2, TC3, TC4, and TC5 groups. Figure 7 shows cumulative (A) and class-individual (B) cross-validation errors of the PAM model. In Table 7 are reported the classification probabilities for the class prediction of all Zhan's samples.
The stratified gene expression database from Zhan et al was also directly analyzed using PAM. The classifier was first trained on the entire data set (eg, 4,077 probes after filtering out data with Absent calls in all of the arrays; no filtering procedure was applied to the intensity levels) to select relevant genes marking TC1, TC2, TC4, and TC5 groups. The training samples were randomly divided 10 times into 10 approximately equal-size, class-balanced parts, the model fitted on 90% of the samples and used to predict the class labels of the remaining 10%. This procedure was repeated 10 times, with each part playing the role of the test samples, and the errors on all 10 parts added together to compute the overall error as a function of . Details of PAM statistical analysis applied on the four TC groups are reported in Figures 8 and 9.
The description and class scores of the 135 unique HuGeneFL probes retained by PAM model constructed on 4 TC groups are reported in Table 8.
Appendix D
Meta-analysis on independent data set (II)
The 112 probe sets expression signature identified in the proprietary database as characterizing TC1, TC2, TC4 and TC5 groups, was validated through a meta-analysis on an additional independent set of 173 MM cases (validation set) profiled on Affymetrix HG-U95Av2 microarrays (see References, Tian et al, 2003). Using Affymetrix HG-U95 to HG-U133 Best Match spreadsheet, the 112 HG-U133A probe sets identified in the proprietary database were converted into 91 HG-U95Av2 best match probes. In Table 9 is reported the full list of the inter-platform common probes.
The data set from Tian et al (Gene Expression Omnibus, accession number GSE755) contained signal intensities and detection calls for 173 MM samples. The signal intensities were calculated using Affymetrix MAS 5.01 algorithm. No cytogenetic characteristics or data for normal PCs were available. However, the samples were from the same laboratory of Zhan et al, 2002 (Appendix C) and 54 out of 173 reported the very same description label and cyclins D characteristics as in Zhan et al. Thus, anchoring on these 54 common cases, putative TC class was assigned thresholding intensity signal of cyclins D probe sets (Fig 10) and controlling known marker genes of chromosomal translocations (Mattioli et al, 2005). A total of 32 samples over-expressed CCND1 or CCND3 and were assigned to TC1; 56 presented CCND1 expression but no CCND2 or CCND3 and were thus labeled as TC2; 20 cases were assigned to TC4 (based on CCND2 and FGFR3 expression levels), and 5 overexpressing CCND2, ITG?7 and CX3CR1 to TC5. The remaining 60 specimens were either labeled as TC3 (n = 48) or left unassigned (n = 12).
A PAM model was designed using all the 91 matching probes (ie, always selecting a threshold parameter = 0), trained on the proprietary samples, and used to classify the validation set samples into TC groups. Figure 11 shows cumulative (A) and class-individual (B) cross-validation errors of the PAM model. In Table 10 are reported the classification accuracy measures for TC1, TC2, TC4 and TC5 class (113 samples, excluded TC3) and in Table 11 the classification probabilities for the class prediction of the 161 out of 173 Tian et al samples for which was possible to assign a putative TC.
The stratified gene expression database from Tian et al was also directly analyzed using PAM. The classifier was first trained on the entire data set (eg, 10599 probes after filtering out data with Absent calls in all of the arrays; no filtering procedure was applied to the intensity levels) to select relevant genes marking TC1, TC2, TC4, and TC5 groups. The training samples were randomly divided 10 times into 10 approximately equal-size, class-balanced parts, the model fitted on 90% of the samples and used to predict the class labels of the remaining 10%. This procedure was repeated 10 times, with each part playing the role of the test samples, and the errors on all 10 parts added together to compute the overall error as a function of . Details of PAM statistical analysis applied on the four TC groups are reported in Figures 12 and 13.
The description and class scores of the 204 unique HG-U95Av2 probes retained by PAM model constructed on 4 TC groups are reported in Table 12.
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
NOTES
Supported by a grant from the Associazione Italiana Ricerca sul Cancro (AIRC) to A.N. and by the Fondazione Assistenza e Studio Malati Ematologici and the Italian Ministry of Health; by Fondo per gli Investimeni della Ricerca de Base Ministero Istruzione Università e Ricerca Grants No. RBNE01TZZ8 and RBAU01935A to S.B.
L.A. and S.B. contributed equally to this study.
Terms in blue are defined in the glossary, found at the end of this issue and online at www.jco.org.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Seidl S, Kaufmann H, Drach J: New insights into the pathophysiology of multiple myeloma. Lancet Oncol 4:557-564, 2003
Fonseca R, Barlogie B, Bataille R, et al: Genetics and cytogenetics of multiple myeloma: A workshop report. Cancer Res 64:1546-1558, 2004
Chesi M, Bergsagel PL, Brents LA, et al: Dysregulation of cyclin D1 by translocation into an IgH gamma switch region in two multiple myeloma cell lines. Blood 88:674-681, 1996
Ronchetti D, Finelli P, Richelda R, et al: Molecular analysis of 11q13 breakpoints in multiple myeloma. Blood 93:1330-1337, 1999
Chesi M, Nardini E, Brents LA, et al: Frequent translocation t(4;14)(p16.3;q32.3) in multiple myeloma is associated with increased expression and activating mutations of fibroblast growth factor receptor 3. Nat Genet 16:260-264, 1997
Richelda R, Ronchetti D, Baldini L, et al: A novel chromosomal translocation t(4; 14)(p16.3; q32) in multiple myeloma involves the fibroblast growth-factor receptor 3 gene. Blood 90:4062-4070, 1997
Chesi M, Nardini E, Lim RS, et al: The t(4;14) translocation in myeloma dysregulates both FGFR3 and a novel gene, MMSET, resulting in IgH/MMSET hybrid transcripts. Blood 92:3025-3034, 1998
Shaughnessy J Jr, Gabrea A, Qi Y, et al: Cyclin D3 at 6p21 is dysregulated by recurrent chromosomal translocations to immunoglobulin loci in multiple myeloma. Blood 98:217-223, 2001
Chesi M, Bergsagel PL, Shonukan OO, et al: Frequent dysregulation of the c-maf proto-oncogene at 16q23 by translocation to an Ig locus in multiple myeloma. Blood 91:4457-4463, 1998
Hanamura I, Iida S, Akano Y, et al: Ectopic expression of MAFB gene in human myeloma cells carrying (14;20)(q32;q11) chromosomal translocations. Jpn J Cancer Res 92:638-644, 2001
Malumbres M, Barbacid M: To cycle or not to cycle: A critical decision in cancer. Nat Rev Cancer 1:222-231, 2001
Pruneri G, Fabris S, Baldini L, et al: Immunohistochemical analysis of cyclin D1 shows deregulated expression in multiple myeloma with the t(11;14). Am J Pathol 156:1505-1513, 2000
Specht K, Haralambieva E, Bink K, et al: Different mechanisms of cyclin D1 overexpression in multiple myeloma revealed by fluorescence in situ hybridization and quantitative analysis of mRNA levels. Blood 104:1120-1126, 2004
Hurt EM, Wiestner A, Rosenwald A, et al: Overexpression of c-maf is a frequent oncogenic event in multiple myeloma that promotes proliferation and pathological interactions with bone marrow stroma. Cancer Cell 5:191-199, 2004
Hideshima T, Bergsagel PL, Kuehl WM, et al: Advances in biology of multiple myeloma: Clinical applications. Blood 104:607-618, 2004
Zhan F, Hardin J, Kordsmeier B, et al: Global gene expression profiling of multiple myeloma, monoclonal gammopathy of undetermined significance, and normal bone marrow plasma cells. Blood 99:1745-1757, 2002
Tian E, Zhan F, Walker R, et al: The role of the Wnt-signaling antagonist DKK1 in the development of osteolytic lesions in multiple myeloma. N Engl J Med 349:2483-2494, 2003
Fabris S, Agnelli L, Mattioli M, et al: Characterization of oncogene dysregulation in multiple myeloma by combined FISH and DNA microarray analyses. Genes Chromosomes Cancer 42:117-127, 2005
Mattioli M, Agnelli L, Fabris S, et al: Gene expression profiling of plasma cell dyscrasias reveals molecular patterns associated with distinct IGH translocations in multiple myeloma. Oncogene 24:2461-2473, 2005
Durie BG, Salmon SE: A clinical staging system for multiple myeloma: Correlation of measured myeloma cell mass with presenting clinical features, response to treatment, and survival. Cancer 36:842-854, 1975
Irizarry RA, Hobbs B, Collin F, et al: Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 4:249-264, 2003
Tibshirani R, Hastie T, Narasimhan B, et al: Diagnosis of multiple cancer types by shrunken centroids of gene expression. Proc Natl Acad Sci U S A 99:6567-6572, 2002
Eisen MB, Spellman PT, Brown PO, et al: Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A 95:14863-14868, 1998
Schadt EE, Li C, Ellis B, et al: Feature extraction and normalization algorithms for high-density oligonucleotide gene expression array data. J Cell Biochem Suppl 37:120-125, 2001
Heid CA, Stevens J, Livak KJ, et al: Real time quantitative PCR. Genome Res 6:986-994, 1996
Elnenaei MO, Hamoudi RA, Swansbury J, et al: Delineation of the minimal region of loss at 13q14 in multiple myeloma. Genes Chromosomes Cancer 36:99-106, 2003
Robbiani DF, Chesi M, Bergsagel PL: Bone lesions in molecular subtypes of multiple myeloma. N Engl J Med 351:197-198, 2004
Falini B, Mecucci C, Tiacci E, et al: Cytoplasmic nucleophosmin in acute myelogenous leukemia with a normal karyotype. N Engl J Med 352:254-266, 2005
Bergsagel PL, Kuehl WM: Critical roles for immunoglobulin translocations and cyclin D dysregulation in multiple myeloma. Immunol Rev 194:96-104, 2003
Rosenwald IB: The role of translation in neoplastic transformation from a pathologist's point of view. Oncogene 23:3230-3247, 2004
Donze O, Jagus R, Koromilas AE, et al: Abrogation of translation initiation factor eIF-2 phosphorylation causes malignant transformation of NIH 3T3 cells. EMBO J 14:3828-3834, 1995
Mayeur GL, Hershey JW: Malignant transformation by the eukaryotic translation initiation factor 3 subunit p48 (eIF3e). FEBS Lett 514:49-54, 2002
Morley SJ: The regulation of eIF4F during cell growth and cell death. Prog Mol Subcell Biol 27:1-37, 2001
Thornton S, Anand N, Purcell D, et al: Not just for housekeeping: Protein initiation and elongation factors in cell growth and tumorigenesis. J Mol Med 81:536-548, 2003
Guttridge DC, Albanese C, Reuther JY, et al: NF-kappaB controls cell growth and differentiation through transcriptional regulation of cyclin D1. Mol Cell Biol 19:5785-5799, 1999
Hinz M, Krappmann D, Eichten A, et al: NF-kappaB function in growth control: Regulation of cyclin D1 expression and G0/G1-to-S-phase transition. Mol Cell Biol 19:2690-2698, 1999
Welsh JB, Zarrinkar PP, Sapinoso LM, et al: Analysis of gene expression profiles in normal and neoplastic ovarian tissue samples identifies candidate molecular markers of epithelial ovarian cancer. Proc Natl Acad Sci U S A 98:1176-1181, 2001
Durig J, Nuckel H, Huttmann A, et al: Expression of ribosomal and translation-associated genes is correlated with a favorable clinical course in chronic lymphocytic leukemia. Blood 101:2748-2755, 2003
Soverini S, Cavo M, Cellini C, et al: Cyclin D1 overexpression is a favorable prognostic variable for newly diagnosed multiple myeloma patients treated with high-dose chemotherapy and single or double autologous transplantation. Blood 102:1588-1594, 2003
Fonseca R, Blood E, Rue M, et al: Clinical and biologic implications of recurrent genomic aberrations in myeloma. Blood 101:4569-4575, 2003
Collins T, Stone JR, Williams AJ: All in the family: The BTB/POZ, KRAB, and SCAN domains. Mol Cell Biol 21:3609-3615, 2001
Polo JM, Dell'oso T, Ranuncolo SM, et al: Specific peptide interference reveals BCL6 transcriptional and oncogenic mechanisms in B-cell lymphoma cells. Nat Med 10:1329-1335, 2004
Delmer A, Ajchenbaum-Cymbalista F, Tang R, et al: Overexpression of cyclin D2 in chronic B-cell malignancies. Blood 85:2870-2876, 1995(Luca Agnelli, Silvio Bicc)
Dipartimento di Processi Chimici dell'Ingegneria, Università degli Studi di Padova, Padova
the Divisione di Ematologia e Centro Trapianto di Midollo, Azienda Ospedaliera Bianchi-Melacrinò-Morelli, Reggio Calabria, Italy
ABSTRACT
PURPOSE: The deregulation of CCND1, CCND2 and CCND3 genes represents a common event in multiple myeloma (MM). A recently proposed classification grouped MM patients into five classes on the basis of their cyclin D expression profiles and the presence of the main translocations involving the immunoglobulin heavy chain locus (IGH) at 14q32. In this study, we provide a molecular characterization of the identified translocations/cyclins (TC) groups.
MATERIALS AND METHODS: The gene expression profiles of purified plasma cells from 50 MM cases were used to stratify the samples into the five TC classes and identify their transcriptional fingerprints. The cyclin D expression data were validated by means of real-time quantitative polymerase chain reaction analysis; fluorescence in situ hybridization was used to investigate the cyclin D loci arrangements, and to detect the main IGH translocations and the chromosome 13q deletion.
RESULTS: Class-prediction analysis identified 112 probe sets as characterizing the TC1, TC2, TC4 and TC5 groups, whereas the TC3 samples showed heterogeneous phenotypes and no marker genes. The TC2 group, which showed extra copies of the CCND1 locus and no IGH translocations or the chromosome 13q deletion, was characterized by the overexpression of genes involved in protein biosynthesis at the translational level. A meta-analysis of published data sets validated the identified gene expression signatures.
CONCLUSION: Our data contribute to the understanding of the molecular and biologic features of distinct MM subtypes. The identification of a distinctive gene expression pattern in TC2 patients may improve risk stratification and indicate novel therapeutic targets.
INTRODUCTION
Cytogenetic and molecular studies have provided evidence that multiple myeloma (MM) is characterized by markedly heterogeneous chromosomal aberrations.1 In particular, it has been demonstrated that translocations involving the immunoglobulin heavy chain (IGH) locus at 14q32 and different chromosomal partners occur in approximately 60% of MM cases.2 The most recurrent IGH translocations in MM patients include t(11;14)(q13;q32), t(4;14)(p16.3;q32), t(6;14)(p21;q32), t(14;16)(q32;q23), and t(14;20)(q32;q11) which respectively deregulate the CCND1,3,4 FGFR3 and MMSET/WHSC1,5-7 CCND3,8 MAF,9 and MAFB10 genes.
Together with CCND2, CCND1 and CCND3 are members of the cyclin D family involved in a complex pathway that closely regulates physiologic cell cycle progression from G1 to S phase.11 Various interphase fluorescence in situ hybridization (FISH) studies have indicated that t(11;14) and t(6;14) are found in 15% to 20% and less than 5% of MM patients, respectively.2 However, CCND1 overexpression investigated by real-time quantitative polymerase chain reaction (Q-RT-PCR) and immunohistochemistry has been found in 25% to 50% of MM patients, suggesting that mechanisms other than t(11;14) translocation, such as gene amplification or polysomy, are involved.12,13 In addition, it has been reported that CCND2 is frequently deregulated in MM cases with t(14;16) and t(14;20), and recent experimental evidence indicates that CCND2 is a transcriptional target of the MAF protein.14
It has been recently proposed that the deregulation of cyclin D genes may represent a common event in MM because at least one of them is overexpressed in almost all cases. This has allowed the proposal of a molecular classification of MM based on the presence of the recurrent IGH chromosomal translocations and cyclins D expression, named TC classification.15 MM patients were stratified into five different groups: TC1, characterized by the t(11;14) or t(6;14) translocation, with the consequent overexpression of CCND1 or CCND3, and a nonhyperdiploid status; TC2, showing low to moderate levels of the CCND1 gene in the absence of any primary IGH translocation but associated with a hyperdiploid status; TC3, including tumors that do not fall into any of the other groups, most of which express CCND2; TC4, showing high CCND2 levels and the presence of the t(4;14) translocation; and TC5, expressing the highest levels of CCND2 in association with either the t(14;16) or t(14;20) translocation.
The aim of this study was to provide insights into the potential role of D-type cyclins and IGH translocations in the molecular classification of MM. To this end, a panel of 50 MM cases was investigated using combined DNA microarray, Q-RT-PCR, and FISH analyses, and the identified transcriptional fingerprints were validated using two independent gene expression datasets.16,17
MATERIALS AND METHODS
Patients and Samples Preparation
Bone marrow specimens from four healthy donors and pathologic samples from 50 untreated MM patients (39 of whom are described in our previous reports)18,19 were obtained during standard diagnostic procedures after informed consent. Twenty-eight patients were male; the median age was 66 years (range, 39 to 85 years). Thirty-two patients had an immunoglobulin (Ig-) G protein monoclonal component, eleven IgA, two IgG/IgA, and one IgD protein. Four patients had the light chain . The / ratio was 1.4. The diagnosis and clinical staging of MM were made according to previously described criteria.20 Seventeen patients were in stage IA, 16 in stage IIA/B and 17 in stage IIIA/B. No conventional cytogenetic (G-banding) analyses were available (Appendix A, online only).
Plasma cells (PCs) were purified from bone marrow samples using CD138 immunomagnetic microbeads (MidiMACS system, Miltenyi Biotec, Auburn, CA) as previously described.18,19 The purity of the positively selected PCs was assessed by means of morphology and flow cytometry and was 90% in all cases.
Gene Expression Profiling
Total RNA was extracted and purified, and biotin-labeled cRNA was synthesized as previously described.19 In accordance with the Affymetrix protocols, 15 μg of fragmented cRNA was hybridized on HG-U133A Probe Arrays (Affymetrix Inc, Santa Clara, CA), and the oligonucleotide arrays were scanned using an Agilent GeneArray Scanner G2500A (Agilent Technologies, Waldbronn, Germany).
Microarray Data Analysis
Probe level data were converted to expression values using the Bioconductor function for Robust Multi-Array average (RMA) procedure,21 in which perfect match values are background adjusted, normalized using quantile-quantile normalization, and log transformed. Detection calls were calculated using the default parameters of Affymetrix MAS 5.0 software package. Data with absent calls in all of the arrays were filtered out; no filtering procedure was applied to the intensity levels. The filtered data led to 50 samples and 15,764 probe sets. The RMA intensity values of six Affymetrix probes specific for CCND1 (208711_s_at, 208712_at), CCND2 (200951_s_at, 200952_s_at, 200953_s_at), and CCND3 (201700_at) were compared in MM cases versus healthy donors. The cyclin genes were considered as being overexpressed only if the intensity of all their representative probes in the MM samples was greater than the mean value plus 3 standard deviations of the corresponding expression in healthy donors (Appendix B). The differentially expressed genes discriminating the various TC classes were identified using Prediction Analysis of Microarrays (PAM; Excel front-end publicly available at http://www-stat.stanford.edu/tibs/PAM), a statistical method for ranking genes and performing multi-class classification on the basis of gene expression data.22 The method is based on the nearest shrunken centroids, and shrinkage is obtained through soft thresholding (ie, reducing the centroid statistic by an amount in order to obtain the minimum cross-validation error). The optimal value of was chosen using a 10-fold cross-validation process. Full details of the statistic analysis are included in Appendix B (online only).
To perform the hierarchical agglomerative clustering of the selected probe lists, Pearson's correlation coefficient and average linkage were respectively used as distance and linkage methods in DNA-Chip Analyzer (dChip) software,23,24 as previously described.19
The transcriptional fingerprints identified in the proprietary database were validated using the publicly available MM gene expression dataset described in Zhan et al,16 which contains 74 MM cases and 31 healthy individuals hybridized on Affymetrix HuGeneFL microarrays (http://lambertlab.uams.edu). As for the proprietary cases, the MM samples were stratified into the five TC groups on the basis of their cytogenetic characteristics and the cyclin D expression. The intensity of the three Affymetrix probes specific for CCND1 (X59798_at), CCND2 (D13639_at), and CCND3 (M92287_at) in the MM samples were tested against the mean values plus 3 standard deviations of the corresponding expression in normal PCs.
To compare the data of Zhan et al with the proprietary expression sets, Affymetrix HuGeneFL and HG-U133A GeneChip designs were matched using Affymetrix pair-wise comparison spreadsheets (http://www.affymetrix.com/support/), thus allowing the identification of a total of 4,919 unique probe sets (Appendix C, online only). The same meta-analysis approach was also used on an additional database (Appendix D, online only).17
Wilcoxon's rank sum test was used for the statistical analyses when non-parametric tests were required; the contingency tables were analyzed using Fisher's exact test. The data discussed in this article have been deposited in National Center for Biotechnology Information's Gene Expression Omnibus (GEO; http://www.ncbi.nlm.mih.gov/geo) and are accessible through GEO Series accession number GSE2912.
Q-RT-PCR
Human CCND1, CCND2, CCND3 and GAPDH expression was analyzed in purified PCs according to a published protocol,25 using an ABI PRISM 7700 Sequence Detection System (Applied Biosystems, Foster City, CA). The following cDNA-specific primers and probes were designed against the GenBank published sequences using Primer Express (Applied Biosystems): CCND1 (NM_053056), Forward (Fw) 5'-ACCTGAGGAGCCCCAACAA-3', Reverse (Rv) 5'-TCTGCTCCTGGCAGGCC-3', Probe FAM-TCCTCATACCGCCTCACACGCTTCCT–TAMRA; CCND2 (NM_001759), Fw 5'-ACCCTACATGCGCAGCAGAATGGT-3', Rv 5'-GACCTCTTCTTCGCACTTCTGTTC-3', Probe FAM-CACCTGGATGCTGGA-MGB; CCND3 (NM_001760), Fw 5'-CAGGCCTTGGTCAAAAAGCA-3', Rv 5'-GGCGGGTACATGGCAAAG-3', Probe FAM-TCTGTGCTACAGAATATA–MGB. GAPDH-specific Pre-Developed Assay Reagent (PDAR; Applied Biosystems) was used as the internal control. RT-PCR amplifications were performed in a final volume of 25 μL containing Universal Master Mix (Applied Biosystems), 50 ng of cDNA equivalents, primers (300 nmol/L each), and 200 nmol/L of the TaqMan Probe (Applied Biosystems, Foster City, CA). Cycling was as follows: 10 minutes at 95°C; 50 bi-phase cycles for 15 seconds at 95°C, and 1 minute at 60°C. The amount of CCND1, CCND2, CCND3 and GAPDH transcripts in each sample was calculated from a standard curve relative to the MM cell lines KMS12, CMA-03 (an interleukin-6 [IL-6] -dependent human myeloma cell line established in our laboratory; Neri et al, submitted for publication) and KMM1, respectively. All of the determinations were made in triplicate. The cyclin/GAPDH ratios of each cyclin D gene in the MM cases and healthy donors were compared following the same procedure used for the microarray data. The Q-RT-PCR data and full details of the comparisons are reported in Appendix B.
FISH
The FISH procedure and the specific probes for the detection of IGH translocations involving the CCND1, CCND3, FGFR3/WHSC1, MAF and MAFB genes have been described previously.18 Abnormalities in the CCND2 locus were investigated using a set of probes located centromerically (BAC 631E11) or telomerically (BAC 828B14) to the gene, spanning a genomic region of approximately 600 Kb. The polysomy of chromosomes 11, 12 and 6 was investigated by means of co-hybridization with the specific chromosome alpha satellite probes (Appligene Oncor, Illkirch, France), and probes specific for the CCND1 (BAC300I6), CCND2 (74J21), and CCND3 (973N23) loci. For the 13q14 deletion analyses, we used two BAC clones, 320G21 and 34F20, the latter located within the minimally deleted region.26 All of the clones were selected by browsing the University of California, Santa Cruz (Santa Cruz, CA) Genome Database (http://genome.ucsc.edu/). Two hundred nuclei taken from six healthy controls were analyzed for each type of co-hybridization. The cutoff values for CCND1, CCND2 and CCND3 extra-signals in interphase nuclei were, respectively, 3.4, 2.8 and 3.2 (mean value plus three standard deviations).
RESULTS
Samples Stratification Into TC Groups
The 50 MM samples were stratified into TC groups15 using molecular characterization and the differential levels of cyclin D expression in the microarray data (see Materials and Methods). The results are summarized in Table 1.
Relevant overexpression of CCND1 gene was found in 22 of the 50 MMs, of CCND2 in 23, and of CCND3 in five. Although only five MM cases showed CCND3 overexpression, it is worth noting that moderate expression levels could be detected in healthy individuals and all of the MM tumors, whereas the normal PCs showed no CCND1 or CCND2 expression. An appreciable combined expression of two cyclin D genes was observed in two cases (MM-067, MM-069), and only MM-089 showed the combined expression of all three. Overall, a significant overexpression of at least one type of cyclin D was found in 46 of 50 MM patients (92%). The samples were stratified as follows (Fig 1A): 11 patients in TC1, including 10 carrying t(11;14) and expressing high levels of CCND1 and one harboring a t(6;14) strongly deregulating CCND3; 11 patients in TC2, all of whom were negative for IGH translocations and overexpressing CCND1 at lower levels than TC1 (P = 1.3 x 10–3 for both probes), but not CCND2 or CCND3; 15 patients in TC3, all of whom were negative for the IGH translocation and CCND1 expression, most of whom (11 of 15) expressed CCND2 at variable levels, and one patient positive for CCND3; 10 patients in TC4, all harboring t(4;14); and three patients in TC5, all carrying deregulated MAF or MAFB genes. The patients in TC4 and TC5 were characterized by a consistent expression of CCND2, with the TC5 patients generally showing the highest levels.
The samples classification determined by the microarray data was validated using Q-RT-PCR in all four healthy donors, and 42 of the MM patients (Fig 1B; Table 1). The correspondence between the results of the two analyses was evaluated by assessing the correlation coefficients of cyclin D expression levels determined by microarray and Q-RT-PCR: the coefficients were 0.90 and 0.94 for the CCND1 probes; 0.74, 0.71 and 0.82 for the CCND2 probes; and 0.98 for the CCND3 probes, thus indicating an almost complete concordance for CCND1 and CCND3, and a satisfactory correspondence for the CCND2 gene (Appendix B). As shown in Figure 2A, Q-RT-PCR data indicated that TC1 cases showed significantly higher CCND1 expression levels than TC2 cases (P = 3.9 x 10–3). Moreover, the TC2 group showed significantly higher CCND1 expression levels than the TC3, TC4 and TC5 (P = 1.39 x 10–4, P = 1.05 x 10–4, and P = .022, respectively). The TC3 patients defined two different subgroups, one characterized by the absence of CCND2 expression and the other by heterogeneous CCND2 expression levels (Fig 2B). Similarly to microarray data, all patients in TC4 and TC5 were characterized by a consistent expression of CCND2, with the TC5 cases reaching the highest levels (Fig 2B).
Finally, concerning the clinical parameters (Appendix A), in our data set the only identified correlation with the determined TC classes was the prevalence of focal bone lesions in the patients overexpressing CCND1 (the TC1 and TC2 groups) in comparison with all the other MMs (P = .0231), thus confirming previously reported data.27
FISH Analysis
The results of the FISH analysis are reported in Table 1. Among the patients with t(11;14) included in the TC1 group, increased CCND2 and CCND3 copy numbers were only found in one of nine and two of nine studied cases, whereas the chromosome 13 deletion was found in four of 10 (40%).
The analysis of the cyclin D loci arrangements in the TC2 group showed that 9 of 11 patients (82%) had extra copies of the CCND1 locus, whereas none were positive for CCND2 polysomy and only 1 of 11 (9%) was positive for extra copies of CCND3. Notably, all but one of the TC2 patients were negative for the chromosome 13q deletion.
Twelve cases in the TC3 group were analyzed: extra copies of CCND1, CCND2 and CCND3 were found in, respectively, four (33%), three (25%) and eight patients (66%), whereas chromosome 13q deletion was found in eight (66%). The FISH analyses of the patients in the TC4 group showed extra copies of only CCND3 in 3 of 9 cases (33%), and 13q deletion in 8 of 9 (89%). Extra copies of all the cyclin D genes were found in both of the TC5 patients for whom material was available; these patients also had a chromosome 13q deletion. It is interesting that none of the 50 patients showed any structural alterations in the CCND2 locus, such as split centromeric and telomeric probes (see Materials and Methods).
A contingency analysis made to determine whether the number of cases with increased copies of cyclin D genes correlated with the assigned classes indicated that the samples with extra copies of CCND1 were more significantly represented in the TC2 than the TC3 (P = 6.3 x 10–3) or TC4 groups (P = 3 x 10–4), whereas extra copies of CCND3 were more significantly represented in the TC3 than the TC1 (P = .0281) or TC2 groups (P = .094). Globally, extra copies of CCND1 more strictly correlated with the TC2 group (9 of 11 with extra copies) when this was compared with all of the other available MM samples (11 of 34 with extra copies, P = 5.9 x 10–3). Only the TC2 group showed polysomy associated with the CCND1 locus but not the CCND2 or CCND3 loci; in the other TC groups, the significant presence of polysomy (eg, CCND3 in TC3 group; P = .0106) was not related to any single type of cyclin D genes.
Finally, chromosome 13q deletion was more represented in the TC4 group (eight of nine) than in all of the other MM samples (15 of 36; P = .0220), whereas its absence significantly correlated with the TC2 group (10 of 11) in comparison with all of the other MM cases (12 of 34; P = 1.7 x 10–3).
Analysis of Gene Expression Profiling Data
The stratified gene expression database was analyzed using the PAM software22 in order to identify the transcriptional fingerprints characterizing the five TC groups. The classifier was first trained on the entire MM data set in order to select the set of genes with the smallest estimated misclassification error by means of a cross-validation process. The minimum error corresponded to 51 differentially expressed probes ( = 5) specific for 36 genes, which globally characterized four of the TC groups (Appendix B). Figure 3 shows the hierarchical clustering of the selected probe sets.
The TC1, TC4 and TC5 groups were characterized mainly by the expression of genes previously identified by us (on a smaller data set and using different algorithms) as part of the specific transcriptional signatures respectively related to t(11;14), t(4;14), and t(14;16)/t(14;20).19
The TC2 group showed the specific expression of five marker transcripts, including one codifying for a novel protein containing a POZ domain (BTBD3) characteristic of transcriptional repressor factors, and two codifying for the L13a and L36 ribosomal proteins involved in ribosome biogenesis (RPL13A and RPL36).
No marker was differentially expressed in the TC3 group, thus suggesting the relative heterogeneity of the TC3 samples. As indicated by the dendrogram in Figure 3, the TC3 patients were divided in two main subgroups that respectively clustered with the TC2 and TC4/TC5 patients. Interestingly, the expression of the CCND2 probe sets was significantly higher in the subgroup clustering with TC4/TC5 (P < .05 for all the three probes).
Because no differentially expressed gene characterized the TC3 group, the 15 TC3 samples were excluded from the analysis in order to evaluate whether the TC1, TC4, TC5 and, particularly, the TC2 group could be better characterized by reducing the heterogeneity of the data set. PAM training and cross-validation results indicated retaining 112 probe sets ( = 4.6), distinguishing the TC1, TC2, TC4 and TC5 groups (Fig 4, left panel; Appendix B). In particular, the TC2 group was characterized by 30 genes, mainly involved in protein biosynthesis at translational level. Twelve ribosomal protein genes related to large (L13a, L36, L28, L18a, L35, L4 and L7a) and small (S19, S9, S15 S16 and S27) ribosome subunits, the translation initiation factor 3 (eIF3k), and the translation elongation factor 2 (EEF2) were overexpressed in the TC2 cases, as well as other genes involved in cell cycle regulation (PIK3R3, ISG20, KCNS3), Wnt signaling (FRZB1) or nuclear-cytoplasmic transport (NPM1). In particular, the putative oncogene NPM1 is involved in chromosomal translocations typical of large cell lymphoma, and in acute promyelocytic leukemia.28 In line with the results obtained by analyzing the five groups, the most differentially expressed probe in TC2 was that specific for the BTBD3 gene (Figs 3 and 4).
The right panel of Figure 4 shows the expression profiles in the TC3 cases of the identified probes, whose heterogeneous expression confirmed the previous observation (Fig 3). Furthermore, when the PAM model was used to classify the TC3 samples (as a test set), four patients were predicted as belonging to TC2, ten to TC4 and one to TC5 class (Appendix B). These data prompted us to verify whether the distribution of TC3 specimens in the various groups could be related to the expression levels of CCND1 or CCND2. A nonparametric test applied to the probes specific for the two genes showed that only CCND2 expression was statistically higher in the TC3 samples predicted as belonging to TC4 than in those predicted as belonging to TC2 (P = 2 x 10–3; Appendix B). Furthermore, the only TC3 sample predicted as belonging to TC5 (MM-036) had CCND2 levels comparable with those of the samples with deregulated MAF genes.
Validation of Gene Expression Signatures by Meta-Analysis of Independent Samples
The expression signatures identified in the proprietary database as characterizing the TC1, TC2, TC4 and TC5 groups were validated by means of a meta-analysis of an independent set of 74 MM cases (validation set) profiled on Affymetrix HuGeneFL microarrays.16
The validation set was stratified as follows: the TC1 samples (n = 15) were identified by the presence of t(11;14) or the ectopic overexpression of the CCND3 gene; the TC2 samples (n = 25) by their relevant CCND1 expression (higher than the mean value plus three standard deviations of the corresponding expression in normal PCs) and the absence of t(11;14); the TC4 samples (n = 9) by their ectopic expression of the FGFR3 gene; the TC5 samples (n = 7) by their combined expression of CCND2, ITG?7 and CX3CR1,29 which were significantly different from the normal PC samples; and the TC3 samples (n = 15) by their CCND2 expression and the absence of CCND1, together with the absence of the genes characterizing the TC4 and TC5 groups.
To allow classification, the probes' IDs in different arrays were correlated using Affymetrix probe match spreadsheets (Appendix C); 50 probes were the best-match HuGeneFL counterparts for the 112 HG-U133A predictor transcripts. A PAM model was designed using all 50 matching probes trained on the proprietary samples, and was then used to classify the validation set samples into the TC1, TC2, TC4, and TC5 groups (because no marker probe characterized the TC3 group, the TC3 cases were excluded from the classification session). The model correctly recognized the "unseen" samples of the validation set, classifying 12 of 15 specimens into TC1 (the three not correctly predicted samples showed CCND3 but not CCND1 gene ectopic expression), 20 of 25 into TC2, nine of nine into TC4 and seven of seven into TC5 (Table 2). The classification accuracy measures (a global classification rate of 85.71%; mean specificity, 85.71%; and mean sensitivity, 90%) suggest that the identified TC expression signature is a highly conserved characteristic of the MM samples that is not affected by cohort- or lab-specific biases. Full details are provided in Appendix C.
In addition, we further verified the robustness of the identified gene expression signatures by directly analyzing the validation set through the same multi-class classification approach as that used for the proprietary data set, and by comparing the lists of transcripts obtained in the two analyses (Appendix C). In particular, also in Zhan's database we identified the specific upregulation in the TC2 group of several genes belonging to the translational machinery. Among the genes found to be particularly well preserved among independent data sets, we identified in TC2 RPS9, RPL35, eIF3k, FBL and GNB2L1, as well as FRZB, ISG20, COX5A, ATP5D and NPM1. Unfortunately, there was no specific probe for BTBD3 (the most differentially expressed transcript in the TC2 group in the multiclass analysis of the proprietary data set) on the HuGeneFL array.
Finally, the same interplatform comparisons were used to validate our results on an additional publicly available database.17 It is worth noting that, in addition to the verification of the translational machinery genes, also the BTBD3 transcript resulted to be significantly upregulated in the TC2 group (Appendix D).
DISCUSSION
This study extends previous evidence that cyclin D overexpression is an almost universal event in MM, and provides new insights into the molecular definition of MM based on cyclin D deregulation and IGH translocations.
Approximately 95% of the cases investigated by means of DNA microarrays overexpressed at least one of the three cyclin D genes, with a globally very good concordance with the Q-RT-PCR data. The TC1 and TC2 groups were characterized by the overexpression of CCND1, which was almost undetectable in the normal controls. In particular, low to moderate levels of CCND1 expression were found in 22% of the samples (TC2 group), which did not show either overexpression of CCND2 and CCND3 or known chromosomal translocations. A remarkable finding in this group was the significant occurrence of chromosome 11 polysomy. Similarly, Specht et al13 have recently reported such a correlation determined by Q-RT-PCR analysis in a series of 14 cases. However, whether polysomy 11 per se directly or indirectly causes increased CCND1 expression remains to be clarified; since in other patients included in our database chromosome 11 polysomy is not associated with the overexpression of the gene, we may suggest that other mechanisms may contribute to CCND1 deregulation.
The gene expression profiling analysis showed that the TC2 group is a well-defined entity at transcriptional level: in particular, the TC2 samples were characterized by the differential expression of 30 genes, most of which are involved in protein biosynthesis at translational level. This result was supported by meta-analyses on two independent MM databases, also showing that the TC2 group is characterized by the specific upregulation of the translational machinery pathway. A number of studies have indicated that the constitutive activation of signal transduction pathways and the inactivation of tumor suppressor genes lead to an abnormal activation of translational components.30 In addition, there are experimental evidences that the constitutive expression of some of these factors (such as eIF-2, eIF-3 and eIF-4) may induce transformation in vitro, as well as protect from apoptosis.31-34 Previous studies have shown that CCND1 overexpression is regulated not only at posttranscriptional, but also at translational level: in particular, under specific experimental conditions, translation factors may regulate the synthesis of factors, such as nuclear factor kappa B (NF-B), known to regulate the transcription of CCND1 positively.35,36 Although the mechanisms by which the translational machinery influences CCND1 expression remain to be fully elucidated, these previous observations may explain the moderate CCND1 expression levels in TC2 patients. Interestingly, the highly coordinated expression of ribosomal and translation-associated genes has been recently reported in human cancers investigated by means of gene expression profiling: particularly, they have been found to be upregulated in well-differentiated ovarian tumors37 and in a subset of chronic lymphocytic leukemia patients characterized by a more favorable clinical course.38 Recent studies have shown that CCND1 overexpression in patients carrying either the t(11;14) or chromosome 11 polysomy is a favorable prognostic factor associated with a prolonged remission and event-free survival after autologous transplantation.2,39 In addition, we have shown that the TC2 group is associated with the absence of the chromosome 13q deletion (known to be a negative prognostic factor).40 Further studies of larger cohorts with an adequate follow-up are needed to determine whether TC2 patients are characterized by a distinct prognosis, a finding that may contribute to the risk stratification of MM patients.
The most specifically modulated transcript in the TC2 group (BTBD3) contains a BTB/POZ domain, a conserved protein-protein interaction domain found in many zinc finger transcription factors. This domain has been reported to be associated with the transcriptional repression mediated by the promyelocytic leukemia zinc finger protein.41 In addition, it has been demonstrated that the BTB/POZ domain mediates the transcriptional repressor activity of the BCL6 oncogene, which is frequently deregulated in B-cell lymphomas.42 On the basis of these considerations, this transcript could play an important role in the pathogenesis of a subtype of MM patients.
Our results confirm the strong association between CCND2 overexpression and the presence of t(4;14) in TC4 or the deregulation of MAF or MAFB genes in TC5. Consistently with recent reports indicating that CCND2 is a transcriptional target of MAF protein,14 the highest expression levels were observed in the TC5 group. The mechanisms responsible for the upregulation of CCND2 in the TC4 group are not clear, but it is worth noting that our study did not identify chromosome 12 polysomy or CCND2 extra copies in any of the 10 cases with t(4;14).
The TC3 group has been defined as a miscellany of MM tumors that do not fall into any of the other four groups and do not show any of the known major translocations. In our study cohort, 10 of 15 patients included in the TC3 group (67%) expressed moderate or high levels of CCND2, whereas no deregulation of any cyclin D was observed in four cases. Furthermore, CCND2 expression levels could be statistically correlated with the gene expression–based prediction of TC3 cases assigned to the TC2 or TC4 groups. Because previous data have indicated that CCND2 overexpression in different types of lymphoid tumors is not associated with chromosomal aberrations,43 it is possible to hypothesize that CCND2 deregulation may be the result of epigenetic mechanisms in a significant fraction of MM cases. Overall, these data suggest that CCND2 deregulation may play a role in the molecular and biologic phenotype of the majority of patients included in the TC3 group, and raise the question of whether TC3 patients should be considered an independent group or distributed into the TC2, TC4 or TC5 groups.
Finally, no correlation between CCND3 overexpression and TC classification was found in our data set as overexpression was only found in the single t(6;14) patient and four additional patients stratified into different TC classes. A larger number of patients harboring t(6;14) or overexpressing CCND3 is required in order to establish whether a specific signature can be associated with CCND3 deregulation, thus leading to the definition of an additional TC subgroup of MM patients. We have recently identified molecular signatures associated with the presence of the main IGH translocations by means of gene expression profiling.19 The genes specifically expressed in the TC1 group only partially overlapped those previously described as part of the t(11;14) signature, a discrepancy that may be due to the presence of one sample carrying the t(6;14) translocation in the TC1 group.
Our data may provide important contributions to the understanding of the molecular and biologic features of distinct MM subtypes associated with different prognoses and treatment responses. In particular, together with the reported shorter survival of t(4;14) (TC4) and t(14;16) (TC5) patients receiving either standard- or high-dose therapy, and the longer and significantly better survival of t(11;14) MM cases (TC1) on high-dose conventional therapy,2 the identification of distinctive patterns of gene expression associated with the TC2 subtype may improve risk stratification and indicate novel therapeutic targets. In this context, Q-RT-PCR analyses of highly purified PCs populations provide a suitable approach to the characterization and stratification of MM patients. Furthermore, validation of the identified TC transcriptional signatures by means of the meta-analysis on almost 250 samples from two different MM data sets highlighted the robustness of microarray expression profiling as an integrated platform, thus suggesting that disease models can be efficiently analyzed through interplatform comparisons.
Appendix A
Appendix B
Stratification of proprietary samples in TC groups
The 50 MM samples were stratified into TC groups using molecular characterization and the differential levels of cyclins D expression in the microarray data (see Materials and Methods). Specifically, RMA intensity values of Affymetrix probes specific for CCND1 (208711_s_at, 208712_at), CCND2 (200951_s_at, 200952_s_at, 200953_s_at), and CCND3 (201700_at) were compared in healthy and MM populations. The cyclin genes were considered as overexpressed only if all of their representative probes presented an intensity level in MM samples higher than the threshold computed as the mean plus 3 standard deviations of the corresponding expression values in healthy donors (indicated by "+" in Table 2).
Similarly, the levels of CCND1, CCND2, and CCND3 over GAPDH ratios obtained from Q-RT-PCR ratios were analyzed in all healthy donors (n = 4) and in 42/50 MM patients, following the same procedure described for microarray data (Table 3).
Comparison of microarray and Q-RT-PCR data for cyclins D probes
The correspondence between TC classification obtained by microarray and Q-RT-PCR was evaluated assessing correlation coefficients of cyclins expression levels determined using the two different methods. Being on different intensity scales, both microarray and Q-RT-PCR data have been scaled in the range 0÷1, subtracting the minimum value from original data and dividing it by the data range (ie, the maximum minus the minimum value). The correlation coefficients of expression signals were 0.90 and 0.94 for CCND1; 0.74, 0.71 and 0.82 for CCND2; and 0.98 for CCND3 probes, respectively. The side-by-side comparison and the linear regression curve of cyclin expression levels obtained by microarray and Q-RT-PCR methods are shown in Figures 1A-C.
Identification of transcriptional fingerprints of TC groups: Details of statistical analysis
The stratified gene expression database was analyzed using Prediction Analysis of Microarrays (PAM). PAM uses the nearest shrunken centroids methods to select differentially expressed genes and perform multi-class classification Tibshirani et al.22 The classifier was first trained on the entire dataset to select relevant and the optimal value of chosen by a 10-fold cross-validation process. The training samples were randomly divided 10 times into 10 approximately equal-size, class-balanced parts, the model fitted on 90% of the samples and used to predict the class labels of the remaining 10%. This procedure was repeated 10 times, with each part playing the role of the test samples, and the errors on all 10 parts added together to compute the overall error as a function of . Details of PAM statistical analysis applied on the five TC groups are reported in Figures 2 and 3.
Details of PAM statistical analysis applied on the TC1, TC2, TC4, and TC5 groups only are reported in Figure 4. The description and class scores of the 112 probe sets retained by PAM model constructed on 4 TC groups are reported in Table 4. In Figure 5 is reported the PAM classification of TC3 samples on the basis of the 112 probes identified by applying the PAM model on four TC groups.
Appendix C
Meta-analysis on independent data set (I)
The 112 probe set expression signature identified in the proprietary database as characterizing TC1, TC2, TC4 and TC5 groups, was validated through a meta-analysis on an independent set of 74 MM cases (validation set) profiled on Affymetrix HuGeneFL microarrays (Zhan et al16). To modify this test dataset into data applicable to the predictor gene list from the HG-U133A arrays, the probe IDs between these different arrays were correlated using Affymetrix probe match spreadsheets. Since no conversion table is currently available for a direct comparison between HuGeneFL and HG-U133A chip designs, HuGeneFL probes were first matched to HG-U95Av2 using the HuGeneFL to Human Genome U95A comparison file, and then linked to the HG-U133A annotations through the HG-U95 to HG-U133 Best Match spreadsheet. This procedure generated a conversion query matching a total of 4,919 unique probe sets out of the original 7,070 HuGeneFL non-control probes and led to 50 best match HuGeneFL counterparts for the 112 HG-U133A predictor transcripts (Table 5).
The distribution of the validation set cases in different TC classes was performed comparing the expression levels of cyclins D and the genes targeted by IGH translocations in MM and in 30 normal samples as described for the proprietary database (ie, molecular characterization and cyclin D genes with an intensity level in MM samples higher than the threshold computed as the mean plus 3 standard deviations of the corresponding expression values in healthy donors). Table 6 reports original expression values (as in Zhan et al) for 30 healthy donors and 74 MM samples and TC assignments. TC1 samples (n = 15) were identified for the presence of t(11;14) or the ectopic overexpression of the CCND3 gene; TC2 (n = 25) samples were identified by relevant level of CCND1 expression and the absence of t(11;14); TC4 (n = 9) samples were identified for the expression of FGFR3 gene; TC5 (n = 7) samples were identified for the combined expression of CCND2, ITGB7 and CX3CR1, significantly different from healthy donor samples; TC3 (n = 15) samples were identified for the CCND2 expression and the absence of CCND1, together with the absence of genes characterizing TC4 and TC5 groups. Three samples could not be assigned due to the simultaneous overexpression of CCND1 and CCND2 genes (Fig 6).
A PAM model was designed using all the 50 matching probes (ie, always selecting a threshold parameter =0), trained on the proprietary samples, and used to classify the validation set expression signals into TC1, TC2, TC3, TC4, and TC5 groups. Figure 7 shows cumulative (A) and class-individual (B) cross-validation errors of the PAM model. In Table 7 are reported the classification probabilities for the class prediction of all Zhan's samples.
The stratified gene expression database from Zhan et al was also directly analyzed using PAM. The classifier was first trained on the entire data set (eg, 4,077 probes after filtering out data with Absent calls in all of the arrays; no filtering procedure was applied to the intensity levels) to select relevant genes marking TC1, TC2, TC4, and TC5 groups. The training samples were randomly divided 10 times into 10 approximately equal-size, class-balanced parts, the model fitted on 90% of the samples and used to predict the class labels of the remaining 10%. This procedure was repeated 10 times, with each part playing the role of the test samples, and the errors on all 10 parts added together to compute the overall error as a function of . Details of PAM statistical analysis applied on the four TC groups are reported in Figures 8 and 9.
The description and class scores of the 135 unique HuGeneFL probes retained by PAM model constructed on 4 TC groups are reported in Table 8.
Appendix D
Meta-analysis on independent data set (II)
The 112 probe sets expression signature identified in the proprietary database as characterizing TC1, TC2, TC4 and TC5 groups, was validated through a meta-analysis on an additional independent set of 173 MM cases (validation set) profiled on Affymetrix HG-U95Av2 microarrays (see References, Tian et al, 2003). Using Affymetrix HG-U95 to HG-U133 Best Match spreadsheet, the 112 HG-U133A probe sets identified in the proprietary database were converted into 91 HG-U95Av2 best match probes. In Table 9 is reported the full list of the inter-platform common probes.
The data set from Tian et al (Gene Expression Omnibus, accession number GSE755) contained signal intensities and detection calls for 173 MM samples. The signal intensities were calculated using Affymetrix MAS 5.01 algorithm. No cytogenetic characteristics or data for normal PCs were available. However, the samples were from the same laboratory of Zhan et al, 2002 (Appendix C) and 54 out of 173 reported the very same description label and cyclins D characteristics as in Zhan et al. Thus, anchoring on these 54 common cases, putative TC class was assigned thresholding intensity signal of cyclins D probe sets (Fig 10) and controlling known marker genes of chromosomal translocations (Mattioli et al, 2005). A total of 32 samples over-expressed CCND1 or CCND3 and were assigned to TC1; 56 presented CCND1 expression but no CCND2 or CCND3 and were thus labeled as TC2; 20 cases were assigned to TC4 (based on CCND2 and FGFR3 expression levels), and 5 overexpressing CCND2, ITG?7 and CX3CR1 to TC5. The remaining 60 specimens were either labeled as TC3 (n = 48) or left unassigned (n = 12).
A PAM model was designed using all the 91 matching probes (ie, always selecting a threshold parameter = 0), trained on the proprietary samples, and used to classify the validation set samples into TC groups. Figure 11 shows cumulative (A) and class-individual (B) cross-validation errors of the PAM model. In Table 10 are reported the classification accuracy measures for TC1, TC2, TC4 and TC5 class (113 samples, excluded TC3) and in Table 11 the classification probabilities for the class prediction of the 161 out of 173 Tian et al samples for which was possible to assign a putative TC.
The stratified gene expression database from Tian et al was also directly analyzed using PAM. The classifier was first trained on the entire data set (eg, 10599 probes after filtering out data with Absent calls in all of the arrays; no filtering procedure was applied to the intensity levels) to select relevant genes marking TC1, TC2, TC4, and TC5 groups. The training samples were randomly divided 10 times into 10 approximately equal-size, class-balanced parts, the model fitted on 90% of the samples and used to predict the class labels of the remaining 10%. This procedure was repeated 10 times, with each part playing the role of the test samples, and the errors on all 10 parts added together to compute the overall error as a function of . Details of PAM statistical analysis applied on the four TC groups are reported in Figures 12 and 13.
The description and class scores of the 204 unique HG-U95Av2 probes retained by PAM model constructed on 4 TC groups are reported in Table 12.
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
NOTES
Supported by a grant from the Associazione Italiana Ricerca sul Cancro (AIRC) to A.N. and by the Fondazione Assistenza e Studio Malati Ematologici and the Italian Ministry of Health; by Fondo per gli Investimeni della Ricerca de Base Ministero Istruzione Università e Ricerca Grants No. RBNE01TZZ8 and RBAU01935A to S.B.
L.A. and S.B. contributed equally to this study.
Terms in blue are defined in the glossary, found at the end of this issue and online at www.jco.org.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Seidl S, Kaufmann H, Drach J: New insights into the pathophysiology of multiple myeloma. Lancet Oncol 4:557-564, 2003
Fonseca R, Barlogie B, Bataille R, et al: Genetics and cytogenetics of multiple myeloma: A workshop report. Cancer Res 64:1546-1558, 2004
Chesi M, Bergsagel PL, Brents LA, et al: Dysregulation of cyclin D1 by translocation into an IgH gamma switch region in two multiple myeloma cell lines. Blood 88:674-681, 1996
Ronchetti D, Finelli P, Richelda R, et al: Molecular analysis of 11q13 breakpoints in multiple myeloma. Blood 93:1330-1337, 1999
Chesi M, Nardini E, Brents LA, et al: Frequent translocation t(4;14)(p16.3;q32.3) in multiple myeloma is associated with increased expression and activating mutations of fibroblast growth factor receptor 3. Nat Genet 16:260-264, 1997
Richelda R, Ronchetti D, Baldini L, et al: A novel chromosomal translocation t(4; 14)(p16.3; q32) in multiple myeloma involves the fibroblast growth-factor receptor 3 gene. Blood 90:4062-4070, 1997
Chesi M, Nardini E, Lim RS, et al: The t(4;14) translocation in myeloma dysregulates both FGFR3 and a novel gene, MMSET, resulting in IgH/MMSET hybrid transcripts. Blood 92:3025-3034, 1998
Shaughnessy J Jr, Gabrea A, Qi Y, et al: Cyclin D3 at 6p21 is dysregulated by recurrent chromosomal translocations to immunoglobulin loci in multiple myeloma. Blood 98:217-223, 2001
Chesi M, Bergsagel PL, Shonukan OO, et al: Frequent dysregulation of the c-maf proto-oncogene at 16q23 by translocation to an Ig locus in multiple myeloma. Blood 91:4457-4463, 1998
Hanamura I, Iida S, Akano Y, et al: Ectopic expression of MAFB gene in human myeloma cells carrying (14;20)(q32;q11) chromosomal translocations. Jpn J Cancer Res 92:638-644, 2001
Malumbres M, Barbacid M: To cycle or not to cycle: A critical decision in cancer. Nat Rev Cancer 1:222-231, 2001
Pruneri G, Fabris S, Baldini L, et al: Immunohistochemical analysis of cyclin D1 shows deregulated expression in multiple myeloma with the t(11;14). Am J Pathol 156:1505-1513, 2000
Specht K, Haralambieva E, Bink K, et al: Different mechanisms of cyclin D1 overexpression in multiple myeloma revealed by fluorescence in situ hybridization and quantitative analysis of mRNA levels. Blood 104:1120-1126, 2004
Hurt EM, Wiestner A, Rosenwald A, et al: Overexpression of c-maf is a frequent oncogenic event in multiple myeloma that promotes proliferation and pathological interactions with bone marrow stroma. Cancer Cell 5:191-199, 2004
Hideshima T, Bergsagel PL, Kuehl WM, et al: Advances in biology of multiple myeloma: Clinical applications. Blood 104:607-618, 2004
Zhan F, Hardin J, Kordsmeier B, et al: Global gene expression profiling of multiple myeloma, monoclonal gammopathy of undetermined significance, and normal bone marrow plasma cells. Blood 99:1745-1757, 2002
Tian E, Zhan F, Walker R, et al: The role of the Wnt-signaling antagonist DKK1 in the development of osteolytic lesions in multiple myeloma. N Engl J Med 349:2483-2494, 2003
Fabris S, Agnelli L, Mattioli M, et al: Characterization of oncogene dysregulation in multiple myeloma by combined FISH and DNA microarray analyses. Genes Chromosomes Cancer 42:117-127, 2005
Mattioli M, Agnelli L, Fabris S, et al: Gene expression profiling of plasma cell dyscrasias reveals molecular patterns associated with distinct IGH translocations in multiple myeloma. Oncogene 24:2461-2473, 2005
Durie BG, Salmon SE: A clinical staging system for multiple myeloma: Correlation of measured myeloma cell mass with presenting clinical features, response to treatment, and survival. Cancer 36:842-854, 1975
Irizarry RA, Hobbs B, Collin F, et al: Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 4:249-264, 2003
Tibshirani R, Hastie T, Narasimhan B, et al: Diagnosis of multiple cancer types by shrunken centroids of gene expression. Proc Natl Acad Sci U S A 99:6567-6572, 2002
Eisen MB, Spellman PT, Brown PO, et al: Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A 95:14863-14868, 1998
Schadt EE, Li C, Ellis B, et al: Feature extraction and normalization algorithms for high-density oligonucleotide gene expression array data. J Cell Biochem Suppl 37:120-125, 2001
Heid CA, Stevens J, Livak KJ, et al: Real time quantitative PCR. Genome Res 6:986-994, 1996
Elnenaei MO, Hamoudi RA, Swansbury J, et al: Delineation of the minimal region of loss at 13q14 in multiple myeloma. Genes Chromosomes Cancer 36:99-106, 2003
Robbiani DF, Chesi M, Bergsagel PL: Bone lesions in molecular subtypes of multiple myeloma. N Engl J Med 351:197-198, 2004
Falini B, Mecucci C, Tiacci E, et al: Cytoplasmic nucleophosmin in acute myelogenous leukemia with a normal karyotype. N Engl J Med 352:254-266, 2005
Bergsagel PL, Kuehl WM: Critical roles for immunoglobulin translocations and cyclin D dysregulation in multiple myeloma. Immunol Rev 194:96-104, 2003
Rosenwald IB: The role of translation in neoplastic transformation from a pathologist's point of view. Oncogene 23:3230-3247, 2004
Donze O, Jagus R, Koromilas AE, et al: Abrogation of translation initiation factor eIF-2 phosphorylation causes malignant transformation of NIH 3T3 cells. EMBO J 14:3828-3834, 1995
Mayeur GL, Hershey JW: Malignant transformation by the eukaryotic translation initiation factor 3 subunit p48 (eIF3e). FEBS Lett 514:49-54, 2002
Morley SJ: The regulation of eIF4F during cell growth and cell death. Prog Mol Subcell Biol 27:1-37, 2001
Thornton S, Anand N, Purcell D, et al: Not just for housekeeping: Protein initiation and elongation factors in cell growth and tumorigenesis. J Mol Med 81:536-548, 2003
Guttridge DC, Albanese C, Reuther JY, et al: NF-kappaB controls cell growth and differentiation through transcriptional regulation of cyclin D1. Mol Cell Biol 19:5785-5799, 1999
Hinz M, Krappmann D, Eichten A, et al: NF-kappaB function in growth control: Regulation of cyclin D1 expression and G0/G1-to-S-phase transition. Mol Cell Biol 19:2690-2698, 1999
Welsh JB, Zarrinkar PP, Sapinoso LM, et al: Analysis of gene expression profiles in normal and neoplastic ovarian tissue samples identifies candidate molecular markers of epithelial ovarian cancer. Proc Natl Acad Sci U S A 98:1176-1181, 2001
Durig J, Nuckel H, Huttmann A, et al: Expression of ribosomal and translation-associated genes is correlated with a favorable clinical course in chronic lymphocytic leukemia. Blood 101:2748-2755, 2003
Soverini S, Cavo M, Cellini C, et al: Cyclin D1 overexpression is a favorable prognostic variable for newly diagnosed multiple myeloma patients treated with high-dose chemotherapy and single or double autologous transplantation. Blood 102:1588-1594, 2003
Fonseca R, Blood E, Rue M, et al: Clinical and biologic implications of recurrent genomic aberrations in myeloma. Blood 101:4569-4575, 2003
Collins T, Stone JR, Williams AJ: All in the family: The BTB/POZ, KRAB, and SCAN domains. Mol Cell Biol 21:3609-3615, 2001
Polo JM, Dell'oso T, Ranuncolo SM, et al: Specific peptide interference reveals BCL6 transcriptional and oncogenic mechanisms in B-cell lymphoma cells. Nat Med 10:1329-1335, 2004
Delmer A, Ajchenbaum-Cymbalista F, Tang R, et al: Overexpression of cyclin D2 in chronic B-cell malignancies. Blood 85:2870-2876, 1995(Luca Agnelli, Silvio Bicc)